1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118
5119
5120
5121
5122
5123
5124
5125
5126
5127
5128
5129
5130
5131
5132
5133
5134
5135
5136
5137
5138
5139
5140
5141
5142
5143
5144
5145
5146
5147
5148
5149
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160
5161
5162
5163
5164
5165
5166
5167
5168
5169
5170
5171
5172
5173
5174
5175
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220
5221
5222
5223
5224
5225
5226
5227
5228
5229
5230
5231
5232
5233
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
5271
5272
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289
5290
5291
5292
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308
5309
5310
5311
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334
5335
5336
5337
5338
5339
5340
5341
5342
5343
5344
5345
5346
5347
5348
5349
5350
5351
5352
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364
5365
5366
5367
5368
5369
5370
5371
5372
5373
5374
5375
5376
5377
5378
5379
5380
5381
5382
5383
5384
5385
5386
5387
5388
5389
5390
5391
5392
5393
5394
5395
5396
5397
5398
5399
5400
5401
5402
5403
5404
5405
5406
5407
5408
5409
5410
5411
5412
5413
5414
5415
5416
5417
5418
5419
5420
5421
5422
5423
5424
5425
5426
5427
5428
5429
5430
5431
5432
5433
5434
5435
5436
5437
5438
5439
5440
5441
5442
5443
5444
5445
5446
5447
5448
5449
5450
5451
5452
5453
5454
5455
5456
5457
5458
5459
5460
5461
5462
5463
5464
5465
5466
5467
5468
5469
5470
5471
5472
5473
5474
5475
5476
5477
5478
5479
5480
5481
5482
5483
5484
5485
5486
5487
5488
5489
5490
5491
5492
5493
5494
5495
5496
5497
5498
5499
5500
5501
5502
5503
5504
5505
5506
5507
5508
5509
5510
5511
5512
5513
5514
5515
5516
5517
5518
5519
5520
5521
5522
5523
5524
5525
5526
5527
5528
5529
5530
5531
5532
5533
5534
5535
5536
5537
5538
5539
5540
5541
5542
5543
5544
5545
5546
5547
5548
5549
5550
5551
5552
5553
5554
5555
5556
5557
5558
5559
5560
5561
5562
5563
5564
5565
5566
5567
5568
5569
5570
5571
5572
5573
5574
5575
5576
5577
5578
5579
5580
5581
5582
5583
5584
5585
5586
5587
5588
5589
5590
5591
5592
5593
5594
5595
5596
5597
5598
5599
5600
5601
5602
5603
5604
5605
5606
5607
5608
5609
5610
5611
5612
5613
5614
5615
5616
5617
5618
5619
5620
5621
5622
5623
5624
5625
5626
5627
5628
5629
5630
5631
5632
5633
5634
5635
5636
5637
5638
5639
5640
5641
5642
5643
5644
5645
5646
5647
5648
5649
5650
5651
5652
5653
5654
5655
5656
5657
5658
5659
5660
5661
5662
5663
5664
5665
5666
5667
5668
5669
5670
5671
5672
5673
5674
5675
5676
5677
5678
5679
5680
5681
5682
5683
5684
5685
5686
5687
5688
5689
5690
5691
5692
5693
5694
5695
5696
5697
5698
5699
5700
5701
5702
5703
5704
5705
5706
5707
5708
5709
5710
5711
5712
5713
5714
5715
5716
5717
5718
5719
5720
5721
5722
5723
5724
5725
5726
5727
5728
5729
5730
5731
5732
5733
5734
5735
5736
5737
5738
5739
5740
5741
5742
5743
5744
5745
5746
5747
5748
5749
5750
5751
5752
5753
5754
5755
5756
5757
5758
5759
5760
5761
5762
5763
5764
5765
5766
5767
5768
5769
5770
5771
5772
5773
5774
5775
5776
5777
5778
5779
5780
5781
5782
5783
5784
5785
5786
5787
5788
5789
5790
5791
5792
5793
5794
5795
5796
5797
5798
5799
5800
5801
5802
5803
5804
5805
5806
5807
5808
5809
5810
5811
5812
5813
5814
5815
5816
5817
5818
5819
5820
5821
5822
5823
5824
5825
5826
5827
5828
5829
5830
5831
5832
5833
5834
5835
5836
5837
5838
5839
5840
5841
5842
5843
5844
5845
5846
5847
5848
5849
5850
5851
5852
5853
5854
5855
5856
5857
5858
5859
5860
5861
5862
5863
5864
5865
5866
5867
5868
5869
5870
5871
5872
5873
5874
5875
5876
5877
5878
5879
5880
5881
5882
5883
5884
5885
5886
5887
5888
5889
5890
5891
5892
5893
5894
5895
5896
5897
5898
5899
5900
5901
5902
5903
5904
5905
5906
5907
5908
5909
5910
5911
5912
5913
5914
5915
5916
5917
5918
5919
5920
5921
5922
5923
5924
5925
5926
5927
5928
5929
5930
5931
5932
5933
5934
5935
5936
5937
5938
5939
5940
5941
5942
5943
5944
5945
5946
5947
5948
5949
5950
5951
5952
5953
5954
5955
5956
5957
5958
5959
5960
5961
5962
5963
5964
5965
5966
5967
5968
5969
5970
5971
5972
5973
5974
5975
5976
5977
5978
5979
5980
5981
5982
5983
5984
5985
5986
5987
5988
5989
5990
5991
5992
5993
5994
5995
5996
5997
5998
5999
6000
6001
6002
6003
6004
6005
6006
6007
6008
6009
6010
6011
6012
6013
6014
6015
6016
6017
6018
6019
6020
6021
6022
6023
6024
6025
6026
6027
6028
6029
6030
6031
6032
6033
6034
6035
6036
6037
6038
6039
6040
6041
6042
6043
6044
6045
6046
6047
6048
6049
6050
6051
6052
6053
6054
6055
6056
6057
6058
6059
6060
6061
6062
6063
6064
6065
6066
6067
6068
6069
6070
6071
6072
6073
6074
6075
6076
6077
6078
6079
6080
6081
6082
6083
6084
6085
6086
6087
6088
6089
6090
6091
6092
6093
6094
6095
6096
6097
6098
6099
6100
6101
6102
6103
6104
6105
6106
6107
6108
6109
6110
6111
6112
6113
6114
6115
6116
6117
6118
6119
6120
6121
6122
6123
6124
6125
6126
6127
6128
6129
6130
6131
6132
6133
6134
6135
6136
6137
6138
6139
6140
6141
6142
6143
6144
6145
6146
6147
6148
6149
6150
6151
6152
6153
6154
6155
6156
6157
6158
6159
6160
6161
6162
6163
6164
6165
6166
6167
6168
6169
6170
6171
6172
6173
6174
6175
6176
6177
6178
6179
6180
6181
6182
6183
6184
6185
6186
6187
6188
6189
6190
6191
6192
6193
6194
6195
6196
6197
6198
6199
6200
6201
6202
6203
6204
6205
6206
6207
6208
6209
6210
6211
6212
6213
6214
6215
6216
6217
6218
6219
6220
6221
6222
6223
6224
6225
6226
6227
6228
6229
6230
6231
6232
6233
6234
6235
6236
6237
6238
6239
6240
6241
6242
6243
6244
6245
6246
6247
6248
6249
6250
6251
6252
6253
6254
6255
6256
6257
6258
6259
6260
6261
6262
6263
6264
6265
6266
6267
6268
6269
6270
6271
6272
6273
6274
6275
6276
6277
6278
6279
6280
6281
6282
6283
6284
6285
6286
6287
6288
6289
6290
6291
6292
6293
6294
6295
6296
6297
6298
6299
6300
6301
6302
6303
6304
6305
6306
6307
6308
6309
6310
6311
6312
6313
6314
6315
6316
6317
6318
6319
6320
6321
6322
6323
6324
6325
6326
6327
6328
6329
6330
6331
6332
6333
6334
6335
6336
6337
6338
6339
6340
6341
6342
6343
6344
6345
6346
6347
6348
6349
6350
6351
6352
6353
6354
6355
6356
6357
6358
6359
6360
6361
6362
6363
6364
6365
6366
6367
6368
6369
6370
6371
6372
6373
6374
6375
6376
6377
6378
6379
6380
6381
6382
6383
6384
6385
6386
6387
6388
6389
6390
6391
6392
6393
6394
6395
6396
6397
6398
6399
6400
6401
6402
6403
6404
6405
6406
6407
6408
6409
6410
6411
6412
6413
6414
6415
6416
6417
6418
6419
6420
6421
6422
6423
6424
6425
6426
6427
6428
6429
6430
6431
6432
6433
6434
6435
6436
6437
6438
6439
6440
6441
6442
6443
6444
6445
6446
6447
6448
6449
6450
6451
6452
6453
6454
6455
6456
6457
6458
6459
6460
6461
6462
6463
6464
6465
6466
6467
6468
6469
6470
6471
6472
6473
6474
6475
6476
6477
6478
6479
6480
6481
6482
6483
6484
6485
6486
6487
6488
6489
6490
6491
6492
6493
6494
6495
6496
6497
6498
6499
6500
6501
6502
6503
6504
6505
6506
6507
6508
6509
6510
6511
6512
6513
6514
6515
6516
6517
6518
6519
6520
6521
6522
6523
6524
6525
6526
6527
6528
6529
6530
6531
6532
6533
6534
6535
6536
6537
6538
6539
6540
6541
6542
6543
6544
6545
6546
6547
6548
6549
6550
6551
6552
6553
6554
6555
6556
6557
6558
6559
6560
6561
6562
6563
6564
6565
6566
6567
6568
6569
6570
6571
6572
6573
6574
6575
6576
6577
6578
6579
6580
6581
6582
6583
6584
6585
6586
6587
6588
6589
6590
6591
6592
6593
6594
6595
6596
6597
6598
6599
6600
6601
6602
6603
6604
6605
6606
6607
6608
6609
6610
6611
6612
6613
6614
6615
6616
6617
6618
6619
6620
6621
6622
6623
6624
6625
6626
6627
6628
6629
6630
6631
6632
6633
6634
6635
6636
6637
6638
6639
6640
6641
6642
6643
6644
6645
6646
6647
6648
6649
6650
6651
6652
6653
6654
6655
6656
6657
6658
6659
6660
6661
6662
6663
6664
6665
6666
6667
6668
6669
6670
6671
6672
6673
6674
6675
6676
6677
6678
6679
6680
6681
6682
6683
6684
6685
6686
6687
6688
6689
6690
6691
6692
6693
6694
6695
6696
6697
6698
6699
6700
6701
6702
6703
6704
6705
6706
6707
6708
6709
6710
6711
6712
6713
6714
6715
6716
6717
6718
6719
6720
6721
6722
6723
6724
6725
6726
6727
6728
6729
6730
6731
6732
6733
6734
6735
6736
6737
6738
6739
6740
6741
6742
6743
6744
6745
6746
6747
6748
6749
6750
6751
6752
6753
6754
6755
6756
6757
6758
6759
6760
6761
6762
6763
6764
6765
6766
6767
6768
6769
6770
6771
6772
6773
6774
6775
6776
6777
6778
6779
6780
6781
6782
6783
6784
6785
6786
6787
6788
6789
6790
6791
6792
6793
6794
6795
6796
6797
6798
6799
6800
6801
6802
6803
6804
6805
6806
6807
6808
6809
6810
6811
6812
6813
6814
6815
6816
6817
6818
6819
6820
6821
6822
6823
6824
6825
6826
6827
6828
6829
6830
6831
6832
6833
6834
6835
6836
6837
6838
6839
6840
6841
6842
6843
6844
6845
6846
6847
6848
6849
6850
6851
6852
6853
6854
6855
6856
6857
6858
6859
6860
6861
6862
6863
6864
6865
6866
6867
6868
6869
6870
6871
6872
6873
6874
6875
6876
6877
6878
6879
6880
6881
6882
6883
6884
6885
6886
6887
6888
6889
6890
6891
6892
6893
6894
6895
6896
6897
6898
6899
6900
6901
6902
6903
6904
6905
6906
6907
6908
6909
6910
6911
6912
6913
6914
6915
6916
6917
6918
6919
6920
6921
6922
6923
6924
6925
6926
6927
6928
6929
6930
6931
6932
6933
6934
6935
6936
6937
6938
6939
6940
6941
6942
6943
6944
6945
6946
6947
6948
6949
6950
6951
6952
6953
6954
6955
6956
6957
6958
6959
6960
6961
6962
6963
6964
6965
6966
6967
6968
6969
6970
6971
6972
6973
6974
6975
6976
6977
6978
6979
6980
6981
6982
6983
6984
6985
6986
6987
6988
6989
6990
6991
6992
6993
6994
6995
6996
6997
6998
6999
7000
7001
7002
7003
7004
7005
7006
7007
7008
7009
7010
7011
7012
7013
7014
7015
7016
7017
7018
7019
7020
7021
7022
7023
7024
7025
7026
7027
7028
7029
7030
7031
7032
7033
7034
7035
7036
7037
7038
7039
7040
7041
7042
7043
7044
7045
7046
7047
7048
7049
7050
7051
7052
7053
7054
7055
7056
7057
7058
7059
7060
7061
7062
7063
7064
7065
7066
7067
7068
7069
7070
7071
7072
7073
7074
7075
7076
7077
7078
7079
7080
7081
7082
7083
7084
7085
7086
7087
7088
7089
7090
7091
7092
7093
7094
7095
7096
7097
7098
7099
7100
7101
7102
7103
7104
7105
7106
7107
7108
7109
7110
7111
7112
7113
7114
7115
7116
7117
7118
7119
7120
7121
7122
7123
7124
7125
7126
7127
7128
7129
7130
7131
7132
7133
7134
7135
7136
7137
7138
7139
7140
7141
7142
7143
7144
7145
7146
7147
7148
7149
7150
7151
7152
7153
7154
7155
7156
7157
7158
7159
7160
7161
7162
7163
7164
7165
7166
7167
7168
7169
7170
7171
7172
7173
7174
7175
7176
7177
7178
7179
7180
7181
7182
7183
7184
7185
7186
7187
7188
7189
7190
7191
7192
7193
7194
7195
7196
7197
7198
7199
7200
7201
7202
7203
7204
7205
7206
7207
7208
7209
7210
7211
7212
7213
7214
7215
7216
7217
7218
7219
7220
7221
7222
7223
7224
7225
7226
7227
7228
7229
7230
7231
7232
7233
7234
7235
7236
7237
7238
7239
7240
7241
7242
7243
7244
7245
7246
7247
7248
7249
7250
7251
7252
7253
7254
7255
7256
7257
7258
7259
7260
7261
7262
7263
7264
7265
7266
7267
7268
7269
7270
7271
7272
7273
7274
7275
7276
7277
7278
7279
7280
7281
7282
7283
7284
7285
7286
7287
7288
7289
7290
7291
7292
7293
7294
7295
7296
7297
7298
7299
7300
7301
7302
7303
7304
7305
7306
7307
7308
7309
7310
7311
7312
7313
7314
7315
7316
7317
7318
7319
7320
7321
7322
7323
7324
7325
7326
7327
7328
7329
7330
7331
7332
7333
7334
7335
7336
7337
7338
7339
7340
7341
7342
7343
7344
7345
7346
7347
7348
7349
7350
7351
7352
7353
7354
7355
7356
7357
7358
7359
7360
7361
7362
7363
7364
7365
7366
7367
7368
7369
7370
7371
7372
7373
7374
7375
7376
7377
7378
7379
7380
7381
7382
7383
7384
7385
7386
7387
7388
7389
7390
7391
7392
7393
7394
7395
7396
7397
7398
7399
7400
7401
7402
7403
7404
7405
7406
7407
7408
7409
7410
7411
7412
7413
7414
7415
7416
7417
7418
7419
7420
7421
7422
7423
7424
7425
7426
7427
7428
7429
7430
7431
7432
7433
7434
7435
7436
7437
7438
7439
7440
7441
7442
7443
7444
7445
7446
7447
7448
7449
7450
7451
7452
7453
7454
7455
7456
7457
7458
7459
7460
7461
7462
7463
7464
7465
7466
7467
7468
7469
7470
7471
7472
7473
7474
7475
7476
7477
7478
7479
7480
7481
7482
7483
7484
7485
7486
7487
7488
7489
7490
7491
7492
7493
7494
7495
7496
7497
7498
7499
7500
7501
7502
7503
7504
7505
7506
7507
7508
7509
7510
7511
7512
7513
7514
7515
7516
7517
7518
7519
7520
7521
7522
7523
7524
7525
7526
7527
7528
7529
7530
7531
7532
7533
7534
7535
7536
7537
7538
7539
7540
7541
7542
7543
7544
7545
7546
7547
7548
7549
7550
7551
7552
7553
7554
7555
7556
7557
7558
7559
7560
7561
7562
7563
7564
7565
7566
7567
7568
7569
7570
7571
7572
7573
7574
7575
7576
7577
7578
7579
7580
7581
7582
7583
7584
7585
7586
7587
7588
7589
7590
7591
7592
7593
7594
7595
7596
7597
7598
7599
7600
7601
7602
7603
7604
7605
7606
7607
7608
7609
7610
7611
7612
7613
7614
7615
7616
7617
7618
7619
7620
7621
7622
7623
7624
7625
7626
7627
7628
7629
7630
7631
7632
7633
7634
7635
7636
7637
7638
7639
7640
7641
7642
7643
7644
7645
7646
7647
7648
7649
7650
7651
7652
7653
7654
7655
7656
7657
7658
7659
7660
7661
7662
7663
7664
7665
7666
7667
7668
7669
7670
7671
7672
7673
7674
7675
7676
7677
7678
7679
7680
7681
7682
7683
7684
7685
7686
7687
7688
7689
7690
7691
7692
7693
7694
7695
7696
7697
7698
7699
7700
7701
7702
7703
7704
7705
7706
7707
7708
7709
7710
7711
7712
7713
7714
7715
7716
7717
7718
7719
7720
7721
7722
7723
7724
7725
7726
7727
7728
7729
7730
7731
7732
7733
7734
7735
7736
7737
7738
7739
7740
7741
7742
7743
7744
7745
7746
7747
7748
7749
7750
7751
7752
7753
7754
7755
7756
7757
7758
7759
7760
7761
7762
7763
7764
7765
7766
7767
7768
7769
7770
7771
7772
7773
7774
7775
7776
7777
7778
7779
7780
7781
7782
7783
7784
7785
7786
7787
7788
7789
7790
7791
7792
7793
7794
7795
7796
7797
7798
7799
7800
7801
7802
7803
7804
7805
7806
7807
7808
7809
7810
7811
7812
7813
7814
7815
7816
7817
7818
7819
7820
7821
7822
7823
7824
7825
7826
7827
7828
7829
7830
7831
7832
7833
7834
7835
7836
7837
7838
7839
7840
7841
7842
7843
7844
7845
7846
7847
7848
7849
7850
7851
7852
7853
7854
7855
7856
7857
7858
7859
7860
7861
7862
7863
7864
7865
7866
7867
7868
7869
7870
7871
7872
7873
7874
7875
7876
7877
7878
7879
7880
7881
7882
7883
7884
7885
7886
7887
7888
7889
7890
7891
7892
7893
7894
7895
7896
7897
7898
7899
7900
7901
7902
7903
7904
7905
7906
7907
7908
7909
7910
7911
7912
7913
7914
7915
7916
7917
7918
7919
7920
7921
7922
7923
7924
7925
7926
7927
7928
7929
7930
7931
7932
7933
7934
7935
7936
7937
7938
7939
7940
7941
7942
7943
7944
7945
7946
7947
7948
7949
7950
7951
7952
7953
7954
7955
7956
7957
7958
7959
7960
7961
7962
7963
7964
7965
7966
7967
7968
7969
7970
7971
7972
7973
7974
7975
7976
7977
7978
7979
7980
7981
7982
7983
7984
7985
7986
7987
7988
7989
7990
7991
7992
7993
7994
7995
7996
7997
7998
7999
8000
8001
8002
8003
8004
8005
8006
8007
8008
8009
8010
8011
8012
8013
8014
8015
8016
8017
8018
8019
8020
8021
8022
8023
8024
8025
8026
8027
8028
8029
8030
8031
8032
8033
8034
8035
8036
8037
8038
8039
8040
8041
8042
8043
8044
8045
8046
8047
8048
8049
8050
8051
8052
8053
8054
8055
8056
8057
8058
8059
8060
8061
8062
8063
8064
8065
8066
8067
8068
8069
8070
8071
8072
8073
8074
8075
8076
8077
8078
8079
8080
8081
8082
8083
8084
8085
8086
8087
8088
8089
8090
8091
8092
8093
8094
8095
8096
8097
8098
8099
8100
8101
8102
8103
8104
8105
8106
8107
8108
8109
8110
8111
8112
8113
8114
8115
8116
8117
8118
8119
8120
8121
8122
8123
8124
8125
8126
8127
8128
8129
8130
8131
8132
8133
8134
8135
8136
8137
8138
8139
8140
8141
8142
8143
8144
8145
8146
8147
8148
8149
8150
8151
8152
8153
8154
8155
8156
8157
8158
8159
8160
8161
8162
8163
8164
8165
8166
8167
8168
8169
8170
8171
8172
8173
8174
8175
8176
8177
8178
8179
8180
8181
8182
8183
8184
8185
8186
8187
8188
8189
8190
8191
8192
8193
8194
8195
8196
8197
8198
8199
8200
8201
8202
8203
8204
8205
8206
8207
8208
8209
8210
8211
8212
8213
8214
8215
8216
8217
8218
8219
8220
8221
8222
8223
8224
8225
8226
8227
8228
8229
8230
8231
8232
8233
8234
8235
8236
8237
8238
8239
8240
8241
8242
8243
8244
8245
8246
8247
8248
8249
8250
8251
8252
8253
8254
8255
8256
8257
8258
8259
8260
8261
8262
8263
8264
8265
8266
8267
8268
8269
8270
8271
8272
8273
8274
8275
8276
8277
8278
8279
8280
8281
8282
8283
8284
8285
8286
8287
8288
8289
8290
8291
8292
8293
8294
8295
8296
8297
8298
8299
8300
8301
8302
8303
8304
8305
8306
8307
8308
8309
8310
8311
8312
8313
8314
8315
8316
8317
8318
8319
8320
8321
8322
8323
8324
8325
8326
8327
8328
8329
8330
8331
8332
8333
8334
8335
8336
8337
8338
8339
8340
8341
8342
8343
8344
8345
8346
8347
8348
8349
8350
8351
8352
8353
8354
8355
8356
8357
8358
8359
8360
8361
8362
8363
8364
8365
8366
8367
8368
8369
8370
8371
8372
8373
8374
8375
8376
8377
8378
8379
8380
8381
8382
8383
8384
8385
8386
8387
8388
8389
8390
8391
8392
8393
8394
8395
8396
8397
8398
8399
8400
8401
8402
8403
8404
8405
8406
8407
8408
8409
8410
8411
8412
8413
8414
8415
8416
8417
8418
8419
8420
8421
8422
8423
8424
8425
8426
8427
8428
8429
8430
8431
8432
8433
8434
8435
8436
8437
8438
8439
8440
8441
8442
8443
8444
8445
8446
8447
8448
8449
8450
8451
8452
8453
8454
8455
8456
8457
8458
8459
8460
8461
8462
8463
8464
8465
8466
8467
8468
8469
8470
8471
8472
8473
8474
8475
8476
8477
8478
8479
8480
8481
8482
8483
8484
8485
8486
8487
8488
8489
8490
8491
8492
8493
8494
8495
8496
8497
8498
8499
8500
8501
8502
8503
8504
8505
8506
8507
8508
8509
8510
8511
8512
8513
8514
8515
8516
8517
8518
8519
8520
8521
8522
8523
8524
8525
8526
8527
8528
8529
8530
8531
8532
8533
8534
8535
8536
8537
8538
8539
8540
8541
8542
8543
8544
8545
8546
8547
8548
8549
8550
8551
8552
8553
8554
8555
8556
8557
8558
8559
8560
8561
8562
8563
8564
8565
8566
8567
8568
8569
8570
8571
8572
8573
8574
8575
8576
8577
8578
8579
8580
8581
8582
8583
8584
8585
8586
8587
8588
8589
8590
8591
8592
8593
8594
8595
8596
8597
8598
8599
8600
8601
8602
8603
8604
8605
8606
8607
8608
8609
8610
8611
8612
8613
8614
8615
8616
8617
8618
8619
8620
8621
8622
8623
8624
8625
8626
8627
8628
8629
8630
8631
8632
8633
8634
8635
8636
8637
8638
8639
8640
8641
8642
8643
8644
8645
8646
8647
8648
8649
8650
8651
8652
8653
8654
8655
8656
8657
8658
8659
8660
8661
8662
8663
8664
8665
8666
8667
8668
8669
8670
8671
8672
8673
8674
8675
8676
8677
8678
8679
8680
8681
8682
8683
8684
8685
8686
8687
8688
8689
8690
8691
8692
8693
8694
8695
8696
8697
8698
8699
8700
8701
8702
8703
8704
8705
8706
8707
8708
8709
8710
8711
8712
8713
8714
8715
8716
8717
8718
8719
8720
8721
8722
8723
8724
8725
8726
8727
8728
8729
8730
8731
8732
8733
8734
8735
8736
8737
8738
8739
8740
8741
8742
8743
8744
8745
8746
8747
8748
8749
8750
8751
8752
8753
8754
8755
8756
8757
8758
8759
8760
8761
8762
8763
8764
8765
8766
8767
8768
8769
8770
8771
8772
8773
8774
8775
8776
8777
8778
8779
8780
8781
8782
8783
8784
8785
8786
8787
8788
8789
8790
8791
8792
8793
8794
8795
8796
8797
8798
8799
8800
8801
8802
8803
8804
8805
8806
8807
8808
8809
8810
8811
8812
8813
8814
8815
8816
8817
8818
8819
8820
8821
8822
8823
8824
8825
8826
8827
8828
8829
8830
8831
8832
8833
8834
8835
8836
8837
8838
8839
8840
8841
8842
8843
8844
8845
8846
8847
8848
8849
8850
8851
8852
8853
8854
8855
8856
8857
8858
8859
8860
8861
8862
8863
8864
8865
8866
8867
8868
8869
8870
8871
8872
8873
8874
8875
8876
8877
8878
8879
8880
8881
8882
8883
8884
8885
8886
8887
8888
8889
8890
8891
8892
8893
8894
8895
8896
8897
8898
8899
8900
8901
8902
8903
8904
8905
8906
8907
8908
8909
8910
8911
8912
8913
8914
8915
8916
8917
8918
8919
8920
8921
8922
8923
8924
8925
8926
8927
8928
8929
8930
8931
8932
8933
8934
8935
8936
8937
8938
8939
8940
8941
8942
8943
8944
8945
8946
8947
8948
8949
8950
8951
8952
8953
8954
8955
8956
8957
8958
8959
8960
8961
8962
8963
8964
8965
8966
8967
8968
8969
8970
8971
8972
8973
8974
8975
8976
8977
8978
8979
8980
8981
8982
8983
8984
8985
8986
8987
8988
8989
8990
8991
8992
8993
8994
8995
8996
8997
8998
8999
9000
9001
9002
9003
9004
9005
9006
9007
9008
9009
9010
9011
9012
9013
9014
9015
9016
9017
9018
9019
9020
9021
9022
9023
9024
9025
9026
9027
9028
9029
9030
9031
9032
9033
9034
9035
9036
9037
9038
9039
9040
9041
9042
9043
9044
9045
9046
9047
9048
9049
9050
9051
9052
9053
9054
9055
9056
9057
9058
9059
9060
9061
9062
9063
9064
9065
9066
9067
9068
9069
9070
9071
9072
9073
9074
9075
9076
9077
9078
9079
9080
9081
9082
9083
9084
9085
9086
9087
9088
9089
9090
9091
9092
9093
9094
9095
9096
9097
9098
9099
9100
9101
9102
9103
9104
9105
9106
9107
9108
9109
9110
9111
9112
9113
9114
9115
9116
9117
9118
9119
9120
9121
9122
9123
9124
9125
9126
9127
9128
9129
9130
9131
9132
9133
9134
9135
9136
9137
9138
9139
9140
9141
9142
9143
9144
9145
9146
9147
9148
9149
9150
9151
9152
9153
9154
9155
9156
9157
9158
9159
9160
9161
9162
9163
9164
9165
9166
9167
9168
9169
9170
9171
9172
9173
9174
9175
9176
9177
9178
9179
9180
9181
9182
9183
9184
9185
9186
9187
9188
9189
9190
9191
9192
9193
9194
9195
9196
9197
9198
9199
9200
9201
9202
9203
9204
9205
9206
9207
9208
9209
9210
9211
9212
9213
9214
9215
9216
9217
9218
9219
9220
9221
9222
9223
9224
9225
9226
9227
9228
9229
9230
9231
9232
9233
9234
9235
9236
9237
9238
9239
9240
9241
9242
9243
9244
9245
9246
9247
9248
9249
9250
9251
9252
9253
9254
9255
9256
9257
9258
9259
9260
9261
9262
9263
9264
9265
9266
9267
9268
9269
9270
9271
9272
9273
9274
9275
9276
9277
9278
9279
9280
9281
9282
9283
9284
9285
9286
9287
9288
9289
9290
9291
9292
9293
9294
9295
9296
9297
9298
9299
9300
9301
9302
9303
9304
9305
9306
9307
9308
9309
9310
9311
9312
9313
9314
9315
9316
9317
9318
9319
9320
9321
9322
9323
9324
9325
9326
9327
9328
9329
9330
9331
9332
9333
9334
9335
9336
9337
9338
9339
9340
9341
9342
9343
9344
9345
9346
9347
9348
9349
9350
9351
9352
9353
9354
9355
9356
9357
9358
9359
9360
9361
9362
9363
9364
9365
9366
9367
9368
9369
9370
9371
9372
9373
9374
9375
9376
9377
9378
9379
9380
9381
9382
9383
9384
9385
9386
9387
9388
9389
9390
9391
9392
9393
9394
9395
9396
9397
9398
9399
9400
9401
9402
9403
9404
9405
9406
9407
9408
9409
9410
9411
9412
9413
9414
9415
9416
9417
9418
9419
9420
9421
9422
9423
9424
9425
9426
9427
9428
9429
9430
9431
9432
9433
9434
9435
9436
9437
9438
9439
9440
9441
9442
9443
9444
9445
9446
9447
9448
9449
9450
9451
9452
9453
9454
9455
9456
9457
9458
9459
9460
9461
9462
9463
9464
9465
9466
9467
9468
9469
9470
9471
9472
9473
9474
9475
9476
9477
9478
9479
9480
9481
9482
9483
9484
9485
9486
9487
9488
9489
9490
9491
9492
9493
9494
9495
9496
9497
9498
9499
9500
9501
9502
9503
9504
9505
9506
9507
9508
9509
9510
9511
9512
9513
9514
9515
9516
9517
9518
9519
9520
9521
9522
9523
9524
9525
9526
9527
9528
9529
9530
9531
9532
9533
9534
9535
9536
9537
9538
9539
9540
9541
9542
9543
9544
9545
9546
9547
9548
9549
9550
9551
9552
9553
9554
9555
9556
9557
9558
9559
9560
9561
9562
9563
9564
9565
9566
9567
9568
9569
9570
9571
9572
9573
9574
9575
9576
9577
9578
9579
9580
9581
9582
9583
9584
9585
9586
9587
9588
9589
9590
9591
9592
9593
9594
9595
9596
9597
9598
9599
9600
9601
9602
9603
9604
9605
9606
9607
9608
9609
9610
9611
9612
9613
9614
9615
9616
9617
9618
9619
9620
9621
9622
9623
9624
9625
9626
9627
9628
9629
9630
9631
9632
9633
9634
9635
9636
9637
9638
9639
9640
9641
9642
9643
9644
9645
9646
9647
9648
9649
9650
9651
9652
9653
9654
9655
9656
9657
9658
9659
9660
9661
9662
9663
9664
9665
9666
9667
9668
9669
9670
9671
9672
9673
9674
9675
9676
9677
9678
9679
9680
9681
9682
9683
9684
9685
9686
9687
9688
9689
9690
9691
9692
9693
9694
9695
9696
9697
9698
9699
9700
9701
9702
9703
9704
9705
9706
9707
9708
9709
9710
9711
9712
9713
9714
9715
9716
9717
9718
9719
9720
9721
9722
9723
9724
9725
9726
9727
9728
9729
9730
9731
9732
9733
9734
9735
9736
9737
9738
9739
9740
9741
9742
9743
9744
9745
9746
9747
9748
9749
9750
9751
9752
9753
9754
9755
9756
9757
9758
9759
9760
9761
9762
9763
9764
9765
9766
9767
9768
9769
9770
9771
9772
9773
9774
9775
9776
9777
9778
9779
9780
9781
9782
9783
9784
9785
9786
9787
9788
9789
9790
9791
9792
9793
9794
9795
9796
9797
9798
9799
9800
9801
9802
9803
9804
9805
9806
9807
9808
9809
9810
9811
9812
9813
9814
9815
9816
9817
9818
9819
9820
9821
9822
9823
9824
9825
9826
9827
9828
9829
9830
9831
9832
9833
9834
9835
9836
9837
9838
9839
9840
9841
9842
9843
9844
9845
9846
9847
9848
9849
9850
9851
9852
9853
9854
9855
9856
9857
9858
9859
9860
9861
9862
9863
9864
9865
9866
9867
9868
9869
9870
9871
9872
9873
9874
9875
9876
9877
9878
9879
9880
9881
9882
9883
9884
9885
9886
9887
9888
9889
9890
9891
9892
9893
9894
9895
9896
9897
9898
9899
9900
9901
9902
9903
9904
9905
9906
9907
9908
9909
9910
9911
9912
9913
9914
9915
9916
9917
9918
9919
9920
9921
9922
9923
9924
9925
9926
9927
9928
9929
9930
9931
9932
9933
9934
9935
9936
9937
9938
9939
9940
9941
9942
9943
9944
9945
9946
9947
9948
9949
9950
9951
9952
9953
9954
9955
9956
9957
9958
9959
9960
9961
9962
9963
9964
9965
9966
9967
9968
9969
9970
9971
9972
9973
9974
9975
9976
9977
9978
9979
9980
9981
9982
9983
9984
9985
9986
9987
9988
9989
9990
9991
9992
9993
9994
9995
9996
9997
9998
9999
10000
10001
10002
10003
10004
10005
10006
10007
10008
10009
10010
10011
10012
10013
10014
10015
10016
10017
10018
10019
10020
10021
10022
10023
10024
10025
10026
10027
10028
10029
10030
10031
10032
10033
10034
10035
10036
10037
10038
10039
10040
10041
10042
10043
10044
10045
10046
10047
10048
10049
10050
10051
10052
10053
10054
10055
10056
10057
10058
10059
10060
10061
10062
10063
10064
10065
10066
10067
10068
10069
10070
10071
10072
10073
10074
10075
10076
10077
10078
10079
10080
10081
10082
10083
10084
10085
10086
10087
10088
10089
10090
10091
10092
10093
10094
10095
10096
10097
10098
10099
10100
10101
10102
10103
10104
10105
10106
10107
10108
10109
10110
10111
10112
10113
10114
10115
10116
10117
10118
10119
10120
10121
10122
10123
10124
10125
10126
10127
10128
10129
10130
10131
10132
10133
10134
10135
10136
10137
10138
10139
10140
10141
10142
10143
10144
10145
10146
10147
10148
10149
10150
10151
10152
10153
10154
10155
10156
10157
10158
10159
10160
10161
10162
10163
10164
10165
10166
10167
10168
10169
10170
10171
10172
10173
10174
10175
10176
10177
10178
10179
10180
10181
10182
10183
10184
10185
10186
10187
10188
10189
10190
10191
10192
10193
10194
10195
10196
10197
10198
10199
10200
10201
10202
10203
10204
10205
10206
10207
10208
10209
10210
10211
10212
10213
10214
10215
10216
10217
10218
10219
10220
10221
10222
10223
10224
10225
10226
10227
10228
10229
10230
10231
10232
10233
10234
10235
10236
10237
10238
10239
10240
10241
10242
10243
10244
10245
10246
10247
10248
10249
10250
10251
10252
10253
10254
10255
10256
10257
10258
10259
10260
10261
10262
10263
10264
10265
10266
10267
10268
10269
10270
10271
10272
10273
10274
10275
10276
10277
10278
10279
10280
10281
10282
10283
10284
10285
10286
10287
10288
10289
10290
10291
10292
10293
10294
10295
10296
10297
10298
10299
10300
10301
10302
10303
10304
10305
10306
10307
10308
10309
10310
10311
10312
10313
10314
10315
10316
10317
10318
10319
10320
10321
10322
10323
10324
10325
10326
10327
10328
10329
10330
10331
10332
10333
10334
10335
10336
10337
10338
10339
10340
10341
10342
10343
10344
10345
10346
10347
10348
10349
10350
10351
10352
10353
10354
10355
10356
10357
10358
10359
10360
10361
10362
10363
10364
10365
10366
10367
10368
10369
10370
10371
10372
10373
10374
10375
10376
10377
10378
10379
10380
10381
10382
10383
10384
10385
10386
10387
10388
10389
10390
10391
10392
10393
10394
10395
10396
10397
10398
10399
10400
10401
10402
10403
10404
10405
10406
10407
10408
10409
10410
10411
10412
10413
10414
10415
10416
10417
10418
10419
10420
10421
10422
10423
10424
10425
10426
10427
10428
10429
10430
10431
10432
10433
10434
10435
10436
10437
10438
10439
10440
10441
10442
10443
10444
10445
10446
10447
10448
10449
10450
10451
10452
10453
10454
10455
10456
10457
10458
10459
10460
10461
10462
10463
10464
10465
10466
10467
10468
10469
10470
10471
10472
10473
10474
10475
10476
10477
10478
10479
10480
10481
10482
10483
10484
10485
10486
10487
10488
10489
10490
10491
10492
10493
10494
10495
10496
10497
10498
10499
10500
10501
10502
10503
10504
10505
10506
10507
10508
10509
10510
10511
10512
10513
10514
10515
10516
10517
10518
10519
10520
10521
10522
10523
10524
10525
10526
10527
10528
10529
10530
10531
10532
10533
10534
10535
10536
10537
10538
10539
10540
10541
10542
10543
10544
10545
10546
10547
10548
10549
10550
10551
10552
10553
10554
10555
10556
10557
10558
10559
10560
10561
10562
10563
10564
10565
10566
10567
10568
10569
10570
10571
10572
10573
10574
10575
10576
10577
10578
10579
10580
10581
10582
10583
10584
10585
10586
10587
10588
10589
10590
10591
10592
10593
10594
10595
10596
10597
10598
10599
10600
10601
10602
10603
10604
10605
10606
10607
10608
10609
10610
10611
10612
10613
10614
10615
10616
10617
10618
10619
10620
10621
10622
10623
10624
10625
10626
10627
10628
10629
10630
10631
10632
10633
10634
10635
10636
10637
10638
10639
10640
10641
10642
10643
10644
10645
10646
10647
10648
10649
10650
10651
10652
10653
10654
10655
10656
10657
10658
10659
10660
10661
10662
10663
10664
10665
10666
10667
10668
10669
10670
10671
10672
10673
10674
10675
10676
10677
10678
10679
10680
10681
10682
10683
10684
10685
10686
10687
10688
10689
10690
10691
10692
10693
10694
10695
10696
10697
10698
10699
10700
10701
10702
10703
10704
10705
10706
10707
10708
10709
10710
10711
10712
10713
10714
10715
10716
10717
10718
10719
10720
10721
10722
10723
10724
10725
10726
10727
10728
10729
10730
10731
10732
10733
10734
10735
10736
10737
10738
10739
10740
10741
10742
10743
10744
10745
10746
10747
10748
10749
10750
10751
10752
10753
10754
10755
10756
10757
10758
10759
10760
10761
10762
10763
10764
10765
10766
10767
10768
10769
10770
10771
10772
10773
10774
10775
10776
10777
10778
10779
10780
10781
10782
10783
10784
10785
10786
10787
10788
10789
10790
10791
10792
10793
10794
10795
10796
10797
10798
10799
10800
10801
10802
10803
10804
10805
10806
10807
10808
10809
10810
10811
10812
10813
10814
10815
10816
10817
10818
10819
10820
10821
10822
10823
10824
10825
10826
10827
10828
10829
10830
10831
10832
10833
10834
10835
10836
10837
10838
10839
10840
10841
10842
10843
10844
10845
10846
10847
10848
10849
10850
10851
10852
10853
10854
10855
10856
10857
10858
10859
10860
10861
10862
10863
10864
10865
10866
10867
10868
10869
10870
10871
10872
10873
10874
10875
10876
10877
10878
10879
10880
10881
10882
10883
10884
10885
10886
10887
10888
10889
10890
10891
10892
10893
10894
10895
10896
10897
10898
10899
10900
10901
10902
10903
10904
10905
10906
10907
10908
10909
10910
10911
10912
10913
10914
10915
10916
10917
10918
10919
10920
10921
10922
10923
10924
10925
10926
10927
10928
10929
10930
10931
10932
10933
10934
10935
10936
10937
10938
10939
10940
10941
10942
10943
10944
10945
10946
10947
10948
10949
10950
10951
10952
10953
10954
10955
10956
10957
10958
10959
10960
10961
10962
10963
10964
10965
10966
10967
10968
10969
10970
10971
10972
10973
10974
10975
10976
10977
10978
10979
10980
10981
10982
10983
10984
10985
10986
10987
10988
10989
10990
10991
10992
10993
10994
10995
10996
10997
10998
10999
11000
11001
11002
11003
11004
11005
11006
11007
11008
11009
11010
11011
11012
11013
11014
11015
11016
11017
11018
11019
11020
11021
11022
11023
11024
11025
11026
11027
11028
11029
11030
11031
11032
11033
11034
11035
11036
11037
11038
11039
11040
11041
11042
11043
11044
11045
11046
11047
11048
11049
11050
11051
11052
11053
11054
11055
11056
11057
11058
11059
11060
11061
11062
11063
11064
11065
11066
11067
11068
11069
11070
11071
11072
11073
11074
11075
11076
11077
11078
11079
11080
11081
11082
11083
11084
11085
11086
11087
11088
11089
11090
11091
11092
11093
11094
11095
11096
11097
11098
11099
11100
11101
11102
11103
11104
11105
11106
11107
11108
11109
11110
11111
11112
11113
11114
11115
11116
11117
11118
11119
11120
11121
11122
11123
11124
11125
11126
11127
11128
11129
11130
11131
11132
11133
11134
11135
11136
11137
11138
11139
11140
11141
11142
11143
11144
11145
11146
11147
11148
11149
11150
11151
11152
11153
11154
11155
11156
11157
11158
11159
11160
11161
11162
11163
11164
11165
11166
11167
11168
11169
11170
11171
11172
11173
11174
11175
11176
11177
11178
11179
11180
11181
11182
11183
11184
11185
11186
11187
11188
11189
11190
11191
11192
11193
11194
11195
11196
11197
11198
11199
11200
11201
11202
11203
11204
11205
11206
11207
11208
11209
11210
11211
11212
11213
11214
11215
11216
11217
11218
11219
11220
11221
11222
11223
11224
11225
11226
11227
11228
11229
11230
11231
11232
11233
11234
11235
11236
11237
11238
11239
11240
11241
11242
11243
11244
11245
11246
11247
11248
11249
11250
11251
11252
11253
11254
11255
11256
11257
11258
11259
11260
11261
11262
11263
11264
11265
11266
11267
11268
11269
11270
11271
11272
11273
11274
11275
11276
11277
11278
11279
11280
11281
11282
11283
11284
11285
11286
11287
11288
11289
11290
11291
11292
11293
11294
11295
11296
11297
11298
11299
11300
11301
11302
11303
11304
11305
11306
11307
11308
11309
11310
11311
11312
11313
11314
11315
11316
11317
11318
11319
11320
11321
11322
11323
11324
11325
11326
11327
11328
11329
11330
11331
11332
11333
11334
11335
11336
11337
11338
11339
11340
11341
11342
11343
11344
11345
11346
11347
11348
11349
11350
11351
11352
11353
11354
11355
11356
11357
11358
11359
11360
11361
11362
11363
11364
11365
11366
11367
11368
11369
11370
11371
11372
11373
11374
11375
11376
11377
11378
11379
11380
11381
11382
11383
11384
11385
11386
11387
11388
11389
11390
11391
11392
11393
11394
11395
11396
11397
11398
11399
11400
11401
11402
11403
11404
11405
11406
11407
11408
11409
11410
11411
11412
11413
11414
11415
11416
11417
11418
11419
11420
11421
11422
11423
11424
11425
11426
11427
11428
11429
11430
11431
11432
11433
11434
11435
11436
11437
11438
11439
11440
11441
11442
11443
11444
11445
11446
11447
11448
11449
11450
11451
11452
11453
11454
11455
11456
11457
11458
11459
11460
11461
11462
11463
11464
11465
11466
11467
11468
11469
11470
11471
11472
11473
11474
11475
11476
11477
11478
11479
11480
11481
11482
11483
11484
11485
11486
11487
11488
11489
11490
11491
11492
11493
11494
11495
11496
11497
11498
11499
11500
11501
11502
11503
11504
11505
11506
11507
11508
11509
11510
11511
11512
11513
11514
11515
11516
11517
11518
11519
11520
11521
11522
11523
11524
11525
11526
11527
11528
11529
11530
11531
11532
11533
11534
11535
11536
11537
11538
11539
11540
11541
11542
11543
11544
11545
11546
11547
11548
11549
11550
11551
11552
11553
11554
11555
11556
11557
11558
11559
11560
11561
11562
11563
11564
11565
11566
11567
11568
11569
11570
11571
11572
11573
11574
11575
11576
11577
11578
11579
11580
11581
11582
11583
11584
11585
11586
11587
11588
11589
11590
11591
11592
11593
11594
11595
11596
11597
11598
11599
11600
11601
11602
11603
11604
11605
11606
11607
11608
11609
11610
11611
11612
11613
11614
11615
11616
11617
11618
11619
11620
11621
11622
11623
11624
11625
11626
11627
11628
11629
11630
11631
11632
11633
11634
11635
11636
11637
11638
11639
11640
11641
11642
11643
11644
11645
11646
11647
11648
11649
11650
11651
11652
11653
11654
11655
11656
11657
11658
11659
11660
11661
11662
11663
11664
11665
11666
11667
11668
11669
11670
11671
11672
11673
11674
11675
11676
11677
11678
11679
11680
11681
11682
11683
11684
11685
11686
11687
11688
11689
11690
11691
11692
11693
11694
11695
11696
11697
11698
11699
11700
11701
11702
11703
11704
11705
11706
11707
11708
11709
11710
11711
11712
11713
11714
11715
11716
11717
11718
11719
11720
11721
11722
11723
11724
11725
11726
11727
11728
11729
11730
11731
11732
11733
11734
11735
11736
11737
11738
11739
11740
11741
11742
11743
11744
11745
11746
11747
11748
11749
11750
11751
11752
11753
11754
11755
11756
11757
11758
11759
11760
11761
11762
11763
11764
11765
11766
11767
11768
11769
11770
11771
11772
11773
11774
11775
11776
11777
11778
11779
11780
11781
11782
11783
11784
11785
11786
11787
11788
11789
11790
11791
11792
11793
11794
11795
11796
11797
11798
11799
11800
11801
11802
11803
11804
11805
11806
11807
11808
11809
11810
11811
11812
11813
11814
11815
11816
11817
11818
11819
11820
11821
11822
11823
11824
11825
11826
11827
11828
11829
11830
11831
11832
11833
11834
11835
11836
11837
11838
11839
11840
11841
11842
11843
11844
11845
11846
11847
11848
11849
11850
11851
11852
11853
11854
11855
11856
11857
11858
11859
11860
11861
11862
11863
11864
11865
11866
11867
11868
11869
11870
11871
11872
11873
11874
11875
11876
11877
11878
11879
11880
11881
11882
11883
11884
11885
11886
11887
11888
11889
11890
11891
11892
11893
11894
11895
11896
11897
11898
11899
11900
11901
11902
11903
11904
11905
11906
11907
11908
11909
11910
11911
11912
11913
11914
11915
11916
11917
11918
11919
11920
11921
11922
11923
11924
11925
11926
11927
11928
11929
11930
11931
11932
11933
11934
11935
11936
11937
11938
11939
11940
11941
11942
11943
11944
11945
11946
11947
11948
11949
11950
11951
11952
11953
11954
11955
11956
11957
11958
11959
11960
11961
11962
11963
11964
11965
11966
11967
11968
11969
11970
11971
11972
11973
11974
11975
11976
11977
11978
11979
11980
11981
11982
11983
11984
11985
11986
11987
11988
11989
11990
11991
11992
11993
11994
11995
11996
11997
11998
11999
12000
12001
12002
12003
12004
12005
12006
12007
12008
12009
12010
12011
12012
12013
12014
12015
12016
12017
12018
12019
12020
12021
12022
12023
12024
12025
12026
12027
12028
12029
12030
12031
12032
12033
12034
12035
12036
12037
12038
12039
12040
12041
12042
12043
12044
12045
12046
12047
12048
12049
12050
12051
12052
12053
12054
12055
12056
12057
12058
12059
12060
12061
12062
12063
12064
12065
12066
12067
12068
12069
12070
12071
12072
12073
12074
12075
12076
12077
12078
12079
12080
12081
12082
12083
12084
12085
12086
12087
12088
12089
12090
12091
12092
12093
12094
12095
12096
12097
12098
12099
12100
12101
12102
12103
12104
12105
12106
12107
12108
12109
12110
12111
12112
12113
12114
12115
12116
12117
12118
12119
12120
12121
12122
12123
12124
12125
12126
12127
12128
12129
12130
12131
12132
12133
12134
12135
12136
12137
12138
12139
12140
12141
12142
12143
12144
12145
12146
12147
12148
12149
12150
12151
12152
12153
12154
12155
12156
12157
12158
12159
12160
12161
12162
12163
12164
12165
12166
12167
12168
12169
12170
12171
12172
12173
12174
12175
12176
12177
12178
12179
12180
12181
12182
12183
12184
12185
12186
12187
12188
12189
12190
12191
12192
12193
12194
12195
12196
12197
12198
12199
12200
12201
12202
12203
12204
12205
12206
12207
12208
12209
12210
12211
12212
12213
12214
12215
12216
12217
12218
12219
12220
12221
12222
12223
12224
12225
12226
12227
12228
12229
12230
12231
12232
12233
12234
12235
12236
12237
12238
12239
12240
12241
12242
12243
12244
12245
12246
12247
12248
12249
12250
12251
12252
12253
12254
12255
12256
12257
12258
12259
12260
12261
12262
12263
12264
12265
12266
12267
12268
12269
12270
12271
12272
12273
12274
12275
12276
12277
12278
12279
12280
12281
12282
12283
12284
12285
12286
12287
12288
12289
12290
12291
12292
12293
12294
12295
12296
12297
12298
12299
12300
12301
12302
12303
12304
12305
12306
12307
12308
12309
12310
12311
12312
12313
12314
12315
12316
12317
12318
12319
12320
12321
12322
12323
12324
12325
12326
12327
12328
12329
12330
12331
12332
12333
12334
12335
12336
12337
12338
12339
12340
12341
12342
12343
12344
12345
12346
12347
12348
12349
12350
12351
12352
12353
12354
12355
12356
12357
12358
12359
12360
12361
12362
12363
12364
12365
12366
12367
12368
12369
12370
12371
12372
12373
12374
12375
12376
12377
12378
12379
12380
12381
12382
12383
12384
12385
12386
12387
12388
12389
12390
12391
12392
12393
12394
12395
12396
12397
12398
12399
12400
12401
12402
12403
12404
12405
12406
12407
12408
12409
12410
12411
12412
12413
12414
12415
12416
12417
12418
12419
12420
12421
12422
12423
12424
12425
12426
12427
12428
12429
12430
12431
12432
12433
12434
12435
12436
12437
12438
12439
12440
12441
12442
12443
12444
12445
12446
12447
12448
12449
12450
12451
12452
12453
12454
12455
12456
12457
12458
12459
12460
12461
12462
12463
12464
12465
12466
12467
12468
12469
12470
12471
12472
12473
12474
12475
12476
12477
12478
12479
12480
12481
12482
12483
12484
12485
12486
12487
12488
12489
12490
12491
12492
12493
12494
12495
12496
12497
12498
12499
12500
12501
12502
12503
12504
12505
12506
12507
12508
12509
12510
12511
12512
12513
12514
12515
12516
12517
12518
12519
12520
12521
12522
12523
12524
12525
12526
12527
12528
12529
12530
12531
12532
12533
12534
12535
12536
12537
12538
12539
12540
12541
12542
12543
12544
12545
12546
12547
12548
12549
12550
12551
12552
12553
12554
12555
12556
12557
12558
12559
12560
12561
12562
12563
12564
12565
12566
12567
12568
12569
12570
12571
12572
12573
12574
12575
12576
12577
12578
12579
12580
12581
12582
12583
12584
12585
12586
12587
12588
12589
12590
12591
12592
12593
12594
12595
12596
12597
12598
12599
12600
12601
12602
12603
12604
12605
12606
12607
12608
12609
12610
12611
12612
12613
12614
12615
12616
12617
12618
12619
12620
12621
12622
12623
12624
12625
12626
12627
12628
12629
12630
12631
12632
12633
12634
12635
12636
12637
12638
12639
12640
12641
12642
12643
12644
12645
12646
12647
12648
12649
12650
12651
12652
12653
12654
12655
12656
12657
12658
12659
12660
12661
12662
12663
12664
12665
12666
12667
12668
12669
12670
12671
12672
12673
12674
12675
12676
12677
12678
12679
12680
12681
12682
12683
12684
12685
12686
12687
12688
12689
12690
12691
12692
12693
12694
12695
12696
12697
12698
12699
12700
12701
12702
12703
12704
12705
12706
12707
12708
12709
12710
12711
12712
12713
12714
12715
12716
12717
12718
12719
12720
12721
12722
12723
12724
12725
12726
12727
12728
12729
12730
12731
12732
12733
12734
12735
12736
12737
12738
12739
12740
12741
12742
12743
12744
12745
12746
12747
12748
12749
12750
12751
12752
12753
12754
12755
12756
12757
12758
12759
12760
12761
12762
12763
12764
12765
12766
12767
12768
12769
12770
12771
12772
12773
12774
12775
12776
12777
12778
12779
12780
12781
12782
12783
12784
12785
12786
12787
12788
12789
12790
12791
12792
12793
12794
12795
12796
12797
12798
12799
12800
12801
12802
12803
12804
12805
12806
12807
12808
12809
12810
12811
12812
12813
12814
12815
12816
12817
12818
12819
12820
12821
12822
12823
12824
12825
12826
12827
12828
12829
12830
12831
12832
12833
12834
12835
12836
12837
12838
12839
12840
12841
12842
12843
12844
12845
12846
12847
12848
12849
12850
12851
12852
12853
12854
12855
12856
12857
12858
12859
12860
12861
12862
12863
12864
12865
12866
12867
12868
12869
12870
12871
12872
12873
12874
12875
12876
12877
12878
12879
12880
12881
12882
12883
12884
12885
12886
12887
12888
12889
12890
12891
12892
12893
12894
12895
12896
12897
12898
12899
12900
12901
12902
12903
12904
12905
12906
12907
12908
12909
12910
12911
12912
12913
12914
12915
12916
12917
12918
12919
12920
12921
12922
12923
12924
12925
12926
12927
12928
12929
12930
12931
12932
12933
12934
12935
12936
12937
12938
12939
12940
12941
12942
12943
12944
12945
12946
12947
12948
12949
12950
12951
12952
12953
12954
12955
12956
12957
12958
12959
12960
12961
12962
12963
12964
12965
12966
12967
12968
12969
12970
12971
12972
12973
12974
12975
12976
12977
12978
12979
12980
12981
12982
12983
12984
12985
12986
12987
12988
12989
12990
12991
12992
12993
12994
12995
12996
12997
12998
12999
13000
13001
13002
13003
13004
13005
13006
13007
13008
13009
13010
13011
13012
13013
13014
13015
13016
13017
13018
13019
13020
13021
13022
13023
13024
13025
13026
13027
13028
13029
13030
13031
13032
13033
13034
13035
13036
13037
13038
13039
13040
13041
13042
13043
13044
13045
13046
13047
13048
13049
13050
13051
13052
13053
13054
13055
13056
13057
13058
13059
13060
13061
13062
13063
13064
13065
13066
13067
13068
13069
13070
13071
13072
13073
13074
13075
13076
13077
13078
13079
13080
13081
13082
13083
13084
13085
13086
13087
13088
13089
13090
13091
13092
13093
13094
13095
13096
13097
13098
13099
13100
13101
13102
13103
13104
13105
13106
13107
13108
13109
13110
13111
13112
13113
13114
13115
13116
13117
13118
13119
13120
13121
13122
13123
13124
13125
13126
13127
13128
13129
13130
13131
13132
13133
13134
13135
13136
13137
13138
13139
13140
13141
13142
13143
13144
13145
13146
13147
13148
13149
13150
13151
13152
13153
13154
13155
13156
13157
13158
13159
13160
13161
13162
13163
13164
13165
13166
13167
13168
13169
13170
13171
13172
13173
13174
13175
13176
13177
13178
13179
13180
13181
13182
13183
13184
13185
13186
13187
13188
13189
13190
13191
13192
13193
13194
13195
13196
13197
13198
13199
13200
13201
13202
13203
13204
13205
13206
13207
13208
13209
13210
13211
13212
13213
13214
13215
13216
13217
13218
13219
13220
13221
13222
13223
13224
13225
13226
13227
13228
13229
13230
13231
13232
13233
13234
13235
13236
13237
13238
13239
13240
13241
13242
13243
13244
13245
13246
13247
13248
13249
13250
13251
13252
13253
13254
13255
13256
13257
13258
13259
13260
13261
13262
13263
13264
13265
13266
13267
13268
13269
13270
13271
13272
13273
13274
13275
13276
13277
13278
13279
13280
13281
13282
13283
13284
13285
13286
13287
13288
13289
13290
13291
13292
13293
13294
13295
13296
13297
13298
13299
13300
13301
13302
13303
13304
13305
13306
13307
13308
13309
13310
13311
13312
13313
13314
13315
13316
13317
13318
13319
13320
13321
13322
13323
13324
13325
13326
13327
13328
13329
13330
13331
13332
13333
13334
13335
13336
13337
13338
13339
13340
13341
13342
13343
13344
13345
13346
13347
13348
13349
13350
13351
13352
13353
13354
13355
13356
13357
13358
13359
13360
13361
13362
13363
13364
13365
13366
13367
13368
13369
13370
13371
13372
13373
13374
13375
13376
13377
13378
13379
13380
13381
13382
13383
13384
13385
13386
13387
13388
13389
13390
13391
13392
13393
13394
13395
13396
13397
13398
13399
13400
13401
13402
13403
13404
13405
13406
13407
13408
13409
13410
13411
13412
13413
13414
13415
13416
13417
13418
13419
13420
13421
13422
13423
13424
13425
13426
13427
13428
13429
13430
13431
13432
13433
13434
13435
13436
13437
13438
13439
13440
13441
13442
13443
13444
13445
13446
13447
13448
13449
13450
13451
13452
13453
13454
13455
13456
13457
13458
13459
13460
13461
13462
13463
13464
13465
13466
13467
13468
13469
13470
13471
13472
13473
13474
13475
13476
13477
13478
13479
13480
13481
13482
13483
13484
13485
13486
13487
13488
13489
13490
13491
13492
13493
13494
13495
13496
13497
13498
13499
13500
13501
13502
13503
13504
13505
13506
13507
13508
13509
13510
13511
13512
13513
13514
13515
13516
13517
13518
13519
13520
13521
13522
13523
13524
13525
13526
13527
13528
13529
13530
13531
13532
13533
13534
13535
13536
13537
13538
13539
13540
13541
13542
13543
13544
13545
13546
13547
13548
13549
13550
13551
13552
13553
13554
13555
13556
13557
13558
13559
13560
13561
13562
13563
13564
13565
13566
13567
13568
13569
13570
13571
13572
13573
13574
13575
13576
13577
13578
13579
13580
13581
13582
13583
13584
13585
13586
13587
13588
13589
13590
13591
13592
13593
13594
13595
13596
13597
13598
13599
13600
13601
13602
13603
13604
13605
13606
13607
13608
13609
13610
13611
13612
13613
13614
13615
13616
13617
13618
13619
13620
13621
13622
13623
13624
13625
13626
13627
13628
13629
13630
13631
13632
13633
13634
13635
13636
13637
13638
13639
13640
13641
13642
13643
13644
13645
13646
13647
13648
13649
13650
13651
13652
13653
13654
13655
13656
13657
13658
13659
13660
13661
13662
13663
13664
13665
13666
13667
13668
13669
13670
13671
13672
13673
13674
13675
13676
13677
13678
13679
13680
13681
13682
13683
13684
13685
13686
13687
13688
13689
13690
13691
13692
13693
13694
13695
13696
13697
13698
13699
13700
13701
13702
13703
13704
13705
13706
13707
13708
13709
13710
13711
13712
13713
13714
13715
13716
13717
13718
13719
13720
13721
13722
13723
13724
13725
13726
13727
13728
13729
13730
13731
13732
13733
13734
13735
13736
13737
13738
13739
13740
13741
13742
13743
13744
13745
13746
13747
13748
13749
13750
13751
13752
13753
13754
13755
13756
13757
13758
13759
13760
13761
13762
13763
13764
13765
13766
13767
13768
13769
13770
13771
13772
13773
13774
13775
13776
13777
13778
13779
13780
13781
13782
13783
13784
13785
13786
13787
13788
13789
13790
13791
13792
13793
13794
13795
13796
13797
13798
13799
13800
13801
13802
13803
13804
13805
13806
13807
13808
13809
13810
13811
13812
13813
13814
13815
13816
13817
13818
13819
13820
13821
13822
13823
13824
13825
13826
13827
13828
13829
13830
13831
13832
13833
13834
13835
13836
13837
13838
13839
13840
13841
13842
13843
13844
13845
13846
13847
13848
13849
13850
13851
13852
13853
13854
13855
13856
13857
13858
13859
13860
13861
13862
13863
13864
13865
13866
13867
13868
13869
13870
13871
13872
13873
13874
13875
13876
13877
13878
13879
13880
13881
13882
13883
13884
13885
13886
13887
13888
13889
13890
13891
13892
13893
13894
13895
13896
13897
13898
13899
13900
13901
13902
13903
13904
13905
13906
13907
13908
13909
13910
13911
13912
13913
13914
13915
13916
13917
13918
13919
13920
13921
13922
13923
13924
13925
13926
13927
13928
13929
13930
13931
13932
13933
13934
13935
13936
13937
13938
13939
13940
13941
13942
13943
13944
13945
13946
13947
13948
13949
13950
13951
13952
13953
13954
13955
13956
13957
13958
13959
13960
13961
13962
13963
13964
13965
13966
13967
13968
13969
13970
13971
13972
13973
13974
13975
13976
13977
13978
13979
13980
13981
13982
13983
13984
13985
13986
13987
13988
13989
13990
13991
13992
13993
13994
13995
13996
13997
13998
13999
14000
14001
14002
14003
14004
14005
14006
14007
14008
14009
14010
14011
14012
14013
14014
14015
14016
14017
14018
14019
14020
14021
14022
14023
14024
14025
14026
14027
14028
14029
14030
14031
14032
14033
14034
14035
14036
14037
14038
14039
14040
14041
14042
14043
14044
14045
14046
14047
14048
14049
14050
14051
14052
14053
14054
14055
14056
14057
14058
14059
14060
14061
14062
14063
14064
14065
14066
14067
14068
14069
14070
14071
14072
14073
14074
14075
14076
14077
14078
14079
14080
14081
14082
14083
14084
14085
14086
14087
14088
14089
14090
14091
14092
14093
14094
14095
14096
14097
14098
14099
14100
14101
14102
14103
14104
14105
14106
14107
14108
14109
14110
14111
14112
14113
14114
14115
14116
14117
14118
14119
14120
14121
14122
14123
14124
14125
14126
14127
14128
14129
14130
14131
14132
14133
14134
14135
14136
14137
14138
14139
14140
14141
14142
14143
14144
14145
14146
14147
14148
14149
14150
14151
14152
14153
14154
14155
14156
14157
14158
14159
14160
14161
14162
14163
14164
14165
14166
14167
14168
14169
14170
14171
14172
14173
14174
14175
14176
14177
14178
14179
14180
14181
14182
14183
14184
14185
14186
14187
14188
14189
14190
14191
14192
14193
14194
14195
14196
14197
14198
14199
14200
14201
14202
14203
14204
14205
14206
14207
14208
14209
14210
14211
14212
14213
14214
14215
14216
14217
14218
14219
14220
14221
14222
14223
14224
14225
14226
14227
14228
14229
14230
14231
14232
14233
14234
14235
14236
14237
14238
14239
14240
14241
14242
14243
14244
14245
14246
14247
14248
14249
14250
14251
14252
14253
14254
14255
14256
14257
14258
14259
14260
14261
14262
14263
14264
14265
14266
14267
14268
14269
14270
14271
14272
14273
14274
14275
14276
14277
14278
14279
14280
14281
14282
14283
14284
14285
14286
14287
14288
14289
14290
14291
14292
14293
14294
14295
14296
14297
14298
14299
14300
14301
14302
14303
14304
14305
14306
14307
14308
14309
14310
14311
14312
14313
14314
14315
14316
14317
14318
14319
14320
14321
14322
14323
14324
14325
14326
14327
14328
14329
14330
14331
14332
14333
14334
14335
14336
14337
14338
14339
14340
14341
14342
14343
14344
14345
14346
14347
14348
14349
14350
14351
14352
14353
14354
14355
14356
14357
14358
14359
14360
14361
14362
14363
14364
14365
14366
14367
14368
14369
14370
14371
14372
14373
14374
14375
14376
14377
14378
14379
14380
14381
14382
14383
14384
14385
14386
14387
14388
14389
14390
14391
14392
14393
14394
14395
14396
14397
14398
14399
14400
14401
14402
14403
14404
14405
14406
14407
14408
14409
14410
14411
14412
14413
14414
14415
14416
14417
14418
14419
14420
14421
14422
14423
14424
14425
14426
14427
14428
14429
14430
14431
14432
14433
14434
14435
14436
14437
14438
14439
14440
14441
14442
14443
14444
14445
14446
14447
14448
14449
14450
14451
14452
14453
14454
14455
14456
14457
14458
14459
14460
14461
14462
14463
14464
14465
14466
14467
14468
14469
14470
14471
14472
14473
14474
14475
14476
14477
14478
14479
14480
14481
14482
14483
14484
14485
14486
14487
14488
14489
14490
14491
14492
14493
14494
14495
14496
14497
14498
14499
14500
14501
14502
14503
14504
14505
14506
14507
14508
14509
14510
14511
14512
14513
14514
14515
14516
14517
14518
14519
14520
14521
14522
14523
14524
14525
14526
14527
14528
14529
14530
14531
14532
14533
14534
14535
14536
14537
14538
14539
14540
14541
14542
14543
14544
14545
14546
14547
14548
14549
14550
14551
14552
14553
14554
14555
14556
14557
14558
14559
14560
14561
14562
14563
14564
14565
14566
14567
14568
14569
14570
14571
14572
14573
14574
14575
14576
14577
14578
14579
14580
14581
14582
14583
14584
14585
14586
14587
14588
14589
14590
14591
14592
14593
14594
14595
14596
14597
14598
14599
14600
14601
14602
14603
14604
14605
14606
14607
14608
14609
14610
14611
14612
14613
14614
14615
14616
14617
14618
14619
14620
14621
14622
14623
14624
14625
14626
14627
14628
14629
14630
14631
14632
14633
14634
14635
14636
14637
14638
14639
14640
14641
14642
14643
14644
14645
14646
14647
14648
14649
14650
14651
14652
14653
14654
14655
14656
14657
14658
14659
14660
14661
14662
14663
14664
14665
14666
14667
14668
14669
14670
14671
14672
14673
14674
14675
14676
14677
14678
14679
14680
14681
14682
14683
14684
14685
14686
14687
14688
14689
14690
14691
14692
14693
14694
14695
14696
14697
14698
14699
14700
14701
14702
14703
14704
14705
14706
14707
14708
14709
14710
14711
14712
14713
14714
14715
14716
14717
14718
14719
14720
14721
14722
14723
14724
14725
14726
14727
14728
14729
14730
14731
14732
14733
14734
14735
14736
14737
14738
14739
14740
14741
14742
14743
14744
14745
14746
14747
14748
14749
14750
14751
14752
14753
14754
14755
14756
14757
14758
14759
14760
14761
14762
14763
14764
14765
14766
14767
14768
14769
14770
14771
14772
14773
14774
14775
14776
14777
14778
14779
14780
14781
14782
14783
14784
14785
14786
14787
14788
14789
14790
14791
14792
14793
14794
14795
14796
14797
14798
14799
14800
14801
14802
14803
14804
14805
14806
14807
14808
14809
14810
14811
14812
14813
14814
14815
14816
14817
14818
14819
14820
14821
14822
14823
14824
14825
14826
14827
14828
14829
14830
14831
14832
14833
14834
14835
14836
14837
14838
14839
14840
14841
14842
14843
14844
14845
14846
14847
14848
14849
14850
14851
14852
14853
14854
14855
14856
14857
14858
14859
14860
14861
14862
14863
14864
14865
14866
14867
14868
14869
14870
14871
14872
14873
14874
14875
14876
14877
14878
14879
14880
14881
14882
14883
14884
14885
14886
14887
14888
14889
14890
14891
14892
14893
14894
14895
14896
14897
14898
14899
14900
14901
14902
14903
14904
14905
14906
14907
14908
14909
14910
14911
14912
14913
14914
14915
14916
14917
14918
14919
14920
14921
14922
14923
14924
14925
14926
14927
14928
14929
14930
14931
14932
14933
14934
14935
14936
14937
14938
14939
14940
14941
14942
14943
14944
14945
14946
14947
14948
14949
14950
14951
14952
14953
14954
14955
14956
14957
14958
14959
14960
14961
14962
14963
14964
14965
14966
14967
14968
14969
14970
14971
14972
14973
14974
14975
14976
14977
14978
14979
14980
14981
14982
14983
14984
14985
14986
14987
14988
14989
14990
14991
14992
14993
14994
14995
14996
14997
14998
14999
15000
15001
15002
15003
15004
15005
15006
15007
15008
15009
15010
15011
15012
15013
15014
15015
15016
15017
15018
15019
15020
15021
15022
15023
15024
15025
15026
15027
15028
15029
15030
15031
15032
15033
15034
15035
15036
15037
15038
15039
15040
15041
15042
15043
15044
15045
15046
15047
15048
15049
15050
15051
15052
15053
15054
15055
15056
15057
15058
15059
15060
15061
15062
15063
15064
15065
15066
15067
15068
15069
15070
15071
15072
15073
15074
15075
15076
15077
15078
15079
15080
15081
15082
15083
15084
15085
15086
15087
15088
15089
15090
15091
15092
15093
15094
15095
15096
15097
15098
15099
15100
15101
15102
15103
15104
15105
15106
15107
15108
15109
15110
15111
15112
15113
15114
15115
15116
15117
15118
15119
15120
15121
15122
15123
15124
15125
15126
15127
15128
15129
15130
15131
15132
15133
15134
15135
15136
15137
15138
15139
15140
15141
15142
15143
15144
15145
15146
15147
15148
15149
15150
15151
15152
15153
15154
15155
15156
15157
15158
15159
15160
15161
15162
15163
15164
15165
15166
15167
15168
15169
15170
15171
15172
15173
15174
15175
15176
15177
15178
15179
15180
15181
15182
15183
15184
15185
15186
15187
15188
15189
15190
15191
15192
15193
15194
15195
15196
15197
15198
15199
15200
15201
15202
15203
15204
15205
15206
15207
15208
15209
15210
15211
15212
15213
15214
15215
15216
15217
15218
15219
15220
15221
15222
15223
15224
15225
15226
15227
15228
15229
15230
15231
15232
15233
15234
15235
15236
15237
15238
15239
15240
15241
15242
15243
15244
15245
15246
15247
15248
15249
15250
15251
15252
15253
15254
15255
15256
15257
15258
15259
15260
15261
15262
15263
15264
15265
15266
15267
15268
15269
15270
15271
15272
15273
15274
15275
15276
15277
15278
15279
15280
15281
15282
15283
15284
15285
15286
15287
15288
15289
15290
15291
15292
15293
15294
15295
15296
15297
15298
15299
15300
15301
15302
15303
15304
15305
15306
15307
15308
15309
15310
15311
15312
15313
15314
15315
15316
15317
15318
15319
15320
15321
15322
15323
15324
15325
15326
15327
15328
15329
15330
15331
15332
15333
15334
15335
15336
15337
15338
15339
15340
15341
15342
15343
15344
15345
15346
15347
15348
15349
15350
15351
15352
15353
15354
15355
15356
15357
15358
15359
15360
15361
15362
15363
15364
15365
15366
15367
15368
15369
15370
15371
15372
15373
15374
15375
15376
15377
15378
15379
15380
15381
15382
15383
15384
15385
15386
15387
15388
15389
15390
15391
15392
15393
15394
15395
15396
15397
15398
15399
15400
15401
15402
15403
15404
15405
15406
15407
15408
15409
15410
15411
15412
15413
15414
15415
15416
15417
15418
15419
15420
15421
15422
15423
15424
15425
15426
15427
15428
15429
15430
15431
15432
15433
15434
15435
15436
15437
15438
15439
15440
15441
15442
15443
15444
15445
15446
15447
15448
15449
15450
15451
15452
15453
15454
15455
15456
15457
15458
15459
15460
15461
15462
15463
15464
15465
15466
15467
15468
15469
15470
15471
15472
15473
15474
15475
15476
15477
15478
15479
15480
15481
15482
15483
15484
15485
15486
15487
15488
15489
15490
15491
15492
15493
15494
15495
15496
15497
15498
15499
15500
15501
15502
15503
15504
15505
15506
15507
15508
15509
15510
15511
15512
15513
15514
15515
15516
15517
15518
15519
15520
15521
15522
15523
15524
15525
15526
15527
15528
15529
15530
15531
15532
15533
15534
15535
15536
15537
15538
15539
15540
15541
15542
15543
15544
15545
15546
15547
15548
15549
15550
15551
15552
15553
15554
15555
15556
15557
15558
15559
15560
15561
15562
15563
15564
15565
15566
15567
15568
15569
15570
15571
15572
15573
15574
15575
15576
15577
15578
15579
15580
15581
15582
15583
15584
15585
15586
15587
15588
15589
15590
15591
15592
15593
15594
15595
15596
15597
15598
15599
15600
15601
15602
15603
15604
15605
15606
15607
15608
15609
15610
15611
15612
15613
15614
15615
15616
15617
15618
15619
15620
15621
15622
15623
15624
15625
15626
15627
15628
15629
15630
15631
15632
15633
15634
15635
15636
15637
15638
15639
15640
15641
15642
15643
15644
15645
15646
15647
15648
15649
15650
15651
15652
15653
15654
15655
15656
15657
15658
15659
15660
15661
15662
15663
15664
15665
15666
15667
15668
15669
15670
15671
15672
15673
15674
15675
15676
15677
15678
15679
15680
15681
15682
15683
15684
15685
15686
15687
15688
15689
15690
15691
15692
15693
15694
15695
15696
15697
15698
15699
15700
15701
15702
15703
15704
15705
15706
15707
15708
15709
15710
15711
15712
15713
15714
15715
15716
15717
15718
15719
15720
15721
15722
15723
15724
15725
15726
15727
15728
15729
15730
15731
15732
15733
15734
15735
15736
15737
15738
15739
15740
15741
15742
15743
15744
15745
15746
15747
15748
15749
15750
15751
15752
15753
15754
15755
15756
15757
15758
15759
15760
15761
15762
15763
15764
15765
15766
15767
15768
15769
15770
15771
15772
15773
15774
15775
15776
15777
15778
15779
15780
15781
15782
15783
15784
15785
15786
15787
15788
15789
15790
15791
15792
15793
15794
15795
15796
15797
15798
15799
15800
15801
15802
15803
15804
15805
15806
15807
15808
15809
15810
15811
15812
15813
15814
15815
15816
15817
15818
15819
15820
15821
15822
15823
15824
15825
15826
15827
15828
15829
15830
15831
15832
15833
15834
15835
15836
15837
15838
15839
15840
15841
15842
15843
15844
15845
15846
15847
15848
15849
15850
15851
15852
15853
15854
15855
15856
15857
15858
15859
15860
15861
15862
15863
15864
15865
15866
15867
15868
15869
15870
15871
15872
15873
15874
15875
15876
15877
15878
15879
15880
15881
15882
15883
15884
15885
15886
15887
15888
15889
15890
15891
15892
15893
15894
15895
15896
15897
15898
15899
15900
15901
15902
15903
15904
15905
15906
15907
15908
15909
15910
15911
15912
15913
15914
15915
15916
15917
15918
15919
15920
15921
15922
15923
15924
15925
15926
15927
15928
15929
15930
15931
15932
15933
15934
15935
15936
15937
15938
15939
15940
15941
15942
15943
15944
15945
15946
15947
15948
15949
15950
15951
15952
15953
15954
15955
15956
15957
15958
15959
15960
15961
15962
15963
15964
15965
15966
15967
15968
15969
15970
15971
15972
15973
15974
15975
15976
15977
15978
15979
15980
15981
15982
15983
15984
15985
15986
15987
15988
15989
15990
15991
15992
15993
15994
15995
15996
15997
15998
15999
16000
16001
16002
16003
16004
16005
16006
16007
16008
16009
16010
16011
16012
16013
16014
16015
16016
16017
16018
16019
16020
16021
16022
16023
16024
16025
16026
16027
16028
16029
16030
16031
16032
16033
16034
16035
16036
16037
16038
16039
16040
16041
16042
16043
16044
16045
16046
16047
16048
16049
16050
16051
16052
16053
16054
16055
16056
16057
16058
16059
16060
16061
16062
16063
16064
16065
16066
16067
16068
16069
16070
16071
16072
16073
16074
16075
16076
16077
16078
16079
16080
16081
16082
16083
16084
16085
16086
16087
16088
16089
16090
16091
16092
16093
16094
16095
16096
16097
16098
16099
16100
16101
16102
16103
16104
16105
16106
16107
16108
16109
16110
16111
16112
16113
16114
16115
16116
16117
16118
16119
16120
16121
16122
16123
16124
16125
16126
16127
16128
16129
16130
16131
16132
16133
16134
16135
16136
16137
16138
16139
16140
16141
16142
16143
16144
16145
16146
16147
16148
16149
16150
16151
16152
16153
16154
16155
16156
16157
16158
16159
16160
16161
16162
16163
16164
16165
16166
16167
16168
16169
16170
16171
16172
16173
16174
16175
16176
16177
16178
16179
16180
16181
16182
16183
16184
16185
16186
16187
16188
16189
16190
16191
16192
16193
16194
16195
16196
16197
16198
16199
16200
16201
16202
16203
16204
16205
16206
16207
16208
16209
16210
16211
16212
16213
16214
16215
16216
16217
16218
16219
16220
16221
16222
16223
16224
16225
16226
16227
16228
16229
16230
16231
16232
16233
16234
16235
16236
16237
16238
16239
16240
16241
16242
16243
16244
16245
16246
16247
16248
16249
16250
16251
16252
16253
16254
16255
16256
16257
16258
16259
16260
16261
16262
16263
16264
16265
16266
16267
16268
16269
16270
16271
16272
16273
16274
16275
16276
16277
16278
16279
16280
16281
16282
16283
16284
16285
16286
16287
16288
16289
16290
16291
16292
16293
16294
16295
16296
16297
16298
16299
16300
16301
16302
16303
16304
16305
16306
16307
16308
16309
16310
16311
16312
16313
16314
16315
16316
16317
16318
16319
16320
16321
16322
16323
16324
16325
16326
16327
16328
16329
16330
16331
16332
16333
16334
16335
16336
16337
16338
16339
16340
16341
16342
16343
16344
16345
16346
16347
16348
16349
16350
16351
16352
16353
16354
16355
16356
16357
16358
16359
16360
16361
16362
16363
16364
16365
16366
16367
16368
16369
16370
16371
16372
16373
16374
16375
16376
16377
16378
16379
16380
16381
16382
16383
16384
16385
16386
16387
16388
16389
16390
16391
16392
16393
16394
16395
16396
16397
16398
16399
16400
16401
16402
16403
16404
16405
16406
16407
16408
16409
16410
16411
16412
16413
16414
16415
16416
16417
16418
16419
16420
16421
16422
16423
16424
16425
16426
16427
16428
16429
16430
16431
16432
16433
16434
16435
16436
16437
16438
16439
16440
16441
16442
16443
16444
16445
16446
16447
16448
16449
16450
16451
16452
16453
16454
16455
16456
16457
16458
16459
16460
16461
16462
16463
16464
16465
16466
16467
16468
16469
16470
16471
16472
16473
16474
16475
16476
16477
16478
16479
16480
16481
16482
16483
16484
16485
16486
16487
16488
16489
16490
16491
16492
16493
16494
16495
16496
16497
16498
16499
16500
16501
16502
16503
16504
16505
16506
16507
16508
16509
16510
16511
16512
16513
16514
16515
16516
16517
16518
16519
16520
16521
16522
16523
16524
16525
16526
16527
16528
16529
16530
16531
16532
16533
16534
16535
16536
16537
16538
16539
16540
16541
16542
16543
16544
16545
16546
16547
16548
16549
16550
16551
16552
16553
16554
16555
16556
16557
16558
16559
16560
16561
16562
16563
16564
16565
16566
16567
16568
16569
16570
16571
16572
16573
16574
16575
16576
16577
16578
16579
16580
16581
16582
16583
16584
16585
16586
16587
16588
16589
16590
16591
16592
16593
16594
16595
16596
16597
16598
16599
16600
16601
16602
16603
16604
16605
16606
16607
16608
16609
16610
16611
16612
16613
16614
16615
16616
16617
16618
16619
16620
16621
16622
16623
16624
16625
16626
16627
16628
16629
16630
16631
16632
16633
16634
16635
16636
16637
16638
16639
16640
16641
16642
16643
16644
16645
16646
16647
16648
16649
16650
16651
16652
16653
16654
16655
16656
16657
16658
16659
16660
16661
16662
16663
16664
16665
16666
16667
16668
16669
16670
16671
16672
16673
16674
16675
16676
16677
16678
16679
16680
16681
16682
16683
16684
16685
16686
16687
16688
16689
16690
16691
16692
16693
16694
16695
16696
16697
16698
16699
16700
16701
16702
16703
16704
16705
16706
16707
16708
16709
16710
16711
16712
16713
16714
16715
16716
16717
16718
16719
16720
16721
16722
16723
16724
16725
16726
16727
16728
16729
16730
16731
16732
16733
16734
16735
16736
16737
16738
16739
16740
16741
16742
16743
16744
16745
16746
16747
16748
16749
16750
16751
16752
16753
16754
16755
16756
16757
16758
16759
16760
16761
16762
16763
16764
16765
16766
16767
16768
16769
16770
16771
16772
16773
16774
16775
16776
16777
16778
16779
16780
16781
16782
16783
16784
16785
16786
16787
16788
16789
16790
16791
16792
16793
16794
16795
16796
16797
16798
16799
16800
16801
16802
16803
16804
16805
16806
16807
16808
16809
16810
16811
16812
16813
16814
16815
16816
16817
16818
16819
16820
16821
16822
16823
16824
16825
16826
16827
16828
16829
16830
16831
16832
16833
16834
16835
16836
16837
16838
16839
16840
16841
16842
16843
16844
16845
16846
16847
16848
16849
16850
16851
16852
16853
16854
16855
16856
16857
16858
16859
16860
16861
16862
16863
16864
16865
16866
16867
16868
16869
16870
16871
16872
16873
16874
16875
16876
16877
16878
16879
16880
16881
16882
16883
16884
16885
16886
16887
16888
16889
16890
16891
16892
16893
16894
16895
16896
16897
16898
16899
16900
16901
16902
16903
16904
16905
16906
16907
16908
16909
16910
16911
16912
16913
16914
16915
16916
16917
16918
16919
16920
16921
16922
16923
16924
16925
16926
16927
16928
16929
16930
16931
16932
16933
16934
16935
16936
16937
16938
16939
16940
16941
16942
16943
16944
16945
16946
16947
16948
16949
16950
16951
16952
16953
16954
16955
16956
16957
16958
16959
16960
16961
16962
16963
16964
16965
16966
16967
16968
16969
16970
16971
16972
16973
16974
16975
16976
16977
16978
16979
16980
16981
16982
16983
16984
16985
16986
16987
16988
16989
16990
16991
16992
16993
16994
16995
16996
16997
16998
16999
17000
17001
17002
17003
17004
17005
17006
17007
17008
17009
17010
17011
17012
17013
17014
17015
17016
17017
17018
17019
17020
17021
17022
17023
17024
17025
17026
17027
17028
17029
17030
17031
17032
17033
17034
17035
17036
17037
17038
17039
17040
17041
17042
17043
17044
17045
17046
17047
17048
17049
17050
17051
17052
17053
17054
17055
17056
17057
17058
17059
17060
17061
17062
17063
17064
17065
17066
17067
17068
17069
17070
17071
17072
17073
17074
17075
17076
17077
17078
17079
17080
17081
17082
17083
17084
17085
17086
17087
17088
17089
17090
17091
17092
17093
17094
17095
17096
17097
17098
17099
17100
17101
17102
17103
17104
17105
17106
17107
17108
17109
17110
17111
17112
17113
17114
17115
17116
17117
17118
17119
17120
17121
17122
17123
17124
17125
17126
17127
17128
17129
17130
17131
17132
17133
17134
17135
17136
17137
17138
17139
17140
17141
17142
17143
17144
17145
17146
17147
17148
17149
17150
17151
17152
17153
17154
17155
17156
17157
17158
17159
17160
17161
17162
17163
17164
17165
17166
17167
17168
17169
17170
17171
17172
17173
17174
17175
17176
17177
17178
17179
17180
17181
17182
17183
17184
17185
17186
17187
17188
17189
17190
17191
17192
17193
17194
17195
17196
17197
17198
17199
17200
17201
17202
17203
17204
17205
17206
17207
17208
17209
17210
17211
17212
17213
17214
17215
17216
17217
17218
17219
17220
17221
17222
17223
17224
17225
17226
17227
17228
17229
17230
17231
17232
17233
17234
17235
17236
17237
17238
17239
17240
17241
17242
17243
17244
17245
17246
17247
17248
17249
17250
17251
17252
17253
17254
17255
17256
17257
17258
17259
17260
17261
17262
17263
17264
17265
17266
17267
17268
17269
17270
17271
17272
17273
17274
17275
17276
17277
17278
17279
17280
17281
17282
17283
17284
17285
17286
17287
17288
17289
17290
17291
17292
17293
17294
17295
17296
17297
17298
17299
17300
17301
17302
17303
17304
17305
17306
17307
17308
17309
17310
17311
17312
17313
17314
17315
17316
17317
17318
17319
17320
17321
17322
17323
17324
17325
17326
17327
17328
17329
17330
17331
17332
17333
17334
17335
17336
17337
17338
17339
17340
17341
17342
17343
17344
17345
17346
17347
17348
17349
17350
17351
17352
17353
17354
17355
17356
17357
17358
17359
17360
17361
17362
17363
17364
17365
17366
17367
17368
17369
17370
17371
17372
17373
17374
17375
17376
17377
17378
17379
17380
17381
17382
17383
17384
17385
17386
17387
17388
17389
17390
17391
17392
17393
17394
17395
17396
17397
17398
17399
17400
17401
17402
17403
17404
17405
17406
17407
17408
17409
17410
17411
17412
17413
17414
17415
17416
17417
17418
17419
17420
17421
17422
17423
17424
17425
17426
17427
17428
17429
17430
17431
17432
17433
17434
17435
17436
17437
17438
17439
17440
17441
17442
17443
17444
17445
17446
17447
17448
17449
17450
17451
17452
17453
17454
17455
17456
17457
17458
17459
17460
17461
17462
17463
17464
17465
17466
17467
17468
17469
17470
17471
17472
17473
17474
17475
17476
17477
17478
17479
17480
17481
17482
17483
17484
17485
17486
17487
17488
17489
17490
17491
17492
17493
17494
17495
17496
17497
17498
17499
17500
17501
17502
17503
17504
17505
17506
17507
17508
17509
17510
17511
17512
17513
17514
17515
17516
17517
17518
17519
17520
17521
17522
17523
17524
17525
17526
17527
17528
17529
17530
17531
17532
17533
17534
17535
17536
17537
17538
17539
17540
17541
17542
17543
17544
17545
17546
17547
17548
17549
17550
17551
17552
17553
17554
17555
17556
17557
17558
17559
17560
17561
17562
17563
17564
17565
17566
17567
17568
17569
17570
17571
17572
17573
17574
17575
17576
17577
17578
17579
17580
17581
17582
17583
17584
17585
17586
17587
17588
17589
17590
17591
17592
17593
17594
17595
17596
17597
17598
17599
17600
17601
17602
17603
17604
17605
17606
17607
17608
17609
17610
17611
17612
17613
17614
17615
17616
17617
17618
17619
17620
17621
17622
17623
17624
17625
17626
17627
17628
17629
17630
17631
17632
17633
17634
17635
17636
17637
17638
17639
17640
17641
17642
17643
17644
17645
17646
17647
17648
17649
17650
17651
17652
17653
17654
17655
17656
17657
17658
17659
17660
17661
17662
17663
17664
17665
17666
17667
17668
17669
17670
17671
17672
17673
17674
17675
17676
17677
17678
17679
17680
17681
17682
17683
17684
17685
17686
17687
17688
17689
17690
17691
17692
17693
17694
17695
17696
17697
17698
17699
17700
17701
17702
17703
17704
17705
17706
17707
17708
17709
17710
17711
17712
17713
17714
17715
17716
17717
17718
17719
17720
17721
17722
17723
17724
17725
17726
17727
17728
17729
17730
17731
17732
17733
17734
17735
17736
17737
17738
17739
17740
17741
17742
17743
17744
17745
17746
17747
17748
17749
17750
17751
17752
17753
17754
17755
17756
17757
17758
17759
17760
17761
17762
17763
17764
17765
17766
17767
17768
17769
17770
17771
17772
17773
17774
17775
17776
17777
17778
17779
17780
17781
17782
17783
17784
17785
17786
17787
17788
17789
17790
17791
17792
17793
17794
17795
17796
17797
17798
17799
17800
17801
17802
17803
17804
17805
17806
17807
17808
17809
17810
17811
17812
17813
17814
17815
17816
17817
17818
17819
17820
17821
17822
17823
17824
17825
17826
17827
17828
17829
17830
17831
17832
17833
17834
17835
17836
17837
17838
17839
17840
17841
17842
17843
17844
17845
17846
17847
17848
17849
17850
17851
17852
17853
17854
17855
17856
17857
17858
17859
17860
17861
17862
17863
17864
17865
17866
17867
17868
17869
17870
17871
17872
17873
17874
17875
17876
17877
17878
17879
17880
17881
17882
17883
17884
17885
17886
17887
17888
17889
17890
17891
17892
17893
17894
17895
17896
17897
17898
17899
17900
17901
17902
17903
17904
17905
17906
17907
17908
17909
17910
17911
17912
17913
17914
17915
17916
17917
17918
17919
17920
17921
17922
17923
17924
17925
17926
17927
17928
17929
17930
17931
17932
17933
17934
17935
17936
17937
17938
17939
17940
17941
17942
17943
17944
17945
17946
17947
17948
17949
17950
17951
17952
17953
17954
17955
17956
17957
17958
17959
17960
17961
17962
17963
17964
17965
17966
17967
17968
17969
17970
17971
17972
17973
17974
17975
17976
17977
17978
17979
17980
17981
17982
17983
17984
17985
17986
17987
17988
17989
17990
17991
17992
17993
17994
17995
17996
17997
17998
17999
18000
18001
18002
18003
18004
18005
18006
18007
18008
18009
18010
18011
18012
18013
18014
18015
18016
18017
18018
18019
18020
18021
18022
18023
18024
18025
18026
18027
18028
18029
18030
18031
18032
18033
18034
18035
18036
18037
18038
18039
18040
18041
18042
18043
18044
18045
18046
18047
18048
18049
18050
18051
18052
18053
18054
18055
18056
18057
18058
18059
18060
18061
18062
18063
18064
18065
18066
18067
18068
18069
18070
18071
18072
18073
18074
18075
18076
18077
18078
18079
18080
18081
18082
18083
18084
18085
18086
18087
18088
18089
18090
18091
18092
18093
18094
18095
18096
18097
18098
18099
18100
18101
18102
18103
18104
18105
18106
18107
18108
18109
18110
18111
18112
18113
18114
18115
18116
18117
18118
18119
18120
18121
18122
18123
18124
18125
18126
18127
18128
18129
18130
18131
18132
18133
18134
18135
18136
18137
18138
18139
18140
18141
18142
18143
18144
18145
18146
18147
18148
18149
18150
18151
18152
18153
18154
18155
18156
18157
18158
18159
18160
18161
18162
18163
18164
18165
18166
18167
18168
18169
18170
18171
18172
18173
18174
18175
18176
18177
18178
18179
18180
18181
18182
18183
18184
18185
18186
18187
18188
18189
18190
18191
18192
18193
18194
18195
18196
18197
18198
18199
18200
18201
18202
18203
18204
18205
18206
18207
18208
18209
18210
18211
18212
18213
18214
18215
18216
18217
18218
18219
18220
18221
18222
18223
18224
18225
18226
18227
18228
18229
18230
18231
18232
18233
18234
18235
18236
18237
18238
18239
18240
18241
18242
18243
18244
18245
18246
18247
18248
18249
18250
18251
18252
18253
18254
18255
18256
18257
18258
18259
18260
18261
18262
18263
18264
18265
18266
18267
18268
18269
18270
18271
18272
18273
18274
18275
18276
18277
18278
18279
18280
18281
18282
18283
18284
18285
18286
18287
18288
18289
18290
18291
18292
18293
18294
18295
18296
18297
18298
18299
18300
18301
18302
18303
18304
18305
18306
18307
18308
18309
18310
18311
18312
18313
18314
18315
18316
18317
18318
18319
18320
18321
18322
18323
18324
18325
18326
18327
18328
18329
18330
18331
18332
18333
18334
18335
18336
18337
18338
18339
18340
18341
18342
18343
18344
18345
18346
18347
18348
18349
18350
18351
18352
18353
18354
18355
18356
18357
18358
18359
18360
18361
18362
18363
18364
18365
18366
18367
18368
18369
18370
18371
18372
18373
18374
18375
18376
18377
18378
18379
18380
18381
18382
18383
18384
18385
18386
18387
18388
18389
18390
18391
18392
18393
18394
18395
18396
18397
18398
18399
18400
18401
18402
18403
18404
18405
18406
18407
18408
18409
18410
18411
18412
18413
18414
18415
18416
18417
18418
18419
18420
18421
18422
18423
18424
18425
18426
18427
18428
18429
18430
18431
18432
18433
18434
18435
18436
18437
18438
18439
18440
18441
18442
18443
18444
18445
18446
18447
18448
18449
18450
18451
18452
18453
18454
18455
18456
18457
18458
18459
18460
18461
18462
18463
18464
18465
18466
18467
18468
18469
18470
18471
18472
18473
18474
18475
18476
18477
18478
18479
18480
18481
18482
18483
18484
18485
18486
18487
18488
18489
18490
18491
18492
18493
18494
18495
18496
18497
18498
18499
18500
18501
18502
18503
18504
18505
18506
18507
18508
18509
18510
18511
18512
18513
18514
18515
18516
18517
18518
18519
18520
18521
18522
18523
18524
18525
18526
18527
18528
18529
18530
18531
18532
18533
18534
18535
18536
18537
18538
18539
18540
18541
18542
18543
18544
18545
18546
18547
18548
18549
18550
18551
18552
18553
18554
18555
18556
18557
18558
18559
18560
18561
18562
18563
18564
18565
18566
18567
18568
18569
18570
18571
18572
18573
18574
18575
18576
18577
18578
18579
18580
18581
18582
18583
18584
18585
18586
18587
18588
18589
18590
18591
18592
18593
18594
18595
18596
18597
18598
18599
18600
18601
18602
18603
18604
18605
18606
18607
18608
18609
18610
18611
18612
18613
18614
18615
18616
18617
18618
18619
18620
18621
18622
18623
18624
18625
18626
18627
18628
18629
18630
18631
18632
18633
18634
18635
18636
18637
18638
18639
18640
18641
18642
18643
18644
18645
18646
18647
18648
18649
18650
18651
18652
18653
18654
18655
18656
18657
18658
18659
18660
18661
18662
18663
18664
18665
18666
18667
18668
18669
18670
18671
18672
18673
18674
18675
18676
18677
18678
18679
18680
18681
18682
18683
18684
18685
18686
18687
18688
18689
18690
18691
18692
18693
18694
18695
18696
18697
18698
18699
18700
18701
18702
18703
18704
18705
18706
18707
18708
18709
18710
18711
18712
18713
18714
18715
18716
18717
18718
18719
18720
18721
18722
18723
18724
18725
18726
18727
18728
18729
18730
18731
18732
18733
18734
18735
18736
18737
18738
18739
18740
18741
18742
18743
18744
18745
18746
18747
18748
18749
18750
18751
18752
18753
18754
18755
18756
18757
18758
18759
18760
18761
18762
18763
18764
18765
18766
18767
18768
18769
18770
18771
18772
18773
18774
18775
18776
18777
18778
18779
18780
18781
18782
18783
18784
18785
18786
18787
18788
18789
18790
18791
18792
18793
18794
18795
18796
18797
18798
18799
18800
18801
18802
18803
18804
18805
18806
18807
18808
18809
18810
18811
18812
18813
18814
18815
18816
18817
18818
18819
18820
18821
18822
18823
18824
18825
18826
18827
18828
18829
18830
18831
18832
18833
18834
18835
18836
18837
18838
18839
18840
18841
18842
18843
18844
18845
18846
18847
18848
18849
18850
18851
18852
18853
18854
18855
18856
18857
18858
18859
18860
18861
18862
18863
18864
18865
18866
18867
18868
18869
18870
18871
18872
18873
18874
18875
18876
18877
18878
18879
18880
18881
18882
18883
18884
18885
18886
18887
18888
18889
18890
18891
18892
18893
18894
18895
18896
18897
18898
18899
18900
18901
18902
18903
18904
18905
18906
18907
18908
18909
18910
18911
18912
18913
18914
18915
18916
18917
18918
18919
18920
18921
18922
18923
18924
18925
18926
18927
18928
18929
18930
18931
18932
18933
18934
18935
18936
18937
18938
18939
18940
18941
18942
18943
18944
18945
18946
18947
18948
18949
18950
18951
18952
18953
18954
18955
18956
18957
18958
18959
18960
18961
18962
18963
18964
18965
18966
18967
18968
18969
18970
18971
18972
18973
18974
18975
18976
18977
18978
18979
18980
18981
18982
18983
18984
18985
18986
18987
18988
18989
18990
18991
18992
18993
18994
18995
18996
18997
18998
18999
19000
19001
19002
19003
19004
19005
19006
19007
19008
19009
19010
19011
19012
19013
19014
19015
19016
19017
19018
19019
19020
19021
19022
19023
19024
19025
19026
19027
19028
19029
19030
19031
19032
19033
19034
19035
19036
19037
19038
19039
19040
19041
19042
19043
19044
19045
19046
19047
19048
19049
19050
19051
19052
19053
19054
19055
19056
19057
19058
19059
19060
19061
19062
19063
19064
19065
19066
19067
19068
19069
19070
19071
19072
19073
19074
19075
19076
19077
19078
19079
19080
19081
19082
19083
19084
19085
19086
19087
19088
19089
19090
19091
19092
19093
19094
19095
19096
19097
19098
19099
19100
19101
19102
19103
19104
19105
19106
19107
19108
19109
19110
19111
19112
19113
19114
19115
19116
19117
19118
19119
19120
19121
19122
19123
19124
19125
19126
19127
19128
19129
19130
19131
19132
19133
19134
19135
19136
19137
19138
19139
19140
19141
19142
19143
19144
19145
19146
19147
19148
19149
19150
19151
19152
19153
19154
19155
19156
19157
19158
19159
19160
19161
19162
19163
19164
19165
19166
19167
19168
19169
19170
19171
19172
19173
19174
19175
19176
19177
19178
19179
19180
19181
19182
19183
19184
19185
19186
19187
19188
19189
19190
19191
19192
19193
19194
19195
19196
19197
19198
19199
19200
19201
19202
19203
19204
19205
19206
19207
19208
19209
19210
19211
19212
19213
19214
19215
19216
19217
19218
19219
19220
19221
19222
19223
19224
19225
19226
19227
19228
19229
19230
19231
19232
19233
19234
19235
19236
19237
19238
19239
19240
19241
19242
19243
19244
19245
19246
19247
19248
19249
19250
19251
19252
19253
19254
19255
19256
19257
19258
19259
19260
19261
19262
19263
19264
19265
19266
19267
19268
19269
19270
19271
19272
19273
19274
19275
19276
19277
19278
19279
19280
19281
19282
19283
19284
19285
19286
19287
19288
19289
19290
19291
19292
19293
19294
19295
19296
19297
19298
19299
19300
19301
19302
19303
19304
19305
19306
19307
19308
19309
19310
19311
19312
19313
19314
19315
19316
19317
19318
19319
19320
19321
19322
19323
19324
19325
19326
19327
19328
19329
19330
19331
19332
19333
19334
19335
19336
19337
19338
19339
19340
19341
19342
19343
19344
19345
19346
19347
19348
19349
19350
19351
19352
19353
19354
19355
19356
19357
19358
19359
19360
19361
19362
19363
19364
19365
19366
19367
19368
19369
19370
19371
19372
19373
19374
19375
19376
19377
19378
19379
19380
19381
19382
19383
19384
19385
19386
19387
19388
19389
19390
19391
19392
19393
19394
19395
19396
19397
19398
19399
19400
19401
19402
19403
19404
19405
19406
19407
19408
19409
19410
19411
19412
19413
19414
19415
19416
19417
19418
19419
19420
19421
19422
19423
19424
19425
19426
19427
19428
19429
19430
19431
19432
19433
19434
19435
19436
19437
19438
19439
19440
19441
19442
19443
19444
19445
19446
19447
19448
19449
19450
19451
19452
19453
19454
19455
19456
19457
19458
19459
19460
19461
19462
19463
19464
19465
19466
19467
19468
19469
19470
19471
19472
19473
19474
19475
19476
19477
19478
19479
19480
19481
19482
19483
19484
19485
19486
19487
19488
19489
19490
19491
19492
19493
19494
19495
19496
19497
19498
19499
19500
19501
19502
19503
19504
19505
19506
19507
19508
19509
19510
19511
19512
19513
19514
19515
19516
19517
19518
19519
19520
19521
19522
19523
19524
19525
19526
19527
19528
19529
19530
19531
19532
19533
19534
19535
19536
19537
19538
19539
19540
19541
19542
19543
19544
19545
19546
19547
19548
19549
19550
19551
19552
19553
19554
19555
19556
19557
19558
19559
19560
19561
19562
19563
19564
19565
19566
19567
19568
19569
19570
19571
19572
19573
19574
19575
19576
19577
19578
19579
19580
19581
19582
19583
19584
19585
19586
19587
19588
19589
19590
19591
19592
19593
19594
19595
19596
19597
19598
19599
19600
19601
19602
19603
19604
19605
19606
19607
19608
19609
19610
19611
19612
19613
19614
19615
19616
19617
19618
19619
19620
19621
19622
19623
19624
19625
19626
19627
19628
19629
19630
19631
19632
19633
19634
19635
19636
19637
19638
19639
19640
19641
19642
19643
19644
19645
19646
19647
19648
19649
19650
19651
19652
19653
19654
19655
19656
19657
19658
19659
19660
19661
19662
19663
19664
19665
19666
19667
19668
19669
19670
19671
19672
19673
19674
19675
19676
19677
19678
19679
19680
19681
19682
19683
19684
19685
19686
19687
19688
19689
19690
19691
19692
19693
19694
19695
19696
19697
19698
19699
19700
19701
19702
19703
19704
19705
19706
19707
19708
19709
19710
19711
19712
19713
19714
19715
19716
19717
19718
19719
19720
19721
19722
19723
19724
19725
19726
19727
19728
19729
19730
19731
19732
19733
19734
19735
19736
19737
19738
19739
19740
19741
19742
19743
19744
19745
19746
19747
19748
19749
19750
19751
19752
19753
19754
19755
19756
19757
19758
19759
19760
19761
19762
19763
19764
19765
19766
19767
19768
19769
19770
19771
19772
19773
19774
19775
19776
19777
19778
19779
19780
19781
19782
19783
19784
19785
19786
19787
19788
19789
19790
19791
19792
19793
19794
19795
19796
19797
19798
19799
19800
19801
19802
19803
19804
19805
19806
19807
19808
19809
19810
19811
19812
19813
19814
19815
19816
19817
19818
19819
19820
19821
19822
19823
19824
19825
19826
19827
19828
19829
19830
19831
19832
19833
19834
19835
19836
19837
19838
19839
19840
19841
19842
19843
19844
19845
19846
19847
19848
19849
19850
19851
19852
19853
19854
19855
19856
19857
19858
19859
19860
19861
19862
19863
19864
19865
19866
19867
19868
19869
19870
19871
19872
19873
19874
19875
19876
19877
19878
19879
19880
19881
19882
19883
19884
19885
19886
19887
19888
19889
19890
19891
19892
19893
19894
19895
19896
19897
19898
19899
19900
19901
19902
19903
19904
19905
19906
19907
19908
19909
19910
19911
19912
19913
19914
19915
19916
19917
19918
19919
19920
19921
19922
19923
19924
19925
19926
19927
19928
19929
19930
19931
19932
19933
19934
19935
19936
19937
19938
19939
19940
19941
19942
19943
19944
19945
19946
19947
19948
19949
19950
19951
19952
19953
19954
19955
19956
19957
19958
19959
19960
19961
19962
19963
19964
19965
19966
19967
19968
19969
19970
19971
19972
19973
19974
19975
19976
19977
19978
19979
19980
19981
19982
19983
19984
19985
19986
19987
19988
19989
19990
19991
19992
19993
19994
19995
19996
19997
19998
19999
20000
20001
20002
20003
20004
20005
20006
20007
20008
20009
20010
20011
20012
20013
20014
20015
20016
20017
20018
20019
20020
20021
20022
20023
20024
20025
20026
20027
20028
20029
20030
20031
20032
20033
20034
20035
20036
20037
20038
20039
20040
20041
20042
20043
20044
20045
20046
20047
20048
20049
20050
20051
20052
20053
20054
20055
20056
20057
20058
20059
20060
20061
20062
20063
20064
20065
20066
20067
20068
20069
20070
20071
20072
20073
20074
20075
20076
20077
20078
20079
20080
20081
20082
20083
20084
20085
20086
20087
20088
20089
20090
20091
20092
20093
20094
20095
20096
20097
20098
20099
20100
20101
20102
20103
20104
20105
20106
20107
20108
20109
20110
20111
20112
20113
20114
20115
20116
20117
20118
20119
20120
20121
20122
20123
20124
20125
20126
20127
20128
20129
20130
20131
20132
20133
20134
20135
20136
20137
20138
20139
20140
20141
20142
20143
20144
20145
20146
20147
20148
20149
20150
20151
20152
20153
20154
20155
20156
20157
20158
20159
20160
20161
20162
20163
20164
20165
20166
20167
20168
20169
20170
20171
20172
20173
20174
20175
20176
20177
20178
20179
20180
20181
20182
20183
20184
20185
20186
20187
20188
20189
20190
20191
20192
20193
20194
20195
20196
20197
20198
20199
20200
20201
20202
20203
20204
20205
20206
20207
20208
20209
20210
20211
20212
20213
20214
20215
20216
20217
20218
20219
20220
20221
20222
20223
20224
20225
20226
20227
20228
20229
20230
20231
20232
20233
20234
20235
20236
20237
20238
20239
20240
20241
20242
20243
20244
20245
20246
20247
20248
20249
20250
20251
20252
20253
20254
20255
20256
20257
20258
20259
20260
20261
20262
20263
20264
20265
20266
20267
20268
20269
20270
20271
20272
20273
20274
20275
20276
20277
20278
20279
20280
20281
20282
20283
20284
20285
20286
20287
20288
20289
20290
20291
20292
20293
20294
20295
20296
20297
20298
20299
20300
20301
20302
20303
20304
20305
20306
20307
20308
20309
20310
20311
20312
20313
20314
20315
20316
20317
20318
20319
20320
20321
20322
20323
20324
20325
20326
20327
20328
20329
20330
20331
20332
20333
20334
20335
20336
20337
20338
20339
20340
20341
20342
20343
20344
20345
20346
20347
20348
20349
20350
20351
20352
20353
20354
20355
20356
20357
20358
20359
20360
20361
20362
20363
20364
20365
20366
20367
20368
20369
20370
20371
20372
20373
20374
20375
20376
20377
20378
20379
20380
20381
20382
20383
20384
20385
20386
20387
20388
20389
20390
20391
20392
20393
20394
20395
20396
20397
20398
20399
20400
20401
20402
20403
20404
20405
20406
20407
20408
20409
20410
20411
20412
20413
20414
20415
20416
20417
20418
20419
20420
20421
20422
20423
20424
20425
20426
20427
20428
20429
20430
20431
20432
20433
20434
20435
20436
20437
20438
20439
20440
20441
20442
20443
20444
20445
20446
20447
20448
20449
20450
20451
20452
20453
20454
20455
20456
20457
20458
20459
20460
20461
20462
20463
20464
20465
20466
20467
20468
20469
20470
20471
20472
20473
20474
20475
20476
20477
20478
20479
20480
20481
20482
20483
20484
20485
20486
20487
20488
20489
20490
20491
20492
20493
20494
20495
20496
20497
20498
20499
20500
20501
20502
20503
20504
20505
20506
20507
20508
20509
20510
20511
20512
20513
20514
20515
20516
20517
20518
20519
20520
20521
20522
20523
20524
20525
20526
20527
20528
20529
20530
20531
20532
20533
20534
20535
20536
20537
20538
20539
20540
20541
20542
20543
20544
20545
20546
20547
20548
20549
20550
20551
20552
20553
20554
20555
20556
20557
20558
20559
20560
20561
20562
20563
20564
20565
20566
20567
20568
20569
20570
20571
20572
20573
20574
20575
20576
20577
20578
20579
20580
20581
20582
20583
20584
20585
20586
20587
20588
20589
20590
20591
20592
20593
20594
20595
20596
20597
20598
20599
20600
20601
20602
20603
20604
20605
20606
20607
20608
20609
20610
20611
20612
20613
20614
20615
20616
20617
20618
20619
20620
20621
20622
20623
20624
20625
20626
20627
20628
20629
20630
20631
20632
20633
20634
20635
20636
20637
20638
20639
20640
20641
20642
20643
20644
20645
20646
20647
20648
20649
20650
20651
20652
20653
20654
20655
20656
20657
20658
20659
20660
20661
20662
20663
20664
20665
20666
20667
20668
20669
20670
20671
20672
20673
20674
20675
20676
20677
20678
20679
20680
20681
20682
20683
20684
20685
20686
20687
20688
20689
20690
20691
20692
20693
20694
20695
20696
20697
20698
20699
20700
20701
20702
20703
20704
20705
20706
20707
20708
20709
20710
20711
20712
20713
20714
20715
20716
20717
20718
20719
20720
20721
20722
20723
20724
20725
20726
20727
20728
20729
20730
20731
20732
20733
20734
20735
20736
20737
20738
20739
20740
20741
20742
20743
20744
20745
20746
20747
20748
20749
20750
20751
20752
20753
20754
20755
20756
20757
20758
20759
20760
20761
20762
20763
20764
20765
20766
20767
20768
20769
20770
20771
20772
20773
20774
20775
20776
20777
20778
20779
20780
20781
20782
20783
20784
20785
20786
20787
20788
20789
20790
20791
20792
20793
20794
20795
20796
20797
20798
20799
20800
20801
20802
20803
20804
20805
20806
20807
20808
20809
20810
20811
20812
20813
20814
20815
20816
20817
20818
20819
20820
20821
20822
20823
20824
20825
20826
20827
20828
20829
20830
20831
20832
20833
20834
20835
20836
20837
20838
20839
20840
20841
20842
20843
20844
20845
20846
20847
20848
20849
20850
20851
20852
20853
20854
20855
20856
20857
20858
20859
20860
20861
20862
20863
20864
20865
20866
20867
20868
20869
20870
20871
20872
20873
20874
20875
20876
20877
20878
20879
20880
20881
20882
20883
20884
20885
20886
20887
20888
20889
20890
20891
20892
20893
20894
20895
20896
20897
20898
20899
20900
20901
20902
20903
20904
20905
20906
20907
20908
20909
20910
20911
20912
20913
20914
20915
20916
20917
20918
20919
20920
20921
20922
20923
20924
20925
20926
20927
20928
20929
20930
20931
20932
20933
20934
20935
20936
20937
20938
20939
20940
20941
20942
20943
20944
20945
20946
20947
20948
20949
20950
20951
20952
20953
20954
20955
20956
20957
20958
20959
20960
20961
20962
20963
20964
20965
20966
20967
20968
20969
20970
20971
20972
20973
20974
20975
20976
20977
20978
20979
20980
20981
20982
20983
20984
20985
20986
20987
20988
20989
20990
20991
20992
20993
20994
20995
20996
20997
20998
20999
21000
21001
21002
21003
21004
21005
21006
21007
21008
21009
21010
21011
21012
21013
21014
21015
21016
21017
21018
21019
21020
21021
21022
21023
21024
21025
21026
21027
21028
21029
21030
21031
21032
21033
21034
21035
21036
21037
21038
21039
21040
21041
21042
21043
21044
21045
21046
21047
21048
21049
21050
21051
21052
21053
21054
21055
21056
21057
21058
21059
21060
21061
21062
21063
21064
21065
21066
21067
21068
21069
21070
21071
21072
21073
21074
21075
21076
21077
21078
21079
21080
21081
21082
21083
21084
21085
21086
21087
21088
21089
21090
21091
21092
21093
21094
21095
21096
21097
21098
21099
21100
21101
21102
21103
21104
21105
21106
21107
21108
21109
21110
21111
21112
21113
21114
21115
21116
21117
21118
21119
21120
21121
21122
21123
21124
21125
21126
21127
21128
21129
21130
21131
21132
21133
21134
21135
21136
21137
21138
21139
21140
21141
21142
21143
21144
21145
21146
21147
21148
21149
21150
21151
21152
21153
21154
21155
21156
21157
21158
21159
21160
21161
21162
21163
21164
21165
21166
21167
21168
21169
21170
21171
21172
21173
21174
21175
21176
21177
21178
21179
21180
21181
21182
21183
21184
21185
21186
21187
21188
21189
21190
21191
21192
21193
21194
21195
21196
21197
21198
21199
21200
21201
21202
21203
21204
21205
21206
21207
21208
21209
21210
21211
21212
21213
21214
21215
21216
21217
21218
21219
21220
21221
21222
21223
21224
21225
21226
21227
21228
21229
21230
21231
21232
21233
21234
21235
21236
21237
21238
21239
21240
21241
21242
21243
21244
21245
21246
21247
21248
21249
21250
21251
21252
21253
21254
21255
21256
21257
21258
21259
21260
21261
21262
21263
21264
21265
21266
21267
21268
21269
21270
21271
21272
21273
21274
21275
21276
21277
21278
21279
21280
21281
21282
21283
21284
21285
21286
21287
21288
21289
21290
21291
21292
21293
21294
21295
21296
21297
21298
21299
21300
21301
21302
21303
21304
21305
21306
21307
21308
21309
21310
21311
21312
21313
21314
21315
21316
21317
21318
21319
21320
21321
21322
21323
21324
21325
21326
21327
21328
21329
21330
21331
21332
21333
21334
21335
21336
21337
21338
21339
21340
21341
21342
21343
21344
21345
21346
21347
21348
21349
21350
21351
21352
21353
21354
21355
21356
21357
21358
21359
21360
21361
21362
21363
21364
21365
21366
21367
21368
21369
21370
21371
21372
21373
21374
21375
21376
21377
21378
21379
21380
21381
21382
21383
21384
21385
21386
21387
21388
21389
21390
21391
21392
21393
21394
21395
21396
21397
21398
21399
21400
21401
21402
21403
21404
21405
21406
21407
21408
21409
21410
21411
21412
21413
21414
21415
21416
21417
21418
21419
21420
21421
21422
21423
21424
21425
21426
21427
21428
21429
21430
21431
21432
21433
21434
21435
21436
21437
21438
21439
21440
21441
21442
21443
21444
21445
21446
21447
21448
21449
21450
21451
21452
21453
21454
21455
21456
21457
21458
21459
21460
21461
21462
21463
21464
21465
21466
21467
21468
21469
21470
21471
21472
21473
21474
21475
21476
21477
21478
21479
21480
21481
21482
21483
21484
21485
21486
21487
21488
21489
21490
21491
21492
21493
21494
21495
21496
21497
21498
21499
21500
21501
21502
21503
21504
21505
21506
21507
21508
21509
21510
21511
21512
21513
21514
21515
21516
21517
21518
21519
21520
21521
21522
21523
21524
21525
21526
21527
21528
21529
21530
21531
21532
21533
21534
21535
21536
21537
21538
21539
21540
21541
21542
21543
21544
21545
21546
21547
21548
21549
21550
21551
21552
21553
21554
21555
21556
21557
21558
21559
21560
21561
21562
21563
21564
21565
21566
21567
21568
21569
21570
21571
21572
21573
21574
21575
21576
21577
21578
21579
21580
21581
21582
21583
21584
21585
21586
21587
21588
21589
21590
21591
21592
21593
21594
21595
21596
21597
21598
21599
21600
21601
21602
21603
21604
21605
21606
21607
21608
21609
21610
21611
21612
21613
21614
21615
21616
21617
21618
21619
21620
21621
21622
21623
21624
21625
21626
21627
21628
21629
21630
21631
21632
21633
21634
21635
21636
21637
21638
21639
21640
21641
21642
21643
21644
21645
21646
21647
21648
21649
21650
21651
21652
21653
21654
21655
21656
21657
21658
21659
21660
21661
21662
21663
21664
21665
21666
21667
21668
21669
21670
21671
21672
21673
21674
21675
21676
21677
21678
21679
21680
21681
21682
21683
21684
21685
21686
21687
21688
21689
21690
21691
21692
21693
21694
21695
21696
21697
21698
21699
21700
21701
21702
21703
21704
21705
21706
21707
21708
21709
21710
21711
21712
21713
21714
21715
21716
21717
21718
21719
21720
21721
21722
21723
21724
21725
21726
21727
21728
21729
21730
21731
21732
21733
21734
21735
21736
21737
21738
21739
21740
21741
21742
21743
21744
21745
21746
21747
21748
21749
21750
21751
21752
21753
21754
21755
21756
21757
21758
21759
21760
21761
21762
21763
21764
21765
21766
21767
21768
21769
21770
21771
21772
21773
21774
21775
21776
21777
21778
21779
21780
21781
21782
21783
21784
21785
21786
21787
21788
21789
21790
21791
21792
21793
21794
21795
21796
21797
21798
21799
21800
21801
21802
21803
21804
21805
21806
21807
21808
21809
21810
21811
21812
21813
21814
21815
21816
21817
21818
21819
21820
21821
21822
21823
21824
21825
21826
21827
21828
21829
21830
21831
21832
21833
21834
21835
21836
21837
21838
21839
21840
21841
21842
21843
21844
21845
21846
21847
21848
21849
21850
21851
21852
21853
21854
21855
21856
21857
21858
21859
21860
21861
21862
21863
21864
21865
21866
21867
21868
21869
21870
21871
21872
21873
21874
21875
21876
21877
21878
21879
21880
21881
21882
21883
21884
21885
21886
21887
21888
21889
21890
21891
21892
21893
21894
21895
21896
21897
21898
21899
21900
21901
21902
21903
21904
21905
21906
21907
21908
21909
21910
21911
21912
21913
21914
21915
21916
21917
21918
21919
21920
21921
21922
21923
21924
21925
21926
21927
21928
21929
21930
21931
21932
21933
21934
21935
21936
21937
21938
21939
21940
21941
21942
21943
21944
21945
21946
21947
21948
21949
21950
21951
21952
21953
21954
21955
21956
21957
21958
21959
21960
21961
21962
21963
21964
21965
21966
21967
21968
21969
21970
21971
21972
21973
21974
21975
21976
21977
21978
21979
21980
21981
21982
21983
21984
21985
21986
21987
21988
21989
21990
21991
21992
21993
21994
21995
21996
21997
21998
21999
22000
22001
22002
22003
22004
22005
22006
22007
22008
22009
22010
22011
22012
22013
22014
22015
22016
22017
22018
22019
22020
22021
22022
22023
22024
22025
22026
22027
22028
22029
22030
22031
22032
22033
22034
22035
22036
22037
22038
22039
22040
22041
22042
22043
22044
22045
22046
22047
22048
22049
22050
22051
22052
22053
22054
22055
22056
22057
22058
22059
22060
22061
22062
22063
22064
22065
22066
22067
22068
22069
22070
22071
22072
22073
22074
22075
22076
22077
22078
22079
22080
22081
22082
22083
22084
22085
22086
22087
22088
22089
22090
22091
22092
22093
22094
22095
22096
22097
22098
22099
22100
22101
22102
22103
22104
22105
22106
22107
22108
22109
22110
22111
22112
22113
22114
22115
22116
22117
22118
22119
22120
22121
22122
22123
22124
22125
22126
22127
22128
22129
22130
22131
22132
22133
22134
22135
22136
22137
22138
22139
22140
22141
22142
22143
22144
22145
22146
22147
22148
22149
22150
22151
22152
22153
22154
22155
22156
22157
22158
22159
22160
22161
22162
22163
22164
22165
22166
22167
22168
22169
22170
22171
22172
22173
22174
22175
22176
22177
22178
22179
22180
22181
22182
22183
22184
22185
22186
22187
22188
22189
22190
22191
22192
22193
22194
22195
22196
22197
22198
22199
22200
22201
22202
22203
22204
22205
22206
22207
22208
22209
22210
22211
22212
22213
22214
22215
22216
22217
22218
22219
22220
22221
22222
22223
22224
22225
22226
22227
22228
22229
22230
22231
22232
22233
22234
22235
22236
22237
22238
22239
22240
22241
22242
22243
22244
22245
22246
22247
22248
22249
22250
22251
22252
22253
22254
22255
22256
22257
22258
22259
22260
22261
22262
22263
22264
22265
22266
22267
22268
22269
22270
22271
22272
22273
22274
22275
22276
22277
22278
22279
22280
22281
22282
22283
22284
22285
22286
22287
22288
22289
22290
22291
22292
22293
22294
22295
22296
22297
22298
22299
22300
22301
22302
22303
22304
22305
22306
22307
22308
22309
22310
22311
22312
22313
22314
22315
22316
22317
22318
22319
22320
22321
22322
22323
22324
22325
22326
22327
22328
22329
22330
22331
22332
22333
22334
22335
22336
22337
22338
22339
22340
22341
22342
22343
22344
22345
22346
22347
22348
22349
22350
22351
22352
22353
22354
22355
22356
22357
22358
22359
22360
22361
22362
22363
22364
22365
22366
22367
22368
22369
22370
22371
22372
22373
22374
22375
22376
22377
22378
22379
22380
22381
22382
22383
22384
22385
22386
22387
22388
22389
22390
22391
22392
22393
22394
22395
22396
22397
22398
22399
22400
22401
22402
22403
22404
22405
22406
22407
22408
22409
22410
22411
22412
22413
22414
22415
22416
22417
22418
22419
22420
22421
22422
22423
22424
22425
22426
22427
22428
22429
22430
22431
22432
22433
22434
22435
22436
22437
22438
22439
22440
22441
22442
22443
22444
22445
22446
22447
22448
22449
22450
22451
22452
22453
22454
22455
22456
22457
22458
22459
22460
22461
22462
22463
22464
22465
22466
22467
22468
22469
22470
22471
22472
22473
22474
22475
22476
22477
22478
22479
22480
22481
22482
22483
22484
22485
22486
22487
22488
22489
22490
22491
22492
22493
22494
22495
22496
22497
22498
22499
22500
22501
22502
22503
22504
22505
22506
22507
22508
22509
22510
22511
22512
22513
22514
22515
22516
22517
22518
22519
22520
22521
22522
22523
22524
22525
22526
22527
22528
22529
22530
22531
22532
22533
22534
22535
22536
22537
22538
22539
22540
22541
22542
22543
22544
22545
22546
22547
22548
22549
22550
22551
22552
22553
22554
22555
22556
22557
22558
22559
22560
22561
22562
22563
22564
22565
22566
22567
22568
22569
22570
22571
22572
22573
22574
22575
22576
22577
22578
22579
22580
22581
22582
22583
22584
22585
22586
22587
22588
22589
22590
22591
22592
22593
22594
22595
22596
22597
22598
22599
22600
22601
22602
22603
22604
22605
22606
22607
22608
22609
22610
22611
22612
22613
22614
22615
22616
22617
22618
22619
22620
22621
22622
22623
22624
22625
22626
22627
22628
22629
22630
22631
22632
22633
22634
22635
22636
22637
22638
22639
22640
22641
22642
22643
22644
22645
22646
22647
22648
22649
22650
22651
22652
22653
22654
22655
22656
22657
22658
22659
22660
22661
22662
22663
22664
22665
22666
22667
22668
22669
22670
22671
22672
22673
22674
22675
22676
22677
22678
22679
22680
22681
22682
22683
22684
22685
22686
22687
22688
22689
22690
22691
22692
22693
22694
22695
22696
22697
22698
22699
22700
22701
22702
22703
22704
22705
22706
22707
22708
22709
22710
22711
22712
22713
22714
22715
22716
22717
22718
22719
22720
22721
22722
22723
22724
22725
22726
22727
22728
22729
22730
22731
22732
22733
22734
22735
22736
22737
22738
22739
22740
22741
22742
22743
22744
22745
22746
22747
22748
22749
22750
22751
22752
22753
22754
22755
22756
22757
22758
22759
22760
22761
22762
22763
22764
22765
22766
22767
22768
22769
22770
22771
22772
22773
22774
22775
22776
22777
22778
22779
22780
22781
22782
22783
22784
22785
22786
22787
22788
22789
22790
22791
22792
22793
22794
22795
22796
22797
22798
22799
22800
22801
22802
22803
22804
22805
22806
22807
22808
22809
22810
22811
22812
22813
22814
22815
22816
22817
22818
22819
22820
22821
22822
22823
22824
22825
22826
22827
22828
22829
22830
22831
22832
22833
22834
22835
22836
22837
22838
22839
22840
22841
22842
22843
22844
22845
22846
22847
22848
22849
22850
22851
22852
22853
22854
22855
22856
22857
22858
22859
22860
22861
22862
22863
22864
22865
22866
22867
22868
22869
22870
22871
22872
22873
22874
22875
22876
22877
22878
22879
22880
22881
22882
22883
22884
22885
22886
22887
22888
22889
22890
22891
22892
22893
22894
22895
22896
22897
22898
22899
22900
22901
22902
22903
22904
22905
22906
22907
22908
22909
22910
22911
22912
22913
22914
22915
22916
22917
22918
22919
22920
22921
22922
22923
22924
22925
22926
22927
22928
22929
22930
22931
22932
22933
22934
22935
22936
22937
22938
22939
22940
22941
22942
22943
22944
22945
22946
22947
22948
22949
22950
22951
22952
22953
22954
22955
22956
22957
22958
22959
22960
22961
22962
22963
22964
22965
22966
22967
22968
22969
22970
22971
22972
22973
22974
22975
22976
22977
22978
22979
22980
22981
22982
22983
22984
22985
22986
22987
22988
22989
22990
22991
22992
22993
22994
22995
22996
22997
22998
22999
23000
23001
23002
23003
23004
23005
23006
23007
23008
23009
23010
23011
23012
23013
23014
23015
23016
23017
23018
23019
23020
23021
23022
23023
23024
23025
23026
23027
23028
23029
23030
23031
23032
23033
23034
23035
23036
23037
23038
23039
23040
23041
23042
23043
23044
23045
23046
23047
23048
23049
23050
23051
23052
23053
23054
23055
23056
23057
23058
23059
23060
23061
23062
23063
23064
23065
23066
23067
23068
23069
23070
23071
23072
23073
23074
23075
23076
23077
23078
23079
23080
23081
23082
23083
23084
23085
23086
23087
23088
23089
23090
23091
23092
23093
23094
23095
23096
23097
23098
23099
23100
23101
23102
23103
23104
23105
23106
23107
23108
23109
23110
23111
23112
23113
23114
23115
23116
23117
23118
23119
23120
23121
23122
23123
23124
23125
23126
23127
23128
23129
23130
23131
23132
23133
23134
23135
23136
23137
23138
23139
23140
23141
23142
23143
23144
23145
23146
23147
23148
23149
23150
23151
23152
23153
23154
23155
23156
23157
23158
23159
23160
23161
23162
23163
23164
23165
23166
23167
23168
23169
23170
23171
23172
23173
23174
23175
23176
23177
23178
23179
23180
23181
23182
23183
23184
23185
23186
23187
23188
23189
23190
23191
23192
23193
23194
23195
23196
23197
23198
23199
23200
23201
23202
23203
23204
23205
23206
23207
23208
23209
23210
23211
23212
23213
23214
23215
23216
23217
23218
23219
23220
23221
23222
23223
23224
23225
23226
23227
23228
23229
23230
23231
23232
23233
23234
23235
23236
23237
23238
23239
23240
23241
23242
23243
23244
23245
23246
23247
23248
23249
23250
23251
23252
23253
23254
23255
23256
23257
23258
23259
23260
23261
23262
23263
23264
23265
23266
23267
23268
23269
23270
23271
23272
23273
23274
23275
23276
23277
23278
23279
23280
23281
23282
23283
23284
23285
23286
23287
23288
23289
23290
23291
23292
23293
23294
23295
23296
23297
23298
23299
23300
23301
23302
23303
23304
23305
23306
23307
23308
23309
23310
23311
23312
23313
23314
23315
23316
23317
23318
23319
23320
23321
23322
23323
23324
23325
23326
23327
23328
23329
23330
23331
23332
23333
23334
23335
23336
23337
23338
23339
23340
23341
23342
23343
23344
23345
23346
23347
23348
23349
23350
23351
23352
23353
23354
23355
23356
23357
23358
23359
23360
23361
23362
23363
23364
23365
23366
23367
23368
23369
23370
23371
23372
23373
23374
23375
23376
23377
23378
23379
23380
23381
23382
23383
23384
23385
23386
23387
23388
23389
23390
23391
23392
23393
23394
23395
23396
23397
23398
23399
23400
23401
23402
23403
23404
23405
23406
23407
23408
23409
23410
23411
23412
23413
23414
23415
23416
23417
23418
23419
23420
23421
23422
23423
23424
23425
23426
23427
23428
23429
23430
23431
23432
23433
23434
23435
23436
23437
23438
23439
23440
23441
23442
23443
23444
23445
23446
23447
23448
23449
23450
23451
23452
23453
23454
23455
23456
23457
23458
23459
23460
23461
23462
23463
23464
23465
23466
23467
23468
23469
23470
23471
23472
23473
23474
23475
23476
23477
23478
23479
23480
23481
23482
23483
23484
23485
23486
23487
23488
23489
23490
23491
23492
23493
23494
23495
23496
23497
23498
23499
23500
23501
23502
23503
23504
23505
23506
23507
23508
23509
23510
23511
23512
23513
23514
23515
23516
23517
23518
23519
23520
23521
23522
23523
23524
23525
23526
23527
23528
23529
23530
23531
23532
23533
23534
23535
23536
23537
23538
23539
23540
23541
23542
23543
23544
23545
23546
23547
23548
23549
23550
23551
23552
23553
23554
23555
23556
23557
23558
23559
23560
23561
23562
23563
23564
23565
23566
23567
23568
23569
23570
23571
23572
23573
23574
23575
23576
23577
23578
23579
23580
23581
23582
23583
23584
23585
23586
23587
23588
23589
23590
23591
23592
23593
23594
23595
23596
23597
23598
23599
23600
23601
23602
23603
23604
23605
23606
23607
23608
23609
23610
23611
23612
23613
23614
23615
23616
23617
23618
23619
23620
23621
23622
23623
23624
23625
23626
23627
23628
23629
23630
23631
23632
23633
23634
23635
23636
23637
23638
23639
23640
23641
23642
23643
23644
23645
23646
23647
23648
23649
23650
23651
23652
23653
23654
23655
23656
23657
23658
23659
23660
23661
23662
23663
23664
23665
23666
23667
23668
23669
23670
23671
23672
23673
23674
23675
23676
23677
23678
23679
23680
23681
23682
23683
23684
23685
23686
23687
23688
23689
23690
23691
23692
23693
23694
23695
23696
23697
23698
23699
23700
23701
23702
23703
23704
23705
23706
23707
23708
23709
23710
23711
23712
23713
23714
23715
23716
23717
23718
23719
23720
23721
23722
23723
23724
23725
23726
23727
23728
23729
23730
23731
23732
23733
23734
23735
23736
23737
23738
23739
23740
23741
23742
23743
23744
23745
23746
23747
23748
23749
23750
23751
23752
23753
23754
23755
23756
23757
23758
23759
23760
23761
23762
23763
23764
23765
23766
23767
23768
23769
23770
23771
23772
23773
23774
23775
23776
23777
23778
23779
23780
23781
23782
23783
23784
23785
23786
23787
23788
23789
23790
23791
23792
23793
23794
23795
23796
23797
23798
23799
23800
23801
23802
23803
23804
23805
23806
23807
23808
23809
23810
23811
23812
23813
23814
23815
23816
23817
23818
23819
23820
23821
23822
23823
23824
23825
23826
23827
23828
23829
23830
23831
23832
23833
23834
23835
23836
23837
23838
23839
23840
23841
23842
23843
23844
23845
23846
23847
23848
23849
23850
23851
23852
23853
23854
23855
23856
23857
23858
23859
23860
23861
23862
23863
23864
23865
23866
23867
23868
23869
23870
23871
23872
23873
23874
23875
23876
23877
23878
23879
23880
23881
23882
23883
23884
23885
23886
23887
23888
23889
23890
23891
23892
23893
23894
23895
23896
23897
23898
23899
23900
23901
23902
23903
23904
23905
23906
23907
23908
23909
23910
23911
23912
23913
23914
23915
23916
23917
23918
23919
23920
23921
23922
23923
23924
23925
23926
23927
23928
23929
23930
23931
23932
23933
23934
23935
23936
23937
23938
23939
23940
23941
23942
23943
23944
23945
23946
23947
23948
23949
23950
23951
23952
23953
23954
23955
23956
23957
23958
23959
23960
23961
23962
23963
23964
23965
23966
23967
23968
23969
23970
23971
23972
23973
23974
23975
23976
23977
23978
23979
23980
23981
23982
23983
23984
23985
23986
23987
23988
23989
23990
23991
23992
23993
23994
23995
23996
23997
23998
23999
24000
24001
24002
24003
24004
24005
24006
24007
24008
24009
24010
24011
24012
24013
24014
24015
24016
24017
24018
24019
24020
24021
24022
24023
24024
24025
24026
24027
24028
24029
24030
24031
24032
24033
24034
24035
24036
24037
24038
24039
24040
24041
24042
24043
24044
24045
24046
24047
24048
24049
24050
24051
24052
24053
24054
24055
24056
24057
24058
24059
24060
24061
24062
24063
24064
24065
24066
24067
24068
24069
24070
24071
24072
24073
24074
24075
24076
24077
24078
24079
24080
24081
24082
24083
24084
24085
24086
24087
24088
24089
24090
24091
24092
24093
24094
24095
24096
24097
24098
24099
24100
24101
24102
24103
24104
24105
24106
24107
24108
24109
24110
24111
24112
24113
24114
24115
24116
24117
24118
24119
24120
24121
24122
24123
24124
24125
24126
24127
24128
24129
24130
24131
24132
24133
24134
24135
24136
24137
24138
24139
24140
24141
24142
24143
24144
24145
24146
24147
24148
24149
24150
24151
24152
24153
24154
24155
24156
24157
24158
24159
24160
24161
24162
24163
24164
24165
24166
24167
24168
24169
24170
24171
24172
24173
24174
24175
24176
24177
24178
24179
24180
24181
24182
24183
24184
24185
24186
24187
24188
24189
24190
24191
24192
24193
24194
24195
24196
24197
24198
24199
24200
24201
24202
24203
24204
24205
24206
24207
24208
24209
24210
24211
24212
24213
24214
24215
24216
24217
24218
24219
24220
24221
24222
24223
24224
24225
24226
24227
24228
24229
24230
24231
24232
24233
24234
24235
24236
24237
24238
24239
24240
24241
24242
24243
24244
24245
24246
24247
24248
24249
24250
24251
24252
24253
24254
24255
24256
24257
24258
24259
24260
24261
24262
24263
24264
24265
24266
24267
24268
24269
24270
24271
24272
24273
24274
24275
24276
24277
24278
24279
24280
24281
24282
24283
24284
24285
24286
24287
24288
24289
24290
24291
24292
24293
24294
24295
24296
24297
24298
24299
24300
24301
24302
24303
24304
24305
24306
24307
24308
24309
24310
24311
24312
24313
24314
24315
24316
24317
24318
24319
24320
24321
24322
24323
24324
24325
24326
24327
24328
24329
24330
24331
24332
24333
24334
24335
24336
24337
24338
24339
24340
24341
24342
24343
24344
24345
24346
24347
24348
24349
24350
24351
24352
24353
24354
24355
24356
24357
24358
24359
24360
24361
24362
24363
24364
24365
24366
24367
24368
24369
24370
24371
24372
24373
24374
24375
24376
24377
24378
24379
24380
24381
24382
24383
24384
24385
24386
24387
24388
24389
24390
24391
24392
24393
24394
24395
24396
24397
24398
24399
24400
24401
24402
24403
24404
24405
24406
24407
24408
24409
24410
24411
24412
24413
24414
24415
24416
24417
24418
24419
24420
24421
24422
24423
24424
24425
24426
24427
24428
24429
24430
24431
24432
24433
24434
24435
24436
24437
24438
24439
24440
24441
24442
24443
24444
24445
24446
24447
24448
24449
24450
24451
24452
24453
24454
24455
24456
24457
24458
24459
24460
24461
24462
24463
24464
24465
24466
24467
24468
24469
24470
24471
24472
24473
24474
24475
24476
24477
24478
24479
24480
24481
24482
24483
24484
24485
24486
24487
24488
24489
24490
24491
24492
24493
24494
24495
24496
24497
24498
24499
24500
24501
24502
24503
24504
24505
24506
24507
24508
24509
24510
24511
24512
24513
24514
24515
24516
24517
24518
24519
24520
24521
24522
24523
24524
24525
24526
24527
24528
24529
24530
24531
24532
24533
24534
24535
24536
24537
24538
24539
24540
24541
24542
24543
24544
24545
24546
24547
24548
24549
24550
24551
24552
24553
24554
24555
24556
24557
24558
24559
24560
24561
24562
24563
24564
24565
24566
24567
24568
24569
24570
24571
24572
24573
24574
24575
24576
24577
24578
24579
24580
24581
24582
24583
24584
24585
24586
24587
24588
24589
24590
24591
24592
24593
24594
24595
24596
24597
24598
24599
24600
24601
24602
24603
24604
24605
24606
24607
24608
24609
24610
24611
24612
24613
24614
24615
24616
24617
24618
24619
24620
24621
24622
24623
24624
24625
24626
24627
24628
24629
24630
24631
24632
24633
24634
24635
24636
24637
24638
24639
24640
24641
24642
24643
24644
24645
24646
24647
24648
24649
24650
24651
24652
24653
24654
24655
24656
24657
24658
24659
24660
24661
24662
24663
24664
24665
24666
24667
24668
24669
24670
24671
24672
24673
24674
24675
24676
24677
24678
24679
24680
24681
24682
24683
24684
24685
24686
24687
24688
24689
24690
24691
24692
24693
24694
24695
24696
24697
24698
24699
24700
24701
24702
24703
24704
24705
24706
24707
24708
24709
24710
24711
24712
24713
24714
24715
24716
24717
24718
24719
24720
24721
24722
24723
24724
24725
24726
24727
24728
24729
24730
24731
24732
24733
24734
24735
24736
24737
24738
24739
24740
24741
24742
24743
24744
24745
24746
24747
24748
24749
24750
24751
24752
24753
24754
24755
24756
24757
24758
24759
24760
24761
24762
24763
24764
24765
24766
24767
24768
24769
24770
24771
24772
24773
24774
24775
24776
24777
24778
24779
24780
24781
24782
24783
24784
24785
24786
24787
24788
24789
24790
24791
24792
24793
24794
24795
24796
24797
24798
24799
24800
24801
24802
24803
24804
24805
24806
24807
24808
24809
24810
24811
24812
24813
24814
24815
24816
24817
24818
24819
24820
24821
24822
24823
24824
24825
24826
24827
24828
24829
24830
24831
24832
24833
24834
24835
24836
24837
24838
24839
24840
24841
24842
24843
24844
24845
24846
24847
24848
24849
24850
24851
24852
24853
24854
24855
24856
24857
24858
24859
24860
24861
24862
24863
24864
24865
24866
24867
24868
24869
24870
24871
24872
24873
24874
24875
24876
24877
24878
24879
24880
24881
24882
24883
24884
24885
24886
24887
24888
24889
24890
24891
24892
24893
24894
24895
24896
24897
24898
24899
24900
24901
24902
24903
24904
24905
24906
24907
24908
24909
24910
24911
24912
24913
24914
24915
24916
24917
24918
24919
24920
24921
24922
24923
24924
24925
24926
24927
24928
24929
24930
24931
24932
24933
24934
24935
24936
24937
24938
24939
24940
24941
24942
24943
24944
24945
24946
24947
24948
24949
24950
24951
24952
24953
24954
24955
24956
24957
24958
24959
24960
24961
24962
24963
24964
24965
24966
24967
24968
24969
24970
24971
24972
24973
24974
24975
24976
24977
24978
24979
24980
24981
24982
24983
24984
24985
24986
24987
24988
24989
24990
24991
24992
24993
24994
24995
24996
24997
24998
24999
25000
25001
25002
25003
25004
25005
25006
25007
25008
25009
25010
25011
25012
25013
25014
25015
25016
25017
25018
25019
25020
25021
25022
25023
25024
25025
25026
25027
25028
25029
25030
25031
25032
25033
25034
25035
25036
25037
25038
25039
25040
25041
25042
25043
25044
25045
25046
25047
25048
25049
25050
25051
25052
25053
25054
25055
25056
25057
25058
25059
25060
25061
25062
25063
25064
25065
25066
25067
25068
25069
25070
25071
25072
25073
25074
25075
25076
25077
25078
25079
25080
25081
25082
25083
25084
25085
25086
25087
25088
25089
25090
25091
25092
25093
25094
25095
25096
25097
25098
25099
25100
25101
25102
25103
25104
25105
25106
25107
25108
25109
25110
25111
25112
25113
25114
25115
25116
25117
25118
25119
25120
25121
25122
25123
25124
25125
25126
25127
25128
25129
25130
25131
25132
25133
25134
25135
25136
25137
25138
25139
25140
25141
25142
25143
25144
25145
25146
25147
25148
25149
25150
25151
25152
25153
25154
25155
25156
25157
25158
25159
25160
25161
25162
25163
25164
25165
25166
25167
25168
25169
25170
25171
25172
25173
25174
25175
25176
25177
25178
25179
25180
25181
25182
25183
25184
25185
25186
25187
25188
25189
25190
25191
25192
25193
25194
25195
25196
25197
25198
25199
25200
25201
25202
25203
25204
25205
25206
25207
25208
25209
25210
25211
25212
25213
25214
25215
25216
25217
25218
25219
25220
25221
25222
25223
25224
25225
25226
25227
25228
25229
25230
25231
25232
25233
25234
25235
25236
25237
25238
25239
25240
25241
25242
25243
25244
25245
25246
25247
25248
25249
25250
25251
25252
25253
25254
25255
25256
25257
25258
25259
25260
25261
25262
25263
25264
25265
25266
25267
25268
25269
25270
25271
25272
25273
25274
25275
25276
25277
25278
25279
25280
25281
25282
25283
25284
25285
25286
25287
25288
25289
25290
25291
25292
25293
25294
25295
25296
25297
25298
25299
25300
25301
25302
25303
25304
25305
25306
25307
25308
25309
25310
25311
25312
25313
25314
25315
25316
25317
25318
25319
25320
25321
25322
25323
25324
25325
25326
25327
25328
25329
25330
25331
25332
25333
25334
25335
25336
25337
25338
25339
25340
25341
25342
25343
25344
25345
25346
25347
25348
25349
25350
25351
25352
25353
25354
25355
25356
25357
25358
25359
25360
25361
25362
25363
25364
25365
25366
25367
25368
25369
25370
25371
25372
25373
25374
25375
25376
25377
25378
25379
25380
25381
25382
25383
25384
25385
25386
25387
25388
25389
25390
25391
25392
25393
25394
25395
25396
25397
25398
25399
25400
25401
25402
25403
25404
25405
25406
25407
25408
25409
25410
25411
25412
25413
25414
25415
25416
25417
25418
25419
25420
25421
25422
25423
25424
25425
25426
25427
25428
25429
25430
25431
25432
25433
25434
25435
25436
25437
25438
25439
25440
25441
25442
25443
25444
25445
25446
25447
25448
25449
25450
25451
25452
25453
25454
25455
25456
25457
25458
25459
25460
25461
25462
25463
25464
25465
25466
25467
25468
25469
25470
25471
25472
25473
25474
25475
25476
25477
25478
25479
25480
25481
25482
25483
25484
25485
25486
25487
25488
25489
25490
25491
25492
25493
25494
25495
25496
25497
25498
25499
25500
25501
25502
25503
25504
25505
25506
25507
25508
25509
25510
25511
25512
25513
25514
25515
25516
25517
25518
25519
25520
25521
25522
25523
25524
25525
25526
25527
25528
25529
25530
25531
25532
25533
25534
25535
25536
25537
25538
25539
25540
25541
25542
25543
25544
25545
25546
25547
25548
25549
25550
25551
25552
25553
25554
25555
25556
25557
25558
25559
25560
25561
25562
25563
25564
25565
25566
25567
25568
25569
25570
25571
25572
25573
25574
25575
25576
25577
25578
25579
25580
25581
25582
25583
25584
25585
25586
25587
25588
25589
25590
25591
25592
25593
25594
25595
25596
25597
25598
25599
25600
25601
25602
25603
25604
25605
25606
25607
25608
25609
25610
25611
25612
25613
25614
25615
25616
25617
25618
25619
25620
25621
25622
25623
25624
25625
25626
25627
25628
25629
25630
25631
25632
25633
25634
25635
25636
25637
25638
25639
25640
25641
25642
25643
25644
25645
25646
25647
25648
25649
25650
25651
25652
25653
25654
25655
25656
25657
25658
25659
25660
25661
25662
25663
25664
25665
25666
25667
25668
25669
25670
25671
25672
25673
25674
25675
25676
25677
25678
25679
25680
25681
25682
25683
25684
25685
25686
25687
25688
25689
25690
25691
25692
25693
25694
25695
25696
25697
25698
25699
25700
25701
25702
25703
25704
25705
25706
25707
25708
25709
25710
25711
25712
25713
25714
25715
25716
25717
25718
25719
25720
25721
25722
25723
25724
25725
25726
25727
25728
25729
25730
25731
25732
25733
25734
25735
25736
25737
25738
25739
25740
25741
25742
25743
25744
25745
25746
25747
25748
25749
25750
25751
25752
25753
25754
25755
25756
25757
25758
25759
25760
25761
25762
25763
25764
25765
25766
25767
25768
25769
25770
25771
25772
25773
25774
25775
25776
25777
25778
25779
25780
25781
25782
25783
25784
25785
25786
25787
25788
25789
25790
25791
25792
25793
25794
25795
25796
25797
25798
25799
25800
25801
25802
25803
25804
25805
25806
25807
25808
25809
25810
25811
25812
25813
25814
25815
25816
25817
25818
25819
25820
25821
25822
25823
25824
25825
25826
25827
25828
25829
25830
25831
25832
25833
25834
25835
25836
25837
25838
25839
25840
25841
25842
25843
25844
25845
25846
25847
25848
25849
25850
25851
25852
25853
25854
25855
25856
25857
25858
25859
25860
25861
25862
25863
25864
25865
25866
25867
25868
25869
25870
25871
25872
25873
25874
25875
25876
25877
25878
25879
25880
25881
25882
25883
25884
25885
25886
25887
25888
25889
25890
25891
25892
25893
25894
25895
25896
25897
25898
25899
25900
25901
25902
25903
25904
25905
25906
25907
25908
25909
25910
25911
25912
25913
25914
25915
25916
25917
25918
25919
25920
25921
25922
25923
25924
25925
25926
25927
25928
25929
25930
25931
25932
25933
25934
25935
25936
25937
25938
25939
25940
25941
25942
25943
25944
25945
25946
25947
25948
25949
25950
25951
25952
25953
25954
25955
25956
25957
25958
25959
25960
25961
25962
25963
25964
25965
25966
25967
25968
25969
25970
25971
25972
25973
25974
25975
25976
25977
25978
25979
25980
25981
25982
25983
25984
25985
25986
25987
25988
25989
25990
25991
25992
25993
25994
25995
25996
25997
25998
25999
26000
26001
26002
26003
26004
26005
26006
26007
26008
26009
26010
26011
26012
26013
26014
26015
26016
26017
26018
26019
26020
26021
26022
26023
26024
26025
26026
26027
26028
26029
26030
26031
26032
26033
26034
26035
26036
26037
26038
26039
26040
26041
26042
26043
26044
26045
26046
26047
26048
26049
26050
26051
26052
26053
26054
26055
26056
26057
26058
26059
26060
26061
26062
26063
26064
26065
26066
26067
26068
26069
26070
26071
26072
26073
26074
26075
26076
26077
26078
26079
26080
26081
26082
26083
26084
26085
26086
26087
26088
26089
26090
26091
26092
26093
26094
26095
26096
26097
26098
26099
26100
26101
26102
26103
26104
26105
26106
26107
26108
26109
26110
26111
26112
26113
26114
26115
26116
26117
26118
26119
26120
26121
26122
26123
26124
26125
26126
26127
26128
26129
26130
26131
26132
26133
26134
26135
26136
26137
26138
26139
26140
26141
26142
26143
26144
26145
26146
26147
26148
26149
26150
26151
26152
26153
26154
26155
26156
26157
26158
26159
26160
26161
26162
26163
26164
26165
26166
26167
26168
26169
26170
26171
26172
26173
26174
26175
26176
26177
26178
26179
26180
26181
26182
26183
26184
26185
26186
26187
26188
26189
26190
26191
26192
26193
26194
26195
26196
26197
26198
26199
26200
26201
26202
26203
26204
26205
26206
26207
26208
26209
26210
26211
26212
26213
26214
26215
26216
26217
26218
26219
26220
26221
26222
26223
26224
26225
26226
26227
26228
26229
26230
26231
26232
26233
26234
26235
26236
26237
26238
26239
26240
26241
26242
26243
26244
26245
26246
26247
26248
26249
26250
26251
26252
26253
26254
26255
26256
26257
26258
26259
26260
26261
26262
26263
26264
26265
26266
26267
26268
26269
26270
26271
26272
26273
26274
26275
26276
26277
26278
26279
26280
26281
26282
26283
26284
26285
26286
26287
26288
26289
26290
26291
26292
26293
26294
26295
26296
26297
26298
26299
26300
26301
26302
26303
26304
26305
26306
26307
26308
26309
26310
26311
26312
26313
26314
26315
26316
26317
26318
26319
26320
26321
26322
26323
26324
26325
26326
26327
26328
26329
26330
26331
26332
26333
26334
26335
26336
26337
26338
26339
26340
26341
26342
26343
26344
26345
26346
26347
26348
26349
26350
26351
26352
26353
26354
26355
26356
26357
26358
26359
26360
26361
26362
26363
26364
26365
26366
26367
26368
26369
26370
26371
26372
26373
26374
26375
26376
26377
26378
26379
26380
26381
26382
26383
26384
26385
26386
26387
26388
26389
26390
26391
26392
26393
26394
26395
26396
26397
26398
26399
26400
26401
26402
26403
26404
26405
26406
26407
26408
26409
26410
26411
26412
26413
26414
26415
26416
26417
26418
26419
26420
26421
26422
26423
26424
26425
26426
26427
26428
26429
26430
26431
26432
26433
26434
26435
26436
26437
26438
26439
26440
26441
26442
26443
26444
26445
26446
26447
26448
26449
26450
26451
26452
26453
26454
26455
26456
26457
26458
26459
26460
26461
26462
26463
26464
26465
26466
26467
26468
26469
26470
26471
26472
26473
26474
26475
26476
26477
26478
26479
26480
26481
26482
26483
26484
26485
26486
26487
26488
26489
26490
26491
26492
26493
26494
26495
26496
26497
26498
26499
26500
26501
26502
26503
26504
26505
26506
26507
26508
26509
26510
26511
26512
26513
26514
26515
26516
26517
26518
26519
26520
26521
26522
26523
26524
26525
26526
26527
26528
26529
26530
26531
26532
26533
26534
26535
26536
26537
26538
26539
26540
26541
26542
26543
26544
26545
26546
26547
26548
26549
26550
26551
26552
26553
26554
26555
26556
26557
26558
26559
26560
26561
26562
26563
26564
26565
26566
26567
26568
26569
26570
26571
26572
26573
26574
26575
26576
26577
26578
26579
26580
26581
26582
26583
26584
26585
26586
26587
26588
26589
26590
26591
26592
26593
26594
26595
26596
26597
26598
26599
26600
26601
26602
26603
26604
26605
26606
26607
26608
26609
26610
26611
26612
26613
26614
26615
26616
26617
26618
26619
26620
26621
26622
26623
26624
26625
26626
26627
26628
26629
26630
26631
26632
26633
26634
26635
26636
26637
26638
26639
26640
26641
26642
26643
26644
26645
26646
26647
26648
26649
26650
26651
26652
26653
26654
26655
26656
26657
26658
26659
26660
26661
26662
26663
26664
26665
26666
26667
26668
26669
26670
26671
26672
26673
26674
26675
26676
26677
26678
26679
26680
26681
26682
26683
26684
26685
26686
26687
26688
26689
26690
26691
26692
26693
26694
26695
26696
26697
26698
26699
26700
26701
26702
26703
26704
26705
26706
26707
26708
26709
26710
26711
26712
26713
26714
26715
26716
26717
26718
26719
26720
26721
26722
26723
26724
26725
26726
26727
26728
26729
26730
26731
26732
26733
26734
26735
26736
26737
26738
26739
26740
26741
26742
26743
26744
26745
26746
26747
26748
26749
26750
26751
26752
26753
26754
26755
26756
26757
26758
26759
26760
26761
26762
26763
26764
26765
26766
26767
26768
26769
26770
26771
26772
26773
26774
26775
26776
26777
26778
26779
26780
26781
26782
26783
26784
26785
26786
26787
26788
26789
26790
26791
26792
26793
26794
26795
26796
26797
26798
26799
26800
26801
26802
26803
26804
26805
26806
26807
26808
26809
26810
26811
26812
26813
26814
26815
26816
26817
26818
26819
26820
26821
26822
26823
26824
26825
26826
26827
26828
26829
26830
26831
26832
26833
26834
26835
26836
26837
26838
26839
26840
26841
26842
26843
26844
26845
26846
26847
26848
26849
26850
26851
26852
26853
26854
26855
26856
26857
26858
26859
26860
26861
26862
26863
26864
26865
26866
26867
26868
26869
26870
26871
26872
26873
26874
26875
26876
26877
26878
26879
26880
26881
26882
26883
26884
26885
26886
26887
26888
26889
26890
26891
26892
26893
26894
26895
26896
26897
26898
26899
26900
26901
26902
26903
26904
26905
26906
26907
26908
26909
26910
26911
26912
26913
26914
26915
26916
26917
26918
26919
26920
26921
26922
26923
26924
26925
26926
26927
26928
26929
26930
26931
26932
26933
26934
26935
26936
26937
26938
26939
26940
26941
26942
26943
26944
26945
26946
26947
26948
26949
26950
26951
26952
26953
26954
26955
26956
26957
26958
26959
26960
26961
26962
26963
26964
26965
26966
26967
26968
26969
26970
26971
26972
26973
26974
26975
26976
26977
26978
26979
26980
26981
26982
26983
26984
26985
26986
26987
26988
26989
26990
26991
26992
26993
26994
26995
26996
26997
26998
26999
27000
27001
27002
27003
27004
27005
27006
27007
27008
27009
27010
27011
27012
27013
27014
27015
27016
27017
27018
27019
27020
27021
27022
27023
27024
27025
27026
27027
27028
27029
27030
27031
27032
27033
27034
27035
27036
27037
27038
27039
27040
27041
27042
27043
27044
27045
27046
27047
27048
27049
27050
27051
27052
27053
27054
27055
27056
27057
27058
27059
27060
27061
27062
27063
27064
27065
27066
27067
27068
27069
27070
27071
27072
27073
27074
27075
27076
27077
27078
27079
27080
27081
27082
27083
27084
27085
27086
27087
27088
27089
27090
27091
27092
27093
27094
27095
27096
27097
27098
27099
27100
27101
27102
27103
27104
27105
27106
27107
27108
27109
27110
27111
27112
27113
27114
27115
27116
27117
27118
27119
27120
27121
27122
27123
27124
27125
27126
27127
27128
27129
27130
27131
27132
27133
27134
27135
27136
27137
27138
27139
27140
27141
27142
27143
27144
27145
27146
27147
27148
27149
27150
27151
27152
27153
27154
27155
27156
27157
27158
27159
27160
27161
27162
27163
27164
27165
27166
27167
27168
27169
27170
27171
27172
27173
27174
27175
27176
27177
27178
27179
27180
27181
27182
27183
27184
27185
27186
27187
27188
27189
27190
27191
27192
27193
27194
27195
27196
27197
27198
27199
27200
27201
27202
27203
27204
27205
27206
27207
27208
27209
27210
27211
27212
27213
27214
27215
27216
27217
27218
27219
27220
27221
27222
27223
27224
27225
27226
27227
27228
27229
27230
27231
27232
27233
27234
27235
27236
27237
27238
27239
27240
27241
27242
27243
27244
27245
27246
27247
27248
27249
27250
27251
27252
27253
27254
27255
27256
27257
27258
27259
27260
27261
27262
27263
27264
27265
27266
27267
27268
27269
27270
27271
27272
27273
27274
27275
27276
27277
27278
27279
27280
27281
27282
27283
27284
27285
27286
27287
27288
27289
27290
27291
27292
27293
27294
27295
27296
27297
27298
27299
27300
27301
27302
27303
27304
27305
27306
27307
27308
27309
27310
27311
27312
27313
27314
27315
27316
27317
27318
27319
27320
27321
27322
27323
27324
27325
27326
27327
27328
27329
27330
27331
27332
27333
27334
27335
27336
27337
27338
27339
27340
27341
27342
27343
27344
27345
27346
27347
27348
27349
27350
27351
27352
27353
27354
27355
27356
27357
27358
27359
27360
27361
27362
27363
27364
27365
27366
27367
27368
27369
27370
27371
27372
27373
27374
27375
27376
27377
27378
27379
27380
27381
27382
27383
27384
27385
27386
27387
27388
27389
27390
27391
27392
27393
27394
27395
27396
27397
27398
27399
27400
27401
27402
27403
27404
27405
27406
27407
27408
27409
27410
27411
27412
27413
27414
27415
27416
27417
27418
27419
27420
27421
27422
27423
27424
27425
27426
27427
27428
27429
27430
27431
27432
27433
27434
27435
27436
27437
27438
27439
27440
27441
27442
27443
27444
27445
27446
27447
27448
27449
27450
27451
27452
27453
27454
27455
27456
27457
27458
27459
27460
27461
27462
27463
27464
27465
27466
27467
27468
27469
27470
27471
27472
27473
27474
27475
27476
27477
27478
27479
27480
27481
27482
27483
27484
27485
27486
27487
27488
27489
27490
27491
27492
27493
27494
27495
27496
27497
27498
27499
27500
27501
27502
27503
27504
27505
27506
27507
27508
27509
27510
27511
27512
27513
27514
27515
27516
27517
27518
27519
27520
27521
27522
27523
27524
27525
27526
27527
27528
27529
27530
27531
27532
27533
27534
27535
27536
27537
27538
27539
27540
27541
27542
27543
27544
27545
27546
27547
27548
27549
27550
27551
27552
27553
27554
27555
27556
27557
27558
27559
27560
27561
27562
27563
27564
27565
27566
27567
27568
27569
27570
27571
27572
27573
27574
27575
27576
27577
27578
27579
27580
27581
27582
27583
27584
27585
27586
27587
27588
27589
27590
27591
27592
27593
27594
27595
27596
27597
27598
27599
27600
27601
27602
27603
27604
27605
27606
27607
27608
27609
27610
27611
27612
27613
27614
27615
27616
27617
27618
27619
27620
27621
27622
27623
27624
27625
27626
27627
27628
27629
27630
27631
27632
27633
27634
27635
27636
27637
27638
27639
27640
27641
27642
27643
27644
27645
27646
27647
27648
27649
27650
27651
27652
27653
27654
27655
27656
27657
27658
27659
27660
27661
27662
27663
27664
27665
27666
27667
27668
27669
27670
27671
27672
27673
27674
27675
27676
27677
27678
27679
27680
27681
27682
27683
27684
27685
27686
27687
27688
27689
27690
27691
27692
27693
27694
27695
27696
27697
27698
27699
27700
27701
27702
27703
27704
27705
27706
27707
27708
27709
27710
27711
27712
27713
27714
27715
27716
27717
27718
27719
27720
27721
27722
27723
27724
27725
27726
27727
27728
27729
27730
27731
27732
27733
27734
27735
27736
27737
27738
27739
27740
27741
27742
27743
27744
27745
27746
27747
27748
27749
27750
27751
27752
27753
27754
27755
27756
27757
27758
27759
27760
27761
27762
27763
27764
27765
27766
27767
27768
27769
27770
27771
27772
27773
27774
27775
27776
27777
27778
27779
27780
27781
27782
27783
27784
27785
27786
27787
27788
27789
27790
27791
27792
27793
27794
27795
27796
27797
27798
27799
27800
27801
27802
27803
27804
27805
27806
27807
27808
27809
27810
27811
27812
27813
27814
27815
27816
27817
27818
27819
27820
27821
27822
27823
27824
27825
27826
27827
27828
27829
27830
27831
27832
27833
27834
27835
27836
27837
27838
27839
27840
27841
27842
27843
27844
27845
27846
27847
27848
27849
27850
27851
27852
27853
27854
27855
27856
27857
27858
27859
27860
27861
27862
27863
27864
27865
27866
27867
27868
27869
27870
27871
27872
27873
27874
27875
27876
27877
27878
27879
27880
27881
27882
27883
27884
27885
27886
27887
27888
27889
27890
27891
27892
27893
27894
27895
27896
27897
27898
27899
27900
27901
27902
27903
27904
27905
27906
27907
27908
27909
27910
27911
27912
27913
27914
27915
27916
27917
27918
27919
27920
27921
27922
27923
27924
27925
27926
27927
27928
27929
27930
27931
27932
27933
27934
27935
27936
27937
27938
27939
27940
27941
27942
27943
27944
27945
27946
27947
27948
27949
27950
27951
27952
27953
27954
27955
27956
27957
27958
27959
27960
27961
27962
27963
27964
27965
27966
27967
27968
27969
27970
27971
27972
27973
27974
27975
27976
27977
27978
27979
27980
27981
27982
27983
27984
27985
27986
27987
27988
27989
27990
27991
27992
27993
27994
27995
27996
27997
27998
27999
28000
28001
28002
28003
28004
28005
28006
28007
28008
28009
28010
28011
28012
28013
28014
28015
28016
28017
28018
28019
28020
28021
28022
28023
28024
28025
28026
28027
28028
28029
28030
28031
28032
28033
28034
28035
28036
28037
28038
28039
28040
28041
28042
28043
28044
28045
28046
28047
28048
28049
28050
28051
28052
28053
28054
28055
28056
28057
28058
28059
28060
28061
28062
28063
28064
28065
28066
28067
28068
28069
28070
28071
28072
28073
28074
28075
28076
28077
28078
28079
28080
28081
28082
28083
28084
28085
28086
28087
28088
28089
28090
28091
28092
28093
28094
28095
28096
28097
28098
28099
28100
28101
28102
28103
28104
28105
28106
28107
28108
28109
28110
28111
28112
28113
28114
28115
28116
28117
28118
28119
28120
28121
28122
28123
28124
28125
28126
28127
28128
28129
28130
28131
28132
28133
28134
28135
28136
28137
28138
28139
28140
28141
28142
28143
28144
28145
28146
28147
28148
28149
28150
28151
28152
28153
28154
28155
28156
28157
28158
28159
28160
28161
28162
28163
28164
28165
28166
28167
28168
28169
28170
28171
28172
28173
28174
28175
28176
28177
28178
28179
28180
28181
28182
28183
28184
28185
28186
28187
28188
28189
28190
28191
28192
28193
28194
28195
28196
28197
28198
28199
28200
28201
28202
28203
28204
28205
28206
28207
28208
28209
28210
28211
28212
28213
28214
28215
28216
28217
28218
28219
28220
28221
28222
28223
28224
28225
28226
28227
28228
28229
28230
28231
28232
28233
28234
28235
28236
28237
28238
28239
28240
28241
28242
28243
28244
28245
28246
28247
28248
28249
28250
28251
28252
28253
28254
28255
28256
28257
28258
28259
28260
28261
28262
28263
28264
28265
28266
28267
28268
28269
28270
28271
28272
28273
28274
28275
28276
28277
28278
28279
28280
28281
28282
28283
28284
28285
28286
28287
28288
28289
28290
28291
28292
28293
28294
28295
28296
28297
28298
28299
28300
28301
28302
28303
28304
28305
28306
28307
28308
28309
28310
28311
28312
28313
28314
28315
28316
28317
28318
28319
28320
28321
28322
28323
28324
28325
28326
28327
28328
28329
28330
28331
28332
28333
28334
28335
28336
28337
28338
28339
28340
28341
28342
28343
28344
28345
28346
28347
28348
28349
28350
28351
28352
28353
28354
28355
28356
28357
28358
28359
28360
28361
28362
28363
28364
28365
28366
28367
28368
28369
28370
28371
28372
28373
28374
28375
28376
28377
28378
28379
28380
28381
28382
28383
28384
28385
28386
28387
28388
28389
28390
28391
28392
28393
28394
28395
28396
28397
28398
28399
28400
28401
28402
28403
28404
28405
28406
28407
28408
28409
28410
28411
28412
28413
28414
28415
28416
28417
28418
28419
28420
28421
28422
28423
28424
28425
28426
28427
28428
28429
28430
28431
28432
28433
28434
28435
28436
28437
28438
28439
28440
28441
28442
28443
28444
28445
28446
28447
28448
28449
28450
28451
28452
28453
28454
28455
28456
28457
28458
28459
28460
28461
28462
28463
28464
28465
28466
28467
28468
28469
28470
28471
28472
28473
28474
28475
28476
28477
28478
28479
28480
28481
28482
28483
28484
28485
28486
28487
28488
28489
28490
28491
28492
28493
28494
28495
28496
28497
28498
28499
28500
28501
28502
28503
28504
28505
28506
28507
28508
28509
28510
28511
28512
28513
28514
28515
28516
28517
28518
28519
28520
28521
28522
28523
28524
28525
28526
28527
28528
28529
28530
28531
28532
28533
28534
28535
28536
28537
28538
28539
28540
28541
28542
28543
28544
28545
28546
28547
28548
28549
28550
28551
28552
28553
28554
28555
28556
28557
28558
28559
28560
28561
28562
28563
28564
28565
28566
28567
28568
28569
28570
28571
28572
28573
28574
28575
28576
28577
28578
28579
28580
28581
28582
28583
28584
28585
28586
28587
28588
28589
28590
28591
28592
28593
28594
28595
28596
28597
28598
28599
28600
28601
28602
28603
28604
28605
28606
28607
28608
28609
28610
28611
28612
28613
28614
28615
28616
28617
28618
28619
28620
28621
28622
28623
28624
28625
28626
28627
28628
28629
28630
28631
28632
28633
28634
28635
28636
28637
28638
28639
28640
28641
28642
28643
28644
28645
28646
28647
28648
28649
28650
28651
28652
28653
28654
28655
28656
28657
28658
28659
28660
28661
28662
28663
28664
28665
28666
28667
28668
28669
28670
28671
28672
28673
28674
28675
28676
28677
28678
28679
28680
28681
28682
28683
28684
28685
28686
28687
28688
28689
28690
28691
28692
28693
28694
28695
28696
28697
28698
28699
28700
28701
28702
28703
28704
28705
28706
28707
28708
28709
28710
28711
28712
28713
28714
28715
28716
28717
28718
28719
28720
28721
28722
28723
28724
28725
28726
28727
28728
28729
28730
28731
28732
28733
28734
28735
28736
28737
28738
28739
28740
28741
28742
28743
28744
28745
28746
28747
28748
28749
28750
28751
28752
28753
28754
28755
28756
28757
28758
28759
28760
28761
28762
28763
28764
28765
28766
28767
28768
28769
28770
28771
28772
28773
28774
28775
28776
28777
28778
28779
28780
28781
28782
28783
28784
28785
28786
28787
28788
28789
28790
28791
28792
28793
28794
28795
28796
28797
28798
28799
28800
28801
28802
28803
28804
28805
28806
28807
28808
28809
28810
28811
28812
28813
28814
28815
28816
28817
28818
28819
28820
28821
28822
28823
28824
28825
28826
28827
28828
28829
28830
28831
28832
28833
28834
28835
28836
28837
28838
28839
28840
28841
28842
28843
28844
28845
28846
28847
28848
28849
28850
28851
28852
28853
28854
28855
28856
28857
28858
28859
28860
28861
28862
28863
28864
28865
28866
28867
28868
28869
28870
28871
28872
28873
28874
28875
28876
28877
28878
28879
28880
28881
28882
28883
28884
28885
28886
28887
28888
28889
28890
28891
28892
28893
28894
28895
28896
28897
28898
28899
28900
28901
28902
28903
28904
28905
28906
28907
28908
28909
28910
28911
28912
28913
28914
28915
28916
28917
28918
28919
28920
28921
28922
28923
28924
28925
28926
28927
28928
28929
28930
28931
28932
28933
28934
28935
28936
28937
28938
28939
28940
28941
28942
28943
28944
28945
28946
28947
28948
28949
28950
28951
28952
28953
28954
28955
28956
28957
28958
28959
28960
28961
28962
28963
28964
28965
28966
28967
28968
28969
28970
28971
28972
28973
28974
28975
28976
28977
28978
28979
28980
28981
28982
28983
28984
28985
28986
28987
28988
28989
28990
28991
28992
28993
28994
28995
28996
28997
28998
28999
29000
29001
29002
29003
29004
29005
29006
29007
29008
29009
29010
29011
29012
29013
29014
29015
29016
29017
29018
29019
29020
29021
29022
29023
29024
29025
29026
29027
29028
29029
29030
29031
29032
29033
29034
29035
29036
29037
29038
29039
29040
29041
29042
29043
29044
29045
29046
29047
29048
29049
29050
29051
29052
29053
29054
29055
29056
29057
29058
29059
29060
29061
29062
29063
29064
29065
29066
29067
29068
29069
29070
29071
29072
29073
29074
29075
29076
29077
29078
29079
29080
29081
29082
29083
29084
29085
29086
29087
29088
29089
29090
29091
29092
29093
29094
29095
29096
29097
29098
29099
29100
29101
29102
29103
29104
29105
29106
29107
29108
29109
29110
29111
29112
29113
29114
29115
29116
29117
29118
29119
29120
29121
29122
29123
29124
29125
29126
29127
29128
29129
29130
29131
29132
29133
29134
29135
29136
29137
29138
29139
29140
29141
29142
29143
29144
29145
29146
29147
29148
29149
29150
29151
29152
29153
29154
29155
29156
29157
29158
29159
29160
29161
29162
29163
29164
29165
29166
29167
29168
29169
29170
29171
29172
29173
29174
29175
29176
29177
29178
29179
29180
29181
29182
29183
29184
29185
29186
29187
29188
29189
29190
29191
29192
29193
29194
29195
29196
29197
29198
29199
29200
29201
29202
29203
29204
29205
29206
29207
29208
29209
29210
29211
29212
29213
29214
29215
29216
29217
29218
29219
29220
29221
29222
29223
29224
29225
29226
29227
29228
29229
29230
29231
29232
29233
29234
29235
29236
29237
29238
29239
29240
29241
29242
29243
29244
29245
29246
29247
29248
29249
29250
29251
29252
29253
29254
29255
29256
29257
29258
29259
29260
29261
29262
29263
29264
29265
29266
29267
29268
29269
29270
29271
29272
29273
29274
29275
29276
29277
29278
29279
29280
29281
29282
29283
29284
29285
29286
29287
29288
29289
29290
29291
29292
29293
29294
29295
29296
29297
29298
29299
29300
29301
29302
29303
29304
29305
29306
29307
29308
29309
29310
29311
29312
29313
29314
29315
29316
29317
29318
29319
29320
29321
29322
29323
29324
29325
29326
29327
29328
29329
29330
29331
29332
29333
29334
29335
29336
29337
29338
29339
29340
29341
29342
29343
29344
29345
29346
29347
29348
29349
29350
29351
29352
29353
29354
29355
29356
29357
29358
29359
29360
29361
29362
29363
29364
29365
29366
29367
29368
29369
29370
29371
29372
29373
29374
29375
29376
29377
29378
29379
29380
29381
29382
29383
29384
29385
29386
29387
29388
29389
29390
29391
29392
29393
29394
29395
29396
29397
29398
29399
29400
29401
29402
29403
29404
29405
29406
29407
29408
29409
29410
29411
29412
29413
29414
29415
29416
29417
29418
29419
29420
29421
29422
29423
29424
29425
29426
29427
29428
29429
29430
29431
29432
29433
29434
29435
29436
29437
29438
29439
29440
29441
29442
29443
29444
29445
29446
29447
29448
29449
29450
29451
29452
29453
29454
29455
29456
29457
29458
29459
29460
29461
29462
29463
29464
29465
29466
29467
29468
29469
29470
29471
29472
29473
29474
29475
29476
29477
29478
29479
29480
29481
29482
29483
29484
29485
29486
29487
29488
29489
29490
29491
29492
29493
29494
29495
29496
29497
29498
29499
29500
29501
29502
29503
29504
29505
29506
29507
29508
29509
29510
29511
29512
29513
29514
29515
29516
29517
29518
29519
29520
29521
29522
29523
29524
29525
29526
29527
29528
29529
29530
29531
29532
29533
29534
29535
29536
29537
29538
29539
29540
29541
29542
29543
29544
29545
29546
29547
29548
29549
29550
29551
29552
29553
29554
29555
29556
29557
29558
29559
29560
29561
29562
29563
29564
29565
29566
29567
29568
29569
29570
29571
29572
29573
29574
29575
29576
29577
29578
29579
29580
29581
29582
29583
29584
29585
29586
29587
29588
29589
29590
29591
29592
29593
29594
29595
29596
29597
29598
29599
29600
29601
29602
29603
29604
29605
29606
29607
29608
29609
29610
29611
29612
29613
29614
29615
29616
29617
29618
29619
29620
29621
29622
29623
29624
29625
29626
29627
29628
29629
29630
29631
29632
29633
29634
29635
29636
29637
29638
29639
29640
29641
29642
29643
29644
29645
29646
29647
29648
29649
29650
29651
29652
29653
29654
29655
29656
29657
29658
29659
29660
29661
29662
29663
29664
29665
29666
29667
29668
29669
29670
29671
29672
29673
29674
29675
29676
29677
29678
29679
29680
29681
29682
29683
29684
29685
29686
29687
29688
29689
29690
29691
29692
29693
29694
29695
29696
29697
29698
29699
29700
29701
29702
29703
29704
29705
29706
29707
29708
29709
29710
29711
29712
29713
29714
29715
29716
29717
29718
29719
29720
29721
29722
29723
29724
29725
29726
29727
29728
29729
29730
29731
29732
29733
29734
29735
29736
29737
29738
29739
29740
29741
29742
29743
29744
29745
29746
29747
29748
29749
29750
29751
29752
29753
29754
29755
29756
29757
29758
29759
29760
29761
29762
29763
29764
29765
29766
29767
29768
29769
29770
29771
29772
29773
29774
29775
29776
29777
29778
29779
29780
29781
29782
29783
29784
29785
29786
29787
29788
29789
29790
29791
29792
29793
29794
29795
29796
29797
29798
29799
29800
29801
29802
29803
29804
29805
29806
29807
29808
29809
29810
29811
29812
29813
29814
29815
29816
29817
29818
29819
29820
29821
29822
29823
29824
29825
29826
29827
29828
29829
29830
29831
29832
29833
29834
29835
29836
29837
29838
29839
29840
29841
29842
29843
29844
29845
29846
29847
29848
29849
29850
29851
29852
29853
29854
29855
29856
29857
29858
29859
29860
29861
29862
29863
29864
29865
29866
29867
29868
29869
29870
29871
29872
29873
29874
29875
29876
29877
29878
29879
29880
29881
29882
29883
29884
29885
29886
29887
29888
29889
29890
29891
29892
29893
29894
29895
29896
29897
29898
29899
29900
29901
29902
29903
29904
29905
29906
29907
29908
29909
29910
29911
29912
29913
29914
29915
29916
29917
29918
29919
29920
29921
29922
29923
29924
29925
29926
29927
29928
29929
29930
29931
29932
29933
29934
29935
29936
29937
29938
29939
29940
29941
29942
29943
29944
29945
29946
29947
29948
29949
29950
29951
29952
29953
29954
29955
29956
29957
29958
29959
29960
29961
29962
29963
29964
29965
29966
29967
29968
29969
29970
29971
29972
29973
29974
29975
29976
29977
29978
29979
29980
29981
29982
29983
29984
29985
29986
29987
29988
29989
29990
29991
29992
29993
29994
29995
29996
29997
29998
29999
30000
30001
30002
30003
30004
30005
30006
30007
30008
30009
30010
30011
30012
30013
30014
30015
30016
30017
30018
30019
30020
30021
30022
30023
30024
30025
30026
30027
30028
30029
30030
30031
30032
30033
30034
30035
30036
30037
30038
30039
30040
30041
30042
30043
30044
30045
30046
30047
30048
30049
30050
30051
30052
30053
30054
30055
30056
30057
30058
30059
30060
30061
30062
30063
30064
30065
30066
30067
30068
30069
30070
30071
30072
30073
30074
30075
30076
30077
30078
30079
30080
30081
30082
30083
30084
30085
30086
30087
30088
30089
30090
30091
30092
30093
30094
30095
30096
30097
30098
30099
30100
30101
30102
30103
30104
30105
30106
30107
30108
30109
30110
30111
30112
30113
30114
30115
30116
30117
30118
30119
30120
30121
30122
30123
30124
30125
30126
30127
30128
30129
30130
30131
30132
30133
30134
30135
30136
30137
30138
30139
30140
30141
30142
30143
30144
30145
30146
30147
30148
30149
30150
30151
30152
30153
30154
30155
30156
30157
30158
30159
30160
30161
30162
30163
30164
30165
30166
30167
30168
30169
30170
30171
30172
30173
30174
30175
30176
30177
30178
30179
30180
30181
30182
30183
30184
30185
30186
30187
30188
30189
30190
30191
30192
30193
30194
30195
30196
30197
30198
30199
30200
30201
30202
30203
30204
30205
30206
30207
30208
30209
30210
30211
30212
30213
30214
30215
30216
30217
30218
30219
30220
30221
30222
30223
30224
30225
30226
30227
30228
30229
30230
30231
30232
30233
30234
30235
30236
30237
30238
30239
30240
30241
30242
30243
30244
30245
30246
30247
30248
30249
30250
30251
30252
30253
30254
30255
30256
30257
30258
30259
30260
30261
30262
30263
30264
30265
30266
30267
30268
30269
30270
30271
30272
30273
30274
30275
30276
30277
30278
30279
30280
30281
30282
30283
30284
30285
30286
30287
30288
30289
30290
30291
30292
30293
30294
30295
30296
30297
30298
30299
30300
30301
30302
30303
30304
30305
30306
30307
30308
30309
30310
30311
30312
30313
30314
30315
30316
30317
30318
30319
30320
30321
30322
30323
30324
30325
30326
30327
30328
30329
30330
30331
30332
30333
30334
30335
30336
30337
30338
30339
30340
30341
30342
30343
30344
30345
30346
30347
30348
30349
30350
30351
30352
30353
30354
30355
30356
30357
30358
30359
30360
30361
30362
30363
30364
30365
30366
30367
30368
30369
30370
30371
30372
30373
30374
30375
30376
30377
30378
30379
30380
30381
30382
30383
30384
30385
30386
30387
30388
30389
30390
30391
30392
30393
30394
30395
30396
30397
30398
30399
30400
30401
30402
30403
30404
30405
30406
30407
30408
30409
30410
30411
30412
30413
30414
30415
30416
30417
30418
30419
30420
30421
30422
30423
30424
30425
30426
30427
30428
30429
30430
30431
30432
30433
30434
30435
30436
30437
30438
30439
30440
30441
30442
30443
30444
30445
30446
30447
30448
30449
30450
30451
30452
30453
30454
30455
30456
30457
30458
30459
30460
30461
30462
30463
30464
30465
30466
30467
30468
30469
30470
30471
30472
30473
30474
30475
30476
30477
30478
30479
30480
30481
30482
30483
30484
30485
30486
30487
30488
30489
30490
30491
30492
30493
30494
30495
30496
30497
30498
30499
30500
30501
30502
30503
30504
30505
30506
30507
30508
30509
30510
30511
30512
30513
30514
30515
30516
30517
30518
30519
30520
30521
30522
30523
30524
30525
30526
30527
30528
30529
30530
30531
30532
30533
30534
30535
30536
30537
30538
30539
30540
30541
30542
30543
30544
30545
30546
30547
30548
30549
30550
30551
30552
30553
30554
30555
30556
30557
30558
30559
30560
30561
30562
30563
30564
30565
30566
30567
30568
30569
30570
30571
30572
30573
30574
30575
30576
30577
30578
30579
30580
30581
30582
30583
30584
30585
30586
30587
30588
30589
30590
30591
30592
30593
30594
30595
30596
30597
30598
30599
30600
30601
30602
30603
30604
30605
30606
30607
30608
30609
30610
30611
30612
30613
30614
30615
30616
30617
30618
30619
30620
30621
30622
30623
30624
30625
30626
30627
30628
30629
30630
30631
30632
30633
30634
30635
30636
30637
30638
30639
30640
30641
30642
30643
30644
30645
30646
30647
30648
30649
30650
30651
30652
30653
30654
30655
30656
30657
30658
30659
30660
30661
30662
30663
30664
30665
30666
30667
30668
30669
30670
30671
30672
30673
30674
30675
30676
30677
30678
30679
30680
30681
30682
30683
30684
30685
30686
30687
30688
30689
30690
30691
30692
30693
30694
30695
30696
30697
30698
30699
30700
30701
30702
30703
30704
30705
30706
30707
30708
30709
30710
30711
30712
30713
30714
30715
30716
30717
30718
30719
30720
30721
30722
30723
30724
30725
30726
30727
30728
30729
30730
30731
30732
30733
30734
30735
30736
30737
30738
30739
30740
30741
30742
30743
30744
30745
30746
30747
30748
30749
30750
30751
30752
30753
30754
30755
30756
30757
30758
30759
30760
30761
30762
30763
30764
30765
30766
30767
30768
30769
30770
30771
30772
30773
30774
30775
30776
30777
30778
30779
30780
30781
30782
30783
30784
30785
30786
30787
30788
30789
30790
30791
30792
30793
30794
30795
30796
30797
30798
30799
30800
30801
30802
30803
30804
30805
30806
30807
30808
30809
30810
30811
30812
30813
30814
30815
30816
30817
30818
30819
30820
30821
30822
30823
30824
30825
30826
30827
30828
30829
30830
30831
30832
30833
30834
30835
30836
30837
30838
30839
30840
30841
30842
30843
30844
30845
30846
30847
30848
30849
30850
30851
30852
30853
30854
30855
30856
30857
30858
30859
30860
30861
30862
30863
30864
30865
30866
30867
30868
30869
30870
30871
30872
30873
30874
30875
30876
30877
30878
30879
30880
30881
30882
30883
30884
30885
30886
30887
30888
30889
30890
30891
30892
30893
30894
30895
30896
30897
30898
30899
30900
30901
30902
30903
30904
30905
30906
30907
30908
30909
30910
30911
30912
30913
30914
30915
30916
30917
30918
30919
30920
30921
30922
30923
30924
30925
30926
30927
30928
30929
30930
30931
30932
30933
30934
30935
30936
30937
30938
30939
30940
30941
30942
30943
30944
30945
30946
30947
30948
30949
30950
30951
30952
30953
30954
30955
30956
30957
30958
30959
30960
30961
30962
30963
30964
30965
30966
30967
30968
30969
30970
30971
30972
30973
30974
30975
30976
30977
30978
30979
30980
30981
30982
30983
30984
30985
30986
30987
30988
30989
30990
30991
30992
30993
30994
30995
30996
30997
30998
30999
31000
31001
31002
31003
31004
31005
31006
31007
31008
31009
31010
31011
31012
31013
31014
31015
31016
31017
31018
31019
31020
31021
31022
31023
31024
31025
31026
31027
31028
31029
31030
31031
31032
31033
31034
31035
31036
31037
31038
31039
31040
31041
31042
31043
31044
31045
31046
31047
31048
31049
31050
31051
31052
31053
31054
31055
31056
31057
31058
31059
31060
31061
31062
31063
31064
31065
31066
31067
31068
31069
31070
31071
31072
31073
31074
31075
31076
31077
31078
31079
31080
31081
31082
31083
31084
31085
31086
31087
31088
31089
31090
31091
31092
31093
31094
31095
31096
31097
31098
31099
31100
31101
31102
31103
31104
31105
31106
31107
31108
31109
31110
31111
31112
31113
31114
31115
31116
31117
31118
31119
31120
31121
31122
31123
31124
31125
31126
31127
31128
31129
31130
31131
31132
31133
31134
31135
31136
31137
31138
31139
31140
31141
31142
31143
31144
31145
31146
31147
31148
31149
31150
31151
31152
31153
31154
31155
31156
31157
31158
31159
31160
31161
31162
31163
31164
31165
31166
31167
31168
31169
31170
31171
31172
31173
31174
31175
31176
31177
31178
31179
31180
31181
31182
31183
31184
31185
31186
31187
31188
31189
31190
31191
31192
31193
31194
31195
31196
31197
31198
31199
31200
31201
31202
31203
31204
31205
31206
31207
31208
31209
31210
31211
31212
31213
31214
31215
31216
31217
31218
31219
31220
31221
31222
31223
31224
31225
31226
31227
31228
31229
31230
31231
31232
31233
31234
31235
31236
31237
31238
31239
31240
31241
31242
31243
31244
31245
31246
31247
31248
31249
31250
31251
31252
31253
31254
31255
31256
31257
31258
31259
31260
31261
31262
31263
31264
31265
31266
31267
31268
31269
31270
31271
31272
31273
31274
31275
31276
31277
31278
31279
31280
31281
31282
31283
31284
31285
31286
31287
31288
31289
31290
31291
31292
31293
31294
31295
31296
31297
31298
31299
31300
31301
31302
31303
31304
31305
31306
31307
31308
31309
31310
31311
31312
31313
31314
31315
31316
31317
31318
31319
31320
31321
31322
31323
31324
31325
31326
31327
31328
31329
31330
31331
31332
31333
31334
31335
31336
31337
31338
31339
31340
31341
31342
31343
31344
31345
31346
31347
31348
31349
31350
31351
31352
31353
31354
31355
31356
31357
31358
31359
31360
31361
31362
31363
31364
31365
31366
31367
31368
31369
31370
31371
31372
31373
31374
31375
31376
31377
31378
31379
31380
31381
31382
31383
31384
31385
31386
31387
31388
31389
31390
31391
31392
31393
31394
31395
31396
31397
31398
31399
31400
31401
31402
31403
31404
31405
31406
31407
31408
31409
31410
31411
31412
31413
31414
31415
31416
31417
31418
31419
31420
31421
31422
31423
31424
31425
31426
31427
31428
31429
31430
31431
31432
31433
31434
31435
31436
31437
31438
31439
31440
31441
31442
31443
31444
31445
31446
31447
31448
31449
31450
31451
31452
31453
31454
31455
31456
31457
31458
31459
31460
31461
31462
31463
31464
31465
31466
31467
31468
31469
31470
31471
31472
31473
31474
31475
31476
31477
31478
31479
31480
31481
31482
31483
31484
31485
31486
31487
31488
31489
31490
31491
31492
31493
31494
31495
31496
31497
31498
31499
31500
31501
31502
31503
31504
31505
31506
31507
31508
31509
31510
31511
31512
31513
31514
31515
31516
31517
31518
31519
31520
31521
31522
31523
31524
31525
31526
31527
31528
31529
31530
31531
31532
31533
31534
31535
31536
31537
31538
31539
31540
31541
31542
31543
31544
31545
31546
31547
31548
31549
31550
31551
31552
31553
31554
31555
31556
31557
31558
31559
31560
31561
31562
31563
31564
31565
31566
31567
31568
31569
31570
31571
31572
31573
31574
31575
31576
31577
31578
31579
31580
31581
31582
31583
31584
31585
31586
31587
31588
31589
31590
31591
31592
31593
31594
31595
31596
31597
31598
31599
31600
31601
31602
31603
31604
31605
31606
31607
31608
31609
31610
31611
31612
31613
31614
31615
31616
31617
31618
31619
31620
31621
31622
31623
31624
31625
31626
31627
31628
31629
31630
31631
31632
31633
31634
31635
31636
31637
31638
31639
31640
31641
31642
31643
31644
31645
31646
31647
31648
31649
31650
31651
31652
31653
31654
31655
31656
31657
31658
31659
31660
31661
31662
31663
31664
31665
31666
31667
31668
31669
31670
31671
31672
31673
31674
31675
31676
31677
31678
31679
31680
31681
31682
31683
31684
31685
31686
31687
31688
31689
31690
31691
31692
31693
31694
31695
31696
31697
31698
31699
31700
31701
31702
31703
31704
31705
31706
31707
31708
31709
31710
31711
31712
31713
31714
31715
31716
31717
31718
31719
31720
31721
31722
31723
31724
31725
31726
31727
31728
31729
31730
31731
31732
31733
31734
31735
31736
31737
31738
31739
31740
31741
31742
31743
31744
31745
31746
31747
31748
31749
31750
31751
31752
31753
31754
31755
31756
31757
31758
31759
31760
31761
31762
31763
31764
31765
31766
31767
31768
31769
31770
31771
31772
31773
31774
31775
31776
31777
31778
31779
31780
31781
31782
31783
31784
31785
31786
31787
31788
31789
31790
31791
31792
31793
31794
31795
31796
31797
31798
31799
31800
31801
31802
31803
31804
31805
31806
31807
31808
31809
31810
31811
31812
31813
31814
31815
31816
31817
31818
31819
31820
31821
31822
31823
31824
31825
31826
31827
31828
31829
31830
31831
31832
31833
31834
31835
31836
31837
31838
31839
31840
31841
31842
31843
31844
31845
31846
31847
31848
31849
31850
31851
31852
31853
31854
31855
31856
31857
31858
31859
31860
31861
31862
31863
31864
31865
31866
31867
31868
31869
31870
31871
31872
31873
31874
31875
31876
31877
31878
31879
31880
31881
31882
31883
31884
31885
31886
31887
31888
31889
31890
31891
31892
31893
31894
31895
31896
31897
31898
31899
31900
31901
31902
31903
31904
31905
31906
31907
31908
31909
31910
31911
31912
31913
31914
31915
31916
31917
31918
31919
31920
31921
31922
31923
31924
31925
31926
31927
31928
31929
31930
31931
31932
31933
31934
31935
31936
31937
31938
31939
31940
31941
31942
31943
31944
31945
31946
31947
31948
31949
31950
31951
31952
31953
31954
31955
31956
31957
31958
31959
31960
31961
31962
31963
31964
31965
31966
31967
31968
31969
31970
31971
31972
31973
31974
31975
31976
31977
31978
31979
31980
31981
31982
31983
31984
31985
31986
31987
31988
31989
31990
31991
31992
31993
31994
31995
31996
31997
31998
31999
32000
32001
32002
32003
32004
32005
32006
32007
32008
32009
32010
32011
32012
32013
32014
32015
32016
32017
32018
32019
32020
32021
32022
32023
32024
32025
32026
32027
32028
32029
32030
32031
32032
32033
32034
32035
32036
32037
32038
32039
32040
32041
32042
32043
32044
32045
32046
32047
32048
32049
32050
32051
32052
32053
32054
32055
32056
32057
32058
32059
32060
32061
32062
32063
32064
32065
32066
32067
32068
32069
32070
32071
32072
32073
32074
32075
32076
32077
32078
32079
32080
32081
32082
32083
32084
32085
32086
32087
32088
32089
32090
32091
32092
32093
32094
32095
32096
32097
32098
32099
32100
32101
32102
32103
32104
32105
32106
32107
32108
32109
32110
32111
32112
32113
32114
32115
32116
32117
32118
32119
32120
32121
32122
32123
32124
32125
32126
32127
32128
32129
32130
32131
32132
32133
32134
32135
32136
32137
32138
32139
32140
32141
32142
32143
32144
32145
32146
32147
32148
32149
32150
32151
32152
32153
32154
32155
32156
32157
32158
32159
32160
32161
32162
32163
32164
32165
32166
32167
32168
32169
32170
32171
32172
32173
32174
32175
32176
32177
32178
32179
32180
32181
32182
32183
32184
32185
32186
32187
32188
32189
32190
32191
32192
32193
32194
32195
32196
32197
32198
32199
32200
32201
32202
32203
32204
32205
32206
32207
32208
32209
32210
32211
32212
32213
32214
32215
32216
32217
32218
32219
32220
32221
32222
32223
32224
32225
32226
32227
32228
32229
32230
32231
32232
32233
32234
32235
32236
32237
32238
32239
32240
32241
32242
32243
32244
32245
32246
32247
32248
32249
32250
32251
32252
32253
32254
32255
32256
32257
32258
32259
32260
32261
32262
32263
32264
32265
32266
32267
32268
32269
32270
32271
32272
32273
32274
32275
32276
32277
32278
32279
32280
32281
32282
32283
32284
32285
32286
32287
32288
32289
32290
32291
32292
32293
32294
32295
32296
32297
32298
32299
32300
32301
32302
32303
32304
32305
32306
32307
32308
32309
32310
32311
32312
32313
32314
32315
32316
32317
32318
32319
32320
32321
32322
32323
32324
32325
32326
32327
32328
32329
32330
32331
32332
32333
32334
32335
32336
32337
32338
32339
32340
32341
32342
32343
32344
32345
32346
32347
32348
32349
32350
32351
32352
32353
32354
32355
32356
32357
32358
32359
32360
32361
32362
32363
32364
32365
32366
32367
32368
32369
32370
32371
32372
32373
32374
32375
32376
32377
32378
32379
32380
32381
32382
32383
32384
32385
32386
32387
32388
32389
32390
32391
32392
32393
32394
32395
32396
32397
32398
32399
32400
32401
32402
32403
32404
32405
32406
32407
32408
32409
32410
32411
32412
32413
32414
32415
32416
32417
32418
32419
32420
32421
32422
32423
32424
32425
32426
32427
32428
32429
32430
32431
32432
32433
32434
32435
32436
32437
32438
32439
32440
32441
32442
32443
32444
32445
32446
32447
32448
32449
32450
32451
32452
32453
32454
32455
32456
32457
32458
32459
32460
32461
32462
32463
32464
32465
32466
32467
32468
32469
32470
32471
32472
32473
32474
32475
32476
32477
32478
32479
32480
32481
32482
32483
32484
32485
32486
32487
32488
32489
32490
32491
32492
32493
32494
32495
32496
32497
32498
32499
32500
32501
32502
32503
32504
32505
32506
32507
32508
32509
32510
32511
32512
32513
32514
32515
32516
32517
32518
32519
32520
32521
32522
32523
32524
32525
32526
32527
32528
32529
32530
32531
32532
32533
32534
32535
32536
32537
32538
32539
32540
32541
32542
32543
32544
32545
32546
32547
32548
32549
32550
32551
32552
32553
32554
32555
32556
32557
32558
32559
32560
32561
32562
32563
32564
32565
32566
32567
32568
32569
32570
32571
32572
32573
32574
32575
32576
32577
32578
32579
32580
32581
32582
32583
32584
32585
32586
32587
32588
32589
32590
32591
32592
32593
32594
32595
32596
32597
32598
32599
32600
32601
32602
32603
32604
32605
32606
32607
32608
32609
32610
32611
32612
32613
32614
32615
32616
32617
32618
32619
32620
32621
32622
32623
32624
32625
32626
32627
32628
32629
32630
32631
32632
32633
32634
32635
32636
32637
32638
32639
32640
32641
32642
32643
32644
32645
32646
32647
32648
32649
32650
32651
32652
32653
32654
32655
32656
32657
32658
32659
32660
32661
32662
32663
32664
32665
32666
32667
32668
32669
32670
32671
32672
32673
32674
32675
32676
32677
32678
32679
32680
32681
32682
32683
32684
32685
32686
32687
32688
32689
32690
32691
32692
32693
32694
32695
32696
32697
32698
32699
32700
32701
32702
32703
32704
32705
32706
32707
32708
32709
32710
32711
32712
32713
32714
32715
32716
32717
32718
32719
32720
32721
32722
32723
32724
32725
32726
32727
32728
32729
32730
32731
32732
32733
32734
32735
32736
32737
32738
32739
32740
32741
32742
32743
32744
32745
32746
32747
32748
32749
32750
32751
32752
32753
32754
32755
32756
32757
32758
32759
32760
32761
32762
32763
32764
32765
32766
32767
32768
32769
32770
32771
32772
32773
32774
32775
32776
32777
32778
32779
32780
32781
32782
32783
32784
32785
32786
32787
32788
32789
32790
32791
32792
32793
32794
32795
32796
32797
32798
32799
32800
32801
32802
32803
32804
32805
32806
32807
32808
32809
32810
32811
32812
32813
32814
32815
32816
32817
32818
32819
32820
32821
32822
32823
32824
32825
32826
32827
32828
32829
32830
32831
32832
32833
32834
32835
32836
32837
32838
32839
32840
32841
32842
32843
32844
32845
32846
32847
32848
32849
32850
32851
32852
32853
32854
32855
32856
32857
32858
32859
32860
32861
32862
32863
32864
32865
32866
32867
32868
32869
32870
32871
32872
32873
32874
32875
32876
32877
32878
32879
32880
32881
32882
32883
32884
32885
32886
32887
32888
32889
32890
32891
32892
32893
32894
32895
32896
32897
32898
32899
32900
32901
32902
32903
32904
32905
32906
32907
32908
32909
32910
32911
32912
32913
32914
32915
32916
32917
32918
32919
32920
32921
32922
32923
32924
32925
32926
32927
32928
32929
32930
32931
32932
32933
32934
32935
32936
32937
32938
32939
32940
32941
32942
32943
32944
32945
32946
32947
32948
32949
32950
32951
32952
32953
32954
32955
32956
32957
32958
32959
32960
32961
32962
32963
32964
32965
32966
32967
32968
32969
32970
32971
32972
32973
32974
32975
32976
32977
32978
32979
32980
32981
32982
32983
32984
32985
32986
32987
32988
32989
32990
32991
32992
32993
32994
32995
32996
32997
32998
32999
33000
33001
33002
33003
33004
33005
33006
33007
33008
33009
33010
33011
33012
33013
33014
33015
33016
33017
33018
33019
33020
33021
33022
33023
33024
33025
33026
33027
33028
33029
33030
33031
33032
33033
33034
33035
33036
33037
33038
33039
33040
33041
33042
33043
33044
33045
33046
33047
33048
33049
33050
33051
33052
33053
33054
33055
33056
33057
33058
33059
33060
33061
33062
33063
33064
33065
33066
33067
33068
33069
33070
33071
33072
33073
33074
33075
33076
33077
33078
33079
33080
33081
33082
33083
33084
33085
33086
33087
33088
33089
33090
33091
33092
33093
33094
33095
33096
33097
33098
33099
33100
33101
33102
33103
33104
33105
33106
33107
33108
33109
33110
33111
33112
33113
33114
33115
33116
33117
33118
33119
33120
33121
33122
33123
33124
33125
33126
33127
33128
33129
33130
33131
33132
33133
33134
33135
33136
33137
33138
33139
33140
33141
33142
33143
33144
33145
33146
33147
33148
33149
33150
33151
33152
33153
33154
33155
33156
33157
33158
33159
33160
33161
33162
33163
33164
33165
33166
33167
33168
33169
33170
33171
33172
33173
33174
33175
33176
33177
33178
33179
33180
33181
33182
33183
33184
33185
33186
33187
33188
33189
33190
33191
33192
33193
33194
33195
33196
33197
33198
33199
33200
33201
33202
33203
33204
33205
33206
33207
33208
33209
33210
33211
33212
33213
33214
33215
33216
33217
33218
33219
33220
33221
33222
33223
33224
33225
33226
33227
33228
33229
33230
33231
33232
33233
33234
33235
33236
33237
33238
33239
33240
33241
33242
33243
33244
33245
33246
33247
33248
33249
33250
33251
33252
33253
33254
33255
33256
33257
33258
33259
33260
33261
33262
33263
33264
33265
33266
33267
33268
33269
33270
33271
33272
33273
33274
33275
33276
33277
33278
33279
33280
33281
33282
33283
33284
33285
33286
33287
33288
33289
33290
33291
33292
33293
33294
33295
33296
33297
33298
33299
33300
33301
33302
33303
33304
33305
33306
33307
33308
33309
33310
33311
33312
33313
33314
33315
33316
33317
33318
33319
33320
33321
33322
33323
33324
33325
33326
33327
33328
33329
33330
33331
33332
33333
33334
33335
33336
33337
33338
33339
33340
33341
33342
33343
33344
33345
33346
33347
33348
33349
33350
33351
33352
33353
33354
33355
33356
33357
33358
33359
33360
33361
33362
33363
33364
33365
33366
33367
33368
33369
33370
33371
33372
33373
33374
33375
33376
33377
33378
33379
33380
33381
33382
33383
33384
33385
33386
33387
33388
33389
33390
33391
33392
33393
33394
33395
33396
33397
33398
33399
33400
33401
33402
33403
33404
33405
33406
33407
33408
33409
33410
33411
33412
33413
33414
33415
33416
33417
33418
33419
33420
33421
33422
33423
33424
33425
33426
33427
33428
33429
33430
33431
33432
33433
33434
33435
33436
33437
33438
33439
33440
33441
33442
33443
33444
33445
33446
33447
33448
33449
33450
33451
33452
33453
33454
33455
33456
33457
33458
33459
33460
33461
33462
33463
33464
33465
33466
33467
33468
33469
33470
33471
33472
33473
33474
33475
33476
33477
33478
33479
33480
33481
33482
33483
33484
33485
33486
33487
33488
33489
33490
33491
33492
33493
33494
33495
33496
33497
33498
33499
33500
33501
33502
33503
33504
33505
33506
33507
33508
33509
33510
33511
33512
33513
33514
33515
33516
33517
33518
33519
33520
33521
33522
33523
33524
33525
33526
33527
33528
33529
33530
33531
33532
33533
33534
33535
33536
33537
33538
33539
33540
33541
33542
33543
33544
33545
33546
33547
33548
33549
33550
33551
33552
33553
33554
33555
33556
33557
33558
33559
33560
33561
33562
33563
33564
33565
33566
33567
33568
33569
33570
33571
33572
33573
33574
33575
33576
33577
33578
33579
33580
33581
33582
33583
33584
33585
33586
33587
33588
33589
33590
33591
33592
33593
33594
33595
33596
33597
33598
33599
33600
33601
33602
33603
33604
33605
33606
33607
33608
33609
33610
33611
33612
33613
33614
33615
33616
33617
33618
33619
33620
33621
33622
33623
33624
33625
33626
33627
33628
33629
33630
33631
33632
33633
33634
33635
33636
33637
33638
33639
33640
33641
33642
33643
33644
33645
33646
33647
33648
33649
33650
33651
33652
33653
33654
33655
33656
33657
33658
33659
33660
33661
33662
33663
33664
33665
33666
33667
33668
33669
33670
33671
33672
33673
33674
33675
33676
33677
33678
33679
33680
33681
33682
33683
33684
33685
33686
33687
33688
33689
33690
33691
33692
33693
33694
33695
33696
33697
33698
33699
33700
33701
33702
33703
33704
33705
33706
33707
33708
33709
33710
33711
33712
33713
33714
33715
33716
33717
33718
33719
33720
33721
33722
33723
33724
33725
33726
33727
33728
33729
33730
33731
33732
33733
33734
33735
33736
33737
33738
33739
33740
33741
33742
33743
33744
33745
33746
33747
33748
33749
33750
33751
33752
33753
33754
33755
33756
33757
33758
33759
33760
33761
33762
33763
33764
33765
33766
33767
33768
33769
33770
33771
33772
33773
33774
33775
33776
33777
33778
33779
33780
33781
33782
33783
33784
33785
33786
33787
33788
33789
33790
33791
33792
33793
33794
33795
33796
33797
33798
33799
33800
33801
33802
33803
33804
33805
33806
33807
33808
33809
33810
33811
33812
33813
33814
33815
33816
33817
33818
33819
33820
33821
33822
33823
33824
33825
33826
33827
33828
33829
33830
33831
33832
33833
33834
33835
33836
33837
33838
33839
33840
33841
33842
33843
33844
33845
33846
33847
33848
33849
33850
33851
33852
33853
33854
33855
33856
33857
33858
33859
33860
33861
33862
33863
33864
33865
33866
33867
33868
33869
33870
33871
33872
33873
33874
33875
33876
33877
33878
33879
33880
33881
33882
33883
33884
33885
33886
33887
33888
33889
33890
33891
33892
33893
33894
33895
33896
33897
33898
33899
33900
33901
33902
33903
33904
33905
33906
33907
33908
33909
33910
33911
33912
33913
33914
33915
33916
33917
33918
33919
33920
33921
33922
33923
33924
33925
33926
33927
33928
33929
33930
33931
33932
33933
33934
33935
33936
33937
33938
33939
33940
33941
33942
33943
33944
33945
33946
33947
33948
33949
33950
33951
33952
33953
33954
33955
33956
33957
33958
33959
33960
33961
33962
33963
33964
33965
33966
33967
33968
33969
33970
33971
33972
33973
33974
33975
33976
33977
33978
33979
33980
33981
33982
33983
33984
33985
33986
33987
33988
33989
33990
33991
33992
33993
33994
33995
33996
33997
33998
33999
34000
34001
34002
34003
34004
34005
34006
34007
34008
34009
34010
34011
34012
34013
34014
34015
34016
34017
34018
34019
34020
34021
34022
34023
34024
34025
34026
34027
34028
34029
34030
34031
34032
34033
34034
34035
34036
34037
34038
34039
34040
34041
34042
34043
34044
34045
34046
34047
34048
34049
34050
34051
34052
34053
34054
34055
34056
34057
34058
34059
34060
34061
34062
34063
34064
34065
34066
34067
34068
34069
34070
34071
34072
34073
34074
34075
34076
34077
34078
34079
34080
34081
34082
34083
34084
34085
34086
34087
34088
34089
34090
34091
34092
34093
34094
34095
34096
34097
34098
34099
34100
34101
34102
34103
34104
34105
34106
34107
34108
34109
34110
34111
34112
34113
34114
34115
34116
34117
34118
34119
34120
34121
34122
34123
34124
34125
34126
34127
34128
34129
34130
34131
34132
34133
34134
34135
34136
34137
34138
34139
34140
34141
34142
34143
34144
34145
34146
34147
34148
34149
34150
34151
34152
34153
34154
34155
34156
34157
34158
34159
34160
34161
34162
34163
34164
34165
34166
34167
34168
34169
34170
34171
34172
34173
34174
34175
34176
34177
34178
34179
34180
34181
34182
34183
34184
34185
34186
34187
34188
34189
34190
34191
34192
34193
34194
34195
34196
34197
34198
34199
34200
34201
34202
34203
34204
34205
34206
34207
34208
34209
34210
34211
34212
34213
34214
34215
34216
34217
34218
34219
34220
34221
34222
34223
34224
34225
34226
34227
34228
34229
34230
34231
34232
34233
34234
34235
34236
34237
34238
34239
34240
34241
34242
34243
34244
34245
34246
34247
34248
34249
34250
34251
34252
34253
34254
34255
34256
34257
34258
34259
34260
34261
34262
34263
34264
34265
34266
34267
34268
34269
34270
34271
34272
34273
34274
34275
34276
34277
34278
34279
34280
34281
34282
34283
34284
34285
34286
34287
34288
34289
34290
34291
34292
34293
34294
34295
34296
34297
34298
34299
34300
34301
34302
34303
34304
34305
34306
34307
34308
34309
34310
34311
34312
34313
34314
34315
34316
34317
34318
34319
34320
34321
34322
34323
34324
34325
34326
34327
34328
34329
34330
34331
34332
34333
34334
34335
34336
34337
34338
34339
34340
34341
34342
34343
34344
34345
34346
34347
34348
34349
34350
34351
34352
34353
34354
34355
34356
34357
34358
34359
34360
34361
34362
34363
34364
34365
34366
34367
34368
34369
34370
34371
34372
34373
34374
34375
34376
34377
34378
34379
34380
34381
34382
34383
34384
34385
34386
34387
34388
34389
34390
34391
34392
34393
34394
34395
34396
34397
34398
34399
34400
34401
34402
34403
34404
34405
34406
34407
34408
34409
34410
34411
34412
34413
34414
34415
34416
34417
34418
34419
34420
34421
34422
34423
34424
34425
34426
34427
34428
34429
34430
34431
34432
34433
34434
34435
34436
34437
34438
34439
34440
34441
34442
34443
34444
34445
34446
34447
34448
34449
34450
34451
34452
34453
34454
34455
34456
34457
34458
34459
34460
34461
34462
34463
34464
34465
34466
34467
34468
34469
34470
34471
34472
34473
34474
34475
34476
34477
34478
34479
34480
34481
34482
34483
34484
34485
34486
34487
34488
34489
34490
34491
34492
34493
34494
34495
34496
34497
34498
34499
34500
34501
34502
34503
34504
34505
34506
34507
34508
34509
34510
34511
34512
34513
34514
34515
34516
34517
34518
34519
34520
34521
34522
34523
34524
34525
34526
34527
34528
34529
34530
34531
34532
34533
34534
34535
34536
34537
34538
34539
34540
34541
34542
34543
34544
34545
34546
34547
34548
34549
34550
34551
34552
34553
34554
34555
34556
34557
34558
34559
34560
34561
34562
34563
34564
34565
34566
34567
34568
34569
34570
34571
34572
34573
34574
34575
34576
34577
34578
34579
34580
34581
34582
34583
34584
34585
34586
34587
34588
34589
34590
34591
34592
34593
34594
34595
34596
34597
34598
34599
34600
34601
34602
34603
34604
34605
34606
34607
34608
34609
34610
34611
34612
34613
34614
34615
34616
34617
34618
34619
34620
34621
34622
34623
34624
34625
34626
34627
34628
34629
34630
34631
34632
34633
34634
34635
34636
34637
34638
34639
34640
34641
34642
34643
34644
34645
34646
34647
34648
34649
34650
34651
34652
34653
34654
34655
34656
34657
34658
34659
34660
34661
34662
34663
34664
34665
34666
34667
34668
34669
34670
34671
34672
34673
34674
34675
34676
34677
34678
34679
34680
34681
34682
34683
34684
34685
34686
34687
34688
34689
34690
34691
34692
34693
34694
34695
34696
34697
34698
34699
34700
34701
34702
34703
34704
34705
34706
34707
34708
34709
34710
34711
34712
34713
34714
34715
34716
34717
34718
34719
34720
34721
34722
34723
34724
34725
34726
34727
34728
34729
34730
34731
34732
34733
34734
34735
34736
34737
34738
34739
34740
34741
34742
34743
34744
34745
34746
34747
34748
34749
34750
34751
34752
34753
34754
34755
34756
34757
34758
34759
34760
34761
34762
34763
34764
34765
34766
34767
34768
34769
34770
34771
34772
34773
34774
34775
34776
34777
34778
34779
34780
34781
34782
34783
34784
34785
34786
34787
34788
34789
34790
34791
34792
34793
34794
34795
34796
34797
34798
34799
34800
34801
34802
34803
34804
34805
34806
34807
34808
34809
34810
34811
34812
34813
34814
34815
34816
34817
34818
34819
34820
34821
34822
34823
34824
34825
34826
34827
34828
34829
34830
34831
34832
34833
34834
34835
34836
34837
34838
34839
34840
34841
34842
34843
34844
34845
34846
34847
34848
34849
34850
34851
34852
34853
34854
34855
34856
34857
34858
34859
34860
34861
34862
34863
34864
34865
34866
34867
34868
34869
34870
34871
34872
34873
34874
34875
34876
34877
34878
34879
34880
34881
34882
34883
34884
34885
34886
34887
34888
34889
34890
34891
34892
34893
34894
34895
34896
34897
34898
34899
34900
34901
34902
34903
34904
34905
34906
34907
34908
34909
34910
34911
34912
34913
34914
34915
34916
34917
34918
34919
34920
34921
34922
34923
34924
34925
34926
34927
34928
34929
34930
34931
34932
34933
34934
34935
34936
34937
34938
34939
34940
34941
34942
34943
34944
34945
34946
34947
34948
34949
34950
34951
34952
34953
34954
34955
34956
34957
34958
34959
34960
34961
34962
34963
34964
34965
34966
34967
34968
34969
34970
34971
34972
34973
34974
34975
34976
34977
34978
34979
34980
34981
34982
34983
34984
34985
34986
34987
34988
34989
34990
34991
34992
34993
34994
34995
34996
34997
34998
34999
35000
35001
35002
35003
35004
35005
35006
35007
35008
35009
35010
35011
35012
35013
35014
35015
35016
35017
35018
35019
35020
35021
35022
35023
35024
35025
35026
35027
35028
35029
35030
35031
35032
35033
35034
35035
35036
35037
35038
35039
35040
35041
35042
35043
35044
35045
35046
35047
35048
35049
35050
35051
35052
35053
35054
35055
35056
35057
35058
35059
35060
35061
35062
35063
35064
35065
35066
35067
35068
35069
35070
35071
35072
35073
35074
35075
35076
35077
35078
35079
35080
35081
35082
35083
35084
35085
35086
35087
35088
35089
35090
35091
35092
35093
35094
35095
35096
35097
35098
35099
35100
35101
35102
35103
35104
35105
35106
35107
35108
35109
35110
35111
35112
35113
35114
35115
35116
35117
35118
35119
35120
35121
35122
35123
35124
35125
35126
35127
35128
35129
35130
35131
35132
35133
35134
35135
35136
35137
35138
35139
35140
35141
35142
35143
35144
35145
35146
35147
35148
35149
35150
35151
35152
35153
35154
35155
35156
35157
35158
35159
35160
35161
35162
35163
35164
35165
35166
35167
35168
35169
35170
35171
35172
35173
35174
35175
35176
35177
35178
35179
35180
35181
35182
35183
35184
35185
35186
35187
35188
35189
35190
35191
35192
35193
35194
35195
35196
35197
35198
35199
35200
35201
35202
35203
35204
35205
35206
35207
35208
35209
35210
35211
35212
35213
35214
35215
35216
35217
35218
35219
35220
35221
35222
35223
35224
35225
35226
35227
35228
35229
35230
35231
35232
35233
35234
35235
35236
35237
35238
35239
35240
35241
35242
35243
35244
35245
35246
35247
35248
35249
35250
35251
35252
35253
35254
35255
35256
35257
35258
35259
35260
35261
35262
35263
35264
35265
35266
35267
35268
35269
35270
35271
35272
35273
35274
35275
35276
35277
35278
35279
35280
35281
35282
35283
35284
35285
35286
35287
35288
35289
35290
35291
35292
35293
35294
35295
35296
35297
35298
35299
35300
35301
35302
35303
35304
35305
35306
35307
35308
35309
35310
35311
35312
35313
35314
35315
35316
35317
35318
35319
35320
35321
35322
35323
35324
35325
35326
35327
35328
35329
35330
35331
35332
35333
35334
35335
35336
35337
35338
35339
35340
35341
35342
35343
35344
35345
35346
35347
35348
35349
35350
35351
35352
35353
35354
35355
35356
35357
35358
35359
35360
35361
35362
35363
35364
35365
35366
35367
35368
35369
35370
35371
35372
35373
35374
35375
35376
35377
35378
35379
35380
35381
35382
35383
35384
35385
35386
35387
35388
35389
35390
35391
35392
35393
35394
35395
35396
35397
35398
35399
35400
35401
35402
35403
35404
35405
35406
35407
35408
35409
35410
35411
35412
35413
35414
35415
35416
35417
35418
35419
35420
35421
35422
35423
35424
35425
35426
35427
35428
35429
35430
35431
35432
35433
35434
35435
35436
35437
35438
35439
35440
35441
35442
35443
35444
35445
35446
35447
35448
35449
35450
35451
35452
35453
35454
35455
35456
35457
35458
35459
35460
35461
35462
35463
35464
35465
35466
35467
35468
35469
35470
35471
35472
35473
35474
35475
35476
35477
35478
35479
35480
35481
35482
35483
35484
35485
35486
35487
35488
35489
35490
35491
35492
35493
35494
35495
35496
35497
35498
35499
35500
35501
35502
35503
35504
35505
35506
35507
35508
35509
35510
35511
35512
35513
35514
35515
35516
35517
35518
35519
35520
35521
35522
35523
35524
35525
35526
35527
35528
35529
35530
35531
35532
35533
35534
35535
35536
35537
35538
35539
35540
35541
35542
35543
35544
35545
35546
35547
35548
35549
35550
35551
35552
35553
35554
35555
35556
35557
35558
35559
35560
35561
35562
35563
35564
35565
35566
35567
35568
35569
35570
35571
35572
35573
35574
35575
35576
35577
35578
35579
35580
35581
35582
35583
35584
35585
35586
35587
35588
35589
35590
35591
35592
35593
35594
35595
35596
35597
35598
35599
35600
35601
35602
35603
35604
35605
35606
35607
35608
35609
35610
35611
35612
35613
35614
35615
35616
35617
35618
35619
35620
35621
35622
35623
35624
35625
35626
35627
35628
35629
35630
35631
35632
35633
35634
35635
35636
35637
35638
35639
35640
35641
35642
35643
35644
35645
35646
35647
35648
35649
35650
35651
35652
35653
35654
35655
35656
35657
35658
35659
35660
35661
35662
35663
35664
35665
35666
35667
35668
35669
35670
35671
35672
35673
35674
35675
35676
35677
35678
35679
35680
35681
35682
35683
35684
35685
35686
35687
35688
35689
35690
35691
35692
35693
35694
35695
35696
35697
35698
35699
35700
35701
35702
35703
35704
35705
35706
35707
35708
35709
35710
35711
35712
35713
35714
35715
35716
35717
35718
35719
35720
35721
35722
35723
35724
35725
35726
35727
35728
35729
35730
35731
35732
35733
35734
35735
35736
35737
35738
35739
35740
35741
35742
35743
35744
35745
35746
35747
35748
35749
35750
35751
35752
35753
35754
35755
35756
35757
35758
35759
35760
35761
35762
35763
35764
35765
35766
35767
35768
35769
35770
35771
35772
35773
35774
35775
35776
35777
35778
35779
35780
35781
35782
35783
35784
35785
35786
35787
35788
35789
35790
35791
35792
35793
35794
35795
35796
35797
35798
35799
35800
35801
35802
35803
35804
35805
35806
35807
35808
35809
35810
35811
35812
35813
35814
35815
35816
35817
35818
35819
35820
35821
35822
35823
35824
35825
35826
35827
35828
35829
35830
35831
35832
35833
35834
35835
35836
35837
35838
35839
35840
35841
35842
35843
35844
35845
35846
35847
35848
35849
35850
35851
35852
35853
35854
35855
35856
35857
35858
35859
35860
35861
35862
35863
35864
35865
35866
35867
35868
35869
35870
35871
35872
35873
35874
35875
35876
35877
35878
35879
35880
35881
35882
35883
35884
35885
35886
35887
35888
35889
35890
35891
35892
35893
35894
35895
35896
35897
35898
35899
35900
35901
35902
35903
35904
35905
35906
35907
35908
35909
35910
35911
35912
35913
35914
35915
35916
35917
35918
35919
35920
35921
35922
35923
35924
35925
35926
35927
35928
35929
35930
35931
35932
35933
35934
35935
35936
35937
35938
35939
35940
35941
35942
35943
35944
35945
35946
35947
35948
35949
35950
35951
35952
35953
35954
35955
35956
35957
35958
35959
35960
35961
35962
35963
35964
35965
35966
35967
35968
35969
35970
35971
35972
35973
35974
35975
35976
35977
35978
35979
35980
35981
35982
35983
35984
35985
35986
35987
35988
35989
35990
35991
35992
35993
35994
35995
35996
35997
35998
35999
36000
36001
36002
36003
36004
36005
36006
36007
36008
36009
36010
36011
36012
36013
36014
36015
36016
36017
36018
36019
36020
36021
36022
36023
36024
36025
36026
36027
36028
36029
36030
36031
36032
36033
36034
36035
36036
36037
36038
36039
36040
36041
36042
36043
36044
36045
36046
36047
36048
36049
36050
36051
36052
36053
36054
36055
36056
36057
36058
36059
36060
36061
36062
36063
36064
36065
36066
36067
36068
36069
36070
36071
36072
36073
36074
36075
36076
36077
36078
36079
36080
36081
36082
36083
36084
36085
36086
36087
36088
36089
36090
36091
36092
36093
36094
36095
36096
36097
36098
36099
36100
36101
36102
36103
36104
36105
36106
36107
36108
36109
36110
36111
36112
36113
36114
36115
36116
36117
36118
36119
36120
36121
36122
36123
36124
36125
36126
36127
36128
36129
36130
36131
36132
36133
36134
36135
36136
36137
36138
36139
36140
36141
36142
36143
36144
36145
36146
36147
36148
36149
36150
36151
36152
36153
36154
36155
36156
36157
36158
36159
36160
36161
36162
36163
36164
36165
36166
36167
36168
36169
36170
36171
36172
36173
36174
36175
36176
36177
36178
36179
36180
36181
36182
36183
36184
36185
36186
36187
36188
36189
36190
36191
36192
36193
36194
36195
36196
36197
36198
36199
36200
36201
36202
36203
36204
36205
36206
36207
36208
36209
36210
36211
36212
36213
36214
36215
36216
36217
36218
36219
36220
36221
36222
36223
36224
36225
36226
36227
36228
36229
36230
36231
36232
36233
36234
36235
36236
36237
36238
36239
36240
36241
36242
36243
36244
36245
36246
36247
36248
36249
36250
36251
36252
36253
36254
36255
36256
36257
36258
36259
36260
36261
36262
36263
36264
36265
36266
36267
36268
36269
36270
36271
36272
36273
36274
36275
36276
36277
36278
36279
36280
36281
36282
36283
36284
36285
36286
36287
36288
36289
36290
36291
36292
36293
36294
36295
36296
36297
36298
36299
36300
36301
36302
36303
36304
36305
36306
36307
36308
36309
36310
36311
36312
36313
36314
36315
36316
36317
36318
36319
36320
36321
36322
36323
36324
36325
36326
36327
36328
36329
36330
36331
36332
36333
36334
36335
36336
36337
36338
36339
36340
36341
36342
36343
36344
36345
36346
36347
36348
36349
36350
36351
36352
36353
36354
36355
36356
36357
36358
36359
36360
36361
36362
36363
36364
36365
36366
36367
36368
36369
36370
36371
36372
36373
36374
36375
36376
36377
36378
36379
36380
36381
36382
36383
36384
36385
36386
36387
36388
36389
36390
36391
36392
36393
36394
36395
36396
36397
36398
36399
36400
36401
36402
36403
36404
36405
36406
36407
36408
36409
36410
36411
36412
36413
36414
36415
36416
36417
36418
36419
36420
36421
36422
36423
36424
36425
36426
36427
36428
36429
36430
36431
36432
36433
36434
36435
36436
36437
36438
36439
36440
36441
36442
36443
36444
36445
36446
36447
36448
36449
36450
36451
36452
36453
36454
36455
36456
36457
36458
36459
36460
36461
36462
36463
36464
36465
36466
36467
36468
36469
36470
36471
36472
36473
36474
36475
36476
36477
36478
36479
36480
36481
36482
36483
36484
36485
36486
36487
36488
36489
36490
36491
36492
36493
36494
36495
36496
36497
36498
36499
36500
36501
36502
36503
36504
36505
36506
36507
36508
36509
36510
36511
36512
36513
36514
36515
36516
36517
36518
36519
36520
36521
36522
36523
36524
36525
36526
36527
36528
36529
36530
36531
36532
36533
36534
36535
36536
36537
36538
36539
36540
36541
36542
36543
36544
36545
36546
36547
36548
36549
36550
36551
36552
36553
36554
36555
36556
36557
36558
36559
36560
36561
36562
36563
36564
36565
36566
36567
36568
36569
36570
36571
36572
36573
36574
36575
36576
36577
36578
36579
36580
36581
36582
36583
36584
36585
36586
36587
36588
36589
36590
36591
36592
36593
36594
36595
36596
36597
36598
36599
36600
36601
36602
36603
36604
36605
36606
36607
36608
36609
36610
36611
36612
36613
36614
36615
36616
36617
36618
36619
36620
36621
36622
36623
36624
36625
36626
36627
36628
36629
36630
36631
36632
36633
36634
36635
36636
36637
36638
36639
36640
36641
36642
36643
36644
36645
36646
36647
36648
36649
36650
36651
36652
36653
36654
36655
36656
36657
36658
36659
36660
36661
36662
36663
36664
36665
36666
36667
36668
36669
36670
36671
36672
36673
36674
36675
36676
36677
36678
36679
36680
36681
36682
36683
36684
36685
36686
36687
36688
36689
36690
36691
36692
36693
36694
36695
36696
36697
36698
36699
36700
36701
36702
36703
36704
36705
36706
36707
36708
36709
36710
36711
36712
36713
36714
36715
36716
36717
36718
36719
36720
36721
36722
36723
36724
36725
36726
36727
36728
36729
36730
36731
36732
36733
36734
36735
36736
36737
36738
36739
36740
36741
36742
36743
36744
36745
36746
36747
36748
36749
36750
36751
36752
36753
36754
36755
36756
36757
36758
36759
36760
36761
36762
36763
36764
36765
36766
36767
36768
36769
36770
36771
36772
36773
36774
36775
36776
36777
36778
36779
36780
36781
36782
36783
36784
36785
36786
36787
36788
36789
36790
36791
36792
36793
36794
36795
36796
36797
36798
36799
36800
36801
36802
36803
36804
36805
36806
36807
36808
36809
36810
36811
36812
36813
36814
36815
36816
36817
36818
36819
36820
36821
36822
36823
36824
36825
36826
36827
36828
36829
36830
36831
36832
36833
36834
36835
36836
36837
36838
36839
36840
36841
36842
36843
36844
36845
36846
36847
36848
36849
36850
36851
36852
36853
36854
36855
36856
36857
36858
36859
36860
36861
36862
36863
36864
36865
36866
36867
36868
36869
36870
36871
36872
36873
36874
36875
36876
36877
36878
36879
36880
36881
36882
36883
36884
36885
36886
36887
36888
36889
36890
36891
36892
36893
36894
36895
36896
36897
36898
36899
36900
36901
36902
36903
36904
36905
36906
36907
36908
36909
36910
36911
36912
36913
36914
36915
36916
36917
36918
36919
36920
36921
36922
36923
36924
36925
36926
36927
36928
36929
36930
36931
36932
36933
36934
36935
36936
36937
36938
36939
36940
36941
36942
36943
36944
36945
36946
36947
36948
36949
36950
36951
36952
36953
36954
36955
36956
36957
36958
36959
36960
36961
36962
36963
36964
36965
36966
36967
36968
36969
36970
36971
36972
36973
36974
36975
36976
36977
36978
36979
36980
36981
36982
36983
36984
36985
36986
36987
36988
36989
36990
36991
36992
36993
36994
36995
36996
36997
36998
36999
37000
37001
37002
37003
37004
37005
37006
37007
37008
37009
37010
37011
37012
37013
37014
37015
37016
37017
37018
37019
37020
37021
37022
37023
37024
37025
37026
37027
37028
37029
37030
37031
37032
37033
37034
37035
37036
37037
37038
37039
37040
37041
37042
37043
37044
37045
37046
37047
37048
37049
37050
37051
37052
37053
37054
37055
37056
37057
37058
37059
37060
37061
37062
37063
37064
37065
37066
37067
37068
37069
37070
37071
37072
37073
37074
37075
37076
37077
37078
37079
37080
37081
37082
37083
37084
37085
37086
37087
37088
37089
37090
37091
37092
37093
37094
37095
37096
37097
37098
37099
37100
37101
37102
37103
37104
37105
37106
37107
37108
37109
37110
37111
37112
37113
37114
37115
37116
37117
37118
37119
37120
37121
37122
37123
37124
37125
37126
37127
37128
37129
37130
37131
37132
37133
37134
37135
37136
37137
37138
37139
37140
37141
37142
37143
37144
37145
37146
37147
37148
37149
37150
37151
37152
37153
37154
37155
37156
37157
37158
37159
37160
37161
37162
37163
37164
37165
37166
37167
37168
37169
37170
37171
37172
37173
37174
37175
37176
37177
37178
37179
37180
37181
37182
37183
37184
37185
37186
37187
37188
37189
37190
37191
37192
37193
37194
37195
37196
37197
37198
37199
37200
37201
37202
37203
37204
37205
37206
37207
37208
37209
37210
37211
37212
37213
37214
37215
37216
37217
37218
37219
37220
37221
37222
37223
37224
37225
37226
37227
37228
37229
37230
37231
37232
37233
37234
37235
37236
37237
37238
37239
37240
37241
37242
37243
37244
37245
37246
37247
37248
37249
37250
37251
37252
37253
37254
37255
37256
37257
37258
37259
37260
37261
37262
37263
37264
37265
37266
37267
37268
37269
37270
37271
37272
37273
37274
37275
37276
37277
37278
37279
37280
37281
37282
37283
37284
37285
37286
37287
37288
37289
37290
37291
37292
37293
37294
37295
37296
37297
37298
37299
37300
37301
37302
37303
37304
37305
37306
37307
37308
37309
37310
37311
37312
37313
37314
37315
37316
37317
37318
37319
37320
37321
37322
37323
37324
37325
37326
37327
37328
37329
37330
37331
37332
37333
37334
37335
37336
37337
37338
37339
37340
37341
37342
37343
37344
37345
37346
37347
37348
37349
37350
37351
37352
37353
37354
37355
37356
37357
37358
37359
37360
37361
37362
37363
37364
37365
37366
37367
37368
37369
37370
37371
37372
37373
37374
37375
37376
37377
37378
37379
37380
37381
37382
37383
37384
37385
37386
37387
37388
37389
37390
37391
37392
37393
37394
37395
37396
37397
37398
37399
37400
37401
37402
37403
37404
37405
37406
37407
37408
37409
37410
37411
37412
37413
37414
37415
37416
37417
37418
37419
37420
37421
37422
37423
37424
37425
37426
37427
37428
37429
37430
37431
37432
37433
37434
37435
37436
37437
37438
37439
37440
37441
37442
37443
37444
37445
37446
37447
37448
37449
37450
37451
37452
37453
37454
37455
37456
37457
37458
37459
37460
37461
37462
37463
37464
37465
37466
37467
37468
37469
37470
37471
37472
37473
37474
37475
37476
37477
37478
37479
37480
37481
37482
37483
37484
37485
37486
37487
37488
37489
37490
37491
37492
37493
37494
37495
37496
37497
37498
37499
37500
37501
37502
37503
37504
37505
37506
37507
37508
37509
37510
37511
37512
37513
37514
37515
37516
37517
37518
37519
37520
37521
37522
37523
37524
37525
37526
37527
37528
37529
37530
37531
37532
37533
37534
37535
37536
37537
37538
37539
37540
37541
37542
37543
37544
37545
37546
37547
37548
37549
37550
37551
37552
37553
37554
37555
37556
37557
37558
37559
37560
37561
37562
37563
37564
37565
37566
37567
37568
37569
37570
37571
37572
37573
37574
37575
37576
37577
37578
37579
37580
37581
37582
37583
37584
37585
37586
37587
37588
37589
37590
37591
37592
37593
37594
37595
37596
37597
37598
37599
37600
37601
37602
37603
37604
37605
37606
37607
37608
37609
37610
37611
37612
37613
37614
37615
37616
37617
37618
37619
37620
37621
37622
37623
37624
37625
37626
37627
37628
37629
37630
37631
37632
37633
37634
37635
37636
37637
37638
37639
37640
37641
37642
37643
37644
37645
37646
37647
37648
37649
37650
37651
37652
37653
37654
37655
37656
37657
37658
37659
37660
37661
37662
37663
37664
37665
37666
37667
37668
37669
37670
37671
37672
37673
37674
37675
37676
37677
37678
37679
37680
37681
37682
37683
37684
37685
37686
37687
37688
37689
37690
37691
37692
37693
37694
37695
37696
37697
37698
37699
37700
37701
37702
37703
37704
37705
37706
37707
37708
37709
37710
37711
37712
37713
37714
37715
37716
37717
37718
37719
37720
37721
37722
37723
37724
37725
37726
37727
37728
37729
37730
37731
37732
37733
37734
37735
37736
37737
37738
37739
37740
37741
37742
37743
37744
37745
37746
37747
37748
37749
37750
37751
37752
37753
37754
37755
37756
37757
37758
37759
37760
37761
37762
37763
37764
37765
37766
37767
37768
37769
37770
37771
37772
37773
37774
37775
37776
37777
37778
37779
37780
37781
37782
37783
37784
37785
37786
37787
37788
37789
37790
37791
37792
37793
37794
37795
37796
37797
37798
37799
37800
37801
37802
37803
37804
37805
37806
37807
37808
37809
37810
37811
37812
37813
37814
37815
37816
37817
37818
37819
37820
37821
37822
37823
37824
37825
37826
37827
37828
37829
37830
37831
37832
37833
37834
37835
37836
37837
37838
37839
37840
37841
37842
37843
37844
37845
37846
37847
37848
37849
37850
37851
37852
37853
37854
37855
37856
37857
37858
37859
37860
37861
37862
37863
37864
37865
37866
37867
37868
37869
37870
37871
37872
37873
37874
37875
37876
37877
37878
37879
37880
37881
37882
37883
37884
37885
37886
37887
37888
37889
37890
37891
37892
37893
37894
37895
37896
37897
37898
37899
37900
37901
37902
37903
37904
37905
37906
37907
37908
37909
37910
37911
37912
37913
37914
37915
37916
37917
37918
37919
37920
37921
37922
37923
37924
37925
37926
37927
37928
37929
37930
37931
37932
37933
37934
37935
37936
37937
37938
37939
37940
37941
37942
37943
37944
37945
37946
37947
37948
37949
37950
37951
37952
37953
37954
37955
37956
37957
37958
37959
37960
37961
37962
37963
37964
37965
37966
37967
37968
37969
37970
37971
37972
37973
37974
37975
37976
37977
37978
37979
37980
37981
37982
37983
37984
37985
37986
37987
37988
37989
37990
37991
37992
37993
37994
37995
37996
37997
37998
37999
38000
38001
38002
38003
38004
38005
38006
38007
38008
38009
38010
38011
38012
38013
38014
38015
38016
38017
38018
38019
38020
38021
38022
38023
38024
38025
38026
38027
38028
38029
38030
38031
38032
38033
38034
38035
38036
38037
38038
38039
38040
38041
38042
38043
38044
38045
38046
38047
38048
38049
38050
38051
38052
38053
38054
38055
38056
38057
38058
38059
38060
38061
38062
38063
38064
38065
38066
38067
38068
38069
38070
38071
38072
38073
38074
38075
38076
38077
38078
38079
38080
38081
38082
38083
38084
38085
38086
38087
38088
38089
38090
38091
38092
38093
38094
38095
38096
38097
38098
38099
38100
38101
38102
38103
38104
38105
38106
38107
38108
38109
38110
38111
38112
38113
38114
38115
38116
38117
38118
38119
38120
38121
38122
38123
38124
38125
38126
38127
38128
38129
38130
38131
38132
38133
38134
38135
38136
38137
38138
38139
38140
38141
38142
38143
38144
38145
38146
38147
38148
38149
38150
38151
38152
38153
38154
38155
38156
38157
38158
38159
38160
38161
38162
38163
38164
38165
38166
38167
38168
38169
38170
38171
38172
38173
38174
38175
38176
38177
38178
38179
38180
38181
38182
38183
38184
38185
38186
38187
38188
38189
38190
38191
38192
38193
38194
38195
38196
38197
38198
38199
38200
38201
38202
38203
38204
38205
38206
38207
38208
38209
38210
38211
38212
38213
38214
38215
38216
38217
38218
38219
38220
38221
38222
38223
38224
38225
38226
38227
38228
38229
38230
38231
38232
38233
38234
38235
38236
38237
38238
38239
38240
38241
38242
38243
38244
38245
38246
38247
38248
38249
38250
38251
38252
38253
38254
38255
38256
38257
38258
38259
38260
38261
38262
38263
38264
38265
38266
38267
38268
38269
38270
38271
38272
38273
38274
38275
38276
38277
38278
38279
38280
38281
38282
38283
38284
38285
38286
38287
38288
38289
38290
38291
38292
38293
38294
38295
38296
38297
38298
38299
38300
38301
38302
38303
38304
38305
38306
38307
38308
38309
38310
38311
38312
38313
38314
38315
38316
38317
38318
38319
38320
38321
38322
38323
38324
38325
38326
38327
38328
38329
38330
38331
38332
38333
38334
38335
38336
38337
38338
38339
38340
38341
38342
38343
38344
38345
38346
38347
38348
38349
38350
38351
38352
38353
38354
38355
38356
38357
38358
38359
38360
38361
38362
38363
38364
38365
38366
38367
38368
38369
38370
38371
38372
38373
38374
38375
38376
38377
38378
38379
38380
38381
38382
38383
38384
38385
38386
38387
38388
38389
38390
38391
38392
38393
38394
38395
38396
38397
38398
38399
38400
38401
38402
38403
38404
38405
38406
38407
38408
38409
38410
38411
38412
38413
38414
38415
38416
38417
38418
38419
38420
38421
38422
38423
38424
38425
38426
38427
38428
38429
38430
38431
38432
38433
38434
38435
38436
38437
38438
38439
38440
38441
38442
38443
38444
38445
38446
38447
38448
38449
38450
38451
38452
38453
38454
38455
38456
38457
38458
38459
38460
38461
38462
38463
38464
38465
38466
38467
38468
38469
38470
38471
38472
38473
38474
38475
38476
38477
38478
38479
38480
38481
38482
38483
38484
38485
38486
38487
38488
38489
38490
38491
38492
38493
38494
38495
38496
38497
38498
38499
38500
38501
38502
38503
38504
38505
38506
38507
38508
38509
38510
38511
38512
38513
38514
38515
38516
38517
38518
38519
38520
38521
38522
38523
38524
38525
38526
38527
38528
38529
38530
38531
38532
38533
38534
38535
38536
38537
38538
38539
38540
38541
38542
38543
38544
38545
38546
38547
38548
38549
38550
38551
38552
38553
38554
38555
38556
38557
38558
38559
38560
38561
38562
38563
38564
38565
38566
38567
38568
38569
38570
38571
38572
38573
38574
38575
38576
38577
38578
38579
38580
38581
38582
38583
38584
38585
38586
38587
38588
38589
38590
38591
38592
38593
38594
38595
38596
38597
38598
38599
38600
38601
38602
38603
38604
38605
38606
38607
38608
38609
38610
38611
38612
38613
38614
38615
38616
38617
38618
38619
38620
38621
38622
38623
38624
38625
38626
38627
38628
38629
38630
38631
38632
38633
38634
38635
38636
38637
38638
38639
38640
38641
38642
38643
38644
38645
38646
38647
38648
38649
38650
38651
38652
38653
38654
38655
38656
38657
38658
38659
38660
38661
38662
38663
38664
38665
38666
38667
38668
38669
38670
38671
38672
38673
38674
38675
38676
38677
38678
38679
38680
38681
38682
38683
38684
38685
38686
38687
38688
38689
38690
38691
38692
38693
38694
38695
38696
38697
38698
38699
38700
38701
38702
38703
38704
38705
38706
38707
38708
38709
38710
38711
38712
38713
38714
38715
38716
38717
38718
38719
38720
38721
38722
38723
38724
38725
38726
38727
38728
38729
38730
38731
38732
38733
38734
38735
38736
38737
38738
38739
38740
38741
38742
38743
38744
38745
38746
38747
38748
38749
38750
38751
38752
38753
38754
38755
38756
38757
38758
38759
38760
38761
38762
38763
38764
38765
38766
38767
38768
38769
38770
38771
38772
38773
38774
38775
38776
38777
38778
38779
38780
38781
38782
38783
38784
38785
38786
38787
38788
38789
38790
38791
38792
38793
38794
38795
38796
38797
38798
38799
38800
38801
38802
38803
38804
38805
38806
38807
38808
38809
38810
38811
38812
38813
38814
38815
38816
38817
38818
38819
38820
38821
38822
38823
38824
38825
38826
38827
38828
38829
38830
38831
38832
38833
38834
38835
38836
38837
38838
38839
38840
38841
38842
38843
38844
38845
38846
38847
38848
38849
38850
38851
38852
38853
38854
38855
38856
38857
38858
38859
38860
38861
38862
38863
38864
38865
38866
38867
38868
38869
38870
38871
38872
38873
38874
38875
38876
38877
38878
38879
38880
38881
38882
38883
38884
38885
38886
38887
38888
38889
38890
38891
38892
38893
38894
38895
38896
38897
38898
38899
38900
38901
38902
38903
38904
38905
38906
38907
38908
38909
38910
38911
38912
38913
38914
38915
38916
38917
38918
38919
38920
38921
38922
38923
38924
38925
38926
38927
38928
38929
38930
38931
38932
38933
38934
38935
38936
38937
38938
38939
38940
38941
38942
38943
38944
38945
38946
38947
38948
38949
38950
38951
38952
38953
38954
38955
38956
38957
38958
38959
38960
38961
38962
38963
38964
38965
38966
38967
38968
38969
38970
38971
38972
38973
38974
38975
38976
38977
38978
38979
38980
38981
38982
38983
38984
38985
38986
38987
38988
38989
38990
38991
38992
38993
38994
38995
38996
38997
38998
38999
39000
39001
39002
39003
39004
39005
39006
39007
39008
39009
39010
39011
39012
39013
39014
39015
39016
39017
39018
39019
39020
39021
39022
39023
39024
39025
39026
39027
39028
39029
39030
39031
39032
39033
39034
39035
39036
39037
39038
39039
39040
39041
39042
39043
39044
39045
39046
39047
39048
39049
39050
39051
39052
39053
39054
39055
39056
39057
39058
39059
39060
39061
39062
39063
39064
39065
39066
39067
39068
39069
39070
39071
39072
39073
39074
39075
39076
39077
39078
39079
39080
39081
39082
39083
39084
39085
39086
39087
39088
39089
39090
39091
39092
39093
39094
39095
39096
39097
39098
39099
39100
39101
39102
39103
39104
39105
39106
39107
39108
39109
39110
39111
39112
39113
39114
39115
39116
39117
39118
39119
39120
39121
39122
39123
39124
39125
39126
39127
39128
39129
39130
39131
39132
39133
39134
39135
39136
39137
39138
39139
39140
39141
39142
39143
39144
39145
39146
39147
39148
39149
39150
39151
39152
39153
39154
39155
39156
39157
39158
39159
39160
39161
39162
39163
39164
39165
39166
39167
39168
39169
39170
39171
39172
39173
39174
39175
39176
39177
39178
39179
39180
39181
39182
39183
39184
39185
39186
39187
39188
39189
39190
39191
39192
39193
39194
39195
39196
39197
39198
39199
39200
39201
39202
39203
39204
39205
39206
39207
39208
39209
39210
39211
39212
39213
39214
39215
39216
39217
39218
39219
39220
39221
39222
39223
39224
39225
39226
39227
39228
39229
39230
39231
39232
39233
39234
39235
39236
39237
39238
39239
39240
39241
39242
39243
39244
39245
39246
39247
39248
39249
39250
39251
39252
39253
39254
39255
39256
39257
39258
39259
39260
39261
39262
39263
39264
39265
39266
39267
39268
39269
39270
39271
39272
39273
39274
39275
39276
39277
39278
39279
39280
39281
39282
39283
39284
39285
39286
39287
39288
39289
39290
39291
39292
39293
39294
39295
39296
39297
39298
39299
39300
39301
39302
39303
39304
39305
39306
39307
39308
39309
39310
39311
39312
39313
39314
39315
39316
39317
39318
39319
39320
39321
39322
39323
39324
39325
39326
39327
39328
39329
39330
39331
39332
39333
39334
39335
39336
39337
39338
39339
39340
39341
39342
39343
39344
39345
39346
39347
39348
39349
39350
39351
39352
39353
39354
39355
39356
39357
39358
39359
39360
39361
39362
39363
39364
39365
39366
39367
39368
39369
39370
39371
39372
39373
39374
39375
39376
39377
39378
39379
39380
39381
39382
39383
39384
39385
39386
39387
39388
39389
39390
39391
39392
39393
39394
39395
39396
39397
39398
39399
39400
39401
39402
39403
39404
39405
39406
39407
39408
39409
39410
39411
39412
39413
39414
39415
39416
39417
39418
39419
39420
39421
39422
39423
39424
39425
39426
39427
39428
39429
39430
39431
39432
39433
39434
39435
39436
39437
39438
39439
39440
39441
39442
39443
39444
39445
39446
39447
39448
39449
39450
39451
39452
39453
39454
39455
39456
39457
39458
39459
39460
39461
39462
39463
39464
39465
39466
39467
39468
39469
39470
39471
39472
39473
39474
39475
39476
39477
39478
39479
39480
39481
39482
39483
39484
39485
39486
39487
39488
39489
39490
39491
39492
39493
39494
39495
39496
39497
39498
39499
39500
39501
39502
39503
39504
39505
39506
39507
39508
39509
39510
39511
39512
39513
39514
39515
39516
39517
39518
39519
39520
39521
39522
39523
39524
39525
39526
39527
39528
39529
39530
39531
39532
39533
39534
39535
39536
39537
39538
39539
39540
39541
39542
39543
39544
39545
39546
39547
39548
39549
39550
39551
39552
39553
39554
39555
39556
|
/* Subroutines used for code generation on IBM RS/6000.
Copyright (C) 1991-2017 Free Software Foundation, Inc.
Contributed by Richard Kenner (kenner@vlsi1.ultra.nyu.edu)
This file is part of GCC.
GCC is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published
by the Free Software Foundation; either version 3, or (at your
option) any later version.
GCC is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
License for more details.
You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING3. If not see
<http://www.gnu.org/licenses/>. */
#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "backend.h"
#include "rtl.h"
#include "tree.h"
#include "memmodel.h"
#include "gimple.h"
#include "cfghooks.h"
#include "cfgloop.h"
#include "df.h"
#include "tm_p.h"
#include "stringpool.h"
#include "expmed.h"
#include "optabs.h"
#include "regs.h"
#include "ira.h"
#include "recog.h"
#include "cgraph.h"
#include "diagnostic-core.h"
#include "insn-attr.h"
#include "flags.h"
#include "alias.h"
#include "fold-const.h"
#include "attribs.h"
#include "stor-layout.h"
#include "calls.h"
#include "print-tree.h"
#include "varasm.h"
#include "explow.h"
#include "expr.h"
#include "output.h"
#include "dbxout.h"
#include "common/common-target.h"
#include "langhooks.h"
#include "reload.h"
#include "sched-int.h"
#include "gimplify.h"
#include "gimple-fold.h"
#include "gimple-iterator.h"
#include "gimple-ssa.h"
#include "gimple-walk.h"
#include "intl.h"
#include "params.h"
#include "tm-constrs.h"
#include "tree-vectorizer.h"
#include "target-globals.h"
#include "builtins.h"
#include "context.h"
#include "tree-pass.h"
#include "except.h"
#if TARGET_XCOFF
#include "xcoffout.h" /* get declarations of xcoff_*_section_name */
#endif
#if TARGET_MACHO
#include "gstab.h" /* for N_SLINE */
#endif
#include "case-cfn-macros.h"
#include "ppc-auxv.h"
#include "tree-ssa-propagate.h"
/* This file should be included last. */
#include "target-def.h"
#ifndef TARGET_NO_PROTOTYPE
#define TARGET_NO_PROTOTYPE 0
#endif
/* Set -mabi=ieeelongdouble on some old targets. In the future, power server
systems will also set long double to be IEEE 128-bit. AIX and Darwin
explicitly redefine TARGET_IEEEQUAD and TARGET_IEEEQUAD_DEFAULT to 0, so
those systems will not pick up this default. This needs to be after all
of the include files, so that POWERPC_LINUX and POWERPC_FREEBSD are
properly defined. */
#ifndef TARGET_IEEEQUAD_DEFAULT
#if !defined (POWERPC_LINUX) && !defined (POWERPC_FREEBSD)
#define TARGET_IEEEQUAD_DEFAULT 1
#else
#define TARGET_IEEEQUAD_DEFAULT 0
#endif
#endif
#define min(A,B) ((A) < (B) ? (A) : (B))
#define max(A,B) ((A) > (B) ? (A) : (B))
static pad_direction rs6000_function_arg_padding (machine_mode, const_tree);
/* Structure used to define the rs6000 stack */
typedef struct rs6000_stack {
int reload_completed; /* stack info won't change from here on */
int first_gp_reg_save; /* first callee saved GP register used */
int first_fp_reg_save; /* first callee saved FP register used */
int first_altivec_reg_save; /* first callee saved AltiVec register used */
int lr_save_p; /* true if the link reg needs to be saved */
int cr_save_p; /* true if the CR reg needs to be saved */
unsigned int vrsave_mask; /* mask of vec registers to save */
int push_p; /* true if we need to allocate stack space */
int calls_p; /* true if the function makes any calls */
int world_save_p; /* true if we're saving *everything*:
r13-r31, cr, f14-f31, vrsave, v20-v31 */
enum rs6000_abi abi; /* which ABI to use */
int gp_save_offset; /* offset to save GP regs from initial SP */
int fp_save_offset; /* offset to save FP regs from initial SP */
int altivec_save_offset; /* offset to save AltiVec regs from initial SP */
int lr_save_offset; /* offset to save LR from initial SP */
int cr_save_offset; /* offset to save CR from initial SP */
int vrsave_save_offset; /* offset to save VRSAVE from initial SP */
int varargs_save_offset; /* offset to save the varargs registers */
int ehrd_offset; /* offset to EH return data */
int ehcr_offset; /* offset to EH CR field data */
int reg_size; /* register size (4 or 8) */
HOST_WIDE_INT vars_size; /* variable save area size */
int parm_size; /* outgoing parameter size */
int save_size; /* save area size */
int fixed_size; /* fixed size of stack frame */
int gp_size; /* size of saved GP registers */
int fp_size; /* size of saved FP registers */
int altivec_size; /* size of saved AltiVec registers */
int cr_size; /* size to hold CR if not in fixed area */
int vrsave_size; /* size to hold VRSAVE */
int altivec_padding_size; /* size of altivec alignment padding */
HOST_WIDE_INT total_size; /* total bytes allocated for stack */
int savres_strategy;
} rs6000_stack_t;
/* A C structure for machine-specific, per-function data.
This is added to the cfun structure. */
typedef struct GTY(()) machine_function
{
/* Flags if __builtin_return_address (n) with n >= 1 was used. */
int ra_needs_full_frame;
/* Flags if __builtin_return_address (0) was used. */
int ra_need_lr;
/* Cache lr_save_p after expansion of builtin_eh_return. */
int lr_save_state;
/* Whether we need to save the TOC to the reserved stack location in the
function prologue. */
bool save_toc_in_prologue;
/* Offset from virtual_stack_vars_rtx to the start of the ABI_V4
varargs save area. */
HOST_WIDE_INT varargs_save_offset;
/* Alternative internal arg pointer for -fsplit-stack. */
rtx split_stack_arg_pointer;
bool split_stack_argp_used;
/* Flag if r2 setup is needed with ELFv2 ABI. */
bool r2_setup_needed;
/* The number of components we use for separate shrink-wrapping. */
int n_components;
/* The components already handled by separate shrink-wrapping, which should
not be considered by the prologue and epilogue. */
bool gpr_is_wrapped_separately[32];
bool fpr_is_wrapped_separately[32];
bool lr_is_wrapped_separately;
bool toc_is_wrapped_separately;
} machine_function;
/* Support targetm.vectorize.builtin_mask_for_load. */
static GTY(()) tree altivec_builtin_mask_for_load;
/* Set to nonzero once AIX common-mode calls have been defined. */
static GTY(()) int common_mode_defined;
/* Label number of label created for -mrelocatable, to call to so we can
get the address of the GOT section */
static int rs6000_pic_labelno;
#ifdef USING_ELFOS_H
/* Counter for labels which are to be placed in .fixup. */
int fixuplabelno = 0;
#endif
/* Whether to use variant of AIX ABI for PowerPC64 Linux. */
int dot_symbols;
/* Specify the machine mode that pointers have. After generation of rtl, the
compiler makes no further distinction between pointers and any other objects
of this machine mode. */
scalar_int_mode rs6000_pmode;
/* Width in bits of a pointer. */
unsigned rs6000_pointer_size;
#ifdef HAVE_AS_GNU_ATTRIBUTE
# ifndef HAVE_LD_PPC_GNU_ATTR_LONG_DOUBLE
# define HAVE_LD_PPC_GNU_ATTR_LONG_DOUBLE 0
# endif
/* Flag whether floating point values have been passed/returned.
Note that this doesn't say whether fprs are used, since the
Tag_GNU_Power_ABI_FP .gnu.attributes value this flag controls
should be set for soft-float values passed in gprs and ieee128
values passed in vsx registers. */
static bool rs6000_passes_float;
static bool rs6000_passes_long_double;
/* Flag whether vector values have been passed/returned. */
static bool rs6000_passes_vector;
/* Flag whether small (<= 8 byte) structures have been returned. */
static bool rs6000_returns_struct;
#endif
/* Value is TRUE if register/mode pair is acceptable. */
static bool rs6000_hard_regno_mode_ok_p
[NUM_MACHINE_MODES][FIRST_PSEUDO_REGISTER];
/* Maximum number of registers needed for a given register class and mode. */
unsigned char rs6000_class_max_nregs[NUM_MACHINE_MODES][LIM_REG_CLASSES];
/* How many registers are needed for a given register and mode. */
unsigned char rs6000_hard_regno_nregs[NUM_MACHINE_MODES][FIRST_PSEUDO_REGISTER];
/* Map register number to register class. */
enum reg_class rs6000_regno_regclass[FIRST_PSEUDO_REGISTER];
static int dbg_cost_ctrl;
/* Built in types. */
tree rs6000_builtin_types[RS6000_BTI_MAX];
tree rs6000_builtin_decls[RS6000_BUILTIN_COUNT];
/* Flag to say the TOC is initialized */
int toc_initialized, need_toc_init;
char toc_label_name[10];
/* Cached value of rs6000_variable_issue. This is cached in
rs6000_variable_issue hook and returned from rs6000_sched_reorder2. */
static short cached_can_issue_more;
static GTY(()) section *read_only_data_section;
static GTY(()) section *private_data_section;
static GTY(()) section *tls_data_section;
static GTY(()) section *tls_private_data_section;
static GTY(()) section *read_only_private_data_section;
static GTY(()) section *sdata2_section;
static GTY(()) section *toc_section;
struct builtin_description
{
const HOST_WIDE_INT mask;
const enum insn_code icode;
const char *const name;
const enum rs6000_builtins code;
};
/* Describe the vector unit used for modes. */
enum rs6000_vector rs6000_vector_unit[NUM_MACHINE_MODES];
enum rs6000_vector rs6000_vector_mem[NUM_MACHINE_MODES];
/* Register classes for various constraints that are based on the target
switches. */
enum reg_class rs6000_constraints[RS6000_CONSTRAINT_MAX];
/* Describe the alignment of a vector. */
int rs6000_vector_align[NUM_MACHINE_MODES];
/* Map selected modes to types for builtins. */
static GTY(()) tree builtin_mode_to_type[MAX_MACHINE_MODE][2];
/* What modes to automatically generate reciprocal divide estimate (fre) and
reciprocal sqrt (frsqrte) for. */
unsigned char rs6000_recip_bits[MAX_MACHINE_MODE];
/* Masks to determine which reciprocal esitmate instructions to generate
automatically. */
enum rs6000_recip_mask {
RECIP_SF_DIV = 0x001, /* Use divide estimate */
RECIP_DF_DIV = 0x002,
RECIP_V4SF_DIV = 0x004,
RECIP_V2DF_DIV = 0x008,
RECIP_SF_RSQRT = 0x010, /* Use reciprocal sqrt estimate. */
RECIP_DF_RSQRT = 0x020,
RECIP_V4SF_RSQRT = 0x040,
RECIP_V2DF_RSQRT = 0x080,
/* Various combination of flags for -mrecip=xxx. */
RECIP_NONE = 0,
RECIP_ALL = (RECIP_SF_DIV | RECIP_DF_DIV | RECIP_V4SF_DIV
| RECIP_V2DF_DIV | RECIP_SF_RSQRT | RECIP_DF_RSQRT
| RECIP_V4SF_RSQRT | RECIP_V2DF_RSQRT),
RECIP_HIGH_PRECISION = RECIP_ALL,
/* On low precision machines like the power5, don't enable double precision
reciprocal square root estimate, since it isn't accurate enough. */
RECIP_LOW_PRECISION = (RECIP_ALL & ~(RECIP_DF_RSQRT | RECIP_V2DF_RSQRT))
};
/* -mrecip options. */
static struct
{
const char *string; /* option name */
unsigned int mask; /* mask bits to set */
} recip_options[] = {
{ "all", RECIP_ALL },
{ "none", RECIP_NONE },
{ "div", (RECIP_SF_DIV | RECIP_DF_DIV | RECIP_V4SF_DIV
| RECIP_V2DF_DIV) },
{ "divf", (RECIP_SF_DIV | RECIP_V4SF_DIV) },
{ "divd", (RECIP_DF_DIV | RECIP_V2DF_DIV) },
{ "rsqrt", (RECIP_SF_RSQRT | RECIP_DF_RSQRT | RECIP_V4SF_RSQRT
| RECIP_V2DF_RSQRT) },
{ "rsqrtf", (RECIP_SF_RSQRT | RECIP_V4SF_RSQRT) },
{ "rsqrtd", (RECIP_DF_RSQRT | RECIP_V2DF_RSQRT) },
};
/* Used by __builtin_cpu_is(), mapping from PLATFORM names to values. */
static const struct
{
const char *cpu;
unsigned int cpuid;
} cpu_is_info[] = {
{ "power9", PPC_PLATFORM_POWER9 },
{ "power8", PPC_PLATFORM_POWER8 },
{ "power7", PPC_PLATFORM_POWER7 },
{ "power6x", PPC_PLATFORM_POWER6X },
{ "power6", PPC_PLATFORM_POWER6 },
{ "power5+", PPC_PLATFORM_POWER5_PLUS },
{ "power5", PPC_PLATFORM_POWER5 },
{ "ppc970", PPC_PLATFORM_PPC970 },
{ "power4", PPC_PLATFORM_POWER4 },
{ "ppca2", PPC_PLATFORM_PPCA2 },
{ "ppc476", PPC_PLATFORM_PPC476 },
{ "ppc464", PPC_PLATFORM_PPC464 },
{ "ppc440", PPC_PLATFORM_PPC440 },
{ "ppc405", PPC_PLATFORM_PPC405 },
{ "ppc-cell-be", PPC_PLATFORM_CELL_BE }
};
/* Used by __builtin_cpu_supports(), mapping from HWCAP names to masks. */
static const struct
{
const char *hwcap;
int mask;
unsigned int id;
} cpu_supports_info[] = {
/* AT_HWCAP masks. */
{ "4xxmac", PPC_FEATURE_HAS_4xxMAC, 0 },
{ "altivec", PPC_FEATURE_HAS_ALTIVEC, 0 },
{ "arch_2_05", PPC_FEATURE_ARCH_2_05, 0 },
{ "arch_2_06", PPC_FEATURE_ARCH_2_06, 0 },
{ "archpmu", PPC_FEATURE_PERFMON_COMPAT, 0 },
{ "booke", PPC_FEATURE_BOOKE, 0 },
{ "cellbe", PPC_FEATURE_CELL_BE, 0 },
{ "dfp", PPC_FEATURE_HAS_DFP, 0 },
{ "efpdouble", PPC_FEATURE_HAS_EFP_DOUBLE, 0 },
{ "efpsingle", PPC_FEATURE_HAS_EFP_SINGLE, 0 },
{ "fpu", PPC_FEATURE_HAS_FPU, 0 },
{ "ic_snoop", PPC_FEATURE_ICACHE_SNOOP, 0 },
{ "mmu", PPC_FEATURE_HAS_MMU, 0 },
{ "notb", PPC_FEATURE_NO_TB, 0 },
{ "pa6t", PPC_FEATURE_PA6T, 0 },
{ "power4", PPC_FEATURE_POWER4, 0 },
{ "power5", PPC_FEATURE_POWER5, 0 },
{ "power5+", PPC_FEATURE_POWER5_PLUS, 0 },
{ "power6x", PPC_FEATURE_POWER6_EXT, 0 },
{ "ppc32", PPC_FEATURE_32, 0 },
{ "ppc601", PPC_FEATURE_601_INSTR, 0 },
{ "ppc64", PPC_FEATURE_64, 0 },
{ "ppcle", PPC_FEATURE_PPC_LE, 0 },
{ "smt", PPC_FEATURE_SMT, 0 },
{ "spe", PPC_FEATURE_HAS_SPE, 0 },
{ "true_le", PPC_FEATURE_TRUE_LE, 0 },
{ "ucache", PPC_FEATURE_UNIFIED_CACHE, 0 },
{ "vsx", PPC_FEATURE_HAS_VSX, 0 },
/* AT_HWCAP2 masks. */
{ "arch_2_07", PPC_FEATURE2_ARCH_2_07, 1 },
{ "dscr", PPC_FEATURE2_HAS_DSCR, 1 },
{ "ebb", PPC_FEATURE2_HAS_EBB, 1 },
{ "htm", PPC_FEATURE2_HAS_HTM, 1 },
{ "htm-nosc", PPC_FEATURE2_HTM_NOSC, 1 },
{ "htm-no-suspend", PPC_FEATURE2_HTM_NO_SUSPEND, 1 },
{ "isel", PPC_FEATURE2_HAS_ISEL, 1 },
{ "tar", PPC_FEATURE2_HAS_TAR, 1 },
{ "vcrypto", PPC_FEATURE2_HAS_VEC_CRYPTO, 1 },
{ "arch_3_00", PPC_FEATURE2_ARCH_3_00, 1 },
{ "ieee128", PPC_FEATURE2_HAS_IEEE128, 1 },
{ "darn", PPC_FEATURE2_DARN, 1 },
{ "scv", PPC_FEATURE2_SCV, 1 }
};
/* On PowerPC, we have a limited number of target clones that we care about
which means we can use an array to hold the options, rather than having more
elaborate data structures to identify each possible variation. Order the
clones from the default to the highest ISA. */
enum {
CLONE_DEFAULT = 0, /* default clone. */
CLONE_ISA_2_05, /* ISA 2.05 (power6). */
CLONE_ISA_2_06, /* ISA 2.06 (power7). */
CLONE_ISA_2_07, /* ISA 2.07 (power8). */
CLONE_ISA_3_00, /* ISA 3.00 (power9). */
CLONE_MAX
};
/* Map compiler ISA bits into HWCAP names. */
struct clone_map {
HOST_WIDE_INT isa_mask; /* rs6000_isa mask */
const char *name; /* name to use in __builtin_cpu_supports. */
};
static const struct clone_map rs6000_clone_map[CLONE_MAX] = {
{ 0, "" }, /* Default options. */
{ OPTION_MASK_CMPB, "arch_2_05" }, /* ISA 2.05 (power6). */
{ OPTION_MASK_POPCNTD, "arch_2_06" }, /* ISA 2.06 (power7). */
{ OPTION_MASK_P8_VECTOR, "arch_2_07" }, /* ISA 2.07 (power8). */
{ OPTION_MASK_P9_VECTOR, "arch_3_00" }, /* ISA 3.00 (power9). */
};
/* Newer LIBCs explicitly export this symbol to declare that they provide
the AT_PLATFORM and AT_HWCAP/AT_HWCAP2 values in the TCB. We emit a
reference to this symbol whenever we expand a CPU builtin, so that
we never link against an old LIBC. */
const char *tcb_verification_symbol = "__parse_hwcap_and_convert_at_platform";
/* True if we have expanded a CPU builtin. */
bool cpu_builtin_p;
/* Pointer to function (in rs6000-c.c) that can define or undefine target
macros that have changed. Languages that don't support the preprocessor
don't link in rs6000-c.c, so we can't call it directly. */
void (*rs6000_target_modify_macros_ptr) (bool, HOST_WIDE_INT, HOST_WIDE_INT);
/* Simplfy register classes into simpler classifications. We assume
GPR_REG_TYPE - FPR_REG_TYPE are ordered so that we can use a simple range
check for standard register classes (gpr/floating/altivec/vsx) and
floating/vector classes (float/altivec/vsx). */
enum rs6000_reg_type {
NO_REG_TYPE,
PSEUDO_REG_TYPE,
GPR_REG_TYPE,
VSX_REG_TYPE,
ALTIVEC_REG_TYPE,
FPR_REG_TYPE,
SPR_REG_TYPE,
CR_REG_TYPE
};
/* Map register class to register type. */
static enum rs6000_reg_type reg_class_to_reg_type[N_REG_CLASSES];
/* First/last register type for the 'normal' register types (i.e. general
purpose, floating point, altivec, and VSX registers). */
#define IS_STD_REG_TYPE(RTYPE) IN_RANGE(RTYPE, GPR_REG_TYPE, FPR_REG_TYPE)
#define IS_FP_VECT_REG_TYPE(RTYPE) IN_RANGE(RTYPE, VSX_REG_TYPE, FPR_REG_TYPE)
/* Register classes we care about in secondary reload or go if legitimate
address. We only need to worry about GPR, FPR, and Altivec registers here,
along an ANY field that is the OR of the 3 register classes. */
enum rs6000_reload_reg_type {
RELOAD_REG_GPR, /* General purpose registers. */
RELOAD_REG_FPR, /* Traditional floating point regs. */
RELOAD_REG_VMX, /* Altivec (VMX) registers. */
RELOAD_REG_ANY, /* OR of GPR, FPR, Altivec masks. */
N_RELOAD_REG
};
/* For setting up register classes, loop through the 3 register classes mapping
into real registers, and skip the ANY class, which is just an OR of the
bits. */
#define FIRST_RELOAD_REG_CLASS RELOAD_REG_GPR
#define LAST_RELOAD_REG_CLASS RELOAD_REG_VMX
/* Map reload register type to a register in the register class. */
struct reload_reg_map_type {
const char *name; /* Register class name. */
int reg; /* Register in the register class. */
};
static const struct reload_reg_map_type reload_reg_map[N_RELOAD_REG] = {
{ "Gpr", FIRST_GPR_REGNO }, /* RELOAD_REG_GPR. */
{ "Fpr", FIRST_FPR_REGNO }, /* RELOAD_REG_FPR. */
{ "VMX", FIRST_ALTIVEC_REGNO }, /* RELOAD_REG_VMX. */
{ "Any", -1 }, /* RELOAD_REG_ANY. */
};
/* Mask bits for each register class, indexed per mode. Historically the
compiler has been more restrictive which types can do PRE_MODIFY instead of
PRE_INC and PRE_DEC, so keep track of sepaate bits for these two. */
typedef unsigned char addr_mask_type;
#define RELOAD_REG_VALID 0x01 /* Mode valid in register.. */
#define RELOAD_REG_MULTIPLE 0x02 /* Mode takes multiple registers. */
#define RELOAD_REG_INDEXED 0x04 /* Reg+reg addressing. */
#define RELOAD_REG_OFFSET 0x08 /* Reg+offset addressing. */
#define RELOAD_REG_PRE_INCDEC 0x10 /* PRE_INC/PRE_DEC valid. */
#define RELOAD_REG_PRE_MODIFY 0x20 /* PRE_MODIFY valid. */
#define RELOAD_REG_AND_M16 0x40 /* AND -16 addressing. */
#define RELOAD_REG_QUAD_OFFSET 0x80 /* quad offset is limited. */
/* Register type masks based on the type, of valid addressing modes. */
struct rs6000_reg_addr {
enum insn_code reload_load; /* INSN to reload for loading. */
enum insn_code reload_store; /* INSN to reload for storing. */
enum insn_code reload_fpr_gpr; /* INSN to move from FPR to GPR. */
enum insn_code reload_gpr_vsx; /* INSN to move from GPR to VSX. */
enum insn_code reload_vsx_gpr; /* INSN to move from VSX to GPR. */
enum insn_code fusion_gpr_ld; /* INSN for fusing gpr ADDIS/loads. */
/* INSNs for fusing addi with loads
or stores for each reg. class. */
enum insn_code fusion_addi_ld[(int)N_RELOAD_REG];
enum insn_code fusion_addi_st[(int)N_RELOAD_REG];
/* INSNs for fusing addis with loads
or stores for each reg. class. */
enum insn_code fusion_addis_ld[(int)N_RELOAD_REG];
enum insn_code fusion_addis_st[(int)N_RELOAD_REG];
addr_mask_type addr_mask[(int)N_RELOAD_REG]; /* Valid address masks. */
bool scalar_in_vmx_p; /* Scalar value can go in VMX. */
bool fused_toc; /* Mode supports TOC fusion. */
};
static struct rs6000_reg_addr reg_addr[NUM_MACHINE_MODES];
/* Helper function to say whether a mode supports PRE_INC or PRE_DEC. */
static inline bool
mode_supports_pre_incdec_p (machine_mode mode)
{
return ((reg_addr[mode].addr_mask[RELOAD_REG_ANY] & RELOAD_REG_PRE_INCDEC)
!= 0);
}
/* Helper function to say whether a mode supports PRE_MODIFY. */
static inline bool
mode_supports_pre_modify_p (machine_mode mode)
{
return ((reg_addr[mode].addr_mask[RELOAD_REG_ANY] & RELOAD_REG_PRE_MODIFY)
!= 0);
}
/* Given that there exists at least one variable that is set (produced)
by OUT_INSN and read (consumed) by IN_INSN, return true iff
IN_INSN represents one or more memory store operations and none of
the variables set by OUT_INSN is used by IN_INSN as the address of a
store operation. If either IN_INSN or OUT_INSN does not represent
a "single" RTL SET expression (as loosely defined by the
implementation of the single_set function) or a PARALLEL with only
SETs, CLOBBERs, and USEs inside, this function returns false.
This rs6000-specific version of store_data_bypass_p checks for
certain conditions that result in assertion failures (and internal
compiler errors) in the generic store_data_bypass_p function and
returns false rather than calling store_data_bypass_p if one of the
problematic conditions is detected. */
int
rs6000_store_data_bypass_p (rtx_insn *out_insn, rtx_insn *in_insn)
{
rtx out_set, in_set;
rtx out_pat, in_pat;
rtx out_exp, in_exp;
int i, j;
in_set = single_set (in_insn);
if (in_set)
{
if (MEM_P (SET_DEST (in_set)))
{
out_set = single_set (out_insn);
if (!out_set)
{
out_pat = PATTERN (out_insn);
if (GET_CODE (out_pat) == PARALLEL)
{
for (i = 0; i < XVECLEN (out_pat, 0); i++)
{
out_exp = XVECEXP (out_pat, 0, i);
if ((GET_CODE (out_exp) == CLOBBER)
|| (GET_CODE (out_exp) == USE))
continue;
else if (GET_CODE (out_exp) != SET)
return false;
}
}
}
}
}
else
{
in_pat = PATTERN (in_insn);
if (GET_CODE (in_pat) != PARALLEL)
return false;
for (i = 0; i < XVECLEN (in_pat, 0); i++)
{
in_exp = XVECEXP (in_pat, 0, i);
if ((GET_CODE (in_exp) == CLOBBER) || (GET_CODE (in_exp) == USE))
continue;
else if (GET_CODE (in_exp) != SET)
return false;
if (MEM_P (SET_DEST (in_exp)))
{
out_set = single_set (out_insn);
if (!out_set)
{
out_pat = PATTERN (out_insn);
if (GET_CODE (out_pat) != PARALLEL)
return false;
for (j = 0; j < XVECLEN (out_pat, 0); j++)
{
out_exp = XVECEXP (out_pat, 0, j);
if ((GET_CODE (out_exp) == CLOBBER)
|| (GET_CODE (out_exp) == USE))
continue;
else if (GET_CODE (out_exp) != SET)
return false;
}
}
}
}
}
return store_data_bypass_p (out_insn, in_insn);
}
/* Return true if we have D-form addressing in altivec registers. */
static inline bool
mode_supports_vmx_dform (machine_mode mode)
{
return ((reg_addr[mode].addr_mask[RELOAD_REG_VMX] & RELOAD_REG_OFFSET) != 0);
}
/* Return true if we have D-form addressing in VSX registers. This addressing
is more limited than normal d-form addressing in that the offset must be
aligned on a 16-byte boundary. */
static inline bool
mode_supports_vsx_dform_quad (machine_mode mode)
{
return ((reg_addr[mode].addr_mask[RELOAD_REG_ANY] & RELOAD_REG_QUAD_OFFSET)
!= 0);
}
/* Processor costs (relative to an add) */
const struct processor_costs *rs6000_cost;
/* Instruction size costs on 32bit processors. */
static const
struct processor_costs size32_cost = {
COSTS_N_INSNS (1), /* mulsi */
COSTS_N_INSNS (1), /* mulsi_const */
COSTS_N_INSNS (1), /* mulsi_const9 */
COSTS_N_INSNS (1), /* muldi */
COSTS_N_INSNS (1), /* divsi */
COSTS_N_INSNS (1), /* divdi */
COSTS_N_INSNS (1), /* fp */
COSTS_N_INSNS (1), /* dmul */
COSTS_N_INSNS (1), /* sdiv */
COSTS_N_INSNS (1), /* ddiv */
32, /* cache line size */
0, /* l1 cache */
0, /* l2 cache */
0, /* streams */
0, /* SF->DF convert */
};
/* Instruction size costs on 64bit processors. */
static const
struct processor_costs size64_cost = {
COSTS_N_INSNS (1), /* mulsi */
COSTS_N_INSNS (1), /* mulsi_const */
COSTS_N_INSNS (1), /* mulsi_const9 */
COSTS_N_INSNS (1), /* muldi */
COSTS_N_INSNS (1), /* divsi */
COSTS_N_INSNS (1), /* divdi */
COSTS_N_INSNS (1), /* fp */
COSTS_N_INSNS (1), /* dmul */
COSTS_N_INSNS (1), /* sdiv */
COSTS_N_INSNS (1), /* ddiv */
128, /* cache line size */
0, /* l1 cache */
0, /* l2 cache */
0, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on RS64A processors. */
static const
struct processor_costs rs64a_cost = {
COSTS_N_INSNS (20), /* mulsi */
COSTS_N_INSNS (12), /* mulsi_const */
COSTS_N_INSNS (8), /* mulsi_const9 */
COSTS_N_INSNS (34), /* muldi */
COSTS_N_INSNS (65), /* divsi */
COSTS_N_INSNS (67), /* divdi */
COSTS_N_INSNS (4), /* fp */
COSTS_N_INSNS (4), /* dmul */
COSTS_N_INSNS (31), /* sdiv */
COSTS_N_INSNS (31), /* ddiv */
128, /* cache line size */
128, /* l1 cache */
2048, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on MPCCORE processors. */
static const
struct processor_costs mpccore_cost = {
COSTS_N_INSNS (2), /* mulsi */
COSTS_N_INSNS (2), /* mulsi_const */
COSTS_N_INSNS (2), /* mulsi_const9 */
COSTS_N_INSNS (2), /* muldi */
COSTS_N_INSNS (6), /* divsi */
COSTS_N_INSNS (6), /* divdi */
COSTS_N_INSNS (4), /* fp */
COSTS_N_INSNS (5), /* dmul */
COSTS_N_INSNS (10), /* sdiv */
COSTS_N_INSNS (17), /* ddiv */
32, /* cache line size */
4, /* l1 cache */
16, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on PPC403 processors. */
static const
struct processor_costs ppc403_cost = {
COSTS_N_INSNS (4), /* mulsi */
COSTS_N_INSNS (4), /* mulsi_const */
COSTS_N_INSNS (4), /* mulsi_const9 */
COSTS_N_INSNS (4), /* muldi */
COSTS_N_INSNS (33), /* divsi */
COSTS_N_INSNS (33), /* divdi */
COSTS_N_INSNS (11), /* fp */
COSTS_N_INSNS (11), /* dmul */
COSTS_N_INSNS (11), /* sdiv */
COSTS_N_INSNS (11), /* ddiv */
32, /* cache line size */
4, /* l1 cache */
16, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on PPC405 processors. */
static const
struct processor_costs ppc405_cost = {
COSTS_N_INSNS (5), /* mulsi */
COSTS_N_INSNS (4), /* mulsi_const */
COSTS_N_INSNS (3), /* mulsi_const9 */
COSTS_N_INSNS (5), /* muldi */
COSTS_N_INSNS (35), /* divsi */
COSTS_N_INSNS (35), /* divdi */
COSTS_N_INSNS (11), /* fp */
COSTS_N_INSNS (11), /* dmul */
COSTS_N_INSNS (11), /* sdiv */
COSTS_N_INSNS (11), /* ddiv */
32, /* cache line size */
16, /* l1 cache */
128, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on PPC440 processors. */
static const
struct processor_costs ppc440_cost = {
COSTS_N_INSNS (3), /* mulsi */
COSTS_N_INSNS (2), /* mulsi_const */
COSTS_N_INSNS (2), /* mulsi_const9 */
COSTS_N_INSNS (3), /* muldi */
COSTS_N_INSNS (34), /* divsi */
COSTS_N_INSNS (34), /* divdi */
COSTS_N_INSNS (5), /* fp */
COSTS_N_INSNS (5), /* dmul */
COSTS_N_INSNS (19), /* sdiv */
COSTS_N_INSNS (33), /* ddiv */
32, /* cache line size */
32, /* l1 cache */
256, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on PPC476 processors. */
static const
struct processor_costs ppc476_cost = {
COSTS_N_INSNS (4), /* mulsi */
COSTS_N_INSNS (4), /* mulsi_const */
COSTS_N_INSNS (4), /* mulsi_const9 */
COSTS_N_INSNS (4), /* muldi */
COSTS_N_INSNS (11), /* divsi */
COSTS_N_INSNS (11), /* divdi */
COSTS_N_INSNS (6), /* fp */
COSTS_N_INSNS (6), /* dmul */
COSTS_N_INSNS (19), /* sdiv */
COSTS_N_INSNS (33), /* ddiv */
32, /* l1 cache line size */
32, /* l1 cache */
512, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on PPC601 processors. */
static const
struct processor_costs ppc601_cost = {
COSTS_N_INSNS (5), /* mulsi */
COSTS_N_INSNS (5), /* mulsi_const */
COSTS_N_INSNS (5), /* mulsi_const9 */
COSTS_N_INSNS (5), /* muldi */
COSTS_N_INSNS (36), /* divsi */
COSTS_N_INSNS (36), /* divdi */
COSTS_N_INSNS (4), /* fp */
COSTS_N_INSNS (5), /* dmul */
COSTS_N_INSNS (17), /* sdiv */
COSTS_N_INSNS (31), /* ddiv */
32, /* cache line size */
32, /* l1 cache */
256, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on PPC603 processors. */
static const
struct processor_costs ppc603_cost = {
COSTS_N_INSNS (5), /* mulsi */
COSTS_N_INSNS (3), /* mulsi_const */
COSTS_N_INSNS (2), /* mulsi_const9 */
COSTS_N_INSNS (5), /* muldi */
COSTS_N_INSNS (37), /* divsi */
COSTS_N_INSNS (37), /* divdi */
COSTS_N_INSNS (3), /* fp */
COSTS_N_INSNS (4), /* dmul */
COSTS_N_INSNS (18), /* sdiv */
COSTS_N_INSNS (33), /* ddiv */
32, /* cache line size */
8, /* l1 cache */
64, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on PPC604 processors. */
static const
struct processor_costs ppc604_cost = {
COSTS_N_INSNS (4), /* mulsi */
COSTS_N_INSNS (4), /* mulsi_const */
COSTS_N_INSNS (4), /* mulsi_const9 */
COSTS_N_INSNS (4), /* muldi */
COSTS_N_INSNS (20), /* divsi */
COSTS_N_INSNS (20), /* divdi */
COSTS_N_INSNS (3), /* fp */
COSTS_N_INSNS (3), /* dmul */
COSTS_N_INSNS (18), /* sdiv */
COSTS_N_INSNS (32), /* ddiv */
32, /* cache line size */
16, /* l1 cache */
512, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on PPC604e processors. */
static const
struct processor_costs ppc604e_cost = {
COSTS_N_INSNS (2), /* mulsi */
COSTS_N_INSNS (2), /* mulsi_const */
COSTS_N_INSNS (2), /* mulsi_const9 */
COSTS_N_INSNS (2), /* muldi */
COSTS_N_INSNS (20), /* divsi */
COSTS_N_INSNS (20), /* divdi */
COSTS_N_INSNS (3), /* fp */
COSTS_N_INSNS (3), /* dmul */
COSTS_N_INSNS (18), /* sdiv */
COSTS_N_INSNS (32), /* ddiv */
32, /* cache line size */
32, /* l1 cache */
1024, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on PPC620 processors. */
static const
struct processor_costs ppc620_cost = {
COSTS_N_INSNS (5), /* mulsi */
COSTS_N_INSNS (4), /* mulsi_const */
COSTS_N_INSNS (3), /* mulsi_const9 */
COSTS_N_INSNS (7), /* muldi */
COSTS_N_INSNS (21), /* divsi */
COSTS_N_INSNS (37), /* divdi */
COSTS_N_INSNS (3), /* fp */
COSTS_N_INSNS (3), /* dmul */
COSTS_N_INSNS (18), /* sdiv */
COSTS_N_INSNS (32), /* ddiv */
128, /* cache line size */
32, /* l1 cache */
1024, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on PPC630 processors. */
static const
struct processor_costs ppc630_cost = {
COSTS_N_INSNS (5), /* mulsi */
COSTS_N_INSNS (4), /* mulsi_const */
COSTS_N_INSNS (3), /* mulsi_const9 */
COSTS_N_INSNS (7), /* muldi */
COSTS_N_INSNS (21), /* divsi */
COSTS_N_INSNS (37), /* divdi */
COSTS_N_INSNS (3), /* fp */
COSTS_N_INSNS (3), /* dmul */
COSTS_N_INSNS (17), /* sdiv */
COSTS_N_INSNS (21), /* ddiv */
128, /* cache line size */
64, /* l1 cache */
1024, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on Cell processor. */
/* COSTS_N_INSNS (1) ~ one add. */
static const
struct processor_costs ppccell_cost = {
COSTS_N_INSNS (9/2)+2, /* mulsi */
COSTS_N_INSNS (6/2), /* mulsi_const */
COSTS_N_INSNS (6/2), /* mulsi_const9 */
COSTS_N_INSNS (15/2)+2, /* muldi */
COSTS_N_INSNS (38/2), /* divsi */
COSTS_N_INSNS (70/2), /* divdi */
COSTS_N_INSNS (10/2), /* fp */
COSTS_N_INSNS (10/2), /* dmul */
COSTS_N_INSNS (74/2), /* sdiv */
COSTS_N_INSNS (74/2), /* ddiv */
128, /* cache line size */
32, /* l1 cache */
512, /* l2 cache */
6, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on PPC750 and PPC7400 processors. */
static const
struct processor_costs ppc750_cost = {
COSTS_N_INSNS (5), /* mulsi */
COSTS_N_INSNS (3), /* mulsi_const */
COSTS_N_INSNS (2), /* mulsi_const9 */
COSTS_N_INSNS (5), /* muldi */
COSTS_N_INSNS (17), /* divsi */
COSTS_N_INSNS (17), /* divdi */
COSTS_N_INSNS (3), /* fp */
COSTS_N_INSNS (3), /* dmul */
COSTS_N_INSNS (17), /* sdiv */
COSTS_N_INSNS (31), /* ddiv */
32, /* cache line size */
32, /* l1 cache */
512, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on PPC7450 processors. */
static const
struct processor_costs ppc7450_cost = {
COSTS_N_INSNS (4), /* mulsi */
COSTS_N_INSNS (3), /* mulsi_const */
COSTS_N_INSNS (3), /* mulsi_const9 */
COSTS_N_INSNS (4), /* muldi */
COSTS_N_INSNS (23), /* divsi */
COSTS_N_INSNS (23), /* divdi */
COSTS_N_INSNS (5), /* fp */
COSTS_N_INSNS (5), /* dmul */
COSTS_N_INSNS (21), /* sdiv */
COSTS_N_INSNS (35), /* ddiv */
32, /* cache line size */
32, /* l1 cache */
1024, /* l2 cache */
1, /* streams */
0, /* SF->DF convert */
};
/* Instruction costs on PPC8540 processors. */
static const
struct processor_costs ppc8540_cost = {
COSTS_N_INSNS (4), /* mulsi */
COSTS_N_INSNS (4), /* mulsi_const */
COSTS_N_INSNS (4), /* mulsi_const9 */
COSTS_N_INSNS (4), /* muldi */
COSTS_N_INSNS (19), /* divsi */
COSTS_N_INSNS (19), /* divdi */
COSTS_N_INSNS (4), /* fp */
COSTS_N_INSNS (4), /* dmul */
COSTS_N_INSNS (29), /* sdiv */
COSTS_N_INSNS (29), /* ddiv */
32, /* cache line size */
32, /* l1 cache */
256, /* l2 cache */
1, /* prefetch streams /*/
0, /* SF->DF convert */
};
/* Instruction costs on E300C2 and E300C3 cores. */
static const
struct processor_costs ppce300c2c3_cost = {
COSTS_N_INSNS (4), /* mulsi */
COSTS_N_INSNS (4), /* mulsi_const */
COSTS_N_INSNS (4), /* mulsi_const9 */
COSTS_N_INSNS (4), /* muldi */
COSTS_N_INSNS (19), /* divsi */
COSTS_N_INSNS (19), /* divdi */
COSTS_N_INSNS (3), /* fp */
COSTS_N_INSNS (4), /* dmul */
COSTS_N_INSNS (18), /* sdiv */
COSTS_N_INSNS (33), /* ddiv */
32,
16, /* l1 cache */
16, /* l2 cache */
1, /* prefetch streams /*/
0, /* SF->DF convert */
};
/* Instruction costs on PPCE500MC processors. */
static const
struct processor_costs ppce500mc_cost = {
COSTS_N_INSNS (4), /* mulsi */
COSTS_N_INSNS (4), /* mulsi_const */
COSTS_N_INSNS (4), /* mulsi_const9 */
COSTS_N_INSNS (4), /* muldi */
COSTS_N_INSNS (14), /* divsi */
COSTS_N_INSNS (14), /* divdi */
COSTS_N_INSNS (8), /* fp */
COSTS_N_INSNS (10), /* dmul */
COSTS_N_INSNS (36), /* sdiv */
COSTS_N_INSNS (66), /* ddiv */
64, /* cache line size */
32, /* l1 cache */
128, /* l2 cache */
1, /* prefetch streams /*/
0, /* SF->DF convert */
};
/* Instruction costs on PPCE500MC64 processors. */
static const
struct processor_costs ppce500mc64_cost = {
COSTS_N_INSNS (4), /* mulsi */
COSTS_N_INSNS (4), /* mulsi_const */
COSTS_N_INSNS (4), /* mulsi_const9 */
COSTS_N_INSNS (4), /* muldi */
COSTS_N_INSNS (14), /* divsi */
COSTS_N_INSNS (14), /* divdi */
COSTS_N_INSNS (4), /* fp */
COSTS_N_INSNS (10), /* dmul */
COSTS_N_INSNS (36), /* sdiv */
COSTS_N_INSNS (66), /* ddiv */
64, /* cache line size */
32, /* l1 cache */
128, /* l2 cache */
1, /* prefetch streams /*/
0, /* SF->DF convert */
};
/* Instruction costs on PPCE5500 processors. */
static const
struct processor_costs ppce5500_cost = {
COSTS_N_INSNS (5), /* mulsi */
COSTS_N_INSNS (5), /* mulsi_const */
COSTS_N_INSNS (4), /* mulsi_const9 */
COSTS_N_INSNS (5), /* muldi */
COSTS_N_INSNS (14), /* divsi */
COSTS_N_INSNS (14), /* divdi */
COSTS_N_INSNS (7), /* fp */
COSTS_N_INSNS (10), /* dmul */
COSTS_N_INSNS (36), /* sdiv */
COSTS_N_INSNS (66), /* ddiv */
64, /* cache line size */
32, /* l1 cache */
128, /* l2 cache */
1, /* prefetch streams /*/
0, /* SF->DF convert */
};
/* Instruction costs on PPCE6500 processors. */
static const
struct processor_costs ppce6500_cost = {
COSTS_N_INSNS (5), /* mulsi */
COSTS_N_INSNS (5), /* mulsi_const */
COSTS_N_INSNS (4), /* mulsi_const9 */
COSTS_N_INSNS (5), /* muldi */
COSTS_N_INSNS (14), /* divsi */
COSTS_N_INSNS (14), /* divdi */
COSTS_N_INSNS (7), /* fp */
COSTS_N_INSNS (10), /* dmul */
COSTS_N_INSNS (36), /* sdiv */
COSTS_N_INSNS (66), /* ddiv */
64, /* cache line size */
32, /* l1 cache */
128, /* l2 cache */
1, /* prefetch streams /*/
0, /* SF->DF convert */
};
/* Instruction costs on AppliedMicro Titan processors. */
static const
struct processor_costs titan_cost = {
COSTS_N_INSNS (5), /* mulsi */
COSTS_N_INSNS (5), /* mulsi_const */
COSTS_N_INSNS (5), /* mulsi_const9 */
COSTS_N_INSNS (5), /* muldi */
COSTS_N_INSNS (18), /* divsi */
COSTS_N_INSNS (18), /* divdi */
COSTS_N_INSNS (10), /* fp */
COSTS_N_INSNS (10), /* dmul */
COSTS_N_INSNS (46), /* sdiv */
COSTS_N_INSNS (72), /* ddiv */
32, /* cache line size */
32, /* l1 cache */
512, /* l2 cache */
1, /* prefetch streams /*/
0, /* SF->DF convert */
};
/* Instruction costs on POWER4 and POWER5 processors. */
static const
struct processor_costs power4_cost = {
COSTS_N_INSNS (3), /* mulsi */
COSTS_N_INSNS (2), /* mulsi_const */
COSTS_N_INSNS (2), /* mulsi_const9 */
COSTS_N_INSNS (4), /* muldi */
COSTS_N_INSNS (18), /* divsi */
COSTS_N_INSNS (34), /* divdi */
COSTS_N_INSNS (3), /* fp */
COSTS_N_INSNS (3), /* dmul */
COSTS_N_INSNS (17), /* sdiv */
COSTS_N_INSNS (17), /* ddiv */
128, /* cache line size */
32, /* l1 cache */
1024, /* l2 cache */
8, /* prefetch streams /*/
0, /* SF->DF convert */
};
/* Instruction costs on POWER6 processors. */
static const
struct processor_costs power6_cost = {
COSTS_N_INSNS (8), /* mulsi */
COSTS_N_INSNS (8), /* mulsi_const */
COSTS_N_INSNS (8), /* mulsi_const9 */
COSTS_N_INSNS (8), /* muldi */
COSTS_N_INSNS (22), /* divsi */
COSTS_N_INSNS (28), /* divdi */
COSTS_N_INSNS (3), /* fp */
COSTS_N_INSNS (3), /* dmul */
COSTS_N_INSNS (13), /* sdiv */
COSTS_N_INSNS (16), /* ddiv */
128, /* cache line size */
64, /* l1 cache */
2048, /* l2 cache */
16, /* prefetch streams */
0, /* SF->DF convert */
};
/* Instruction costs on POWER7 processors. */
static const
struct processor_costs power7_cost = {
COSTS_N_INSNS (2), /* mulsi */
COSTS_N_INSNS (2), /* mulsi_const */
COSTS_N_INSNS (2), /* mulsi_const9 */
COSTS_N_INSNS (2), /* muldi */
COSTS_N_INSNS (18), /* divsi */
COSTS_N_INSNS (34), /* divdi */
COSTS_N_INSNS (3), /* fp */
COSTS_N_INSNS (3), /* dmul */
COSTS_N_INSNS (13), /* sdiv */
COSTS_N_INSNS (16), /* ddiv */
128, /* cache line size */
32, /* l1 cache */
256, /* l2 cache */
12, /* prefetch streams */
COSTS_N_INSNS (3), /* SF->DF convert */
};
/* Instruction costs on POWER8 processors. */
static const
struct processor_costs power8_cost = {
COSTS_N_INSNS (3), /* mulsi */
COSTS_N_INSNS (3), /* mulsi_const */
COSTS_N_INSNS (3), /* mulsi_const9 */
COSTS_N_INSNS (3), /* muldi */
COSTS_N_INSNS (19), /* divsi */
COSTS_N_INSNS (35), /* divdi */
COSTS_N_INSNS (3), /* fp */
COSTS_N_INSNS (3), /* dmul */
COSTS_N_INSNS (14), /* sdiv */
COSTS_N_INSNS (17), /* ddiv */
128, /* cache line size */
32, /* l1 cache */
256, /* l2 cache */
12, /* prefetch streams */
COSTS_N_INSNS (3), /* SF->DF convert */
};
/* Instruction costs on POWER9 processors. */
static const
struct processor_costs power9_cost = {
COSTS_N_INSNS (3), /* mulsi */
COSTS_N_INSNS (3), /* mulsi_const */
COSTS_N_INSNS (3), /* mulsi_const9 */
COSTS_N_INSNS (3), /* muldi */
COSTS_N_INSNS (8), /* divsi */
COSTS_N_INSNS (12), /* divdi */
COSTS_N_INSNS (3), /* fp */
COSTS_N_INSNS (3), /* dmul */
COSTS_N_INSNS (13), /* sdiv */
COSTS_N_INSNS (18), /* ddiv */
128, /* cache line size */
32, /* l1 cache */
512, /* l2 cache */
8, /* prefetch streams */
COSTS_N_INSNS (3), /* SF->DF convert */
};
/* Instruction costs on POWER A2 processors. */
static const
struct processor_costs ppca2_cost = {
COSTS_N_INSNS (16), /* mulsi */
COSTS_N_INSNS (16), /* mulsi_const */
COSTS_N_INSNS (16), /* mulsi_const9 */
COSTS_N_INSNS (16), /* muldi */
COSTS_N_INSNS (22), /* divsi */
COSTS_N_INSNS (28), /* divdi */
COSTS_N_INSNS (3), /* fp */
COSTS_N_INSNS (3), /* dmul */
COSTS_N_INSNS (59), /* sdiv */
COSTS_N_INSNS (72), /* ddiv */
64,
16, /* l1 cache */
2048, /* l2 cache */
16, /* prefetch streams */
0, /* SF->DF convert */
};
/* Table that classifies rs6000 builtin functions (pure, const, etc.). */
#undef RS6000_BUILTIN_0
#undef RS6000_BUILTIN_1
#undef RS6000_BUILTIN_2
#undef RS6000_BUILTIN_3
#undef RS6000_BUILTIN_A
#undef RS6000_BUILTIN_D
#undef RS6000_BUILTIN_H
#undef RS6000_BUILTIN_P
#undef RS6000_BUILTIN_Q
#undef RS6000_BUILTIN_X
#define RS6000_BUILTIN_0(ENUM, NAME, MASK, ATTR, ICODE) \
{ NAME, ICODE, MASK, ATTR },
#define RS6000_BUILTIN_1(ENUM, NAME, MASK, ATTR, ICODE) \
{ NAME, ICODE, MASK, ATTR },
#define RS6000_BUILTIN_2(ENUM, NAME, MASK, ATTR, ICODE) \
{ NAME, ICODE, MASK, ATTR },
#define RS6000_BUILTIN_3(ENUM, NAME, MASK, ATTR, ICODE) \
{ NAME, ICODE, MASK, ATTR },
#define RS6000_BUILTIN_A(ENUM, NAME, MASK, ATTR, ICODE) \
{ NAME, ICODE, MASK, ATTR },
#define RS6000_BUILTIN_D(ENUM, NAME, MASK, ATTR, ICODE) \
{ NAME, ICODE, MASK, ATTR },
#define RS6000_BUILTIN_H(ENUM, NAME, MASK, ATTR, ICODE) \
{ NAME, ICODE, MASK, ATTR },
#define RS6000_BUILTIN_P(ENUM, NAME, MASK, ATTR, ICODE) \
{ NAME, ICODE, MASK, ATTR },
#define RS6000_BUILTIN_Q(ENUM, NAME, MASK, ATTR, ICODE) \
{ NAME, ICODE, MASK, ATTR },
#define RS6000_BUILTIN_X(ENUM, NAME, MASK, ATTR, ICODE) \
{ NAME, ICODE, MASK, ATTR },
struct rs6000_builtin_info_type {
const char *name;
const enum insn_code icode;
const HOST_WIDE_INT mask;
const unsigned attr;
};
static const struct rs6000_builtin_info_type rs6000_builtin_info[] =
{
#include "rs6000-builtin.def"
};
#undef RS6000_BUILTIN_0
#undef RS6000_BUILTIN_1
#undef RS6000_BUILTIN_2
#undef RS6000_BUILTIN_3
#undef RS6000_BUILTIN_A
#undef RS6000_BUILTIN_D
#undef RS6000_BUILTIN_H
#undef RS6000_BUILTIN_P
#undef RS6000_BUILTIN_Q
#undef RS6000_BUILTIN_X
/* Support for -mveclibabi=<xxx> to control which vector library to use. */
static tree (*rs6000_veclib_handler) (combined_fn, tree, tree);
static bool rs6000_debug_legitimate_address_p (machine_mode, rtx, bool);
static struct machine_function * rs6000_init_machine_status (void);
static int rs6000_ra_ever_killed (void);
static tree rs6000_handle_longcall_attribute (tree *, tree, tree, int, bool *);
static tree rs6000_handle_altivec_attribute (tree *, tree, tree, int, bool *);
static tree rs6000_handle_struct_attribute (tree *, tree, tree, int, bool *);
static tree rs6000_builtin_vectorized_libmass (combined_fn, tree, tree);
static void rs6000_emit_set_long_const (rtx, HOST_WIDE_INT);
static int rs6000_memory_move_cost (machine_mode, reg_class_t, bool);
static bool rs6000_debug_rtx_costs (rtx, machine_mode, int, int, int *, bool);
static int rs6000_debug_address_cost (rtx, machine_mode, addr_space_t,
bool);
static int rs6000_debug_adjust_cost (rtx_insn *, int, rtx_insn *, int,
unsigned int);
static bool is_microcoded_insn (rtx_insn *);
static bool is_nonpipeline_insn (rtx_insn *);
static bool is_cracked_insn (rtx_insn *);
static bool is_load_insn (rtx, rtx *);
static bool is_store_insn (rtx, rtx *);
static bool set_to_load_agen (rtx_insn *,rtx_insn *);
static bool insn_terminates_group_p (rtx_insn *, enum group_termination);
static bool insn_must_be_first_in_group (rtx_insn *);
static bool insn_must_be_last_in_group (rtx_insn *);
static void altivec_init_builtins (void);
static tree builtin_function_type (machine_mode, machine_mode,
machine_mode, machine_mode,
enum rs6000_builtins, const char *name);
static void rs6000_common_init_builtins (void);
static void paired_init_builtins (void);
static rtx paired_expand_predicate_builtin (enum insn_code, tree, rtx);
static void htm_init_builtins (void);
static rs6000_stack_t *rs6000_stack_info (void);
static void is_altivec_return_reg (rtx, void *);
int easy_vector_constant (rtx, machine_mode);
static rtx rs6000_debug_legitimize_address (rtx, rtx, machine_mode);
static rtx rs6000_legitimize_tls_address (rtx, enum tls_model);
static rtx rs6000_darwin64_record_arg (CUMULATIVE_ARGS *, const_tree,
bool, bool);
#if TARGET_MACHO
static void macho_branch_islands (void);
#endif
static rtx rs6000_legitimize_reload_address (rtx, machine_mode, int, int,
int, int *);
static rtx rs6000_debug_legitimize_reload_address (rtx, machine_mode, int,
int, int, int *);
static bool rs6000_mode_dependent_address (const_rtx);
static bool rs6000_debug_mode_dependent_address (const_rtx);
static enum reg_class rs6000_secondary_reload_class (enum reg_class,
machine_mode, rtx);
static enum reg_class rs6000_debug_secondary_reload_class (enum reg_class,
machine_mode,
rtx);
static enum reg_class rs6000_preferred_reload_class (rtx, enum reg_class);
static enum reg_class rs6000_debug_preferred_reload_class (rtx,
enum reg_class);
static bool rs6000_debug_secondary_memory_needed (machine_mode,
reg_class_t,
reg_class_t);
static bool rs6000_debug_can_change_mode_class (machine_mode,
machine_mode,
reg_class_t);
static bool rs6000_save_toc_in_prologue_p (void);
static rtx rs6000_internal_arg_pointer (void);
rtx (*rs6000_legitimize_reload_address_ptr) (rtx, machine_mode, int, int,
int, int *)
= rs6000_legitimize_reload_address;
static bool (*rs6000_mode_dependent_address_ptr) (const_rtx)
= rs6000_mode_dependent_address;
enum reg_class (*rs6000_secondary_reload_class_ptr) (enum reg_class,
machine_mode, rtx)
= rs6000_secondary_reload_class;
enum reg_class (*rs6000_preferred_reload_class_ptr) (rtx, enum reg_class)
= rs6000_preferred_reload_class;
const int INSN_NOT_AVAILABLE = -1;
static void rs6000_print_isa_options (FILE *, int, const char *,
HOST_WIDE_INT);
static void rs6000_print_builtin_options (FILE *, int, const char *,
HOST_WIDE_INT);
static HOST_WIDE_INT rs6000_disable_incompatible_switches (void);
static enum rs6000_reg_type register_to_reg_type (rtx, bool *);
static bool rs6000_secondary_reload_move (enum rs6000_reg_type,
enum rs6000_reg_type,
machine_mode,
secondary_reload_info *,
bool);
rtl_opt_pass *make_pass_analyze_swaps (gcc::context*);
static bool rs6000_keep_leaf_when_profiled () __attribute__ ((unused));
static tree rs6000_fold_builtin (tree, int, tree *, bool);
/* Hash table stuff for keeping track of TOC entries. */
struct GTY((for_user)) toc_hash_struct
{
/* `key' will satisfy CONSTANT_P; in fact, it will satisfy
ASM_OUTPUT_SPECIAL_POOL_ENTRY_P. */
rtx key;
machine_mode key_mode;
int labelno;
};
struct toc_hasher : ggc_ptr_hash<toc_hash_struct>
{
static hashval_t hash (toc_hash_struct *);
static bool equal (toc_hash_struct *, toc_hash_struct *);
};
static GTY (()) hash_table<toc_hasher> *toc_hash_table;
/* Hash table to keep track of the argument types for builtin functions. */
struct GTY((for_user)) builtin_hash_struct
{
tree type;
machine_mode mode[4]; /* return value + 3 arguments. */
unsigned char uns_p[4]; /* and whether the types are unsigned. */
};
struct builtin_hasher : ggc_ptr_hash<builtin_hash_struct>
{
static hashval_t hash (builtin_hash_struct *);
static bool equal (builtin_hash_struct *, builtin_hash_struct *);
};
static GTY (()) hash_table<builtin_hasher> *builtin_hash_table;
/* Default register names. */
char rs6000_reg_names[][8] =
{
"0", "1", "2", "3", "4", "5", "6", "7",
"8", "9", "10", "11", "12", "13", "14", "15",
"16", "17", "18", "19", "20", "21", "22", "23",
"24", "25", "26", "27", "28", "29", "30", "31",
"0", "1", "2", "3", "4", "5", "6", "7",
"8", "9", "10", "11", "12", "13", "14", "15",
"16", "17", "18", "19", "20", "21", "22", "23",
"24", "25", "26", "27", "28", "29", "30", "31",
"mq", "lr", "ctr","ap",
"0", "1", "2", "3", "4", "5", "6", "7",
"ca",
/* AltiVec registers. */
"0", "1", "2", "3", "4", "5", "6", "7",
"8", "9", "10", "11", "12", "13", "14", "15",
"16", "17", "18", "19", "20", "21", "22", "23",
"24", "25", "26", "27", "28", "29", "30", "31",
"vrsave", "vscr",
/* Soft frame pointer. */
"sfp",
/* HTM SPR registers. */
"tfhar", "tfiar", "texasr"
};
#ifdef TARGET_REGNAMES
static const char alt_reg_names[][8] =
{
"%r0", "%r1", "%r2", "%r3", "%r4", "%r5", "%r6", "%r7",
"%r8", "%r9", "%r10", "%r11", "%r12", "%r13", "%r14", "%r15",
"%r16", "%r17", "%r18", "%r19", "%r20", "%r21", "%r22", "%r23",
"%r24", "%r25", "%r26", "%r27", "%r28", "%r29", "%r30", "%r31",
"%f0", "%f1", "%f2", "%f3", "%f4", "%f5", "%f6", "%f7",
"%f8", "%f9", "%f10", "%f11", "%f12", "%f13", "%f14", "%f15",
"%f16", "%f17", "%f18", "%f19", "%f20", "%f21", "%f22", "%f23",
"%f24", "%f25", "%f26", "%f27", "%f28", "%f29", "%f30", "%f31",
"mq", "lr", "ctr", "ap",
"%cr0", "%cr1", "%cr2", "%cr3", "%cr4", "%cr5", "%cr6", "%cr7",
"ca",
/* AltiVec registers. */
"%v0", "%v1", "%v2", "%v3", "%v4", "%v5", "%v6", "%v7",
"%v8", "%v9", "%v10", "%v11", "%v12", "%v13", "%v14", "%v15",
"%v16", "%v17", "%v18", "%v19", "%v20", "%v21", "%v22", "%v23",
"%v24", "%v25", "%v26", "%v27", "%v28", "%v29", "%v30", "%v31",
"vrsave", "vscr",
/* Soft frame pointer. */
"sfp",
/* HTM SPR registers. */
"tfhar", "tfiar", "texasr"
};
#endif
/* Table of valid machine attributes. */
static const struct attribute_spec rs6000_attribute_table[] =
{
/* { name, min_len, max_len, decl_req, type_req, fn_type_req,
affects_type_identity, handler, exclude } */
{ "altivec", 1, 1, false, true, false, false,
rs6000_handle_altivec_attribute, NULL },
{ "longcall", 0, 0, false, true, true, false,
rs6000_handle_longcall_attribute, NULL },
{ "shortcall", 0, 0, false, true, true, false,
rs6000_handle_longcall_attribute, NULL },
{ "ms_struct", 0, 0, false, false, false, false,
rs6000_handle_struct_attribute, NULL },
{ "gcc_struct", 0, 0, false, false, false, false,
rs6000_handle_struct_attribute, NULL },
#ifdef SUBTARGET_ATTRIBUTE_TABLE
SUBTARGET_ATTRIBUTE_TABLE,
#endif
{ NULL, 0, 0, false, false, false, false, NULL, NULL }
};
#ifndef TARGET_PROFILE_KERNEL
#define TARGET_PROFILE_KERNEL 0
#endif
/* The VRSAVE bitmask puts bit %v0 as the most significant bit. */
#define ALTIVEC_REG_BIT(REGNO) (0x80000000 >> ((REGNO) - FIRST_ALTIVEC_REGNO))
/* Initialize the GCC target structure. */
#undef TARGET_ATTRIBUTE_TABLE
#define TARGET_ATTRIBUTE_TABLE rs6000_attribute_table
#undef TARGET_SET_DEFAULT_TYPE_ATTRIBUTES
#define TARGET_SET_DEFAULT_TYPE_ATTRIBUTES rs6000_set_default_type_attributes
#undef TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P
#define TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P rs6000_attribute_takes_identifier_p
#undef TARGET_ASM_ALIGNED_DI_OP
#define TARGET_ASM_ALIGNED_DI_OP DOUBLE_INT_ASM_OP
/* Default unaligned ops are only provided for ELF. Find the ops needed
for non-ELF systems. */
#ifndef OBJECT_FORMAT_ELF
#if TARGET_XCOFF
/* For XCOFF. rs6000_assemble_integer will handle unaligned DIs on
64-bit targets. */
#undef TARGET_ASM_UNALIGNED_HI_OP
#define TARGET_ASM_UNALIGNED_HI_OP "\t.vbyte\t2,"
#undef TARGET_ASM_UNALIGNED_SI_OP
#define TARGET_ASM_UNALIGNED_SI_OP "\t.vbyte\t4,"
#undef TARGET_ASM_UNALIGNED_DI_OP
#define TARGET_ASM_UNALIGNED_DI_OP "\t.vbyte\t8,"
#else
/* For Darwin. */
#undef TARGET_ASM_UNALIGNED_HI_OP
#define TARGET_ASM_UNALIGNED_HI_OP "\t.short\t"
#undef TARGET_ASM_UNALIGNED_SI_OP
#define TARGET_ASM_UNALIGNED_SI_OP "\t.long\t"
#undef TARGET_ASM_UNALIGNED_DI_OP
#define TARGET_ASM_UNALIGNED_DI_OP "\t.quad\t"
#undef TARGET_ASM_ALIGNED_DI_OP
#define TARGET_ASM_ALIGNED_DI_OP "\t.quad\t"
#endif
#endif
/* This hook deals with fixups for relocatable code and DI-mode objects
in 64-bit code. */
#undef TARGET_ASM_INTEGER
#define TARGET_ASM_INTEGER rs6000_assemble_integer
#if defined (HAVE_GAS_HIDDEN) && !TARGET_MACHO
#undef TARGET_ASM_ASSEMBLE_VISIBILITY
#define TARGET_ASM_ASSEMBLE_VISIBILITY rs6000_assemble_visibility
#endif
#undef TARGET_SET_UP_BY_PROLOGUE
#define TARGET_SET_UP_BY_PROLOGUE rs6000_set_up_by_prologue
#undef TARGET_SHRINK_WRAP_GET_SEPARATE_COMPONENTS
#define TARGET_SHRINK_WRAP_GET_SEPARATE_COMPONENTS rs6000_get_separate_components
#undef TARGET_SHRINK_WRAP_COMPONENTS_FOR_BB
#define TARGET_SHRINK_WRAP_COMPONENTS_FOR_BB rs6000_components_for_bb
#undef TARGET_SHRINK_WRAP_DISQUALIFY_COMPONENTS
#define TARGET_SHRINK_WRAP_DISQUALIFY_COMPONENTS rs6000_disqualify_components
#undef TARGET_SHRINK_WRAP_EMIT_PROLOGUE_COMPONENTS
#define TARGET_SHRINK_WRAP_EMIT_PROLOGUE_COMPONENTS rs6000_emit_prologue_components
#undef TARGET_SHRINK_WRAP_EMIT_EPILOGUE_COMPONENTS
#define TARGET_SHRINK_WRAP_EMIT_EPILOGUE_COMPONENTS rs6000_emit_epilogue_components
#undef TARGET_SHRINK_WRAP_SET_HANDLED_COMPONENTS
#define TARGET_SHRINK_WRAP_SET_HANDLED_COMPONENTS rs6000_set_handled_components
#undef TARGET_EXTRA_LIVE_ON_ENTRY
#define TARGET_EXTRA_LIVE_ON_ENTRY rs6000_live_on_entry
#undef TARGET_INTERNAL_ARG_POINTER
#define TARGET_INTERNAL_ARG_POINTER rs6000_internal_arg_pointer
#undef TARGET_HAVE_TLS
#define TARGET_HAVE_TLS HAVE_AS_TLS
#undef TARGET_CANNOT_FORCE_CONST_MEM
#define TARGET_CANNOT_FORCE_CONST_MEM rs6000_cannot_force_const_mem
#undef TARGET_DELEGITIMIZE_ADDRESS
#define TARGET_DELEGITIMIZE_ADDRESS rs6000_delegitimize_address
#undef TARGET_CONST_NOT_OK_FOR_DEBUG_P
#define TARGET_CONST_NOT_OK_FOR_DEBUG_P rs6000_const_not_ok_for_debug_p
#undef TARGET_LEGITIMATE_COMBINED_INSN
#define TARGET_LEGITIMATE_COMBINED_INSN rs6000_legitimate_combined_insn
#undef TARGET_ASM_FUNCTION_PROLOGUE
#define TARGET_ASM_FUNCTION_PROLOGUE rs6000_output_function_prologue
#undef TARGET_ASM_FUNCTION_EPILOGUE
#define TARGET_ASM_FUNCTION_EPILOGUE rs6000_output_function_epilogue
#undef TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA
#define TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA rs6000_output_addr_const_extra
#undef TARGET_LEGITIMIZE_ADDRESS
#define TARGET_LEGITIMIZE_ADDRESS rs6000_legitimize_address
#undef TARGET_SCHED_VARIABLE_ISSUE
#define TARGET_SCHED_VARIABLE_ISSUE rs6000_variable_issue
#undef TARGET_SCHED_ISSUE_RATE
#define TARGET_SCHED_ISSUE_RATE rs6000_issue_rate
#undef TARGET_SCHED_ADJUST_COST
#define TARGET_SCHED_ADJUST_COST rs6000_adjust_cost
#undef TARGET_SCHED_ADJUST_PRIORITY
#define TARGET_SCHED_ADJUST_PRIORITY rs6000_adjust_priority
#undef TARGET_SCHED_IS_COSTLY_DEPENDENCE
#define TARGET_SCHED_IS_COSTLY_DEPENDENCE rs6000_is_costly_dependence
#undef TARGET_SCHED_INIT
#define TARGET_SCHED_INIT rs6000_sched_init
#undef TARGET_SCHED_FINISH
#define TARGET_SCHED_FINISH rs6000_sched_finish
#undef TARGET_SCHED_REORDER
#define TARGET_SCHED_REORDER rs6000_sched_reorder
#undef TARGET_SCHED_REORDER2
#define TARGET_SCHED_REORDER2 rs6000_sched_reorder2
#undef TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD
#define TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD rs6000_use_sched_lookahead
#undef TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD
#define TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD rs6000_use_sched_lookahead_guard
#undef TARGET_SCHED_ALLOC_SCHED_CONTEXT
#define TARGET_SCHED_ALLOC_SCHED_CONTEXT rs6000_alloc_sched_context
#undef TARGET_SCHED_INIT_SCHED_CONTEXT
#define TARGET_SCHED_INIT_SCHED_CONTEXT rs6000_init_sched_context
#undef TARGET_SCHED_SET_SCHED_CONTEXT
#define TARGET_SCHED_SET_SCHED_CONTEXT rs6000_set_sched_context
#undef TARGET_SCHED_FREE_SCHED_CONTEXT
#define TARGET_SCHED_FREE_SCHED_CONTEXT rs6000_free_sched_context
#undef TARGET_SCHED_CAN_SPECULATE_INSN
#define TARGET_SCHED_CAN_SPECULATE_INSN rs6000_sched_can_speculate_insn
#undef TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD
#define TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD rs6000_builtin_mask_for_load
#undef TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT
#define TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT \
rs6000_builtin_support_vector_misalignment
#undef TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE
#define TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE rs6000_vector_alignment_reachable
#undef TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST
#define TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST \
rs6000_builtin_vectorization_cost
#undef TARGET_VECTORIZE_PREFERRED_SIMD_MODE
#define TARGET_VECTORIZE_PREFERRED_SIMD_MODE \
rs6000_preferred_simd_mode
#undef TARGET_VECTORIZE_INIT_COST
#define TARGET_VECTORIZE_INIT_COST rs6000_init_cost
#undef TARGET_VECTORIZE_ADD_STMT_COST
#define TARGET_VECTORIZE_ADD_STMT_COST rs6000_add_stmt_cost
#undef TARGET_VECTORIZE_FINISH_COST
#define TARGET_VECTORIZE_FINISH_COST rs6000_finish_cost
#undef TARGET_VECTORIZE_DESTROY_COST_DATA
#define TARGET_VECTORIZE_DESTROY_COST_DATA rs6000_destroy_cost_data
#undef TARGET_INIT_BUILTINS
#define TARGET_INIT_BUILTINS rs6000_init_builtins
#undef TARGET_BUILTIN_DECL
#define TARGET_BUILTIN_DECL rs6000_builtin_decl
#undef TARGET_FOLD_BUILTIN
#define TARGET_FOLD_BUILTIN rs6000_fold_builtin
#undef TARGET_GIMPLE_FOLD_BUILTIN
#define TARGET_GIMPLE_FOLD_BUILTIN rs6000_gimple_fold_builtin
#undef TARGET_EXPAND_BUILTIN
#define TARGET_EXPAND_BUILTIN rs6000_expand_builtin
#undef TARGET_MANGLE_TYPE
#define TARGET_MANGLE_TYPE rs6000_mangle_type
#undef TARGET_INIT_LIBFUNCS
#define TARGET_INIT_LIBFUNCS rs6000_init_libfuncs
#if TARGET_MACHO
#undef TARGET_BINDS_LOCAL_P
#define TARGET_BINDS_LOCAL_P darwin_binds_local_p
#endif
#undef TARGET_MS_BITFIELD_LAYOUT_P
#define TARGET_MS_BITFIELD_LAYOUT_P rs6000_ms_bitfield_layout_p
#undef TARGET_ASM_OUTPUT_MI_THUNK
#define TARGET_ASM_OUTPUT_MI_THUNK rs6000_output_mi_thunk
#undef TARGET_ASM_CAN_OUTPUT_MI_THUNK
#define TARGET_ASM_CAN_OUTPUT_MI_THUNK hook_bool_const_tree_hwi_hwi_const_tree_true
#undef TARGET_FUNCTION_OK_FOR_SIBCALL
#define TARGET_FUNCTION_OK_FOR_SIBCALL rs6000_function_ok_for_sibcall
#undef TARGET_REGISTER_MOVE_COST
#define TARGET_REGISTER_MOVE_COST rs6000_register_move_cost
#undef TARGET_MEMORY_MOVE_COST
#define TARGET_MEMORY_MOVE_COST rs6000_memory_move_cost
#undef TARGET_CANNOT_COPY_INSN_P
#define TARGET_CANNOT_COPY_INSN_P rs6000_cannot_copy_insn_p
#undef TARGET_RTX_COSTS
#define TARGET_RTX_COSTS rs6000_rtx_costs
#undef TARGET_ADDRESS_COST
#define TARGET_ADDRESS_COST hook_int_rtx_mode_as_bool_0
#undef TARGET_INSN_COST
#define TARGET_INSN_COST rs6000_insn_cost
#undef TARGET_INIT_DWARF_REG_SIZES_EXTRA
#define TARGET_INIT_DWARF_REG_SIZES_EXTRA rs6000_init_dwarf_reg_sizes_extra
#undef TARGET_PROMOTE_FUNCTION_MODE
#define TARGET_PROMOTE_FUNCTION_MODE rs6000_promote_function_mode
#undef TARGET_RETURN_IN_MEMORY
#define TARGET_RETURN_IN_MEMORY rs6000_return_in_memory
#undef TARGET_RETURN_IN_MSB
#define TARGET_RETURN_IN_MSB rs6000_return_in_msb
#undef TARGET_SETUP_INCOMING_VARARGS
#define TARGET_SETUP_INCOMING_VARARGS setup_incoming_varargs
/* Always strict argument naming on rs6000. */
#undef TARGET_STRICT_ARGUMENT_NAMING
#define TARGET_STRICT_ARGUMENT_NAMING hook_bool_CUMULATIVE_ARGS_true
#undef TARGET_PRETEND_OUTGOING_VARARGS_NAMED
#define TARGET_PRETEND_OUTGOING_VARARGS_NAMED hook_bool_CUMULATIVE_ARGS_true
#undef TARGET_SPLIT_COMPLEX_ARG
#define TARGET_SPLIT_COMPLEX_ARG hook_bool_const_tree_true
#undef TARGET_MUST_PASS_IN_STACK
#define TARGET_MUST_PASS_IN_STACK rs6000_must_pass_in_stack
#undef TARGET_PASS_BY_REFERENCE
#define TARGET_PASS_BY_REFERENCE rs6000_pass_by_reference
#undef TARGET_ARG_PARTIAL_BYTES
#define TARGET_ARG_PARTIAL_BYTES rs6000_arg_partial_bytes
#undef TARGET_FUNCTION_ARG_ADVANCE
#define TARGET_FUNCTION_ARG_ADVANCE rs6000_function_arg_advance
#undef TARGET_FUNCTION_ARG
#define TARGET_FUNCTION_ARG rs6000_function_arg
#undef TARGET_FUNCTION_ARG_PADDING
#define TARGET_FUNCTION_ARG_PADDING rs6000_function_arg_padding
#undef TARGET_FUNCTION_ARG_BOUNDARY
#define TARGET_FUNCTION_ARG_BOUNDARY rs6000_function_arg_boundary
#undef TARGET_BUILD_BUILTIN_VA_LIST
#define TARGET_BUILD_BUILTIN_VA_LIST rs6000_build_builtin_va_list
#undef TARGET_EXPAND_BUILTIN_VA_START
#define TARGET_EXPAND_BUILTIN_VA_START rs6000_va_start
#undef TARGET_GIMPLIFY_VA_ARG_EXPR
#define TARGET_GIMPLIFY_VA_ARG_EXPR rs6000_gimplify_va_arg
#undef TARGET_EH_RETURN_FILTER_MODE
#define TARGET_EH_RETURN_FILTER_MODE rs6000_eh_return_filter_mode
#undef TARGET_SCALAR_MODE_SUPPORTED_P
#define TARGET_SCALAR_MODE_SUPPORTED_P rs6000_scalar_mode_supported_p
#undef TARGET_VECTOR_MODE_SUPPORTED_P
#define TARGET_VECTOR_MODE_SUPPORTED_P rs6000_vector_mode_supported_p
#undef TARGET_FLOATN_MODE
#define TARGET_FLOATN_MODE rs6000_floatn_mode
#undef TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN
#define TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN invalid_arg_for_unprototyped_fn
#undef TARGET_ASM_LOOP_ALIGN_MAX_SKIP
#define TARGET_ASM_LOOP_ALIGN_MAX_SKIP rs6000_loop_align_max_skip
#undef TARGET_MD_ASM_ADJUST
#define TARGET_MD_ASM_ADJUST rs6000_md_asm_adjust
#undef TARGET_OPTION_OVERRIDE
#define TARGET_OPTION_OVERRIDE rs6000_option_override
#undef TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION
#define TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION \
rs6000_builtin_vectorized_function
#undef TARGET_VECTORIZE_BUILTIN_MD_VECTORIZED_FUNCTION
#define TARGET_VECTORIZE_BUILTIN_MD_VECTORIZED_FUNCTION \
rs6000_builtin_md_vectorized_function
#undef TARGET_STACK_PROTECT_GUARD
#define TARGET_STACK_PROTECT_GUARD rs6000_init_stack_protect_guard
#if !TARGET_MACHO
#undef TARGET_STACK_PROTECT_FAIL
#define TARGET_STACK_PROTECT_FAIL rs6000_stack_protect_fail
#endif
#ifdef HAVE_AS_TLS
#undef TARGET_ASM_OUTPUT_DWARF_DTPREL
#define TARGET_ASM_OUTPUT_DWARF_DTPREL rs6000_output_dwarf_dtprel
#endif
/* Use a 32-bit anchor range. This leads to sequences like:
addis tmp,anchor,high
add dest,tmp,low
where tmp itself acts as an anchor, and can be shared between
accesses to the same 64k page. */
#undef TARGET_MIN_ANCHOR_OFFSET
#define TARGET_MIN_ANCHOR_OFFSET -0x7fffffff - 1
#undef TARGET_MAX_ANCHOR_OFFSET
#define TARGET_MAX_ANCHOR_OFFSET 0x7fffffff
#undef TARGET_USE_BLOCKS_FOR_CONSTANT_P
#define TARGET_USE_BLOCKS_FOR_CONSTANT_P rs6000_use_blocks_for_constant_p
#undef TARGET_USE_BLOCKS_FOR_DECL_P
#define TARGET_USE_BLOCKS_FOR_DECL_P rs6000_use_blocks_for_decl_p
#undef TARGET_BUILTIN_RECIPROCAL
#define TARGET_BUILTIN_RECIPROCAL rs6000_builtin_reciprocal
#undef TARGET_SECONDARY_RELOAD
#define TARGET_SECONDARY_RELOAD rs6000_secondary_reload
#undef TARGET_SECONDARY_MEMORY_NEEDED
#define TARGET_SECONDARY_MEMORY_NEEDED rs6000_secondary_memory_needed
#undef TARGET_SECONDARY_MEMORY_NEEDED_MODE
#define TARGET_SECONDARY_MEMORY_NEEDED_MODE rs6000_secondary_memory_needed_mode
#undef TARGET_LEGITIMATE_ADDRESS_P
#define TARGET_LEGITIMATE_ADDRESS_P rs6000_legitimate_address_p
#undef TARGET_MODE_DEPENDENT_ADDRESS_P
#define TARGET_MODE_DEPENDENT_ADDRESS_P rs6000_mode_dependent_address_p
#undef TARGET_COMPUTE_PRESSURE_CLASSES
#define TARGET_COMPUTE_PRESSURE_CLASSES rs6000_compute_pressure_classes
#undef TARGET_CAN_ELIMINATE
#define TARGET_CAN_ELIMINATE rs6000_can_eliminate
#undef TARGET_CONDITIONAL_REGISTER_USAGE
#define TARGET_CONDITIONAL_REGISTER_USAGE rs6000_conditional_register_usage
#undef TARGET_SCHED_REASSOCIATION_WIDTH
#define TARGET_SCHED_REASSOCIATION_WIDTH rs6000_reassociation_width
#undef TARGET_TRAMPOLINE_INIT
#define TARGET_TRAMPOLINE_INIT rs6000_trampoline_init
#undef TARGET_FUNCTION_VALUE
#define TARGET_FUNCTION_VALUE rs6000_function_value
#undef TARGET_OPTION_VALID_ATTRIBUTE_P
#define TARGET_OPTION_VALID_ATTRIBUTE_P rs6000_valid_attribute_p
#undef TARGET_OPTION_SAVE
#define TARGET_OPTION_SAVE rs6000_function_specific_save
#undef TARGET_OPTION_RESTORE
#define TARGET_OPTION_RESTORE rs6000_function_specific_restore
#undef TARGET_OPTION_PRINT
#define TARGET_OPTION_PRINT rs6000_function_specific_print
#undef TARGET_CAN_INLINE_P
#define TARGET_CAN_INLINE_P rs6000_can_inline_p
#undef TARGET_SET_CURRENT_FUNCTION
#define TARGET_SET_CURRENT_FUNCTION rs6000_set_current_function
#undef TARGET_LEGITIMATE_CONSTANT_P
#define TARGET_LEGITIMATE_CONSTANT_P rs6000_legitimate_constant_p
#undef TARGET_VECTORIZE_VEC_PERM_CONST_OK
#define TARGET_VECTORIZE_VEC_PERM_CONST_OK rs6000_vectorize_vec_perm_const_ok
#undef TARGET_CAN_USE_DOLOOP_P
#define TARGET_CAN_USE_DOLOOP_P can_use_doloop_if_innermost
#undef TARGET_ATOMIC_ASSIGN_EXPAND_FENV
#define TARGET_ATOMIC_ASSIGN_EXPAND_FENV rs6000_atomic_assign_expand_fenv
#undef TARGET_LIBGCC_CMP_RETURN_MODE
#define TARGET_LIBGCC_CMP_RETURN_MODE rs6000_abi_word_mode
#undef TARGET_LIBGCC_SHIFT_COUNT_MODE
#define TARGET_LIBGCC_SHIFT_COUNT_MODE rs6000_abi_word_mode
#undef TARGET_UNWIND_WORD_MODE
#define TARGET_UNWIND_WORD_MODE rs6000_abi_word_mode
#undef TARGET_OFFLOAD_OPTIONS
#define TARGET_OFFLOAD_OPTIONS rs6000_offload_options
#undef TARGET_C_MODE_FOR_SUFFIX
#define TARGET_C_MODE_FOR_SUFFIX rs6000_c_mode_for_suffix
#undef TARGET_INVALID_BINARY_OP
#define TARGET_INVALID_BINARY_OP rs6000_invalid_binary_op
#undef TARGET_OPTAB_SUPPORTED_P
#define TARGET_OPTAB_SUPPORTED_P rs6000_optab_supported_p
#undef TARGET_CUSTOM_FUNCTION_DESCRIPTORS
#define TARGET_CUSTOM_FUNCTION_DESCRIPTORS 1
#undef TARGET_COMPARE_VERSION_PRIORITY
#define TARGET_COMPARE_VERSION_PRIORITY rs6000_compare_version_priority
#undef TARGET_GENERATE_VERSION_DISPATCHER_BODY
#define TARGET_GENERATE_VERSION_DISPATCHER_BODY \
rs6000_generate_version_dispatcher_body
#undef TARGET_GET_FUNCTION_VERSIONS_DISPATCHER
#define TARGET_GET_FUNCTION_VERSIONS_DISPATCHER \
rs6000_get_function_versions_dispatcher
#undef TARGET_OPTION_FUNCTION_VERSIONS
#define TARGET_OPTION_FUNCTION_VERSIONS common_function_versions
#undef TARGET_HARD_REGNO_NREGS
#define TARGET_HARD_REGNO_NREGS rs6000_hard_regno_nregs_hook
#undef TARGET_HARD_REGNO_MODE_OK
#define TARGET_HARD_REGNO_MODE_OK rs6000_hard_regno_mode_ok
#undef TARGET_MODES_TIEABLE_P
#define TARGET_MODES_TIEABLE_P rs6000_modes_tieable_p
#undef TARGET_HARD_REGNO_CALL_PART_CLOBBERED
#define TARGET_HARD_REGNO_CALL_PART_CLOBBERED \
rs6000_hard_regno_call_part_clobbered
#undef TARGET_SLOW_UNALIGNED_ACCESS
#define TARGET_SLOW_UNALIGNED_ACCESS rs6000_slow_unaligned_access
#undef TARGET_CAN_CHANGE_MODE_CLASS
#define TARGET_CAN_CHANGE_MODE_CLASS rs6000_can_change_mode_class
#undef TARGET_CONSTANT_ALIGNMENT
#define TARGET_CONSTANT_ALIGNMENT rs6000_constant_alignment
#undef TARGET_STARTING_FRAME_OFFSET
#define TARGET_STARTING_FRAME_OFFSET rs6000_starting_frame_offset
/* Processor table. */
struct rs6000_ptt
{
const char *const name; /* Canonical processor name. */
const enum processor_type processor; /* Processor type enum value. */
const HOST_WIDE_INT target_enable; /* Target flags to enable. */
};
static struct rs6000_ptt const processor_target_table[] =
{
#define RS6000_CPU(NAME, CPU, FLAGS) { NAME, CPU, FLAGS },
#include "rs6000-cpus.def"
#undef RS6000_CPU
};
/* Look up a processor name for -mcpu=xxx and -mtune=xxx. Return -1 if the
name is invalid. */
static int
rs6000_cpu_name_lookup (const char *name)
{
size_t i;
if (name != NULL)
{
for (i = 0; i < ARRAY_SIZE (processor_target_table); i++)
if (! strcmp (name, processor_target_table[i].name))
return (int)i;
}
return -1;
}
/* Return number of consecutive hard regs needed starting at reg REGNO
to hold something of mode MODE.
This is ordinarily the length in words of a value of mode MODE
but can be less for certain modes in special long registers.
POWER and PowerPC GPRs hold 32 bits worth;
PowerPC64 GPRs and FPRs point register holds 64 bits worth. */
static int
rs6000_hard_regno_nregs_internal (int regno, machine_mode mode)
{
unsigned HOST_WIDE_INT reg_size;
/* 128-bit floating point usually takes 2 registers, unless it is IEEE
128-bit floating point that can go in vector registers, which has VSX
memory addressing. */
if (FP_REGNO_P (regno))
reg_size = (VECTOR_MEM_VSX_P (mode) || FLOAT128_VECTOR_P (mode)
? UNITS_PER_VSX_WORD
: UNITS_PER_FP_WORD);
else if (ALTIVEC_REGNO_P (regno))
reg_size = UNITS_PER_ALTIVEC_WORD;
else
reg_size = UNITS_PER_WORD;
return (GET_MODE_SIZE (mode) + reg_size - 1) / reg_size;
}
/* Value is 1 if hard register REGNO can hold a value of machine-mode
MODE. */
static int
rs6000_hard_regno_mode_ok_uncached (int regno, machine_mode mode)
{
int last_regno = regno + rs6000_hard_regno_nregs[mode][regno] - 1;
if (COMPLEX_MODE_P (mode))
mode = GET_MODE_INNER (mode);
/* PTImode can only go in GPRs. Quad word memory operations require even/odd
register combinations, and use PTImode where we need to deal with quad
word memory operations. Don't allow quad words in the argument or frame
pointer registers, just registers 0..31. */
if (mode == PTImode)
return (IN_RANGE (regno, FIRST_GPR_REGNO, LAST_GPR_REGNO)
&& IN_RANGE (last_regno, FIRST_GPR_REGNO, LAST_GPR_REGNO)
&& ((regno & 1) == 0));
/* VSX registers that overlap the FPR registers are larger than for non-VSX
implementations. Don't allow an item to be split between a FP register
and an Altivec register. Allow TImode in all VSX registers if the user
asked for it. */
if (TARGET_VSX && VSX_REGNO_P (regno)
&& (VECTOR_MEM_VSX_P (mode)
|| FLOAT128_VECTOR_P (mode)
|| reg_addr[mode].scalar_in_vmx_p
|| mode == TImode
|| (TARGET_VADDUQM && mode == V1TImode)))
{
if (FP_REGNO_P (regno))
return FP_REGNO_P (last_regno);
if (ALTIVEC_REGNO_P (regno))
{
if (GET_MODE_SIZE (mode) != 16 && !reg_addr[mode].scalar_in_vmx_p)
return 0;
return ALTIVEC_REGNO_P (last_regno);
}
}
/* The GPRs can hold any mode, but values bigger than one register
cannot go past R31. */
if (INT_REGNO_P (regno))
return INT_REGNO_P (last_regno);
/* The float registers (except for VSX vector modes) can only hold floating
modes and DImode. */
if (FP_REGNO_P (regno))
{
if (FLOAT128_VECTOR_P (mode))
return false;
if (SCALAR_FLOAT_MODE_P (mode)
&& (mode != TDmode || (regno % 2) == 0)
&& FP_REGNO_P (last_regno))
return 1;
if (GET_MODE_CLASS (mode) == MODE_INT)
{
if(GET_MODE_SIZE (mode) == UNITS_PER_FP_WORD)
return 1;
if (TARGET_P8_VECTOR && (mode == SImode))
return 1;
if (TARGET_P9_VECTOR && (mode == QImode || mode == HImode))
return 1;
}
if (PAIRED_SIMD_REGNO_P (regno) && TARGET_PAIRED_FLOAT
&& PAIRED_VECTOR_MODE (mode))
return 1;
return 0;
}
/* The CR register can only hold CC modes. */
if (CR_REGNO_P (regno))
return GET_MODE_CLASS (mode) == MODE_CC;
if (CA_REGNO_P (regno))
return mode == Pmode || mode == SImode;
/* AltiVec only in AldyVec registers. */
if (ALTIVEC_REGNO_P (regno))
return (VECTOR_MEM_ALTIVEC_OR_VSX_P (mode)
|| mode == V1TImode);
/* We cannot put non-VSX TImode or PTImode anywhere except general register
and it must be able to fit within the register set. */
return GET_MODE_SIZE (mode) <= UNITS_PER_WORD;
}
/* Implement TARGET_HARD_REGNO_NREGS. */
static unsigned int
rs6000_hard_regno_nregs_hook (unsigned int regno, machine_mode mode)
{
return rs6000_hard_regno_nregs[mode][regno];
}
/* Implement TARGET_HARD_REGNO_MODE_OK. */
static bool
rs6000_hard_regno_mode_ok (unsigned int regno, machine_mode mode)
{
return rs6000_hard_regno_mode_ok_p[mode][regno];
}
/* Implement TARGET_MODES_TIEABLE_P.
PTImode cannot tie with other modes because PTImode is restricted to even
GPR registers, and TImode can go in any GPR as well as VSX registers (PR
57744).
Altivec/VSX vector tests were moved ahead of scalar float mode, so that IEEE
128-bit floating point on VSX systems ties with other vectors. */
static bool
rs6000_modes_tieable_p (machine_mode mode1, machine_mode mode2)
{
if (mode1 == PTImode)
return mode2 == PTImode;
if (mode2 == PTImode)
return false;
if (ALTIVEC_OR_VSX_VECTOR_MODE (mode1))
return ALTIVEC_OR_VSX_VECTOR_MODE (mode2);
if (ALTIVEC_OR_VSX_VECTOR_MODE (mode2))
return false;
if (SCALAR_FLOAT_MODE_P (mode1))
return SCALAR_FLOAT_MODE_P (mode2);
if (SCALAR_FLOAT_MODE_P (mode2))
return false;
if (GET_MODE_CLASS (mode1) == MODE_CC)
return GET_MODE_CLASS (mode2) == MODE_CC;
if (GET_MODE_CLASS (mode2) == MODE_CC)
return false;
if (PAIRED_VECTOR_MODE (mode1))
return PAIRED_VECTOR_MODE (mode2);
if (PAIRED_VECTOR_MODE (mode2))
return false;
return true;
}
/* Implement TARGET_HARD_REGNO_CALL_PART_CLOBBERED. */
static bool
rs6000_hard_regno_call_part_clobbered (unsigned int regno, machine_mode mode)
{
if (TARGET_32BIT
&& TARGET_POWERPC64
&& GET_MODE_SIZE (mode) > 4
&& INT_REGNO_P (regno))
return true;
if (TARGET_VSX
&& FP_REGNO_P (regno)
&& GET_MODE_SIZE (mode) > 8
&& !FLOAT128_2REG_P (mode))
return true;
return false;
}
/* Print interesting facts about registers. */
static void
rs6000_debug_reg_print (int first_regno, int last_regno, const char *reg_name)
{
int r, m;
for (r = first_regno; r <= last_regno; ++r)
{
const char *comma = "";
int len;
if (first_regno == last_regno)
fprintf (stderr, "%s:\t", reg_name);
else
fprintf (stderr, "%s%d:\t", reg_name, r - first_regno);
len = 8;
for (m = 0; m < NUM_MACHINE_MODES; ++m)
if (rs6000_hard_regno_mode_ok_p[m][r] && rs6000_hard_regno_nregs[m][r])
{
if (len > 70)
{
fprintf (stderr, ",\n\t");
len = 8;
comma = "";
}
if (rs6000_hard_regno_nregs[m][r] > 1)
len += fprintf (stderr, "%s%s/%d", comma, GET_MODE_NAME (m),
rs6000_hard_regno_nregs[m][r]);
else
len += fprintf (stderr, "%s%s", comma, GET_MODE_NAME (m));
comma = ", ";
}
if (call_used_regs[r])
{
if (len > 70)
{
fprintf (stderr, ",\n\t");
len = 8;
comma = "";
}
len += fprintf (stderr, "%s%s", comma, "call-used");
comma = ", ";
}
if (fixed_regs[r])
{
if (len > 70)
{
fprintf (stderr, ",\n\t");
len = 8;
comma = "";
}
len += fprintf (stderr, "%s%s", comma, "fixed");
comma = ", ";
}
if (len > 70)
{
fprintf (stderr, ",\n\t");
comma = "";
}
len += fprintf (stderr, "%sreg-class = %s", comma,
reg_class_names[(int)rs6000_regno_regclass[r]]);
comma = ", ";
if (len > 70)
{
fprintf (stderr, ",\n\t");
comma = "";
}
fprintf (stderr, "%sregno = %d\n", comma, r);
}
}
static const char *
rs6000_debug_vector_unit (enum rs6000_vector v)
{
const char *ret;
switch (v)
{
case VECTOR_NONE: ret = "none"; break;
case VECTOR_ALTIVEC: ret = "altivec"; break;
case VECTOR_VSX: ret = "vsx"; break;
case VECTOR_P8_VECTOR: ret = "p8_vector"; break;
case VECTOR_PAIRED: ret = "paired"; break;
case VECTOR_OTHER: ret = "other"; break;
default: ret = "unknown"; break;
}
return ret;
}
/* Inner function printing just the address mask for a particular reload
register class. */
DEBUG_FUNCTION char *
rs6000_debug_addr_mask (addr_mask_type mask, bool keep_spaces)
{
static char ret[8];
char *p = ret;
if ((mask & RELOAD_REG_VALID) != 0)
*p++ = 'v';
else if (keep_spaces)
*p++ = ' ';
if ((mask & RELOAD_REG_MULTIPLE) != 0)
*p++ = 'm';
else if (keep_spaces)
*p++ = ' ';
if ((mask & RELOAD_REG_INDEXED) != 0)
*p++ = 'i';
else if (keep_spaces)
*p++ = ' ';
if ((mask & RELOAD_REG_QUAD_OFFSET) != 0)
*p++ = 'O';
else if ((mask & RELOAD_REG_OFFSET) != 0)
*p++ = 'o';
else if (keep_spaces)
*p++ = ' ';
if ((mask & RELOAD_REG_PRE_INCDEC) != 0)
*p++ = '+';
else if (keep_spaces)
*p++ = ' ';
if ((mask & RELOAD_REG_PRE_MODIFY) != 0)
*p++ = '+';
else if (keep_spaces)
*p++ = ' ';
if ((mask & RELOAD_REG_AND_M16) != 0)
*p++ = '&';
else if (keep_spaces)
*p++ = ' ';
*p = '\0';
return ret;
}
/* Print the address masks in a human readble fashion. */
DEBUG_FUNCTION void
rs6000_debug_print_mode (ssize_t m)
{
ssize_t rc;
int spaces = 0;
bool fuse_extra_p;
fprintf (stderr, "Mode: %-5s", GET_MODE_NAME (m));
for (rc = 0; rc < N_RELOAD_REG; rc++)
fprintf (stderr, " %s: %s", reload_reg_map[rc].name,
rs6000_debug_addr_mask (reg_addr[m].addr_mask[rc], true));
if ((reg_addr[m].reload_store != CODE_FOR_nothing)
|| (reg_addr[m].reload_load != CODE_FOR_nothing))
fprintf (stderr, " Reload=%c%c",
(reg_addr[m].reload_store != CODE_FOR_nothing) ? 's' : '*',
(reg_addr[m].reload_load != CODE_FOR_nothing) ? 'l' : '*');
else
spaces += sizeof (" Reload=sl") - 1;
if (reg_addr[m].scalar_in_vmx_p)
{
fprintf (stderr, "%*s Upper=y", spaces, "");
spaces = 0;
}
else
spaces += sizeof (" Upper=y") - 1;
fuse_extra_p = ((reg_addr[m].fusion_gpr_ld != CODE_FOR_nothing)
|| reg_addr[m].fused_toc);
if (!fuse_extra_p)
{
for (rc = 0; rc < N_RELOAD_REG; rc++)
{
if (rc != RELOAD_REG_ANY)
{
if (reg_addr[m].fusion_addi_ld[rc] != CODE_FOR_nothing
|| reg_addr[m].fusion_addi_ld[rc] != CODE_FOR_nothing
|| reg_addr[m].fusion_addi_st[rc] != CODE_FOR_nothing
|| reg_addr[m].fusion_addis_ld[rc] != CODE_FOR_nothing
|| reg_addr[m].fusion_addis_st[rc] != CODE_FOR_nothing)
{
fuse_extra_p = true;
break;
}
}
}
}
if (fuse_extra_p)
{
fprintf (stderr, "%*s Fuse:", spaces, "");
spaces = 0;
for (rc = 0; rc < N_RELOAD_REG; rc++)
{
if (rc != RELOAD_REG_ANY)
{
char load, store;
if (reg_addr[m].fusion_addis_ld[rc] != CODE_FOR_nothing)
load = 'l';
else if (reg_addr[m].fusion_addi_ld[rc] != CODE_FOR_nothing)
load = 'L';
else
load = '-';
if (reg_addr[m].fusion_addis_st[rc] != CODE_FOR_nothing)
store = 's';
else if (reg_addr[m].fusion_addi_st[rc] != CODE_FOR_nothing)
store = 'S';
else
store = '-';
if (load == '-' && store == '-')
spaces += 5;
else
{
fprintf (stderr, "%*s%c=%c%c", (spaces + 1), "",
reload_reg_map[rc].name[0], load, store);
spaces = 0;
}
}
}
if (reg_addr[m].fusion_gpr_ld != CODE_FOR_nothing)
{
fprintf (stderr, "%*sP8gpr", (spaces + 1), "");
spaces = 0;
}
else
spaces += sizeof (" P8gpr") - 1;
if (reg_addr[m].fused_toc)
{
fprintf (stderr, "%*sToc", (spaces + 1), "");
spaces = 0;
}
else
spaces += sizeof (" Toc") - 1;
}
else
spaces += sizeof (" Fuse: G=ls F=ls v=ls P8gpr Toc") - 1;
if (rs6000_vector_unit[m] != VECTOR_NONE
|| rs6000_vector_mem[m] != VECTOR_NONE)
{
fprintf (stderr, "%*s vector: arith=%-10s mem=%s",
spaces, "",
rs6000_debug_vector_unit (rs6000_vector_unit[m]),
rs6000_debug_vector_unit (rs6000_vector_mem[m]));
}
fputs ("\n", stderr);
}
#define DEBUG_FMT_ID "%-32s= "
#define DEBUG_FMT_D DEBUG_FMT_ID "%d\n"
#define DEBUG_FMT_WX DEBUG_FMT_ID "%#.12" HOST_WIDE_INT_PRINT "x: "
#define DEBUG_FMT_S DEBUG_FMT_ID "%s\n"
/* Print various interesting information with -mdebug=reg. */
static void
rs6000_debug_reg_global (void)
{
static const char *const tf[2] = { "false", "true" };
const char *nl = (const char *)0;
int m;
size_t m1, m2, v;
char costly_num[20];
char nop_num[20];
char flags_buffer[40];
const char *costly_str;
const char *nop_str;
const char *trace_str;
const char *abi_str;
const char *cmodel_str;
struct cl_target_option cl_opts;
/* Modes we want tieable information on. */
static const machine_mode print_tieable_modes[] = {
QImode,
HImode,
SImode,
DImode,
TImode,
PTImode,
SFmode,
DFmode,
TFmode,
IFmode,
KFmode,
SDmode,
DDmode,
TDmode,
V2SImode,
V16QImode,
V8HImode,
V4SImode,
V2DImode,
V1TImode,
V32QImode,
V16HImode,
V8SImode,
V4DImode,
V2TImode,
V2SFmode,
V4SFmode,
V2DFmode,
V8SFmode,
V4DFmode,
CCmode,
CCUNSmode,
CCEQmode,
};
/* Virtual regs we are interested in. */
const static struct {
int regno; /* register number. */
const char *name; /* register name. */
} virtual_regs[] = {
{ STACK_POINTER_REGNUM, "stack pointer:" },
{ TOC_REGNUM, "toc: " },
{ STATIC_CHAIN_REGNUM, "static chain: " },
{ RS6000_PIC_OFFSET_TABLE_REGNUM, "pic offset: " },
{ HARD_FRAME_POINTER_REGNUM, "hard frame: " },
{ ARG_POINTER_REGNUM, "arg pointer: " },
{ FRAME_POINTER_REGNUM, "frame pointer:" },
{ FIRST_PSEUDO_REGISTER, "first pseudo: " },
{ FIRST_VIRTUAL_REGISTER, "first virtual:" },
{ VIRTUAL_INCOMING_ARGS_REGNUM, "incoming_args:" },
{ VIRTUAL_STACK_VARS_REGNUM, "stack_vars: " },
{ VIRTUAL_STACK_DYNAMIC_REGNUM, "stack_dynamic:" },
{ VIRTUAL_OUTGOING_ARGS_REGNUM, "outgoing_args:" },
{ VIRTUAL_CFA_REGNUM, "cfa (frame): " },
{ VIRTUAL_PREFERRED_STACK_BOUNDARY_REGNUM, "stack boundry:" },
{ LAST_VIRTUAL_REGISTER, "last virtual: " },
};
fputs ("\nHard register information:\n", stderr);
rs6000_debug_reg_print (FIRST_GPR_REGNO, LAST_GPR_REGNO, "gr");
rs6000_debug_reg_print (FIRST_FPR_REGNO, LAST_FPR_REGNO, "fp");
rs6000_debug_reg_print (FIRST_ALTIVEC_REGNO,
LAST_ALTIVEC_REGNO,
"vs");
rs6000_debug_reg_print (LR_REGNO, LR_REGNO, "lr");
rs6000_debug_reg_print (CTR_REGNO, CTR_REGNO, "ctr");
rs6000_debug_reg_print (CR0_REGNO, CR7_REGNO, "cr");
rs6000_debug_reg_print (CA_REGNO, CA_REGNO, "ca");
rs6000_debug_reg_print (VRSAVE_REGNO, VRSAVE_REGNO, "vrsave");
rs6000_debug_reg_print (VSCR_REGNO, VSCR_REGNO, "vscr");
fputs ("\nVirtual/stack/frame registers:\n", stderr);
for (v = 0; v < ARRAY_SIZE (virtual_regs); v++)
fprintf (stderr, "%s regno = %3d\n", virtual_regs[v].name, virtual_regs[v].regno);
fprintf (stderr,
"\n"
"d reg_class = %s\n"
"f reg_class = %s\n"
"v reg_class = %s\n"
"wa reg_class = %s\n"
"wb reg_class = %s\n"
"wd reg_class = %s\n"
"we reg_class = %s\n"
"wf reg_class = %s\n"
"wg reg_class = %s\n"
"wh reg_class = %s\n"
"wi reg_class = %s\n"
"wj reg_class = %s\n"
"wk reg_class = %s\n"
"wl reg_class = %s\n"
"wm reg_class = %s\n"
"wo reg_class = %s\n"
"wp reg_class = %s\n"
"wq reg_class = %s\n"
"wr reg_class = %s\n"
"ws reg_class = %s\n"
"wt reg_class = %s\n"
"wu reg_class = %s\n"
"wv reg_class = %s\n"
"ww reg_class = %s\n"
"wx reg_class = %s\n"
"wy reg_class = %s\n"
"wz reg_class = %s\n"
"wA reg_class = %s\n"
"wH reg_class = %s\n"
"wI reg_class = %s\n"
"wJ reg_class = %s\n"
"wK reg_class = %s\n"
"\n",
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_d]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_f]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_v]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wa]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wb]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wd]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_we]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wf]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wg]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wh]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wi]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wj]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wk]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wl]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wm]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wo]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wp]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wq]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wr]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_ws]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wt]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wu]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wv]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_ww]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wx]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wy]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wz]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wA]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wH]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wI]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wJ]],
reg_class_names[rs6000_constraints[RS6000_CONSTRAINT_wK]]);
nl = "\n";
for (m = 0; m < NUM_MACHINE_MODES; ++m)
rs6000_debug_print_mode (m);
fputs ("\n", stderr);
for (m1 = 0; m1 < ARRAY_SIZE (print_tieable_modes); m1++)
{
machine_mode mode1 = print_tieable_modes[m1];
bool first_time = true;
nl = (const char *)0;
for (m2 = 0; m2 < ARRAY_SIZE (print_tieable_modes); m2++)
{
machine_mode mode2 = print_tieable_modes[m2];
if (mode1 != mode2 && rs6000_modes_tieable_p (mode1, mode2))
{
if (first_time)
{
fprintf (stderr, "Tieable modes %s:", GET_MODE_NAME (mode1));
nl = "\n";
first_time = false;
}
fprintf (stderr, " %s", GET_MODE_NAME (mode2));
}
}
if (!first_time)
fputs ("\n", stderr);
}
if (nl)
fputs (nl, stderr);
if (rs6000_recip_control)
{
fprintf (stderr, "\nReciprocal mask = 0x%x\n", rs6000_recip_control);
for (m = 0; m < NUM_MACHINE_MODES; ++m)
if (rs6000_recip_bits[m])
{
fprintf (stderr,
"Reciprocal estimate mode: %-5s divide: %s rsqrt: %s\n",
GET_MODE_NAME (m),
(RS6000_RECIP_AUTO_RE_P (m)
? "auto"
: (RS6000_RECIP_HAVE_RE_P (m) ? "have" : "none")),
(RS6000_RECIP_AUTO_RSQRTE_P (m)
? "auto"
: (RS6000_RECIP_HAVE_RSQRTE_P (m) ? "have" : "none")));
}
fputs ("\n", stderr);
}
if (rs6000_cpu_index >= 0)
{
const char *name = processor_target_table[rs6000_cpu_index].name;
HOST_WIDE_INT flags
= processor_target_table[rs6000_cpu_index].target_enable;
sprintf (flags_buffer, "-mcpu=%s flags", name);
rs6000_print_isa_options (stderr, 0, flags_buffer, flags);
}
else
fprintf (stderr, DEBUG_FMT_S, "cpu", "<none>");
if (rs6000_tune_index >= 0)
{
const char *name = processor_target_table[rs6000_tune_index].name;
HOST_WIDE_INT flags
= processor_target_table[rs6000_tune_index].target_enable;
sprintf (flags_buffer, "-mtune=%s flags", name);
rs6000_print_isa_options (stderr, 0, flags_buffer, flags);
}
else
fprintf (stderr, DEBUG_FMT_S, "tune", "<none>");
cl_target_option_save (&cl_opts, &global_options);
rs6000_print_isa_options (stderr, 0, "rs6000_isa_flags",
rs6000_isa_flags);
rs6000_print_isa_options (stderr, 0, "rs6000_isa_flags_explicit",
rs6000_isa_flags_explicit);
rs6000_print_builtin_options (stderr, 0, "rs6000_builtin_mask",
rs6000_builtin_mask);
rs6000_print_isa_options (stderr, 0, "TARGET_DEFAULT", TARGET_DEFAULT);
fprintf (stderr, DEBUG_FMT_S, "--with-cpu default",
OPTION_TARGET_CPU_DEFAULT ? OPTION_TARGET_CPU_DEFAULT : "<none>");
switch (rs6000_sched_costly_dep)
{
case max_dep_latency:
costly_str = "max_dep_latency";
break;
case no_dep_costly:
costly_str = "no_dep_costly";
break;
case all_deps_costly:
costly_str = "all_deps_costly";
break;
case true_store_to_load_dep_costly:
costly_str = "true_store_to_load_dep_costly";
break;
case store_to_load_dep_costly:
costly_str = "store_to_load_dep_costly";
break;
default:
costly_str = costly_num;
sprintf (costly_num, "%d", (int)rs6000_sched_costly_dep);
break;
}
fprintf (stderr, DEBUG_FMT_S, "sched_costly_dep", costly_str);
switch (rs6000_sched_insert_nops)
{
case sched_finish_regroup_exact:
nop_str = "sched_finish_regroup_exact";
break;
case sched_finish_pad_groups:
nop_str = "sched_finish_pad_groups";
break;
case sched_finish_none:
nop_str = "sched_finish_none";
break;
default:
nop_str = nop_num;
sprintf (nop_num, "%d", (int)rs6000_sched_insert_nops);
break;
}
fprintf (stderr, DEBUG_FMT_S, "sched_insert_nops", nop_str);
switch (rs6000_sdata)
{
default:
case SDATA_NONE:
break;
case SDATA_DATA:
fprintf (stderr, DEBUG_FMT_S, "sdata", "data");
break;
case SDATA_SYSV:
fprintf (stderr, DEBUG_FMT_S, "sdata", "sysv");
break;
case SDATA_EABI:
fprintf (stderr, DEBUG_FMT_S, "sdata", "eabi");
break;
}
switch (rs6000_traceback)
{
case traceback_default: trace_str = "default"; break;
case traceback_none: trace_str = "none"; break;
case traceback_part: trace_str = "part"; break;
case traceback_full: trace_str = "full"; break;
default: trace_str = "unknown"; break;
}
fprintf (stderr, DEBUG_FMT_S, "traceback", trace_str);
switch (rs6000_current_cmodel)
{
case CMODEL_SMALL: cmodel_str = "small"; break;
case CMODEL_MEDIUM: cmodel_str = "medium"; break;
case CMODEL_LARGE: cmodel_str = "large"; break;
default: cmodel_str = "unknown"; break;
}
fprintf (stderr, DEBUG_FMT_S, "cmodel", cmodel_str);
switch (rs6000_current_abi)
{
case ABI_NONE: abi_str = "none"; break;
case ABI_AIX: abi_str = "aix"; break;
case ABI_ELFv2: abi_str = "ELFv2"; break;
case ABI_V4: abi_str = "V4"; break;
case ABI_DARWIN: abi_str = "darwin"; break;
default: abi_str = "unknown"; break;
}
fprintf (stderr, DEBUG_FMT_S, "abi", abi_str);
if (rs6000_altivec_abi)
fprintf (stderr, DEBUG_FMT_S, "altivec_abi", "true");
if (rs6000_darwin64_abi)
fprintf (stderr, DEBUG_FMT_S, "darwin64_abi", "true");
fprintf (stderr, DEBUG_FMT_S, "single_float",
(TARGET_SINGLE_FLOAT ? "true" : "false"));
fprintf (stderr, DEBUG_FMT_S, "double_float",
(TARGET_DOUBLE_FLOAT ? "true" : "false"));
fprintf (stderr, DEBUG_FMT_S, "soft_float",
(TARGET_SOFT_FLOAT ? "true" : "false"));
if (TARGET_LINK_STACK)
fprintf (stderr, DEBUG_FMT_S, "link_stack", "true");
if (TARGET_P8_FUSION)
{
char options[80];
strcpy (options, (TARGET_P9_FUSION) ? "power9" : "power8");
if (TARGET_TOC_FUSION)
strcat (options, ", toc");
if (TARGET_P8_FUSION_SIGN)
strcat (options, ", sign");
fprintf (stderr, DEBUG_FMT_S, "fusion", options);
}
fprintf (stderr, DEBUG_FMT_S, "plt-format",
TARGET_SECURE_PLT ? "secure" : "bss");
fprintf (stderr, DEBUG_FMT_S, "struct-return",
aix_struct_return ? "aix" : "sysv");
fprintf (stderr, DEBUG_FMT_S, "always_hint", tf[!!rs6000_always_hint]);
fprintf (stderr, DEBUG_FMT_S, "sched_groups", tf[!!rs6000_sched_groups]);
fprintf (stderr, DEBUG_FMT_S, "align_branch",
tf[!!rs6000_align_branch_targets]);
fprintf (stderr, DEBUG_FMT_D, "tls_size", rs6000_tls_size);
fprintf (stderr, DEBUG_FMT_D, "long_double_size",
rs6000_long_double_type_size);
if (rs6000_long_double_type_size == 128)
{
fprintf (stderr, DEBUG_FMT_S, "long double type",
TARGET_IEEEQUAD ? "IEEE" : "IBM");
fprintf (stderr, DEBUG_FMT_S, "default long double type",
TARGET_IEEEQUAD_DEFAULT ? "IEEE" : "IBM");
}
fprintf (stderr, DEBUG_FMT_D, "sched_restricted_insns_priority",
(int)rs6000_sched_restricted_insns_priority);
fprintf (stderr, DEBUG_FMT_D, "Number of standard builtins",
(int)END_BUILTINS);
fprintf (stderr, DEBUG_FMT_D, "Number of rs6000 builtins",
(int)RS6000_BUILTIN_COUNT);
fprintf (stderr, DEBUG_FMT_D, "Enable float128 on VSX",
(int)TARGET_FLOAT128_ENABLE_TYPE);
if (TARGET_VSX)
fprintf (stderr, DEBUG_FMT_D, "VSX easy 64-bit scalar element",
(int)VECTOR_ELEMENT_SCALAR_64BIT);
if (TARGET_DIRECT_MOVE_128)
fprintf (stderr, DEBUG_FMT_D, "VSX easy 64-bit mfvsrld element",
(int)VECTOR_ELEMENT_MFVSRLD_64BIT);
}
/* Update the addr mask bits in reg_addr to help secondary reload and go if
legitimate address support to figure out the appropriate addressing to
use. */
static void
rs6000_setup_reg_addr_masks (void)
{
ssize_t rc, reg, m, nregs;
addr_mask_type any_addr_mask, addr_mask;
for (m = 0; m < NUM_MACHINE_MODES; ++m)
{
machine_mode m2 = (machine_mode) m;
bool complex_p = false;
bool small_int_p = (m2 == QImode || m2 == HImode || m2 == SImode);
size_t msize;
if (COMPLEX_MODE_P (m2))
{
complex_p = true;
m2 = GET_MODE_INNER (m2);
}
msize = GET_MODE_SIZE (m2);
/* SDmode is special in that we want to access it only via REG+REG
addressing on power7 and above, since we want to use the LFIWZX and
STFIWZX instructions to load it. */
bool indexed_only_p = (m == SDmode && TARGET_NO_SDMODE_STACK);
any_addr_mask = 0;
for (rc = FIRST_RELOAD_REG_CLASS; rc <= LAST_RELOAD_REG_CLASS; rc++)
{
addr_mask = 0;
reg = reload_reg_map[rc].reg;
/* Can mode values go in the GPR/FPR/Altivec registers? */
if (reg >= 0 && rs6000_hard_regno_mode_ok_p[m][reg])
{
bool small_int_vsx_p = (small_int_p
&& (rc == RELOAD_REG_FPR
|| rc == RELOAD_REG_VMX));
nregs = rs6000_hard_regno_nregs[m][reg];
addr_mask |= RELOAD_REG_VALID;
/* Indicate if the mode takes more than 1 physical register. If
it takes a single register, indicate it can do REG+REG
addressing. Small integers in VSX registers can only do
REG+REG addressing. */
if (small_int_vsx_p)
addr_mask |= RELOAD_REG_INDEXED;
else if (nregs > 1 || m == BLKmode || complex_p)
addr_mask |= RELOAD_REG_MULTIPLE;
else
addr_mask |= RELOAD_REG_INDEXED;
/* Figure out if we can do PRE_INC, PRE_DEC, or PRE_MODIFY
addressing. If we allow scalars into Altivec registers,
don't allow PRE_INC, PRE_DEC, or PRE_MODIFY. */
if (TARGET_UPDATE
&& (rc == RELOAD_REG_GPR || rc == RELOAD_REG_FPR)
&& msize <= 8
&& !VECTOR_MODE_P (m2)
&& !FLOAT128_VECTOR_P (m2)
&& !complex_p
&& !small_int_vsx_p)
{
addr_mask |= RELOAD_REG_PRE_INCDEC;
/* PRE_MODIFY is more restricted than PRE_INC/PRE_DEC in that
we don't allow PRE_MODIFY for some multi-register
operations. */
switch (m)
{
default:
addr_mask |= RELOAD_REG_PRE_MODIFY;
break;
case E_DImode:
if (TARGET_POWERPC64)
addr_mask |= RELOAD_REG_PRE_MODIFY;
break;
case E_DFmode:
case E_DDmode:
if (TARGET_DF_INSN)
addr_mask |= RELOAD_REG_PRE_MODIFY;
break;
}
}
}
/* GPR and FPR registers can do REG+OFFSET addressing, except
possibly for SDmode. ISA 3.0 (i.e. power9) adds D-form addressing
for 64-bit scalars and 32-bit SFmode to altivec registers. */
if ((addr_mask != 0) && !indexed_only_p
&& msize <= 8
&& (rc == RELOAD_REG_GPR
|| ((msize == 8 || m2 == SFmode)
&& (rc == RELOAD_REG_FPR
|| (rc == RELOAD_REG_VMX && TARGET_P9_VECTOR)))))
addr_mask |= RELOAD_REG_OFFSET;
/* VSX registers can do REG+OFFSET addresssing if ISA 3.0
instructions are enabled. The offset for 128-bit VSX registers is
only 12-bits. While GPRs can handle the full offset range, VSX
registers can only handle the restricted range. */
else if ((addr_mask != 0) && !indexed_only_p
&& msize == 16 && TARGET_P9_VECTOR
&& (ALTIVEC_OR_VSX_VECTOR_MODE (m2)
|| (m2 == TImode && TARGET_VSX)))
{
addr_mask |= RELOAD_REG_OFFSET;
if (rc == RELOAD_REG_FPR || rc == RELOAD_REG_VMX)
addr_mask |= RELOAD_REG_QUAD_OFFSET;
}
/* VMX registers can do (REG & -16) and ((REG+REG) & -16)
addressing on 128-bit types. */
if (rc == RELOAD_REG_VMX && msize == 16
&& (addr_mask & RELOAD_REG_VALID) != 0)
addr_mask |= RELOAD_REG_AND_M16;
reg_addr[m].addr_mask[rc] = addr_mask;
any_addr_mask |= addr_mask;
}
reg_addr[m].addr_mask[RELOAD_REG_ANY] = any_addr_mask;
}
}
/* Initialize the various global tables that are based on register size. */
static void
rs6000_init_hard_regno_mode_ok (bool global_init_p)
{
ssize_t r, m, c;
int align64;
int align32;
/* Precalculate REGNO_REG_CLASS. */
rs6000_regno_regclass[0] = GENERAL_REGS;
for (r = 1; r < 32; ++r)
rs6000_regno_regclass[r] = BASE_REGS;
for (r = 32; r < 64; ++r)
rs6000_regno_regclass[r] = FLOAT_REGS;
for (r = 64; r < FIRST_PSEUDO_REGISTER; ++r)
rs6000_regno_regclass[r] = NO_REGS;
for (r = FIRST_ALTIVEC_REGNO; r <= LAST_ALTIVEC_REGNO; ++r)
rs6000_regno_regclass[r] = ALTIVEC_REGS;
rs6000_regno_regclass[CR0_REGNO] = CR0_REGS;
for (r = CR1_REGNO; r <= CR7_REGNO; ++r)
rs6000_regno_regclass[r] = CR_REGS;
rs6000_regno_regclass[LR_REGNO] = LINK_REGS;
rs6000_regno_regclass[CTR_REGNO] = CTR_REGS;
rs6000_regno_regclass[CA_REGNO] = NO_REGS;
rs6000_regno_regclass[VRSAVE_REGNO] = VRSAVE_REGS;
rs6000_regno_regclass[VSCR_REGNO] = VRSAVE_REGS;
rs6000_regno_regclass[TFHAR_REGNO] = SPR_REGS;
rs6000_regno_regclass[TFIAR_REGNO] = SPR_REGS;
rs6000_regno_regclass[TEXASR_REGNO] = SPR_REGS;
rs6000_regno_regclass[ARG_POINTER_REGNUM] = BASE_REGS;
rs6000_regno_regclass[FRAME_POINTER_REGNUM] = BASE_REGS;
/* Precalculate register class to simpler reload register class. We don't
need all of the register classes that are combinations of different
classes, just the simple ones that have constraint letters. */
for (c = 0; c < N_REG_CLASSES; c++)
reg_class_to_reg_type[c] = NO_REG_TYPE;
reg_class_to_reg_type[(int)GENERAL_REGS] = GPR_REG_TYPE;
reg_class_to_reg_type[(int)BASE_REGS] = GPR_REG_TYPE;
reg_class_to_reg_type[(int)VSX_REGS] = VSX_REG_TYPE;
reg_class_to_reg_type[(int)VRSAVE_REGS] = SPR_REG_TYPE;
reg_class_to_reg_type[(int)VSCR_REGS] = SPR_REG_TYPE;
reg_class_to_reg_type[(int)LINK_REGS] = SPR_REG_TYPE;
reg_class_to_reg_type[(int)CTR_REGS] = SPR_REG_TYPE;
reg_class_to_reg_type[(int)LINK_OR_CTR_REGS] = SPR_REG_TYPE;
reg_class_to_reg_type[(int)CR_REGS] = CR_REG_TYPE;
reg_class_to_reg_type[(int)CR0_REGS] = CR_REG_TYPE;
if (TARGET_VSX)
{
reg_class_to_reg_type[(int)FLOAT_REGS] = VSX_REG_TYPE;
reg_class_to_reg_type[(int)ALTIVEC_REGS] = VSX_REG_TYPE;
}
else
{
reg_class_to_reg_type[(int)FLOAT_REGS] = FPR_REG_TYPE;
reg_class_to_reg_type[(int)ALTIVEC_REGS] = ALTIVEC_REG_TYPE;
}
/* Precalculate the valid memory formats as well as the vector information,
this must be set up before the rs6000_hard_regno_nregs_internal calls
below. */
gcc_assert ((int)VECTOR_NONE == 0);
memset ((void *) &rs6000_vector_unit[0], '\0', sizeof (rs6000_vector_unit));
memset ((void *) &rs6000_vector_mem[0], '\0', sizeof (rs6000_vector_unit));
gcc_assert ((int)CODE_FOR_nothing == 0);
memset ((void *) ®_addr[0], '\0', sizeof (reg_addr));
gcc_assert ((int)NO_REGS == 0);
memset ((void *) &rs6000_constraints[0], '\0', sizeof (rs6000_constraints));
/* The VSX hardware allows native alignment for vectors, but control whether the compiler
believes it can use native alignment or still uses 128-bit alignment. */
if (TARGET_VSX && !TARGET_VSX_ALIGN_128)
{
align64 = 64;
align32 = 32;
}
else
{
align64 = 128;
align32 = 128;
}
/* KF mode (IEEE 128-bit in VSX registers). We do not have arithmetic, so
only set the memory modes. Include TFmode if -mabi=ieeelongdouble. */
if (TARGET_FLOAT128_TYPE)
{
rs6000_vector_mem[KFmode] = VECTOR_VSX;
rs6000_vector_align[KFmode] = 128;
if (FLOAT128_IEEE_P (TFmode))
{
rs6000_vector_mem[TFmode] = VECTOR_VSX;
rs6000_vector_align[TFmode] = 128;
}
}
/* V2DF mode, VSX only. */
if (TARGET_VSX)
{
rs6000_vector_unit[V2DFmode] = VECTOR_VSX;
rs6000_vector_mem[V2DFmode] = VECTOR_VSX;
rs6000_vector_align[V2DFmode] = align64;
}
/* V4SF mode, either VSX or Altivec. */
if (TARGET_VSX)
{
rs6000_vector_unit[V4SFmode] = VECTOR_VSX;
rs6000_vector_mem[V4SFmode] = VECTOR_VSX;
rs6000_vector_align[V4SFmode] = align32;
}
else if (TARGET_ALTIVEC)
{
rs6000_vector_unit[V4SFmode] = VECTOR_ALTIVEC;
rs6000_vector_mem[V4SFmode] = VECTOR_ALTIVEC;
rs6000_vector_align[V4SFmode] = align32;
}
/* V16QImode, V8HImode, V4SImode are Altivec only, but possibly do VSX loads
and stores. */
if (TARGET_ALTIVEC)
{
rs6000_vector_unit[V4SImode] = VECTOR_ALTIVEC;
rs6000_vector_unit[V8HImode] = VECTOR_ALTIVEC;
rs6000_vector_unit[V16QImode] = VECTOR_ALTIVEC;
rs6000_vector_align[V4SImode] = align32;
rs6000_vector_align[V8HImode] = align32;
rs6000_vector_align[V16QImode] = align32;
if (TARGET_VSX)
{
rs6000_vector_mem[V4SImode] = VECTOR_VSX;
rs6000_vector_mem[V8HImode] = VECTOR_VSX;
rs6000_vector_mem[V16QImode] = VECTOR_VSX;
}
else
{
rs6000_vector_mem[V4SImode] = VECTOR_ALTIVEC;
rs6000_vector_mem[V8HImode] = VECTOR_ALTIVEC;
rs6000_vector_mem[V16QImode] = VECTOR_ALTIVEC;
}
}
/* V2DImode, full mode depends on ISA 2.07 vector mode. Allow under VSX to
do insert/splat/extract. Altivec doesn't have 64-bit integer support. */
if (TARGET_VSX)
{
rs6000_vector_mem[V2DImode] = VECTOR_VSX;
rs6000_vector_unit[V2DImode]
= (TARGET_P8_VECTOR) ? VECTOR_P8_VECTOR : VECTOR_NONE;
rs6000_vector_align[V2DImode] = align64;
rs6000_vector_mem[V1TImode] = VECTOR_VSX;
rs6000_vector_unit[V1TImode]
= (TARGET_P8_VECTOR) ? VECTOR_P8_VECTOR : VECTOR_NONE;
rs6000_vector_align[V1TImode] = 128;
}
/* DFmode, see if we want to use the VSX unit. Memory is handled
differently, so don't set rs6000_vector_mem. */
if (TARGET_VSX)
{
rs6000_vector_unit[DFmode] = VECTOR_VSX;
rs6000_vector_align[DFmode] = 64;
}
/* SFmode, see if we want to use the VSX unit. */
if (TARGET_P8_VECTOR)
{
rs6000_vector_unit[SFmode] = VECTOR_VSX;
rs6000_vector_align[SFmode] = 32;
}
/* Allow TImode in VSX register and set the VSX memory macros. */
if (TARGET_VSX)
{
rs6000_vector_mem[TImode] = VECTOR_VSX;
rs6000_vector_align[TImode] = align64;
}
/* TODO add paired floating point vector support. */
/* Register class constraints for the constraints that depend on compile
switches. When the VSX code was added, different constraints were added
based on the type (DFmode, V2DFmode, V4SFmode). For the vector types, all
of the VSX registers are used. The register classes for scalar floating
point types is set, based on whether we allow that type into the upper
(Altivec) registers. GCC has register classes to target the Altivec
registers for load/store operations, to select using a VSX memory
operation instead of the traditional floating point operation. The
constraints are:
d - Register class to use with traditional DFmode instructions.
f - Register class to use with traditional SFmode instructions.
v - Altivec register.
wa - Any VSX register.
wc - Reserved to represent individual CR bits (used in LLVM).
wd - Preferred register class for V2DFmode.
wf - Preferred register class for V4SFmode.
wg - Float register for power6x move insns.
wh - FP register for direct move instructions.
wi - FP or VSX register to hold 64-bit integers for VSX insns.
wj - FP or VSX register to hold 64-bit integers for direct moves.
wk - FP or VSX register to hold 64-bit doubles for direct moves.
wl - Float register if we can do 32-bit signed int loads.
wm - VSX register for ISA 2.07 direct move operations.
wn - always NO_REGS.
wr - GPR if 64-bit mode is permitted.
ws - Register class to do ISA 2.06 DF operations.
wt - VSX register for TImode in VSX registers.
wu - Altivec register for ISA 2.07 VSX SF/SI load/stores.
wv - Altivec register for ISA 2.06 VSX DF/DI load/stores.
ww - Register class to do SF conversions in with VSX operations.
wx - Float register if we can do 32-bit int stores.
wy - Register class to do ISA 2.07 SF operations.
wz - Float register if we can do 32-bit unsigned int loads.
wH - Altivec register if SImode is allowed in VSX registers.
wI - VSX register if SImode is allowed in VSX registers.
wJ - VSX register if QImode/HImode are allowed in VSX registers.
wK - Altivec register if QImode/HImode are allowed in VSX registers. */
if (TARGET_HARD_FLOAT)
rs6000_constraints[RS6000_CONSTRAINT_f] = FLOAT_REGS; /* SFmode */
if (TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT)
rs6000_constraints[RS6000_CONSTRAINT_d] = FLOAT_REGS; /* DFmode */
if (TARGET_VSX)
{
rs6000_constraints[RS6000_CONSTRAINT_wa] = VSX_REGS;
rs6000_constraints[RS6000_CONSTRAINT_wd] = VSX_REGS; /* V2DFmode */
rs6000_constraints[RS6000_CONSTRAINT_wf] = VSX_REGS; /* V4SFmode */
rs6000_constraints[RS6000_CONSTRAINT_ws] = VSX_REGS; /* DFmode */
rs6000_constraints[RS6000_CONSTRAINT_wv] = ALTIVEC_REGS; /* DFmode */
rs6000_constraints[RS6000_CONSTRAINT_wi] = VSX_REGS; /* DImode */
rs6000_constraints[RS6000_CONSTRAINT_wt] = VSX_REGS; /* TImode */
}
/* Add conditional constraints based on various options, to allow us to
collapse multiple insn patterns. */
if (TARGET_ALTIVEC)
rs6000_constraints[RS6000_CONSTRAINT_v] = ALTIVEC_REGS;
if (TARGET_MFPGPR) /* DFmode */
rs6000_constraints[RS6000_CONSTRAINT_wg] = FLOAT_REGS;
if (TARGET_LFIWAX)
rs6000_constraints[RS6000_CONSTRAINT_wl] = FLOAT_REGS; /* DImode */
if (TARGET_DIRECT_MOVE)
{
rs6000_constraints[RS6000_CONSTRAINT_wh] = FLOAT_REGS;
rs6000_constraints[RS6000_CONSTRAINT_wj] /* DImode */
= rs6000_constraints[RS6000_CONSTRAINT_wi];
rs6000_constraints[RS6000_CONSTRAINT_wk] /* DFmode */
= rs6000_constraints[RS6000_CONSTRAINT_ws];
rs6000_constraints[RS6000_CONSTRAINT_wm] = VSX_REGS;
}
if (TARGET_POWERPC64)
{
rs6000_constraints[RS6000_CONSTRAINT_wr] = GENERAL_REGS;
rs6000_constraints[RS6000_CONSTRAINT_wA] = BASE_REGS;
}
if (TARGET_P8_VECTOR) /* SFmode */
{
rs6000_constraints[RS6000_CONSTRAINT_wu] = ALTIVEC_REGS;
rs6000_constraints[RS6000_CONSTRAINT_wy] = VSX_REGS;
rs6000_constraints[RS6000_CONSTRAINT_ww] = VSX_REGS;
}
else if (TARGET_VSX)
rs6000_constraints[RS6000_CONSTRAINT_ww] = FLOAT_REGS;
if (TARGET_STFIWX)
rs6000_constraints[RS6000_CONSTRAINT_wx] = FLOAT_REGS; /* DImode */
if (TARGET_LFIWZX)
rs6000_constraints[RS6000_CONSTRAINT_wz] = FLOAT_REGS; /* DImode */
if (TARGET_FLOAT128_TYPE)
{
rs6000_constraints[RS6000_CONSTRAINT_wq] = VSX_REGS; /* KFmode */
if (FLOAT128_IEEE_P (TFmode))
rs6000_constraints[RS6000_CONSTRAINT_wp] = VSX_REGS; /* TFmode */
}
if (TARGET_P9_VECTOR)
{
/* Support for new D-form instructions. */
rs6000_constraints[RS6000_CONSTRAINT_wb] = ALTIVEC_REGS;
/* Support for ISA 3.0 (power9) vectors. */
rs6000_constraints[RS6000_CONSTRAINT_wo] = VSX_REGS;
}
/* Support for new direct moves (ISA 3.0 + 64bit). */
if (TARGET_DIRECT_MOVE_128)
rs6000_constraints[RS6000_CONSTRAINT_we] = VSX_REGS;
/* Support small integers in VSX registers. */
if (TARGET_P8_VECTOR)
{
rs6000_constraints[RS6000_CONSTRAINT_wH] = ALTIVEC_REGS;
rs6000_constraints[RS6000_CONSTRAINT_wI] = FLOAT_REGS;
if (TARGET_P9_VECTOR)
{
rs6000_constraints[RS6000_CONSTRAINT_wJ] = FLOAT_REGS;
rs6000_constraints[RS6000_CONSTRAINT_wK] = ALTIVEC_REGS;
}
}
/* Set up the reload helper and direct move functions. */
if (TARGET_VSX || TARGET_ALTIVEC)
{
if (TARGET_64BIT)
{
reg_addr[V16QImode].reload_store = CODE_FOR_reload_v16qi_di_store;
reg_addr[V16QImode].reload_load = CODE_FOR_reload_v16qi_di_load;
reg_addr[V8HImode].reload_store = CODE_FOR_reload_v8hi_di_store;
reg_addr[V8HImode].reload_load = CODE_FOR_reload_v8hi_di_load;
reg_addr[V4SImode].reload_store = CODE_FOR_reload_v4si_di_store;
reg_addr[V4SImode].reload_load = CODE_FOR_reload_v4si_di_load;
reg_addr[V2DImode].reload_store = CODE_FOR_reload_v2di_di_store;
reg_addr[V2DImode].reload_load = CODE_FOR_reload_v2di_di_load;
reg_addr[V1TImode].reload_store = CODE_FOR_reload_v1ti_di_store;
reg_addr[V1TImode].reload_load = CODE_FOR_reload_v1ti_di_load;
reg_addr[V4SFmode].reload_store = CODE_FOR_reload_v4sf_di_store;
reg_addr[V4SFmode].reload_load = CODE_FOR_reload_v4sf_di_load;
reg_addr[V2DFmode].reload_store = CODE_FOR_reload_v2df_di_store;
reg_addr[V2DFmode].reload_load = CODE_FOR_reload_v2df_di_load;
reg_addr[DFmode].reload_store = CODE_FOR_reload_df_di_store;
reg_addr[DFmode].reload_load = CODE_FOR_reload_df_di_load;
reg_addr[DDmode].reload_store = CODE_FOR_reload_dd_di_store;
reg_addr[DDmode].reload_load = CODE_FOR_reload_dd_di_load;
reg_addr[SFmode].reload_store = CODE_FOR_reload_sf_di_store;
reg_addr[SFmode].reload_load = CODE_FOR_reload_sf_di_load;
if (FLOAT128_VECTOR_P (KFmode))
{
reg_addr[KFmode].reload_store = CODE_FOR_reload_kf_di_store;
reg_addr[KFmode].reload_load = CODE_FOR_reload_kf_di_load;
}
if (FLOAT128_VECTOR_P (TFmode))
{
reg_addr[TFmode].reload_store = CODE_FOR_reload_tf_di_store;
reg_addr[TFmode].reload_load = CODE_FOR_reload_tf_di_load;
}
/* Only provide a reload handler for SDmode if lfiwzx/stfiwx are
available. */
if (TARGET_NO_SDMODE_STACK)
{
reg_addr[SDmode].reload_store = CODE_FOR_reload_sd_di_store;
reg_addr[SDmode].reload_load = CODE_FOR_reload_sd_di_load;
}
if (TARGET_VSX)
{
reg_addr[TImode].reload_store = CODE_FOR_reload_ti_di_store;
reg_addr[TImode].reload_load = CODE_FOR_reload_ti_di_load;
}
if (TARGET_DIRECT_MOVE && !TARGET_DIRECT_MOVE_128)
{
reg_addr[TImode].reload_gpr_vsx = CODE_FOR_reload_gpr_from_vsxti;
reg_addr[V1TImode].reload_gpr_vsx = CODE_FOR_reload_gpr_from_vsxv1ti;
reg_addr[V2DFmode].reload_gpr_vsx = CODE_FOR_reload_gpr_from_vsxv2df;
reg_addr[V2DImode].reload_gpr_vsx = CODE_FOR_reload_gpr_from_vsxv2di;
reg_addr[V4SFmode].reload_gpr_vsx = CODE_FOR_reload_gpr_from_vsxv4sf;
reg_addr[V4SImode].reload_gpr_vsx = CODE_FOR_reload_gpr_from_vsxv4si;
reg_addr[V8HImode].reload_gpr_vsx = CODE_FOR_reload_gpr_from_vsxv8hi;
reg_addr[V16QImode].reload_gpr_vsx = CODE_FOR_reload_gpr_from_vsxv16qi;
reg_addr[SFmode].reload_gpr_vsx = CODE_FOR_reload_gpr_from_vsxsf;
reg_addr[TImode].reload_vsx_gpr = CODE_FOR_reload_vsx_from_gprti;
reg_addr[V1TImode].reload_vsx_gpr = CODE_FOR_reload_vsx_from_gprv1ti;
reg_addr[V2DFmode].reload_vsx_gpr = CODE_FOR_reload_vsx_from_gprv2df;
reg_addr[V2DImode].reload_vsx_gpr = CODE_FOR_reload_vsx_from_gprv2di;
reg_addr[V4SFmode].reload_vsx_gpr = CODE_FOR_reload_vsx_from_gprv4sf;
reg_addr[V4SImode].reload_vsx_gpr = CODE_FOR_reload_vsx_from_gprv4si;
reg_addr[V8HImode].reload_vsx_gpr = CODE_FOR_reload_vsx_from_gprv8hi;
reg_addr[V16QImode].reload_vsx_gpr = CODE_FOR_reload_vsx_from_gprv16qi;
reg_addr[SFmode].reload_vsx_gpr = CODE_FOR_reload_vsx_from_gprsf;
if (FLOAT128_VECTOR_P (KFmode))
{
reg_addr[KFmode].reload_gpr_vsx = CODE_FOR_reload_gpr_from_vsxkf;
reg_addr[KFmode].reload_vsx_gpr = CODE_FOR_reload_vsx_from_gprkf;
}
if (FLOAT128_VECTOR_P (TFmode))
{
reg_addr[TFmode].reload_gpr_vsx = CODE_FOR_reload_gpr_from_vsxtf;
reg_addr[TFmode].reload_vsx_gpr = CODE_FOR_reload_vsx_from_gprtf;
}
}
}
else
{
reg_addr[V16QImode].reload_store = CODE_FOR_reload_v16qi_si_store;
reg_addr[V16QImode].reload_load = CODE_FOR_reload_v16qi_si_load;
reg_addr[V8HImode].reload_store = CODE_FOR_reload_v8hi_si_store;
reg_addr[V8HImode].reload_load = CODE_FOR_reload_v8hi_si_load;
reg_addr[V4SImode].reload_store = CODE_FOR_reload_v4si_si_store;
reg_addr[V4SImode].reload_load = CODE_FOR_reload_v4si_si_load;
reg_addr[V2DImode].reload_store = CODE_FOR_reload_v2di_si_store;
reg_addr[V2DImode].reload_load = CODE_FOR_reload_v2di_si_load;
reg_addr[V1TImode].reload_store = CODE_FOR_reload_v1ti_si_store;
reg_addr[V1TImode].reload_load = CODE_FOR_reload_v1ti_si_load;
reg_addr[V4SFmode].reload_store = CODE_FOR_reload_v4sf_si_store;
reg_addr[V4SFmode].reload_load = CODE_FOR_reload_v4sf_si_load;
reg_addr[V2DFmode].reload_store = CODE_FOR_reload_v2df_si_store;
reg_addr[V2DFmode].reload_load = CODE_FOR_reload_v2df_si_load;
reg_addr[DFmode].reload_store = CODE_FOR_reload_df_si_store;
reg_addr[DFmode].reload_load = CODE_FOR_reload_df_si_load;
reg_addr[DDmode].reload_store = CODE_FOR_reload_dd_si_store;
reg_addr[DDmode].reload_load = CODE_FOR_reload_dd_si_load;
reg_addr[SFmode].reload_store = CODE_FOR_reload_sf_si_store;
reg_addr[SFmode].reload_load = CODE_FOR_reload_sf_si_load;
if (FLOAT128_VECTOR_P (KFmode))
{
reg_addr[KFmode].reload_store = CODE_FOR_reload_kf_si_store;
reg_addr[KFmode].reload_load = CODE_FOR_reload_kf_si_load;
}
if (FLOAT128_IEEE_P (TFmode))
{
reg_addr[TFmode].reload_store = CODE_FOR_reload_tf_si_store;
reg_addr[TFmode].reload_load = CODE_FOR_reload_tf_si_load;
}
/* Only provide a reload handler for SDmode if lfiwzx/stfiwx are
available. */
if (TARGET_NO_SDMODE_STACK)
{
reg_addr[SDmode].reload_store = CODE_FOR_reload_sd_si_store;
reg_addr[SDmode].reload_load = CODE_FOR_reload_sd_si_load;
}
if (TARGET_VSX)
{
reg_addr[TImode].reload_store = CODE_FOR_reload_ti_si_store;
reg_addr[TImode].reload_load = CODE_FOR_reload_ti_si_load;
}
if (TARGET_DIRECT_MOVE)
{
reg_addr[DImode].reload_fpr_gpr = CODE_FOR_reload_fpr_from_gprdi;
reg_addr[DDmode].reload_fpr_gpr = CODE_FOR_reload_fpr_from_gprdd;
reg_addr[DFmode].reload_fpr_gpr = CODE_FOR_reload_fpr_from_gprdf;
}
}
reg_addr[DFmode].scalar_in_vmx_p = true;
reg_addr[DImode].scalar_in_vmx_p = true;
if (TARGET_P8_VECTOR)
{
reg_addr[SFmode].scalar_in_vmx_p = true;
reg_addr[SImode].scalar_in_vmx_p = true;
if (TARGET_P9_VECTOR)
{
reg_addr[HImode].scalar_in_vmx_p = true;
reg_addr[QImode].scalar_in_vmx_p = true;
}
}
}
/* Setup the fusion operations. */
if (TARGET_P8_FUSION)
{
reg_addr[QImode].fusion_gpr_ld = CODE_FOR_fusion_gpr_load_qi;
reg_addr[HImode].fusion_gpr_ld = CODE_FOR_fusion_gpr_load_hi;
reg_addr[SImode].fusion_gpr_ld = CODE_FOR_fusion_gpr_load_si;
if (TARGET_64BIT)
reg_addr[DImode].fusion_gpr_ld = CODE_FOR_fusion_gpr_load_di;
}
if (TARGET_P9_FUSION)
{
struct fuse_insns {
enum machine_mode mode; /* mode of the fused type. */
enum machine_mode pmode; /* pointer mode. */
enum rs6000_reload_reg_type rtype; /* register type. */
enum insn_code load; /* load insn. */
enum insn_code store; /* store insn. */
};
static const struct fuse_insns addis_insns[] = {
{ E_SFmode, E_DImode, RELOAD_REG_FPR,
CODE_FOR_fusion_vsx_di_sf_load,
CODE_FOR_fusion_vsx_di_sf_store },
{ E_SFmode, E_SImode, RELOAD_REG_FPR,
CODE_FOR_fusion_vsx_si_sf_load,
CODE_FOR_fusion_vsx_si_sf_store },
{ E_DFmode, E_DImode, RELOAD_REG_FPR,
CODE_FOR_fusion_vsx_di_df_load,
CODE_FOR_fusion_vsx_di_df_store },
{ E_DFmode, E_SImode, RELOAD_REG_FPR,
CODE_FOR_fusion_vsx_si_df_load,
CODE_FOR_fusion_vsx_si_df_store },
{ E_DImode, E_DImode, RELOAD_REG_FPR,
CODE_FOR_fusion_vsx_di_di_load,
CODE_FOR_fusion_vsx_di_di_store },
{ E_DImode, E_SImode, RELOAD_REG_FPR,
CODE_FOR_fusion_vsx_si_di_load,
CODE_FOR_fusion_vsx_si_di_store },
{ E_QImode, E_DImode, RELOAD_REG_GPR,
CODE_FOR_fusion_gpr_di_qi_load,
CODE_FOR_fusion_gpr_di_qi_store },
{ E_QImode, E_SImode, RELOAD_REG_GPR,
CODE_FOR_fusion_gpr_si_qi_load,
CODE_FOR_fusion_gpr_si_qi_store },
{ E_HImode, E_DImode, RELOAD_REG_GPR,
CODE_FOR_fusion_gpr_di_hi_load,
CODE_FOR_fusion_gpr_di_hi_store },
{ E_HImode, E_SImode, RELOAD_REG_GPR,
CODE_FOR_fusion_gpr_si_hi_load,
CODE_FOR_fusion_gpr_si_hi_store },
{ E_SImode, E_DImode, RELOAD_REG_GPR,
CODE_FOR_fusion_gpr_di_si_load,
CODE_FOR_fusion_gpr_di_si_store },
{ E_SImode, E_SImode, RELOAD_REG_GPR,
CODE_FOR_fusion_gpr_si_si_load,
CODE_FOR_fusion_gpr_si_si_store },
{ E_SFmode, E_DImode, RELOAD_REG_GPR,
CODE_FOR_fusion_gpr_di_sf_load,
CODE_FOR_fusion_gpr_di_sf_store },
{ E_SFmode, E_SImode, RELOAD_REG_GPR,
CODE_FOR_fusion_gpr_si_sf_load,
CODE_FOR_fusion_gpr_si_sf_store },
{ E_DImode, E_DImode, RELOAD_REG_GPR,
CODE_FOR_fusion_gpr_di_di_load,
CODE_FOR_fusion_gpr_di_di_store },
{ E_DFmode, E_DImode, RELOAD_REG_GPR,
CODE_FOR_fusion_gpr_di_df_load,
CODE_FOR_fusion_gpr_di_df_store },
};
machine_mode cur_pmode = Pmode;
size_t i;
for (i = 0; i < ARRAY_SIZE (addis_insns); i++)
{
machine_mode xmode = addis_insns[i].mode;
enum rs6000_reload_reg_type rtype = addis_insns[i].rtype;
if (addis_insns[i].pmode != cur_pmode)
continue;
if (rtype == RELOAD_REG_FPR && !TARGET_HARD_FLOAT)
continue;
reg_addr[xmode].fusion_addis_ld[rtype] = addis_insns[i].load;
reg_addr[xmode].fusion_addis_st[rtype] = addis_insns[i].store;
if (rtype == RELOAD_REG_FPR && TARGET_P9_VECTOR)
{
reg_addr[xmode].fusion_addis_ld[RELOAD_REG_VMX]
= addis_insns[i].load;
reg_addr[xmode].fusion_addis_st[RELOAD_REG_VMX]
= addis_insns[i].store;
}
}
}
/* Note which types we support fusing TOC setup plus memory insn. We only do
fused TOCs for medium/large code models. */
if (TARGET_P8_FUSION && TARGET_TOC_FUSION && TARGET_POWERPC64
&& (TARGET_CMODEL != CMODEL_SMALL))
{
reg_addr[QImode].fused_toc = true;
reg_addr[HImode].fused_toc = true;
reg_addr[SImode].fused_toc = true;
reg_addr[DImode].fused_toc = true;
if (TARGET_HARD_FLOAT)
{
if (TARGET_SINGLE_FLOAT)
reg_addr[SFmode].fused_toc = true;
if (TARGET_DOUBLE_FLOAT)
reg_addr[DFmode].fused_toc = true;
}
}
/* Precalculate HARD_REGNO_NREGS. */
for (r = 0; r < FIRST_PSEUDO_REGISTER; ++r)
for (m = 0; m < NUM_MACHINE_MODES; ++m)
rs6000_hard_regno_nregs[m][r]
= rs6000_hard_regno_nregs_internal (r, (machine_mode)m);
/* Precalculate TARGET_HARD_REGNO_MODE_OK. */
for (r = 0; r < FIRST_PSEUDO_REGISTER; ++r)
for (m = 0; m < NUM_MACHINE_MODES; ++m)
if (rs6000_hard_regno_mode_ok_uncached (r, (machine_mode)m))
rs6000_hard_regno_mode_ok_p[m][r] = true;
/* Precalculate CLASS_MAX_NREGS sizes. */
for (c = 0; c < LIM_REG_CLASSES; ++c)
{
int reg_size;
if (TARGET_VSX && VSX_REG_CLASS_P (c))
reg_size = UNITS_PER_VSX_WORD;
else if (c == ALTIVEC_REGS)
reg_size = UNITS_PER_ALTIVEC_WORD;
else if (c == FLOAT_REGS)
reg_size = UNITS_PER_FP_WORD;
else
reg_size = UNITS_PER_WORD;
for (m = 0; m < NUM_MACHINE_MODES; ++m)
{
machine_mode m2 = (machine_mode)m;
int reg_size2 = reg_size;
/* TDmode & IBM 128-bit floating point always takes 2 registers, even
in VSX. */
if (TARGET_VSX && VSX_REG_CLASS_P (c) && FLOAT128_2REG_P (m))
reg_size2 = UNITS_PER_FP_WORD;
rs6000_class_max_nregs[m][c]
= (GET_MODE_SIZE (m2) + reg_size2 - 1) / reg_size2;
}
}
/* Calculate which modes to automatically generate code to use a the
reciprocal divide and square root instructions. In the future, possibly
automatically generate the instructions even if the user did not specify
-mrecip. The older machines double precision reciprocal sqrt estimate is
not accurate enough. */
memset (rs6000_recip_bits, 0, sizeof (rs6000_recip_bits));
if (TARGET_FRES)
rs6000_recip_bits[SFmode] = RS6000_RECIP_MASK_HAVE_RE;
if (TARGET_FRE)
rs6000_recip_bits[DFmode] = RS6000_RECIP_MASK_HAVE_RE;
if (VECTOR_UNIT_ALTIVEC_OR_VSX_P (V4SFmode))
rs6000_recip_bits[V4SFmode] = RS6000_RECIP_MASK_HAVE_RE;
if (VECTOR_UNIT_VSX_P (V2DFmode))
rs6000_recip_bits[V2DFmode] = RS6000_RECIP_MASK_HAVE_RE;
if (TARGET_FRSQRTES)
rs6000_recip_bits[SFmode] |= RS6000_RECIP_MASK_HAVE_RSQRTE;
if (TARGET_FRSQRTE)
rs6000_recip_bits[DFmode] |= RS6000_RECIP_MASK_HAVE_RSQRTE;
if (VECTOR_UNIT_ALTIVEC_OR_VSX_P (V4SFmode))
rs6000_recip_bits[V4SFmode] |= RS6000_RECIP_MASK_HAVE_RSQRTE;
if (VECTOR_UNIT_VSX_P (V2DFmode))
rs6000_recip_bits[V2DFmode] |= RS6000_RECIP_MASK_HAVE_RSQRTE;
if (rs6000_recip_control)
{
if (!flag_finite_math_only)
warning (0, "%qs requires %qs or %qs", "-mrecip", "-ffinite-math",
"-ffast-math");
if (flag_trapping_math)
warning (0, "%qs requires %qs or %qs", "-mrecip",
"-fno-trapping-math", "-ffast-math");
if (!flag_reciprocal_math)
warning (0, "%qs requires %qs or %qs", "-mrecip", "-freciprocal-math",
"-ffast-math");
if (flag_finite_math_only && !flag_trapping_math && flag_reciprocal_math)
{
if (RS6000_RECIP_HAVE_RE_P (SFmode)
&& (rs6000_recip_control & RECIP_SF_DIV) != 0)
rs6000_recip_bits[SFmode] |= RS6000_RECIP_MASK_AUTO_RE;
if (RS6000_RECIP_HAVE_RE_P (DFmode)
&& (rs6000_recip_control & RECIP_DF_DIV) != 0)
rs6000_recip_bits[DFmode] |= RS6000_RECIP_MASK_AUTO_RE;
if (RS6000_RECIP_HAVE_RE_P (V4SFmode)
&& (rs6000_recip_control & RECIP_V4SF_DIV) != 0)
rs6000_recip_bits[V4SFmode] |= RS6000_RECIP_MASK_AUTO_RE;
if (RS6000_RECIP_HAVE_RE_P (V2DFmode)
&& (rs6000_recip_control & RECIP_V2DF_DIV) != 0)
rs6000_recip_bits[V2DFmode] |= RS6000_RECIP_MASK_AUTO_RE;
if (RS6000_RECIP_HAVE_RSQRTE_P (SFmode)
&& (rs6000_recip_control & RECIP_SF_RSQRT) != 0)
rs6000_recip_bits[SFmode] |= RS6000_RECIP_MASK_AUTO_RSQRTE;
if (RS6000_RECIP_HAVE_RSQRTE_P (DFmode)
&& (rs6000_recip_control & RECIP_DF_RSQRT) != 0)
rs6000_recip_bits[DFmode] |= RS6000_RECIP_MASK_AUTO_RSQRTE;
if (RS6000_RECIP_HAVE_RSQRTE_P (V4SFmode)
&& (rs6000_recip_control & RECIP_V4SF_RSQRT) != 0)
rs6000_recip_bits[V4SFmode] |= RS6000_RECIP_MASK_AUTO_RSQRTE;
if (RS6000_RECIP_HAVE_RSQRTE_P (V2DFmode)
&& (rs6000_recip_control & RECIP_V2DF_RSQRT) != 0)
rs6000_recip_bits[V2DFmode] |= RS6000_RECIP_MASK_AUTO_RSQRTE;
}
}
/* Update the addr mask bits in reg_addr to help secondary reload and go if
legitimate address support to figure out the appropriate addressing to
use. */
rs6000_setup_reg_addr_masks ();
if (global_init_p || TARGET_DEBUG_TARGET)
{
if (TARGET_DEBUG_REG)
rs6000_debug_reg_global ();
if (TARGET_DEBUG_COST || TARGET_DEBUG_REG)
fprintf (stderr,
"SImode variable mult cost = %d\n"
"SImode constant mult cost = %d\n"
"SImode short constant mult cost = %d\n"
"DImode multipliciation cost = %d\n"
"SImode division cost = %d\n"
"DImode division cost = %d\n"
"Simple fp operation cost = %d\n"
"DFmode multiplication cost = %d\n"
"SFmode division cost = %d\n"
"DFmode division cost = %d\n"
"cache line size = %d\n"
"l1 cache size = %d\n"
"l2 cache size = %d\n"
"simultaneous prefetches = %d\n"
"\n",
rs6000_cost->mulsi,
rs6000_cost->mulsi_const,
rs6000_cost->mulsi_const9,
rs6000_cost->muldi,
rs6000_cost->divsi,
rs6000_cost->divdi,
rs6000_cost->fp,
rs6000_cost->dmul,
rs6000_cost->sdiv,
rs6000_cost->ddiv,
rs6000_cost->cache_line_size,
rs6000_cost->l1_cache_size,
rs6000_cost->l2_cache_size,
rs6000_cost->simultaneous_prefetches);
}
}
#if TARGET_MACHO
/* The Darwin version of SUBTARGET_OVERRIDE_OPTIONS. */
static void
darwin_rs6000_override_options (void)
{
/* The Darwin ABI always includes AltiVec, can't be (validly) turned
off. */
rs6000_altivec_abi = 1;
TARGET_ALTIVEC_VRSAVE = 1;
rs6000_current_abi = ABI_DARWIN;
if (DEFAULT_ABI == ABI_DARWIN
&& TARGET_64BIT)
darwin_one_byte_bool = 1;
if (TARGET_64BIT && ! TARGET_POWERPC64)
{
rs6000_isa_flags |= OPTION_MASK_POWERPC64;
warning (0, "%qs requires PowerPC64 architecture, enabling", "-m64");
}
if (flag_mkernel)
{
rs6000_default_long_calls = 1;
rs6000_isa_flags |= OPTION_MASK_SOFT_FLOAT;
}
/* Make -m64 imply -maltivec. Darwin's 64-bit ABI includes
Altivec. */
if (!flag_mkernel && !flag_apple_kext
&& TARGET_64BIT
&& ! (rs6000_isa_flags_explicit & OPTION_MASK_ALTIVEC))
rs6000_isa_flags |= OPTION_MASK_ALTIVEC;
/* Unless the user (not the configurer) has explicitly overridden
it with -mcpu=G3 or -mno-altivec, then 10.5+ targets default to
G4 unless targeting the kernel. */
if (!flag_mkernel
&& !flag_apple_kext
&& strverscmp (darwin_macosx_version_min, "10.5") >= 0
&& ! (rs6000_isa_flags_explicit & OPTION_MASK_ALTIVEC)
&& ! global_options_set.x_rs6000_cpu_index)
{
rs6000_isa_flags |= OPTION_MASK_ALTIVEC;
}
}
#endif
/* If not otherwise specified by a target, make 'long double' equivalent to
'double'. */
#ifndef RS6000_DEFAULT_LONG_DOUBLE_SIZE
#define RS6000_DEFAULT_LONG_DOUBLE_SIZE 64
#endif
/* Return the builtin mask of the various options used that could affect which
builtins were used. In the past we used target_flags, but we've run out of
bits, and some options like PAIRED are no longer in target_flags. */
HOST_WIDE_INT
rs6000_builtin_mask_calculate (void)
{
return (((TARGET_ALTIVEC) ? RS6000_BTM_ALTIVEC : 0)
| ((TARGET_CMPB) ? RS6000_BTM_CMPB : 0)
| ((TARGET_VSX) ? RS6000_BTM_VSX : 0)
| ((TARGET_PAIRED_FLOAT) ? RS6000_BTM_PAIRED : 0)
| ((TARGET_FRE) ? RS6000_BTM_FRE : 0)
| ((TARGET_FRES) ? RS6000_BTM_FRES : 0)
| ((TARGET_FRSQRTE) ? RS6000_BTM_FRSQRTE : 0)
| ((TARGET_FRSQRTES) ? RS6000_BTM_FRSQRTES : 0)
| ((TARGET_POPCNTD) ? RS6000_BTM_POPCNTD : 0)
| ((rs6000_cpu == PROCESSOR_CELL) ? RS6000_BTM_CELL : 0)
| ((TARGET_P8_VECTOR) ? RS6000_BTM_P8_VECTOR : 0)
| ((TARGET_P9_VECTOR) ? RS6000_BTM_P9_VECTOR : 0)
| ((TARGET_P9_MISC) ? RS6000_BTM_P9_MISC : 0)
| ((TARGET_MODULO) ? RS6000_BTM_MODULO : 0)
| ((TARGET_64BIT) ? RS6000_BTM_64BIT : 0)
| ((TARGET_CRYPTO) ? RS6000_BTM_CRYPTO : 0)
| ((TARGET_HTM) ? RS6000_BTM_HTM : 0)
| ((TARGET_DFP) ? RS6000_BTM_DFP : 0)
| ((TARGET_HARD_FLOAT) ? RS6000_BTM_HARD_FLOAT : 0)
| ((TARGET_LONG_DOUBLE_128) ? RS6000_BTM_LDBL128 : 0)
| ((TARGET_FLOAT128_TYPE) ? RS6000_BTM_FLOAT128 : 0)
| ((TARGET_FLOAT128_HW) ? RS6000_BTM_FLOAT128_HW : 0));
}
/* Implement TARGET_MD_ASM_ADJUST. All asm statements are considered
to clobber the XER[CA] bit because clobbering that bit without telling
the compiler worked just fine with versions of GCC before GCC 5, and
breaking a lot of older code in ways that are hard to track down is
not such a great idea. */
static rtx_insn *
rs6000_md_asm_adjust (vec<rtx> &/*outputs*/, vec<rtx> &/*inputs*/,
vec<const char *> &/*constraints*/,
vec<rtx> &clobbers, HARD_REG_SET &clobbered_regs)
{
clobbers.safe_push (gen_rtx_REG (SImode, CA_REGNO));
SET_HARD_REG_BIT (clobbered_regs, CA_REGNO);
return NULL;
}
/* Override command line options.
Combine build-specific configuration information with options
specified on the command line to set various state variables which
influence code generation, optimization, and expansion of built-in
functions. Assure that command-line configuration preferences are
compatible with each other and with the build configuration; issue
warnings while adjusting configuration or error messages while
rejecting configuration.
Upon entry to this function:
This function is called once at the beginning of
compilation, and then again at the start and end of compiling
each section of code that has a different configuration, as
indicated, for example, by adding the
__attribute__((__target__("cpu=power9")))
qualifier to a function definition or, for example, by bracketing
code between
#pragma GCC target("altivec")
and
#pragma GCC reset_options
directives. Parameter global_init_p is true for the initial
invocation, which initializes global variables, and false for all
subsequent invocations.
Various global state information is assumed to be valid. This
includes OPTION_TARGET_CPU_DEFAULT, representing the name of the
default CPU specified at build configure time, TARGET_DEFAULT,
representing the default set of option flags for the default
target, and global_options_set.x_rs6000_isa_flags, representing
which options were requested on the command line.
Upon return from this function:
rs6000_isa_flags_explicit has a non-zero bit for each flag that
was set by name on the command line. Additionally, if certain
attributes are automatically enabled or disabled by this function
in order to assure compatibility between options and
configuration, the flags associated with those attributes are
also set. By setting these "explicit bits", we avoid the risk
that other code might accidentally overwrite these particular
attributes with "default values".
The various bits of rs6000_isa_flags are set to indicate the
target options that have been selected for the most current
compilation efforts. This has the effect of also turning on the
associated TARGET_XXX values since these are macros which are
generally defined to test the corresponding bit of the
rs6000_isa_flags variable.
The variable rs6000_builtin_mask is set to represent the target
options for the most current compilation efforts, consistent with
the current contents of rs6000_isa_flags. This variable controls
expansion of built-in functions.
Various other global variables and fields of global structures
(over 50 in all) are initialized to reflect the desired options
for the most current compilation efforts. */
static bool
rs6000_option_override_internal (bool global_init_p)
{
bool ret = true;
HOST_WIDE_INT set_masks;
HOST_WIDE_INT ignore_masks;
int cpu_index = -1;
int tune_index;
struct cl_target_option *main_target_opt
= ((global_init_p || target_option_default_node == NULL)
? NULL : TREE_TARGET_OPTION (target_option_default_node));
/* Print defaults. */
if ((TARGET_DEBUG_REG || TARGET_DEBUG_TARGET) && global_init_p)
rs6000_print_isa_options (stderr, 0, "TARGET_DEFAULT", TARGET_DEFAULT);
/* Remember the explicit arguments. */
if (global_init_p)
rs6000_isa_flags_explicit = global_options_set.x_rs6000_isa_flags;
/* On 64-bit Darwin, power alignment is ABI-incompatible with some C
library functions, so warn about it. The flag may be useful for
performance studies from time to time though, so don't disable it
entirely. */
if (global_options_set.x_rs6000_alignment_flags
&& rs6000_alignment_flags == MASK_ALIGN_POWER
&& DEFAULT_ABI == ABI_DARWIN
&& TARGET_64BIT)
warning (0, "%qs is not supported for 64-bit Darwin;"
" it is incompatible with the installed C and C++ libraries",
"-malign-power");
/* Numerous experiment shows that IRA based loop pressure
calculation works better for RTL loop invariant motion on targets
with enough (>= 32) registers. It is an expensive optimization.
So it is on only for peak performance. */
if (optimize >= 3 && global_init_p
&& !global_options_set.x_flag_ira_loop_pressure)
flag_ira_loop_pressure = 1;
/* -fsanitize=address needs to turn on -fasynchronous-unwind-tables in order
for tracebacks to be complete but not if any -fasynchronous-unwind-tables
options were already specified. */
if (flag_sanitize & SANITIZE_USER_ADDRESS
&& !global_options_set.x_flag_asynchronous_unwind_tables)
flag_asynchronous_unwind_tables = 1;
/* Set the pointer size. */
if (TARGET_64BIT)
{
rs6000_pmode = DImode;
rs6000_pointer_size = 64;
}
else
{
rs6000_pmode = SImode;
rs6000_pointer_size = 32;
}
/* Some OSs don't support saving the high part of 64-bit registers on context
switch. Other OSs don't support saving Altivec registers. On those OSs,
we don't touch the OPTION_MASK_POWERPC64 or OPTION_MASK_ALTIVEC settings;
if the user wants either, the user must explicitly specify them and we
won't interfere with the user's specification. */
set_masks = POWERPC_MASKS;
#ifdef OS_MISSING_POWERPC64
if (OS_MISSING_POWERPC64)
set_masks &= ~OPTION_MASK_POWERPC64;
#endif
#ifdef OS_MISSING_ALTIVEC
if (OS_MISSING_ALTIVEC)
set_masks &= ~(OPTION_MASK_ALTIVEC | OPTION_MASK_VSX
| OTHER_VSX_VECTOR_MASKS);
#endif
/* Don't override by the processor default if given explicitly. */
set_masks &= ~rs6000_isa_flags_explicit;
/* Process the -mcpu=<xxx> and -mtune=<xxx> argument. If the user changed
the cpu in a target attribute or pragma, but did not specify a tuning
option, use the cpu for the tuning option rather than the option specified
with -mtune on the command line. Process a '--with-cpu' configuration
request as an implicit --cpu. */
if (rs6000_cpu_index >= 0)
cpu_index = rs6000_cpu_index;
else if (main_target_opt != NULL && main_target_opt->x_rs6000_cpu_index >= 0)
cpu_index = main_target_opt->x_rs6000_cpu_index;
else if (OPTION_TARGET_CPU_DEFAULT)
cpu_index = rs6000_cpu_name_lookup (OPTION_TARGET_CPU_DEFAULT);
if (cpu_index >= 0)
{
const char *unavailable_cpu = NULL;
switch (processor_target_table[cpu_index].processor)
{
#ifndef HAVE_AS_POWER9
case PROCESSOR_POWER9:
unavailable_cpu = "power9";
break;
#endif
#ifndef HAVE_AS_POWER8
case PROCESSOR_POWER8:
unavailable_cpu = "power8";
break;
#endif
#ifndef HAVE_AS_POPCNTD
case PROCESSOR_POWER7:
unavailable_cpu = "power7";
break;
#endif
#ifndef HAVE_AS_DFP
case PROCESSOR_POWER6:
unavailable_cpu = "power6";
break;
#endif
#ifndef HAVE_AS_POPCNTB
case PROCESSOR_POWER5:
unavailable_cpu = "power5";
break;
#endif
default:
break;
}
if (unavailable_cpu)
{
cpu_index = -1;
warning (0, "will not generate %qs instructions because "
"assembler lacks %qs support", unavailable_cpu,
unavailable_cpu);
}
}
/* If we have a cpu, either through an explicit -mcpu=<xxx> or if the
compiler was configured with --with-cpu=<xxx>, replace all of the ISA bits
with those from the cpu, except for options that were explicitly set. If
we don't have a cpu, do not override the target bits set in
TARGET_DEFAULT. */
if (cpu_index >= 0)
{
rs6000_cpu_index = cpu_index;
rs6000_isa_flags &= ~set_masks;
rs6000_isa_flags |= (processor_target_table[cpu_index].target_enable
& set_masks);
}
else
{
/* If no -mcpu=<xxx>, inherit any default options that were cleared via
POWERPC_MASKS. Originally, TARGET_DEFAULT was used to initialize
target_flags via the TARGET_DEFAULT_TARGET_FLAGS hook. When we switched
to using rs6000_isa_flags, we need to do the initialization here.
If there is a TARGET_DEFAULT, use that. Otherwise fall back to using
-mcpu=powerpc, -mcpu=powerpc64, or -mcpu=powerpc64le defaults. */
HOST_WIDE_INT flags;
if (TARGET_DEFAULT)
flags = TARGET_DEFAULT;
else
{
/* PowerPC 64-bit LE requires at least ISA 2.07. */
const char *default_cpu = (!TARGET_POWERPC64
? "powerpc"
: (BYTES_BIG_ENDIAN
? "powerpc64"
: "powerpc64le"));
int default_cpu_index = rs6000_cpu_name_lookup (default_cpu);
flags = processor_target_table[default_cpu_index].target_enable;
}
rs6000_isa_flags |= (flags & ~rs6000_isa_flags_explicit);
}
if (rs6000_tune_index >= 0)
tune_index = rs6000_tune_index;
else if (cpu_index >= 0)
rs6000_tune_index = tune_index = cpu_index;
else
{
size_t i;
enum processor_type tune_proc
= (TARGET_POWERPC64 ? PROCESSOR_DEFAULT64 : PROCESSOR_DEFAULT);
tune_index = -1;
for (i = 0; i < ARRAY_SIZE (processor_target_table); i++)
if (processor_target_table[i].processor == tune_proc)
{
tune_index = i;
break;
}
}
if (cpu_index >= 0)
rs6000_cpu = processor_target_table[cpu_index].processor;
else
rs6000_cpu = TARGET_POWERPC64 ? PROCESSOR_DEFAULT64 : PROCESSOR_DEFAULT;
gcc_assert (tune_index >= 0);
rs6000_tune = processor_target_table[tune_index].processor;
if (rs6000_cpu == PROCESSOR_PPCE300C2 || rs6000_cpu == PROCESSOR_PPCE300C3
|| rs6000_cpu == PROCESSOR_PPCE500MC || rs6000_cpu == PROCESSOR_PPCE500MC64
|| rs6000_cpu == PROCESSOR_PPCE5500)
{
if (TARGET_ALTIVEC)
error ("AltiVec not supported in this target");
}
/* If we are optimizing big endian systems for space, use the load/store
multiple and string instructions. */
if (BYTES_BIG_ENDIAN && optimize_size)
rs6000_isa_flags |= ~rs6000_isa_flags_explicit & (OPTION_MASK_MULTIPLE
| OPTION_MASK_STRING);
/* Don't allow -mmultiple or -mstring on little endian systems
unless the cpu is a 750, because the hardware doesn't support the
instructions used in little endian mode, and causes an alignment
trap. The 750 does not cause an alignment trap (except when the
target is unaligned). */
if (!BYTES_BIG_ENDIAN && rs6000_cpu != PROCESSOR_PPC750)
{
if (TARGET_MULTIPLE)
{
rs6000_isa_flags &= ~OPTION_MASK_MULTIPLE;
if ((rs6000_isa_flags_explicit & OPTION_MASK_MULTIPLE) != 0)
warning (0, "%qs is not supported on little endian systems",
"-mmultiple");
}
if (TARGET_STRING)
{
rs6000_isa_flags &= ~OPTION_MASK_STRING;
if ((rs6000_isa_flags_explicit & OPTION_MASK_STRING) != 0)
warning (0, "%qs is not supported on little endian systems",
"-mstring");
}
}
/* If little-endian, default to -mstrict-align on older processors.
Testing for htm matches power8 and later. */
if (!BYTES_BIG_ENDIAN
&& !(processor_target_table[tune_index].target_enable & OPTION_MASK_HTM))
rs6000_isa_flags |= ~rs6000_isa_flags_explicit & OPTION_MASK_STRICT_ALIGN;
/* -maltivec={le,be} implies -maltivec. */
if (rs6000_altivec_element_order != 0)
rs6000_isa_flags |= OPTION_MASK_ALTIVEC;
/* Disallow -maltivec=le in big endian mode for now. This is not
known to be useful for anyone. */
if (BYTES_BIG_ENDIAN && rs6000_altivec_element_order == 1)
{
warning (0, N_("-maltivec=le not allowed for big-endian targets"));
rs6000_altivec_element_order = 0;
}
if (!rs6000_fold_gimple)
fprintf (stderr,
"gimple folding of rs6000 builtins has been disabled.\n");
/* Add some warnings for VSX. */
if (TARGET_VSX)
{
const char *msg = NULL;
if (!TARGET_HARD_FLOAT || !TARGET_SINGLE_FLOAT || !TARGET_DOUBLE_FLOAT)
{
if (rs6000_isa_flags_explicit & OPTION_MASK_VSX)
msg = N_("-mvsx requires hardware floating point");
else
{
rs6000_isa_flags &= ~ OPTION_MASK_VSX;
rs6000_isa_flags_explicit |= OPTION_MASK_VSX;
}
}
else if (TARGET_PAIRED_FLOAT)
msg = N_("-mvsx and -mpaired are incompatible");
else if (TARGET_AVOID_XFORM > 0)
msg = N_("-mvsx needs indexed addressing");
else if (!TARGET_ALTIVEC && (rs6000_isa_flags_explicit
& OPTION_MASK_ALTIVEC))
{
if (rs6000_isa_flags_explicit & OPTION_MASK_VSX)
msg = N_("-mvsx and -mno-altivec are incompatible");
else
msg = N_("-mno-altivec disables vsx");
}
if (msg)
{
warning (0, msg);
rs6000_isa_flags &= ~ OPTION_MASK_VSX;
rs6000_isa_flags_explicit |= OPTION_MASK_VSX;
}
}
/* If hard-float/altivec/vsx were explicitly turned off then don't allow
the -mcpu setting to enable options that conflict. */
if ((!TARGET_HARD_FLOAT || !TARGET_ALTIVEC || !TARGET_VSX)
&& (rs6000_isa_flags_explicit & (OPTION_MASK_SOFT_FLOAT
| OPTION_MASK_ALTIVEC
| OPTION_MASK_VSX)) != 0)
rs6000_isa_flags &= ~((OPTION_MASK_P8_VECTOR | OPTION_MASK_CRYPTO
| OPTION_MASK_DIRECT_MOVE)
& ~rs6000_isa_flags_explicit);
if (TARGET_DEBUG_REG || TARGET_DEBUG_TARGET)
rs6000_print_isa_options (stderr, 0, "before defaults", rs6000_isa_flags);
/* Handle explicit -mno-{altivec,vsx,power8-vector,power9-vector} and turn
off all of the options that depend on those flags. */
ignore_masks = rs6000_disable_incompatible_switches ();
/* For the newer switches (vsx, dfp, etc.) set some of the older options,
unless the user explicitly used the -mno-<option> to disable the code. */
if (TARGET_P9_VECTOR || TARGET_MODULO || TARGET_P9_MISC)
rs6000_isa_flags |= (ISA_3_0_MASKS_SERVER & ~ignore_masks);
else if (TARGET_P9_MINMAX)
{
if (cpu_index >= 0)
{
if (cpu_index == PROCESSOR_POWER9)
{
/* legacy behavior: allow -mcpu=power9 with certain
capabilities explicitly disabled. */
rs6000_isa_flags |= (ISA_3_0_MASKS_SERVER & ~ignore_masks);
}
else
error ("power9 target option is incompatible with %<%s=<xxx>%> "
"for <xxx> less than power9", "-mcpu");
}
else if ((ISA_3_0_MASKS_SERVER & rs6000_isa_flags_explicit)
!= (ISA_3_0_MASKS_SERVER & rs6000_isa_flags
& rs6000_isa_flags_explicit))
/* Enforce that none of the ISA_3_0_MASKS_SERVER flags
were explicitly cleared. */
error ("%qs incompatible with explicitly disabled options",
"-mpower9-minmax");
else
rs6000_isa_flags |= ISA_3_0_MASKS_SERVER;
}
else if (TARGET_P8_VECTOR || TARGET_DIRECT_MOVE || TARGET_CRYPTO)
rs6000_isa_flags |= (ISA_2_7_MASKS_SERVER & ~ignore_masks);
else if (TARGET_VSX)
rs6000_isa_flags |= (ISA_2_6_MASKS_SERVER & ~ignore_masks);
else if (TARGET_POPCNTD)
rs6000_isa_flags |= (ISA_2_6_MASKS_EMBEDDED & ~ignore_masks);
else if (TARGET_DFP)
rs6000_isa_flags |= (ISA_2_5_MASKS_SERVER & ~ignore_masks);
else if (TARGET_CMPB)
rs6000_isa_flags |= (ISA_2_5_MASKS_EMBEDDED & ~ignore_masks);
else if (TARGET_FPRND)
rs6000_isa_flags |= (ISA_2_4_MASKS & ~ignore_masks);
else if (TARGET_POPCNTB)
rs6000_isa_flags |= (ISA_2_2_MASKS & ~ignore_masks);
else if (TARGET_ALTIVEC)
rs6000_isa_flags |= (OPTION_MASK_PPC_GFXOPT & ~ignore_masks);
if (TARGET_CRYPTO && !TARGET_ALTIVEC)
{
if (rs6000_isa_flags_explicit & OPTION_MASK_CRYPTO)
error ("%qs requires %qs", "-mcrypto", "-maltivec");
rs6000_isa_flags &= ~OPTION_MASK_CRYPTO;
}
if (TARGET_DIRECT_MOVE && !TARGET_VSX)
{
if (rs6000_isa_flags_explicit & OPTION_MASK_DIRECT_MOVE)
error ("%qs requires %qs", "-mdirect-move", "-mvsx");
rs6000_isa_flags &= ~OPTION_MASK_DIRECT_MOVE;
}
if (TARGET_P8_VECTOR && !TARGET_ALTIVEC)
{
if (rs6000_isa_flags_explicit & OPTION_MASK_P8_VECTOR)
error ("%qs requires %qs", "-mpower8-vector", "-maltivec");
rs6000_isa_flags &= ~OPTION_MASK_P8_VECTOR;
}
if (TARGET_P8_VECTOR && !TARGET_VSX)
{
if ((rs6000_isa_flags_explicit & OPTION_MASK_P8_VECTOR)
&& (rs6000_isa_flags_explicit & OPTION_MASK_VSX))
error ("%qs requires %qs", "-mpower8-vector", "-mvsx");
else if ((rs6000_isa_flags_explicit & OPTION_MASK_P8_VECTOR) == 0)
{
rs6000_isa_flags &= ~OPTION_MASK_P8_VECTOR;
if (rs6000_isa_flags_explicit & OPTION_MASK_VSX)
rs6000_isa_flags_explicit |= OPTION_MASK_P8_VECTOR;
}
else
{
/* OPTION_MASK_P8_VECTOR is explicit, and OPTION_MASK_VSX is
not explicit. */
rs6000_isa_flags |= OPTION_MASK_VSX;
rs6000_isa_flags_explicit |= OPTION_MASK_VSX;
}
}
if (TARGET_DFP && !TARGET_HARD_FLOAT)
{
if (rs6000_isa_flags_explicit & OPTION_MASK_DFP)
error ("%qs requires %qs", "-mhard-dfp", "-mhard-float");
rs6000_isa_flags &= ~OPTION_MASK_DFP;
}
/* The quad memory instructions only works in 64-bit mode. In 32-bit mode,
silently turn off quad memory mode. */
if ((TARGET_QUAD_MEMORY || TARGET_QUAD_MEMORY_ATOMIC) && !TARGET_POWERPC64)
{
if ((rs6000_isa_flags_explicit & OPTION_MASK_QUAD_MEMORY) != 0)
warning (0, N_("-mquad-memory requires 64-bit mode"));
if ((rs6000_isa_flags_explicit & OPTION_MASK_QUAD_MEMORY_ATOMIC) != 0)
warning (0, N_("-mquad-memory-atomic requires 64-bit mode"));
rs6000_isa_flags &= ~(OPTION_MASK_QUAD_MEMORY
| OPTION_MASK_QUAD_MEMORY_ATOMIC);
}
/* Non-atomic quad memory load/store are disabled for little endian, since
the words are reversed, but atomic operations can still be done by
swapping the words. */
if (TARGET_QUAD_MEMORY && !WORDS_BIG_ENDIAN)
{
if ((rs6000_isa_flags_explicit & OPTION_MASK_QUAD_MEMORY) != 0)
warning (0, N_("-mquad-memory is not available in little endian "
"mode"));
rs6000_isa_flags &= ~OPTION_MASK_QUAD_MEMORY;
}
/* Assume if the user asked for normal quad memory instructions, they want
the atomic versions as well, unless they explicity told us not to use quad
word atomic instructions. */
if (TARGET_QUAD_MEMORY
&& !TARGET_QUAD_MEMORY_ATOMIC
&& ((rs6000_isa_flags_explicit & OPTION_MASK_QUAD_MEMORY_ATOMIC) == 0))
rs6000_isa_flags |= OPTION_MASK_QUAD_MEMORY_ATOMIC;
/* If we can shrink-wrap the TOC register save separately, then use
-msave-toc-indirect unless explicitly disabled. */
if ((rs6000_isa_flags_explicit & OPTION_MASK_SAVE_TOC_INDIRECT) == 0
&& flag_shrink_wrap_separate
&& optimize_function_for_speed_p (cfun))
rs6000_isa_flags |= OPTION_MASK_SAVE_TOC_INDIRECT;
/* Enable power8 fusion if we are tuning for power8, even if we aren't
generating power8 instructions. */
if (!(rs6000_isa_flags_explicit & OPTION_MASK_P8_FUSION))
rs6000_isa_flags |= (processor_target_table[tune_index].target_enable
& OPTION_MASK_P8_FUSION);
/* Setting additional fusion flags turns on base fusion. */
if (!TARGET_P8_FUSION && (TARGET_P8_FUSION_SIGN || TARGET_TOC_FUSION))
{
if (rs6000_isa_flags_explicit & OPTION_MASK_P8_FUSION)
{
if (TARGET_P8_FUSION_SIGN)
error ("%qs requires %qs", "-mpower8-fusion-sign",
"-mpower8-fusion");
if (TARGET_TOC_FUSION)
error ("%qs requires %qs", "-mtoc-fusion", "-mpower8-fusion");
rs6000_isa_flags &= ~OPTION_MASK_P8_FUSION;
}
else
rs6000_isa_flags |= OPTION_MASK_P8_FUSION;
}
/* Power9 fusion is a superset over power8 fusion. */
if (TARGET_P9_FUSION && !TARGET_P8_FUSION)
{
if (rs6000_isa_flags_explicit & OPTION_MASK_P8_FUSION)
{
/* We prefer to not mention undocumented options in
error messages. However, if users have managed to select
power9-fusion without selecting power8-fusion, they
already know about undocumented flags. */
error ("%qs requires %qs", "-mpower9-fusion", "-mpower8-fusion");
rs6000_isa_flags &= ~OPTION_MASK_P9_FUSION;
}
else
rs6000_isa_flags |= OPTION_MASK_P8_FUSION;
}
/* Enable power9 fusion if we are tuning for power9, even if we aren't
generating power9 instructions. */
if (!(rs6000_isa_flags_explicit & OPTION_MASK_P9_FUSION))
rs6000_isa_flags |= (processor_target_table[tune_index].target_enable
& OPTION_MASK_P9_FUSION);
/* Power8 does not fuse sign extended loads with the addis. If we are
optimizing at high levels for speed, convert a sign extended load into a
zero extending load, and an explicit sign extension. */
if (TARGET_P8_FUSION
&& !(rs6000_isa_flags_explicit & OPTION_MASK_P8_FUSION_SIGN)
&& optimize_function_for_speed_p (cfun)
&& optimize >= 3)
rs6000_isa_flags |= OPTION_MASK_P8_FUSION_SIGN;
/* TOC fusion requires 64-bit and medium/large code model. */
if (TARGET_TOC_FUSION && !TARGET_POWERPC64)
{
rs6000_isa_flags &= ~OPTION_MASK_TOC_FUSION;
if ((rs6000_isa_flags_explicit & OPTION_MASK_TOC_FUSION) != 0)
warning (0, N_("-mtoc-fusion requires 64-bit"));
}
if (TARGET_TOC_FUSION && (TARGET_CMODEL == CMODEL_SMALL))
{
rs6000_isa_flags &= ~OPTION_MASK_TOC_FUSION;
if ((rs6000_isa_flags_explicit & OPTION_MASK_TOC_FUSION) != 0)
warning (0, N_("-mtoc-fusion requires medium/large code model"));
}
/* Turn on -mtoc-fusion by default if p8-fusion and 64-bit medium/large code
model. */
if (TARGET_P8_FUSION && !TARGET_TOC_FUSION && TARGET_POWERPC64
&& (TARGET_CMODEL != CMODEL_SMALL)
&& !(rs6000_isa_flags_explicit & OPTION_MASK_TOC_FUSION))
rs6000_isa_flags |= OPTION_MASK_TOC_FUSION;
/* ISA 3.0 vector instructions include ISA 2.07. */
if (TARGET_P9_VECTOR && !TARGET_P8_VECTOR)
{
/* We prefer to not mention undocumented options in
error messages. However, if users have managed to select
power9-vector without selecting power8-vector, they
already know about undocumented flags. */
if ((rs6000_isa_flags_explicit & OPTION_MASK_P9_VECTOR) &&
(rs6000_isa_flags_explicit & OPTION_MASK_P8_VECTOR))
error ("%qs requires %qs", "-mpower9-vector", "-mpower8-vector");
else if ((rs6000_isa_flags_explicit & OPTION_MASK_P9_VECTOR) == 0)
{
rs6000_isa_flags &= ~OPTION_MASK_P9_VECTOR;
if (rs6000_isa_flags_explicit & OPTION_MASK_P8_VECTOR)
rs6000_isa_flags_explicit |= OPTION_MASK_P9_VECTOR;
}
else
{
/* OPTION_MASK_P9_VECTOR is explicit and
OPTION_MASK_P8_VECTOR is not explicit. */
rs6000_isa_flags |= OPTION_MASK_P8_VECTOR;
rs6000_isa_flags_explicit |= OPTION_MASK_P8_VECTOR;
}
}
/* Set -mallow-movmisalign to explicitly on if we have full ISA 2.07
support. If we only have ISA 2.06 support, and the user did not specify
the switch, leave it set to -1 so the movmisalign patterns are enabled,
but we don't enable the full vectorization support */
if (TARGET_ALLOW_MOVMISALIGN == -1 && TARGET_P8_VECTOR && TARGET_DIRECT_MOVE)
TARGET_ALLOW_MOVMISALIGN = 1;
else if (TARGET_ALLOW_MOVMISALIGN && !TARGET_VSX)
{
if (TARGET_ALLOW_MOVMISALIGN > 0
&& global_options_set.x_TARGET_ALLOW_MOVMISALIGN)
error ("%qs requires %qs", "-mallow-movmisalign", "-mvsx");
TARGET_ALLOW_MOVMISALIGN = 0;
}
/* Determine when unaligned vector accesses are permitted, and when
they are preferred over masked Altivec loads. Note that if
TARGET_ALLOW_MOVMISALIGN has been disabled by the user, then
TARGET_EFFICIENT_UNALIGNED_VSX must be as well. The converse is
not true. */
if (TARGET_EFFICIENT_UNALIGNED_VSX)
{
if (!TARGET_VSX)
{
if (rs6000_isa_flags_explicit & OPTION_MASK_EFFICIENT_UNALIGNED_VSX)
error ("%qs requires %qs", "-mefficient-unaligned-vsx", "-mvsx");
rs6000_isa_flags &= ~OPTION_MASK_EFFICIENT_UNALIGNED_VSX;
}
else if (!TARGET_ALLOW_MOVMISALIGN)
{
if (rs6000_isa_flags_explicit & OPTION_MASK_EFFICIENT_UNALIGNED_VSX)
error ("%qs requires %qs", "-munefficient-unaligned-vsx",
"-mallow-movmisalign");
rs6000_isa_flags &= ~OPTION_MASK_EFFICIENT_UNALIGNED_VSX;
}
}
/* Set long double size before the IEEE 128-bit tests. */
if (!global_options_set.x_rs6000_long_double_type_size)
{
if (main_target_opt != NULL
&& (main_target_opt->x_rs6000_long_double_type_size
!= RS6000_DEFAULT_LONG_DOUBLE_SIZE))
error ("target attribute or pragma changes long double size");
else
rs6000_long_double_type_size = RS6000_DEFAULT_LONG_DOUBLE_SIZE;
}
/* Set -mabi=ieeelongdouble on some old targets. In the future, power server
systems will also set long double to be IEEE 128-bit. AIX and Darwin
explicitly redefine TARGET_IEEEQUAD and TARGET_IEEEQUAD_DEFAULT to 0, so
those systems will not pick up this default. Warn if the user changes the
default unless -Wno-psabi. */
if (!global_options_set.x_rs6000_ieeequad)
rs6000_ieeequad = TARGET_IEEEQUAD_DEFAULT;
else if (rs6000_ieeequad != TARGET_IEEEQUAD_DEFAULT && TARGET_LONG_DOUBLE_128)
{
static bool warned_change_long_double;
if (!warned_change_long_double)
{
warned_change_long_double = true;
if (TARGET_IEEEQUAD)
warning (OPT_Wpsabi, "Using IEEE extended precision long double");
else
warning (OPT_Wpsabi, "Using IBM extended precision long double");
}
}
/* Enable the default support for IEEE 128-bit floating point on Linux VSX
sytems. In GCC 7, we would enable the the IEEE 128-bit floating point
infrastructure (-mfloat128-type) but not enable the actual __float128 type
unless the user used the explicit -mfloat128. In GCC 8, we enable both
the keyword as well as the type. */
TARGET_FLOAT128_TYPE = TARGET_FLOAT128_ENABLE_TYPE && TARGET_VSX;
/* IEEE 128-bit floating point requires VSX support. */
if (TARGET_FLOAT128_KEYWORD)
{
if (!TARGET_VSX)
{
if ((rs6000_isa_flags_explicit & OPTION_MASK_FLOAT128_KEYWORD) != 0)
error ("%qs requires VSX support", "-mfloat128");
TARGET_FLOAT128_TYPE = 0;
rs6000_isa_flags &= ~(OPTION_MASK_FLOAT128_KEYWORD
| OPTION_MASK_FLOAT128_HW);
}
else if (!TARGET_FLOAT128_TYPE)
{
TARGET_FLOAT128_TYPE = 1;
warning (0, "The -mfloat128 option may not be fully supported");
}
}
/* Enable the __float128 keyword under Linux by default. */
if (TARGET_FLOAT128_TYPE && !TARGET_FLOAT128_KEYWORD
&& (rs6000_isa_flags_explicit & OPTION_MASK_FLOAT128_KEYWORD) == 0)
rs6000_isa_flags |= OPTION_MASK_FLOAT128_KEYWORD;
/* If we have are supporting the float128 type and full ISA 3.0 support,
enable -mfloat128-hardware by default. However, don't enable the
__float128 keyword if it was explicitly turned off. 64-bit mode is needed
because sometimes the compiler wants to put things in an integer
container, and if we don't have __int128 support, it is impossible. */
if (TARGET_FLOAT128_TYPE && !TARGET_FLOAT128_HW && TARGET_64BIT
&& (rs6000_isa_flags & ISA_3_0_MASKS_IEEE) == ISA_3_0_MASKS_IEEE
&& !(rs6000_isa_flags_explicit & OPTION_MASK_FLOAT128_HW))
rs6000_isa_flags |= OPTION_MASK_FLOAT128_HW;
if (TARGET_FLOAT128_HW
&& (rs6000_isa_flags & ISA_3_0_MASKS_IEEE) != ISA_3_0_MASKS_IEEE)
{
if ((rs6000_isa_flags_explicit & OPTION_MASK_FLOAT128_HW) != 0)
error ("%qs requires full ISA 3.0 support", "-mfloat128-hardware");
rs6000_isa_flags &= ~OPTION_MASK_FLOAT128_HW;
}
if (TARGET_FLOAT128_HW && !TARGET_64BIT)
{
if ((rs6000_isa_flags_explicit & OPTION_MASK_FLOAT128_HW) != 0)
error ("%qs requires %qs", "-mfloat128-hardware", "-m64");
rs6000_isa_flags &= ~OPTION_MASK_FLOAT128_HW;
}
/* Print the options after updating the defaults. */
if (TARGET_DEBUG_REG || TARGET_DEBUG_TARGET)
rs6000_print_isa_options (stderr, 0, "after defaults", rs6000_isa_flags);
/* E500mc does "better" if we inline more aggressively. Respect the
user's opinion, though. */
if (rs6000_block_move_inline_limit == 0
&& (rs6000_tune == PROCESSOR_PPCE500MC
|| rs6000_tune == PROCESSOR_PPCE500MC64
|| rs6000_tune == PROCESSOR_PPCE5500
|| rs6000_tune == PROCESSOR_PPCE6500))
rs6000_block_move_inline_limit = 128;
/* store_one_arg depends on expand_block_move to handle at least the
size of reg_parm_stack_space. */
if (rs6000_block_move_inline_limit < (TARGET_POWERPC64 ? 64 : 32))
rs6000_block_move_inline_limit = (TARGET_POWERPC64 ? 64 : 32);
if (global_init_p)
{
/* If the appropriate debug option is enabled, replace the target hooks
with debug versions that call the real version and then prints
debugging information. */
if (TARGET_DEBUG_COST)
{
targetm.rtx_costs = rs6000_debug_rtx_costs;
targetm.address_cost = rs6000_debug_address_cost;
targetm.sched.adjust_cost = rs6000_debug_adjust_cost;
}
if (TARGET_DEBUG_ADDR)
{
targetm.legitimate_address_p = rs6000_debug_legitimate_address_p;
targetm.legitimize_address = rs6000_debug_legitimize_address;
rs6000_secondary_reload_class_ptr
= rs6000_debug_secondary_reload_class;
targetm.secondary_memory_needed
= rs6000_debug_secondary_memory_needed;
targetm.can_change_mode_class
= rs6000_debug_can_change_mode_class;
rs6000_preferred_reload_class_ptr
= rs6000_debug_preferred_reload_class;
rs6000_legitimize_reload_address_ptr
= rs6000_debug_legitimize_reload_address;
rs6000_mode_dependent_address_ptr
= rs6000_debug_mode_dependent_address;
}
if (rs6000_veclibabi_name)
{
if (strcmp (rs6000_veclibabi_name, "mass") == 0)
rs6000_veclib_handler = rs6000_builtin_vectorized_libmass;
else
{
error ("unknown vectorization library ABI type (%qs) for "
"%qs switch", rs6000_veclibabi_name, "-mveclibabi=");
ret = false;
}
}
}
/* Disable VSX and Altivec silently if the user switched cpus to power7 in a
target attribute or pragma which automatically enables both options,
unless the altivec ABI was set. This is set by default for 64-bit, but
not for 32-bit. */
if (main_target_opt != NULL && !main_target_opt->x_rs6000_altivec_abi)
{
TARGET_FLOAT128_TYPE = 0;
rs6000_isa_flags &= ~((OPTION_MASK_VSX | OPTION_MASK_ALTIVEC
| OPTION_MASK_FLOAT128_KEYWORD)
& ~rs6000_isa_flags_explicit);
}
/* Enable Altivec ABI for AIX -maltivec. */
if (TARGET_XCOFF && (TARGET_ALTIVEC || TARGET_VSX))
{
if (main_target_opt != NULL && !main_target_opt->x_rs6000_altivec_abi)
error ("target attribute or pragma changes AltiVec ABI");
else
rs6000_altivec_abi = 1;
}
/* The AltiVec ABI is the default for PowerPC-64 GNU/Linux. For
PowerPC-32 GNU/Linux, -maltivec implies the AltiVec ABI. It can
be explicitly overridden in either case. */
if (TARGET_ELF)
{
if (!global_options_set.x_rs6000_altivec_abi
&& (TARGET_64BIT || TARGET_ALTIVEC || TARGET_VSX))
{
if (main_target_opt != NULL &&
!main_target_opt->x_rs6000_altivec_abi)
error ("target attribute or pragma changes AltiVec ABI");
else
rs6000_altivec_abi = 1;
}
}
/* Set the Darwin64 ABI as default for 64-bit Darwin.
So far, the only darwin64 targets are also MACH-O. */
if (TARGET_MACHO
&& DEFAULT_ABI == ABI_DARWIN
&& TARGET_64BIT)
{
if (main_target_opt != NULL && !main_target_opt->x_rs6000_darwin64_abi)
error ("target attribute or pragma changes darwin64 ABI");
else
{
rs6000_darwin64_abi = 1;
/* Default to natural alignment, for better performance. */
rs6000_alignment_flags = MASK_ALIGN_NATURAL;
}
}
/* Place FP constants in the constant pool instead of TOC
if section anchors enabled. */
if (flag_section_anchors
&& !global_options_set.x_TARGET_NO_FP_IN_TOC)
TARGET_NO_FP_IN_TOC = 1;
if (TARGET_DEBUG_REG || TARGET_DEBUG_TARGET)
rs6000_print_isa_options (stderr, 0, "before subtarget", rs6000_isa_flags);
#ifdef SUBTARGET_OVERRIDE_OPTIONS
SUBTARGET_OVERRIDE_OPTIONS;
#endif
#ifdef SUBSUBTARGET_OVERRIDE_OPTIONS
SUBSUBTARGET_OVERRIDE_OPTIONS;
#endif
#ifdef SUB3TARGET_OVERRIDE_OPTIONS
SUB3TARGET_OVERRIDE_OPTIONS;
#endif
if (TARGET_DEBUG_REG || TARGET_DEBUG_TARGET)
rs6000_print_isa_options (stderr, 0, "after subtarget", rs6000_isa_flags);
/* For the E500 family of cores, reset the single/double FP flags to let us
check that they remain constant across attributes or pragmas. Also,
clear a possible request for string instructions, not supported and which
we might have silently queried above for -Os. */
switch (rs6000_cpu)
{
case PROCESSOR_PPC8540:
case PROCESSOR_PPC8548:
case PROCESSOR_PPCE500MC:
case PROCESSOR_PPCE500MC64:
case PROCESSOR_PPCE5500:
case PROCESSOR_PPCE6500:
rs6000_single_float = 0;
rs6000_double_float = 0;
rs6000_isa_flags &= ~OPTION_MASK_STRING;
break;
default:
break;
}
if (main_target_opt)
{
if (main_target_opt->x_rs6000_single_float != rs6000_single_float)
error ("target attribute or pragma changes single precision floating "
"point");
if (main_target_opt->x_rs6000_double_float != rs6000_double_float)
error ("target attribute or pragma changes double precision floating "
"point");
}
rs6000_always_hint = (rs6000_tune != PROCESSOR_POWER4
&& rs6000_tune != PROCESSOR_POWER5
&& rs6000_tune != PROCESSOR_POWER6
&& rs6000_tune != PROCESSOR_POWER7
&& rs6000_tune != PROCESSOR_POWER8
&& rs6000_tune != PROCESSOR_POWER9
&& rs6000_tune != PROCESSOR_PPCA2
&& rs6000_tune != PROCESSOR_CELL
&& rs6000_tune != PROCESSOR_PPC476);
rs6000_sched_groups = (rs6000_tune == PROCESSOR_POWER4
|| rs6000_tune == PROCESSOR_POWER5
|| rs6000_tune == PROCESSOR_POWER7
|| rs6000_tune == PROCESSOR_POWER8);
rs6000_align_branch_targets = (rs6000_tune == PROCESSOR_POWER4
|| rs6000_tune == PROCESSOR_POWER5
|| rs6000_tune == PROCESSOR_POWER6
|| rs6000_tune == PROCESSOR_POWER7
|| rs6000_tune == PROCESSOR_POWER8
|| rs6000_tune == PROCESSOR_POWER9
|| rs6000_tune == PROCESSOR_PPCE500MC
|| rs6000_tune == PROCESSOR_PPCE500MC64
|| rs6000_tune == PROCESSOR_PPCE5500
|| rs6000_tune == PROCESSOR_PPCE6500);
/* Allow debug switches to override the above settings. These are set to -1
in rs6000.opt to indicate the user hasn't directly set the switch. */
if (TARGET_ALWAYS_HINT >= 0)
rs6000_always_hint = TARGET_ALWAYS_HINT;
if (TARGET_SCHED_GROUPS >= 0)
rs6000_sched_groups = TARGET_SCHED_GROUPS;
if (TARGET_ALIGN_BRANCH_TARGETS >= 0)
rs6000_align_branch_targets = TARGET_ALIGN_BRANCH_TARGETS;
rs6000_sched_restricted_insns_priority
= (rs6000_sched_groups ? 1 : 0);
/* Handle -msched-costly-dep option. */
rs6000_sched_costly_dep
= (rs6000_sched_groups ? true_store_to_load_dep_costly : no_dep_costly);
if (rs6000_sched_costly_dep_str)
{
if (! strcmp (rs6000_sched_costly_dep_str, "no"))
rs6000_sched_costly_dep = no_dep_costly;
else if (! strcmp (rs6000_sched_costly_dep_str, "all"))
rs6000_sched_costly_dep = all_deps_costly;
else if (! strcmp (rs6000_sched_costly_dep_str, "true_store_to_load"))
rs6000_sched_costly_dep = true_store_to_load_dep_costly;
else if (! strcmp (rs6000_sched_costly_dep_str, "store_to_load"))
rs6000_sched_costly_dep = store_to_load_dep_costly;
else
rs6000_sched_costly_dep = ((enum rs6000_dependence_cost)
atoi (rs6000_sched_costly_dep_str));
}
/* Handle -minsert-sched-nops option. */
rs6000_sched_insert_nops
= (rs6000_sched_groups ? sched_finish_regroup_exact : sched_finish_none);
if (rs6000_sched_insert_nops_str)
{
if (! strcmp (rs6000_sched_insert_nops_str, "no"))
rs6000_sched_insert_nops = sched_finish_none;
else if (! strcmp (rs6000_sched_insert_nops_str, "pad"))
rs6000_sched_insert_nops = sched_finish_pad_groups;
else if (! strcmp (rs6000_sched_insert_nops_str, "regroup_exact"))
rs6000_sched_insert_nops = sched_finish_regroup_exact;
else
rs6000_sched_insert_nops = ((enum rs6000_nop_insertion)
atoi (rs6000_sched_insert_nops_str));
}
/* Handle stack protector */
if (!global_options_set.x_rs6000_stack_protector_guard)
#ifdef TARGET_THREAD_SSP_OFFSET
rs6000_stack_protector_guard = SSP_TLS;
#else
rs6000_stack_protector_guard = SSP_GLOBAL;
#endif
#ifdef TARGET_THREAD_SSP_OFFSET
rs6000_stack_protector_guard_offset = TARGET_THREAD_SSP_OFFSET;
rs6000_stack_protector_guard_reg = TARGET_64BIT ? 13 : 2;
#endif
if (global_options_set.x_rs6000_stack_protector_guard_offset_str)
{
char *endp;
const char *str = rs6000_stack_protector_guard_offset_str;
errno = 0;
long offset = strtol (str, &endp, 0);
if (!*str || *endp || errno)
error ("%qs is not a valid number in %qs", str,
"-mstack-protector-guard-offset=");
if (!IN_RANGE (offset, -0x8000, 0x7fff)
|| (TARGET_64BIT && (offset & 3)))
error ("%qs is not a valid offset in %qs", str,
"-mstack-protector-guard-offset=");
rs6000_stack_protector_guard_offset = offset;
}
if (global_options_set.x_rs6000_stack_protector_guard_reg_str)
{
const char *str = rs6000_stack_protector_guard_reg_str;
int reg = decode_reg_name (str);
if (!IN_RANGE (reg, 1, 31))
error ("%qs is not a valid base register in %qs", str,
"-mstack-protector-guard-reg=");
rs6000_stack_protector_guard_reg = reg;
}
if (rs6000_stack_protector_guard == SSP_TLS
&& !IN_RANGE (rs6000_stack_protector_guard_reg, 1, 31))
error ("%qs needs a valid base register", "-mstack-protector-guard=tls");
if (global_init_p)
{
#ifdef TARGET_REGNAMES
/* If the user desires alternate register names, copy in the
alternate names now. */
if (TARGET_REGNAMES)
memcpy (rs6000_reg_names, alt_reg_names, sizeof (rs6000_reg_names));
#endif
/* Set aix_struct_return last, after the ABI is determined.
If -maix-struct-return or -msvr4-struct-return was explicitly
used, don't override with the ABI default. */
if (!global_options_set.x_aix_struct_return)
aix_struct_return = (DEFAULT_ABI != ABI_V4 || DRAFT_V4_STRUCT_RET);
#if 0
/* IBM XL compiler defaults to unsigned bitfields. */
if (TARGET_XL_COMPAT)
flag_signed_bitfields = 0;
#endif
if (TARGET_LONG_DOUBLE_128 && !TARGET_IEEEQUAD)
REAL_MODE_FORMAT (TFmode) = &ibm_extended_format;
ASM_GENERATE_INTERNAL_LABEL (toc_label_name, "LCTOC", 1);
/* We can only guarantee the availability of DI pseudo-ops when
assembling for 64-bit targets. */
if (!TARGET_64BIT)
{
targetm.asm_out.aligned_op.di = NULL;
targetm.asm_out.unaligned_op.di = NULL;
}
/* Set branch target alignment, if not optimizing for size. */
if (!optimize_size)
{
/* Cell wants to be aligned 8byte for dual issue. Titan wants to be
aligned 8byte to avoid misprediction by the branch predictor. */
if (rs6000_tune == PROCESSOR_TITAN
|| rs6000_tune == PROCESSOR_CELL)
{
if (align_functions <= 0)
align_functions = 8;
if (align_jumps <= 0)
align_jumps = 8;
if (align_loops <= 0)
align_loops = 8;
}
if (rs6000_align_branch_targets)
{
if (align_functions <= 0)
align_functions = 16;
if (align_jumps <= 0)
align_jumps = 16;
if (align_loops <= 0)
{
can_override_loop_align = 1;
align_loops = 16;
}
}
if (align_jumps_max_skip <= 0)
align_jumps_max_skip = 15;
if (align_loops_max_skip <= 0)
align_loops_max_skip = 15;
}
/* Arrange to save and restore machine status around nested functions. */
init_machine_status = rs6000_init_machine_status;
/* We should always be splitting complex arguments, but we can't break
Linux and Darwin ABIs at the moment. For now, only AIX is fixed. */
if (DEFAULT_ABI == ABI_V4 || DEFAULT_ABI == ABI_DARWIN)
targetm.calls.split_complex_arg = NULL;
/* The AIX and ELFv1 ABIs define standard function descriptors. */
if (DEFAULT_ABI == ABI_AIX)
targetm.calls.custom_function_descriptors = 0;
}
/* Initialize rs6000_cost with the appropriate target costs. */
if (optimize_size)
rs6000_cost = TARGET_POWERPC64 ? &size64_cost : &size32_cost;
else
switch (rs6000_tune)
{
case PROCESSOR_RS64A:
rs6000_cost = &rs64a_cost;
break;
case PROCESSOR_MPCCORE:
rs6000_cost = &mpccore_cost;
break;
case PROCESSOR_PPC403:
rs6000_cost = &ppc403_cost;
break;
case PROCESSOR_PPC405:
rs6000_cost = &ppc405_cost;
break;
case PROCESSOR_PPC440:
rs6000_cost = &ppc440_cost;
break;
case PROCESSOR_PPC476:
rs6000_cost = &ppc476_cost;
break;
case PROCESSOR_PPC601:
rs6000_cost = &ppc601_cost;
break;
case PROCESSOR_PPC603:
rs6000_cost = &ppc603_cost;
break;
case PROCESSOR_PPC604:
rs6000_cost = &ppc604_cost;
break;
case PROCESSOR_PPC604e:
rs6000_cost = &ppc604e_cost;
break;
case PROCESSOR_PPC620:
rs6000_cost = &ppc620_cost;
break;
case PROCESSOR_PPC630:
rs6000_cost = &ppc630_cost;
break;
case PROCESSOR_CELL:
rs6000_cost = &ppccell_cost;
break;
case PROCESSOR_PPC750:
case PROCESSOR_PPC7400:
rs6000_cost = &ppc750_cost;
break;
case PROCESSOR_PPC7450:
rs6000_cost = &ppc7450_cost;
break;
case PROCESSOR_PPC8540:
case PROCESSOR_PPC8548:
rs6000_cost = &ppc8540_cost;
break;
case PROCESSOR_PPCE300C2:
case PROCESSOR_PPCE300C3:
rs6000_cost = &ppce300c2c3_cost;
break;
case PROCESSOR_PPCE500MC:
rs6000_cost = &ppce500mc_cost;
break;
case PROCESSOR_PPCE500MC64:
rs6000_cost = &ppce500mc64_cost;
break;
case PROCESSOR_PPCE5500:
rs6000_cost = &ppce5500_cost;
break;
case PROCESSOR_PPCE6500:
rs6000_cost = &ppce6500_cost;
break;
case PROCESSOR_TITAN:
rs6000_cost = &titan_cost;
break;
case PROCESSOR_POWER4:
case PROCESSOR_POWER5:
rs6000_cost = &power4_cost;
break;
case PROCESSOR_POWER6:
rs6000_cost = &power6_cost;
break;
case PROCESSOR_POWER7:
rs6000_cost = &power7_cost;
break;
case PROCESSOR_POWER8:
rs6000_cost = &power8_cost;
break;
case PROCESSOR_POWER9:
rs6000_cost = &power9_cost;
break;
case PROCESSOR_PPCA2:
rs6000_cost = &ppca2_cost;
break;
default:
gcc_unreachable ();
}
if (global_init_p)
{
maybe_set_param_value (PARAM_SIMULTANEOUS_PREFETCHES,
rs6000_cost->simultaneous_prefetches,
global_options.x_param_values,
global_options_set.x_param_values);
maybe_set_param_value (PARAM_L1_CACHE_SIZE, rs6000_cost->l1_cache_size,
global_options.x_param_values,
global_options_set.x_param_values);
maybe_set_param_value (PARAM_L1_CACHE_LINE_SIZE,
rs6000_cost->cache_line_size,
global_options.x_param_values,
global_options_set.x_param_values);
maybe_set_param_value (PARAM_L2_CACHE_SIZE, rs6000_cost->l2_cache_size,
global_options.x_param_values,
global_options_set.x_param_values);
/* Increase loop peeling limits based on performance analysis. */
maybe_set_param_value (PARAM_MAX_PEELED_INSNS, 400,
global_options.x_param_values,
global_options_set.x_param_values);
maybe_set_param_value (PARAM_MAX_COMPLETELY_PEELED_INSNS, 400,
global_options.x_param_values,
global_options_set.x_param_values);
/* Use the 'model' -fsched-pressure algorithm by default. */
maybe_set_param_value (PARAM_SCHED_PRESSURE_ALGORITHM,
SCHED_PRESSURE_MODEL,
global_options.x_param_values,
global_options_set.x_param_values);
/* If using typedef char *va_list, signal that
__builtin_va_start (&ap, 0) can be optimized to
ap = __builtin_next_arg (0). */
if (DEFAULT_ABI != ABI_V4)
targetm.expand_builtin_va_start = NULL;
}
/* Set up single/double float flags.
If TARGET_HARD_FLOAT is set, but neither single or double is set,
then set both flags. */
if (TARGET_HARD_FLOAT && rs6000_single_float == 0 && rs6000_double_float == 0)
rs6000_single_float = rs6000_double_float = 1;
/* If not explicitly specified via option, decide whether to generate indexed
load/store instructions. A value of -1 indicates that the
initial value of this variable has not been overwritten. During
compilation, TARGET_AVOID_XFORM is either 0 or 1. */
if (TARGET_AVOID_XFORM == -1)
/* Avoid indexed addressing when targeting Power6 in order to avoid the
DERAT mispredict penalty. However the LVE and STVE altivec instructions
need indexed accesses and the type used is the scalar type of the element
being loaded or stored. */
TARGET_AVOID_XFORM = (rs6000_tune == PROCESSOR_POWER6 && TARGET_CMPB
&& !TARGET_ALTIVEC);
/* Set the -mrecip options. */
if (rs6000_recip_name)
{
char *p = ASTRDUP (rs6000_recip_name);
char *q;
unsigned int mask, i;
bool invert;
while ((q = strtok (p, ",")) != NULL)
{
p = NULL;
if (*q == '!')
{
invert = true;
q++;
}
else
invert = false;
if (!strcmp (q, "default"))
mask = ((TARGET_RECIP_PRECISION)
? RECIP_HIGH_PRECISION : RECIP_LOW_PRECISION);
else
{
for (i = 0; i < ARRAY_SIZE (recip_options); i++)
if (!strcmp (q, recip_options[i].string))
{
mask = recip_options[i].mask;
break;
}
if (i == ARRAY_SIZE (recip_options))
{
error ("unknown option for %<%s=%s%>", "-mrecip", q);
invert = false;
mask = 0;
ret = false;
}
}
if (invert)
rs6000_recip_control &= ~mask;
else
rs6000_recip_control |= mask;
}
}
/* Set the builtin mask of the various options used that could affect which
builtins were used. In the past we used target_flags, but we've run out
of bits, and some options like PAIRED are no longer in target_flags. */
rs6000_builtin_mask = rs6000_builtin_mask_calculate ();
if (TARGET_DEBUG_BUILTIN || TARGET_DEBUG_TARGET)
rs6000_print_builtin_options (stderr, 0, "builtin mask",
rs6000_builtin_mask);
/* Initialize all of the registers. */
rs6000_init_hard_regno_mode_ok (global_init_p);
/* Save the initial options in case the user does function specific options */
if (global_init_p)
target_option_default_node = target_option_current_node
= build_target_option_node (&global_options);
/* If not explicitly specified via option, decide whether to generate the
extra blr's required to preserve the link stack on some cpus (eg, 476). */
if (TARGET_LINK_STACK == -1)
SET_TARGET_LINK_STACK (rs6000_tune == PROCESSOR_PPC476 && flag_pic);
return ret;
}
/* Implement TARGET_OPTION_OVERRIDE. On the RS/6000 this is used to
define the target cpu type. */
static void
rs6000_option_override (void)
{
(void) rs6000_option_override_internal (true);
}
/* Implement targetm.vectorize.builtin_mask_for_load. */
static tree
rs6000_builtin_mask_for_load (void)
{
/* Don't use lvsl/vperm for P8 and similarly efficient machines. */
if ((TARGET_ALTIVEC && !TARGET_VSX)
|| (TARGET_VSX && !TARGET_EFFICIENT_UNALIGNED_VSX))
return altivec_builtin_mask_for_load;
else
return 0;
}
/* Implement LOOP_ALIGN. */
int
rs6000_loop_align (rtx label)
{
basic_block bb;
int ninsns;
/* Don't override loop alignment if -falign-loops was specified. */
if (!can_override_loop_align)
return align_loops_log;
bb = BLOCK_FOR_INSN (label);
ninsns = num_loop_insns(bb->loop_father);
/* Align small loops to 32 bytes to fit in an icache sector, otherwise return default. */
if (ninsns > 4 && ninsns <= 8
&& (rs6000_tune == PROCESSOR_POWER4
|| rs6000_tune == PROCESSOR_POWER5
|| rs6000_tune == PROCESSOR_POWER6
|| rs6000_tune == PROCESSOR_POWER7
|| rs6000_tune == PROCESSOR_POWER8
|| rs6000_tune == PROCESSOR_POWER9))
return 5;
else
return align_loops_log;
}
/* Implement TARGET_LOOP_ALIGN_MAX_SKIP. */
static int
rs6000_loop_align_max_skip (rtx_insn *label)
{
return (1 << rs6000_loop_align (label)) - 1;
}
/* Return true iff, data reference of TYPE can reach vector alignment (16)
after applying N number of iterations. This routine does not determine
how may iterations are required to reach desired alignment. */
static bool
rs6000_vector_alignment_reachable (const_tree type ATTRIBUTE_UNUSED, bool is_packed)
{
if (is_packed)
return false;
if (TARGET_32BIT)
{
if (rs6000_alignment_flags == MASK_ALIGN_NATURAL)
return true;
if (rs6000_alignment_flags == MASK_ALIGN_POWER)
return true;
return false;
}
else
{
if (TARGET_MACHO)
return false;
/* Assuming that all other types are naturally aligned. CHECKME! */
return true;
}
}
/* Return true if the vector misalignment factor is supported by the
target. */
static bool
rs6000_builtin_support_vector_misalignment (machine_mode mode,
const_tree type,
int misalignment,
bool is_packed)
{
if (TARGET_VSX)
{
if (TARGET_EFFICIENT_UNALIGNED_VSX)
return true;
/* Return if movmisalign pattern is not supported for this mode. */
if (optab_handler (movmisalign_optab, mode) == CODE_FOR_nothing)
return false;
if (misalignment == -1)
{
/* Misalignment factor is unknown at compile time but we know
it's word aligned. */
if (rs6000_vector_alignment_reachable (type, is_packed))
{
int element_size = TREE_INT_CST_LOW (TYPE_SIZE (type));
if (element_size == 64 || element_size == 32)
return true;
}
return false;
}
/* VSX supports word-aligned vector. */
if (misalignment % 4 == 0)
return true;
}
return false;
}
/* Implement targetm.vectorize.builtin_vectorization_cost. */
static int
rs6000_builtin_vectorization_cost (enum vect_cost_for_stmt type_of_cost,
tree vectype, int misalign)
{
unsigned elements;
tree elem_type;
switch (type_of_cost)
{
case scalar_stmt:
case scalar_load:
case scalar_store:
case vector_stmt:
case vector_load:
case vector_store:
case vec_to_scalar:
case scalar_to_vec:
case cond_branch_not_taken:
return 1;
case vec_perm:
if (TARGET_VSX)
return 3;
else
return 1;
case vec_promote_demote:
if (TARGET_VSX)
return 4;
else
return 1;
case cond_branch_taken:
return 3;
case unaligned_load:
case vector_gather_load:
if (TARGET_EFFICIENT_UNALIGNED_VSX)
return 1;
if (TARGET_VSX && TARGET_ALLOW_MOVMISALIGN)
{
elements = TYPE_VECTOR_SUBPARTS (vectype);
if (elements == 2)
/* Double word aligned. */
return 2;
if (elements == 4)
{
switch (misalign)
{
case 8:
/* Double word aligned. */
return 2;
case -1:
/* Unknown misalignment. */
case 4:
case 12:
/* Word aligned. */
return 22;
default:
gcc_unreachable ();
}
}
}
if (TARGET_ALTIVEC)
/* Misaligned loads are not supported. */
gcc_unreachable ();
return 2;
case unaligned_store:
case vector_scatter_store:
if (TARGET_EFFICIENT_UNALIGNED_VSX)
return 1;
if (TARGET_VSX && TARGET_ALLOW_MOVMISALIGN)
{
elements = TYPE_VECTOR_SUBPARTS (vectype);
if (elements == 2)
/* Double word aligned. */
return 2;
if (elements == 4)
{
switch (misalign)
{
case 8:
/* Double word aligned. */
return 2;
case -1:
/* Unknown misalignment. */
case 4:
case 12:
/* Word aligned. */
return 23;
default:
gcc_unreachable ();
}
}
}
if (TARGET_ALTIVEC)
/* Misaligned stores are not supported. */
gcc_unreachable ();
return 2;
case vec_construct:
/* This is a rough approximation assuming non-constant elements
constructed into a vector via element insertion. FIXME:
vec_construct is not granular enough for uniformly good
decisions. If the initialization is a splat, this is
cheaper than we estimate. Improve this someday. */
elem_type = TREE_TYPE (vectype);
/* 32-bit vectors loaded into registers are stored as double
precision, so we need 2 permutes, 2 converts, and 1 merge
to construct a vector of short floats from them. */
if (SCALAR_FLOAT_TYPE_P (elem_type)
&& TYPE_PRECISION (elem_type) == 32)
return 5;
/* On POWER9, integer vector types are built up in GPRs and then
use a direct move (2 cycles). For POWER8 this is even worse,
as we need two direct moves and a merge, and the direct moves
are five cycles. */
else if (INTEGRAL_TYPE_P (elem_type))
{
if (TARGET_P9_VECTOR)
return TYPE_VECTOR_SUBPARTS (vectype) - 1 + 2;
else
return TYPE_VECTOR_SUBPARTS (vectype) - 1 + 5;
}
else
/* V2DFmode doesn't need a direct move. */
return 2;
default:
gcc_unreachable ();
}
}
/* Implement targetm.vectorize.preferred_simd_mode. */
static machine_mode
rs6000_preferred_simd_mode (scalar_mode mode)
{
if (TARGET_VSX)
switch (mode)
{
case E_DFmode:
return V2DFmode;
default:;
}
if (TARGET_ALTIVEC || TARGET_VSX)
switch (mode)
{
case E_SFmode:
return V4SFmode;
case E_TImode:
return V1TImode;
case E_DImode:
return V2DImode;
case E_SImode:
return V4SImode;
case E_HImode:
return V8HImode;
case E_QImode:
return V16QImode;
default:;
}
if (TARGET_PAIRED_FLOAT
&& mode == SFmode)
return V2SFmode;
return word_mode;
}
typedef struct _rs6000_cost_data
{
struct loop *loop_info;
unsigned cost[3];
} rs6000_cost_data;
/* Test for likely overcommitment of vector hardware resources. If a
loop iteration is relatively large, and too large a percentage of
instructions in the loop are vectorized, the cost model may not
adequately reflect delays from unavailable vector resources.
Penalize the loop body cost for this case. */
static void
rs6000_density_test (rs6000_cost_data *data)
{
const int DENSITY_PCT_THRESHOLD = 85;
const int DENSITY_SIZE_THRESHOLD = 70;
const int DENSITY_PENALTY = 10;
struct loop *loop = data->loop_info;
basic_block *bbs = get_loop_body (loop);
int nbbs = loop->num_nodes;
int vec_cost = data->cost[vect_body], not_vec_cost = 0;
int i, density_pct;
for (i = 0; i < nbbs; i++)
{
basic_block bb = bbs[i];
gimple_stmt_iterator gsi;
for (gsi = gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi))
{
gimple *stmt = gsi_stmt (gsi);
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
if (!STMT_VINFO_RELEVANT_P (stmt_info)
&& !STMT_VINFO_IN_PATTERN_P (stmt_info))
not_vec_cost++;
}
}
free (bbs);
density_pct = (vec_cost * 100) / (vec_cost + not_vec_cost);
if (density_pct > DENSITY_PCT_THRESHOLD
&& vec_cost + not_vec_cost > DENSITY_SIZE_THRESHOLD)
{
data->cost[vect_body] = vec_cost * (100 + DENSITY_PENALTY) / 100;
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"density %d%%, cost %d exceeds threshold, penalizing "
"loop body cost by %d%%", density_pct,
vec_cost + not_vec_cost, DENSITY_PENALTY);
}
}
/* Implement targetm.vectorize.init_cost. */
/* For each vectorized loop, this var holds TRUE iff a non-memory vector
instruction is needed by the vectorization. */
static bool rs6000_vect_nonmem;
static void *
rs6000_init_cost (struct loop *loop_info)
{
rs6000_cost_data *data = XNEW (struct _rs6000_cost_data);
data->loop_info = loop_info;
data->cost[vect_prologue] = 0;
data->cost[vect_body] = 0;
data->cost[vect_epilogue] = 0;
rs6000_vect_nonmem = false;
return data;
}
/* Implement targetm.vectorize.add_stmt_cost. */
static unsigned
rs6000_add_stmt_cost (void *data, int count, enum vect_cost_for_stmt kind,
struct _stmt_vec_info *stmt_info, int misalign,
enum vect_cost_model_location where)
{
rs6000_cost_data *cost_data = (rs6000_cost_data*) data;
unsigned retval = 0;
if (flag_vect_cost_model)
{
tree vectype = stmt_info ? stmt_vectype (stmt_info) : NULL_TREE;
int stmt_cost = rs6000_builtin_vectorization_cost (kind, vectype,
misalign);
/* Statements in an inner loop relative to the loop being
vectorized are weighted more heavily. The value here is
arbitrary and could potentially be improved with analysis. */
if (where == vect_body && stmt_info && stmt_in_inner_loop_p (stmt_info))
count *= 50; /* FIXME. */
retval = (unsigned) (count * stmt_cost);
cost_data->cost[where] += retval;
/* Check whether we're doing something other than just a copy loop.
Not all such loops may be profitably vectorized; see
rs6000_finish_cost. */
if ((kind == vec_to_scalar || kind == vec_perm
|| kind == vec_promote_demote || kind == vec_construct
|| kind == scalar_to_vec)
|| (where == vect_body && kind == vector_stmt))
rs6000_vect_nonmem = true;
}
return retval;
}
/* Implement targetm.vectorize.finish_cost. */
static void
rs6000_finish_cost (void *data, unsigned *prologue_cost,
unsigned *body_cost, unsigned *epilogue_cost)
{
rs6000_cost_data *cost_data = (rs6000_cost_data*) data;
if (cost_data->loop_info)
rs6000_density_test (cost_data);
/* Don't vectorize minimum-vectorization-factor, simple copy loops
that require versioning for any reason. The vectorization is at
best a wash inside the loop, and the versioning checks make
profitability highly unlikely and potentially quite harmful. */
if (cost_data->loop_info)
{
loop_vec_info vec_info = loop_vec_info_for_loop (cost_data->loop_info);
if (!rs6000_vect_nonmem
&& LOOP_VINFO_VECT_FACTOR (vec_info) == 2
&& LOOP_REQUIRES_VERSIONING (vec_info))
cost_data->cost[vect_body] += 10000;
}
*prologue_cost = cost_data->cost[vect_prologue];
*body_cost = cost_data->cost[vect_body];
*epilogue_cost = cost_data->cost[vect_epilogue];
}
/* Implement targetm.vectorize.destroy_cost_data. */
static void
rs6000_destroy_cost_data (void *data)
{
free (data);
}
/* Handler for the Mathematical Acceleration Subsystem (mass) interface to a
library with vectorized intrinsics. */
static tree
rs6000_builtin_vectorized_libmass (combined_fn fn, tree type_out,
tree type_in)
{
char name[32];
const char *suffix = NULL;
tree fntype, new_fndecl, bdecl = NULL_TREE;
int n_args = 1;
const char *bname;
machine_mode el_mode, in_mode;
int n, in_n;
/* Libmass is suitable for unsafe math only as it does not correctly support
parts of IEEE with the required precision such as denormals. Only support
it if we have VSX to use the simd d2 or f4 functions.
XXX: Add variable length support. */
if (!flag_unsafe_math_optimizations || !TARGET_VSX)
return NULL_TREE;
el_mode = TYPE_MODE (TREE_TYPE (type_out));
n = TYPE_VECTOR_SUBPARTS (type_out);
in_mode = TYPE_MODE (TREE_TYPE (type_in));
in_n = TYPE_VECTOR_SUBPARTS (type_in);
if (el_mode != in_mode
|| n != in_n)
return NULL_TREE;
switch (fn)
{
CASE_CFN_ATAN2:
CASE_CFN_HYPOT:
CASE_CFN_POW:
n_args = 2;
gcc_fallthrough ();
CASE_CFN_ACOS:
CASE_CFN_ACOSH:
CASE_CFN_ASIN:
CASE_CFN_ASINH:
CASE_CFN_ATAN:
CASE_CFN_ATANH:
CASE_CFN_CBRT:
CASE_CFN_COS:
CASE_CFN_COSH:
CASE_CFN_ERF:
CASE_CFN_ERFC:
CASE_CFN_EXP2:
CASE_CFN_EXP:
CASE_CFN_EXPM1:
CASE_CFN_LGAMMA:
CASE_CFN_LOG10:
CASE_CFN_LOG1P:
CASE_CFN_LOG2:
CASE_CFN_LOG:
CASE_CFN_SIN:
CASE_CFN_SINH:
CASE_CFN_SQRT:
CASE_CFN_TAN:
CASE_CFN_TANH:
if (el_mode == DFmode && n == 2)
{
bdecl = mathfn_built_in (double_type_node, fn);
suffix = "d2"; /* pow -> powd2 */
}
else if (el_mode == SFmode && n == 4)
{
bdecl = mathfn_built_in (float_type_node, fn);
suffix = "4"; /* powf -> powf4 */
}
else
return NULL_TREE;
if (!bdecl)
return NULL_TREE;
break;
default:
return NULL_TREE;
}
gcc_assert (suffix != NULL);
bname = IDENTIFIER_POINTER (DECL_NAME (bdecl));
if (!bname)
return NULL_TREE;
strcpy (name, bname + sizeof ("__builtin_") - 1);
strcat (name, suffix);
if (n_args == 1)
fntype = build_function_type_list (type_out, type_in, NULL);
else if (n_args == 2)
fntype = build_function_type_list (type_out, type_in, type_in, NULL);
else
gcc_unreachable ();
/* Build a function declaration for the vectorized function. */
new_fndecl = build_decl (BUILTINS_LOCATION,
FUNCTION_DECL, get_identifier (name), fntype);
TREE_PUBLIC (new_fndecl) = 1;
DECL_EXTERNAL (new_fndecl) = 1;
DECL_IS_NOVOPS (new_fndecl) = 1;
TREE_READONLY (new_fndecl) = 1;
return new_fndecl;
}
/* Returns a function decl for a vectorized version of the builtin function
with builtin function code FN and the result vector type TYPE, or NULL_TREE
if it is not available. */
static tree
rs6000_builtin_vectorized_function (unsigned int fn, tree type_out,
tree type_in)
{
machine_mode in_mode, out_mode;
int in_n, out_n;
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin_vectorized_function (%s, %s, %s)\n",
combined_fn_name (combined_fn (fn)),
GET_MODE_NAME (TYPE_MODE (type_out)),
GET_MODE_NAME (TYPE_MODE (type_in)));
if (TREE_CODE (type_out) != VECTOR_TYPE
|| TREE_CODE (type_in) != VECTOR_TYPE)
return NULL_TREE;
out_mode = TYPE_MODE (TREE_TYPE (type_out));
out_n = TYPE_VECTOR_SUBPARTS (type_out);
in_mode = TYPE_MODE (TREE_TYPE (type_in));
in_n = TYPE_VECTOR_SUBPARTS (type_in);
switch (fn)
{
CASE_CFN_COPYSIGN:
if (VECTOR_UNIT_VSX_P (V2DFmode)
&& out_mode == DFmode && out_n == 2
&& in_mode == DFmode && in_n == 2)
return rs6000_builtin_decls[VSX_BUILTIN_CPSGNDP];
if (VECTOR_UNIT_VSX_P (V4SFmode)
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[VSX_BUILTIN_CPSGNSP];
if (VECTOR_UNIT_ALTIVEC_P (V4SFmode)
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[ALTIVEC_BUILTIN_COPYSIGN_V4SF];
break;
CASE_CFN_CEIL:
if (VECTOR_UNIT_VSX_P (V2DFmode)
&& out_mode == DFmode && out_n == 2
&& in_mode == DFmode && in_n == 2)
return rs6000_builtin_decls[VSX_BUILTIN_XVRDPIP];
if (VECTOR_UNIT_VSX_P (V4SFmode)
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[VSX_BUILTIN_XVRSPIP];
if (VECTOR_UNIT_ALTIVEC_P (V4SFmode)
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[ALTIVEC_BUILTIN_VRFIP];
break;
CASE_CFN_FLOOR:
if (VECTOR_UNIT_VSX_P (V2DFmode)
&& out_mode == DFmode && out_n == 2
&& in_mode == DFmode && in_n == 2)
return rs6000_builtin_decls[VSX_BUILTIN_XVRDPIM];
if (VECTOR_UNIT_VSX_P (V4SFmode)
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[VSX_BUILTIN_XVRSPIM];
if (VECTOR_UNIT_ALTIVEC_P (V4SFmode)
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[ALTIVEC_BUILTIN_VRFIM];
break;
CASE_CFN_FMA:
if (VECTOR_UNIT_VSX_P (V2DFmode)
&& out_mode == DFmode && out_n == 2
&& in_mode == DFmode && in_n == 2)
return rs6000_builtin_decls[VSX_BUILTIN_XVMADDDP];
if (VECTOR_UNIT_VSX_P (V4SFmode)
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[VSX_BUILTIN_XVMADDSP];
if (VECTOR_UNIT_ALTIVEC_P (V4SFmode)
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[ALTIVEC_BUILTIN_VMADDFP];
break;
CASE_CFN_TRUNC:
if (VECTOR_UNIT_VSX_P (V2DFmode)
&& out_mode == DFmode && out_n == 2
&& in_mode == DFmode && in_n == 2)
return rs6000_builtin_decls[VSX_BUILTIN_XVRDPIZ];
if (VECTOR_UNIT_VSX_P (V4SFmode)
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[VSX_BUILTIN_XVRSPIZ];
if (VECTOR_UNIT_ALTIVEC_P (V4SFmode)
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[ALTIVEC_BUILTIN_VRFIZ];
break;
CASE_CFN_NEARBYINT:
if (VECTOR_UNIT_VSX_P (V2DFmode)
&& flag_unsafe_math_optimizations
&& out_mode == DFmode && out_n == 2
&& in_mode == DFmode && in_n == 2)
return rs6000_builtin_decls[VSX_BUILTIN_XVRDPI];
if (VECTOR_UNIT_VSX_P (V4SFmode)
&& flag_unsafe_math_optimizations
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[VSX_BUILTIN_XVRSPI];
break;
CASE_CFN_RINT:
if (VECTOR_UNIT_VSX_P (V2DFmode)
&& !flag_trapping_math
&& out_mode == DFmode && out_n == 2
&& in_mode == DFmode && in_n == 2)
return rs6000_builtin_decls[VSX_BUILTIN_XVRDPIC];
if (VECTOR_UNIT_VSX_P (V4SFmode)
&& !flag_trapping_math
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[VSX_BUILTIN_XVRSPIC];
break;
default:
break;
}
/* Generate calls to libmass if appropriate. */
if (rs6000_veclib_handler)
return rs6000_veclib_handler (combined_fn (fn), type_out, type_in);
return NULL_TREE;
}
/* Implement TARGET_VECTORIZE_BUILTIN_MD_VECTORIZED_FUNCTION. */
static tree
rs6000_builtin_md_vectorized_function (tree fndecl, tree type_out,
tree type_in)
{
machine_mode in_mode, out_mode;
int in_n, out_n;
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin_md_vectorized_function (%s, %s, %s)\n",
IDENTIFIER_POINTER (DECL_NAME (fndecl)),
GET_MODE_NAME (TYPE_MODE (type_out)),
GET_MODE_NAME (TYPE_MODE (type_in)));
if (TREE_CODE (type_out) != VECTOR_TYPE
|| TREE_CODE (type_in) != VECTOR_TYPE)
return NULL_TREE;
out_mode = TYPE_MODE (TREE_TYPE (type_out));
out_n = TYPE_VECTOR_SUBPARTS (type_out);
in_mode = TYPE_MODE (TREE_TYPE (type_in));
in_n = TYPE_VECTOR_SUBPARTS (type_in);
enum rs6000_builtins fn
= (enum rs6000_builtins) DECL_FUNCTION_CODE (fndecl);
switch (fn)
{
case RS6000_BUILTIN_RSQRTF:
if (VECTOR_UNIT_ALTIVEC_OR_VSX_P (V4SFmode)
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[ALTIVEC_BUILTIN_VRSQRTFP];
break;
case RS6000_BUILTIN_RSQRT:
if (VECTOR_UNIT_VSX_P (V2DFmode)
&& out_mode == DFmode && out_n == 2
&& in_mode == DFmode && in_n == 2)
return rs6000_builtin_decls[VSX_BUILTIN_RSQRT_2DF];
break;
case RS6000_BUILTIN_RECIPF:
if (VECTOR_UNIT_ALTIVEC_OR_VSX_P (V4SFmode)
&& out_mode == SFmode && out_n == 4
&& in_mode == SFmode && in_n == 4)
return rs6000_builtin_decls[ALTIVEC_BUILTIN_VRECIPFP];
break;
case RS6000_BUILTIN_RECIP:
if (VECTOR_UNIT_VSX_P (V2DFmode)
&& out_mode == DFmode && out_n == 2
&& in_mode == DFmode && in_n == 2)
return rs6000_builtin_decls[VSX_BUILTIN_RECIP_V2DF];
break;
default:
break;
}
return NULL_TREE;
}
/* Default CPU string for rs6000*_file_start functions. */
static const char *rs6000_default_cpu;
/* Do anything needed at the start of the asm file. */
static void
rs6000_file_start (void)
{
char buffer[80];
const char *start = buffer;
FILE *file = asm_out_file;
rs6000_default_cpu = TARGET_CPU_DEFAULT;
default_file_start ();
if (flag_verbose_asm)
{
sprintf (buffer, "\n%s rs6000/powerpc options:", ASM_COMMENT_START);
if (rs6000_default_cpu != 0 && rs6000_default_cpu[0] != '\0')
{
fprintf (file, "%s --with-cpu=%s", start, rs6000_default_cpu);
start = "";
}
if (global_options_set.x_rs6000_cpu_index)
{
fprintf (file, "%s -mcpu=%s", start,
processor_target_table[rs6000_cpu_index].name);
start = "";
}
if (global_options_set.x_rs6000_tune_index)
{
fprintf (file, "%s -mtune=%s", start,
processor_target_table[rs6000_tune_index].name);
start = "";
}
if (PPC405_ERRATUM77)
{
fprintf (file, "%s PPC405CR_ERRATUM77", start);
start = "";
}
#ifdef USING_ELFOS_H
switch (rs6000_sdata)
{
case SDATA_NONE: fprintf (file, "%s -msdata=none", start); start = ""; break;
case SDATA_DATA: fprintf (file, "%s -msdata=data", start); start = ""; break;
case SDATA_SYSV: fprintf (file, "%s -msdata=sysv", start); start = ""; break;
case SDATA_EABI: fprintf (file, "%s -msdata=eabi", start); start = ""; break;
}
if (rs6000_sdata && g_switch_value)
{
fprintf (file, "%s -G %d", start,
g_switch_value);
start = "";
}
#endif
if (*start == '\0')
putc ('\n', file);
}
#ifdef USING_ELFOS_H
if (!(rs6000_default_cpu && rs6000_default_cpu[0])
&& !global_options_set.x_rs6000_cpu_index)
{
fputs ("\t.machine ", asm_out_file);
if ((rs6000_isa_flags & OPTION_MASK_MODULO) != 0)
fputs ("power9\n", asm_out_file);
else if ((rs6000_isa_flags & OPTION_MASK_DIRECT_MOVE) != 0)
fputs ("power8\n", asm_out_file);
else if ((rs6000_isa_flags & OPTION_MASK_POPCNTD) != 0)
fputs ("power7\n", asm_out_file);
else if ((rs6000_isa_flags & OPTION_MASK_CMPB) != 0)
fputs ("power6\n", asm_out_file);
else if ((rs6000_isa_flags & OPTION_MASK_POPCNTB) != 0)
fputs ("power5\n", asm_out_file);
else if ((rs6000_isa_flags & OPTION_MASK_MFCRF) != 0)
fputs ("power4\n", asm_out_file);
else if ((rs6000_isa_flags & OPTION_MASK_POWERPC64) != 0)
fputs ("ppc64\n", asm_out_file);
else
fputs ("ppc\n", asm_out_file);
}
#endif
if (DEFAULT_ABI == ABI_ELFv2)
fprintf (file, "\t.abiversion 2\n");
}
/* Return nonzero if this function is known to have a null epilogue. */
int
direct_return (void)
{
if (reload_completed)
{
rs6000_stack_t *info = rs6000_stack_info ();
if (info->first_gp_reg_save == 32
&& info->first_fp_reg_save == 64
&& info->first_altivec_reg_save == LAST_ALTIVEC_REGNO + 1
&& ! info->lr_save_p
&& ! info->cr_save_p
&& info->vrsave_size == 0
&& ! info->push_p)
return 1;
}
return 0;
}
/* Return the number of instructions it takes to form a constant in an
integer register. */
int
num_insns_constant_wide (HOST_WIDE_INT value)
{
/* signed constant loadable with addi */
if (((unsigned HOST_WIDE_INT) value + 0x8000) < 0x10000)
return 1;
/* constant loadable with addis */
else if ((value & 0xffff) == 0
&& (value >> 31 == -1 || value >> 31 == 0))
return 1;
else if (TARGET_POWERPC64)
{
HOST_WIDE_INT low = ((value & 0xffffffff) ^ 0x80000000) - 0x80000000;
HOST_WIDE_INT high = value >> 31;
if (high == 0 || high == -1)
return 2;
high >>= 1;
if (low == 0)
return num_insns_constant_wide (high) + 1;
else if (high == 0)
return num_insns_constant_wide (low) + 1;
else
return (num_insns_constant_wide (high)
+ num_insns_constant_wide (low) + 1);
}
else
return 2;
}
int
num_insns_constant (rtx op, machine_mode mode)
{
HOST_WIDE_INT low, high;
switch (GET_CODE (op))
{
case CONST_INT:
if ((INTVAL (op) >> 31) != 0 && (INTVAL (op) >> 31) != -1
&& rs6000_is_valid_and_mask (op, mode))
return 2;
else
return num_insns_constant_wide (INTVAL (op));
case CONST_WIDE_INT:
{
int i;
int ins = CONST_WIDE_INT_NUNITS (op) - 1;
for (i = 0; i < CONST_WIDE_INT_NUNITS (op); i++)
ins += num_insns_constant_wide (CONST_WIDE_INT_ELT (op, i));
return ins;
}
case CONST_DOUBLE:
if (mode == SFmode || mode == SDmode)
{
long l;
if (DECIMAL_FLOAT_MODE_P (mode))
REAL_VALUE_TO_TARGET_DECIMAL32
(*CONST_DOUBLE_REAL_VALUE (op), l);
else
REAL_VALUE_TO_TARGET_SINGLE (*CONST_DOUBLE_REAL_VALUE (op), l);
return num_insns_constant_wide ((HOST_WIDE_INT) l);
}
long l[2];
if (DECIMAL_FLOAT_MODE_P (mode))
REAL_VALUE_TO_TARGET_DECIMAL64 (*CONST_DOUBLE_REAL_VALUE (op), l);
else
REAL_VALUE_TO_TARGET_DOUBLE (*CONST_DOUBLE_REAL_VALUE (op), l);
high = l[WORDS_BIG_ENDIAN == 0];
low = l[WORDS_BIG_ENDIAN != 0];
if (TARGET_32BIT)
return (num_insns_constant_wide (low)
+ num_insns_constant_wide (high));
else
{
if ((high == 0 && low >= 0)
|| (high == -1 && low < 0))
return num_insns_constant_wide (low);
else if (rs6000_is_valid_and_mask (op, mode))
return 2;
else if (low == 0)
return num_insns_constant_wide (high) + 1;
else
return (num_insns_constant_wide (high)
+ num_insns_constant_wide (low) + 1);
}
default:
gcc_unreachable ();
}
}
/* Interpret element ELT of the CONST_VECTOR OP as an integer value.
If the mode of OP is MODE_VECTOR_INT, this simply returns the
corresponding element of the vector, but for V4SFmode and V2SFmode,
the corresponding "float" is interpreted as an SImode integer. */
HOST_WIDE_INT
const_vector_elt_as_int (rtx op, unsigned int elt)
{
rtx tmp;
/* We can't handle V2DImode and V2DFmode vector constants here yet. */
gcc_assert (GET_MODE (op) != V2DImode
&& GET_MODE (op) != V2DFmode);
tmp = CONST_VECTOR_ELT (op, elt);
if (GET_MODE (op) == V4SFmode
|| GET_MODE (op) == V2SFmode)
tmp = gen_lowpart (SImode, tmp);
return INTVAL (tmp);
}
/* Return true if OP can be synthesized with a particular vspltisb, vspltish
or vspltisw instruction. OP is a CONST_VECTOR. Which instruction is used
depends on STEP and COPIES, one of which will be 1. If COPIES > 1,
all items are set to the same value and contain COPIES replicas of the
vsplt's operand; if STEP > 1, one in STEP elements is set to the vsplt's
operand and the others are set to the value of the operand's msb. */
static bool
vspltis_constant (rtx op, unsigned step, unsigned copies)
{
machine_mode mode = GET_MODE (op);
machine_mode inner = GET_MODE_INNER (mode);
unsigned i;
unsigned nunits;
unsigned bitsize;
unsigned mask;
HOST_WIDE_INT val;
HOST_WIDE_INT splat_val;
HOST_WIDE_INT msb_val;
if (mode == V2DImode || mode == V2DFmode || mode == V1TImode)
return false;
nunits = GET_MODE_NUNITS (mode);
bitsize = GET_MODE_BITSIZE (inner);
mask = GET_MODE_MASK (inner);
val = const_vector_elt_as_int (op, BYTES_BIG_ENDIAN ? nunits - 1 : 0);
splat_val = val;
msb_val = val >= 0 ? 0 : -1;
/* Construct the value to be splatted, if possible. If not, return 0. */
for (i = 2; i <= copies; i *= 2)
{
HOST_WIDE_INT small_val;
bitsize /= 2;
small_val = splat_val >> bitsize;
mask >>= bitsize;
if (splat_val != ((HOST_WIDE_INT)
((unsigned HOST_WIDE_INT) small_val << bitsize)
| (small_val & mask)))
return false;
splat_val = small_val;
}
/* Check if SPLAT_VAL can really be the operand of a vspltis[bhw]. */
if (EASY_VECTOR_15 (splat_val))
;
/* Also check if we can splat, and then add the result to itself. Do so if
the value is positive, of if the splat instruction is using OP's mode;
for splat_val < 0, the splat and the add should use the same mode. */
else if (EASY_VECTOR_15_ADD_SELF (splat_val)
&& (splat_val >= 0 || (step == 1 && copies == 1)))
;
/* Also check if are loading up the most significant bit which can be done by
loading up -1 and shifting the value left by -1. */
else if (EASY_VECTOR_MSB (splat_val, inner))
;
else
return false;
/* Check if VAL is present in every STEP-th element, and the
other elements are filled with its most significant bit. */
for (i = 1; i < nunits; ++i)
{
HOST_WIDE_INT desired_val;
unsigned elt = BYTES_BIG_ENDIAN ? nunits - 1 - i : i;
if ((i & (step - 1)) == 0)
desired_val = val;
else
desired_val = msb_val;
if (desired_val != const_vector_elt_as_int (op, elt))
return false;
}
return true;
}
/* Like vsplitis_constant, but allow the value to be shifted left with a VSLDOI
instruction, filling in the bottom elements with 0 or -1.
Return 0 if the constant cannot be generated with VSLDOI. Return positive
for the number of zeroes to shift in, or negative for the number of 0xff
bytes to shift in.
OP is a CONST_VECTOR. */
int
vspltis_shifted (rtx op)
{
machine_mode mode = GET_MODE (op);
machine_mode inner = GET_MODE_INNER (mode);
unsigned i, j;
unsigned nunits;
unsigned mask;
HOST_WIDE_INT val;
if (mode != V16QImode && mode != V8HImode && mode != V4SImode)
return false;
/* We need to create pseudo registers to do the shift, so don't recognize
shift vector constants after reload. */
if (!can_create_pseudo_p ())
return false;
nunits = GET_MODE_NUNITS (mode);
mask = GET_MODE_MASK (inner);
val = const_vector_elt_as_int (op, BYTES_BIG_ENDIAN ? 0 : nunits - 1);
/* Check if the value can really be the operand of a vspltis[bhw]. */
if (EASY_VECTOR_15 (val))
;
/* Also check if we are loading up the most significant bit which can be done
by loading up -1 and shifting the value left by -1. */
else if (EASY_VECTOR_MSB (val, inner))
;
else
return 0;
/* Check if VAL is present in every STEP-th element until we find elements
that are 0 or all 1 bits. */
for (i = 1; i < nunits; ++i)
{
unsigned elt = BYTES_BIG_ENDIAN ? i : nunits - 1 - i;
HOST_WIDE_INT elt_val = const_vector_elt_as_int (op, elt);
/* If the value isn't the splat value, check for the remaining elements
being 0/-1. */
if (val != elt_val)
{
if (elt_val == 0)
{
for (j = i+1; j < nunits; ++j)
{
unsigned elt2 = BYTES_BIG_ENDIAN ? j : nunits - 1 - j;
if (const_vector_elt_as_int (op, elt2) != 0)
return 0;
}
return (nunits - i) * GET_MODE_SIZE (inner);
}
else if ((elt_val & mask) == mask)
{
for (j = i+1; j < nunits; ++j)
{
unsigned elt2 = BYTES_BIG_ENDIAN ? j : nunits - 1 - j;
if ((const_vector_elt_as_int (op, elt2) & mask) != mask)
return 0;
}
return -((nunits - i) * GET_MODE_SIZE (inner));
}
else
return 0;
}
}
/* If all elements are equal, we don't need to do VLSDOI. */
return 0;
}
/* Return true if OP is of the given MODE and can be synthesized
with a vspltisb, vspltish or vspltisw. */
bool
easy_altivec_constant (rtx op, machine_mode mode)
{
unsigned step, copies;
if (mode == VOIDmode)
mode = GET_MODE (op);
else if (mode != GET_MODE (op))
return false;
/* V2DI/V2DF was added with VSX. Only allow 0 and all 1's as easy
constants. */
if (mode == V2DFmode)
return zero_constant (op, mode);
else if (mode == V2DImode)
{
if (GET_CODE (CONST_VECTOR_ELT (op, 0)) != CONST_INT
|| GET_CODE (CONST_VECTOR_ELT (op, 1)) != CONST_INT)
return false;
if (zero_constant (op, mode))
return true;
if (INTVAL (CONST_VECTOR_ELT (op, 0)) == -1
&& INTVAL (CONST_VECTOR_ELT (op, 1)) == -1)
return true;
return false;
}
/* V1TImode is a special container for TImode. Ignore for now. */
else if (mode == V1TImode)
return false;
/* Start with a vspltisw. */
step = GET_MODE_NUNITS (mode) / 4;
copies = 1;
if (vspltis_constant (op, step, copies))
return true;
/* Then try with a vspltish. */
if (step == 1)
copies <<= 1;
else
step >>= 1;
if (vspltis_constant (op, step, copies))
return true;
/* And finally a vspltisb. */
if (step == 1)
copies <<= 1;
else
step >>= 1;
if (vspltis_constant (op, step, copies))
return true;
if (vspltis_shifted (op) != 0)
return true;
return false;
}
/* Generate a VEC_DUPLICATE representing a vspltis[bhw] instruction whose
result is OP. Abort if it is not possible. */
rtx
gen_easy_altivec_constant (rtx op)
{
machine_mode mode = GET_MODE (op);
int nunits = GET_MODE_NUNITS (mode);
rtx val = CONST_VECTOR_ELT (op, BYTES_BIG_ENDIAN ? nunits - 1 : 0);
unsigned step = nunits / 4;
unsigned copies = 1;
/* Start with a vspltisw. */
if (vspltis_constant (op, step, copies))
return gen_rtx_VEC_DUPLICATE (V4SImode, gen_lowpart (SImode, val));
/* Then try with a vspltish. */
if (step == 1)
copies <<= 1;
else
step >>= 1;
if (vspltis_constant (op, step, copies))
return gen_rtx_VEC_DUPLICATE (V8HImode, gen_lowpart (HImode, val));
/* And finally a vspltisb. */
if (step == 1)
copies <<= 1;
else
step >>= 1;
if (vspltis_constant (op, step, copies))
return gen_rtx_VEC_DUPLICATE (V16QImode, gen_lowpart (QImode, val));
gcc_unreachable ();
}
/* Return true if OP is of the given MODE and can be synthesized with ISA 3.0
instructions (xxspltib, vupkhsb/vextsb2w/vextb2d).
Return the number of instructions needed (1 or 2) into the address pointed
via NUM_INSNS_PTR.
Return the constant that is being split via CONSTANT_PTR. */
bool
xxspltib_constant_p (rtx op,
machine_mode mode,
int *num_insns_ptr,
int *constant_ptr)
{
size_t nunits = GET_MODE_NUNITS (mode);
size_t i;
HOST_WIDE_INT value;
rtx element;
/* Set the returned values to out of bound values. */
*num_insns_ptr = -1;
*constant_ptr = 256;
if (!TARGET_P9_VECTOR)
return false;
if (mode == VOIDmode)
mode = GET_MODE (op);
else if (mode != GET_MODE (op) && GET_MODE (op) != VOIDmode)
return false;
/* Handle (vec_duplicate <constant>). */
if (GET_CODE (op) == VEC_DUPLICATE)
{
if (mode != V16QImode && mode != V8HImode && mode != V4SImode
&& mode != V2DImode)
return false;
element = XEXP (op, 0);
if (!CONST_INT_P (element))
return false;
value = INTVAL (element);
if (!IN_RANGE (value, -128, 127))
return false;
}
/* Handle (const_vector [...]). */
else if (GET_CODE (op) == CONST_VECTOR)
{
if (mode != V16QImode && mode != V8HImode && mode != V4SImode
&& mode != V2DImode)
return false;
element = CONST_VECTOR_ELT (op, 0);
if (!CONST_INT_P (element))
return false;
value = INTVAL (element);
if (!IN_RANGE (value, -128, 127))
return false;
for (i = 1; i < nunits; i++)
{
element = CONST_VECTOR_ELT (op, i);
if (!CONST_INT_P (element))
return false;
if (value != INTVAL (element))
return false;
}
}
/* Handle integer constants being loaded into the upper part of the VSX
register as a scalar. If the value isn't 0/-1, only allow it if the mode
can go in Altivec registers. Prefer VSPLTISW/VUPKHSW over XXSPLITIB. */
else if (CONST_INT_P (op))
{
if (!SCALAR_INT_MODE_P (mode))
return false;
value = INTVAL (op);
if (!IN_RANGE (value, -128, 127))
return false;
if (!IN_RANGE (value, -1, 0))
{
if (!(reg_addr[mode].addr_mask[RELOAD_REG_VMX] & RELOAD_REG_VALID))
return false;
if (EASY_VECTOR_15 (value))
return false;
}
}
else
return false;
/* See if we could generate vspltisw/vspltish directly instead of xxspltib +
sign extend. Special case 0/-1 to allow getting any VSX register instead
of an Altivec register. */
if ((mode == V4SImode || mode == V8HImode) && !IN_RANGE (value, -1, 0)
&& EASY_VECTOR_15 (value))
return false;
/* Return # of instructions and the constant byte for XXSPLTIB. */
if (mode == V16QImode)
*num_insns_ptr = 1;
else if (IN_RANGE (value, -1, 0))
*num_insns_ptr = 1;
else
*num_insns_ptr = 2;
*constant_ptr = (int) value;
return true;
}
const char *
output_vec_const_move (rtx *operands)
{
int shift;
machine_mode mode;
rtx dest, vec;
dest = operands[0];
vec = operands[1];
mode = GET_MODE (dest);
if (TARGET_VSX)
{
bool dest_vmx_p = ALTIVEC_REGNO_P (REGNO (dest));
int xxspltib_value = 256;
int num_insns = -1;
if (zero_constant (vec, mode))
{
if (TARGET_P9_VECTOR)
return "xxspltib %x0,0";
else if (dest_vmx_p)
return "vspltisw %0,0";
else
return "xxlxor %x0,%x0,%x0";
}
if (all_ones_constant (vec, mode))
{
if (TARGET_P9_VECTOR)
return "xxspltib %x0,255";
else if (dest_vmx_p)
return "vspltisw %0,-1";
else if (TARGET_P8_VECTOR)
return "xxlorc %x0,%x0,%x0";
else
gcc_unreachable ();
}
if (TARGET_P9_VECTOR
&& xxspltib_constant_p (vec, mode, &num_insns, &xxspltib_value))
{
if (num_insns == 1)
{
operands[2] = GEN_INT (xxspltib_value & 0xff);
return "xxspltib %x0,%2";
}
return "#";
}
}
if (TARGET_ALTIVEC)
{
rtx splat_vec;
gcc_assert (ALTIVEC_REGNO_P (REGNO (dest)));
if (zero_constant (vec, mode))
return "vspltisw %0,0";
if (all_ones_constant (vec, mode))
return "vspltisw %0,-1";
/* Do we need to construct a value using VSLDOI? */
shift = vspltis_shifted (vec);
if (shift != 0)
return "#";
splat_vec = gen_easy_altivec_constant (vec);
gcc_assert (GET_CODE (splat_vec) == VEC_DUPLICATE);
operands[1] = XEXP (splat_vec, 0);
if (!EASY_VECTOR_15 (INTVAL (operands[1])))
return "#";
switch (GET_MODE (splat_vec))
{
case E_V4SImode:
return "vspltisw %0,%1";
case E_V8HImode:
return "vspltish %0,%1";
case E_V16QImode:
return "vspltisb %0,%1";
default:
gcc_unreachable ();
}
}
gcc_unreachable ();
}
/* Initialize TARGET of vector PAIRED to VALS. */
void
paired_expand_vector_init (rtx target, rtx vals)
{
machine_mode mode = GET_MODE (target);
int n_elts = GET_MODE_NUNITS (mode);
int n_var = 0;
rtx x, new_rtx, tmp, constant_op, op1, op2;
int i;
for (i = 0; i < n_elts; ++i)
{
x = XVECEXP (vals, 0, i);
if (!(CONST_SCALAR_INT_P (x) || CONST_DOUBLE_P (x) || CONST_FIXED_P (x)))
++n_var;
}
if (n_var == 0)
{
/* Load from constant pool. */
emit_move_insn (target, gen_rtx_CONST_VECTOR (mode, XVEC (vals, 0)));
return;
}
if (n_var == 2)
{
/* The vector is initialized only with non-constants. */
new_rtx = gen_rtx_VEC_CONCAT (V2SFmode, XVECEXP (vals, 0, 0),
XVECEXP (vals, 0, 1));
emit_move_insn (target, new_rtx);
return;
}
/* One field is non-constant and the other one is a constant. Load the
constant from the constant pool and use ps_merge instruction to
construct the whole vector. */
op1 = XVECEXP (vals, 0, 0);
op2 = XVECEXP (vals, 0, 1);
constant_op = (CONSTANT_P (op1)) ? op1 : op2;
tmp = gen_reg_rtx (GET_MODE (constant_op));
emit_move_insn (tmp, constant_op);
if (CONSTANT_P (op1))
new_rtx = gen_rtx_VEC_CONCAT (V2SFmode, tmp, op2);
else
new_rtx = gen_rtx_VEC_CONCAT (V2SFmode, op1, tmp);
emit_move_insn (target, new_rtx);
}
void
paired_expand_vector_move (rtx operands[])
{
rtx op0 = operands[0], op1 = operands[1];
emit_move_insn (op0, op1);
}
/* Emit vector compare for code RCODE. DEST is destination, OP1 and
OP2 are two VEC_COND_EXPR operands, CC_OP0 and CC_OP1 are the two
operands for the relation operation COND. This is a recursive
function. */
static void
paired_emit_vector_compare (enum rtx_code rcode,
rtx dest, rtx op0, rtx op1,
rtx cc_op0, rtx cc_op1)
{
rtx tmp = gen_reg_rtx (V2SFmode);
rtx tmp1, max, min;
gcc_assert (TARGET_PAIRED_FLOAT);
gcc_assert (GET_MODE (op0) == GET_MODE (op1));
switch (rcode)
{
case LT:
case LTU:
paired_emit_vector_compare (GE, dest, op1, op0, cc_op0, cc_op1);
return;
case GE:
case GEU:
emit_insn (gen_subv2sf3 (tmp, cc_op0, cc_op1));
emit_insn (gen_selv2sf4 (dest, tmp, op0, op1, CONST0_RTX (SFmode)));
return;
case LE:
case LEU:
paired_emit_vector_compare (GE, dest, op0, op1, cc_op1, cc_op0);
return;
case GT:
paired_emit_vector_compare (LE, dest, op1, op0, cc_op0, cc_op1);
return;
case EQ:
tmp1 = gen_reg_rtx (V2SFmode);
max = gen_reg_rtx (V2SFmode);
min = gen_reg_rtx (V2SFmode);
gen_reg_rtx (V2SFmode);
emit_insn (gen_subv2sf3 (tmp, cc_op0, cc_op1));
emit_insn (gen_selv2sf4
(max, tmp, cc_op0, cc_op1, CONST0_RTX (SFmode)));
emit_insn (gen_subv2sf3 (tmp, cc_op1, cc_op0));
emit_insn (gen_selv2sf4
(min, tmp, cc_op0, cc_op1, CONST0_RTX (SFmode)));
emit_insn (gen_subv2sf3 (tmp1, min, max));
emit_insn (gen_selv2sf4 (dest, tmp1, op0, op1, CONST0_RTX (SFmode)));
return;
case NE:
paired_emit_vector_compare (EQ, dest, op1, op0, cc_op0, cc_op1);
return;
case UNLE:
paired_emit_vector_compare (LE, dest, op1, op0, cc_op0, cc_op1);
return;
case UNLT:
paired_emit_vector_compare (LT, dest, op1, op0, cc_op0, cc_op1);
return;
case UNGE:
paired_emit_vector_compare (GE, dest, op1, op0, cc_op0, cc_op1);
return;
case UNGT:
paired_emit_vector_compare (GT, dest, op1, op0, cc_op0, cc_op1);
return;
default:
gcc_unreachable ();
}
return;
}
/* Emit vector conditional expression.
DEST is destination. OP1 and OP2 are two VEC_COND_EXPR operands.
CC_OP0 and CC_OP1 are the two operands for the relation operation COND. */
int
paired_emit_vector_cond_expr (rtx dest, rtx op1, rtx op2,
rtx cond, rtx cc_op0, rtx cc_op1)
{
enum rtx_code rcode = GET_CODE (cond);
if (!TARGET_PAIRED_FLOAT)
return 0;
paired_emit_vector_compare (rcode, dest, op1, op2, cc_op0, cc_op1);
return 1;
}
/* Initialize vector TARGET to VALS. */
void
rs6000_expand_vector_init (rtx target, rtx vals)
{
machine_mode mode = GET_MODE (target);
machine_mode inner_mode = GET_MODE_INNER (mode);
int n_elts = GET_MODE_NUNITS (mode);
int n_var = 0, one_var = -1;
bool all_same = true, all_const_zero = true;
rtx x, mem;
int i;
for (i = 0; i < n_elts; ++i)
{
x = XVECEXP (vals, 0, i);
if (!(CONST_SCALAR_INT_P (x) || CONST_DOUBLE_P (x) || CONST_FIXED_P (x)))
++n_var, one_var = i;
else if (x != CONST0_RTX (inner_mode))
all_const_zero = false;
if (i > 0 && !rtx_equal_p (x, XVECEXP (vals, 0, 0)))
all_same = false;
}
if (n_var == 0)
{
rtx const_vec = gen_rtx_CONST_VECTOR (mode, XVEC (vals, 0));
bool int_vector_p = (GET_MODE_CLASS (mode) == MODE_VECTOR_INT);
if ((int_vector_p || TARGET_VSX) && all_const_zero)
{
/* Zero register. */
emit_move_insn (target, CONST0_RTX (mode));
return;
}
else if (int_vector_p && easy_vector_constant (const_vec, mode))
{
/* Splat immediate. */
emit_insn (gen_rtx_SET (target, const_vec));
return;
}
else
{
/* Load from constant pool. */
emit_move_insn (target, const_vec);
return;
}
}
/* Double word values on VSX can use xxpermdi or lxvdsx. */
if (VECTOR_MEM_VSX_P (mode) && (mode == V2DFmode || mode == V2DImode))
{
rtx op[2];
size_t i;
size_t num_elements = all_same ? 1 : 2;
for (i = 0; i < num_elements; i++)
{
op[i] = XVECEXP (vals, 0, i);
/* Just in case there is a SUBREG with a smaller mode, do a
conversion. */
if (GET_MODE (op[i]) != inner_mode)
{
rtx tmp = gen_reg_rtx (inner_mode);
convert_move (tmp, op[i], 0);
op[i] = tmp;
}
/* Allow load with splat double word. */
else if (MEM_P (op[i]))
{
if (!all_same)
op[i] = force_reg (inner_mode, op[i]);
}
else if (!REG_P (op[i]))
op[i] = force_reg (inner_mode, op[i]);
}
if (all_same)
{
if (mode == V2DFmode)
emit_insn (gen_vsx_splat_v2df (target, op[0]));
else
emit_insn (gen_vsx_splat_v2di (target, op[0]));
}
else
{
if (mode == V2DFmode)
emit_insn (gen_vsx_concat_v2df (target, op[0], op[1]));
else
emit_insn (gen_vsx_concat_v2di (target, op[0], op[1]));
}
return;
}
/* Special case initializing vector int if we are on 64-bit systems with
direct move or we have the ISA 3.0 instructions. */
if (mode == V4SImode && VECTOR_MEM_VSX_P (V4SImode)
&& TARGET_DIRECT_MOVE_64BIT)
{
if (all_same)
{
rtx element0 = XVECEXP (vals, 0, 0);
if (MEM_P (element0))
element0 = rs6000_address_for_fpconvert (element0);
else
element0 = force_reg (SImode, element0);
if (TARGET_P9_VECTOR)
emit_insn (gen_vsx_splat_v4si (target, element0));
else
{
rtx tmp = gen_reg_rtx (DImode);
emit_insn (gen_zero_extendsidi2 (tmp, element0));
emit_insn (gen_vsx_splat_v4si_di (target, tmp));
}
return;
}
else
{
rtx elements[4];
size_t i;
for (i = 0; i < 4; i++)
{
elements[i] = XVECEXP (vals, 0, i);
if (!CONST_INT_P (elements[i]) && !REG_P (elements[i]))
elements[i] = copy_to_mode_reg (SImode, elements[i]);
}
emit_insn (gen_vsx_init_v4si (target, elements[0], elements[1],
elements[2], elements[3]));
return;
}
}
/* With single precision floating point on VSX, know that internally single
precision is actually represented as a double, and either make 2 V2DF
vectors, and convert these vectors to single precision, or do one
conversion, and splat the result to the other elements. */
if (mode == V4SFmode && VECTOR_MEM_VSX_P (V4SFmode))
{
if (all_same)
{
rtx element0 = XVECEXP (vals, 0, 0);
if (TARGET_P9_VECTOR)
{
if (MEM_P (element0))
element0 = rs6000_address_for_fpconvert (element0);
emit_insn (gen_vsx_splat_v4sf (target, element0));
}
else
{
rtx freg = gen_reg_rtx (V4SFmode);
rtx sreg = force_reg (SFmode, element0);
rtx cvt = (TARGET_XSCVDPSPN
? gen_vsx_xscvdpspn_scalar (freg, sreg)
: gen_vsx_xscvdpsp_scalar (freg, sreg));
emit_insn (cvt);
emit_insn (gen_vsx_xxspltw_v4sf_direct (target, freg,
const0_rtx));
}
}
else
{
rtx dbl_even = gen_reg_rtx (V2DFmode);
rtx dbl_odd = gen_reg_rtx (V2DFmode);
rtx flt_even = gen_reg_rtx (V4SFmode);
rtx flt_odd = gen_reg_rtx (V4SFmode);
rtx op0 = force_reg (SFmode, XVECEXP (vals, 0, 0));
rtx op1 = force_reg (SFmode, XVECEXP (vals, 0, 1));
rtx op2 = force_reg (SFmode, XVECEXP (vals, 0, 2));
rtx op3 = force_reg (SFmode, XVECEXP (vals, 0, 3));
/* Use VMRGEW if we can instead of doing a permute. */
if (TARGET_P8_VECTOR)
{
emit_insn (gen_vsx_concat_v2sf (dbl_even, op0, op2));
emit_insn (gen_vsx_concat_v2sf (dbl_odd, op1, op3));
emit_insn (gen_vsx_xvcvdpsp (flt_even, dbl_even));
emit_insn (gen_vsx_xvcvdpsp (flt_odd, dbl_odd));
if (BYTES_BIG_ENDIAN)
emit_insn (gen_p8_vmrgew_v4sf_direct (target, flt_even, flt_odd));
else
emit_insn (gen_p8_vmrgew_v4sf_direct (target, flt_odd, flt_even));
}
else
{
emit_insn (gen_vsx_concat_v2sf (dbl_even, op0, op1));
emit_insn (gen_vsx_concat_v2sf (dbl_odd, op2, op3));
emit_insn (gen_vsx_xvcvdpsp (flt_even, dbl_even));
emit_insn (gen_vsx_xvcvdpsp (flt_odd, dbl_odd));
rs6000_expand_extract_even (target, flt_even, flt_odd);
}
}
return;
}
/* Special case initializing vector short/char that are splats if we are on
64-bit systems with direct move. */
if (all_same && TARGET_DIRECT_MOVE_64BIT
&& (mode == V16QImode || mode == V8HImode))
{
rtx op0 = XVECEXP (vals, 0, 0);
rtx di_tmp = gen_reg_rtx (DImode);
if (!REG_P (op0))
op0 = force_reg (GET_MODE_INNER (mode), op0);
if (mode == V16QImode)
{
emit_insn (gen_zero_extendqidi2 (di_tmp, op0));
emit_insn (gen_vsx_vspltb_di (target, di_tmp));
return;
}
if (mode == V8HImode)
{
emit_insn (gen_zero_extendhidi2 (di_tmp, op0));
emit_insn (gen_vsx_vsplth_di (target, di_tmp));
return;
}
}
/* Store value to stack temp. Load vector element. Splat. However, splat
of 64-bit items is not supported on Altivec. */
if (all_same && GET_MODE_SIZE (inner_mode) <= 4)
{
mem = assign_stack_temp (mode, GET_MODE_SIZE (inner_mode));
emit_move_insn (adjust_address_nv (mem, inner_mode, 0),
XVECEXP (vals, 0, 0));
x = gen_rtx_UNSPEC (VOIDmode,
gen_rtvec (1, const0_rtx), UNSPEC_LVE);
emit_insn (gen_rtx_PARALLEL (VOIDmode,
gen_rtvec (2,
gen_rtx_SET (target, mem),
x)));
x = gen_rtx_VEC_SELECT (inner_mode, target,
gen_rtx_PARALLEL (VOIDmode,
gen_rtvec (1, const0_rtx)));
emit_insn (gen_rtx_SET (target, gen_rtx_VEC_DUPLICATE (mode, x)));
return;
}
/* One field is non-constant. Load constant then overwrite
varying field. */
if (n_var == 1)
{
rtx copy = copy_rtx (vals);
/* Load constant part of vector, substitute neighboring value for
varying element. */
XVECEXP (copy, 0, one_var) = XVECEXP (vals, 0, (one_var + 1) % n_elts);
rs6000_expand_vector_init (target, copy);
/* Insert variable. */
rs6000_expand_vector_set (target, XVECEXP (vals, 0, one_var), one_var);
return;
}
/* Construct the vector in memory one field at a time
and load the whole vector. */
mem = assign_stack_temp (mode, GET_MODE_SIZE (mode));
for (i = 0; i < n_elts; i++)
emit_move_insn (adjust_address_nv (mem, inner_mode,
i * GET_MODE_SIZE (inner_mode)),
XVECEXP (vals, 0, i));
emit_move_insn (target, mem);
}
/* Set field ELT of TARGET to VAL. */
void
rs6000_expand_vector_set (rtx target, rtx val, int elt)
{
machine_mode mode = GET_MODE (target);
machine_mode inner_mode = GET_MODE_INNER (mode);
rtx reg = gen_reg_rtx (mode);
rtx mask, mem, x;
int width = GET_MODE_SIZE (inner_mode);
int i;
val = force_reg (GET_MODE (val), val);
if (VECTOR_MEM_VSX_P (mode))
{
rtx insn = NULL_RTX;
rtx elt_rtx = GEN_INT (elt);
if (mode == V2DFmode)
insn = gen_vsx_set_v2df (target, target, val, elt_rtx);
else if (mode == V2DImode)
insn = gen_vsx_set_v2di (target, target, val, elt_rtx);
else if (TARGET_P9_VECTOR && TARGET_POWERPC64)
{
if (mode == V4SImode)
insn = gen_vsx_set_v4si_p9 (target, target, val, elt_rtx);
else if (mode == V8HImode)
insn = gen_vsx_set_v8hi_p9 (target, target, val, elt_rtx);
else if (mode == V16QImode)
insn = gen_vsx_set_v16qi_p9 (target, target, val, elt_rtx);
else if (mode == V4SFmode)
insn = gen_vsx_set_v4sf_p9 (target, target, val, elt_rtx);
}
if (insn)
{
emit_insn (insn);
return;
}
}
/* Simplify setting single element vectors like V1TImode. */
if (GET_MODE_SIZE (mode) == GET_MODE_SIZE (inner_mode) && elt == 0)
{
emit_move_insn (target, gen_lowpart (mode, val));
return;
}
/* Load single variable value. */
mem = assign_stack_temp (mode, GET_MODE_SIZE (inner_mode));
emit_move_insn (adjust_address_nv (mem, inner_mode, 0), val);
x = gen_rtx_UNSPEC (VOIDmode,
gen_rtvec (1, const0_rtx), UNSPEC_LVE);
emit_insn (gen_rtx_PARALLEL (VOIDmode,
gen_rtvec (2,
gen_rtx_SET (reg, mem),
x)));
/* Linear sequence. */
mask = gen_rtx_PARALLEL (V16QImode, rtvec_alloc (16));
for (i = 0; i < 16; ++i)
XVECEXP (mask, 0, i) = GEN_INT (i);
/* Set permute mask to insert element into target. */
for (i = 0; i < width; ++i)
XVECEXP (mask, 0, elt*width + i)
= GEN_INT (i + 0x10);
x = gen_rtx_CONST_VECTOR (V16QImode, XVEC (mask, 0));
if (BYTES_BIG_ENDIAN)
x = gen_rtx_UNSPEC (mode,
gen_rtvec (3, target, reg,
force_reg (V16QImode, x)),
UNSPEC_VPERM);
else
{
if (TARGET_P9_VECTOR)
x = gen_rtx_UNSPEC (mode,
gen_rtvec (3, target, reg,
force_reg (V16QImode, x)),
UNSPEC_VPERMR);
else
{
/* Invert selector. We prefer to generate VNAND on P8 so
that future fusion opportunities can kick in, but must
generate VNOR elsewhere. */
rtx notx = gen_rtx_NOT (V16QImode, force_reg (V16QImode, x));
rtx iorx = (TARGET_P8_VECTOR
? gen_rtx_IOR (V16QImode, notx, notx)
: gen_rtx_AND (V16QImode, notx, notx));
rtx tmp = gen_reg_rtx (V16QImode);
emit_insn (gen_rtx_SET (tmp, iorx));
/* Permute with operands reversed and adjusted selector. */
x = gen_rtx_UNSPEC (mode, gen_rtvec (3, reg, target, tmp),
UNSPEC_VPERM);
}
}
emit_insn (gen_rtx_SET (target, x));
}
/* Extract field ELT from VEC into TARGET. */
void
rs6000_expand_vector_extract (rtx target, rtx vec, rtx elt)
{
machine_mode mode = GET_MODE (vec);
machine_mode inner_mode = GET_MODE_INNER (mode);
rtx mem;
if (VECTOR_MEM_VSX_P (mode) && CONST_INT_P (elt))
{
switch (mode)
{
default:
break;
case E_V1TImode:
gcc_assert (INTVAL (elt) == 0 && inner_mode == TImode);
emit_move_insn (target, gen_lowpart (TImode, vec));
break;
case E_V2DFmode:
emit_insn (gen_vsx_extract_v2df (target, vec, elt));
return;
case E_V2DImode:
emit_insn (gen_vsx_extract_v2di (target, vec, elt));
return;
case E_V4SFmode:
emit_insn (gen_vsx_extract_v4sf (target, vec, elt));
return;
case E_V16QImode:
if (TARGET_DIRECT_MOVE_64BIT)
{
emit_insn (gen_vsx_extract_v16qi (target, vec, elt));
return;
}
else
break;
case E_V8HImode:
if (TARGET_DIRECT_MOVE_64BIT)
{
emit_insn (gen_vsx_extract_v8hi (target, vec, elt));
return;
}
else
break;
case E_V4SImode:
if (TARGET_DIRECT_MOVE_64BIT)
{
emit_insn (gen_vsx_extract_v4si (target, vec, elt));
return;
}
break;
}
}
else if (VECTOR_MEM_VSX_P (mode) && !CONST_INT_P (elt)
&& TARGET_DIRECT_MOVE_64BIT)
{
if (GET_MODE (elt) != DImode)
{
rtx tmp = gen_reg_rtx (DImode);
convert_move (tmp, elt, 0);
elt = tmp;
}
else if (!REG_P (elt))
elt = force_reg (DImode, elt);
switch (mode)
{
case E_V2DFmode:
emit_insn (gen_vsx_extract_v2df_var (target, vec, elt));
return;
case E_V2DImode:
emit_insn (gen_vsx_extract_v2di_var (target, vec, elt));
return;
case E_V4SFmode:
emit_insn (gen_vsx_extract_v4sf_var (target, vec, elt));
return;
case E_V4SImode:
emit_insn (gen_vsx_extract_v4si_var (target, vec, elt));
return;
case E_V8HImode:
emit_insn (gen_vsx_extract_v8hi_var (target, vec, elt));
return;
case E_V16QImode:
emit_insn (gen_vsx_extract_v16qi_var (target, vec, elt));
return;
default:
gcc_unreachable ();
}
}
gcc_assert (CONST_INT_P (elt));
/* Allocate mode-sized buffer. */
mem = assign_stack_temp (mode, GET_MODE_SIZE (mode));
emit_move_insn (mem, vec);
/* Add offset to field within buffer matching vector element. */
mem = adjust_address_nv (mem, inner_mode,
INTVAL (elt) * GET_MODE_SIZE (inner_mode));
emit_move_insn (target, adjust_address_nv (mem, inner_mode, 0));
}
/* Helper function to return the register number of a RTX. */
static inline int
regno_or_subregno (rtx op)
{
if (REG_P (op))
return REGNO (op);
else if (SUBREG_P (op))
return subreg_regno (op);
else
gcc_unreachable ();
}
/* Adjust a memory address (MEM) of a vector type to point to a scalar field
within the vector (ELEMENT) with a mode (SCALAR_MODE). Use a base register
temporary (BASE_TMP) to fixup the address. Return the new memory address
that is valid for reads or writes to a given register (SCALAR_REG). */
rtx
rs6000_adjust_vec_address (rtx scalar_reg,
rtx mem,
rtx element,
rtx base_tmp,
machine_mode scalar_mode)
{
unsigned scalar_size = GET_MODE_SIZE (scalar_mode);
rtx addr = XEXP (mem, 0);
rtx element_offset;
rtx new_addr;
bool valid_addr_p;
/* Vector addresses should not have PRE_INC, PRE_DEC, or PRE_MODIFY. */
gcc_assert (GET_RTX_CLASS (GET_CODE (addr)) != RTX_AUTOINC);
/* Calculate what we need to add to the address to get the element
address. */
if (CONST_INT_P (element))
element_offset = GEN_INT (INTVAL (element) * scalar_size);
else
{
int byte_shift = exact_log2 (scalar_size);
gcc_assert (byte_shift >= 0);
if (byte_shift == 0)
element_offset = element;
else
{
if (TARGET_POWERPC64)
emit_insn (gen_ashldi3 (base_tmp, element, GEN_INT (byte_shift)));
else
emit_insn (gen_ashlsi3 (base_tmp, element, GEN_INT (byte_shift)));
element_offset = base_tmp;
}
}
/* Create the new address pointing to the element within the vector. If we
are adding 0, we don't have to change the address. */
if (element_offset == const0_rtx)
new_addr = addr;
/* A simple indirect address can be converted into a reg + offset
address. */
else if (REG_P (addr) || SUBREG_P (addr))
new_addr = gen_rtx_PLUS (Pmode, addr, element_offset);
/* Optimize D-FORM addresses with constant offset with a constant element, to
include the element offset in the address directly. */
else if (GET_CODE (addr) == PLUS)
{
rtx op0 = XEXP (addr, 0);
rtx op1 = XEXP (addr, 1);
rtx insn;
gcc_assert (REG_P (op0) || SUBREG_P (op0));
if (CONST_INT_P (op1) && CONST_INT_P (element_offset))
{
HOST_WIDE_INT offset = INTVAL (op1) + INTVAL (element_offset);
rtx offset_rtx = GEN_INT (offset);
if (IN_RANGE (offset, -32768, 32767)
&& (scalar_size < 8 || (offset & 0x3) == 0))
new_addr = gen_rtx_PLUS (Pmode, op0, offset_rtx);
else
{
emit_move_insn (base_tmp, offset_rtx);
new_addr = gen_rtx_PLUS (Pmode, op0, base_tmp);
}
}
else
{
bool op1_reg_p = (REG_P (op1) || SUBREG_P (op1));
bool ele_reg_p = (REG_P (element_offset) || SUBREG_P (element_offset));
/* Note, ADDI requires the register being added to be a base
register. If the register was R0, load it up into the temporary
and do the add. */
if (op1_reg_p
&& (ele_reg_p || reg_or_subregno (op1) != FIRST_GPR_REGNO))
{
insn = gen_add3_insn (base_tmp, op1, element_offset);
gcc_assert (insn != NULL_RTX);
emit_insn (insn);
}
else if (ele_reg_p
&& reg_or_subregno (element_offset) != FIRST_GPR_REGNO)
{
insn = gen_add3_insn (base_tmp, element_offset, op1);
gcc_assert (insn != NULL_RTX);
emit_insn (insn);
}
else
{
emit_move_insn (base_tmp, op1);
emit_insn (gen_add2_insn (base_tmp, element_offset));
}
new_addr = gen_rtx_PLUS (Pmode, op0, base_tmp);
}
}
else
{
emit_move_insn (base_tmp, addr);
new_addr = gen_rtx_PLUS (Pmode, base_tmp, element_offset);
}
/* If we have a PLUS, we need to see whether the particular register class
allows for D-FORM or X-FORM addressing. */
if (GET_CODE (new_addr) == PLUS)
{
rtx op1 = XEXP (new_addr, 1);
addr_mask_type addr_mask;
int scalar_regno = regno_or_subregno (scalar_reg);
gcc_assert (scalar_regno < FIRST_PSEUDO_REGISTER);
if (INT_REGNO_P (scalar_regno))
addr_mask = reg_addr[scalar_mode].addr_mask[RELOAD_REG_GPR];
else if (FP_REGNO_P (scalar_regno))
addr_mask = reg_addr[scalar_mode].addr_mask[RELOAD_REG_FPR];
else if (ALTIVEC_REGNO_P (scalar_regno))
addr_mask = reg_addr[scalar_mode].addr_mask[RELOAD_REG_VMX];
else
gcc_unreachable ();
if (REG_P (op1) || SUBREG_P (op1))
valid_addr_p = (addr_mask & RELOAD_REG_INDEXED) != 0;
else
valid_addr_p = (addr_mask & RELOAD_REG_OFFSET) != 0;
}
else if (REG_P (new_addr) || SUBREG_P (new_addr))
valid_addr_p = true;
else
valid_addr_p = false;
if (!valid_addr_p)
{
emit_move_insn (base_tmp, new_addr);
new_addr = base_tmp;
}
return change_address (mem, scalar_mode, new_addr);
}
/* Split a variable vec_extract operation into the component instructions. */
void
rs6000_split_vec_extract_var (rtx dest, rtx src, rtx element, rtx tmp_gpr,
rtx tmp_altivec)
{
machine_mode mode = GET_MODE (src);
machine_mode scalar_mode = GET_MODE (dest);
unsigned scalar_size = GET_MODE_SIZE (scalar_mode);
int byte_shift = exact_log2 (scalar_size);
gcc_assert (byte_shift >= 0);
/* If we are given a memory address, optimize to load just the element. We
don't have to adjust the vector element number on little endian
systems. */
if (MEM_P (src))
{
gcc_assert (REG_P (tmp_gpr));
emit_move_insn (dest, rs6000_adjust_vec_address (dest, src, element,
tmp_gpr, scalar_mode));
return;
}
else if (REG_P (src) || SUBREG_P (src))
{
int bit_shift = byte_shift + 3;
rtx element2;
int dest_regno = regno_or_subregno (dest);
int src_regno = regno_or_subregno (src);
int element_regno = regno_or_subregno (element);
gcc_assert (REG_P (tmp_gpr));
/* See if we want to generate VEXTU{B,H,W}{L,R}X if the destination is in
a general purpose register. */
if (TARGET_P9_VECTOR
&& (mode == V16QImode || mode == V8HImode || mode == V4SImode)
&& INT_REGNO_P (dest_regno)
&& ALTIVEC_REGNO_P (src_regno)
&& INT_REGNO_P (element_regno))
{
rtx dest_si = gen_rtx_REG (SImode, dest_regno);
rtx element_si = gen_rtx_REG (SImode, element_regno);
if (mode == V16QImode)
emit_insn (VECTOR_ELT_ORDER_BIG
? gen_vextublx (dest_si, element_si, src)
: gen_vextubrx (dest_si, element_si, src));
else if (mode == V8HImode)
{
rtx tmp_gpr_si = gen_rtx_REG (SImode, REGNO (tmp_gpr));
emit_insn (gen_ashlsi3 (tmp_gpr_si, element_si, const1_rtx));
emit_insn (VECTOR_ELT_ORDER_BIG
? gen_vextuhlx (dest_si, tmp_gpr_si, src)
: gen_vextuhrx (dest_si, tmp_gpr_si, src));
}
else
{
rtx tmp_gpr_si = gen_rtx_REG (SImode, REGNO (tmp_gpr));
emit_insn (gen_ashlsi3 (tmp_gpr_si, element_si, const2_rtx));
emit_insn (VECTOR_ELT_ORDER_BIG
? gen_vextuwlx (dest_si, tmp_gpr_si, src)
: gen_vextuwrx (dest_si, tmp_gpr_si, src));
}
return;
}
gcc_assert (REG_P (tmp_altivec));
/* For little endian, adjust element ordering. For V2DI/V2DF, we can use
an XOR, otherwise we need to subtract. The shift amount is so VSLO
will shift the element into the upper position (adding 3 to convert a
byte shift into a bit shift). */
if (scalar_size == 8)
{
if (!VECTOR_ELT_ORDER_BIG)
{
emit_insn (gen_xordi3 (tmp_gpr, element, const1_rtx));
element2 = tmp_gpr;
}
else
element2 = element;
/* Generate RLDIC directly to shift left 6 bits and retrieve 1
bit. */
emit_insn (gen_rtx_SET (tmp_gpr,
gen_rtx_AND (DImode,
gen_rtx_ASHIFT (DImode,
element2,
GEN_INT (6)),
GEN_INT (64))));
}
else
{
if (!VECTOR_ELT_ORDER_BIG)
{
rtx num_ele_m1 = GEN_INT (GET_MODE_NUNITS (mode) - 1);
emit_insn (gen_anddi3 (tmp_gpr, element, num_ele_m1));
emit_insn (gen_subdi3 (tmp_gpr, num_ele_m1, tmp_gpr));
element2 = tmp_gpr;
}
else
element2 = element;
emit_insn (gen_ashldi3 (tmp_gpr, element2, GEN_INT (bit_shift)));
}
/* Get the value into the lower byte of the Altivec register where VSLO
expects it. */
if (TARGET_P9_VECTOR)
emit_insn (gen_vsx_splat_v2di (tmp_altivec, tmp_gpr));
else if (can_create_pseudo_p ())
emit_insn (gen_vsx_concat_v2di (tmp_altivec, tmp_gpr, tmp_gpr));
else
{
rtx tmp_di = gen_rtx_REG (DImode, REGNO (tmp_altivec));
emit_move_insn (tmp_di, tmp_gpr);
emit_insn (gen_vsx_concat_v2di (tmp_altivec, tmp_di, tmp_di));
}
/* Do the VSLO to get the value into the final location. */
switch (mode)
{
case E_V2DFmode:
emit_insn (gen_vsx_vslo_v2df (dest, src, tmp_altivec));
return;
case E_V2DImode:
emit_insn (gen_vsx_vslo_v2di (dest, src, tmp_altivec));
return;
case E_V4SFmode:
{
rtx tmp_altivec_di = gen_rtx_REG (DImode, REGNO (tmp_altivec));
rtx tmp_altivec_v4sf = gen_rtx_REG (V4SFmode, REGNO (tmp_altivec));
rtx src_v2di = gen_rtx_REG (V2DImode, REGNO (src));
emit_insn (gen_vsx_vslo_v2di (tmp_altivec_di, src_v2di,
tmp_altivec));
emit_insn (gen_vsx_xscvspdp_scalar2 (dest, tmp_altivec_v4sf));
return;
}
case E_V4SImode:
case E_V8HImode:
case E_V16QImode:
{
rtx tmp_altivec_di = gen_rtx_REG (DImode, REGNO (tmp_altivec));
rtx src_v2di = gen_rtx_REG (V2DImode, REGNO (src));
rtx tmp_gpr_di = gen_rtx_REG (DImode, REGNO (dest));
emit_insn (gen_vsx_vslo_v2di (tmp_altivec_di, src_v2di,
tmp_altivec));
emit_move_insn (tmp_gpr_di, tmp_altivec_di);
emit_insn (gen_ashrdi3 (tmp_gpr_di, tmp_gpr_di,
GEN_INT (64 - (8 * scalar_size))));
return;
}
default:
gcc_unreachable ();
}
return;
}
else
gcc_unreachable ();
}
/* Helper function for rs6000_split_v4si_init to build up a DImode value from
two SImode values. */
static void
rs6000_split_v4si_init_di_reg (rtx dest, rtx si1, rtx si2, rtx tmp)
{
const unsigned HOST_WIDE_INT mask_32bit = HOST_WIDE_INT_C (0xffffffff);
if (CONST_INT_P (si1) && CONST_INT_P (si2))
{
unsigned HOST_WIDE_INT const1 = (UINTVAL (si1) & mask_32bit) << 32;
unsigned HOST_WIDE_INT const2 = UINTVAL (si2) & mask_32bit;
emit_move_insn (dest, GEN_INT (const1 | const2));
return;
}
/* Put si1 into upper 32-bits of dest. */
if (CONST_INT_P (si1))
emit_move_insn (dest, GEN_INT ((UINTVAL (si1) & mask_32bit) << 32));
else
{
/* Generate RLDIC. */
rtx si1_di = gen_rtx_REG (DImode, regno_or_subregno (si1));
rtx shift_rtx = gen_rtx_ASHIFT (DImode, si1_di, GEN_INT (32));
rtx mask_rtx = GEN_INT (mask_32bit << 32);
rtx and_rtx = gen_rtx_AND (DImode, shift_rtx, mask_rtx);
gcc_assert (!reg_overlap_mentioned_p (dest, si1));
emit_insn (gen_rtx_SET (dest, and_rtx));
}
/* Put si2 into the temporary. */
gcc_assert (!reg_overlap_mentioned_p (dest, tmp));
if (CONST_INT_P (si2))
emit_move_insn (tmp, GEN_INT (UINTVAL (si2) & mask_32bit));
else
emit_insn (gen_zero_extendsidi2 (tmp, si2));
/* Combine the two parts. */
emit_insn (gen_iordi3 (dest, dest, tmp));
return;
}
/* Split a V4SI initialization. */
void
rs6000_split_v4si_init (rtx operands[])
{
rtx dest = operands[0];
/* Destination is a GPR, build up the two DImode parts in place. */
if (REG_P (dest) || SUBREG_P (dest))
{
int d_regno = regno_or_subregno (dest);
rtx scalar1 = operands[1];
rtx scalar2 = operands[2];
rtx scalar3 = operands[3];
rtx scalar4 = operands[4];
rtx tmp1 = operands[5];
rtx tmp2 = operands[6];
/* Even though we only need one temporary (plus the destination, which
has an early clobber constraint, try to use two temporaries, one for
each double word created. That way the 2nd insn scheduling pass can
rearrange things so the two parts are done in parallel. */
if (BYTES_BIG_ENDIAN)
{
rtx di_lo = gen_rtx_REG (DImode, d_regno);
rtx di_hi = gen_rtx_REG (DImode, d_regno + 1);
rs6000_split_v4si_init_di_reg (di_lo, scalar1, scalar2, tmp1);
rs6000_split_v4si_init_di_reg (di_hi, scalar3, scalar4, tmp2);
}
else
{
rtx di_lo = gen_rtx_REG (DImode, d_regno + 1);
rtx di_hi = gen_rtx_REG (DImode, d_regno);
gcc_assert (!VECTOR_ELT_ORDER_BIG);
rs6000_split_v4si_init_di_reg (di_lo, scalar4, scalar3, tmp1);
rs6000_split_v4si_init_di_reg (di_hi, scalar2, scalar1, tmp2);
}
return;
}
else
gcc_unreachable ();
}
/* Return alignment of TYPE. Existing alignment is ALIGN. HOW
selects whether the alignment is abi mandated, optional, or
both abi and optional alignment. */
unsigned int
rs6000_data_alignment (tree type, unsigned int align, enum data_align how)
{
if (how != align_opt)
{
if (TREE_CODE (type) == VECTOR_TYPE)
{
if (TARGET_PAIRED_FLOAT && PAIRED_VECTOR_MODE (TYPE_MODE (type)))
{
if (align < 64)
align = 64;
}
else if (align < 128)
align = 128;
}
}
if (how != align_abi)
{
if (TREE_CODE (type) == ARRAY_TYPE
&& TYPE_MODE (TREE_TYPE (type)) == QImode)
{
if (align < BITS_PER_WORD)
align = BITS_PER_WORD;
}
}
return align;
}
/* Implement TARGET_SLOW_UNALIGNED_ACCESS. Altivec vector memory
instructions simply ignore the low bits; VSX memory instructions
are aligned to 4 or 8 bytes. */
static bool
rs6000_slow_unaligned_access (machine_mode mode, unsigned int align)
{
return (STRICT_ALIGNMENT
|| (!TARGET_EFFICIENT_UNALIGNED_VSX
&& ((SCALAR_FLOAT_MODE_NOT_VECTOR_P (mode) && align < 32)
|| ((VECTOR_MODE_P (mode) || FLOAT128_VECTOR_P (mode))
&& (int) align < VECTOR_ALIGN (mode)))));
}
/* Previous GCC releases forced all vector types to have 16-byte alignment. */
bool
rs6000_special_adjust_field_align_p (tree type, unsigned int computed)
{
if (TARGET_ALTIVEC && TREE_CODE (type) == VECTOR_TYPE)
{
if (computed != 128)
{
static bool warned;
if (!warned && warn_psabi)
{
warned = true;
inform (input_location,
"the layout of aggregates containing vectors with"
" %d-byte alignment has changed in GCC 5",
computed / BITS_PER_UNIT);
}
}
/* In current GCC there is no special case. */
return false;
}
return false;
}
/* AIX increases natural record alignment to doubleword if the first
field is an FP double while the FP fields remain word aligned. */
unsigned int
rs6000_special_round_type_align (tree type, unsigned int computed,
unsigned int specified)
{
unsigned int align = MAX (computed, specified);
tree field = TYPE_FIELDS (type);
/* Skip all non field decls */
while (field != NULL && TREE_CODE (field) != FIELD_DECL)
field = DECL_CHAIN (field);
if (field != NULL && field != type)
{
type = TREE_TYPE (field);
while (TREE_CODE (type) == ARRAY_TYPE)
type = TREE_TYPE (type);
if (type != error_mark_node && TYPE_MODE (type) == DFmode)
align = MAX (align, 64);
}
return align;
}
/* Darwin increases record alignment to the natural alignment of
the first field. */
unsigned int
darwin_rs6000_special_round_type_align (tree type, unsigned int computed,
unsigned int specified)
{
unsigned int align = MAX (computed, specified);
if (TYPE_PACKED (type))
return align;
/* Find the first field, looking down into aggregates. */
do {
tree field = TYPE_FIELDS (type);
/* Skip all non field decls */
while (field != NULL && TREE_CODE (field) != FIELD_DECL)
field = DECL_CHAIN (field);
if (! field)
break;
/* A packed field does not contribute any extra alignment. */
if (DECL_PACKED (field))
return align;
type = TREE_TYPE (field);
while (TREE_CODE (type) == ARRAY_TYPE)
type = TREE_TYPE (type);
} while (AGGREGATE_TYPE_P (type));
if (! AGGREGATE_TYPE_P (type) && type != error_mark_node)
align = MAX (align, TYPE_ALIGN (type));
return align;
}
/* Return 1 for an operand in small memory on V.4/eabi. */
int
small_data_operand (rtx op ATTRIBUTE_UNUSED,
machine_mode mode ATTRIBUTE_UNUSED)
{
#if TARGET_ELF
rtx sym_ref;
if (rs6000_sdata == SDATA_NONE || rs6000_sdata == SDATA_DATA)
return 0;
if (DEFAULT_ABI != ABI_V4)
return 0;
if (GET_CODE (op) == SYMBOL_REF)
sym_ref = op;
else if (GET_CODE (op) != CONST
|| GET_CODE (XEXP (op, 0)) != PLUS
|| GET_CODE (XEXP (XEXP (op, 0), 0)) != SYMBOL_REF
|| GET_CODE (XEXP (XEXP (op, 0), 1)) != CONST_INT)
return 0;
else
{
rtx sum = XEXP (op, 0);
HOST_WIDE_INT summand;
/* We have to be careful here, because it is the referenced address
that must be 32k from _SDA_BASE_, not just the symbol. */
summand = INTVAL (XEXP (sum, 1));
if (summand < 0 || summand > g_switch_value)
return 0;
sym_ref = XEXP (sum, 0);
}
return SYMBOL_REF_SMALL_P (sym_ref);
#else
return 0;
#endif
}
/* Return true if either operand is a general purpose register. */
bool
gpr_or_gpr_p (rtx op0, rtx op1)
{
return ((REG_P (op0) && INT_REGNO_P (REGNO (op0)))
|| (REG_P (op1) && INT_REGNO_P (REGNO (op1))));
}
/* Return true if this is a move direct operation between GPR registers and
floating point/VSX registers. */
bool
direct_move_p (rtx op0, rtx op1)
{
int regno0, regno1;
if (!REG_P (op0) || !REG_P (op1))
return false;
if (!TARGET_DIRECT_MOVE && !TARGET_MFPGPR)
return false;
regno0 = REGNO (op0);
regno1 = REGNO (op1);
if (regno0 >= FIRST_PSEUDO_REGISTER || regno1 >= FIRST_PSEUDO_REGISTER)
return false;
if (INT_REGNO_P (regno0))
return (TARGET_DIRECT_MOVE) ? VSX_REGNO_P (regno1) : FP_REGNO_P (regno1);
else if (INT_REGNO_P (regno1))
{
if (TARGET_MFPGPR && FP_REGNO_P (regno0))
return true;
else if (TARGET_DIRECT_MOVE && VSX_REGNO_P (regno0))
return true;
}
return false;
}
/* Return true if the OFFSET is valid for the quad address instructions that
use d-form (register + offset) addressing. */
static inline bool
quad_address_offset_p (HOST_WIDE_INT offset)
{
return (IN_RANGE (offset, -32768, 32767) && ((offset) & 0xf) == 0);
}
/* Return true if the ADDR is an acceptable address for a quad memory
operation of mode MODE (either LQ/STQ for general purpose registers, or
LXV/STXV for vector registers under ISA 3.0. GPR_P is true if this address
is intended for LQ/STQ. If it is false, the address is intended for the ISA
3.0 LXV/STXV instruction. */
bool
quad_address_p (rtx addr, machine_mode mode, bool strict)
{
rtx op0, op1;
if (GET_MODE_SIZE (mode) != 16)
return false;
if (legitimate_indirect_address_p (addr, strict))
return true;
if (VECTOR_MODE_P (mode) && !mode_supports_vsx_dform_quad (mode))
return false;
if (GET_CODE (addr) != PLUS)
return false;
op0 = XEXP (addr, 0);
if (!REG_P (op0) || !INT_REG_OK_FOR_BASE_P (op0, strict))
return false;
op1 = XEXP (addr, 1);
if (!CONST_INT_P (op1))
return false;
return quad_address_offset_p (INTVAL (op1));
}
/* Return true if this is a load or store quad operation. This function does
not handle the atomic quad memory instructions. */
bool
quad_load_store_p (rtx op0, rtx op1)
{
bool ret;
if (!TARGET_QUAD_MEMORY)
ret = false;
else if (REG_P (op0) && MEM_P (op1))
ret = (quad_int_reg_operand (op0, GET_MODE (op0))
&& quad_memory_operand (op1, GET_MODE (op1))
&& !reg_overlap_mentioned_p (op0, op1));
else if (MEM_P (op0) && REG_P (op1))
ret = (quad_memory_operand (op0, GET_MODE (op0))
&& quad_int_reg_operand (op1, GET_MODE (op1)));
else
ret = false;
if (TARGET_DEBUG_ADDR)
{
fprintf (stderr, "\n========== quad_load_store, return %s\n",
ret ? "true" : "false");
debug_rtx (gen_rtx_SET (op0, op1));
}
return ret;
}
/* Given an address, return a constant offset term if one exists. */
static rtx
address_offset (rtx op)
{
if (GET_CODE (op) == PRE_INC
|| GET_CODE (op) == PRE_DEC)
op = XEXP (op, 0);
else if (GET_CODE (op) == PRE_MODIFY
|| GET_CODE (op) == LO_SUM)
op = XEXP (op, 1);
if (GET_CODE (op) == CONST)
op = XEXP (op, 0);
if (GET_CODE (op) == PLUS)
op = XEXP (op, 1);
if (CONST_INT_P (op))
return op;
return NULL_RTX;
}
/* Return true if the MEM operand is a memory operand suitable for use
with a (full width, possibly multiple) gpr load/store. On
powerpc64 this means the offset must be divisible by 4.
Implements 'Y' constraint.
Accept direct, indexed, offset, lo_sum and tocref. Since this is
a constraint function we know the operand has satisfied a suitable
memory predicate. Also accept some odd rtl generated by reload
(see rs6000_legitimize_reload_address for various forms). It is
important that reload rtl be accepted by appropriate constraints
but not by the operand predicate.
Offsetting a lo_sum should not be allowed, except where we know by
alignment that a 32k boundary is not crossed, but see the ???
comment in rs6000_legitimize_reload_address. Note that by
"offsetting" here we mean a further offset to access parts of the
MEM. It's fine to have a lo_sum where the inner address is offset
from a sym, since the same sym+offset will appear in the high part
of the address calculation. */
bool
mem_operand_gpr (rtx op, machine_mode mode)
{
unsigned HOST_WIDE_INT offset;
int extra;
rtx addr = XEXP (op, 0);
op = address_offset (addr);
if (op == NULL_RTX)
return true;
offset = INTVAL (op);
if (TARGET_POWERPC64 && (offset & 3) != 0)
return false;
extra = GET_MODE_SIZE (mode) - UNITS_PER_WORD;
if (extra < 0)
extra = 0;
if (GET_CODE (addr) == LO_SUM)
/* For lo_sum addresses, we must allow any offset except one that
causes a wrap, so test only the low 16 bits. */
offset = ((offset & 0xffff) ^ 0x8000) - 0x8000;
return offset + 0x8000 < 0x10000u - extra;
}
/* As above, but for DS-FORM VSX insns. Unlike mem_operand_gpr,
enforce an offset divisible by 4 even for 32-bit. */
bool
mem_operand_ds_form (rtx op, machine_mode mode)
{
unsigned HOST_WIDE_INT offset;
int extra;
rtx addr = XEXP (op, 0);
if (!offsettable_address_p (false, mode, addr))
return false;
op = address_offset (addr);
if (op == NULL_RTX)
return true;
offset = INTVAL (op);
if ((offset & 3) != 0)
return false;
extra = GET_MODE_SIZE (mode) - UNITS_PER_WORD;
if (extra < 0)
extra = 0;
if (GET_CODE (addr) == LO_SUM)
/* For lo_sum addresses, we must allow any offset except one that
causes a wrap, so test only the low 16 bits. */
offset = ((offset & 0xffff) ^ 0x8000) - 0x8000;
return offset + 0x8000 < 0x10000u - extra;
}
/* Subroutines of rs6000_legitimize_address and rs6000_legitimate_address_p. */
static bool
reg_offset_addressing_ok_p (machine_mode mode)
{
switch (mode)
{
case E_V16QImode:
case E_V8HImode:
case E_V4SFmode:
case E_V4SImode:
case E_V2DFmode:
case E_V2DImode:
case E_V1TImode:
case E_TImode:
case E_TFmode:
case E_KFmode:
/* AltiVec/VSX vector modes. Only reg+reg addressing was valid until the
ISA 3.0 vector d-form addressing mode was added. While TImode is not
a vector mode, if we want to use the VSX registers to move it around,
we need to restrict ourselves to reg+reg addressing. Similarly for
IEEE 128-bit floating point that is passed in a single vector
register. */
if (VECTOR_MEM_ALTIVEC_OR_VSX_P (mode))
return mode_supports_vsx_dform_quad (mode);
break;
case E_V2SImode:
case E_V2SFmode:
/* Paired vector modes. Only reg+reg addressing is valid. */
if (TARGET_PAIRED_FLOAT)
return false;
break;
case E_SDmode:
/* If we can do direct load/stores of SDmode, restrict it to reg+reg
addressing for the LFIWZX and STFIWX instructions. */
if (TARGET_NO_SDMODE_STACK)
return false;
break;
default:
break;
}
return true;
}
static bool
virtual_stack_registers_memory_p (rtx op)
{
int regnum;
if (GET_CODE (op) == REG)
regnum = REGNO (op);
else if (GET_CODE (op) == PLUS
&& GET_CODE (XEXP (op, 0)) == REG
&& GET_CODE (XEXP (op, 1)) == CONST_INT)
regnum = REGNO (XEXP (op, 0));
else
return false;
return (regnum >= FIRST_VIRTUAL_REGISTER
&& regnum <= LAST_VIRTUAL_POINTER_REGISTER);
}
/* Return true if a MODE sized memory accesses to OP plus OFFSET
is known to not straddle a 32k boundary. This function is used
to determine whether -mcmodel=medium code can use TOC pointer
relative addressing for OP. This means the alignment of the TOC
pointer must also be taken into account, and unfortunately that is
only 8 bytes. */
#ifndef POWERPC64_TOC_POINTER_ALIGNMENT
#define POWERPC64_TOC_POINTER_ALIGNMENT 8
#endif
static bool
offsettable_ok_by_alignment (rtx op, HOST_WIDE_INT offset,
machine_mode mode)
{
tree decl;
unsigned HOST_WIDE_INT dsize, dalign, lsb, mask;
if (GET_CODE (op) != SYMBOL_REF)
return false;
/* ISA 3.0 vector d-form addressing is restricted, don't allow
SYMBOL_REF. */
if (mode_supports_vsx_dform_quad (mode))
return false;
dsize = GET_MODE_SIZE (mode);
decl = SYMBOL_REF_DECL (op);
if (!decl)
{
if (dsize == 0)
return false;
/* -fsection-anchors loses the original SYMBOL_REF_DECL when
replacing memory addresses with an anchor plus offset. We
could find the decl by rummaging around in the block->objects
VEC for the given offset but that seems like too much work. */
dalign = BITS_PER_UNIT;
if (SYMBOL_REF_HAS_BLOCK_INFO_P (op)
&& SYMBOL_REF_ANCHOR_P (op)
&& SYMBOL_REF_BLOCK (op) != NULL)
{
struct object_block *block = SYMBOL_REF_BLOCK (op);
dalign = block->alignment;
offset += SYMBOL_REF_BLOCK_OFFSET (op);
}
else if (CONSTANT_POOL_ADDRESS_P (op))
{
/* It would be nice to have get_pool_align().. */
machine_mode cmode = get_pool_mode (op);
dalign = GET_MODE_ALIGNMENT (cmode);
}
}
else if (DECL_P (decl))
{
dalign = DECL_ALIGN (decl);
if (dsize == 0)
{
/* Allow BLKmode when the entire object is known to not
cross a 32k boundary. */
if (!DECL_SIZE_UNIT (decl))
return false;
if (!tree_fits_uhwi_p (DECL_SIZE_UNIT (decl)))
return false;
dsize = tree_to_uhwi (DECL_SIZE_UNIT (decl));
if (dsize > 32768)
return false;
dalign /= BITS_PER_UNIT;
if (dalign > POWERPC64_TOC_POINTER_ALIGNMENT)
dalign = POWERPC64_TOC_POINTER_ALIGNMENT;
return dalign >= dsize;
}
}
else
gcc_unreachable ();
/* Find how many bits of the alignment we know for this access. */
dalign /= BITS_PER_UNIT;
if (dalign > POWERPC64_TOC_POINTER_ALIGNMENT)
dalign = POWERPC64_TOC_POINTER_ALIGNMENT;
mask = dalign - 1;
lsb = offset & -offset;
mask &= lsb - 1;
dalign = mask + 1;
return dalign >= dsize;
}
static bool
constant_pool_expr_p (rtx op)
{
rtx base, offset;
split_const (op, &base, &offset);
return (GET_CODE (base) == SYMBOL_REF
&& CONSTANT_POOL_ADDRESS_P (base)
&& ASM_OUTPUT_SPECIAL_POOL_ENTRY_P (get_pool_constant (base), Pmode));
}
/* These are only used to pass through from print_operand/print_operand_address
to rs6000_output_addr_const_extra over the intervening function
output_addr_const which is not target code. */
static const_rtx tocrel_base_oac, tocrel_offset_oac;
/* Return true if OP is a toc pointer relative address (the output
of create_TOC_reference). If STRICT, do not match non-split
-mcmodel=large/medium toc pointer relative addresses. If the pointers
are non-NULL, place base and offset pieces in TOCREL_BASE_RET and
TOCREL_OFFSET_RET respectively. */
bool
toc_relative_expr_p (const_rtx op, bool strict, const_rtx *tocrel_base_ret,
const_rtx *tocrel_offset_ret)
{
if (!TARGET_TOC)
return false;
if (TARGET_CMODEL != CMODEL_SMALL)
{
/* When strict ensure we have everything tidy. */
if (strict
&& !(GET_CODE (op) == LO_SUM
&& REG_P (XEXP (op, 0))
&& INT_REG_OK_FOR_BASE_P (XEXP (op, 0), strict)))
return false;
/* When not strict, allow non-split TOC addresses and also allow
(lo_sum (high ..)) TOC addresses created during reload. */
if (GET_CODE (op) == LO_SUM)
op = XEXP (op, 1);
}
const_rtx tocrel_base = op;
const_rtx tocrel_offset = const0_rtx;
if (GET_CODE (op) == PLUS && add_cint_operand (XEXP (op, 1), GET_MODE (op)))
{
tocrel_base = XEXP (op, 0);
tocrel_offset = XEXP (op, 1);
}
if (tocrel_base_ret)
*tocrel_base_ret = tocrel_base;
if (tocrel_offset_ret)
*tocrel_offset_ret = tocrel_offset;
return (GET_CODE (tocrel_base) == UNSPEC
&& XINT (tocrel_base, 1) == UNSPEC_TOCREL);
}
/* Return true if X is a constant pool address, and also for cmodel=medium
if X is a toc-relative address known to be offsettable within MODE. */
bool
legitimate_constant_pool_address_p (const_rtx x, machine_mode mode,
bool strict)
{
const_rtx tocrel_base, tocrel_offset;
return (toc_relative_expr_p (x, strict, &tocrel_base, &tocrel_offset)
&& (TARGET_CMODEL != CMODEL_MEDIUM
|| constant_pool_expr_p (XVECEXP (tocrel_base, 0, 0))
|| mode == QImode
|| offsettable_ok_by_alignment (XVECEXP (tocrel_base, 0, 0),
INTVAL (tocrel_offset), mode)));
}
static bool
legitimate_small_data_p (machine_mode mode, rtx x)
{
return (DEFAULT_ABI == ABI_V4
&& !flag_pic && !TARGET_TOC
&& (GET_CODE (x) == SYMBOL_REF || GET_CODE (x) == CONST)
&& small_data_operand (x, mode));
}
bool
rs6000_legitimate_offset_address_p (machine_mode mode, rtx x,
bool strict, bool worst_case)
{
unsigned HOST_WIDE_INT offset;
unsigned int extra;
if (GET_CODE (x) != PLUS)
return false;
if (!REG_P (XEXP (x, 0)))
return false;
if (!INT_REG_OK_FOR_BASE_P (XEXP (x, 0), strict))
return false;
if (mode_supports_vsx_dform_quad (mode))
return quad_address_p (x, mode, strict);
if (!reg_offset_addressing_ok_p (mode))
return virtual_stack_registers_memory_p (x);
if (legitimate_constant_pool_address_p (x, mode, strict || lra_in_progress))
return true;
if (GET_CODE (XEXP (x, 1)) != CONST_INT)
return false;
offset = INTVAL (XEXP (x, 1));
extra = 0;
switch (mode)
{
case E_V2SImode:
case E_V2SFmode:
/* Paired single modes: offset addressing isn't valid. */
return false;
case E_DFmode:
case E_DDmode:
case E_DImode:
/* If we are using VSX scalar loads, restrict ourselves to reg+reg
addressing. */
if (VECTOR_MEM_VSX_P (mode))
return false;
if (!worst_case)
break;
if (!TARGET_POWERPC64)
extra = 4;
else if (offset & 3)
return false;
break;
case E_TFmode:
case E_IFmode:
case E_KFmode:
case E_TDmode:
case E_TImode:
case E_PTImode:
extra = 8;
if (!worst_case)
break;
if (!TARGET_POWERPC64)
extra = 12;
else if (offset & 3)
return false;
break;
default:
break;
}
offset += 0x8000;
return offset < 0x10000 - extra;
}
bool
legitimate_indexed_address_p (rtx x, int strict)
{
rtx op0, op1;
if (GET_CODE (x) != PLUS)
return false;
op0 = XEXP (x, 0);
op1 = XEXP (x, 1);
return (REG_P (op0) && REG_P (op1)
&& ((INT_REG_OK_FOR_BASE_P (op0, strict)
&& INT_REG_OK_FOR_INDEX_P (op1, strict))
|| (INT_REG_OK_FOR_BASE_P (op1, strict)
&& INT_REG_OK_FOR_INDEX_P (op0, strict))));
}
bool
avoiding_indexed_address_p (machine_mode mode)
{
/* Avoid indexed addressing for modes that have non-indexed
load/store instruction forms. */
return (TARGET_AVOID_XFORM && VECTOR_MEM_NONE_P (mode));
}
bool
legitimate_indirect_address_p (rtx x, int strict)
{
return GET_CODE (x) == REG && INT_REG_OK_FOR_BASE_P (x, strict);
}
bool
macho_lo_sum_memory_operand (rtx x, machine_mode mode)
{
if (!TARGET_MACHO || !flag_pic
|| mode != SImode || GET_CODE (x) != MEM)
return false;
x = XEXP (x, 0);
if (GET_CODE (x) != LO_SUM)
return false;
if (GET_CODE (XEXP (x, 0)) != REG)
return false;
if (!INT_REG_OK_FOR_BASE_P (XEXP (x, 0), 0))
return false;
x = XEXP (x, 1);
return CONSTANT_P (x);
}
static bool
legitimate_lo_sum_address_p (machine_mode mode, rtx x, int strict)
{
if (GET_CODE (x) != LO_SUM)
return false;
if (GET_CODE (XEXP (x, 0)) != REG)
return false;
if (!INT_REG_OK_FOR_BASE_P (XEXP (x, 0), strict))
return false;
/* quad word addresses are restricted, and we can't use LO_SUM. */
if (mode_supports_vsx_dform_quad (mode))
return false;
x = XEXP (x, 1);
if (TARGET_ELF || TARGET_MACHO)
{
bool large_toc_ok;
if (DEFAULT_ABI == ABI_V4 && flag_pic)
return false;
/* LRA doesn't use LEGITIMIZE_RELOAD_ADDRESS as it usually calls
push_reload from reload pass code. LEGITIMIZE_RELOAD_ADDRESS
recognizes some LO_SUM addresses as valid although this
function says opposite. In most cases, LRA through different
transformations can generate correct code for address reloads.
It can not manage only some LO_SUM cases. So we need to add
code analogous to one in rs6000_legitimize_reload_address for
LOW_SUM here saying that some addresses are still valid. */
large_toc_ok = (lra_in_progress && TARGET_CMODEL != CMODEL_SMALL
&& small_toc_ref (x, VOIDmode));
if (TARGET_TOC && ! large_toc_ok)
return false;
if (GET_MODE_NUNITS (mode) != 1)
return false;
if (GET_MODE_SIZE (mode) > UNITS_PER_WORD
&& !(/* ??? Assume floating point reg based on mode? */
TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT
&& (mode == DFmode || mode == DDmode)))
return false;
return CONSTANT_P (x) || large_toc_ok;
}
return false;
}
/* Try machine-dependent ways of modifying an illegitimate address
to be legitimate. If we find one, return the new, valid address.
This is used from only one place: `memory_address' in explow.c.
OLDX is the address as it was before break_out_memory_refs was
called. In some cases it is useful to look at this to decide what
needs to be done.
It is always safe for this function to do nothing. It exists to
recognize opportunities to optimize the output.
On RS/6000, first check for the sum of a register with a constant
integer that is out of range. If so, generate code to add the
constant with the low-order 16 bits masked to the register and force
this result into another register (this can be done with `cau').
Then generate an address of REG+(CONST&0xffff), allowing for the
possibility of bit 16 being a one.
Then check for the sum of a register and something not constant, try to
load the other things into a register and return the sum. */
static rtx
rs6000_legitimize_address (rtx x, rtx oldx ATTRIBUTE_UNUSED,
machine_mode mode)
{
unsigned int extra;
if (!reg_offset_addressing_ok_p (mode)
|| mode_supports_vsx_dform_quad (mode))
{
if (virtual_stack_registers_memory_p (x))
return x;
/* In theory we should not be seeing addresses of the form reg+0,
but just in case it is generated, optimize it away. */
if (GET_CODE (x) == PLUS && XEXP (x, 1) == const0_rtx)
return force_reg (Pmode, XEXP (x, 0));
/* For TImode with load/store quad, restrict addresses to just a single
pointer, so it works with both GPRs and VSX registers. */
/* Make sure both operands are registers. */
else if (GET_CODE (x) == PLUS
&& (mode != TImode || !TARGET_VSX))
return gen_rtx_PLUS (Pmode,
force_reg (Pmode, XEXP (x, 0)),
force_reg (Pmode, XEXP (x, 1)));
else
return force_reg (Pmode, x);
}
if (GET_CODE (x) == SYMBOL_REF)
{
enum tls_model model = SYMBOL_REF_TLS_MODEL (x);
if (model != 0)
return rs6000_legitimize_tls_address (x, model);
}
extra = 0;
switch (mode)
{
case E_TFmode:
case E_TDmode:
case E_TImode:
case E_PTImode:
case E_IFmode:
case E_KFmode:
/* As in legitimate_offset_address_p we do not assume
worst-case. The mode here is just a hint as to the registers
used. A TImode is usually in gprs, but may actually be in
fprs. Leave worst-case scenario for reload to handle via
insn constraints. PTImode is only GPRs. */
extra = 8;
break;
default:
break;
}
if (GET_CODE (x) == PLUS
&& GET_CODE (XEXP (x, 0)) == REG
&& GET_CODE (XEXP (x, 1)) == CONST_INT
&& ((unsigned HOST_WIDE_INT) (INTVAL (XEXP (x, 1)) + 0x8000)
>= 0x10000 - extra)
&& !PAIRED_VECTOR_MODE (mode))
{
HOST_WIDE_INT high_int, low_int;
rtx sum;
low_int = ((INTVAL (XEXP (x, 1)) & 0xffff) ^ 0x8000) - 0x8000;
if (low_int >= 0x8000 - extra)
low_int = 0;
high_int = INTVAL (XEXP (x, 1)) - low_int;
sum = force_operand (gen_rtx_PLUS (Pmode, XEXP (x, 0),
GEN_INT (high_int)), 0);
return plus_constant (Pmode, sum, low_int);
}
else if (GET_CODE (x) == PLUS
&& GET_CODE (XEXP (x, 0)) == REG
&& GET_CODE (XEXP (x, 1)) != CONST_INT
&& GET_MODE_NUNITS (mode) == 1
&& (GET_MODE_SIZE (mode) <= UNITS_PER_WORD
|| (/* ??? Assume floating point reg based on mode? */
(TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT)
&& (mode == DFmode || mode == DDmode)))
&& !avoiding_indexed_address_p (mode))
{
return gen_rtx_PLUS (Pmode, XEXP (x, 0),
force_reg (Pmode, force_operand (XEXP (x, 1), 0)));
}
else if (PAIRED_VECTOR_MODE (mode))
{
if (mode == DImode)
return x;
/* We accept [reg + reg]. */
if (GET_CODE (x) == PLUS)
{
rtx op1 = XEXP (x, 0);
rtx op2 = XEXP (x, 1);
rtx y;
op1 = force_reg (Pmode, op1);
op2 = force_reg (Pmode, op2);
/* We can't always do [reg + reg] for these, because [reg +
reg + offset] is not a legitimate addressing mode. */
y = gen_rtx_PLUS (Pmode, op1, op2);
if ((GET_MODE_SIZE (mode) > 8 || mode == DDmode) && REG_P (op2))
return force_reg (Pmode, y);
else
return y;
}
return force_reg (Pmode, x);
}
else if ((TARGET_ELF
#if TARGET_MACHO
|| !MACHO_DYNAMIC_NO_PIC_P
#endif
)
&& TARGET_32BIT
&& TARGET_NO_TOC
&& ! flag_pic
&& GET_CODE (x) != CONST_INT
&& GET_CODE (x) != CONST_WIDE_INT
&& GET_CODE (x) != CONST_DOUBLE
&& CONSTANT_P (x)
&& GET_MODE_NUNITS (mode) == 1
&& (GET_MODE_SIZE (mode) <= UNITS_PER_WORD
|| (/* ??? Assume floating point reg based on mode? */
(TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT)
&& (mode == DFmode || mode == DDmode))))
{
rtx reg = gen_reg_rtx (Pmode);
if (TARGET_ELF)
emit_insn (gen_elf_high (reg, x));
else
emit_insn (gen_macho_high (reg, x));
return gen_rtx_LO_SUM (Pmode, reg, x);
}
else if (TARGET_TOC
&& GET_CODE (x) == SYMBOL_REF
&& constant_pool_expr_p (x)
&& ASM_OUTPUT_SPECIAL_POOL_ENTRY_P (get_pool_constant (x), Pmode))
return create_TOC_reference (x, NULL_RTX);
else
return x;
}
/* Debug version of rs6000_legitimize_address. */
static rtx
rs6000_debug_legitimize_address (rtx x, rtx oldx, machine_mode mode)
{
rtx ret;
rtx_insn *insns;
start_sequence ();
ret = rs6000_legitimize_address (x, oldx, mode);
insns = get_insns ();
end_sequence ();
if (ret != x)
{
fprintf (stderr,
"\nrs6000_legitimize_address: mode %s, old code %s, "
"new code %s, modified\n",
GET_MODE_NAME (mode), GET_RTX_NAME (GET_CODE (x)),
GET_RTX_NAME (GET_CODE (ret)));
fprintf (stderr, "Original address:\n");
debug_rtx (x);
fprintf (stderr, "oldx:\n");
debug_rtx (oldx);
fprintf (stderr, "New address:\n");
debug_rtx (ret);
if (insns)
{
fprintf (stderr, "Insns added:\n");
debug_rtx_list (insns, 20);
}
}
else
{
fprintf (stderr,
"\nrs6000_legitimize_address: mode %s, code %s, no change:\n",
GET_MODE_NAME (mode), GET_RTX_NAME (GET_CODE (x)));
debug_rtx (x);
}
if (insns)
emit_insn (insns);
return ret;
}
/* This is called from dwarf2out.c via TARGET_ASM_OUTPUT_DWARF_DTPREL.
We need to emit DTP-relative relocations. */
static void rs6000_output_dwarf_dtprel (FILE *, int, rtx) ATTRIBUTE_UNUSED;
static void
rs6000_output_dwarf_dtprel (FILE *file, int size, rtx x)
{
switch (size)
{
case 4:
fputs ("\t.long\t", file);
break;
case 8:
fputs (DOUBLE_INT_ASM_OP, file);
break;
default:
gcc_unreachable ();
}
output_addr_const (file, x);
if (TARGET_ELF)
fputs ("@dtprel+0x8000", file);
else if (TARGET_XCOFF && GET_CODE (x) == SYMBOL_REF)
{
switch (SYMBOL_REF_TLS_MODEL (x))
{
case 0:
break;
case TLS_MODEL_LOCAL_EXEC:
fputs ("@le", file);
break;
case TLS_MODEL_INITIAL_EXEC:
fputs ("@ie", file);
break;
case TLS_MODEL_GLOBAL_DYNAMIC:
case TLS_MODEL_LOCAL_DYNAMIC:
fputs ("@m", file);
break;
default:
gcc_unreachable ();
}
}
}
/* Return true if X is a symbol that refers to real (rather than emulated)
TLS. */
static bool
rs6000_real_tls_symbol_ref_p (rtx x)
{
return (GET_CODE (x) == SYMBOL_REF
&& SYMBOL_REF_TLS_MODEL (x) >= TLS_MODEL_REAL);
}
/* In the name of slightly smaller debug output, and to cater to
general assembler lossage, recognize various UNSPEC sequences
and turn them back into a direct symbol reference. */
static rtx
rs6000_delegitimize_address (rtx orig_x)
{
rtx x, y, offset;
orig_x = delegitimize_mem_from_attrs (orig_x);
x = orig_x;
if (MEM_P (x))
x = XEXP (x, 0);
y = x;
if (TARGET_CMODEL != CMODEL_SMALL
&& GET_CODE (y) == LO_SUM)
y = XEXP (y, 1);
offset = NULL_RTX;
if (GET_CODE (y) == PLUS
&& GET_MODE (y) == Pmode
&& CONST_INT_P (XEXP (y, 1)))
{
offset = XEXP (y, 1);
y = XEXP (y, 0);
}
if (GET_CODE (y) == UNSPEC
&& XINT (y, 1) == UNSPEC_TOCREL)
{
y = XVECEXP (y, 0, 0);
#ifdef HAVE_AS_TLS
/* Do not associate thread-local symbols with the original
constant pool symbol. */
if (TARGET_XCOFF
&& GET_CODE (y) == SYMBOL_REF
&& CONSTANT_POOL_ADDRESS_P (y)
&& rs6000_real_tls_symbol_ref_p (get_pool_constant (y)))
return orig_x;
#endif
if (offset != NULL_RTX)
y = gen_rtx_PLUS (Pmode, y, offset);
if (!MEM_P (orig_x))
return y;
else
return replace_equiv_address_nv (orig_x, y);
}
if (TARGET_MACHO
&& GET_CODE (orig_x) == LO_SUM
&& GET_CODE (XEXP (orig_x, 1)) == CONST)
{
y = XEXP (XEXP (orig_x, 1), 0);
if (GET_CODE (y) == UNSPEC
&& XINT (y, 1) == UNSPEC_MACHOPIC_OFFSET)
return XVECEXP (y, 0, 0);
}
return orig_x;
}
/* Return true if X shouldn't be emitted into the debug info.
The linker doesn't like .toc section references from
.debug_* sections, so reject .toc section symbols. */
static bool
rs6000_const_not_ok_for_debug_p (rtx x)
{
if (GET_CODE (x) == UNSPEC)
return true;
if (GET_CODE (x) == SYMBOL_REF
&& CONSTANT_POOL_ADDRESS_P (x))
{
rtx c = get_pool_constant (x);
machine_mode cmode = get_pool_mode (x);
if (ASM_OUTPUT_SPECIAL_POOL_ENTRY_P (c, cmode))
return true;
}
return false;
}
/* Implement the TARGET_LEGITIMATE_COMBINED_INSN hook. */
static bool
rs6000_legitimate_combined_insn (rtx_insn *insn)
{
int icode = INSN_CODE (insn);
/* Reject creating doloop insns. Combine should not be allowed
to create these for a number of reasons:
1) In a nested loop, if combine creates one of these in an
outer loop and the register allocator happens to allocate ctr
to the outer loop insn, then the inner loop can't use ctr.
Inner loops ought to be more highly optimized.
2) Combine often wants to create one of these from what was
originally a three insn sequence, first combining the three
insns to two, then to ctrsi/ctrdi. When ctrsi/ctrdi is not
allocated ctr, the splitter takes use back to the three insn
sequence. It's better to stop combine at the two insn
sequence.
3) Faced with not being able to allocate ctr for ctrsi/crtdi
insns, the register allocator sometimes uses floating point
or vector registers for the pseudo. Since ctrsi/ctrdi is a
jump insn and output reloads are not implemented for jumps,
the ctrsi/ctrdi splitters need to handle all possible cases.
That's a pain, and it gets to be seriously difficult when a
splitter that runs after reload needs memory to transfer from
a gpr to fpr. See PR70098 and PR71763 which are not fixed
for the difficult case. It's better to not create problems
in the first place. */
if (icode != CODE_FOR_nothing
&& (icode == CODE_FOR_ctrsi_internal1
|| icode == CODE_FOR_ctrdi_internal1
|| icode == CODE_FOR_ctrsi_internal2
|| icode == CODE_FOR_ctrdi_internal2))
return false;
return true;
}
/* Construct the SYMBOL_REF for the tls_get_addr function. */
static GTY(()) rtx rs6000_tls_symbol;
static rtx
rs6000_tls_get_addr (void)
{
if (!rs6000_tls_symbol)
rs6000_tls_symbol = init_one_libfunc ("__tls_get_addr");
return rs6000_tls_symbol;
}
/* Construct the SYMBOL_REF for TLS GOT references. */
static GTY(()) rtx rs6000_got_symbol;
static rtx
rs6000_got_sym (void)
{
if (!rs6000_got_symbol)
{
rs6000_got_symbol = gen_rtx_SYMBOL_REF (Pmode, "_GLOBAL_OFFSET_TABLE_");
SYMBOL_REF_FLAGS (rs6000_got_symbol) |= SYMBOL_FLAG_LOCAL;
SYMBOL_REF_FLAGS (rs6000_got_symbol) |= SYMBOL_FLAG_EXTERNAL;
}
return rs6000_got_symbol;
}
/* AIX Thread-Local Address support. */
static rtx
rs6000_legitimize_tls_address_aix (rtx addr, enum tls_model model)
{
rtx sym, mem, tocref, tlsreg, tmpreg, dest, tlsaddr;
const char *name;
char *tlsname;
name = XSTR (addr, 0);
/* Append TLS CSECT qualifier, unless the symbol already is qualified
or the symbol will be in TLS private data section. */
if (name[strlen (name) - 1] != ']'
&& (TREE_PUBLIC (SYMBOL_REF_DECL (addr))
|| bss_initializer_p (SYMBOL_REF_DECL (addr))))
{
tlsname = XALLOCAVEC (char, strlen (name) + 4);
strcpy (tlsname, name);
strcat (tlsname,
bss_initializer_p (SYMBOL_REF_DECL (addr)) ? "[UL]" : "[TL]");
tlsaddr = copy_rtx (addr);
XSTR (tlsaddr, 0) = ggc_strdup (tlsname);
}
else
tlsaddr = addr;
/* Place addr into TOC constant pool. */
sym = force_const_mem (GET_MODE (tlsaddr), tlsaddr);
/* Output the TOC entry and create the MEM referencing the value. */
if (constant_pool_expr_p (XEXP (sym, 0))
&& ASM_OUTPUT_SPECIAL_POOL_ENTRY_P (get_pool_constant (XEXP (sym, 0)), Pmode))
{
tocref = create_TOC_reference (XEXP (sym, 0), NULL_RTX);
mem = gen_const_mem (Pmode, tocref);
set_mem_alias_set (mem, get_TOC_alias_set ());
}
else
return sym;
/* Use global-dynamic for local-dynamic. */
if (model == TLS_MODEL_GLOBAL_DYNAMIC
|| model == TLS_MODEL_LOCAL_DYNAMIC)
{
/* Create new TOC reference for @m symbol. */
name = XSTR (XVECEXP (XEXP (mem, 0), 0, 0), 0);
tlsname = XALLOCAVEC (char, strlen (name) + 1);
strcpy (tlsname, "*LCM");
strcat (tlsname, name + 3);
rtx modaddr = gen_rtx_SYMBOL_REF (Pmode, ggc_strdup (tlsname));
SYMBOL_REF_FLAGS (modaddr) |= SYMBOL_FLAG_LOCAL;
tocref = create_TOC_reference (modaddr, NULL_RTX);
rtx modmem = gen_const_mem (Pmode, tocref);
set_mem_alias_set (modmem, get_TOC_alias_set ());
rtx modreg = gen_reg_rtx (Pmode);
emit_insn (gen_rtx_SET (modreg, modmem));
tmpreg = gen_reg_rtx (Pmode);
emit_insn (gen_rtx_SET (tmpreg, mem));
dest = gen_reg_rtx (Pmode);
if (TARGET_32BIT)
emit_insn (gen_tls_get_addrsi (dest, modreg, tmpreg));
else
emit_insn (gen_tls_get_addrdi (dest, modreg, tmpreg));
return dest;
}
/* Obtain TLS pointer: 32 bit call or 64 bit GPR 13. */
else if (TARGET_32BIT)
{
tlsreg = gen_reg_rtx (SImode);
emit_insn (gen_tls_get_tpointer (tlsreg));
}
else
tlsreg = gen_rtx_REG (DImode, 13);
/* Load the TOC value into temporary register. */
tmpreg = gen_reg_rtx (Pmode);
emit_insn (gen_rtx_SET (tmpreg, mem));
set_unique_reg_note (get_last_insn (), REG_EQUAL,
gen_rtx_MINUS (Pmode, addr, tlsreg));
/* Add TOC symbol value to TLS pointer. */
dest = force_reg (Pmode, gen_rtx_PLUS (Pmode, tmpreg, tlsreg));
return dest;
}
/* ADDR contains a thread-local SYMBOL_REF. Generate code to compute
this (thread-local) address. */
static rtx
rs6000_legitimize_tls_address (rtx addr, enum tls_model model)
{
rtx dest, insn;
if (TARGET_XCOFF)
return rs6000_legitimize_tls_address_aix (addr, model);
dest = gen_reg_rtx (Pmode);
if (model == TLS_MODEL_LOCAL_EXEC && rs6000_tls_size == 16)
{
rtx tlsreg;
if (TARGET_64BIT)
{
tlsreg = gen_rtx_REG (Pmode, 13);
insn = gen_tls_tprel_64 (dest, tlsreg, addr);
}
else
{
tlsreg = gen_rtx_REG (Pmode, 2);
insn = gen_tls_tprel_32 (dest, tlsreg, addr);
}
emit_insn (insn);
}
else if (model == TLS_MODEL_LOCAL_EXEC && rs6000_tls_size == 32)
{
rtx tlsreg, tmp;
tmp = gen_reg_rtx (Pmode);
if (TARGET_64BIT)
{
tlsreg = gen_rtx_REG (Pmode, 13);
insn = gen_tls_tprel_ha_64 (tmp, tlsreg, addr);
}
else
{
tlsreg = gen_rtx_REG (Pmode, 2);
insn = gen_tls_tprel_ha_32 (tmp, tlsreg, addr);
}
emit_insn (insn);
if (TARGET_64BIT)
insn = gen_tls_tprel_lo_64 (dest, tmp, addr);
else
insn = gen_tls_tprel_lo_32 (dest, tmp, addr);
emit_insn (insn);
}
else
{
rtx r3, got, tga, tmp1, tmp2, call_insn;
/* We currently use relocations like @got@tlsgd for tls, which
means the linker will handle allocation of tls entries, placing
them in the .got section. So use a pointer to the .got section,
not one to secondary TOC sections used by 64-bit -mminimal-toc,
or to secondary GOT sections used by 32-bit -fPIC. */
if (TARGET_64BIT)
got = gen_rtx_REG (Pmode, 2);
else
{
if (flag_pic == 1)
got = gen_rtx_REG (Pmode, RS6000_PIC_OFFSET_TABLE_REGNUM);
else
{
rtx gsym = rs6000_got_sym ();
got = gen_reg_rtx (Pmode);
if (flag_pic == 0)
rs6000_emit_move (got, gsym, Pmode);
else
{
rtx mem, lab;
tmp1 = gen_reg_rtx (Pmode);
tmp2 = gen_reg_rtx (Pmode);
mem = gen_const_mem (Pmode, tmp1);
lab = gen_label_rtx ();
emit_insn (gen_load_toc_v4_PIC_1b (gsym, lab));
emit_move_insn (tmp1, gen_rtx_REG (Pmode, LR_REGNO));
if (TARGET_LINK_STACK)
emit_insn (gen_addsi3 (tmp1, tmp1, GEN_INT (4)));
emit_move_insn (tmp2, mem);
rtx_insn *last = emit_insn (gen_addsi3 (got, tmp1, tmp2));
set_unique_reg_note (last, REG_EQUAL, gsym);
}
}
}
if (model == TLS_MODEL_GLOBAL_DYNAMIC)
{
tga = rs6000_tls_get_addr ();
emit_library_call_value (tga, dest, LCT_CONST, Pmode,
const0_rtx, Pmode);
r3 = gen_rtx_REG (Pmode, 3);
if (DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
{
if (TARGET_64BIT)
insn = gen_tls_gd_aix64 (r3, got, addr, tga, const0_rtx);
else
insn = gen_tls_gd_aix32 (r3, got, addr, tga, const0_rtx);
}
else if (DEFAULT_ABI == ABI_V4)
insn = gen_tls_gd_sysvsi (r3, got, addr, tga, const0_rtx);
else
gcc_unreachable ();
call_insn = last_call_insn ();
PATTERN (call_insn) = insn;
if (DEFAULT_ABI == ABI_V4 && TARGET_SECURE_PLT && flag_pic)
use_reg (&CALL_INSN_FUNCTION_USAGE (call_insn),
pic_offset_table_rtx);
}
else if (model == TLS_MODEL_LOCAL_DYNAMIC)
{
tga = rs6000_tls_get_addr ();
tmp1 = gen_reg_rtx (Pmode);
emit_library_call_value (tga, tmp1, LCT_CONST, Pmode,
const0_rtx, Pmode);
r3 = gen_rtx_REG (Pmode, 3);
if (DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
{
if (TARGET_64BIT)
insn = gen_tls_ld_aix64 (r3, got, tga, const0_rtx);
else
insn = gen_tls_ld_aix32 (r3, got, tga, const0_rtx);
}
else if (DEFAULT_ABI == ABI_V4)
insn = gen_tls_ld_sysvsi (r3, got, tga, const0_rtx);
else
gcc_unreachable ();
call_insn = last_call_insn ();
PATTERN (call_insn) = insn;
if (DEFAULT_ABI == ABI_V4 && TARGET_SECURE_PLT && flag_pic)
use_reg (&CALL_INSN_FUNCTION_USAGE (call_insn),
pic_offset_table_rtx);
if (rs6000_tls_size == 16)
{
if (TARGET_64BIT)
insn = gen_tls_dtprel_64 (dest, tmp1, addr);
else
insn = gen_tls_dtprel_32 (dest, tmp1, addr);
}
else if (rs6000_tls_size == 32)
{
tmp2 = gen_reg_rtx (Pmode);
if (TARGET_64BIT)
insn = gen_tls_dtprel_ha_64 (tmp2, tmp1, addr);
else
insn = gen_tls_dtprel_ha_32 (tmp2, tmp1, addr);
emit_insn (insn);
if (TARGET_64BIT)
insn = gen_tls_dtprel_lo_64 (dest, tmp2, addr);
else
insn = gen_tls_dtprel_lo_32 (dest, tmp2, addr);
}
else
{
tmp2 = gen_reg_rtx (Pmode);
if (TARGET_64BIT)
insn = gen_tls_got_dtprel_64 (tmp2, got, addr);
else
insn = gen_tls_got_dtprel_32 (tmp2, got, addr);
emit_insn (insn);
insn = gen_rtx_SET (dest, gen_rtx_PLUS (Pmode, tmp2, tmp1));
}
emit_insn (insn);
}
else
{
/* IE, or 64-bit offset LE. */
tmp2 = gen_reg_rtx (Pmode);
if (TARGET_64BIT)
insn = gen_tls_got_tprel_64 (tmp2, got, addr);
else
insn = gen_tls_got_tprel_32 (tmp2, got, addr);
emit_insn (insn);
if (TARGET_64BIT)
insn = gen_tls_tls_64 (dest, tmp2, addr);
else
insn = gen_tls_tls_32 (dest, tmp2, addr);
emit_insn (insn);
}
}
return dest;
}
/* Only create the global variable for the stack protect guard if we are using
the global flavor of that guard. */
static tree
rs6000_init_stack_protect_guard (void)
{
if (rs6000_stack_protector_guard == SSP_GLOBAL)
return default_stack_protect_guard ();
return NULL_TREE;
}
/* Implement TARGET_CANNOT_FORCE_CONST_MEM. */
static bool
rs6000_cannot_force_const_mem (machine_mode mode ATTRIBUTE_UNUSED, rtx x)
{
if (GET_CODE (x) == HIGH
&& GET_CODE (XEXP (x, 0)) == UNSPEC)
return true;
/* A TLS symbol in the TOC cannot contain a sum. */
if (GET_CODE (x) == CONST
&& GET_CODE (XEXP (x, 0)) == PLUS
&& GET_CODE (XEXP (XEXP (x, 0), 0)) == SYMBOL_REF
&& SYMBOL_REF_TLS_MODEL (XEXP (XEXP (x, 0), 0)) != 0)
return true;
/* Do not place an ELF TLS symbol in the constant pool. */
return TARGET_ELF && tls_referenced_p (x);
}
/* Return true iff the given SYMBOL_REF refers to a constant pool entry
that we have put in the TOC, or for cmodel=medium, if the SYMBOL_REF
can be addressed relative to the toc pointer. */
static bool
use_toc_relative_ref (rtx sym, machine_mode mode)
{
return ((constant_pool_expr_p (sym)
&& ASM_OUTPUT_SPECIAL_POOL_ENTRY_P (get_pool_constant (sym),
get_pool_mode (sym)))
|| (TARGET_CMODEL == CMODEL_MEDIUM
&& SYMBOL_REF_LOCAL_P (sym)
&& GET_MODE_SIZE (mode) <= POWERPC64_TOC_POINTER_ALIGNMENT));
}
/* Our implementation of LEGITIMIZE_RELOAD_ADDRESS. Returns a value to
replace the input X, or the original X if no replacement is called for.
The output parameter *WIN is 1 if the calling macro should goto WIN,
0 if it should not.
For RS/6000, we wish to handle large displacements off a base
register by splitting the addend across an addiu/addis and the mem insn.
This cuts number of extra insns needed from 3 to 1.
On Darwin, we use this to generate code for floating point constants.
A movsf_low is generated so we wind up with 2 instructions rather than 3.
The Darwin code is inside #if TARGET_MACHO because only then are the
machopic_* functions defined. */
static rtx
rs6000_legitimize_reload_address (rtx x, machine_mode mode,
int opnum, int type,
int ind_levels ATTRIBUTE_UNUSED, int *win)
{
bool reg_offset_p = reg_offset_addressing_ok_p (mode);
bool quad_offset_p = mode_supports_vsx_dform_quad (mode);
/* Nasty hack for vsx_splat_v2df/v2di load from mem, which takes a
DFmode/DImode MEM. Ditto for ISA 3.0 vsx_splat_v4sf/v4si. */
if (reg_offset_p
&& opnum == 1
&& ((mode == DFmode && recog_data.operand_mode[0] == V2DFmode)
|| (mode == DImode && recog_data.operand_mode[0] == V2DImode)
|| (mode == SFmode && recog_data.operand_mode[0] == V4SFmode
&& TARGET_P9_VECTOR)
|| (mode == SImode && recog_data.operand_mode[0] == V4SImode
&& TARGET_P9_VECTOR)))
reg_offset_p = false;
/* We must recognize output that we have already generated ourselves. */
if (GET_CODE (x) == PLUS
&& GET_CODE (XEXP (x, 0)) == PLUS
&& GET_CODE (XEXP (XEXP (x, 0), 0)) == REG
&& GET_CODE (XEXP (XEXP (x, 0), 1)) == CONST_INT
&& GET_CODE (XEXP (x, 1)) == CONST_INT)
{
if (TARGET_DEBUG_ADDR)
{
fprintf (stderr, "\nlegitimize_reload_address push_reload #1:\n");
debug_rtx (x);
}
push_reload (XEXP (x, 0), NULL_RTX, &XEXP (x, 0), NULL,
BASE_REG_CLASS, GET_MODE (x), VOIDmode, 0, 0,
opnum, (enum reload_type) type);
*win = 1;
return x;
}
/* Likewise for (lo_sum (high ...) ...) output we have generated. */
if (GET_CODE (x) == LO_SUM
&& GET_CODE (XEXP (x, 0)) == HIGH)
{
if (TARGET_DEBUG_ADDR)
{
fprintf (stderr, "\nlegitimize_reload_address push_reload #2:\n");
debug_rtx (x);
}
push_reload (XEXP (x, 0), NULL_RTX, &XEXP (x, 0), NULL,
BASE_REG_CLASS, Pmode, VOIDmode, 0, 0,
opnum, (enum reload_type) type);
*win = 1;
return x;
}
#if TARGET_MACHO
if (DEFAULT_ABI == ABI_DARWIN && flag_pic
&& GET_CODE (x) == LO_SUM
&& GET_CODE (XEXP (x, 0)) == PLUS
&& XEXP (XEXP (x, 0), 0) == pic_offset_table_rtx
&& GET_CODE (XEXP (XEXP (x, 0), 1)) == HIGH
&& XEXP (XEXP (XEXP (x, 0), 1), 0) == XEXP (x, 1)
&& machopic_operand_p (XEXP (x, 1)))
{
/* Result of previous invocation of this function on Darwin
floating point constant. */
push_reload (XEXP (x, 0), NULL_RTX, &XEXP (x, 0), NULL,
BASE_REG_CLASS, Pmode, VOIDmode, 0, 0,
opnum, (enum reload_type) type);
*win = 1;
return x;
}
#endif
if (TARGET_CMODEL != CMODEL_SMALL
&& reg_offset_p
&& !quad_offset_p
&& small_toc_ref (x, VOIDmode))
{
rtx hi = gen_rtx_HIGH (Pmode, copy_rtx (x));
x = gen_rtx_LO_SUM (Pmode, hi, x);
if (TARGET_DEBUG_ADDR)
{
fprintf (stderr, "\nlegitimize_reload_address push_reload #3:\n");
debug_rtx (x);
}
push_reload (XEXP (x, 0), NULL_RTX, &XEXP (x, 0), NULL,
BASE_REG_CLASS, Pmode, VOIDmode, 0, 0,
opnum, (enum reload_type) type);
*win = 1;
return x;
}
if (GET_CODE (x) == PLUS
&& REG_P (XEXP (x, 0))
&& REGNO (XEXP (x, 0)) < FIRST_PSEUDO_REGISTER
&& INT_REG_OK_FOR_BASE_P (XEXP (x, 0), 1)
&& CONST_INT_P (XEXP (x, 1))
&& reg_offset_p
&& !PAIRED_VECTOR_MODE (mode)
&& (quad_offset_p || !VECTOR_MODE_P (mode) || VECTOR_MEM_NONE_P (mode)))
{
HOST_WIDE_INT val = INTVAL (XEXP (x, 1));
HOST_WIDE_INT low = ((val & 0xffff) ^ 0x8000) - 0x8000;
HOST_WIDE_INT high
= (((val - low) & 0xffffffff) ^ 0x80000000) - 0x80000000;
/* Check for 32-bit overflow or quad addresses with one of the
four least significant bits set. */
if (high + low != val
|| (quad_offset_p && (low & 0xf)))
{
*win = 0;
return x;
}
/* Reload the high part into a base reg; leave the low part
in the mem directly. */
x = gen_rtx_PLUS (GET_MODE (x),
gen_rtx_PLUS (GET_MODE (x), XEXP (x, 0),
GEN_INT (high)),
GEN_INT (low));
if (TARGET_DEBUG_ADDR)
{
fprintf (stderr, "\nlegitimize_reload_address push_reload #4:\n");
debug_rtx (x);
}
push_reload (XEXP (x, 0), NULL_RTX, &XEXP (x, 0), NULL,
BASE_REG_CLASS, GET_MODE (x), VOIDmode, 0, 0,
opnum, (enum reload_type) type);
*win = 1;
return x;
}
if (GET_CODE (x) == SYMBOL_REF
&& reg_offset_p
&& !quad_offset_p
&& (!VECTOR_MODE_P (mode) || VECTOR_MEM_NONE_P (mode))
&& !PAIRED_VECTOR_MODE (mode)
#if TARGET_MACHO
&& DEFAULT_ABI == ABI_DARWIN
&& (flag_pic || MACHO_DYNAMIC_NO_PIC_P)
&& machopic_symbol_defined_p (x)
#else
&& DEFAULT_ABI == ABI_V4
&& !flag_pic
#endif
/* Don't do this for TFmode or TDmode, since the result isn't offsettable.
The same goes for DImode without 64-bit gprs and DFmode and DDmode
without fprs.
??? Assume floating point reg based on mode? This assumption is
violated by eg. powerpc-linux -m32 compile of gcc.dg/pr28796-2.c
where reload ends up doing a DFmode load of a constant from
mem using two gprs. Unfortunately, at this point reload
hasn't yet selected regs so poking around in reload data
won't help and even if we could figure out the regs reliably,
we'd still want to allow this transformation when the mem is
naturally aligned. Since we say the address is good here, we
can't disable offsets from LO_SUMs in mem_operand_gpr.
FIXME: Allow offset from lo_sum for other modes too, when
mem is sufficiently aligned.
Also disallow this if the type can go in VMX/Altivec registers, since
those registers do not have d-form (reg+offset) address modes. */
&& !reg_addr[mode].scalar_in_vmx_p
&& mode != TFmode
&& mode != TDmode
&& mode != IFmode
&& mode != KFmode
&& (mode != TImode || !TARGET_VSX)
&& mode != PTImode
&& (mode != DImode || TARGET_POWERPC64)
&& ((mode != DFmode && mode != DDmode) || TARGET_POWERPC64
|| (TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT)))
{
#if TARGET_MACHO
if (flag_pic)
{
rtx offset = machopic_gen_offset (x);
x = gen_rtx_LO_SUM (GET_MODE (x),
gen_rtx_PLUS (Pmode, pic_offset_table_rtx,
gen_rtx_HIGH (Pmode, offset)), offset);
}
else
#endif
x = gen_rtx_LO_SUM (GET_MODE (x),
gen_rtx_HIGH (Pmode, x), x);
if (TARGET_DEBUG_ADDR)
{
fprintf (stderr, "\nlegitimize_reload_address push_reload #5:\n");
debug_rtx (x);
}
push_reload (XEXP (x, 0), NULL_RTX, &XEXP (x, 0), NULL,
BASE_REG_CLASS, Pmode, VOIDmode, 0, 0,
opnum, (enum reload_type) type);
*win = 1;
return x;
}
/* Reload an offset address wrapped by an AND that represents the
masking of the lower bits. Strip the outer AND and let reload
convert the offset address into an indirect address. For VSX,
force reload to create the address with an AND in a separate
register, because we can't guarantee an altivec register will
be used. */
if (VECTOR_MEM_ALTIVEC_P (mode)
&& GET_CODE (x) == AND
&& GET_CODE (XEXP (x, 0)) == PLUS
&& GET_CODE (XEXP (XEXP (x, 0), 0)) == REG
&& GET_CODE (XEXP (XEXP (x, 0), 1)) == CONST_INT
&& GET_CODE (XEXP (x, 1)) == CONST_INT
&& INTVAL (XEXP (x, 1)) == -16)
{
x = XEXP (x, 0);
*win = 1;
return x;
}
if (TARGET_TOC
&& reg_offset_p
&& !quad_offset_p
&& GET_CODE (x) == SYMBOL_REF
&& use_toc_relative_ref (x, mode))
{
x = create_TOC_reference (x, NULL_RTX);
if (TARGET_CMODEL != CMODEL_SMALL)
{
if (TARGET_DEBUG_ADDR)
{
fprintf (stderr, "\nlegitimize_reload_address push_reload #6:\n");
debug_rtx (x);
}
push_reload (XEXP (x, 0), NULL_RTX, &XEXP (x, 0), NULL,
BASE_REG_CLASS, Pmode, VOIDmode, 0, 0,
opnum, (enum reload_type) type);
}
*win = 1;
return x;
}
*win = 0;
return x;
}
/* Debug version of rs6000_legitimize_reload_address. */
static rtx
rs6000_debug_legitimize_reload_address (rtx x, machine_mode mode,
int opnum, int type,
int ind_levels, int *win)
{
rtx ret = rs6000_legitimize_reload_address (x, mode, opnum, type,
ind_levels, win);
fprintf (stderr,
"\nrs6000_legitimize_reload_address: mode = %s, opnum = %d, "
"type = %d, ind_levels = %d, win = %d, original addr:\n",
GET_MODE_NAME (mode), opnum, type, ind_levels, *win);
debug_rtx (x);
if (x == ret)
fprintf (stderr, "Same address returned\n");
else if (!ret)
fprintf (stderr, "NULL returned\n");
else
{
fprintf (stderr, "New address:\n");
debug_rtx (ret);
}
return ret;
}
/* TARGET_LEGITIMATE_ADDRESS_P recognizes an RTL expression
that is a valid memory address for an instruction.
The MODE argument is the machine mode for the MEM expression
that wants to use this address.
On the RS/6000, there are four valid address: a SYMBOL_REF that
refers to a constant pool entry of an address (or the sum of it
plus a constant), a short (16-bit signed) constant plus a register,
the sum of two registers, or a register indirect, possibly with an
auto-increment. For DFmode, DDmode and DImode with a constant plus
register, we must ensure that both words are addressable or PowerPC64
with offset word aligned.
For modes spanning multiple registers (DFmode and DDmode in 32-bit GPRs,
32-bit DImode, TImode, TFmode, TDmode), indexed addressing cannot be used
because adjacent memory cells are accessed by adding word-sized offsets
during assembly output. */
static bool
rs6000_legitimate_address_p (machine_mode mode, rtx x, bool reg_ok_strict)
{
bool reg_offset_p = reg_offset_addressing_ok_p (mode);
bool quad_offset_p = mode_supports_vsx_dform_quad (mode);
/* If this is an unaligned stvx/ldvx type address, discard the outer AND. */
if (VECTOR_MEM_ALTIVEC_P (mode)
&& GET_CODE (x) == AND
&& GET_CODE (XEXP (x, 1)) == CONST_INT
&& INTVAL (XEXP (x, 1)) == -16)
x = XEXP (x, 0);
if (TARGET_ELF && RS6000_SYMBOL_REF_TLS_P (x))
return 0;
if (legitimate_indirect_address_p (x, reg_ok_strict))
return 1;
if (TARGET_UPDATE
&& (GET_CODE (x) == PRE_INC || GET_CODE (x) == PRE_DEC)
&& mode_supports_pre_incdec_p (mode)
&& legitimate_indirect_address_p (XEXP (x, 0), reg_ok_strict))
return 1;
/* Handle restricted vector d-form offsets in ISA 3.0. */
if (quad_offset_p)
{
if (quad_address_p (x, mode, reg_ok_strict))
return 1;
}
else if (virtual_stack_registers_memory_p (x))
return 1;
else if (reg_offset_p)
{
if (legitimate_small_data_p (mode, x))
return 1;
if (legitimate_constant_pool_address_p (x, mode,
reg_ok_strict || lra_in_progress))
return 1;
if (reg_addr[mode].fused_toc && GET_CODE (x) == UNSPEC
&& XINT (x, 1) == UNSPEC_FUSION_ADDIS)
return 1;
}
/* For TImode, if we have TImode in VSX registers, only allow register
indirect addresses. This will allow the values to go in either GPRs
or VSX registers without reloading. The vector types would tend to
go into VSX registers, so we allow REG+REG, while TImode seems
somewhat split, in that some uses are GPR based, and some VSX based. */
/* FIXME: We could loosen this by changing the following to
if (mode == TImode && TARGET_QUAD_MEMORY && TARGET_VSX)
but currently we cannot allow REG+REG addressing for TImode. See
PR72827 for complete details on how this ends up hoodwinking DSE. */
if (mode == TImode && TARGET_VSX)
return 0;
/* If not REG_OK_STRICT (before reload) let pass any stack offset. */
if (! reg_ok_strict
&& reg_offset_p
&& GET_CODE (x) == PLUS
&& GET_CODE (XEXP (x, 0)) == REG
&& (XEXP (x, 0) == virtual_stack_vars_rtx
|| XEXP (x, 0) == arg_pointer_rtx)
&& GET_CODE (XEXP (x, 1)) == CONST_INT)
return 1;
if (rs6000_legitimate_offset_address_p (mode, x, reg_ok_strict, false))
return 1;
if (!FLOAT128_2REG_P (mode)
&& ((TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT)
|| TARGET_POWERPC64
|| (mode != DFmode && mode != DDmode))
&& (TARGET_POWERPC64 || mode != DImode)
&& (mode != TImode || VECTOR_MEM_VSX_P (TImode))
&& mode != PTImode
&& !avoiding_indexed_address_p (mode)
&& legitimate_indexed_address_p (x, reg_ok_strict))
return 1;
if (TARGET_UPDATE && GET_CODE (x) == PRE_MODIFY
&& mode_supports_pre_modify_p (mode)
&& legitimate_indirect_address_p (XEXP (x, 0), reg_ok_strict)
&& (rs6000_legitimate_offset_address_p (mode, XEXP (x, 1),
reg_ok_strict, false)
|| (!avoiding_indexed_address_p (mode)
&& legitimate_indexed_address_p (XEXP (x, 1), reg_ok_strict)))
&& rtx_equal_p (XEXP (XEXP (x, 1), 0), XEXP (x, 0)))
return 1;
if (reg_offset_p && !quad_offset_p
&& legitimate_lo_sum_address_p (mode, x, reg_ok_strict))
return 1;
return 0;
}
/* Debug version of rs6000_legitimate_address_p. */
static bool
rs6000_debug_legitimate_address_p (machine_mode mode, rtx x,
bool reg_ok_strict)
{
bool ret = rs6000_legitimate_address_p (mode, x, reg_ok_strict);
fprintf (stderr,
"\nrs6000_legitimate_address_p: return = %s, mode = %s, "
"strict = %d, reload = %s, code = %s\n",
ret ? "true" : "false",
GET_MODE_NAME (mode),
reg_ok_strict,
(reload_completed ? "after" : "before"),
GET_RTX_NAME (GET_CODE (x)));
debug_rtx (x);
return ret;
}
/* Implement TARGET_MODE_DEPENDENT_ADDRESS_P. */
static bool
rs6000_mode_dependent_address_p (const_rtx addr,
addr_space_t as ATTRIBUTE_UNUSED)
{
return rs6000_mode_dependent_address_ptr (addr);
}
/* Go to LABEL if ADDR (a legitimate address expression)
has an effect that depends on the machine mode it is used for.
On the RS/6000 this is true of all integral offsets (since AltiVec
and VSX modes don't allow them) or is a pre-increment or decrement.
??? Except that due to conceptual problems in offsettable_address_p
we can't really report the problems of integral offsets. So leave
this assuming that the adjustable offset must be valid for the
sub-words of a TFmode operand, which is what we had before. */
static bool
rs6000_mode_dependent_address (const_rtx addr)
{
switch (GET_CODE (addr))
{
case PLUS:
/* Any offset from virtual_stack_vars_rtx and arg_pointer_rtx
is considered a legitimate address before reload, so there
are no offset restrictions in that case. Note that this
condition is safe in strict mode because any address involving
virtual_stack_vars_rtx or arg_pointer_rtx would already have
been rejected as illegitimate. */
if (XEXP (addr, 0) != virtual_stack_vars_rtx
&& XEXP (addr, 0) != arg_pointer_rtx
&& GET_CODE (XEXP (addr, 1)) == CONST_INT)
{
unsigned HOST_WIDE_INT val = INTVAL (XEXP (addr, 1));
return val + 0x8000 >= 0x10000 - (TARGET_POWERPC64 ? 8 : 12);
}
break;
case LO_SUM:
/* Anything in the constant pool is sufficiently aligned that
all bytes have the same high part address. */
return !legitimate_constant_pool_address_p (addr, QImode, false);
/* Auto-increment cases are now treated generically in recog.c. */
case PRE_MODIFY:
return TARGET_UPDATE;
/* AND is only allowed in Altivec loads. */
case AND:
return true;
default:
break;
}
return false;
}
/* Debug version of rs6000_mode_dependent_address. */
static bool
rs6000_debug_mode_dependent_address (const_rtx addr)
{
bool ret = rs6000_mode_dependent_address (addr);
fprintf (stderr, "\nrs6000_mode_dependent_address: ret = %s\n",
ret ? "true" : "false");
debug_rtx (addr);
return ret;
}
/* Implement FIND_BASE_TERM. */
rtx
rs6000_find_base_term (rtx op)
{
rtx base;
base = op;
if (GET_CODE (base) == CONST)
base = XEXP (base, 0);
if (GET_CODE (base) == PLUS)
base = XEXP (base, 0);
if (GET_CODE (base) == UNSPEC)
switch (XINT (base, 1))
{
case UNSPEC_TOCREL:
case UNSPEC_MACHOPIC_OFFSET:
/* OP represents SYM [+ OFFSET] - ANCHOR. SYM is the base term
for aliasing purposes. */
return XVECEXP (base, 0, 0);
}
return op;
}
/* More elaborate version of recog's offsettable_memref_p predicate
that works around the ??? note of rs6000_mode_dependent_address.
In particular it accepts
(mem:DI (plus:SI (reg/f:SI 31 31) (const_int 32760 [0x7ff8])))
in 32-bit mode, that the recog predicate rejects. */
static bool
rs6000_offsettable_memref_p (rtx op, machine_mode reg_mode)
{
bool worst_case;
if (!MEM_P (op))
return false;
/* First mimic offsettable_memref_p. */
if (offsettable_address_p (true, GET_MODE (op), XEXP (op, 0)))
return true;
/* offsettable_address_p invokes rs6000_mode_dependent_address, but
the latter predicate knows nothing about the mode of the memory
reference and, therefore, assumes that it is the largest supported
mode (TFmode). As a consequence, legitimate offsettable memory
references are rejected. rs6000_legitimate_offset_address_p contains
the correct logic for the PLUS case of rs6000_mode_dependent_address,
at least with a little bit of help here given that we know the
actual registers used. */
worst_case = ((TARGET_POWERPC64 && GET_MODE_CLASS (reg_mode) == MODE_INT)
|| GET_MODE_SIZE (reg_mode) == 4);
return rs6000_legitimate_offset_address_p (GET_MODE (op), XEXP (op, 0),
true, worst_case);
}
/* Determine the reassociation width to be used in reassociate_bb.
This takes into account how many parallel operations we
can actually do of a given type, and also the latency.
P8:
int add/sub 6/cycle
mul 2/cycle
vect add/sub/mul 2/cycle
fp add/sub/mul 2/cycle
dfp 1/cycle
*/
static int
rs6000_reassociation_width (unsigned int opc ATTRIBUTE_UNUSED,
machine_mode mode)
{
switch (rs6000_tune)
{
case PROCESSOR_POWER8:
case PROCESSOR_POWER9:
if (DECIMAL_FLOAT_MODE_P (mode))
return 1;
if (VECTOR_MODE_P (mode))
return 4;
if (INTEGRAL_MODE_P (mode))
return opc == MULT_EXPR ? 4 : 6;
if (FLOAT_MODE_P (mode))
return 4;
break;
default:
break;
}
return 1;
}
/* Change register usage conditional on target flags. */
static void
rs6000_conditional_register_usage (void)
{
int i;
if (TARGET_DEBUG_TARGET)
fprintf (stderr, "rs6000_conditional_register_usage called\n");
/* Set MQ register fixed (already call_used) so that it will not be
allocated. */
fixed_regs[64] = 1;
/* 64-bit AIX and Linux reserve GPR13 for thread-private data. */
if (TARGET_64BIT)
fixed_regs[13] = call_used_regs[13]
= call_really_used_regs[13] = 1;
/* Conditionally disable FPRs. */
if (TARGET_SOFT_FLOAT)
for (i = 32; i < 64; i++)
fixed_regs[i] = call_used_regs[i]
= call_really_used_regs[i] = 1;
/* The TOC register is not killed across calls in a way that is
visible to the compiler. */
if (DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
call_really_used_regs[2] = 0;
if (DEFAULT_ABI == ABI_V4 && flag_pic == 2)
fixed_regs[RS6000_PIC_OFFSET_TABLE_REGNUM] = 1;
if (DEFAULT_ABI == ABI_V4 && flag_pic == 1)
fixed_regs[RS6000_PIC_OFFSET_TABLE_REGNUM]
= call_used_regs[RS6000_PIC_OFFSET_TABLE_REGNUM]
= call_really_used_regs[RS6000_PIC_OFFSET_TABLE_REGNUM] = 1;
if (DEFAULT_ABI == ABI_DARWIN && flag_pic)
fixed_regs[RS6000_PIC_OFFSET_TABLE_REGNUM]
= call_used_regs[RS6000_PIC_OFFSET_TABLE_REGNUM]
= call_really_used_regs[RS6000_PIC_OFFSET_TABLE_REGNUM] = 1;
if (TARGET_TOC && TARGET_MINIMAL_TOC)
fixed_regs[RS6000_PIC_OFFSET_TABLE_REGNUM]
= call_used_regs[RS6000_PIC_OFFSET_TABLE_REGNUM] = 1;
if (!TARGET_ALTIVEC && !TARGET_VSX)
{
for (i = FIRST_ALTIVEC_REGNO; i <= LAST_ALTIVEC_REGNO; ++i)
fixed_regs[i] = call_used_regs[i] = call_really_used_regs[i] = 1;
call_really_used_regs[VRSAVE_REGNO] = 1;
}
if (TARGET_ALTIVEC || TARGET_VSX)
global_regs[VSCR_REGNO] = 1;
if (TARGET_ALTIVEC_ABI)
{
for (i = FIRST_ALTIVEC_REGNO; i < FIRST_ALTIVEC_REGNO + 20; ++i)
call_used_regs[i] = call_really_used_regs[i] = 1;
/* AIX reserves VR20:31 in non-extended ABI mode. */
if (TARGET_XCOFF)
for (i = FIRST_ALTIVEC_REGNO + 20; i < FIRST_ALTIVEC_REGNO + 32; ++i)
fixed_regs[i] = call_used_regs[i] = call_really_used_regs[i] = 1;
}
}
/* Output insns to set DEST equal to the constant SOURCE as a series of
lis, ori and shl instructions and return TRUE. */
bool
rs6000_emit_set_const (rtx dest, rtx source)
{
machine_mode mode = GET_MODE (dest);
rtx temp, set;
rtx_insn *insn;
HOST_WIDE_INT c;
gcc_checking_assert (CONST_INT_P (source));
c = INTVAL (source);
switch (mode)
{
case E_QImode:
case E_HImode:
emit_insn (gen_rtx_SET (dest, source));
return true;
case E_SImode:
temp = !can_create_pseudo_p () ? dest : gen_reg_rtx (SImode);
emit_insn (gen_rtx_SET (copy_rtx (temp),
GEN_INT (c & ~(HOST_WIDE_INT) 0xffff)));
emit_insn (gen_rtx_SET (dest,
gen_rtx_IOR (SImode, copy_rtx (temp),
GEN_INT (c & 0xffff))));
break;
case E_DImode:
if (!TARGET_POWERPC64)
{
rtx hi, lo;
hi = operand_subword_force (copy_rtx (dest), WORDS_BIG_ENDIAN == 0,
DImode);
lo = operand_subword_force (dest, WORDS_BIG_ENDIAN != 0,
DImode);
emit_move_insn (hi, GEN_INT (c >> 32));
c = ((c & 0xffffffff) ^ 0x80000000) - 0x80000000;
emit_move_insn (lo, GEN_INT (c));
}
else
rs6000_emit_set_long_const (dest, c);
break;
default:
gcc_unreachable ();
}
insn = get_last_insn ();
set = single_set (insn);
if (! CONSTANT_P (SET_SRC (set)))
set_unique_reg_note (insn, REG_EQUAL, GEN_INT (c));
return true;
}
/* Subroutine of rs6000_emit_set_const, handling PowerPC64 DImode.
Output insns to set DEST equal to the constant C as a series of
lis, ori and shl instructions. */
static void
rs6000_emit_set_long_const (rtx dest, HOST_WIDE_INT c)
{
rtx temp;
HOST_WIDE_INT ud1, ud2, ud3, ud4;
ud1 = c & 0xffff;
c = c >> 16;
ud2 = c & 0xffff;
c = c >> 16;
ud3 = c & 0xffff;
c = c >> 16;
ud4 = c & 0xffff;
if ((ud4 == 0xffff && ud3 == 0xffff && ud2 == 0xffff && (ud1 & 0x8000))
|| (ud4 == 0 && ud3 == 0 && ud2 == 0 && ! (ud1 & 0x8000)))
emit_move_insn (dest, GEN_INT ((ud1 ^ 0x8000) - 0x8000));
else if ((ud4 == 0xffff && ud3 == 0xffff && (ud2 & 0x8000))
|| (ud4 == 0 && ud3 == 0 && ! (ud2 & 0x8000)))
{
temp = !can_create_pseudo_p () ? dest : gen_reg_rtx (DImode);
emit_move_insn (ud1 != 0 ? copy_rtx (temp) : dest,
GEN_INT (((ud2 << 16) ^ 0x80000000) - 0x80000000));
if (ud1 != 0)
emit_move_insn (dest,
gen_rtx_IOR (DImode, copy_rtx (temp),
GEN_INT (ud1)));
}
else if (ud3 == 0 && ud4 == 0)
{
temp = !can_create_pseudo_p () ? dest : gen_reg_rtx (DImode);
gcc_assert (ud2 & 0x8000);
emit_move_insn (copy_rtx (temp),
GEN_INT (((ud2 << 16) ^ 0x80000000) - 0x80000000));
if (ud1 != 0)
emit_move_insn (copy_rtx (temp),
gen_rtx_IOR (DImode, copy_rtx (temp),
GEN_INT (ud1)));
emit_move_insn (dest,
gen_rtx_ZERO_EXTEND (DImode,
gen_lowpart (SImode,
copy_rtx (temp))));
}
else if ((ud4 == 0xffff && (ud3 & 0x8000))
|| (ud4 == 0 && ! (ud3 & 0x8000)))
{
temp = !can_create_pseudo_p () ? dest : gen_reg_rtx (DImode);
emit_move_insn (copy_rtx (temp),
GEN_INT (((ud3 << 16) ^ 0x80000000) - 0x80000000));
if (ud2 != 0)
emit_move_insn (copy_rtx (temp),
gen_rtx_IOR (DImode, copy_rtx (temp),
GEN_INT (ud2)));
emit_move_insn (ud1 != 0 ? copy_rtx (temp) : dest,
gen_rtx_ASHIFT (DImode, copy_rtx (temp),
GEN_INT (16)));
if (ud1 != 0)
emit_move_insn (dest,
gen_rtx_IOR (DImode, copy_rtx (temp),
GEN_INT (ud1)));
}
else
{
temp = !can_create_pseudo_p () ? dest : gen_reg_rtx (DImode);
emit_move_insn (copy_rtx (temp),
GEN_INT (((ud4 << 16) ^ 0x80000000) - 0x80000000));
if (ud3 != 0)
emit_move_insn (copy_rtx (temp),
gen_rtx_IOR (DImode, copy_rtx (temp),
GEN_INT (ud3)));
emit_move_insn (ud2 != 0 || ud1 != 0 ? copy_rtx (temp) : dest,
gen_rtx_ASHIFT (DImode, copy_rtx (temp),
GEN_INT (32)));
if (ud2 != 0)
emit_move_insn (ud1 != 0 ? copy_rtx (temp) : dest,
gen_rtx_IOR (DImode, copy_rtx (temp),
GEN_INT (ud2 << 16)));
if (ud1 != 0)
emit_move_insn (dest,
gen_rtx_IOR (DImode, copy_rtx (temp),
GEN_INT (ud1)));
}
}
/* Helper for the following. Get rid of [r+r] memory refs
in cases where it won't work (TImode, TFmode, TDmode, PTImode). */
static void
rs6000_eliminate_indexed_memrefs (rtx operands[2])
{
if (GET_CODE (operands[0]) == MEM
&& GET_CODE (XEXP (operands[0], 0)) != REG
&& ! legitimate_constant_pool_address_p (XEXP (operands[0], 0),
GET_MODE (operands[0]), false))
operands[0]
= replace_equiv_address (operands[0],
copy_addr_to_reg (XEXP (operands[0], 0)));
if (GET_CODE (operands[1]) == MEM
&& GET_CODE (XEXP (operands[1], 0)) != REG
&& ! legitimate_constant_pool_address_p (XEXP (operands[1], 0),
GET_MODE (operands[1]), false))
operands[1]
= replace_equiv_address (operands[1],
copy_addr_to_reg (XEXP (operands[1], 0)));
}
/* Generate a vector of constants to permute MODE for a little-endian
storage operation by swapping the two halves of a vector. */
static rtvec
rs6000_const_vec (machine_mode mode)
{
int i, subparts;
rtvec v;
switch (mode)
{
case E_V1TImode:
subparts = 1;
break;
case E_V2DFmode:
case E_V2DImode:
subparts = 2;
break;
case E_V4SFmode:
case E_V4SImode:
subparts = 4;
break;
case E_V8HImode:
subparts = 8;
break;
case E_V16QImode:
subparts = 16;
break;
default:
gcc_unreachable();
}
v = rtvec_alloc (subparts);
for (i = 0; i < subparts / 2; ++i)
RTVEC_ELT (v, i) = gen_rtx_CONST_INT (DImode, i + subparts / 2);
for (i = subparts / 2; i < subparts; ++i)
RTVEC_ELT (v, i) = gen_rtx_CONST_INT (DImode, i - subparts / 2);
return v;
}
/* Emit an lxvd2x, stxvd2x, or xxpermdi instruction for a VSX load or
store operation. */
void
rs6000_emit_le_vsx_permute (rtx dest, rtx source, machine_mode mode)
{
/* Scalar permutations are easier to express in integer modes rather than
floating-point modes, so cast them here. We use V1TImode instead
of TImode to ensure that the values don't go through GPRs. */
if (FLOAT128_VECTOR_P (mode))
{
dest = gen_lowpart (V1TImode, dest);
source = gen_lowpart (V1TImode, source);
mode = V1TImode;
}
/* Use ROTATE instead of VEC_SELECT if the mode contains only a single
scalar. */
if (mode == TImode || mode == V1TImode)
emit_insn (gen_rtx_SET (dest, gen_rtx_ROTATE (mode, source,
GEN_INT (64))));
else
{
rtx par = gen_rtx_PARALLEL (VOIDmode, rs6000_const_vec (mode));
emit_insn (gen_rtx_SET (dest, gen_rtx_VEC_SELECT (mode, source, par)));
}
}
/* Emit a little-endian load from vector memory location SOURCE to VSX
register DEST in mode MODE. The load is done with two permuting
insn's that represent an lxvd2x and xxpermdi. */
void
rs6000_emit_le_vsx_load (rtx dest, rtx source, machine_mode mode)
{
/* Use V2DImode to do swaps of types with 128-bit scalare parts (TImode,
V1TImode). */
if (mode == TImode || mode == V1TImode)
{
mode = V2DImode;
dest = gen_lowpart (V2DImode, dest);
source = adjust_address (source, V2DImode, 0);
}
rtx tmp = can_create_pseudo_p () ? gen_reg_rtx_and_attrs (dest) : dest;
rs6000_emit_le_vsx_permute (tmp, source, mode);
rs6000_emit_le_vsx_permute (dest, tmp, mode);
}
/* Emit a little-endian store to vector memory location DEST from VSX
register SOURCE in mode MODE. The store is done with two permuting
insn's that represent an xxpermdi and an stxvd2x. */
void
rs6000_emit_le_vsx_store (rtx dest, rtx source, machine_mode mode)
{
/* This should never be called during or after LRA, because it does
not re-permute the source register. It is intended only for use
during expand. */
gcc_assert (!lra_in_progress && !reload_completed);
/* Use V2DImode to do swaps of types with 128-bit scalar parts (TImode,
V1TImode). */
if (mode == TImode || mode == V1TImode)
{
mode = V2DImode;
dest = adjust_address (dest, V2DImode, 0);
source = gen_lowpart (V2DImode, source);
}
rtx tmp = can_create_pseudo_p () ? gen_reg_rtx_and_attrs (source) : source;
rs6000_emit_le_vsx_permute (tmp, source, mode);
rs6000_emit_le_vsx_permute (dest, tmp, mode);
}
/* Emit a sequence representing a little-endian VSX load or store,
moving data from SOURCE to DEST in mode MODE. This is done
separately from rs6000_emit_move to ensure it is called only
during expand. LE VSX loads and stores introduced later are
handled with a split. The expand-time RTL generation allows
us to optimize away redundant pairs of register-permutes. */
void
rs6000_emit_le_vsx_move (rtx dest, rtx source, machine_mode mode)
{
gcc_assert (!BYTES_BIG_ENDIAN
&& VECTOR_MEM_VSX_P (mode)
&& !TARGET_P9_VECTOR
&& !gpr_or_gpr_p (dest, source)
&& (MEM_P (source) ^ MEM_P (dest)));
if (MEM_P (source))
{
gcc_assert (REG_P (dest) || GET_CODE (dest) == SUBREG);
rs6000_emit_le_vsx_load (dest, source, mode);
}
else
{
if (!REG_P (source))
source = force_reg (mode, source);
rs6000_emit_le_vsx_store (dest, source, mode);
}
}
/* Return whether a SFmode or SImode move can be done without converting one
mode to another. This arrises when we have:
(SUBREG:SF (REG:SI ...))
(SUBREG:SI (REG:SF ...))
and one of the values is in a floating point/vector register, where SFmode
scalars are stored in DFmode format. */
bool
valid_sf_si_move (rtx dest, rtx src, machine_mode mode)
{
if (TARGET_ALLOW_SF_SUBREG)
return true;
if (mode != SFmode && GET_MODE_CLASS (mode) != MODE_INT)
return true;
if (!SUBREG_P (src) || !sf_subreg_operand (src, mode))
return true;
/*. Allow (set (SUBREG:SI (REG:SF)) (SUBREG:SI (REG:SF))). */
if (SUBREG_P (dest))
{
rtx dest_subreg = SUBREG_REG (dest);
rtx src_subreg = SUBREG_REG (src);
return GET_MODE (dest_subreg) == GET_MODE (src_subreg);
}
return false;
}
/* Helper function to change moves with:
(SUBREG:SF (REG:SI)) and
(SUBREG:SI (REG:SF))
into separate UNSPEC insns. In the PowerPC architecture, scalar SFmode
values are stored as DFmode values in the VSX registers. We need to convert
the bits before we can use a direct move or operate on the bits in the
vector register as an integer type.
Skip things like (set (SUBREG:SI (...) (SUBREG:SI (...)). */
static bool
rs6000_emit_move_si_sf_subreg (rtx dest, rtx source, machine_mode mode)
{
if (TARGET_DIRECT_MOVE_64BIT && !lra_in_progress && !reload_completed
&& (!SUBREG_P (dest) || !sf_subreg_operand (dest, mode))
&& SUBREG_P (source) && sf_subreg_operand (source, mode))
{
rtx inner_source = SUBREG_REG (source);
machine_mode inner_mode = GET_MODE (inner_source);
if (mode == SImode && inner_mode == SFmode)
{
emit_insn (gen_movsi_from_sf (dest, inner_source));
return true;
}
if (mode == SFmode && inner_mode == SImode)
{
emit_insn (gen_movsf_from_si (dest, inner_source));
return true;
}
}
return false;
}
/* Emit a move from SOURCE to DEST in mode MODE. */
void
rs6000_emit_move (rtx dest, rtx source, machine_mode mode)
{
rtx operands[2];
operands[0] = dest;
operands[1] = source;
if (TARGET_DEBUG_ADDR)
{
fprintf (stderr,
"\nrs6000_emit_move: mode = %s, lra_in_progress = %d, "
"reload_completed = %d, can_create_pseudos = %d.\ndest:\n",
GET_MODE_NAME (mode),
lra_in_progress,
reload_completed,
can_create_pseudo_p ());
debug_rtx (dest);
fprintf (stderr, "source:\n");
debug_rtx (source);
}
/* Sanity checks. Check that we get CONST_DOUBLE only when we should. */
if (CONST_WIDE_INT_P (operands[1])
&& GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT)
{
/* This should be fixed with the introduction of CONST_WIDE_INT. */
gcc_unreachable ();
}
/* See if we need to special case SImode/SFmode SUBREG moves. */
if ((mode == SImode || mode == SFmode) && SUBREG_P (source)
&& rs6000_emit_move_si_sf_subreg (dest, source, mode))
return;
/* Check if GCC is setting up a block move that will end up using FP
registers as temporaries. We must make sure this is acceptable. */
if (GET_CODE (operands[0]) == MEM
&& GET_CODE (operands[1]) == MEM
&& mode == DImode
&& (rs6000_slow_unaligned_access (DImode, MEM_ALIGN (operands[0]))
|| rs6000_slow_unaligned_access (DImode, MEM_ALIGN (operands[1])))
&& ! (rs6000_slow_unaligned_access (SImode,
(MEM_ALIGN (operands[0]) > 32
? 32 : MEM_ALIGN (operands[0])))
|| rs6000_slow_unaligned_access (SImode,
(MEM_ALIGN (operands[1]) > 32
? 32 : MEM_ALIGN (operands[1]))))
&& ! MEM_VOLATILE_P (operands [0])
&& ! MEM_VOLATILE_P (operands [1]))
{
emit_move_insn (adjust_address (operands[0], SImode, 0),
adjust_address (operands[1], SImode, 0));
emit_move_insn (adjust_address (copy_rtx (operands[0]), SImode, 4),
adjust_address (copy_rtx (operands[1]), SImode, 4));
return;
}
if (can_create_pseudo_p () && GET_CODE (operands[0]) == MEM
&& !gpc_reg_operand (operands[1], mode))
operands[1] = force_reg (mode, operands[1]);
/* Recognize the case where operand[1] is a reference to thread-local
data and load its address to a register. */
if (tls_referenced_p (operands[1]))
{
enum tls_model model;
rtx tmp = operands[1];
rtx addend = NULL;
if (GET_CODE (tmp) == CONST && GET_CODE (XEXP (tmp, 0)) == PLUS)
{
addend = XEXP (XEXP (tmp, 0), 1);
tmp = XEXP (XEXP (tmp, 0), 0);
}
gcc_assert (GET_CODE (tmp) == SYMBOL_REF);
model = SYMBOL_REF_TLS_MODEL (tmp);
gcc_assert (model != 0);
tmp = rs6000_legitimize_tls_address (tmp, model);
if (addend)
{
tmp = gen_rtx_PLUS (mode, tmp, addend);
tmp = force_operand (tmp, operands[0]);
}
operands[1] = tmp;
}
/* 128-bit constant floating-point values on Darwin should really be loaded
as two parts. However, this premature splitting is a problem when DFmode
values can go into Altivec registers. */
if (FLOAT128_IBM_P (mode) && !reg_addr[DFmode].scalar_in_vmx_p
&& GET_CODE (operands[1]) == CONST_DOUBLE)
{
rs6000_emit_move (simplify_gen_subreg (DFmode, operands[0], mode, 0),
simplify_gen_subreg (DFmode, operands[1], mode, 0),
DFmode);
rs6000_emit_move (simplify_gen_subreg (DFmode, operands[0], mode,
GET_MODE_SIZE (DFmode)),
simplify_gen_subreg (DFmode, operands[1], mode,
GET_MODE_SIZE (DFmode)),
DFmode);
return;
}
/* Transform (p0:DD, (SUBREG:DD p1:SD)) to ((SUBREG:SD p0:DD),
p1:SD) if p1 is not of floating point class and p0 is spilled as
we can have no analogous movsd_store for this. */
if (lra_in_progress && mode == DDmode
&& REG_P (operands[0]) && REGNO (operands[0]) >= FIRST_PSEUDO_REGISTER
&& reg_preferred_class (REGNO (operands[0])) == NO_REGS
&& GET_CODE (operands[1]) == SUBREG && REG_P (SUBREG_REG (operands[1]))
&& GET_MODE (SUBREG_REG (operands[1])) == SDmode)
{
enum reg_class cl;
int regno = REGNO (SUBREG_REG (operands[1]));
if (regno >= FIRST_PSEUDO_REGISTER)
{
cl = reg_preferred_class (regno);
regno = cl == NO_REGS ? -1 : ira_class_hard_regs[cl][1];
}
if (regno >= 0 && ! FP_REGNO_P (regno))
{
mode = SDmode;
operands[0] = gen_lowpart_SUBREG (SDmode, operands[0]);
operands[1] = SUBREG_REG (operands[1]);
}
}
if (lra_in_progress
&& mode == SDmode
&& REG_P (operands[0]) && REGNO (operands[0]) >= FIRST_PSEUDO_REGISTER
&& reg_preferred_class (REGNO (operands[0])) == NO_REGS
&& (REG_P (operands[1])
|| (GET_CODE (operands[1]) == SUBREG
&& REG_P (SUBREG_REG (operands[1])))))
{
int regno = REGNO (GET_CODE (operands[1]) == SUBREG
? SUBREG_REG (operands[1]) : operands[1]);
enum reg_class cl;
if (regno >= FIRST_PSEUDO_REGISTER)
{
cl = reg_preferred_class (regno);
gcc_assert (cl != NO_REGS);
regno = ira_class_hard_regs[cl][0];
}
if (FP_REGNO_P (regno))
{
if (GET_MODE (operands[0]) != DDmode)
operands[0] = gen_rtx_SUBREG (DDmode, operands[0], 0);
emit_insn (gen_movsd_store (operands[0], operands[1]));
}
else if (INT_REGNO_P (regno))
emit_insn (gen_movsd_hardfloat (operands[0], operands[1]));
else
gcc_unreachable();
return;
}
/* Transform ((SUBREG:DD p0:SD), p1:DD) to (p0:SD, (SUBREG:SD
p:DD)) if p0 is not of floating point class and p1 is spilled as
we can have no analogous movsd_load for this. */
if (lra_in_progress && mode == DDmode
&& GET_CODE (operands[0]) == SUBREG && REG_P (SUBREG_REG (operands[0]))
&& GET_MODE (SUBREG_REG (operands[0])) == SDmode
&& REG_P (operands[1]) && REGNO (operands[1]) >= FIRST_PSEUDO_REGISTER
&& reg_preferred_class (REGNO (operands[1])) == NO_REGS)
{
enum reg_class cl;
int regno = REGNO (SUBREG_REG (operands[0]));
if (regno >= FIRST_PSEUDO_REGISTER)
{
cl = reg_preferred_class (regno);
regno = cl == NO_REGS ? -1 : ira_class_hard_regs[cl][0];
}
if (regno >= 0 && ! FP_REGNO_P (regno))
{
mode = SDmode;
operands[0] = SUBREG_REG (operands[0]);
operands[1] = gen_lowpart_SUBREG (SDmode, operands[1]);
}
}
if (lra_in_progress
&& mode == SDmode
&& (REG_P (operands[0])
|| (GET_CODE (operands[0]) == SUBREG
&& REG_P (SUBREG_REG (operands[0]))))
&& REG_P (operands[1]) && REGNO (operands[1]) >= FIRST_PSEUDO_REGISTER
&& reg_preferred_class (REGNO (operands[1])) == NO_REGS)
{
int regno = REGNO (GET_CODE (operands[0]) == SUBREG
? SUBREG_REG (operands[0]) : operands[0]);
enum reg_class cl;
if (regno >= FIRST_PSEUDO_REGISTER)
{
cl = reg_preferred_class (regno);
gcc_assert (cl != NO_REGS);
regno = ira_class_hard_regs[cl][0];
}
if (FP_REGNO_P (regno))
{
if (GET_MODE (operands[1]) != DDmode)
operands[1] = gen_rtx_SUBREG (DDmode, operands[1], 0);
emit_insn (gen_movsd_load (operands[0], operands[1]));
}
else if (INT_REGNO_P (regno))
emit_insn (gen_movsd_hardfloat (operands[0], operands[1]));
else
gcc_unreachable();
return;
}
/* FIXME: In the long term, this switch statement should go away
and be replaced by a sequence of tests based on things like
mode == Pmode. */
switch (mode)
{
case E_HImode:
case E_QImode:
if (CONSTANT_P (operands[1])
&& GET_CODE (operands[1]) != CONST_INT)
operands[1] = force_const_mem (mode, operands[1]);
break;
case E_TFmode:
case E_TDmode:
case E_IFmode:
case E_KFmode:
if (FLOAT128_2REG_P (mode))
rs6000_eliminate_indexed_memrefs (operands);
/* fall through */
case E_DFmode:
case E_DDmode:
case E_SFmode:
case E_SDmode:
if (CONSTANT_P (operands[1])
&& ! easy_fp_constant (operands[1], mode))
operands[1] = force_const_mem (mode, operands[1]);
break;
case E_V16QImode:
case E_V8HImode:
case E_V4SFmode:
case E_V4SImode:
case E_V2SFmode:
case E_V2SImode:
case E_V2DFmode:
case E_V2DImode:
case E_V1TImode:
if (CONSTANT_P (operands[1])
&& !easy_vector_constant (operands[1], mode))
operands[1] = force_const_mem (mode, operands[1]);
break;
case E_SImode:
case E_DImode:
/* Use default pattern for address of ELF small data */
if (TARGET_ELF
&& mode == Pmode
&& DEFAULT_ABI == ABI_V4
&& (GET_CODE (operands[1]) == SYMBOL_REF
|| GET_CODE (operands[1]) == CONST)
&& small_data_operand (operands[1], mode))
{
emit_insn (gen_rtx_SET (operands[0], operands[1]));
return;
}
if (DEFAULT_ABI == ABI_V4
&& mode == Pmode && mode == SImode
&& flag_pic == 1 && got_operand (operands[1], mode))
{
emit_insn (gen_movsi_got (operands[0], operands[1]));
return;
}
if ((TARGET_ELF || DEFAULT_ABI == ABI_DARWIN)
&& TARGET_NO_TOC
&& ! flag_pic
&& mode == Pmode
&& CONSTANT_P (operands[1])
&& GET_CODE (operands[1]) != HIGH
&& GET_CODE (operands[1]) != CONST_INT)
{
rtx target = (!can_create_pseudo_p ()
? operands[0]
: gen_reg_rtx (mode));
/* If this is a function address on -mcall-aixdesc,
convert it to the address of the descriptor. */
if (DEFAULT_ABI == ABI_AIX
&& GET_CODE (operands[1]) == SYMBOL_REF
&& XSTR (operands[1], 0)[0] == '.')
{
const char *name = XSTR (operands[1], 0);
rtx new_ref;
while (*name == '.')
name++;
new_ref = gen_rtx_SYMBOL_REF (Pmode, name);
CONSTANT_POOL_ADDRESS_P (new_ref)
= CONSTANT_POOL_ADDRESS_P (operands[1]);
SYMBOL_REF_FLAGS (new_ref) = SYMBOL_REF_FLAGS (operands[1]);
SYMBOL_REF_USED (new_ref) = SYMBOL_REF_USED (operands[1]);
SYMBOL_REF_DATA (new_ref) = SYMBOL_REF_DATA (operands[1]);
operands[1] = new_ref;
}
if (DEFAULT_ABI == ABI_DARWIN)
{
#if TARGET_MACHO
if (MACHO_DYNAMIC_NO_PIC_P)
{
/* Take care of any required data indirection. */
operands[1] = rs6000_machopic_legitimize_pic_address (
operands[1], mode, operands[0]);
if (operands[0] != operands[1])
emit_insn (gen_rtx_SET (operands[0], operands[1]));
return;
}
#endif
emit_insn (gen_macho_high (target, operands[1]));
emit_insn (gen_macho_low (operands[0], target, operands[1]));
return;
}
emit_insn (gen_elf_high (target, operands[1]));
emit_insn (gen_elf_low (operands[0], target, operands[1]));
return;
}
/* If this is a SYMBOL_REF that refers to a constant pool entry,
and we have put it in the TOC, we just need to make a TOC-relative
reference to it. */
if (TARGET_TOC
&& GET_CODE (operands[1]) == SYMBOL_REF
&& use_toc_relative_ref (operands[1], mode))
operands[1] = create_TOC_reference (operands[1], operands[0]);
else if (mode == Pmode
&& CONSTANT_P (operands[1])
&& GET_CODE (operands[1]) != HIGH
&& ((GET_CODE (operands[1]) != CONST_INT
&& ! easy_fp_constant (operands[1], mode))
|| (GET_CODE (operands[1]) == CONST_INT
&& (num_insns_constant (operands[1], mode)
> (TARGET_CMODEL != CMODEL_SMALL ? 3 : 2)))
|| (GET_CODE (operands[0]) == REG
&& FP_REGNO_P (REGNO (operands[0]))))
&& !toc_relative_expr_p (operands[1], false, NULL, NULL)
&& (TARGET_CMODEL == CMODEL_SMALL
|| can_create_pseudo_p ()
|| (REG_P (operands[0])
&& INT_REG_OK_FOR_BASE_P (operands[0], true))))
{
#if TARGET_MACHO
/* Darwin uses a special PIC legitimizer. */
if (DEFAULT_ABI == ABI_DARWIN && MACHOPIC_INDIRECT)
{
operands[1] =
rs6000_machopic_legitimize_pic_address (operands[1], mode,
operands[0]);
if (operands[0] != operands[1])
emit_insn (gen_rtx_SET (operands[0], operands[1]));
return;
}
#endif
/* If we are to limit the number of things we put in the TOC and
this is a symbol plus a constant we can add in one insn,
just put the symbol in the TOC and add the constant. */
if (GET_CODE (operands[1]) == CONST
&& TARGET_NO_SUM_IN_TOC
&& GET_CODE (XEXP (operands[1], 0)) == PLUS
&& add_operand (XEXP (XEXP (operands[1], 0), 1), mode)
&& (GET_CODE (XEXP (XEXP (operands[1], 0), 0)) == LABEL_REF
|| GET_CODE (XEXP (XEXP (operands[1], 0), 0)) == SYMBOL_REF)
&& ! side_effects_p (operands[0]))
{
rtx sym =
force_const_mem (mode, XEXP (XEXP (operands[1], 0), 0));
rtx other = XEXP (XEXP (operands[1], 0), 1);
sym = force_reg (mode, sym);
emit_insn (gen_add3_insn (operands[0], sym, other));
return;
}
operands[1] = force_const_mem (mode, operands[1]);
if (TARGET_TOC
&& GET_CODE (XEXP (operands[1], 0)) == SYMBOL_REF
&& use_toc_relative_ref (XEXP (operands[1], 0), mode))
{
rtx tocref = create_TOC_reference (XEXP (operands[1], 0),
operands[0]);
operands[1] = gen_const_mem (mode, tocref);
set_mem_alias_set (operands[1], get_TOC_alias_set ());
}
}
break;
case E_TImode:
if (!VECTOR_MEM_VSX_P (TImode))
rs6000_eliminate_indexed_memrefs (operands);
break;
case E_PTImode:
rs6000_eliminate_indexed_memrefs (operands);
break;
default:
fatal_insn ("bad move", gen_rtx_SET (dest, source));
}
/* Above, we may have called force_const_mem which may have returned
an invalid address. If we can, fix this up; otherwise, reload will
have to deal with it. */
if (GET_CODE (operands[1]) == MEM)
operands[1] = validize_mem (operands[1]);
emit_insn (gen_rtx_SET (operands[0], operands[1]));
}
/* Nonzero if we can use a floating-point register to pass this arg. */
#define USE_FP_FOR_ARG_P(CUM,MODE) \
(SCALAR_FLOAT_MODE_NOT_VECTOR_P (MODE) \
&& (CUM)->fregno <= FP_ARG_MAX_REG \
&& TARGET_HARD_FLOAT)
/* Nonzero if we can use an AltiVec register to pass this arg. */
#define USE_ALTIVEC_FOR_ARG_P(CUM,MODE,NAMED) \
(ALTIVEC_OR_VSX_VECTOR_MODE (MODE) \
&& (CUM)->vregno <= ALTIVEC_ARG_MAX_REG \
&& TARGET_ALTIVEC_ABI \
&& (NAMED))
/* Walk down the type tree of TYPE counting consecutive base elements.
If *MODEP is VOIDmode, then set it to the first valid floating point
or vector type. If a non-floating point or vector type is found, or
if a floating point or vector type that doesn't match a non-VOIDmode
*MODEP is found, then return -1, otherwise return the count in the
sub-tree. */
static int
rs6000_aggregate_candidate (const_tree type, machine_mode *modep)
{
machine_mode mode;
HOST_WIDE_INT size;
switch (TREE_CODE (type))
{
case REAL_TYPE:
mode = TYPE_MODE (type);
if (!SCALAR_FLOAT_MODE_P (mode))
return -1;
if (*modep == VOIDmode)
*modep = mode;
if (*modep == mode)
return 1;
break;
case COMPLEX_TYPE:
mode = TYPE_MODE (TREE_TYPE (type));
if (!SCALAR_FLOAT_MODE_P (mode))
return -1;
if (*modep == VOIDmode)
*modep = mode;
if (*modep == mode)
return 2;
break;
case VECTOR_TYPE:
if (!TARGET_ALTIVEC_ABI || !TARGET_ALTIVEC)
return -1;
/* Use V4SImode as representative of all 128-bit vector types. */
size = int_size_in_bytes (type);
switch (size)
{
case 16:
mode = V4SImode;
break;
default:
return -1;
}
if (*modep == VOIDmode)
*modep = mode;
/* Vector modes are considered to be opaque: two vectors are
equivalent for the purposes of being homogeneous aggregates
if they are the same size. */
if (*modep == mode)
return 1;
break;
case ARRAY_TYPE:
{
int count;
tree index = TYPE_DOMAIN (type);
/* Can't handle incomplete types nor sizes that are not
fixed. */
if (!COMPLETE_TYPE_P (type)
|| TREE_CODE (TYPE_SIZE (type)) != INTEGER_CST)
return -1;
count = rs6000_aggregate_candidate (TREE_TYPE (type), modep);
if (count == -1
|| !index
|| !TYPE_MAX_VALUE (index)
|| !tree_fits_uhwi_p (TYPE_MAX_VALUE (index))
|| !TYPE_MIN_VALUE (index)
|| !tree_fits_uhwi_p (TYPE_MIN_VALUE (index))
|| count < 0)
return -1;
count *= (1 + tree_to_uhwi (TYPE_MAX_VALUE (index))
- tree_to_uhwi (TYPE_MIN_VALUE (index)));
/* There must be no padding. */
if (wi::to_wide (TYPE_SIZE (type))
!= count * GET_MODE_BITSIZE (*modep))
return -1;
return count;
}
case RECORD_TYPE:
{
int count = 0;
int sub_count;
tree field;
/* Can't handle incomplete types nor sizes that are not
fixed. */
if (!COMPLETE_TYPE_P (type)
|| TREE_CODE (TYPE_SIZE (type)) != INTEGER_CST)
return -1;
for (field = TYPE_FIELDS (type); field; field = TREE_CHAIN (field))
{
if (TREE_CODE (field) != FIELD_DECL)
continue;
sub_count = rs6000_aggregate_candidate (TREE_TYPE (field), modep);
if (sub_count < 0)
return -1;
count += sub_count;
}
/* There must be no padding. */
if (wi::to_wide (TYPE_SIZE (type))
!= count * GET_MODE_BITSIZE (*modep))
return -1;
return count;
}
case UNION_TYPE:
case QUAL_UNION_TYPE:
{
/* These aren't very interesting except in a degenerate case. */
int count = 0;
int sub_count;
tree field;
/* Can't handle incomplete types nor sizes that are not
fixed. */
if (!COMPLETE_TYPE_P (type)
|| TREE_CODE (TYPE_SIZE (type)) != INTEGER_CST)
return -1;
for (field = TYPE_FIELDS (type); field; field = TREE_CHAIN (field))
{
if (TREE_CODE (field) != FIELD_DECL)
continue;
sub_count = rs6000_aggregate_candidate (TREE_TYPE (field), modep);
if (sub_count < 0)
return -1;
count = count > sub_count ? count : sub_count;
}
/* There must be no padding. */
if (wi::to_wide (TYPE_SIZE (type))
!= count * GET_MODE_BITSIZE (*modep))
return -1;
return count;
}
default:
break;
}
return -1;
}
/* If an argument, whose type is described by TYPE and MODE, is a homogeneous
float or vector aggregate that shall be passed in FP/vector registers
according to the ELFv2 ABI, return the homogeneous element mode in
*ELT_MODE and the number of elements in *N_ELTS, and return TRUE.
Otherwise, set *ELT_MODE to MODE and *N_ELTS to 1, and return FALSE. */
static bool
rs6000_discover_homogeneous_aggregate (machine_mode mode, const_tree type,
machine_mode *elt_mode,
int *n_elts)
{
/* Note that we do not accept complex types at the top level as
homogeneous aggregates; these types are handled via the
targetm.calls.split_complex_arg mechanism. Complex types
can be elements of homogeneous aggregates, however. */
if (DEFAULT_ABI == ABI_ELFv2 && type && AGGREGATE_TYPE_P (type))
{
machine_mode field_mode = VOIDmode;
int field_count = rs6000_aggregate_candidate (type, &field_mode);
if (field_count > 0)
{
int n_regs = (SCALAR_FLOAT_MODE_P (field_mode) ?
(GET_MODE_SIZE (field_mode) + 7) >> 3 : 1);
/* The ELFv2 ABI allows homogeneous aggregates to occupy
up to AGGR_ARG_NUM_REG registers. */
if (field_count * n_regs <= AGGR_ARG_NUM_REG)
{
if (elt_mode)
*elt_mode = field_mode;
if (n_elts)
*n_elts = field_count;
return true;
}
}
}
if (elt_mode)
*elt_mode = mode;
if (n_elts)
*n_elts = 1;
return false;
}
/* Return a nonzero value to say to return the function value in
memory, just as large structures are always returned. TYPE will be
the data type of the value, and FNTYPE will be the type of the
function doing the returning, or @code{NULL} for libcalls.
The AIX ABI for the RS/6000 specifies that all structures are
returned in memory. The Darwin ABI does the same.
For the Darwin 64 Bit ABI, a function result can be returned in
registers or in memory, depending on the size of the return data
type. If it is returned in registers, the value occupies the same
registers as it would if it were the first and only function
argument. Otherwise, the function places its result in memory at
the location pointed to by GPR3.
The SVR4 ABI specifies that structures <= 8 bytes are returned in r3/r4,
but a draft put them in memory, and GCC used to implement the draft
instead of the final standard. Therefore, aix_struct_return
controls this instead of DEFAULT_ABI; V.4 targets needing backward
compatibility can change DRAFT_V4_STRUCT_RET to override the
default, and -m switches get the final word. See
rs6000_option_override_internal for more details.
The PPC32 SVR4 ABI uses IEEE double extended for long double, if 128-bit
long double support is enabled. These values are returned in memory.
int_size_in_bytes returns -1 for variable size objects, which go in
memory always. The cast to unsigned makes -1 > 8. */
static bool
rs6000_return_in_memory (const_tree type, const_tree fntype ATTRIBUTE_UNUSED)
{
/* For the Darwin64 ABI, test if we can fit the return value in regs. */
if (TARGET_MACHO
&& rs6000_darwin64_abi
&& TREE_CODE (type) == RECORD_TYPE
&& int_size_in_bytes (type) > 0)
{
CUMULATIVE_ARGS valcum;
rtx valret;
valcum.words = 0;
valcum.fregno = FP_ARG_MIN_REG;
valcum.vregno = ALTIVEC_ARG_MIN_REG;
/* Do a trial code generation as if this were going to be passed
as an argument; if any part goes in memory, we return NULL. */
valret = rs6000_darwin64_record_arg (&valcum, type, true, true);
if (valret)
return false;
/* Otherwise fall through to more conventional ABI rules. */
}
/* The ELFv2 ABI returns homogeneous VFP aggregates in registers */
if (rs6000_discover_homogeneous_aggregate (TYPE_MODE (type), type,
NULL, NULL))
return false;
/* The ELFv2 ABI returns aggregates up to 16B in registers */
if (DEFAULT_ABI == ABI_ELFv2 && AGGREGATE_TYPE_P (type)
&& (unsigned HOST_WIDE_INT) int_size_in_bytes (type) <= 16)
return false;
if (AGGREGATE_TYPE_P (type)
&& (aix_struct_return
|| (unsigned HOST_WIDE_INT) int_size_in_bytes (type) > 8))
return true;
/* Allow -maltivec -mabi=no-altivec without warning. Altivec vector
modes only exist for GCC vector types if -maltivec. */
if (TARGET_32BIT && !TARGET_ALTIVEC_ABI
&& ALTIVEC_VECTOR_MODE (TYPE_MODE (type)))
return false;
/* Return synthetic vectors in memory. */
if (TREE_CODE (type) == VECTOR_TYPE
&& int_size_in_bytes (type) > (TARGET_ALTIVEC_ABI ? 16 : 8))
{
static bool warned_for_return_big_vectors = false;
if (!warned_for_return_big_vectors)
{
warning (OPT_Wpsabi, "GCC vector returned by reference: "
"non-standard ABI extension with no compatibility "
"guarantee");
warned_for_return_big_vectors = true;
}
return true;
}
if (DEFAULT_ABI == ABI_V4 && TARGET_IEEEQUAD
&& FLOAT128_IEEE_P (TYPE_MODE (type)))
return true;
return false;
}
/* Specify whether values returned in registers should be at the most
significant end of a register. We want aggregates returned by
value to match the way aggregates are passed to functions. */
static bool
rs6000_return_in_msb (const_tree valtype)
{
return (DEFAULT_ABI == ABI_ELFv2
&& BYTES_BIG_ENDIAN
&& AGGREGATE_TYPE_P (valtype)
&& (rs6000_function_arg_padding (TYPE_MODE (valtype), valtype)
== PAD_UPWARD));
}
#ifdef HAVE_AS_GNU_ATTRIBUTE
/* Return TRUE if a call to function FNDECL may be one that
potentially affects the function calling ABI of the object file. */
static bool
call_ABI_of_interest (tree fndecl)
{
if (rs6000_gnu_attr && symtab->state == EXPANSION)
{
struct cgraph_node *c_node;
/* Libcalls are always interesting. */
if (fndecl == NULL_TREE)
return true;
/* Any call to an external function is interesting. */
if (DECL_EXTERNAL (fndecl))
return true;
/* Interesting functions that we are emitting in this object file. */
c_node = cgraph_node::get (fndecl);
c_node = c_node->ultimate_alias_target ();
return !c_node->only_called_directly_p ();
}
return false;
}
#endif
/* Initialize a variable CUM of type CUMULATIVE_ARGS
for a call to a function whose data type is FNTYPE.
For a library call, FNTYPE is 0 and RETURN_MODE the return value mode.
For incoming args we set the number of arguments in the prototype large
so we never return a PARALLEL. */
void
init_cumulative_args (CUMULATIVE_ARGS *cum, tree fntype,
rtx libname ATTRIBUTE_UNUSED, int incoming,
int libcall, int n_named_args,
tree fndecl ATTRIBUTE_UNUSED,
machine_mode return_mode ATTRIBUTE_UNUSED)
{
static CUMULATIVE_ARGS zero_cumulative;
*cum = zero_cumulative;
cum->words = 0;
cum->fregno = FP_ARG_MIN_REG;
cum->vregno = ALTIVEC_ARG_MIN_REG;
cum->prototype = (fntype && prototype_p (fntype));
cum->call_cookie = ((DEFAULT_ABI == ABI_V4 && libcall)
? CALL_LIBCALL : CALL_NORMAL);
cum->sysv_gregno = GP_ARG_MIN_REG;
cum->stdarg = stdarg_p (fntype);
cum->libcall = libcall;
cum->nargs_prototype = 0;
if (incoming || cum->prototype)
cum->nargs_prototype = n_named_args;
/* Check for a longcall attribute. */
if ((!fntype && rs6000_default_long_calls)
|| (fntype
&& lookup_attribute ("longcall", TYPE_ATTRIBUTES (fntype))
&& !lookup_attribute ("shortcall", TYPE_ATTRIBUTES (fntype))))
cum->call_cookie |= CALL_LONG;
if (TARGET_DEBUG_ARG)
{
fprintf (stderr, "\ninit_cumulative_args:");
if (fntype)
{
tree ret_type = TREE_TYPE (fntype);
fprintf (stderr, " ret code = %s,",
get_tree_code_name (TREE_CODE (ret_type)));
}
if (cum->call_cookie & CALL_LONG)
fprintf (stderr, " longcall,");
fprintf (stderr, " proto = %d, nargs = %d\n",
cum->prototype, cum->nargs_prototype);
}
#ifdef HAVE_AS_GNU_ATTRIBUTE
if (TARGET_ELF && (TARGET_64BIT || DEFAULT_ABI == ABI_V4))
{
cum->escapes = call_ABI_of_interest (fndecl);
if (cum->escapes)
{
tree return_type;
if (fntype)
{
return_type = TREE_TYPE (fntype);
return_mode = TYPE_MODE (return_type);
}
else
return_type = lang_hooks.types.type_for_mode (return_mode, 0);
if (return_type != NULL)
{
if (TREE_CODE (return_type) == RECORD_TYPE
&& TYPE_TRANSPARENT_AGGR (return_type))
{
return_type = TREE_TYPE (first_field (return_type));
return_mode = TYPE_MODE (return_type);
}
if (AGGREGATE_TYPE_P (return_type)
&& ((unsigned HOST_WIDE_INT) int_size_in_bytes (return_type)
<= 8))
rs6000_returns_struct = true;
}
if (SCALAR_FLOAT_MODE_P (return_mode))
{
rs6000_passes_float = true;
if ((HAVE_LD_PPC_GNU_ATTR_LONG_DOUBLE || TARGET_64BIT)
&& (FLOAT128_IBM_P (return_mode)
|| FLOAT128_IEEE_P (return_mode)
|| (return_type != NULL
&& (TYPE_MAIN_VARIANT (return_type)
== long_double_type_node))))
rs6000_passes_long_double = true;
}
if (ALTIVEC_OR_VSX_VECTOR_MODE (return_mode)
|| PAIRED_VECTOR_MODE (return_mode))
rs6000_passes_vector = true;
}
}
#endif
if (fntype
&& !TARGET_ALTIVEC
&& TARGET_ALTIVEC_ABI
&& ALTIVEC_VECTOR_MODE (TYPE_MODE (TREE_TYPE (fntype))))
{
error ("cannot return value in vector register because"
" altivec instructions are disabled, use %qs"
" to enable them", "-maltivec");
}
}
/* The mode the ABI uses for a word. This is not the same as word_mode
for -m32 -mpowerpc64. This is used to implement various target hooks. */
static scalar_int_mode
rs6000_abi_word_mode (void)
{
return TARGET_32BIT ? SImode : DImode;
}
/* Implement the TARGET_OFFLOAD_OPTIONS hook. */
static char *
rs6000_offload_options (void)
{
if (TARGET_64BIT)
return xstrdup ("-foffload-abi=lp64");
else
return xstrdup ("-foffload-abi=ilp32");
}
/* On rs6000, function arguments are promoted, as are function return
values. */
static machine_mode
rs6000_promote_function_mode (const_tree type ATTRIBUTE_UNUSED,
machine_mode mode,
int *punsignedp ATTRIBUTE_UNUSED,
const_tree, int)
{
PROMOTE_MODE (mode, *punsignedp, type);
return mode;
}
/* Return true if TYPE must be passed on the stack and not in registers. */
static bool
rs6000_must_pass_in_stack (machine_mode mode, const_tree type)
{
if (DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2 || TARGET_64BIT)
return must_pass_in_stack_var_size (mode, type);
else
return must_pass_in_stack_var_size_or_pad (mode, type);
}
static inline bool
is_complex_IBM_long_double (machine_mode mode)
{
return mode == ICmode || (!TARGET_IEEEQUAD && mode == TCmode);
}
/* Whether ABI_V4 passes MODE args to a function in floating point
registers. */
static bool
abi_v4_pass_in_fpr (machine_mode mode)
{
if (!TARGET_HARD_FLOAT)
return false;
if (TARGET_SINGLE_FLOAT && mode == SFmode)
return true;
if (TARGET_DOUBLE_FLOAT && mode == DFmode)
return true;
/* ABI_V4 passes complex IBM long double in 8 gprs.
Stupid, but we can't change the ABI now. */
if (is_complex_IBM_long_double (mode))
return false;
if (FLOAT128_2REG_P (mode))
return true;
if (DECIMAL_FLOAT_MODE_P (mode))
return true;
return false;
}
/* Implement TARGET_FUNCTION_ARG_PADDING.
For the AIX ABI structs are always stored left shifted in their
argument slot. */
static pad_direction
rs6000_function_arg_padding (machine_mode mode, const_tree type)
{
#ifndef AGGREGATE_PADDING_FIXED
#define AGGREGATE_PADDING_FIXED 0
#endif
#ifndef AGGREGATES_PAD_UPWARD_ALWAYS
#define AGGREGATES_PAD_UPWARD_ALWAYS 0
#endif
if (!AGGREGATE_PADDING_FIXED)
{
/* GCC used to pass structures of the same size as integer types as
if they were in fact integers, ignoring TARGET_FUNCTION_ARG_PADDING.
i.e. Structures of size 1 or 2 (or 4 when TARGET_64BIT) were
passed padded downward, except that -mstrict-align further
muddied the water in that multi-component structures of 2 and 4
bytes in size were passed padded upward.
The following arranges for best compatibility with previous
versions of gcc, but removes the -mstrict-align dependency. */
if (BYTES_BIG_ENDIAN)
{
HOST_WIDE_INT size = 0;
if (mode == BLKmode)
{
if (type && TREE_CODE (TYPE_SIZE (type)) == INTEGER_CST)
size = int_size_in_bytes (type);
}
else
size = GET_MODE_SIZE (mode);
if (size == 1 || size == 2 || size == 4)
return PAD_DOWNWARD;
}
return PAD_UPWARD;
}
if (AGGREGATES_PAD_UPWARD_ALWAYS)
{
if (type != 0 && AGGREGATE_TYPE_P (type))
return PAD_UPWARD;
}
/* Fall back to the default. */
return default_function_arg_padding (mode, type);
}
/* If defined, a C expression that gives the alignment boundary, in bits,
of an argument with the specified mode and type. If it is not defined,
PARM_BOUNDARY is used for all arguments.
V.4 wants long longs and doubles to be double word aligned. Just
testing the mode size is a boneheaded way to do this as it means
that other types such as complex int are also double word aligned.
However, we're stuck with this because changing the ABI might break
existing library interfaces.
Quadword align Altivec/VSX vectors.
Quadword align large synthetic vector types. */
static unsigned int
rs6000_function_arg_boundary (machine_mode mode, const_tree type)
{
machine_mode elt_mode;
int n_elts;
rs6000_discover_homogeneous_aggregate (mode, type, &elt_mode, &n_elts);
if (DEFAULT_ABI == ABI_V4
&& (GET_MODE_SIZE (mode) == 8
|| (TARGET_HARD_FLOAT
&& !is_complex_IBM_long_double (mode)
&& FLOAT128_2REG_P (mode))))
return 64;
else if (FLOAT128_VECTOR_P (mode))
return 128;
else if (PAIRED_VECTOR_MODE (mode)
|| (type && TREE_CODE (type) == VECTOR_TYPE
&& int_size_in_bytes (type) >= 8
&& int_size_in_bytes (type) < 16))
return 64;
else if (ALTIVEC_OR_VSX_VECTOR_MODE (elt_mode)
|| (type && TREE_CODE (type) == VECTOR_TYPE
&& int_size_in_bytes (type) >= 16))
return 128;
/* Aggregate types that need > 8 byte alignment are quadword-aligned
in the parameter area in the ELFv2 ABI, and in the AIX ABI unless
-mcompat-align-parm is used. */
if (((DEFAULT_ABI == ABI_AIX && !rs6000_compat_align_parm)
|| DEFAULT_ABI == ABI_ELFv2)
&& type && TYPE_ALIGN (type) > 64)
{
/* "Aggregate" means any AGGREGATE_TYPE except for single-element
or homogeneous float/vector aggregates here. We already handled
vector aggregates above, but still need to check for float here. */
bool aggregate_p = (AGGREGATE_TYPE_P (type)
&& !SCALAR_FLOAT_MODE_P (elt_mode));
/* We used to check for BLKmode instead of the above aggregate type
check. Warn when this results in any difference to the ABI. */
if (aggregate_p != (mode == BLKmode))
{
static bool warned;
if (!warned && warn_psabi)
{
warned = true;
inform (input_location,
"the ABI of passing aggregates with %d-byte alignment"
" has changed in GCC 5",
(int) TYPE_ALIGN (type) / BITS_PER_UNIT);
}
}
if (aggregate_p)
return 128;
}
/* Similar for the Darwin64 ABI. Note that for historical reasons we
implement the "aggregate type" check as a BLKmode check here; this
means certain aggregate types are in fact not aligned. */
if (TARGET_MACHO && rs6000_darwin64_abi
&& mode == BLKmode
&& type && TYPE_ALIGN (type) > 64)
return 128;
return PARM_BOUNDARY;
}
/* The offset in words to the start of the parameter save area. */
static unsigned int
rs6000_parm_offset (void)
{
return (DEFAULT_ABI == ABI_V4 ? 2
: DEFAULT_ABI == ABI_ELFv2 ? 4
: 6);
}
/* For a function parm of MODE and TYPE, return the starting word in
the parameter area. NWORDS of the parameter area are already used. */
static unsigned int
rs6000_parm_start (machine_mode mode, const_tree type,
unsigned int nwords)
{
unsigned int align;
align = rs6000_function_arg_boundary (mode, type) / PARM_BOUNDARY - 1;
return nwords + (-(rs6000_parm_offset () + nwords) & align);
}
/* Compute the size (in words) of a function argument. */
static unsigned long
rs6000_arg_size (machine_mode mode, const_tree type)
{
unsigned long size;
if (mode != BLKmode)
size = GET_MODE_SIZE (mode);
else
size = int_size_in_bytes (type);
if (TARGET_32BIT)
return (size + 3) >> 2;
else
return (size + 7) >> 3;
}
/* Use this to flush pending int fields. */
static void
rs6000_darwin64_record_arg_advance_flush (CUMULATIVE_ARGS *cum,
HOST_WIDE_INT bitpos, int final)
{
unsigned int startbit, endbit;
int intregs, intoffset;
/* Handle the situations where a float is taking up the first half
of the GPR, and the other half is empty (typically due to
alignment restrictions). We can detect this by a 8-byte-aligned
int field, or by seeing that this is the final flush for this
argument. Count the word and continue on. */
if (cum->floats_in_gpr == 1
&& (cum->intoffset % 64 == 0
|| (cum->intoffset == -1 && final)))
{
cum->words++;
cum->floats_in_gpr = 0;
}
if (cum->intoffset == -1)
return;
intoffset = cum->intoffset;
cum->intoffset = -1;
cum->floats_in_gpr = 0;
if (intoffset % BITS_PER_WORD != 0)
{
unsigned int bits = BITS_PER_WORD - intoffset % BITS_PER_WORD;
if (!int_mode_for_size (bits, 0).exists ())
{
/* We couldn't find an appropriate mode, which happens,
e.g., in packed structs when there are 3 bytes to load.
Back intoffset back to the beginning of the word in this
case. */
intoffset = ROUND_DOWN (intoffset, BITS_PER_WORD);
}
}
startbit = ROUND_DOWN (intoffset, BITS_PER_WORD);
endbit = ROUND_UP (bitpos, BITS_PER_WORD);
intregs = (endbit - startbit) / BITS_PER_WORD;
cum->words += intregs;
/* words should be unsigned. */
if ((unsigned)cum->words < (endbit/BITS_PER_WORD))
{
int pad = (endbit/BITS_PER_WORD) - cum->words;
cum->words += pad;
}
}
/* The darwin64 ABI calls for us to recurse down through structs,
looking for elements passed in registers. Unfortunately, we have
to track int register count here also because of misalignments
in powerpc alignment mode. */
static void
rs6000_darwin64_record_arg_advance_recurse (CUMULATIVE_ARGS *cum,
const_tree type,
HOST_WIDE_INT startbitpos)
{
tree f;
for (f = TYPE_FIELDS (type); f ; f = DECL_CHAIN (f))
if (TREE_CODE (f) == FIELD_DECL)
{
HOST_WIDE_INT bitpos = startbitpos;
tree ftype = TREE_TYPE (f);
machine_mode mode;
if (ftype == error_mark_node)
continue;
mode = TYPE_MODE (ftype);
if (DECL_SIZE (f) != 0
&& tree_fits_uhwi_p (bit_position (f)))
bitpos += int_bit_position (f);
/* ??? FIXME: else assume zero offset. */
if (TREE_CODE (ftype) == RECORD_TYPE)
rs6000_darwin64_record_arg_advance_recurse (cum, ftype, bitpos);
else if (USE_FP_FOR_ARG_P (cum, mode))
{
unsigned n_fpregs = (GET_MODE_SIZE (mode) + 7) >> 3;
rs6000_darwin64_record_arg_advance_flush (cum, bitpos, 0);
cum->fregno += n_fpregs;
/* Single-precision floats present a special problem for
us, because they are smaller than an 8-byte GPR, and so
the structure-packing rules combined with the standard
varargs behavior mean that we want to pack float/float
and float/int combinations into a single register's
space. This is complicated by the arg advance flushing,
which works on arbitrarily large groups of int-type
fields. */
if (mode == SFmode)
{
if (cum->floats_in_gpr == 1)
{
/* Two floats in a word; count the word and reset
the float count. */
cum->words++;
cum->floats_in_gpr = 0;
}
else if (bitpos % 64 == 0)
{
/* A float at the beginning of an 8-byte word;
count it and put off adjusting cum->words until
we see if a arg advance flush is going to do it
for us. */
cum->floats_in_gpr++;
}
else
{
/* The float is at the end of a word, preceded
by integer fields, so the arg advance flush
just above has already set cum->words and
everything is taken care of. */
}
}
else
cum->words += n_fpregs;
}
else if (USE_ALTIVEC_FOR_ARG_P (cum, mode, 1))
{
rs6000_darwin64_record_arg_advance_flush (cum, bitpos, 0);
cum->vregno++;
cum->words += 2;
}
else if (cum->intoffset == -1)
cum->intoffset = bitpos;
}
}
/* Check for an item that needs to be considered specially under the darwin 64
bit ABI. These are record types where the mode is BLK or the structure is
8 bytes in size. */
static int
rs6000_darwin64_struct_check_p (machine_mode mode, const_tree type)
{
return rs6000_darwin64_abi
&& ((mode == BLKmode
&& TREE_CODE (type) == RECORD_TYPE
&& int_size_in_bytes (type) > 0)
|| (type && TREE_CODE (type) == RECORD_TYPE
&& int_size_in_bytes (type) == 8)) ? 1 : 0;
}
/* Update the data in CUM to advance over an argument
of mode MODE and data type TYPE.
(TYPE is null for libcalls where that information may not be available.)
Note that for args passed by reference, function_arg will be called
with MODE and TYPE set to that of the pointer to the arg, not the arg
itself. */
static void
rs6000_function_arg_advance_1 (CUMULATIVE_ARGS *cum, machine_mode mode,
const_tree type, bool named, int depth)
{
machine_mode elt_mode;
int n_elts;
rs6000_discover_homogeneous_aggregate (mode, type, &elt_mode, &n_elts);
/* Only tick off an argument if we're not recursing. */
if (depth == 0)
cum->nargs_prototype--;
#ifdef HAVE_AS_GNU_ATTRIBUTE
if (TARGET_ELF && (TARGET_64BIT || DEFAULT_ABI == ABI_V4)
&& cum->escapes)
{
if (SCALAR_FLOAT_MODE_P (mode))
{
rs6000_passes_float = true;
if ((HAVE_LD_PPC_GNU_ATTR_LONG_DOUBLE || TARGET_64BIT)
&& (FLOAT128_IBM_P (mode)
|| FLOAT128_IEEE_P (mode)
|| (type != NULL
&& TYPE_MAIN_VARIANT (type) == long_double_type_node)))
rs6000_passes_long_double = true;
}
if ((named && ALTIVEC_OR_VSX_VECTOR_MODE (mode))
|| (PAIRED_VECTOR_MODE (mode)
&& !cum->stdarg
&& cum->sysv_gregno <= GP_ARG_MAX_REG))
rs6000_passes_vector = true;
}
#endif
if (TARGET_ALTIVEC_ABI
&& (ALTIVEC_OR_VSX_VECTOR_MODE (elt_mode)
|| (type && TREE_CODE (type) == VECTOR_TYPE
&& int_size_in_bytes (type) == 16)))
{
bool stack = false;
if (USE_ALTIVEC_FOR_ARG_P (cum, elt_mode, named))
{
cum->vregno += n_elts;
if (!TARGET_ALTIVEC)
error ("cannot pass argument in vector register because"
" altivec instructions are disabled, use %qs"
" to enable them", "-maltivec");
/* PowerPC64 Linux and AIX allocate GPRs for a vector argument
even if it is going to be passed in a vector register.
Darwin does the same for variable-argument functions. */
if (((DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
&& TARGET_64BIT)
|| (cum->stdarg && DEFAULT_ABI != ABI_V4))
stack = true;
}
else
stack = true;
if (stack)
{
int align;
/* Vector parameters must be 16-byte aligned. In 32-bit
mode this means we need to take into account the offset
to the parameter save area. In 64-bit mode, they just
have to start on an even word, since the parameter save
area is 16-byte aligned. */
if (TARGET_32BIT)
align = -(rs6000_parm_offset () + cum->words) & 3;
else
align = cum->words & 1;
cum->words += align + rs6000_arg_size (mode, type);
if (TARGET_DEBUG_ARG)
{
fprintf (stderr, "function_adv: words = %2d, align=%d, ",
cum->words, align);
fprintf (stderr, "nargs = %4d, proto = %d, mode = %4s\n",
cum->nargs_prototype, cum->prototype,
GET_MODE_NAME (mode));
}
}
}
else if (TARGET_MACHO && rs6000_darwin64_struct_check_p (mode, type))
{
int size = int_size_in_bytes (type);
/* Variable sized types have size == -1 and are
treated as if consisting entirely of ints.
Pad to 16 byte boundary if needed. */
if (TYPE_ALIGN (type) >= 2 * BITS_PER_WORD
&& (cum->words % 2) != 0)
cum->words++;
/* For varargs, we can just go up by the size of the struct. */
if (!named)
cum->words += (size + 7) / 8;
else
{
/* It is tempting to say int register count just goes up by
sizeof(type)/8, but this is wrong in a case such as
{ int; double; int; } [powerpc alignment]. We have to
grovel through the fields for these too. */
cum->intoffset = 0;
cum->floats_in_gpr = 0;
rs6000_darwin64_record_arg_advance_recurse (cum, type, 0);
rs6000_darwin64_record_arg_advance_flush (cum,
size * BITS_PER_UNIT, 1);
}
if (TARGET_DEBUG_ARG)
{
fprintf (stderr, "function_adv: words = %2d, align=%d, size=%d",
cum->words, TYPE_ALIGN (type), size);
fprintf (stderr,
"nargs = %4d, proto = %d, mode = %4s (darwin64 abi)\n",
cum->nargs_prototype, cum->prototype,
GET_MODE_NAME (mode));
}
}
else if (DEFAULT_ABI == ABI_V4)
{
if (abi_v4_pass_in_fpr (mode))
{
/* _Decimal128 must use an even/odd register pair. This assumes
that the register number is odd when fregno is odd. */
if (mode == TDmode && (cum->fregno % 2) == 1)
cum->fregno++;
if (cum->fregno + (FLOAT128_2REG_P (mode) ? 1 : 0)
<= FP_ARG_V4_MAX_REG)
cum->fregno += (GET_MODE_SIZE (mode) + 7) >> 3;
else
{
cum->fregno = FP_ARG_V4_MAX_REG + 1;
if (mode == DFmode || FLOAT128_IBM_P (mode)
|| mode == DDmode || mode == TDmode)
cum->words += cum->words & 1;
cum->words += rs6000_arg_size (mode, type);
}
}
else
{
int n_words = rs6000_arg_size (mode, type);
int gregno = cum->sysv_gregno;
/* Long long is put in (r3,r4), (r5,r6), (r7,r8) or (r9,r10).
As does any other 2 word item such as complex int due to a
historical mistake. */
if (n_words == 2)
gregno += (1 - gregno) & 1;
/* Multi-reg args are not split between registers and stack. */
if (gregno + n_words - 1 > GP_ARG_MAX_REG)
{
/* Long long is aligned on the stack. So are other 2 word
items such as complex int due to a historical mistake. */
if (n_words == 2)
cum->words += cum->words & 1;
cum->words += n_words;
}
/* Note: continuing to accumulate gregno past when we've started
spilling to the stack indicates the fact that we've started
spilling to the stack to expand_builtin_saveregs. */
cum->sysv_gregno = gregno + n_words;
}
if (TARGET_DEBUG_ARG)
{
fprintf (stderr, "function_adv: words = %2d, fregno = %2d, ",
cum->words, cum->fregno);
fprintf (stderr, "gregno = %2d, nargs = %4d, proto = %d, ",
cum->sysv_gregno, cum->nargs_prototype, cum->prototype);
fprintf (stderr, "mode = %4s, named = %d\n",
GET_MODE_NAME (mode), named);
}
}
else
{
int n_words = rs6000_arg_size (mode, type);
int start_words = cum->words;
int align_words = rs6000_parm_start (mode, type, start_words);
cum->words = align_words + n_words;
if (SCALAR_FLOAT_MODE_P (elt_mode) && TARGET_HARD_FLOAT)
{
/* _Decimal128 must be passed in an even/odd float register pair.
This assumes that the register number is odd when fregno is
odd. */
if (elt_mode == TDmode && (cum->fregno % 2) == 1)
cum->fregno++;
cum->fregno += n_elts * ((GET_MODE_SIZE (elt_mode) + 7) >> 3);
}
if (TARGET_DEBUG_ARG)
{
fprintf (stderr, "function_adv: words = %2d, fregno = %2d, ",
cum->words, cum->fregno);
fprintf (stderr, "nargs = %4d, proto = %d, mode = %4s, ",
cum->nargs_prototype, cum->prototype, GET_MODE_NAME (mode));
fprintf (stderr, "named = %d, align = %d, depth = %d\n",
named, align_words - start_words, depth);
}
}
}
static void
rs6000_function_arg_advance (cumulative_args_t cum, machine_mode mode,
const_tree type, bool named)
{
rs6000_function_arg_advance_1 (get_cumulative_args (cum), mode, type, named,
0);
}
/* A subroutine of rs6000_darwin64_record_arg. Assign the bits of the
structure between cum->intoffset and bitpos to integer registers. */
static void
rs6000_darwin64_record_arg_flush (CUMULATIVE_ARGS *cum,
HOST_WIDE_INT bitpos, rtx rvec[], int *k)
{
machine_mode mode;
unsigned int regno;
unsigned int startbit, endbit;
int this_regno, intregs, intoffset;
rtx reg;
if (cum->intoffset == -1)
return;
intoffset = cum->intoffset;
cum->intoffset = -1;
/* If this is the trailing part of a word, try to only load that
much into the register. Otherwise load the whole register. Note
that in the latter case we may pick up unwanted bits. It's not a
problem at the moment but may wish to revisit. */
if (intoffset % BITS_PER_WORD != 0)
{
unsigned int bits = BITS_PER_WORD - intoffset % BITS_PER_WORD;
if (!int_mode_for_size (bits, 0).exists (&mode))
{
/* We couldn't find an appropriate mode, which happens,
e.g., in packed structs when there are 3 bytes to load.
Back intoffset back to the beginning of the word in this
case. */
intoffset = ROUND_DOWN (intoffset, BITS_PER_WORD);
mode = word_mode;
}
}
else
mode = word_mode;
startbit = ROUND_DOWN (intoffset, BITS_PER_WORD);
endbit = ROUND_UP (bitpos, BITS_PER_WORD);
intregs = (endbit - startbit) / BITS_PER_WORD;
this_regno = cum->words + intoffset / BITS_PER_WORD;
if (intregs > 0 && intregs > GP_ARG_NUM_REG - this_regno)
cum->use_stack = 1;
intregs = MIN (intregs, GP_ARG_NUM_REG - this_regno);
if (intregs <= 0)
return;
intoffset /= BITS_PER_UNIT;
do
{
regno = GP_ARG_MIN_REG + this_regno;
reg = gen_rtx_REG (mode, regno);
rvec[(*k)++] =
gen_rtx_EXPR_LIST (VOIDmode, reg, GEN_INT (intoffset));
this_regno += 1;
intoffset = (intoffset | (UNITS_PER_WORD-1)) + 1;
mode = word_mode;
intregs -= 1;
}
while (intregs > 0);
}
/* Recursive workhorse for the following. */
static void
rs6000_darwin64_record_arg_recurse (CUMULATIVE_ARGS *cum, const_tree type,
HOST_WIDE_INT startbitpos, rtx rvec[],
int *k)
{
tree f;
for (f = TYPE_FIELDS (type); f ; f = DECL_CHAIN (f))
if (TREE_CODE (f) == FIELD_DECL)
{
HOST_WIDE_INT bitpos = startbitpos;
tree ftype = TREE_TYPE (f);
machine_mode mode;
if (ftype == error_mark_node)
continue;
mode = TYPE_MODE (ftype);
if (DECL_SIZE (f) != 0
&& tree_fits_uhwi_p (bit_position (f)))
bitpos += int_bit_position (f);
/* ??? FIXME: else assume zero offset. */
if (TREE_CODE (ftype) == RECORD_TYPE)
rs6000_darwin64_record_arg_recurse (cum, ftype, bitpos, rvec, k);
else if (cum->named && USE_FP_FOR_ARG_P (cum, mode))
{
unsigned n_fpreg = (GET_MODE_SIZE (mode) + 7) >> 3;
#if 0
switch (mode)
{
case E_SCmode: mode = SFmode; break;
case E_DCmode: mode = DFmode; break;
case E_TCmode: mode = TFmode; break;
default: break;
}
#endif
rs6000_darwin64_record_arg_flush (cum, bitpos, rvec, k);
if (cum->fregno + n_fpreg > FP_ARG_MAX_REG + 1)
{
gcc_assert (cum->fregno == FP_ARG_MAX_REG
&& (mode == TFmode || mode == TDmode));
/* Long double or _Decimal128 split over regs and memory. */
mode = DECIMAL_FLOAT_MODE_P (mode) ? DDmode : DFmode;
cum->use_stack=1;
}
rvec[(*k)++]
= gen_rtx_EXPR_LIST (VOIDmode,
gen_rtx_REG (mode, cum->fregno++),
GEN_INT (bitpos / BITS_PER_UNIT));
if (FLOAT128_2REG_P (mode))
cum->fregno++;
}
else if (cum->named && USE_ALTIVEC_FOR_ARG_P (cum, mode, 1))
{
rs6000_darwin64_record_arg_flush (cum, bitpos, rvec, k);
rvec[(*k)++]
= gen_rtx_EXPR_LIST (VOIDmode,
gen_rtx_REG (mode, cum->vregno++),
GEN_INT (bitpos / BITS_PER_UNIT));
}
else if (cum->intoffset == -1)
cum->intoffset = bitpos;
}
}
/* For the darwin64 ABI, we want to construct a PARALLEL consisting of
the register(s) to be used for each field and subfield of a struct
being passed by value, along with the offset of where the
register's value may be found in the block. FP fields go in FP
register, vector fields go in vector registers, and everything
else goes in int registers, packed as in memory.
This code is also used for function return values. RETVAL indicates
whether this is the case.
Much of this is taken from the SPARC V9 port, which has a similar
calling convention. */
static rtx
rs6000_darwin64_record_arg (CUMULATIVE_ARGS *orig_cum, const_tree type,
bool named, bool retval)
{
rtx rvec[FIRST_PSEUDO_REGISTER];
int k = 1, kbase = 1;
HOST_WIDE_INT typesize = int_size_in_bytes (type);
/* This is a copy; modifications are not visible to our caller. */
CUMULATIVE_ARGS copy_cum = *orig_cum;
CUMULATIVE_ARGS *cum = ©_cum;
/* Pad to 16 byte boundary if needed. */
if (!retval && TYPE_ALIGN (type) >= 2 * BITS_PER_WORD
&& (cum->words % 2) != 0)
cum->words++;
cum->intoffset = 0;
cum->use_stack = 0;
cum->named = named;
/* Put entries into rvec[] for individual FP and vector fields, and
for the chunks of memory that go in int regs. Note we start at
element 1; 0 is reserved for an indication of using memory, and
may or may not be filled in below. */
rs6000_darwin64_record_arg_recurse (cum, type, /* startbit pos= */ 0, rvec, &k);
rs6000_darwin64_record_arg_flush (cum, typesize * BITS_PER_UNIT, rvec, &k);
/* If any part of the struct went on the stack put all of it there.
This hack is because the generic code for
FUNCTION_ARG_PARTIAL_NREGS cannot handle cases where the register
parts of the struct are not at the beginning. */
if (cum->use_stack)
{
if (retval)
return NULL_RTX; /* doesn't go in registers at all */
kbase = 0;
rvec[0] = gen_rtx_EXPR_LIST (VOIDmode, NULL_RTX, const0_rtx);
}
if (k > 1 || cum->use_stack)
return gen_rtx_PARALLEL (BLKmode, gen_rtvec_v (k - kbase, &rvec[kbase]));
else
return NULL_RTX;
}
/* Determine where to place an argument in 64-bit mode with 32-bit ABI. */
static rtx
rs6000_mixed_function_arg (machine_mode mode, const_tree type,
int align_words)
{
int n_units;
int i, k;
rtx rvec[GP_ARG_NUM_REG + 1];
if (align_words >= GP_ARG_NUM_REG)
return NULL_RTX;
n_units = rs6000_arg_size (mode, type);
/* Optimize the simple case where the arg fits in one gpr, except in
the case of BLKmode due to assign_parms assuming that registers are
BITS_PER_WORD wide. */
if (n_units == 0
|| (n_units == 1 && mode != BLKmode))
return gen_rtx_REG (mode, GP_ARG_MIN_REG + align_words);
k = 0;
if (align_words + n_units > GP_ARG_NUM_REG)
/* Not all of the arg fits in gprs. Say that it goes in memory too,
using a magic NULL_RTX component.
This is not strictly correct. Only some of the arg belongs in
memory, not all of it. However, the normal scheme using
function_arg_partial_nregs can result in unusual subregs, eg.
(subreg:SI (reg:DF) 4), which are not handled well. The code to
store the whole arg to memory is often more efficient than code
to store pieces, and we know that space is available in the right
place for the whole arg. */
rvec[k++] = gen_rtx_EXPR_LIST (VOIDmode, NULL_RTX, const0_rtx);
i = 0;
do
{
rtx r = gen_rtx_REG (SImode, GP_ARG_MIN_REG + align_words);
rtx off = GEN_INT (i++ * 4);
rvec[k++] = gen_rtx_EXPR_LIST (VOIDmode, r, off);
}
while (++align_words < GP_ARG_NUM_REG && --n_units != 0);
return gen_rtx_PARALLEL (mode, gen_rtvec_v (k, rvec));
}
/* We have an argument of MODE and TYPE that goes into FPRs or VRs,
but must also be copied into the parameter save area starting at
offset ALIGN_WORDS. Fill in RVEC with the elements corresponding
to the GPRs and/or memory. Return the number of elements used. */
static int
rs6000_psave_function_arg (machine_mode mode, const_tree type,
int align_words, rtx *rvec)
{
int k = 0;
if (align_words < GP_ARG_NUM_REG)
{
int n_words = rs6000_arg_size (mode, type);
if (align_words + n_words > GP_ARG_NUM_REG
|| mode == BLKmode
|| (TARGET_32BIT && TARGET_POWERPC64))
{
/* If this is partially on the stack, then we only
include the portion actually in registers here. */
machine_mode rmode = TARGET_32BIT ? SImode : DImode;
int i = 0;
if (align_words + n_words > GP_ARG_NUM_REG)
{
/* Not all of the arg fits in gprs. Say that it goes in memory
too, using a magic NULL_RTX component. Also see comment in
rs6000_mixed_function_arg for why the normal
function_arg_partial_nregs scheme doesn't work in this case. */
rvec[k++] = gen_rtx_EXPR_LIST (VOIDmode, NULL_RTX, const0_rtx);
}
do
{
rtx r = gen_rtx_REG (rmode, GP_ARG_MIN_REG + align_words);
rtx off = GEN_INT (i++ * GET_MODE_SIZE (rmode));
rvec[k++] = gen_rtx_EXPR_LIST (VOIDmode, r, off);
}
while (++align_words < GP_ARG_NUM_REG && --n_words != 0);
}
else
{
/* The whole arg fits in gprs. */
rtx r = gen_rtx_REG (mode, GP_ARG_MIN_REG + align_words);
rvec[k++] = gen_rtx_EXPR_LIST (VOIDmode, r, const0_rtx);
}
}
else
{
/* It's entirely in memory. */
rvec[k++] = gen_rtx_EXPR_LIST (VOIDmode, NULL_RTX, const0_rtx);
}
return k;
}
/* RVEC is a vector of K components of an argument of mode MODE.
Construct the final function_arg return value from it. */
static rtx
rs6000_finish_function_arg (machine_mode mode, rtx *rvec, int k)
{
gcc_assert (k >= 1);
/* Avoid returning a PARALLEL in the trivial cases. */
if (k == 1)
{
if (XEXP (rvec[0], 0) == NULL_RTX)
return NULL_RTX;
if (GET_MODE (XEXP (rvec[0], 0)) == mode)
return XEXP (rvec[0], 0);
}
return gen_rtx_PARALLEL (mode, gen_rtvec_v (k, rvec));
}
/* Determine where to put an argument to a function.
Value is zero to push the argument on the stack,
or a hard register in which to store the argument.
MODE is the argument's machine mode.
TYPE is the data type of the argument (as a tree).
This is null for libcalls where that information may
not be available.
CUM is a variable of type CUMULATIVE_ARGS which gives info about
the preceding args and about the function being called. It is
not modified in this routine.
NAMED is nonzero if this argument is a named parameter
(otherwise it is an extra parameter matching an ellipsis).
On RS/6000 the first eight words of non-FP are normally in registers
and the rest are pushed. Under AIX, the first 13 FP args are in registers.
Under V.4, the first 8 FP args are in registers.
If this is floating-point and no prototype is specified, we use
both an FP and integer register (or possibly FP reg and stack). Library
functions (when CALL_LIBCALL is set) always have the proper types for args,
so we can pass the FP value just in one register. emit_library_function
doesn't support PARALLEL anyway.
Note that for args passed by reference, function_arg will be called
with MODE and TYPE set to that of the pointer to the arg, not the arg
itself. */
static rtx
rs6000_function_arg (cumulative_args_t cum_v, machine_mode mode,
const_tree type, bool named)
{
CUMULATIVE_ARGS *cum = get_cumulative_args (cum_v);
enum rs6000_abi abi = DEFAULT_ABI;
machine_mode elt_mode;
int n_elts;
/* Return a marker to indicate whether CR1 needs to set or clear the
bit that V.4 uses to say fp args were passed in registers.
Assume that we don't need the marker for software floating point,
or compiler generated library calls. */
if (mode == VOIDmode)
{
if (abi == ABI_V4
&& (cum->call_cookie & CALL_LIBCALL) == 0
&& (cum->stdarg
|| (cum->nargs_prototype < 0
&& (cum->prototype || TARGET_NO_PROTOTYPE)))
&& TARGET_HARD_FLOAT)
return GEN_INT (cum->call_cookie
| ((cum->fregno == FP_ARG_MIN_REG)
? CALL_V4_SET_FP_ARGS
: CALL_V4_CLEAR_FP_ARGS));
return GEN_INT (cum->call_cookie & ~CALL_LIBCALL);
}
rs6000_discover_homogeneous_aggregate (mode, type, &elt_mode, &n_elts);
if (TARGET_MACHO && rs6000_darwin64_struct_check_p (mode, type))
{
rtx rslt = rs6000_darwin64_record_arg (cum, type, named, /*retval= */false);
if (rslt != NULL_RTX)
return rslt;
/* Else fall through to usual handling. */
}
if (USE_ALTIVEC_FOR_ARG_P (cum, elt_mode, named))
{
rtx rvec[GP_ARG_NUM_REG + AGGR_ARG_NUM_REG + 1];
rtx r, off;
int i, k = 0;
/* Do we also need to pass this argument in the parameter save area?
Library support functions for IEEE 128-bit are assumed to not need the
value passed both in GPRs and in vector registers. */
if (TARGET_64BIT && !cum->prototype
&& (!cum->libcall || !FLOAT128_VECTOR_P (elt_mode)))
{
int align_words = ROUND_UP (cum->words, 2);
k = rs6000_psave_function_arg (mode, type, align_words, rvec);
}
/* Describe where this argument goes in the vector registers. */
for (i = 0; i < n_elts && cum->vregno + i <= ALTIVEC_ARG_MAX_REG; i++)
{
r = gen_rtx_REG (elt_mode, cum->vregno + i);
off = GEN_INT (i * GET_MODE_SIZE (elt_mode));
rvec[k++] = gen_rtx_EXPR_LIST (VOIDmode, r, off);
}
return rs6000_finish_function_arg (mode, rvec, k);
}
else if (TARGET_ALTIVEC_ABI
&& (ALTIVEC_OR_VSX_VECTOR_MODE (mode)
|| (type && TREE_CODE (type) == VECTOR_TYPE
&& int_size_in_bytes (type) == 16)))
{
if (named || abi == ABI_V4)
return NULL_RTX;
else
{
/* Vector parameters to varargs functions under AIX or Darwin
get passed in memory and possibly also in GPRs. */
int align, align_words, n_words;
machine_mode part_mode;
/* Vector parameters must be 16-byte aligned. In 32-bit
mode this means we need to take into account the offset
to the parameter save area. In 64-bit mode, they just
have to start on an even word, since the parameter save
area is 16-byte aligned. */
if (TARGET_32BIT)
align = -(rs6000_parm_offset () + cum->words) & 3;
else
align = cum->words & 1;
align_words = cum->words + align;
/* Out of registers? Memory, then. */
if (align_words >= GP_ARG_NUM_REG)
return NULL_RTX;
if (TARGET_32BIT && TARGET_POWERPC64)
return rs6000_mixed_function_arg (mode, type, align_words);
/* The vector value goes in GPRs. Only the part of the
value in GPRs is reported here. */
part_mode = mode;
n_words = rs6000_arg_size (mode, type);
if (align_words + n_words > GP_ARG_NUM_REG)
/* Fortunately, there are only two possibilities, the value
is either wholly in GPRs or half in GPRs and half not. */
part_mode = DImode;
return gen_rtx_REG (part_mode, GP_ARG_MIN_REG + align_words);
}
}
else if (abi == ABI_V4)
{
if (abi_v4_pass_in_fpr (mode))
{
/* _Decimal128 must use an even/odd register pair. This assumes
that the register number is odd when fregno is odd. */
if (mode == TDmode && (cum->fregno % 2) == 1)
cum->fregno++;
if (cum->fregno + (FLOAT128_2REG_P (mode) ? 1 : 0)
<= FP_ARG_V4_MAX_REG)
return gen_rtx_REG (mode, cum->fregno);
else
return NULL_RTX;
}
else
{
int n_words = rs6000_arg_size (mode, type);
int gregno = cum->sysv_gregno;
/* Long long is put in (r3,r4), (r5,r6), (r7,r8) or (r9,r10).
As does any other 2 word item such as complex int due to a
historical mistake. */
if (n_words == 2)
gregno += (1 - gregno) & 1;
/* Multi-reg args are not split between registers and stack. */
if (gregno + n_words - 1 > GP_ARG_MAX_REG)
return NULL_RTX;
if (TARGET_32BIT && TARGET_POWERPC64)
return rs6000_mixed_function_arg (mode, type,
gregno - GP_ARG_MIN_REG);
return gen_rtx_REG (mode, gregno);
}
}
else
{
int align_words = rs6000_parm_start (mode, type, cum->words);
/* _Decimal128 must be passed in an even/odd float register pair.
This assumes that the register number is odd when fregno is odd. */
if (elt_mode == TDmode && (cum->fregno % 2) == 1)
cum->fregno++;
if (USE_FP_FOR_ARG_P (cum, elt_mode))
{
rtx rvec[GP_ARG_NUM_REG + AGGR_ARG_NUM_REG + 1];
rtx r, off;
int i, k = 0;
unsigned long n_fpreg = (GET_MODE_SIZE (elt_mode) + 7) >> 3;
int fpr_words;
/* Do we also need to pass this argument in the parameter
save area? */
if (type && (cum->nargs_prototype <= 0
|| ((DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
&& TARGET_XL_COMPAT
&& align_words >= GP_ARG_NUM_REG)))
k = rs6000_psave_function_arg (mode, type, align_words, rvec);
/* Describe where this argument goes in the fprs. */
for (i = 0; i < n_elts
&& cum->fregno + i * n_fpreg <= FP_ARG_MAX_REG; i++)
{
/* Check if the argument is split over registers and memory.
This can only ever happen for long double or _Decimal128;
complex types are handled via split_complex_arg. */
machine_mode fmode = elt_mode;
if (cum->fregno + (i + 1) * n_fpreg > FP_ARG_MAX_REG + 1)
{
gcc_assert (FLOAT128_2REG_P (fmode));
fmode = DECIMAL_FLOAT_MODE_P (fmode) ? DDmode : DFmode;
}
r = gen_rtx_REG (fmode, cum->fregno + i * n_fpreg);
off = GEN_INT (i * GET_MODE_SIZE (elt_mode));
rvec[k++] = gen_rtx_EXPR_LIST (VOIDmode, r, off);
}
/* If there were not enough FPRs to hold the argument, the rest
usually goes into memory. However, if the current position
is still within the register parameter area, a portion may
actually have to go into GPRs.
Note that it may happen that the portion of the argument
passed in the first "half" of the first GPR was already
passed in the last FPR as well.
For unnamed arguments, we already set up GPRs to cover the
whole argument in rs6000_psave_function_arg, so there is
nothing further to do at this point. */
fpr_words = (i * GET_MODE_SIZE (elt_mode)) / (TARGET_32BIT ? 4 : 8);
if (i < n_elts && align_words + fpr_words < GP_ARG_NUM_REG
&& cum->nargs_prototype > 0)
{
static bool warned;
machine_mode rmode = TARGET_32BIT ? SImode : DImode;
int n_words = rs6000_arg_size (mode, type);
align_words += fpr_words;
n_words -= fpr_words;
do
{
r = gen_rtx_REG (rmode, GP_ARG_MIN_REG + align_words);
off = GEN_INT (fpr_words++ * GET_MODE_SIZE (rmode));
rvec[k++] = gen_rtx_EXPR_LIST (VOIDmode, r, off);
}
while (++align_words < GP_ARG_NUM_REG && --n_words != 0);
if (!warned && warn_psabi)
{
warned = true;
inform (input_location,
"the ABI of passing homogeneous float aggregates"
" has changed in GCC 5");
}
}
return rs6000_finish_function_arg (mode, rvec, k);
}
else if (align_words < GP_ARG_NUM_REG)
{
if (TARGET_32BIT && TARGET_POWERPC64)
return rs6000_mixed_function_arg (mode, type, align_words);
return gen_rtx_REG (mode, GP_ARG_MIN_REG + align_words);
}
else
return NULL_RTX;
}
}
/* For an arg passed partly in registers and partly in memory, this is
the number of bytes passed in registers. For args passed entirely in
registers or entirely in memory, zero. When an arg is described by a
PARALLEL, perhaps using more than one register type, this function
returns the number of bytes used by the first element of the PARALLEL. */
static int
rs6000_arg_partial_bytes (cumulative_args_t cum_v, machine_mode mode,
tree type, bool named)
{
CUMULATIVE_ARGS *cum = get_cumulative_args (cum_v);
bool passed_in_gprs = true;
int ret = 0;
int align_words;
machine_mode elt_mode;
int n_elts;
rs6000_discover_homogeneous_aggregate (mode, type, &elt_mode, &n_elts);
if (DEFAULT_ABI == ABI_V4)
return 0;
if (USE_ALTIVEC_FOR_ARG_P (cum, elt_mode, named))
{
/* If we are passing this arg in the fixed parameter save area (gprs or
memory) as well as VRs, we do not use the partial bytes mechanism;
instead, rs6000_function_arg will return a PARALLEL including a memory
element as necessary. Library support functions for IEEE 128-bit are
assumed to not need the value passed both in GPRs and in vector
registers. */
if (TARGET_64BIT && !cum->prototype
&& (!cum->libcall || !FLOAT128_VECTOR_P (elt_mode)))
return 0;
/* Otherwise, we pass in VRs only. Check for partial copies. */
passed_in_gprs = false;
if (cum->vregno + n_elts > ALTIVEC_ARG_MAX_REG + 1)
ret = (ALTIVEC_ARG_MAX_REG + 1 - cum->vregno) * 16;
}
/* In this complicated case we just disable the partial_nregs code. */
if (TARGET_MACHO && rs6000_darwin64_struct_check_p (mode, type))
return 0;
align_words = rs6000_parm_start (mode, type, cum->words);
if (USE_FP_FOR_ARG_P (cum, elt_mode))
{
unsigned long n_fpreg = (GET_MODE_SIZE (elt_mode) + 7) >> 3;
/* If we are passing this arg in the fixed parameter save area
(gprs or memory) as well as FPRs, we do not use the partial
bytes mechanism; instead, rs6000_function_arg will return a
PARALLEL including a memory element as necessary. */
if (type
&& (cum->nargs_prototype <= 0
|| ((DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
&& TARGET_XL_COMPAT
&& align_words >= GP_ARG_NUM_REG)))
return 0;
/* Otherwise, we pass in FPRs only. Check for partial copies. */
passed_in_gprs = false;
if (cum->fregno + n_elts * n_fpreg > FP_ARG_MAX_REG + 1)
{
/* Compute number of bytes / words passed in FPRs. If there
is still space available in the register parameter area
*after* that amount, a part of the argument will be passed
in GPRs. In that case, the total amount passed in any
registers is equal to the amount that would have been passed
in GPRs if everything were passed there, so we fall back to
the GPR code below to compute the appropriate value. */
int fpr = ((FP_ARG_MAX_REG + 1 - cum->fregno)
* MIN (8, GET_MODE_SIZE (elt_mode)));
int fpr_words = fpr / (TARGET_32BIT ? 4 : 8);
if (align_words + fpr_words < GP_ARG_NUM_REG)
passed_in_gprs = true;
else
ret = fpr;
}
}
if (passed_in_gprs
&& align_words < GP_ARG_NUM_REG
&& GP_ARG_NUM_REG < align_words + rs6000_arg_size (mode, type))
ret = (GP_ARG_NUM_REG - align_words) * (TARGET_32BIT ? 4 : 8);
if (ret != 0 && TARGET_DEBUG_ARG)
fprintf (stderr, "rs6000_arg_partial_bytes: %d\n", ret);
return ret;
}
/* A C expression that indicates when an argument must be passed by
reference. If nonzero for an argument, a copy of that argument is
made in memory and a pointer to the argument is passed instead of
the argument itself. The pointer is passed in whatever way is
appropriate for passing a pointer to that type.
Under V.4, aggregates and long double are passed by reference.
As an extension to all 32-bit ABIs, AltiVec vectors are passed by
reference unless the AltiVec vector extension ABI is in force.
As an extension to all ABIs, variable sized types are passed by
reference. */
static bool
rs6000_pass_by_reference (cumulative_args_t cum ATTRIBUTE_UNUSED,
machine_mode mode, const_tree type,
bool named ATTRIBUTE_UNUSED)
{
if (!type)
return 0;
if (DEFAULT_ABI == ABI_V4 && TARGET_IEEEQUAD
&& FLOAT128_IEEE_P (TYPE_MODE (type)))
{
if (TARGET_DEBUG_ARG)
fprintf (stderr, "function_arg_pass_by_reference: V4 IEEE 128-bit\n");
return 1;
}
if (DEFAULT_ABI == ABI_V4 && AGGREGATE_TYPE_P (type))
{
if (TARGET_DEBUG_ARG)
fprintf (stderr, "function_arg_pass_by_reference: V4 aggregate\n");
return 1;
}
if (int_size_in_bytes (type) < 0)
{
if (TARGET_DEBUG_ARG)
fprintf (stderr, "function_arg_pass_by_reference: variable size\n");
return 1;
}
/* Allow -maltivec -mabi=no-altivec without warning. Altivec vector
modes only exist for GCC vector types if -maltivec. */
if (TARGET_32BIT && !TARGET_ALTIVEC_ABI && ALTIVEC_VECTOR_MODE (mode))
{
if (TARGET_DEBUG_ARG)
fprintf (stderr, "function_arg_pass_by_reference: AltiVec\n");
return 1;
}
/* Pass synthetic vectors in memory. */
if (TREE_CODE (type) == VECTOR_TYPE
&& int_size_in_bytes (type) > (TARGET_ALTIVEC_ABI ? 16 : 8))
{
static bool warned_for_pass_big_vectors = false;
if (TARGET_DEBUG_ARG)
fprintf (stderr, "function_arg_pass_by_reference: synthetic vector\n");
if (!warned_for_pass_big_vectors)
{
warning (OPT_Wpsabi, "GCC vector passed by reference: "
"non-standard ABI extension with no compatibility "
"guarantee");
warned_for_pass_big_vectors = true;
}
return 1;
}
return 0;
}
/* Process parameter of type TYPE after ARGS_SO_FAR parameters were
already processes. Return true if the parameter must be passed
(fully or partially) on the stack. */
static bool
rs6000_parm_needs_stack (cumulative_args_t args_so_far, tree type)
{
machine_mode mode;
int unsignedp;
rtx entry_parm;
/* Catch errors. */
if (type == NULL || type == error_mark_node)
return true;
/* Handle types with no storage requirement. */
if (TYPE_MODE (type) == VOIDmode)
return false;
/* Handle complex types. */
if (TREE_CODE (type) == COMPLEX_TYPE)
return (rs6000_parm_needs_stack (args_so_far, TREE_TYPE (type))
|| rs6000_parm_needs_stack (args_so_far, TREE_TYPE (type)));
/* Handle transparent aggregates. */
if ((TREE_CODE (type) == UNION_TYPE || TREE_CODE (type) == RECORD_TYPE)
&& TYPE_TRANSPARENT_AGGR (type))
type = TREE_TYPE (first_field (type));
/* See if this arg was passed by invisible reference. */
if (pass_by_reference (get_cumulative_args (args_so_far),
TYPE_MODE (type), type, true))
type = build_pointer_type (type);
/* Find mode as it is passed by the ABI. */
unsignedp = TYPE_UNSIGNED (type);
mode = promote_mode (type, TYPE_MODE (type), &unsignedp);
/* If we must pass in stack, we need a stack. */
if (rs6000_must_pass_in_stack (mode, type))
return true;
/* If there is no incoming register, we need a stack. */
entry_parm = rs6000_function_arg (args_so_far, mode, type, true);
if (entry_parm == NULL)
return true;
/* Likewise if we need to pass both in registers and on the stack. */
if (GET_CODE (entry_parm) == PARALLEL
&& XEXP (XVECEXP (entry_parm, 0, 0), 0) == NULL_RTX)
return true;
/* Also true if we're partially in registers and partially not. */
if (rs6000_arg_partial_bytes (args_so_far, mode, type, true) != 0)
return true;
/* Update info on where next arg arrives in registers. */
rs6000_function_arg_advance (args_so_far, mode, type, true);
return false;
}
/* Return true if FUN has no prototype, has a variable argument
list, or passes any parameter in memory. */
static bool
rs6000_function_parms_need_stack (tree fun, bool incoming)
{
tree fntype, result;
CUMULATIVE_ARGS args_so_far_v;
cumulative_args_t args_so_far;
if (!fun)
/* Must be a libcall, all of which only use reg parms. */
return false;
fntype = fun;
if (!TYPE_P (fun))
fntype = TREE_TYPE (fun);
/* Varargs functions need the parameter save area. */
if ((!incoming && !prototype_p (fntype)) || stdarg_p (fntype))
return true;
INIT_CUMULATIVE_INCOMING_ARGS (args_so_far_v, fntype, NULL_RTX);
args_so_far = pack_cumulative_args (&args_so_far_v);
/* When incoming, we will have been passed the function decl.
It is necessary to use the decl to handle K&R style functions,
where TYPE_ARG_TYPES may not be available. */
if (incoming)
{
gcc_assert (DECL_P (fun));
result = DECL_RESULT (fun);
}
else
result = TREE_TYPE (fntype);
if (result && aggregate_value_p (result, fntype))
{
if (!TYPE_P (result))
result = TREE_TYPE (result);
result = build_pointer_type (result);
rs6000_parm_needs_stack (args_so_far, result);
}
if (incoming)
{
tree parm;
for (parm = DECL_ARGUMENTS (fun);
parm && parm != void_list_node;
parm = TREE_CHAIN (parm))
if (rs6000_parm_needs_stack (args_so_far, TREE_TYPE (parm)))
return true;
}
else
{
function_args_iterator args_iter;
tree arg_type;
FOREACH_FUNCTION_ARGS (fntype, arg_type, args_iter)
if (rs6000_parm_needs_stack (args_so_far, arg_type))
return true;
}
return false;
}
/* Return the size of the REG_PARM_STACK_SPACE are for FUN. This is
usually a constant depending on the ABI. However, in the ELFv2 ABI
the register parameter area is optional when calling a function that
has a prototype is scope, has no variable argument list, and passes
all parameters in registers. */
int
rs6000_reg_parm_stack_space (tree fun, bool incoming)
{
int reg_parm_stack_space;
switch (DEFAULT_ABI)
{
default:
reg_parm_stack_space = 0;
break;
case ABI_AIX:
case ABI_DARWIN:
reg_parm_stack_space = TARGET_64BIT ? 64 : 32;
break;
case ABI_ELFv2:
/* ??? Recomputing this every time is a bit expensive. Is there
a place to cache this information? */
if (rs6000_function_parms_need_stack (fun, incoming))
reg_parm_stack_space = TARGET_64BIT ? 64 : 32;
else
reg_parm_stack_space = 0;
break;
}
return reg_parm_stack_space;
}
static void
rs6000_move_block_from_reg (int regno, rtx x, int nregs)
{
int i;
machine_mode reg_mode = TARGET_32BIT ? SImode : DImode;
if (nregs == 0)
return;
for (i = 0; i < nregs; i++)
{
rtx tem = adjust_address_nv (x, reg_mode, i * GET_MODE_SIZE (reg_mode));
if (reload_completed)
{
if (! strict_memory_address_p (reg_mode, XEXP (tem, 0)))
tem = NULL_RTX;
else
tem = simplify_gen_subreg (reg_mode, x, BLKmode,
i * GET_MODE_SIZE (reg_mode));
}
else
tem = replace_equiv_address (tem, XEXP (tem, 0));
gcc_assert (tem);
emit_move_insn (tem, gen_rtx_REG (reg_mode, regno + i));
}
}
/* Perform any needed actions needed for a function that is receiving a
variable number of arguments.
CUM is as above.
MODE and TYPE are the mode and type of the current parameter.
PRETEND_SIZE is a variable that should be set to the amount of stack
that must be pushed by the prolog to pretend that our caller pushed
it.
Normally, this macro will push all remaining incoming registers on the
stack and set PRETEND_SIZE to the length of the registers pushed. */
static void
setup_incoming_varargs (cumulative_args_t cum, machine_mode mode,
tree type, int *pretend_size ATTRIBUTE_UNUSED,
int no_rtl)
{
CUMULATIVE_ARGS next_cum;
int reg_size = TARGET_32BIT ? 4 : 8;
rtx save_area = NULL_RTX, mem;
int first_reg_offset;
alias_set_type set;
/* Skip the last named argument. */
next_cum = *get_cumulative_args (cum);
rs6000_function_arg_advance_1 (&next_cum, mode, type, true, 0);
if (DEFAULT_ABI == ABI_V4)
{
first_reg_offset = next_cum.sysv_gregno - GP_ARG_MIN_REG;
if (! no_rtl)
{
int gpr_reg_num = 0, gpr_size = 0, fpr_size = 0;
HOST_WIDE_INT offset = 0;
/* Try to optimize the size of the varargs save area.
The ABI requires that ap.reg_save_area is doubleword
aligned, but we don't need to allocate space for all
the bytes, only those to which we actually will save
anything. */
if (cfun->va_list_gpr_size && first_reg_offset < GP_ARG_NUM_REG)
gpr_reg_num = GP_ARG_NUM_REG - first_reg_offset;
if (TARGET_HARD_FLOAT
&& next_cum.fregno <= FP_ARG_V4_MAX_REG
&& cfun->va_list_fpr_size)
{
if (gpr_reg_num)
fpr_size = (next_cum.fregno - FP_ARG_MIN_REG)
* UNITS_PER_FP_WORD;
if (cfun->va_list_fpr_size
< FP_ARG_V4_MAX_REG + 1 - next_cum.fregno)
fpr_size += cfun->va_list_fpr_size * UNITS_PER_FP_WORD;
else
fpr_size += (FP_ARG_V4_MAX_REG + 1 - next_cum.fregno)
* UNITS_PER_FP_WORD;
}
if (gpr_reg_num)
{
offset = -((first_reg_offset * reg_size) & ~7);
if (!fpr_size && gpr_reg_num > cfun->va_list_gpr_size)
{
gpr_reg_num = cfun->va_list_gpr_size;
if (reg_size == 4 && (first_reg_offset & 1))
gpr_reg_num++;
}
gpr_size = (gpr_reg_num * reg_size + 7) & ~7;
}
else if (fpr_size)
offset = - (int) (next_cum.fregno - FP_ARG_MIN_REG)
* UNITS_PER_FP_WORD
- (int) (GP_ARG_NUM_REG * reg_size);
if (gpr_size + fpr_size)
{
rtx reg_save_area
= assign_stack_local (BLKmode, gpr_size + fpr_size, 64);
gcc_assert (GET_CODE (reg_save_area) == MEM);
reg_save_area = XEXP (reg_save_area, 0);
if (GET_CODE (reg_save_area) == PLUS)
{
gcc_assert (XEXP (reg_save_area, 0)
== virtual_stack_vars_rtx);
gcc_assert (GET_CODE (XEXP (reg_save_area, 1)) == CONST_INT);
offset += INTVAL (XEXP (reg_save_area, 1));
}
else
gcc_assert (reg_save_area == virtual_stack_vars_rtx);
}
cfun->machine->varargs_save_offset = offset;
save_area = plus_constant (Pmode, virtual_stack_vars_rtx, offset);
}
}
else
{
first_reg_offset = next_cum.words;
save_area = crtl->args.internal_arg_pointer;
if (targetm.calls.must_pass_in_stack (mode, type))
first_reg_offset += rs6000_arg_size (TYPE_MODE (type), type);
}
set = get_varargs_alias_set ();
if (! no_rtl && first_reg_offset < GP_ARG_NUM_REG
&& cfun->va_list_gpr_size)
{
int n_gpr, nregs = GP_ARG_NUM_REG - first_reg_offset;
if (va_list_gpr_counter_field)
/* V4 va_list_gpr_size counts number of registers needed. */
n_gpr = cfun->va_list_gpr_size;
else
/* char * va_list instead counts number of bytes needed. */
n_gpr = (cfun->va_list_gpr_size + reg_size - 1) / reg_size;
if (nregs > n_gpr)
nregs = n_gpr;
mem = gen_rtx_MEM (BLKmode,
plus_constant (Pmode, save_area,
first_reg_offset * reg_size));
MEM_NOTRAP_P (mem) = 1;
set_mem_alias_set (mem, set);
set_mem_align (mem, BITS_PER_WORD);
rs6000_move_block_from_reg (GP_ARG_MIN_REG + first_reg_offset, mem,
nregs);
}
/* Save FP registers if needed. */
if (DEFAULT_ABI == ABI_V4
&& TARGET_HARD_FLOAT
&& ! no_rtl
&& next_cum.fregno <= FP_ARG_V4_MAX_REG
&& cfun->va_list_fpr_size)
{
int fregno = next_cum.fregno, nregs;
rtx cr1 = gen_rtx_REG (CCmode, CR1_REGNO);
rtx lab = gen_label_rtx ();
int off = (GP_ARG_NUM_REG * reg_size) + ((fregno - FP_ARG_MIN_REG)
* UNITS_PER_FP_WORD);
emit_jump_insn
(gen_rtx_SET (pc_rtx,
gen_rtx_IF_THEN_ELSE (VOIDmode,
gen_rtx_NE (VOIDmode, cr1,
const0_rtx),
gen_rtx_LABEL_REF (VOIDmode, lab),
pc_rtx)));
for (nregs = 0;
fregno <= FP_ARG_V4_MAX_REG && nregs < cfun->va_list_fpr_size;
fregno++, off += UNITS_PER_FP_WORD, nregs++)
{
mem = gen_rtx_MEM ((TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT)
? DFmode : SFmode,
plus_constant (Pmode, save_area, off));
MEM_NOTRAP_P (mem) = 1;
set_mem_alias_set (mem, set);
set_mem_align (mem, GET_MODE_ALIGNMENT (
(TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT)
? DFmode : SFmode));
emit_move_insn (mem, gen_rtx_REG (
(TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT)
? DFmode : SFmode, fregno));
}
emit_label (lab);
}
}
/* Create the va_list data type. */
static tree
rs6000_build_builtin_va_list (void)
{
tree f_gpr, f_fpr, f_res, f_ovf, f_sav, record, type_decl;
/* For AIX, prefer 'char *' because that's what the system
header files like. */
if (DEFAULT_ABI != ABI_V4)
return build_pointer_type (char_type_node);
record = (*lang_hooks.types.make_type) (RECORD_TYPE);
type_decl = build_decl (BUILTINS_LOCATION, TYPE_DECL,
get_identifier ("__va_list_tag"), record);
f_gpr = build_decl (BUILTINS_LOCATION, FIELD_DECL, get_identifier ("gpr"),
unsigned_char_type_node);
f_fpr = build_decl (BUILTINS_LOCATION, FIELD_DECL, get_identifier ("fpr"),
unsigned_char_type_node);
/* Give the two bytes of padding a name, so that -Wpadded won't warn on
every user file. */
f_res = build_decl (BUILTINS_LOCATION, FIELD_DECL,
get_identifier ("reserved"), short_unsigned_type_node);
f_ovf = build_decl (BUILTINS_LOCATION, FIELD_DECL,
get_identifier ("overflow_arg_area"),
ptr_type_node);
f_sav = build_decl (BUILTINS_LOCATION, FIELD_DECL,
get_identifier ("reg_save_area"),
ptr_type_node);
va_list_gpr_counter_field = f_gpr;
va_list_fpr_counter_field = f_fpr;
DECL_FIELD_CONTEXT (f_gpr) = record;
DECL_FIELD_CONTEXT (f_fpr) = record;
DECL_FIELD_CONTEXT (f_res) = record;
DECL_FIELD_CONTEXT (f_ovf) = record;
DECL_FIELD_CONTEXT (f_sav) = record;
TYPE_STUB_DECL (record) = type_decl;
TYPE_NAME (record) = type_decl;
TYPE_FIELDS (record) = f_gpr;
DECL_CHAIN (f_gpr) = f_fpr;
DECL_CHAIN (f_fpr) = f_res;
DECL_CHAIN (f_res) = f_ovf;
DECL_CHAIN (f_ovf) = f_sav;
layout_type (record);
/* The correct type is an array type of one element. */
return build_array_type (record, build_index_type (size_zero_node));
}
/* Implement va_start. */
static void
rs6000_va_start (tree valist, rtx nextarg)
{
HOST_WIDE_INT words, n_gpr, n_fpr;
tree f_gpr, f_fpr, f_res, f_ovf, f_sav;
tree gpr, fpr, ovf, sav, t;
/* Only SVR4 needs something special. */
if (DEFAULT_ABI != ABI_V4)
{
std_expand_builtin_va_start (valist, nextarg);
return;
}
f_gpr = TYPE_FIELDS (TREE_TYPE (va_list_type_node));
f_fpr = DECL_CHAIN (f_gpr);
f_res = DECL_CHAIN (f_fpr);
f_ovf = DECL_CHAIN (f_res);
f_sav = DECL_CHAIN (f_ovf);
valist = build_simple_mem_ref (valist);
gpr = build3 (COMPONENT_REF, TREE_TYPE (f_gpr), valist, f_gpr, NULL_TREE);
fpr = build3 (COMPONENT_REF, TREE_TYPE (f_fpr), unshare_expr (valist),
f_fpr, NULL_TREE);
ovf = build3 (COMPONENT_REF, TREE_TYPE (f_ovf), unshare_expr (valist),
f_ovf, NULL_TREE);
sav = build3 (COMPONENT_REF, TREE_TYPE (f_sav), unshare_expr (valist),
f_sav, NULL_TREE);
/* Count number of gp and fp argument registers used. */
words = crtl->args.info.words;
n_gpr = MIN (crtl->args.info.sysv_gregno - GP_ARG_MIN_REG,
GP_ARG_NUM_REG);
n_fpr = MIN (crtl->args.info.fregno - FP_ARG_MIN_REG,
FP_ARG_NUM_REG);
if (TARGET_DEBUG_ARG)
fprintf (stderr, "va_start: words = " HOST_WIDE_INT_PRINT_DEC", n_gpr = "
HOST_WIDE_INT_PRINT_DEC", n_fpr = " HOST_WIDE_INT_PRINT_DEC"\n",
words, n_gpr, n_fpr);
if (cfun->va_list_gpr_size)
{
t = build2 (MODIFY_EXPR, TREE_TYPE (gpr), gpr,
build_int_cst (NULL_TREE, n_gpr));
TREE_SIDE_EFFECTS (t) = 1;
expand_expr (t, const0_rtx, VOIDmode, EXPAND_NORMAL);
}
if (cfun->va_list_fpr_size)
{
t = build2 (MODIFY_EXPR, TREE_TYPE (fpr), fpr,
build_int_cst (NULL_TREE, n_fpr));
TREE_SIDE_EFFECTS (t) = 1;
expand_expr (t, const0_rtx, VOIDmode, EXPAND_NORMAL);
#ifdef HAVE_AS_GNU_ATTRIBUTE
if (call_ABI_of_interest (cfun->decl))
rs6000_passes_float = true;
#endif
}
/* Find the overflow area. */
t = make_tree (TREE_TYPE (ovf), crtl->args.internal_arg_pointer);
if (words != 0)
t = fold_build_pointer_plus_hwi (t, words * MIN_UNITS_PER_WORD);
t = build2 (MODIFY_EXPR, TREE_TYPE (ovf), ovf, t);
TREE_SIDE_EFFECTS (t) = 1;
expand_expr (t, const0_rtx, VOIDmode, EXPAND_NORMAL);
/* If there were no va_arg invocations, don't set up the register
save area. */
if (!cfun->va_list_gpr_size
&& !cfun->va_list_fpr_size
&& n_gpr < GP_ARG_NUM_REG
&& n_fpr < FP_ARG_V4_MAX_REG)
return;
/* Find the register save area. */
t = make_tree (TREE_TYPE (sav), virtual_stack_vars_rtx);
if (cfun->machine->varargs_save_offset)
t = fold_build_pointer_plus_hwi (t, cfun->machine->varargs_save_offset);
t = build2 (MODIFY_EXPR, TREE_TYPE (sav), sav, t);
TREE_SIDE_EFFECTS (t) = 1;
expand_expr (t, const0_rtx, VOIDmode, EXPAND_NORMAL);
}
/* Implement va_arg. */
static tree
rs6000_gimplify_va_arg (tree valist, tree type, gimple_seq *pre_p,
gimple_seq *post_p)
{
tree f_gpr, f_fpr, f_res, f_ovf, f_sav;
tree gpr, fpr, ovf, sav, reg, t, u;
int size, rsize, n_reg, sav_ofs, sav_scale;
tree lab_false, lab_over, addr;
int align;
tree ptrtype = build_pointer_type_for_mode (type, ptr_mode, true);
int regalign = 0;
gimple *stmt;
if (pass_by_reference (NULL, TYPE_MODE (type), type, false))
{
t = rs6000_gimplify_va_arg (valist, ptrtype, pre_p, post_p);
return build_va_arg_indirect_ref (t);
}
/* We need to deal with the fact that the darwin ppc64 ABI is defined by an
earlier version of gcc, with the property that it always applied alignment
adjustments to the va-args (even for zero-sized types). The cheapest way
to deal with this is to replicate the effect of the part of
std_gimplify_va_arg_expr that carries out the align adjust, for the case
of relevance.
We don't need to check for pass-by-reference because of the test above.
We can return a simplifed answer, since we know there's no offset to add. */
if (((TARGET_MACHO
&& rs6000_darwin64_abi)
|| DEFAULT_ABI == ABI_ELFv2
|| (DEFAULT_ABI == ABI_AIX && !rs6000_compat_align_parm))
&& integer_zerop (TYPE_SIZE (type)))
{
unsigned HOST_WIDE_INT align, boundary;
tree valist_tmp = get_initialized_tmp_var (valist, pre_p, NULL);
align = PARM_BOUNDARY / BITS_PER_UNIT;
boundary = rs6000_function_arg_boundary (TYPE_MODE (type), type);
if (boundary > MAX_SUPPORTED_STACK_ALIGNMENT)
boundary = MAX_SUPPORTED_STACK_ALIGNMENT;
boundary /= BITS_PER_UNIT;
if (boundary > align)
{
tree t ;
/* This updates arg ptr by the amount that would be necessary
to align the zero-sized (but not zero-alignment) item. */
t = build2 (MODIFY_EXPR, TREE_TYPE (valist), valist_tmp,
fold_build_pointer_plus_hwi (valist_tmp, boundary - 1));
gimplify_and_add (t, pre_p);
t = fold_convert (sizetype, valist_tmp);
t = build2 (MODIFY_EXPR, TREE_TYPE (valist), valist_tmp,
fold_convert (TREE_TYPE (valist),
fold_build2 (BIT_AND_EXPR, sizetype, t,
size_int (-boundary))));
t = build2 (MODIFY_EXPR, TREE_TYPE (valist), valist, t);
gimplify_and_add (t, pre_p);
}
/* Since it is zero-sized there's no increment for the item itself. */
valist_tmp = fold_convert (build_pointer_type (type), valist_tmp);
return build_va_arg_indirect_ref (valist_tmp);
}
if (DEFAULT_ABI != ABI_V4)
{
if (targetm.calls.split_complex_arg && TREE_CODE (type) == COMPLEX_TYPE)
{
tree elem_type = TREE_TYPE (type);
machine_mode elem_mode = TYPE_MODE (elem_type);
int elem_size = GET_MODE_SIZE (elem_mode);
if (elem_size < UNITS_PER_WORD)
{
tree real_part, imag_part;
gimple_seq post = NULL;
real_part = rs6000_gimplify_va_arg (valist, elem_type, pre_p,
&post);
/* Copy the value into a temporary, lest the formal temporary
be reused out from under us. */
real_part = get_initialized_tmp_var (real_part, pre_p, &post);
gimple_seq_add_seq (pre_p, post);
imag_part = rs6000_gimplify_va_arg (valist, elem_type, pre_p,
post_p);
return build2 (COMPLEX_EXPR, type, real_part, imag_part);
}
}
return std_gimplify_va_arg_expr (valist, type, pre_p, post_p);
}
f_gpr = TYPE_FIELDS (TREE_TYPE (va_list_type_node));
f_fpr = DECL_CHAIN (f_gpr);
f_res = DECL_CHAIN (f_fpr);
f_ovf = DECL_CHAIN (f_res);
f_sav = DECL_CHAIN (f_ovf);
gpr = build3 (COMPONENT_REF, TREE_TYPE (f_gpr), valist, f_gpr, NULL_TREE);
fpr = build3 (COMPONENT_REF, TREE_TYPE (f_fpr), unshare_expr (valist),
f_fpr, NULL_TREE);
ovf = build3 (COMPONENT_REF, TREE_TYPE (f_ovf), unshare_expr (valist),
f_ovf, NULL_TREE);
sav = build3 (COMPONENT_REF, TREE_TYPE (f_sav), unshare_expr (valist),
f_sav, NULL_TREE);
size = int_size_in_bytes (type);
rsize = (size + 3) / 4;
int pad = 4 * rsize - size;
align = 1;
machine_mode mode = TYPE_MODE (type);
if (abi_v4_pass_in_fpr (mode))
{
/* FP args go in FP registers, if present. */
reg = fpr;
n_reg = (size + 7) / 8;
sav_ofs = ((TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT) ? 8 : 4) * 4;
sav_scale = ((TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT) ? 8 : 4);
if (mode != SFmode && mode != SDmode)
align = 8;
}
else
{
/* Otherwise into GP registers. */
reg = gpr;
n_reg = rsize;
sav_ofs = 0;
sav_scale = 4;
if (n_reg == 2)
align = 8;
}
/* Pull the value out of the saved registers.... */
lab_over = NULL;
addr = create_tmp_var (ptr_type_node, "addr");
/* AltiVec vectors never go in registers when -mabi=altivec. */
if (TARGET_ALTIVEC_ABI && ALTIVEC_VECTOR_MODE (mode))
align = 16;
else
{
lab_false = create_artificial_label (input_location);
lab_over = create_artificial_label (input_location);
/* Long long is aligned in the registers. As are any other 2 gpr
item such as complex int due to a historical mistake. */
u = reg;
if (n_reg == 2 && reg == gpr)
{
regalign = 1;
u = build2 (BIT_AND_EXPR, TREE_TYPE (reg), unshare_expr (reg),
build_int_cst (TREE_TYPE (reg), n_reg - 1));
u = build2 (POSTINCREMENT_EXPR, TREE_TYPE (reg),
unshare_expr (reg), u);
}
/* _Decimal128 is passed in even/odd fpr pairs; the stored
reg number is 0 for f1, so we want to make it odd. */
else if (reg == fpr && mode == TDmode)
{
t = build2 (BIT_IOR_EXPR, TREE_TYPE (reg), unshare_expr (reg),
build_int_cst (TREE_TYPE (reg), 1));
u = build2 (MODIFY_EXPR, void_type_node, unshare_expr (reg), t);
}
t = fold_convert (TREE_TYPE (reg), size_int (8 - n_reg + 1));
t = build2 (GE_EXPR, boolean_type_node, u, t);
u = build1 (GOTO_EXPR, void_type_node, lab_false);
t = build3 (COND_EXPR, void_type_node, t, u, NULL_TREE);
gimplify_and_add (t, pre_p);
t = sav;
if (sav_ofs)
t = fold_build_pointer_plus_hwi (sav, sav_ofs);
u = build2 (POSTINCREMENT_EXPR, TREE_TYPE (reg), unshare_expr (reg),
build_int_cst (TREE_TYPE (reg), n_reg));
u = fold_convert (sizetype, u);
u = build2 (MULT_EXPR, sizetype, u, size_int (sav_scale));
t = fold_build_pointer_plus (t, u);
/* _Decimal32 varargs are located in the second word of the 64-bit
FP register for 32-bit binaries. */
if (TARGET_32BIT && TARGET_HARD_FLOAT && mode == SDmode)
t = fold_build_pointer_plus_hwi (t, size);
/* Args are passed right-aligned. */
if (BYTES_BIG_ENDIAN)
t = fold_build_pointer_plus_hwi (t, pad);
gimplify_assign (addr, t, pre_p);
gimple_seq_add_stmt (pre_p, gimple_build_goto (lab_over));
stmt = gimple_build_label (lab_false);
gimple_seq_add_stmt (pre_p, stmt);
if ((n_reg == 2 && !regalign) || n_reg > 2)
{
/* Ensure that we don't find any more args in regs.
Alignment has taken care of for special cases. */
gimplify_assign (reg, build_int_cst (TREE_TYPE (reg), 8), pre_p);
}
}
/* ... otherwise out of the overflow area. */
/* Care for on-stack alignment if needed. */
t = ovf;
if (align != 1)
{
t = fold_build_pointer_plus_hwi (t, align - 1);
t = build2 (BIT_AND_EXPR, TREE_TYPE (t), t,
build_int_cst (TREE_TYPE (t), -align));
}
/* Args are passed right-aligned. */
if (BYTES_BIG_ENDIAN)
t = fold_build_pointer_plus_hwi (t, pad);
gimplify_expr (&t, pre_p, NULL, is_gimple_val, fb_rvalue);
gimplify_assign (unshare_expr (addr), t, pre_p);
t = fold_build_pointer_plus_hwi (t, size);
gimplify_assign (unshare_expr (ovf), t, pre_p);
if (lab_over)
{
stmt = gimple_build_label (lab_over);
gimple_seq_add_stmt (pre_p, stmt);
}
if (STRICT_ALIGNMENT
&& (TYPE_ALIGN (type)
> (unsigned) BITS_PER_UNIT * (align < 4 ? 4 : align)))
{
/* The value (of type complex double, for example) may not be
aligned in memory in the saved registers, so copy via a
temporary. (This is the same code as used for SPARC.) */
tree tmp = create_tmp_var (type, "va_arg_tmp");
tree dest_addr = build_fold_addr_expr (tmp);
tree copy = build_call_expr (builtin_decl_implicit (BUILT_IN_MEMCPY),
3, dest_addr, addr, size_int (rsize * 4));
gimplify_and_add (copy, pre_p);
addr = dest_addr;
}
addr = fold_convert (ptrtype, addr);
return build_va_arg_indirect_ref (addr);
}
/* Builtins. */
static void
def_builtin (const char *name, tree type, enum rs6000_builtins code)
{
tree t;
unsigned classify = rs6000_builtin_info[(int)code].attr;
const char *attr_string = "";
gcc_assert (name != NULL);
gcc_assert (IN_RANGE ((int)code, 0, (int)RS6000_BUILTIN_COUNT));
if (rs6000_builtin_decls[(int)code])
fatal_error (input_location,
"internal error: builtin function %qs already processed",
name);
rs6000_builtin_decls[(int)code] = t =
add_builtin_function (name, type, (int)code, BUILT_IN_MD, NULL, NULL_TREE);
/* Set any special attributes. */
if ((classify & RS6000_BTC_CONST) != 0)
{
/* const function, function only depends on the inputs. */
TREE_READONLY (t) = 1;
TREE_NOTHROW (t) = 1;
attr_string = ", const";
}
else if ((classify & RS6000_BTC_PURE) != 0)
{
/* pure function, function can read global memory, but does not set any
external state. */
DECL_PURE_P (t) = 1;
TREE_NOTHROW (t) = 1;
attr_string = ", pure";
}
else if ((classify & RS6000_BTC_FP) != 0)
{
/* Function is a math function. If rounding mode is on, then treat the
function as not reading global memory, but it can have arbitrary side
effects. If it is off, then assume the function is a const function.
This mimics the ATTR_MATHFN_FPROUNDING attribute in
builtin-attribute.def that is used for the math functions. */
TREE_NOTHROW (t) = 1;
if (flag_rounding_math)
{
DECL_PURE_P (t) = 1;
DECL_IS_NOVOPS (t) = 1;
attr_string = ", fp, pure";
}
else
{
TREE_READONLY (t) = 1;
attr_string = ", fp, const";
}
}
else if ((classify & RS6000_BTC_ATTR_MASK) != 0)
gcc_unreachable ();
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin, code = %4d, %s%s\n",
(int)code, name, attr_string);
}
/* Simple ternary operations: VECd = foo (VECa, VECb, VECc). */
#undef RS6000_BUILTIN_0
#undef RS6000_BUILTIN_1
#undef RS6000_BUILTIN_2
#undef RS6000_BUILTIN_3
#undef RS6000_BUILTIN_A
#undef RS6000_BUILTIN_D
#undef RS6000_BUILTIN_H
#undef RS6000_BUILTIN_P
#undef RS6000_BUILTIN_Q
#undef RS6000_BUILTIN_X
#define RS6000_BUILTIN_0(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_1(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_2(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_3(ENUM, NAME, MASK, ATTR, ICODE) \
{ MASK, ICODE, NAME, ENUM },
#define RS6000_BUILTIN_A(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_D(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_H(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_P(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_Q(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_X(ENUM, NAME, MASK, ATTR, ICODE)
static const struct builtin_description bdesc_3arg[] =
{
#include "rs6000-builtin.def"
};
/* DST operations: void foo (void *, const int, const char). */
#undef RS6000_BUILTIN_0
#undef RS6000_BUILTIN_1
#undef RS6000_BUILTIN_2
#undef RS6000_BUILTIN_3
#undef RS6000_BUILTIN_A
#undef RS6000_BUILTIN_D
#undef RS6000_BUILTIN_H
#undef RS6000_BUILTIN_P
#undef RS6000_BUILTIN_Q
#undef RS6000_BUILTIN_X
#define RS6000_BUILTIN_0(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_1(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_2(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_3(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_A(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_D(ENUM, NAME, MASK, ATTR, ICODE) \
{ MASK, ICODE, NAME, ENUM },
#define RS6000_BUILTIN_H(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_P(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_Q(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_X(ENUM, NAME, MASK, ATTR, ICODE)
static const struct builtin_description bdesc_dst[] =
{
#include "rs6000-builtin.def"
};
/* Simple binary operations: VECc = foo (VECa, VECb). */
#undef RS6000_BUILTIN_0
#undef RS6000_BUILTIN_1
#undef RS6000_BUILTIN_2
#undef RS6000_BUILTIN_3
#undef RS6000_BUILTIN_A
#undef RS6000_BUILTIN_D
#undef RS6000_BUILTIN_H
#undef RS6000_BUILTIN_P
#undef RS6000_BUILTIN_Q
#undef RS6000_BUILTIN_X
#define RS6000_BUILTIN_0(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_1(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_2(ENUM, NAME, MASK, ATTR, ICODE) \
{ MASK, ICODE, NAME, ENUM },
#define RS6000_BUILTIN_3(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_A(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_D(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_H(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_P(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_Q(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_X(ENUM, NAME, MASK, ATTR, ICODE)
static const struct builtin_description bdesc_2arg[] =
{
#include "rs6000-builtin.def"
};
#undef RS6000_BUILTIN_0
#undef RS6000_BUILTIN_1
#undef RS6000_BUILTIN_2
#undef RS6000_BUILTIN_3
#undef RS6000_BUILTIN_A
#undef RS6000_BUILTIN_D
#undef RS6000_BUILTIN_H
#undef RS6000_BUILTIN_P
#undef RS6000_BUILTIN_Q
#undef RS6000_BUILTIN_X
#define RS6000_BUILTIN_0(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_1(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_2(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_3(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_A(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_D(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_H(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_P(ENUM, NAME, MASK, ATTR, ICODE) \
{ MASK, ICODE, NAME, ENUM },
#define RS6000_BUILTIN_Q(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_X(ENUM, NAME, MASK, ATTR, ICODE)
/* AltiVec predicates. */
static const struct builtin_description bdesc_altivec_preds[] =
{
#include "rs6000-builtin.def"
};
/* PAIRED predicates. */
#undef RS6000_BUILTIN_0
#undef RS6000_BUILTIN_1
#undef RS6000_BUILTIN_2
#undef RS6000_BUILTIN_3
#undef RS6000_BUILTIN_A
#undef RS6000_BUILTIN_D
#undef RS6000_BUILTIN_H
#undef RS6000_BUILTIN_P
#undef RS6000_BUILTIN_Q
#undef RS6000_BUILTIN_X
#define RS6000_BUILTIN_0(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_1(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_2(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_3(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_A(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_D(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_H(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_P(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_Q(ENUM, NAME, MASK, ATTR, ICODE) \
{ MASK, ICODE, NAME, ENUM },
#define RS6000_BUILTIN_X(ENUM, NAME, MASK, ATTR, ICODE)
static const struct builtin_description bdesc_paired_preds[] =
{
#include "rs6000-builtin.def"
};
/* ABS* operations. */
#undef RS6000_BUILTIN_0
#undef RS6000_BUILTIN_1
#undef RS6000_BUILTIN_2
#undef RS6000_BUILTIN_3
#undef RS6000_BUILTIN_A
#undef RS6000_BUILTIN_D
#undef RS6000_BUILTIN_H
#undef RS6000_BUILTIN_P
#undef RS6000_BUILTIN_Q
#undef RS6000_BUILTIN_X
#define RS6000_BUILTIN_0(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_1(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_2(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_3(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_A(ENUM, NAME, MASK, ATTR, ICODE) \
{ MASK, ICODE, NAME, ENUM },
#define RS6000_BUILTIN_D(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_H(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_P(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_Q(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_X(ENUM, NAME, MASK, ATTR, ICODE)
static const struct builtin_description bdesc_abs[] =
{
#include "rs6000-builtin.def"
};
/* Simple unary operations: VECb = foo (unsigned literal) or VECb =
foo (VECa). */
#undef RS6000_BUILTIN_0
#undef RS6000_BUILTIN_1
#undef RS6000_BUILTIN_2
#undef RS6000_BUILTIN_3
#undef RS6000_BUILTIN_A
#undef RS6000_BUILTIN_D
#undef RS6000_BUILTIN_H
#undef RS6000_BUILTIN_P
#undef RS6000_BUILTIN_Q
#undef RS6000_BUILTIN_X
#define RS6000_BUILTIN_0(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_1(ENUM, NAME, MASK, ATTR, ICODE) \
{ MASK, ICODE, NAME, ENUM },
#define RS6000_BUILTIN_2(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_3(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_A(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_D(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_H(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_P(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_Q(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_X(ENUM, NAME, MASK, ATTR, ICODE)
static const struct builtin_description bdesc_1arg[] =
{
#include "rs6000-builtin.def"
};
/* Simple no-argument operations: result = __builtin_darn_32 () */
#undef RS6000_BUILTIN_0
#undef RS6000_BUILTIN_1
#undef RS6000_BUILTIN_2
#undef RS6000_BUILTIN_3
#undef RS6000_BUILTIN_A
#undef RS6000_BUILTIN_D
#undef RS6000_BUILTIN_H
#undef RS6000_BUILTIN_P
#undef RS6000_BUILTIN_Q
#undef RS6000_BUILTIN_X
#define RS6000_BUILTIN_0(ENUM, NAME, MASK, ATTR, ICODE) \
{ MASK, ICODE, NAME, ENUM },
#define RS6000_BUILTIN_1(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_2(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_3(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_A(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_D(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_H(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_P(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_Q(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_X(ENUM, NAME, MASK, ATTR, ICODE)
static const struct builtin_description bdesc_0arg[] =
{
#include "rs6000-builtin.def"
};
/* HTM builtins. */
#undef RS6000_BUILTIN_0
#undef RS6000_BUILTIN_1
#undef RS6000_BUILTIN_2
#undef RS6000_BUILTIN_3
#undef RS6000_BUILTIN_A
#undef RS6000_BUILTIN_D
#undef RS6000_BUILTIN_H
#undef RS6000_BUILTIN_P
#undef RS6000_BUILTIN_Q
#undef RS6000_BUILTIN_X
#define RS6000_BUILTIN_0(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_1(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_2(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_3(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_A(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_D(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_H(ENUM, NAME, MASK, ATTR, ICODE) \
{ MASK, ICODE, NAME, ENUM },
#define RS6000_BUILTIN_P(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_Q(ENUM, NAME, MASK, ATTR, ICODE)
#define RS6000_BUILTIN_X(ENUM, NAME, MASK, ATTR, ICODE)
static const struct builtin_description bdesc_htm[] =
{
#include "rs6000-builtin.def"
};
#undef RS6000_BUILTIN_0
#undef RS6000_BUILTIN_1
#undef RS6000_BUILTIN_2
#undef RS6000_BUILTIN_3
#undef RS6000_BUILTIN_A
#undef RS6000_BUILTIN_D
#undef RS6000_BUILTIN_H
#undef RS6000_BUILTIN_P
#undef RS6000_BUILTIN_Q
/* Return true if a builtin function is overloaded. */
bool
rs6000_overloaded_builtin_p (enum rs6000_builtins fncode)
{
return (rs6000_builtin_info[(int)fncode].attr & RS6000_BTC_OVERLOADED) != 0;
}
const char *
rs6000_overloaded_builtin_name (enum rs6000_builtins fncode)
{
return rs6000_builtin_info[(int)fncode].name;
}
/* Expand an expression EXP that calls a builtin without arguments. */
static rtx
rs6000_expand_zeroop_builtin (enum insn_code icode, rtx target)
{
rtx pat;
machine_mode tmode = insn_data[icode].operand[0].mode;
if (icode == CODE_FOR_nothing)
/* Builtin not supported on this processor. */
return 0;
if (target == 0
|| GET_MODE (target) != tmode
|| ! (*insn_data[icode].operand[0].predicate) (target, tmode))
target = gen_reg_rtx (tmode);
pat = GEN_FCN (icode) (target);
if (! pat)
return 0;
emit_insn (pat);
return target;
}
static rtx
rs6000_expand_mtfsf_builtin (enum insn_code icode, tree exp)
{
rtx pat;
tree arg0 = CALL_EXPR_ARG (exp, 0);
tree arg1 = CALL_EXPR_ARG (exp, 1);
rtx op0 = expand_normal (arg0);
rtx op1 = expand_normal (arg1);
machine_mode mode0 = insn_data[icode].operand[0].mode;
machine_mode mode1 = insn_data[icode].operand[1].mode;
if (icode == CODE_FOR_nothing)
/* Builtin not supported on this processor. */
return 0;
/* If we got invalid arguments bail out before generating bad rtl. */
if (arg0 == error_mark_node || arg1 == error_mark_node)
return const0_rtx;
if (GET_CODE (op0) != CONST_INT
|| INTVAL (op0) > 255
|| INTVAL (op0) < 0)
{
error ("argument 1 must be an 8-bit field value");
return const0_rtx;
}
if (! (*insn_data[icode].operand[0].predicate) (op0, mode0))
op0 = copy_to_mode_reg (mode0, op0);
if (! (*insn_data[icode].operand[1].predicate) (op1, mode1))
op1 = copy_to_mode_reg (mode1, op1);
pat = GEN_FCN (icode) (op0, op1);
if (! pat)
return const0_rtx;
emit_insn (pat);
return NULL_RTX;
}
static rtx
rs6000_expand_unop_builtin (enum insn_code icode, tree exp, rtx target)
{
rtx pat;
tree arg0 = CALL_EXPR_ARG (exp, 0);
rtx op0 = expand_normal (arg0);
machine_mode tmode = insn_data[icode].operand[0].mode;
machine_mode mode0 = insn_data[icode].operand[1].mode;
if (icode == CODE_FOR_nothing)
/* Builtin not supported on this processor. */
return 0;
/* If we got invalid arguments bail out before generating bad rtl. */
if (arg0 == error_mark_node)
return const0_rtx;
if (icode == CODE_FOR_altivec_vspltisb
|| icode == CODE_FOR_altivec_vspltish
|| icode == CODE_FOR_altivec_vspltisw)
{
/* Only allow 5-bit *signed* literals. */
if (GET_CODE (op0) != CONST_INT
|| INTVAL (op0) > 15
|| INTVAL (op0) < -16)
{
error ("argument 1 must be a 5-bit signed literal");
return CONST0_RTX (tmode);
}
}
if (target == 0
|| GET_MODE (target) != tmode
|| ! (*insn_data[icode].operand[0].predicate) (target, tmode))
target = gen_reg_rtx (tmode);
if (! (*insn_data[icode].operand[1].predicate) (op0, mode0))
op0 = copy_to_mode_reg (mode0, op0);
pat = GEN_FCN (icode) (target, op0);
if (! pat)
return 0;
emit_insn (pat);
return target;
}
static rtx
altivec_expand_abs_builtin (enum insn_code icode, tree exp, rtx target)
{
rtx pat, scratch1, scratch2;
tree arg0 = CALL_EXPR_ARG (exp, 0);
rtx op0 = expand_normal (arg0);
machine_mode tmode = insn_data[icode].operand[0].mode;
machine_mode mode0 = insn_data[icode].operand[1].mode;
/* If we have invalid arguments, bail out before generating bad rtl. */
if (arg0 == error_mark_node)
return const0_rtx;
if (target == 0
|| GET_MODE (target) != tmode
|| ! (*insn_data[icode].operand[0].predicate) (target, tmode))
target = gen_reg_rtx (tmode);
if (! (*insn_data[icode].operand[1].predicate) (op0, mode0))
op0 = copy_to_mode_reg (mode0, op0);
scratch1 = gen_reg_rtx (mode0);
scratch2 = gen_reg_rtx (mode0);
pat = GEN_FCN (icode) (target, op0, scratch1, scratch2);
if (! pat)
return 0;
emit_insn (pat);
return target;
}
static rtx
rs6000_expand_binop_builtin (enum insn_code icode, tree exp, rtx target)
{
rtx pat;
tree arg0 = CALL_EXPR_ARG (exp, 0);
tree arg1 = CALL_EXPR_ARG (exp, 1);
rtx op0 = expand_normal (arg0);
rtx op1 = expand_normal (arg1);
machine_mode tmode = insn_data[icode].operand[0].mode;
machine_mode mode0 = insn_data[icode].operand[1].mode;
machine_mode mode1 = insn_data[icode].operand[2].mode;
if (icode == CODE_FOR_nothing)
/* Builtin not supported on this processor. */
return 0;
/* If we got invalid arguments bail out before generating bad rtl. */
if (arg0 == error_mark_node || arg1 == error_mark_node)
return const0_rtx;
if (icode == CODE_FOR_altivec_vcfux
|| icode == CODE_FOR_altivec_vcfsx
|| icode == CODE_FOR_altivec_vctsxs
|| icode == CODE_FOR_altivec_vctuxs
|| icode == CODE_FOR_altivec_vspltb
|| icode == CODE_FOR_altivec_vsplth
|| icode == CODE_FOR_altivec_vspltw)
{
/* Only allow 5-bit unsigned literals. */
STRIP_NOPS (arg1);
if (TREE_CODE (arg1) != INTEGER_CST
|| TREE_INT_CST_LOW (arg1) & ~0x1f)
{
error ("argument 2 must be a 5-bit unsigned literal");
return CONST0_RTX (tmode);
}
}
else if (icode == CODE_FOR_dfptstsfi_eq_dd
|| icode == CODE_FOR_dfptstsfi_lt_dd
|| icode == CODE_FOR_dfptstsfi_gt_dd
|| icode == CODE_FOR_dfptstsfi_unordered_dd
|| icode == CODE_FOR_dfptstsfi_eq_td
|| icode == CODE_FOR_dfptstsfi_lt_td
|| icode == CODE_FOR_dfptstsfi_gt_td
|| icode == CODE_FOR_dfptstsfi_unordered_td)
{
/* Only allow 6-bit unsigned literals. */
STRIP_NOPS (arg0);
if (TREE_CODE (arg0) != INTEGER_CST
|| !IN_RANGE (TREE_INT_CST_LOW (arg0), 0, 63))
{
error ("argument 1 must be a 6-bit unsigned literal");
return CONST0_RTX (tmode);
}
}
else if (icode == CODE_FOR_xststdcqp_kf
|| icode == CODE_FOR_xststdcqp_tf
|| icode == CODE_FOR_xststdcdp
|| icode == CODE_FOR_xststdcsp
|| icode == CODE_FOR_xvtstdcdp
|| icode == CODE_FOR_xvtstdcsp)
{
/* Only allow 7-bit unsigned literals. */
STRIP_NOPS (arg1);
if (TREE_CODE (arg1) != INTEGER_CST
|| !IN_RANGE (TREE_INT_CST_LOW (arg1), 0, 127))
{
error ("argument 2 must be a 7-bit unsigned literal");
return CONST0_RTX (tmode);
}
}
else if (icode == CODE_FOR_unpackv1ti
|| icode == CODE_FOR_unpackkf
|| icode == CODE_FOR_unpacktf
|| icode == CODE_FOR_unpackif
|| icode == CODE_FOR_unpacktd)
{
/* Only allow 1-bit unsigned literals. */
STRIP_NOPS (arg1);
if (TREE_CODE (arg1) != INTEGER_CST
|| !IN_RANGE (TREE_INT_CST_LOW (arg1), 0, 1))
{
error ("argument 2 must be a 1-bit unsigned literal");
return CONST0_RTX (tmode);
}
}
if (target == 0
|| GET_MODE (target) != tmode
|| ! (*insn_data[icode].operand[0].predicate) (target, tmode))
target = gen_reg_rtx (tmode);
if (! (*insn_data[icode].operand[1].predicate) (op0, mode0))
op0 = copy_to_mode_reg (mode0, op0);
if (! (*insn_data[icode].operand[2].predicate) (op1, mode1))
op1 = copy_to_mode_reg (mode1, op1);
pat = GEN_FCN (icode) (target, op0, op1);
if (! pat)
return 0;
emit_insn (pat);
return target;
}
static rtx
altivec_expand_predicate_builtin (enum insn_code icode, tree exp, rtx target)
{
rtx pat, scratch;
tree cr6_form = CALL_EXPR_ARG (exp, 0);
tree arg0 = CALL_EXPR_ARG (exp, 1);
tree arg1 = CALL_EXPR_ARG (exp, 2);
rtx op0 = expand_normal (arg0);
rtx op1 = expand_normal (arg1);
machine_mode tmode = SImode;
machine_mode mode0 = insn_data[icode].operand[1].mode;
machine_mode mode1 = insn_data[icode].operand[2].mode;
int cr6_form_int;
if (TREE_CODE (cr6_form) != INTEGER_CST)
{
error ("argument 1 of %qs must be a constant",
"__builtin_altivec_predicate");
return const0_rtx;
}
else
cr6_form_int = TREE_INT_CST_LOW (cr6_form);
gcc_assert (mode0 == mode1);
/* If we have invalid arguments, bail out before generating bad rtl. */
if (arg0 == error_mark_node || arg1 == error_mark_node)
return const0_rtx;
if (target == 0
|| GET_MODE (target) != tmode
|| ! (*insn_data[icode].operand[0].predicate) (target, tmode))
target = gen_reg_rtx (tmode);
if (! (*insn_data[icode].operand[1].predicate) (op0, mode0))
op0 = copy_to_mode_reg (mode0, op0);
if (! (*insn_data[icode].operand[2].predicate) (op1, mode1))
op1 = copy_to_mode_reg (mode1, op1);
/* Note that for many of the relevant operations (e.g. cmpne or
cmpeq) with float or double operands, it makes more sense for the
mode of the allocated scratch register to select a vector of
integer. But the choice to copy the mode of operand 0 was made
long ago and there are no plans to change it. */
scratch = gen_reg_rtx (mode0);
pat = GEN_FCN (icode) (scratch, op0, op1);
if (! pat)
return 0;
emit_insn (pat);
/* The vec_any* and vec_all* predicates use the same opcodes for two
different operations, but the bits in CR6 will be different
depending on what information we want. So we have to play tricks
with CR6 to get the right bits out.
If you think this is disgusting, look at the specs for the
AltiVec predicates. */
switch (cr6_form_int)
{
case 0:
emit_insn (gen_cr6_test_for_zero (target));
break;
case 1:
emit_insn (gen_cr6_test_for_zero_reverse (target));
break;
case 2:
emit_insn (gen_cr6_test_for_lt (target));
break;
case 3:
emit_insn (gen_cr6_test_for_lt_reverse (target));
break;
default:
error ("argument 1 of %qs is out of range",
"__builtin_altivec_predicate");
break;
}
return target;
}
static rtx
paired_expand_lv_builtin (enum insn_code icode, tree exp, rtx target)
{
rtx pat, addr;
tree arg0 = CALL_EXPR_ARG (exp, 0);
tree arg1 = CALL_EXPR_ARG (exp, 1);
machine_mode tmode = insn_data[icode].operand[0].mode;
machine_mode mode0 = Pmode;
machine_mode mode1 = Pmode;
rtx op0 = expand_normal (arg0);
rtx op1 = expand_normal (arg1);
if (icode == CODE_FOR_nothing)
/* Builtin not supported on this processor. */
return 0;
/* If we got invalid arguments bail out before generating bad rtl. */
if (arg0 == error_mark_node || arg1 == error_mark_node)
return const0_rtx;
if (target == 0
|| GET_MODE (target) != tmode
|| ! (*insn_data[icode].operand[0].predicate) (target, tmode))
target = gen_reg_rtx (tmode);
op1 = copy_to_mode_reg (mode1, op1);
if (op0 == const0_rtx)
{
addr = gen_rtx_MEM (tmode, op1);
}
else
{
op0 = copy_to_mode_reg (mode0, op0);
addr = gen_rtx_MEM (tmode, gen_rtx_PLUS (Pmode, op0, op1));
}
pat = GEN_FCN (icode) (target, addr);
if (! pat)
return 0;
emit_insn (pat);
return target;
}
/* Return a constant vector for use as a little-endian permute control vector
to reverse the order of elements of the given vector mode. */
static rtx
swap_selector_for_mode (machine_mode mode)
{
/* These are little endian vectors, so their elements are reversed
from what you would normally expect for a permute control vector. */
unsigned int swap2[16] = {7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8};
unsigned int swap4[16] = {3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12};
unsigned int swap8[16] = {1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14};
unsigned int swap16[16] = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
unsigned int *swaparray, i;
rtx perm[16];
switch (mode)
{
case E_V2DFmode:
case E_V2DImode:
swaparray = swap2;
break;
case E_V4SFmode:
case E_V4SImode:
swaparray = swap4;
break;
case E_V8HImode:
swaparray = swap8;
break;
case E_V16QImode:
swaparray = swap16;
break;
default:
gcc_unreachable ();
}
for (i = 0; i < 16; ++i)
perm[i] = GEN_INT (swaparray[i]);
return force_reg (V16QImode, gen_rtx_CONST_VECTOR (V16QImode, gen_rtvec_v (16, perm)));
}
rtx
swap_endian_selector_for_mode (machine_mode mode)
{
unsigned int swap1[16] = {15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0};
unsigned int swap2[16] = {7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8};
unsigned int swap4[16] = {3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12};
unsigned int swap8[16] = {1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14};
unsigned int *swaparray, i;
rtx perm[16];
switch (mode)
{
case E_V1TImode:
swaparray = swap1;
break;
case E_V2DFmode:
case E_V2DImode:
swaparray = swap2;
break;
case E_V4SFmode:
case E_V4SImode:
swaparray = swap4;
break;
case E_V8HImode:
swaparray = swap8;
break;
default:
gcc_unreachable ();
}
for (i = 0; i < 16; ++i)
perm[i] = GEN_INT (swaparray[i]);
return force_reg (V16QImode, gen_rtx_CONST_VECTOR (V16QImode,
gen_rtvec_v (16, perm)));
}
/* Generate code for an "lvxl", or "lve*x" built-in for a little endian target
with -maltivec=be specified. Issue the load followed by an element-
reversing permute. */
void
altivec_expand_lvx_be (rtx op0, rtx op1, machine_mode mode, unsigned unspec)
{
rtx tmp = gen_reg_rtx (mode);
rtx load = gen_rtx_SET (tmp, op1);
rtx lvx = gen_rtx_UNSPEC (mode, gen_rtvec (1, const0_rtx), unspec);
rtx par = gen_rtx_PARALLEL (mode, gen_rtvec (2, load, lvx));
rtx sel = swap_selector_for_mode (mode);
rtx vperm = gen_rtx_UNSPEC (mode, gen_rtvec (3, tmp, tmp, sel), UNSPEC_VPERM);
gcc_assert (REG_P (op0));
emit_insn (par);
emit_insn (gen_rtx_SET (op0, vperm));
}
/* Generate code for a "stvxl" built-in for a little endian target with
-maltivec=be specified. Issue the store preceded by an element-reversing
permute. */
void
altivec_expand_stvx_be (rtx op0, rtx op1, machine_mode mode, unsigned unspec)
{
rtx tmp = gen_reg_rtx (mode);
rtx store = gen_rtx_SET (op0, tmp);
rtx stvx = gen_rtx_UNSPEC (mode, gen_rtvec (1, const0_rtx), unspec);
rtx par = gen_rtx_PARALLEL (mode, gen_rtvec (2, store, stvx));
rtx sel = swap_selector_for_mode (mode);
rtx vperm;
gcc_assert (REG_P (op1));
vperm = gen_rtx_UNSPEC (mode, gen_rtvec (3, op1, op1, sel), UNSPEC_VPERM);
emit_insn (gen_rtx_SET (tmp, vperm));
emit_insn (par);
}
/* Generate code for a "stve*x" built-in for a little endian target with -maltivec=be
specified. Issue the store preceded by an element-reversing permute. */
void
altivec_expand_stvex_be (rtx op0, rtx op1, machine_mode mode, unsigned unspec)
{
machine_mode inner_mode = GET_MODE_INNER (mode);
rtx tmp = gen_reg_rtx (mode);
rtx stvx = gen_rtx_UNSPEC (inner_mode, gen_rtvec (1, tmp), unspec);
rtx sel = swap_selector_for_mode (mode);
rtx vperm;
gcc_assert (REG_P (op1));
vperm = gen_rtx_UNSPEC (mode, gen_rtvec (3, op1, op1, sel), UNSPEC_VPERM);
emit_insn (gen_rtx_SET (tmp, vperm));
emit_insn (gen_rtx_SET (op0, stvx));
}
static rtx
altivec_expand_lv_builtin (enum insn_code icode, tree exp, rtx target, bool blk)
{
rtx pat, addr;
tree arg0 = CALL_EXPR_ARG (exp, 0);
tree arg1 = CALL_EXPR_ARG (exp, 1);
machine_mode tmode = insn_data[icode].operand[0].mode;
machine_mode mode0 = Pmode;
machine_mode mode1 = Pmode;
rtx op0 = expand_normal (arg0);
rtx op1 = expand_normal (arg1);
if (icode == CODE_FOR_nothing)
/* Builtin not supported on this processor. */
return 0;
/* If we got invalid arguments bail out before generating bad rtl. */
if (arg0 == error_mark_node || arg1 == error_mark_node)
return const0_rtx;
if (target == 0
|| GET_MODE (target) != tmode
|| ! (*insn_data[icode].operand[0].predicate) (target, tmode))
target = gen_reg_rtx (tmode);
op1 = copy_to_mode_reg (mode1, op1);
/* For LVX, express the RTL accurately by ANDing the address with -16.
LVXL and LVE*X expand to use UNSPECs to hide their special behavior,
so the raw address is fine. */
if (icode == CODE_FOR_altivec_lvx_v2df_2op
|| icode == CODE_FOR_altivec_lvx_v2di_2op
|| icode == CODE_FOR_altivec_lvx_v4sf_2op
|| icode == CODE_FOR_altivec_lvx_v4si_2op
|| icode == CODE_FOR_altivec_lvx_v8hi_2op
|| icode == CODE_FOR_altivec_lvx_v16qi_2op)
{
rtx rawaddr;
if (op0 == const0_rtx)
rawaddr = op1;
else
{
op0 = copy_to_mode_reg (mode0, op0);
rawaddr = gen_rtx_PLUS (Pmode, op1, op0);
}
addr = gen_rtx_AND (Pmode, rawaddr, gen_rtx_CONST_INT (Pmode, -16));
addr = gen_rtx_MEM (blk ? BLKmode : tmode, addr);
/* For -maltivec=be, emit the load and follow it up with a
permute to swap the elements. */
if (!BYTES_BIG_ENDIAN && VECTOR_ELT_ORDER_BIG)
{
rtx temp = gen_reg_rtx (tmode);
emit_insn (gen_rtx_SET (temp, addr));
rtx sel = swap_selector_for_mode (tmode);
rtx vperm = gen_rtx_UNSPEC (tmode, gen_rtvec (3, temp, temp, sel),
UNSPEC_VPERM);
emit_insn (gen_rtx_SET (target, vperm));
}
else
emit_insn (gen_rtx_SET (target, addr));
}
else
{
if (op0 == const0_rtx)
addr = gen_rtx_MEM (blk ? BLKmode : tmode, op1);
else
{
op0 = copy_to_mode_reg (mode0, op0);
addr = gen_rtx_MEM (blk ? BLKmode : tmode,
gen_rtx_PLUS (Pmode, op1, op0));
}
pat = GEN_FCN (icode) (target, addr);
if (! pat)
return 0;
emit_insn (pat);
}
return target;
}
static rtx
paired_expand_stv_builtin (enum insn_code icode, tree exp)
{
tree arg0 = CALL_EXPR_ARG (exp, 0);
tree arg1 = CALL_EXPR_ARG (exp, 1);
tree arg2 = CALL_EXPR_ARG (exp, 2);
rtx op0 = expand_normal (arg0);
rtx op1 = expand_normal (arg1);
rtx op2 = expand_normal (arg2);
rtx pat, addr;
machine_mode tmode = insn_data[icode].operand[0].mode;
machine_mode mode1 = Pmode;
machine_mode mode2 = Pmode;
/* Invalid arguments. Bail before doing anything stoopid! */
if (arg0 == error_mark_node
|| arg1 == error_mark_node
|| arg2 == error_mark_node)
return const0_rtx;
if (! (*insn_data[icode].operand[1].predicate) (op0, tmode))
op0 = copy_to_mode_reg (tmode, op0);
op2 = copy_to_mode_reg (mode2, op2);
if (op1 == const0_rtx)
{
addr = gen_rtx_MEM (tmode, op2);
}
else
{
op1 = copy_to_mode_reg (mode1, op1);
addr = gen_rtx_MEM (tmode, gen_rtx_PLUS (Pmode, op1, op2));
}
pat = GEN_FCN (icode) (addr, op0);
if (pat)
emit_insn (pat);
return NULL_RTX;
}
static rtx
altivec_expand_stxvl_builtin (enum insn_code icode, tree exp)
{
rtx pat;
tree arg0 = CALL_EXPR_ARG (exp, 0);
tree arg1 = CALL_EXPR_ARG (exp, 1);
tree arg2 = CALL_EXPR_ARG (exp, 2);
rtx op0 = expand_normal (arg0);
rtx op1 = expand_normal (arg1);
rtx op2 = expand_normal (arg2);
machine_mode mode0 = insn_data[icode].operand[0].mode;
machine_mode mode1 = insn_data[icode].operand[1].mode;
machine_mode mode2 = insn_data[icode].operand[2].mode;
if (icode == CODE_FOR_nothing)
/* Builtin not supported on this processor. */
return NULL_RTX;
/* If we got invalid arguments bail out before generating bad rtl. */
if (arg0 == error_mark_node
|| arg1 == error_mark_node
|| arg2 == error_mark_node)
return NULL_RTX;
if (! (*insn_data[icode].operand[1].predicate) (op0, mode0))
op0 = copy_to_mode_reg (mode0, op0);
if (! (*insn_data[icode].operand[2].predicate) (op1, mode1))
op1 = copy_to_mode_reg (mode1, op1);
if (! (*insn_data[icode].operand[3].predicate) (op2, mode2))
op2 = copy_to_mode_reg (mode2, op2);
pat = GEN_FCN (icode) (op0, op1, op2);
if (pat)
emit_insn (pat);
return NULL_RTX;
}
static rtx
altivec_expand_stv_builtin (enum insn_code icode, tree exp)
{
tree arg0 = CALL_EXPR_ARG (exp, 0);
tree arg1 = CALL_EXPR_ARG (exp, 1);
tree arg2 = CALL_EXPR_ARG (exp, 2);
rtx op0 = expand_normal (arg0);
rtx op1 = expand_normal (arg1);
rtx op2 = expand_normal (arg2);
rtx pat, addr, rawaddr;
machine_mode tmode = insn_data[icode].operand[0].mode;
machine_mode smode = insn_data[icode].operand[1].mode;
machine_mode mode1 = Pmode;
machine_mode mode2 = Pmode;
/* Invalid arguments. Bail before doing anything stoopid! */
if (arg0 == error_mark_node
|| arg1 == error_mark_node
|| arg2 == error_mark_node)
return const0_rtx;
op2 = copy_to_mode_reg (mode2, op2);
/* For STVX, express the RTL accurately by ANDing the address with -16.
STVXL and STVE*X expand to use UNSPECs to hide their special behavior,
so the raw address is fine. */
if (icode == CODE_FOR_altivec_stvx_v2df_2op
|| icode == CODE_FOR_altivec_stvx_v2di_2op
|| icode == CODE_FOR_altivec_stvx_v4sf_2op
|| icode == CODE_FOR_altivec_stvx_v4si_2op
|| icode == CODE_FOR_altivec_stvx_v8hi_2op
|| icode == CODE_FOR_altivec_stvx_v16qi_2op)
{
if (op1 == const0_rtx)
rawaddr = op2;
else
{
op1 = copy_to_mode_reg (mode1, op1);
rawaddr = gen_rtx_PLUS (Pmode, op2, op1);
}
addr = gen_rtx_AND (Pmode, rawaddr, gen_rtx_CONST_INT (Pmode, -16));
addr = gen_rtx_MEM (tmode, addr);
op0 = copy_to_mode_reg (tmode, op0);
/* For -maltivec=be, emit a permute to swap the elements, followed
by the store. */
if (!BYTES_BIG_ENDIAN && VECTOR_ELT_ORDER_BIG)
{
rtx temp = gen_reg_rtx (tmode);
rtx sel = swap_selector_for_mode (tmode);
rtx vperm = gen_rtx_UNSPEC (tmode, gen_rtvec (3, op0, op0, sel),
UNSPEC_VPERM);
emit_insn (gen_rtx_SET (temp, vperm));
emit_insn (gen_rtx_SET (addr, temp));
}
else
emit_insn (gen_rtx_SET (addr, op0));
}
else
{
if (! (*insn_data[icode].operand[1].predicate) (op0, smode))
op0 = copy_to_mode_reg (smode, op0);
if (op1 == const0_rtx)
addr = gen_rtx_MEM (tmode, op2);
else
{
op1 = copy_to_mode_reg (mode1, op1);
addr = gen_rtx_MEM (tmode, gen_rtx_PLUS (Pmode, op2, op1));
}
pat = GEN_FCN (icode) (addr, op0);
if (pat)
emit_insn (pat);
}
return NULL_RTX;
}
/* Return the appropriate SPR number associated with the given builtin. */
static inline HOST_WIDE_INT
htm_spr_num (enum rs6000_builtins code)
{
if (code == HTM_BUILTIN_GET_TFHAR
|| code == HTM_BUILTIN_SET_TFHAR)
return TFHAR_SPR;
else if (code == HTM_BUILTIN_GET_TFIAR
|| code == HTM_BUILTIN_SET_TFIAR)
return TFIAR_SPR;
else if (code == HTM_BUILTIN_GET_TEXASR
|| code == HTM_BUILTIN_SET_TEXASR)
return TEXASR_SPR;
gcc_assert (code == HTM_BUILTIN_GET_TEXASRU
|| code == HTM_BUILTIN_SET_TEXASRU);
return TEXASRU_SPR;
}
/* Return the appropriate SPR regno associated with the given builtin. */
static inline HOST_WIDE_INT
htm_spr_regno (enum rs6000_builtins code)
{
if (code == HTM_BUILTIN_GET_TFHAR
|| code == HTM_BUILTIN_SET_TFHAR)
return TFHAR_REGNO;
else if (code == HTM_BUILTIN_GET_TFIAR
|| code == HTM_BUILTIN_SET_TFIAR)
return TFIAR_REGNO;
gcc_assert (code == HTM_BUILTIN_GET_TEXASR
|| code == HTM_BUILTIN_SET_TEXASR
|| code == HTM_BUILTIN_GET_TEXASRU
|| code == HTM_BUILTIN_SET_TEXASRU);
return TEXASR_REGNO;
}
/* Return the correct ICODE value depending on whether we are
setting or reading the HTM SPRs. */
static inline enum insn_code
rs6000_htm_spr_icode (bool nonvoid)
{
if (nonvoid)
return (TARGET_POWERPC64) ? CODE_FOR_htm_mfspr_di : CODE_FOR_htm_mfspr_si;
else
return (TARGET_POWERPC64) ? CODE_FOR_htm_mtspr_di : CODE_FOR_htm_mtspr_si;
}
/* Expand the HTM builtin in EXP and store the result in TARGET.
Store true in *EXPANDEDP if we found a builtin to expand. */
static rtx
htm_expand_builtin (tree exp, rtx target, bool * expandedp)
{
tree fndecl = TREE_OPERAND (CALL_EXPR_FN (exp), 0);
bool nonvoid = TREE_TYPE (TREE_TYPE (fndecl)) != void_type_node;
enum rs6000_builtins fcode = (enum rs6000_builtins) DECL_FUNCTION_CODE (fndecl);
const struct builtin_description *d;
size_t i;
*expandedp = true;
if (!TARGET_POWERPC64
&& (fcode == HTM_BUILTIN_TABORTDC
|| fcode == HTM_BUILTIN_TABORTDCI))
{
size_t uns_fcode = (size_t)fcode;
const char *name = rs6000_builtin_info[uns_fcode].name;
error ("builtin %qs is only valid in 64-bit mode", name);
return const0_rtx;
}
/* Expand the HTM builtins. */
d = bdesc_htm;
for (i = 0; i < ARRAY_SIZE (bdesc_htm); i++, d++)
if (d->code == fcode)
{
rtx op[MAX_HTM_OPERANDS], pat;
int nopnds = 0;
tree arg;
call_expr_arg_iterator iter;
unsigned attr = rs6000_builtin_info[fcode].attr;
enum insn_code icode = d->icode;
const struct insn_operand_data *insn_op;
bool uses_spr = (attr & RS6000_BTC_SPR);
rtx cr = NULL_RTX;
if (uses_spr)
icode = rs6000_htm_spr_icode (nonvoid);
insn_op = &insn_data[icode].operand[0];
if (nonvoid)
{
machine_mode tmode = (uses_spr) ? insn_op->mode : E_SImode;
if (!target
|| GET_MODE (target) != tmode
|| (uses_spr && !(*insn_op->predicate) (target, tmode)))
target = gen_reg_rtx (tmode);
if (uses_spr)
op[nopnds++] = target;
}
FOR_EACH_CALL_EXPR_ARG (arg, iter, exp)
{
if (arg == error_mark_node || nopnds >= MAX_HTM_OPERANDS)
return const0_rtx;
insn_op = &insn_data[icode].operand[nopnds];
op[nopnds] = expand_normal (arg);
if (!(*insn_op->predicate) (op[nopnds], insn_op->mode))
{
if (!strcmp (insn_op->constraint, "n"))
{
int arg_num = (nonvoid) ? nopnds : nopnds + 1;
if (!CONST_INT_P (op[nopnds]))
error ("argument %d must be an unsigned literal", arg_num);
else
error ("argument %d is an unsigned literal that is "
"out of range", arg_num);
return const0_rtx;
}
op[nopnds] = copy_to_mode_reg (insn_op->mode, op[nopnds]);
}
nopnds++;
}
/* Handle the builtins for extended mnemonics. These accept
no arguments, but map to builtins that take arguments. */
switch (fcode)
{
case HTM_BUILTIN_TENDALL: /* Alias for: tend. 1 */
case HTM_BUILTIN_TRESUME: /* Alias for: tsr. 1 */
op[nopnds++] = GEN_INT (1);
if (flag_checking)
attr |= RS6000_BTC_UNARY;
break;
case HTM_BUILTIN_TSUSPEND: /* Alias for: tsr. 0 */
op[nopnds++] = GEN_INT (0);
if (flag_checking)
attr |= RS6000_BTC_UNARY;
break;
default:
break;
}
/* If this builtin accesses SPRs, then pass in the appropriate
SPR number and SPR regno as the last two operands. */
if (uses_spr)
{
machine_mode mode = (TARGET_POWERPC64) ? DImode : SImode;
op[nopnds++] = gen_rtx_CONST_INT (mode, htm_spr_num (fcode));
op[nopnds++] = gen_rtx_REG (mode, htm_spr_regno (fcode));
}
/* If this builtin accesses a CR, then pass in a scratch
CR as the last operand. */
else if (attr & RS6000_BTC_CR)
{ cr = gen_reg_rtx (CCmode);
op[nopnds++] = cr;
}
if (flag_checking)
{
int expected_nopnds = 0;
if ((attr & RS6000_BTC_TYPE_MASK) == RS6000_BTC_UNARY)
expected_nopnds = 1;
else if ((attr & RS6000_BTC_TYPE_MASK) == RS6000_BTC_BINARY)
expected_nopnds = 2;
else if ((attr & RS6000_BTC_TYPE_MASK) == RS6000_BTC_TERNARY)
expected_nopnds = 3;
if (!(attr & RS6000_BTC_VOID))
expected_nopnds += 1;
if (uses_spr)
expected_nopnds += 2;
gcc_assert (nopnds == expected_nopnds
&& nopnds <= MAX_HTM_OPERANDS);
}
switch (nopnds)
{
case 1:
pat = GEN_FCN (icode) (op[0]);
break;
case 2:
pat = GEN_FCN (icode) (op[0], op[1]);
break;
case 3:
pat = GEN_FCN (icode) (op[0], op[1], op[2]);
break;
case 4:
pat = GEN_FCN (icode) (op[0], op[1], op[2], op[3]);
break;
default:
gcc_unreachable ();
}
if (!pat)
return NULL_RTX;
emit_insn (pat);
if (attr & RS6000_BTC_CR)
{
if (fcode == HTM_BUILTIN_TBEGIN)
{
/* Emit code to set TARGET to true or false depending on
whether the tbegin. instruction successfully or failed
to start a transaction. We do this by placing the 1's
complement of CR's EQ bit into TARGET. */
rtx scratch = gen_reg_rtx (SImode);
emit_insn (gen_rtx_SET (scratch,
gen_rtx_EQ (SImode, cr,
const0_rtx)));
emit_insn (gen_rtx_SET (target,
gen_rtx_XOR (SImode, scratch,
GEN_INT (1))));
}
else
{
/* Emit code to copy the 4-bit condition register field
CR into the least significant end of register TARGET. */
rtx scratch1 = gen_reg_rtx (SImode);
rtx scratch2 = gen_reg_rtx (SImode);
rtx subreg = simplify_gen_subreg (CCmode, scratch1, SImode, 0);
emit_insn (gen_movcc (subreg, cr));
emit_insn (gen_lshrsi3 (scratch2, scratch1, GEN_INT (28)));
emit_insn (gen_andsi3 (target, scratch2, GEN_INT (0xf)));
}
}
if (nonvoid)
return target;
return const0_rtx;
}
*expandedp = false;
return NULL_RTX;
}
/* Expand the CPU builtin in FCODE and store the result in TARGET. */
static rtx
cpu_expand_builtin (enum rs6000_builtins fcode, tree exp ATTRIBUTE_UNUSED,
rtx target)
{
/* __builtin_cpu_init () is a nop, so expand to nothing. */
if (fcode == RS6000_BUILTIN_CPU_INIT)
return const0_rtx;
if (target == 0 || GET_MODE (target) != SImode)
target = gen_reg_rtx (SImode);
#ifdef TARGET_LIBC_PROVIDES_HWCAP_IN_TCB
tree arg = TREE_OPERAND (CALL_EXPR_ARG (exp, 0), 0);
/* Target clones creates an ARRAY_REF instead of STRING_CST, convert it back
to a STRING_CST. */
if (TREE_CODE (arg) == ARRAY_REF
&& TREE_CODE (TREE_OPERAND (arg, 0)) == STRING_CST
&& TREE_CODE (TREE_OPERAND (arg, 1)) == INTEGER_CST
&& compare_tree_int (TREE_OPERAND (arg, 1), 0) == 0)
arg = TREE_OPERAND (arg, 0);
if (TREE_CODE (arg) != STRING_CST)
{
error ("builtin %qs only accepts a string argument",
rs6000_builtin_info[(size_t) fcode].name);
return const0_rtx;
}
if (fcode == RS6000_BUILTIN_CPU_IS)
{
const char *cpu = TREE_STRING_POINTER (arg);
rtx cpuid = NULL_RTX;
for (size_t i = 0; i < ARRAY_SIZE (cpu_is_info); i++)
if (strcmp (cpu, cpu_is_info[i].cpu) == 0)
{
/* The CPUID value in the TCB is offset by _DL_FIRST_PLATFORM. */
cpuid = GEN_INT (cpu_is_info[i].cpuid + _DL_FIRST_PLATFORM);
break;
}
if (cpuid == NULL_RTX)
{
/* Invalid CPU argument. */
error ("cpu %qs is an invalid argument to builtin %qs",
cpu, rs6000_builtin_info[(size_t) fcode].name);
return const0_rtx;
}
rtx platform = gen_reg_rtx (SImode);
rtx tcbmem = gen_const_mem (SImode,
gen_rtx_PLUS (Pmode,
gen_rtx_REG (Pmode, TLS_REGNUM),
GEN_INT (TCB_PLATFORM_OFFSET)));
emit_move_insn (platform, tcbmem);
emit_insn (gen_eqsi3 (target, platform, cpuid));
}
else if (fcode == RS6000_BUILTIN_CPU_SUPPORTS)
{
const char *hwcap = TREE_STRING_POINTER (arg);
rtx mask = NULL_RTX;
int hwcap_offset;
for (size_t i = 0; i < ARRAY_SIZE (cpu_supports_info); i++)
if (strcmp (hwcap, cpu_supports_info[i].hwcap) == 0)
{
mask = GEN_INT (cpu_supports_info[i].mask);
hwcap_offset = TCB_HWCAP_OFFSET (cpu_supports_info[i].id);
break;
}
if (mask == NULL_RTX)
{
/* Invalid HWCAP argument. */
error ("%s %qs is an invalid argument to builtin %qs",
"hwcap", hwcap, rs6000_builtin_info[(size_t) fcode].name);
return const0_rtx;
}
rtx tcb_hwcap = gen_reg_rtx (SImode);
rtx tcbmem = gen_const_mem (SImode,
gen_rtx_PLUS (Pmode,
gen_rtx_REG (Pmode, TLS_REGNUM),
GEN_INT (hwcap_offset)));
emit_move_insn (tcb_hwcap, tcbmem);
rtx scratch1 = gen_reg_rtx (SImode);
emit_insn (gen_rtx_SET (scratch1, gen_rtx_AND (SImode, tcb_hwcap, mask)));
rtx scratch2 = gen_reg_rtx (SImode);
emit_insn (gen_eqsi3 (scratch2, scratch1, const0_rtx));
emit_insn (gen_rtx_SET (target, gen_rtx_XOR (SImode, scratch2, const1_rtx)));
}
else
gcc_unreachable ();
/* Record that we have expanded a CPU builtin, so that we can later
emit a reference to the special symbol exported by LIBC to ensure we
do not link against an old LIBC that doesn't support this feature. */
cpu_builtin_p = true;
#else
warning (0, "builtin %qs needs GLIBC (2.23 and newer) that exports hardware "
"capability bits", rs6000_builtin_info[(size_t) fcode].name);
/* For old LIBCs, always return FALSE. */
emit_move_insn (target, GEN_INT (0));
#endif /* TARGET_LIBC_PROVIDES_HWCAP_IN_TCB */
return target;
}
static rtx
rs6000_expand_ternop_builtin (enum insn_code icode, tree exp, rtx target)
{
rtx pat;
tree arg0 = CALL_EXPR_ARG (exp, 0);
tree arg1 = CALL_EXPR_ARG (exp, 1);
tree arg2 = CALL_EXPR_ARG (exp, 2);
rtx op0 = expand_normal (arg0);
rtx op1 = expand_normal (arg1);
rtx op2 = expand_normal (arg2);
machine_mode tmode = insn_data[icode].operand[0].mode;
machine_mode mode0 = insn_data[icode].operand[1].mode;
machine_mode mode1 = insn_data[icode].operand[2].mode;
machine_mode mode2 = insn_data[icode].operand[3].mode;
if (icode == CODE_FOR_nothing)
/* Builtin not supported on this processor. */
return 0;
/* If we got invalid arguments bail out before generating bad rtl. */
if (arg0 == error_mark_node
|| arg1 == error_mark_node
|| arg2 == error_mark_node)
return const0_rtx;
/* Check and prepare argument depending on the instruction code.
Note that a switch statement instead of the sequence of tests
would be incorrect as many of the CODE_FOR values could be
CODE_FOR_nothing and that would yield multiple alternatives
with identical values. We'd never reach here at runtime in
this case. */
if (icode == CODE_FOR_altivec_vsldoi_v4sf
|| icode == CODE_FOR_altivec_vsldoi_v2df
|| icode == CODE_FOR_altivec_vsldoi_v4si
|| icode == CODE_FOR_altivec_vsldoi_v8hi
|| icode == CODE_FOR_altivec_vsldoi_v16qi)
{
/* Only allow 4-bit unsigned literals. */
STRIP_NOPS (arg2);
if (TREE_CODE (arg2) != INTEGER_CST
|| TREE_INT_CST_LOW (arg2) & ~0xf)
{
error ("argument 3 must be a 4-bit unsigned literal");
return CONST0_RTX (tmode);
}
}
else if (icode == CODE_FOR_vsx_xxpermdi_v2df
|| icode == CODE_FOR_vsx_xxpermdi_v2di
|| icode == CODE_FOR_vsx_xxpermdi_v2df_be
|| icode == CODE_FOR_vsx_xxpermdi_v2di_be
|| icode == CODE_FOR_vsx_xxpermdi_v1ti
|| icode == CODE_FOR_vsx_xxpermdi_v4sf
|| icode == CODE_FOR_vsx_xxpermdi_v4si
|| icode == CODE_FOR_vsx_xxpermdi_v8hi
|| icode == CODE_FOR_vsx_xxpermdi_v16qi
|| icode == CODE_FOR_vsx_xxsldwi_v16qi
|| icode == CODE_FOR_vsx_xxsldwi_v8hi
|| icode == CODE_FOR_vsx_xxsldwi_v4si
|| icode == CODE_FOR_vsx_xxsldwi_v4sf
|| icode == CODE_FOR_vsx_xxsldwi_v2di
|| icode == CODE_FOR_vsx_xxsldwi_v2df)
{
/* Only allow 2-bit unsigned literals. */
STRIP_NOPS (arg2);
if (TREE_CODE (arg2) != INTEGER_CST
|| TREE_INT_CST_LOW (arg2) & ~0x3)
{
error ("argument 3 must be a 2-bit unsigned literal");
return CONST0_RTX (tmode);
}
}
else if (icode == CODE_FOR_vsx_set_v2df
|| icode == CODE_FOR_vsx_set_v2di
|| icode == CODE_FOR_bcdadd
|| icode == CODE_FOR_bcdadd_lt
|| icode == CODE_FOR_bcdadd_eq
|| icode == CODE_FOR_bcdadd_gt
|| icode == CODE_FOR_bcdsub
|| icode == CODE_FOR_bcdsub_lt
|| icode == CODE_FOR_bcdsub_eq
|| icode == CODE_FOR_bcdsub_gt)
{
/* Only allow 1-bit unsigned literals. */
STRIP_NOPS (arg2);
if (TREE_CODE (arg2) != INTEGER_CST
|| TREE_INT_CST_LOW (arg2) & ~0x1)
{
error ("argument 3 must be a 1-bit unsigned literal");
return CONST0_RTX (tmode);
}
}
else if (icode == CODE_FOR_dfp_ddedpd_dd
|| icode == CODE_FOR_dfp_ddedpd_td)
{
/* Only allow 2-bit unsigned literals where the value is 0 or 2. */
STRIP_NOPS (arg0);
if (TREE_CODE (arg0) != INTEGER_CST
|| TREE_INT_CST_LOW (arg2) & ~0x3)
{
error ("argument 1 must be 0 or 2");
return CONST0_RTX (tmode);
}
}
else if (icode == CODE_FOR_dfp_denbcd_dd
|| icode == CODE_FOR_dfp_denbcd_td)
{
/* Only allow 1-bit unsigned literals. */
STRIP_NOPS (arg0);
if (TREE_CODE (arg0) != INTEGER_CST
|| TREE_INT_CST_LOW (arg0) & ~0x1)
{
error ("argument 1 must be a 1-bit unsigned literal");
return CONST0_RTX (tmode);
}
}
else if (icode == CODE_FOR_dfp_dscli_dd
|| icode == CODE_FOR_dfp_dscli_td
|| icode == CODE_FOR_dfp_dscri_dd
|| icode == CODE_FOR_dfp_dscri_td)
{
/* Only allow 6-bit unsigned literals. */
STRIP_NOPS (arg1);
if (TREE_CODE (arg1) != INTEGER_CST
|| TREE_INT_CST_LOW (arg1) & ~0x3f)
{
error ("argument 2 must be a 6-bit unsigned literal");
return CONST0_RTX (tmode);
}
}
else if (icode == CODE_FOR_crypto_vshasigmaw
|| icode == CODE_FOR_crypto_vshasigmad)
{
/* Check whether the 2nd and 3rd arguments are integer constants and in
range and prepare arguments. */
STRIP_NOPS (arg1);
if (TREE_CODE (arg1) != INTEGER_CST || wi::geu_p (wi::to_wide (arg1), 2))
{
error ("argument 2 must be 0 or 1");
return CONST0_RTX (tmode);
}
STRIP_NOPS (arg2);
if (TREE_CODE (arg2) != INTEGER_CST
|| wi::geu_p (wi::to_wide (arg2), 16))
{
error ("argument 3 must be in the range 0..15");
return CONST0_RTX (tmode);
}
}
if (target == 0
|| GET_MODE (target) != tmode
|| ! (*insn_data[icode].operand[0].predicate) (target, tmode))
target = gen_reg_rtx (tmode);
if (! (*insn_data[icode].operand[1].predicate) (op0, mode0))
op0 = copy_to_mode_reg (mode0, op0);
if (! (*insn_data[icode].operand[2].predicate) (op1, mode1))
op1 = copy_to_mode_reg (mode1, op1);
if (! (*insn_data[icode].operand[3].predicate) (op2, mode2))
op2 = copy_to_mode_reg (mode2, op2);
if (TARGET_PAIRED_FLOAT && icode == CODE_FOR_selv2sf4)
pat = GEN_FCN (icode) (target, op0, op1, op2, CONST0_RTX (SFmode));
else
pat = GEN_FCN (icode) (target, op0, op1, op2);
if (! pat)
return 0;
emit_insn (pat);
return target;
}
/* Expand the lvx builtins. */
static rtx
altivec_expand_ld_builtin (tree exp, rtx target, bool *expandedp)
{
tree fndecl = TREE_OPERAND (CALL_EXPR_FN (exp), 0);
unsigned int fcode = DECL_FUNCTION_CODE (fndecl);
tree arg0;
machine_mode tmode, mode0;
rtx pat, op0;
enum insn_code icode;
switch (fcode)
{
case ALTIVEC_BUILTIN_LD_INTERNAL_16qi:
icode = CODE_FOR_vector_altivec_load_v16qi;
break;
case ALTIVEC_BUILTIN_LD_INTERNAL_8hi:
icode = CODE_FOR_vector_altivec_load_v8hi;
break;
case ALTIVEC_BUILTIN_LD_INTERNAL_4si:
icode = CODE_FOR_vector_altivec_load_v4si;
break;
case ALTIVEC_BUILTIN_LD_INTERNAL_4sf:
icode = CODE_FOR_vector_altivec_load_v4sf;
break;
case ALTIVEC_BUILTIN_LD_INTERNAL_2df:
icode = CODE_FOR_vector_altivec_load_v2df;
break;
case ALTIVEC_BUILTIN_LD_INTERNAL_2di:
icode = CODE_FOR_vector_altivec_load_v2di;
break;
case ALTIVEC_BUILTIN_LD_INTERNAL_1ti:
icode = CODE_FOR_vector_altivec_load_v1ti;
break;
default:
*expandedp = false;
return NULL_RTX;
}
*expandedp = true;
arg0 = CALL_EXPR_ARG (exp, 0);
op0 = expand_normal (arg0);
tmode = insn_data[icode].operand[0].mode;
mode0 = insn_data[icode].operand[1].mode;
if (target == 0
|| GET_MODE (target) != tmode
|| ! (*insn_data[icode].operand[0].predicate) (target, tmode))
target = gen_reg_rtx (tmode);
if (! (*insn_data[icode].operand[1].predicate) (op0, mode0))
op0 = gen_rtx_MEM (mode0, copy_to_mode_reg (Pmode, op0));
pat = GEN_FCN (icode) (target, op0);
if (! pat)
return 0;
emit_insn (pat);
return target;
}
/* Expand the stvx builtins. */
static rtx
altivec_expand_st_builtin (tree exp, rtx target ATTRIBUTE_UNUSED,
bool *expandedp)
{
tree fndecl = TREE_OPERAND (CALL_EXPR_FN (exp), 0);
unsigned int fcode = DECL_FUNCTION_CODE (fndecl);
tree arg0, arg1;
machine_mode mode0, mode1;
rtx pat, op0, op1;
enum insn_code icode;
switch (fcode)
{
case ALTIVEC_BUILTIN_ST_INTERNAL_16qi:
icode = CODE_FOR_vector_altivec_store_v16qi;
break;
case ALTIVEC_BUILTIN_ST_INTERNAL_8hi:
icode = CODE_FOR_vector_altivec_store_v8hi;
break;
case ALTIVEC_BUILTIN_ST_INTERNAL_4si:
icode = CODE_FOR_vector_altivec_store_v4si;
break;
case ALTIVEC_BUILTIN_ST_INTERNAL_4sf:
icode = CODE_FOR_vector_altivec_store_v4sf;
break;
case ALTIVEC_BUILTIN_ST_INTERNAL_2df:
icode = CODE_FOR_vector_altivec_store_v2df;
break;
case ALTIVEC_BUILTIN_ST_INTERNAL_2di:
icode = CODE_FOR_vector_altivec_store_v2di;
break;
case ALTIVEC_BUILTIN_ST_INTERNAL_1ti:
icode = CODE_FOR_vector_altivec_store_v1ti;
break;
default:
*expandedp = false;
return NULL_RTX;
}
arg0 = CALL_EXPR_ARG (exp, 0);
arg1 = CALL_EXPR_ARG (exp, 1);
op0 = expand_normal (arg0);
op1 = expand_normal (arg1);
mode0 = insn_data[icode].operand[0].mode;
mode1 = insn_data[icode].operand[1].mode;
if (! (*insn_data[icode].operand[0].predicate) (op0, mode0))
op0 = gen_rtx_MEM (mode0, copy_to_mode_reg (Pmode, op0));
if (! (*insn_data[icode].operand[1].predicate) (op1, mode1))
op1 = copy_to_mode_reg (mode1, op1);
pat = GEN_FCN (icode) (op0, op1);
if (pat)
emit_insn (pat);
*expandedp = true;
return NULL_RTX;
}
/* Expand the dst builtins. */
static rtx
altivec_expand_dst_builtin (tree exp, rtx target ATTRIBUTE_UNUSED,
bool *expandedp)
{
tree fndecl = TREE_OPERAND (CALL_EXPR_FN (exp), 0);
enum rs6000_builtins fcode = (enum rs6000_builtins) DECL_FUNCTION_CODE (fndecl);
tree arg0, arg1, arg2;
machine_mode mode0, mode1;
rtx pat, op0, op1, op2;
const struct builtin_description *d;
size_t i;
*expandedp = false;
/* Handle DST variants. */
d = bdesc_dst;
for (i = 0; i < ARRAY_SIZE (bdesc_dst); i++, d++)
if (d->code == fcode)
{
arg0 = CALL_EXPR_ARG (exp, 0);
arg1 = CALL_EXPR_ARG (exp, 1);
arg2 = CALL_EXPR_ARG (exp, 2);
op0 = expand_normal (arg0);
op1 = expand_normal (arg1);
op2 = expand_normal (arg2);
mode0 = insn_data[d->icode].operand[0].mode;
mode1 = insn_data[d->icode].operand[1].mode;
/* Invalid arguments, bail out before generating bad rtl. */
if (arg0 == error_mark_node
|| arg1 == error_mark_node
|| arg2 == error_mark_node)
return const0_rtx;
*expandedp = true;
STRIP_NOPS (arg2);
if (TREE_CODE (arg2) != INTEGER_CST
|| TREE_INT_CST_LOW (arg2) & ~0x3)
{
error ("argument to %qs must be a 2-bit unsigned literal", d->name);
return const0_rtx;
}
if (! (*insn_data[d->icode].operand[0].predicate) (op0, mode0))
op0 = copy_to_mode_reg (Pmode, op0);
if (! (*insn_data[d->icode].operand[1].predicate) (op1, mode1))
op1 = copy_to_mode_reg (mode1, op1);
pat = GEN_FCN (d->icode) (op0, op1, op2);
if (pat != 0)
emit_insn (pat);
return NULL_RTX;
}
return NULL_RTX;
}
/* Expand vec_init builtin. */
static rtx
altivec_expand_vec_init_builtin (tree type, tree exp, rtx target)
{
machine_mode tmode = TYPE_MODE (type);
machine_mode inner_mode = GET_MODE_INNER (tmode);
int i, n_elt = GET_MODE_NUNITS (tmode);
gcc_assert (VECTOR_MODE_P (tmode));
gcc_assert (n_elt == call_expr_nargs (exp));
if (!target || !register_operand (target, tmode))
target = gen_reg_rtx (tmode);
/* If we have a vector compromised of a single element, such as V1TImode, do
the initialization directly. */
if (n_elt == 1 && GET_MODE_SIZE (tmode) == GET_MODE_SIZE (inner_mode))
{
rtx x = expand_normal (CALL_EXPR_ARG (exp, 0));
emit_move_insn (target, gen_lowpart (tmode, x));
}
else
{
rtvec v = rtvec_alloc (n_elt);
for (i = 0; i < n_elt; ++i)
{
rtx x = expand_normal (CALL_EXPR_ARG (exp, i));
RTVEC_ELT (v, i) = gen_lowpart (inner_mode, x);
}
rs6000_expand_vector_init (target, gen_rtx_PARALLEL (tmode, v));
}
return target;
}
/* Return the integer constant in ARG. Constrain it to be in the range
of the subparts of VEC_TYPE; issue an error if not. */
static int
get_element_number (tree vec_type, tree arg)
{
unsigned HOST_WIDE_INT elt, max = TYPE_VECTOR_SUBPARTS (vec_type) - 1;
if (!tree_fits_uhwi_p (arg)
|| (elt = tree_to_uhwi (arg), elt > max))
{
error ("selector must be an integer constant in the range 0..%wi", max);
return 0;
}
return elt;
}
/* Expand vec_set builtin. */
static rtx
altivec_expand_vec_set_builtin (tree exp)
{
machine_mode tmode, mode1;
tree arg0, arg1, arg2;
int elt;
rtx op0, op1;
arg0 = CALL_EXPR_ARG (exp, 0);
arg1 = CALL_EXPR_ARG (exp, 1);
arg2 = CALL_EXPR_ARG (exp, 2);
tmode = TYPE_MODE (TREE_TYPE (arg0));
mode1 = TYPE_MODE (TREE_TYPE (TREE_TYPE (arg0)));
gcc_assert (VECTOR_MODE_P (tmode));
op0 = expand_expr (arg0, NULL_RTX, tmode, EXPAND_NORMAL);
op1 = expand_expr (arg1, NULL_RTX, mode1, EXPAND_NORMAL);
elt = get_element_number (TREE_TYPE (arg0), arg2);
if (GET_MODE (op1) != mode1 && GET_MODE (op1) != VOIDmode)
op1 = convert_modes (mode1, GET_MODE (op1), op1, true);
op0 = force_reg (tmode, op0);
op1 = force_reg (mode1, op1);
rs6000_expand_vector_set (op0, op1, elt);
return op0;
}
/* Expand vec_ext builtin. */
static rtx
altivec_expand_vec_ext_builtin (tree exp, rtx target)
{
machine_mode tmode, mode0;
tree arg0, arg1;
rtx op0;
rtx op1;
arg0 = CALL_EXPR_ARG (exp, 0);
arg1 = CALL_EXPR_ARG (exp, 1);
op0 = expand_normal (arg0);
op1 = expand_normal (arg1);
/* Call get_element_number to validate arg1 if it is a constant. */
if (TREE_CODE (arg1) == INTEGER_CST)
(void) get_element_number (TREE_TYPE (arg0), arg1);
tmode = TYPE_MODE (TREE_TYPE (TREE_TYPE (arg0)));
mode0 = TYPE_MODE (TREE_TYPE (arg0));
gcc_assert (VECTOR_MODE_P (mode0));
op0 = force_reg (mode0, op0);
if (optimize || !target || !register_operand (target, tmode))
target = gen_reg_rtx (tmode);
rs6000_expand_vector_extract (target, op0, op1);
return target;
}
/* Expand the builtin in EXP and store the result in TARGET. Store
true in *EXPANDEDP if we found a builtin to expand. */
static rtx
altivec_expand_builtin (tree exp, rtx target, bool *expandedp)
{
const struct builtin_description *d;
size_t i;
enum insn_code icode;
tree fndecl = TREE_OPERAND (CALL_EXPR_FN (exp), 0);
tree arg0, arg1, arg2;
rtx op0, pat;
machine_mode tmode, mode0;
enum rs6000_builtins fcode
= (enum rs6000_builtins) DECL_FUNCTION_CODE (fndecl);
if (rs6000_overloaded_builtin_p (fcode))
{
*expandedp = true;
error ("unresolved overload for Altivec builtin %qF", fndecl);
/* Given it is invalid, just generate a normal call. */
return expand_call (exp, target, false);
}
target = altivec_expand_ld_builtin (exp, target, expandedp);
if (*expandedp)
return target;
target = altivec_expand_st_builtin (exp, target, expandedp);
if (*expandedp)
return target;
target = altivec_expand_dst_builtin (exp, target, expandedp);
if (*expandedp)
return target;
*expandedp = true;
switch (fcode)
{
case ALTIVEC_BUILTIN_STVX_V2DF:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvx_v2df_2op, exp);
case ALTIVEC_BUILTIN_STVX_V2DI:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvx_v2di_2op, exp);
case ALTIVEC_BUILTIN_STVX_V4SF:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvx_v4sf_2op, exp);
case ALTIVEC_BUILTIN_STVX:
case ALTIVEC_BUILTIN_STVX_V4SI:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvx_v4si_2op, exp);
case ALTIVEC_BUILTIN_STVX_V8HI:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvx_v8hi_2op, exp);
case ALTIVEC_BUILTIN_STVX_V16QI:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvx_v16qi_2op, exp);
case ALTIVEC_BUILTIN_STVEBX:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvebx, exp);
case ALTIVEC_BUILTIN_STVEHX:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvehx, exp);
case ALTIVEC_BUILTIN_STVEWX:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvewx, exp);
case ALTIVEC_BUILTIN_STVXL_V2DF:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvxl_v2df, exp);
case ALTIVEC_BUILTIN_STVXL_V2DI:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvxl_v2di, exp);
case ALTIVEC_BUILTIN_STVXL_V4SF:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvxl_v4sf, exp);
case ALTIVEC_BUILTIN_STVXL:
case ALTIVEC_BUILTIN_STVXL_V4SI:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvxl_v4si, exp);
case ALTIVEC_BUILTIN_STVXL_V8HI:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvxl_v8hi, exp);
case ALTIVEC_BUILTIN_STVXL_V16QI:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvxl_v16qi, exp);
case ALTIVEC_BUILTIN_STVLX:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvlx, exp);
case ALTIVEC_BUILTIN_STVLXL:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvlxl, exp);
case ALTIVEC_BUILTIN_STVRX:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvrx, exp);
case ALTIVEC_BUILTIN_STVRXL:
return altivec_expand_stv_builtin (CODE_FOR_altivec_stvrxl, exp);
case P9V_BUILTIN_STXVL:
return altivec_expand_stxvl_builtin (CODE_FOR_stxvl, exp);
case P9V_BUILTIN_XST_LEN_R:
return altivec_expand_stxvl_builtin (CODE_FOR_xst_len_r, exp);
case VSX_BUILTIN_STXVD2X_V1TI:
return altivec_expand_stv_builtin (CODE_FOR_vsx_store_v1ti, exp);
case VSX_BUILTIN_STXVD2X_V2DF:
return altivec_expand_stv_builtin (CODE_FOR_vsx_store_v2df, exp);
case VSX_BUILTIN_STXVD2X_V2DI:
return altivec_expand_stv_builtin (CODE_FOR_vsx_store_v2di, exp);
case VSX_BUILTIN_STXVW4X_V4SF:
return altivec_expand_stv_builtin (CODE_FOR_vsx_store_v4sf, exp);
case VSX_BUILTIN_STXVW4X_V4SI:
return altivec_expand_stv_builtin (CODE_FOR_vsx_store_v4si, exp);
case VSX_BUILTIN_STXVW4X_V8HI:
return altivec_expand_stv_builtin (CODE_FOR_vsx_store_v8hi, exp);
case VSX_BUILTIN_STXVW4X_V16QI:
return altivec_expand_stv_builtin (CODE_FOR_vsx_store_v16qi, exp);
/* For the following on big endian, it's ok to use any appropriate
unaligned-supporting store, so use a generic expander. For
little-endian, the exact element-reversing instruction must
be used. */
case VSX_BUILTIN_ST_ELEMREV_V2DF:
{
enum insn_code code = (BYTES_BIG_ENDIAN ? CODE_FOR_vsx_store_v2df
: CODE_FOR_vsx_st_elemrev_v2df);
return altivec_expand_stv_builtin (code, exp);
}
case VSX_BUILTIN_ST_ELEMREV_V2DI:
{
enum insn_code code = (BYTES_BIG_ENDIAN ? CODE_FOR_vsx_store_v2di
: CODE_FOR_vsx_st_elemrev_v2di);
return altivec_expand_stv_builtin (code, exp);
}
case VSX_BUILTIN_ST_ELEMREV_V4SF:
{
enum insn_code code = (BYTES_BIG_ENDIAN ? CODE_FOR_vsx_store_v4sf
: CODE_FOR_vsx_st_elemrev_v4sf);
return altivec_expand_stv_builtin (code, exp);
}
case VSX_BUILTIN_ST_ELEMREV_V4SI:
{
enum insn_code code = (BYTES_BIG_ENDIAN ? CODE_FOR_vsx_store_v4si
: CODE_FOR_vsx_st_elemrev_v4si);
return altivec_expand_stv_builtin (code, exp);
}
case VSX_BUILTIN_ST_ELEMREV_V8HI:
{
enum insn_code code = (BYTES_BIG_ENDIAN ? CODE_FOR_vsx_store_v8hi
: CODE_FOR_vsx_st_elemrev_v8hi);
return altivec_expand_stv_builtin (code, exp);
}
case VSX_BUILTIN_ST_ELEMREV_V16QI:
{
enum insn_code code = (BYTES_BIG_ENDIAN ? CODE_FOR_vsx_store_v16qi
: CODE_FOR_vsx_st_elemrev_v16qi);
return altivec_expand_stv_builtin (code, exp);
}
case ALTIVEC_BUILTIN_MFVSCR:
icode = CODE_FOR_altivec_mfvscr;
tmode = insn_data[icode].operand[0].mode;
if (target == 0
|| GET_MODE (target) != tmode
|| ! (*insn_data[icode].operand[0].predicate) (target, tmode))
target = gen_reg_rtx (tmode);
pat = GEN_FCN (icode) (target);
if (! pat)
return 0;
emit_insn (pat);
return target;
case ALTIVEC_BUILTIN_MTVSCR:
icode = CODE_FOR_altivec_mtvscr;
arg0 = CALL_EXPR_ARG (exp, 0);
op0 = expand_normal (arg0);
mode0 = insn_data[icode].operand[0].mode;
/* If we got invalid arguments bail out before generating bad rtl. */
if (arg0 == error_mark_node)
return const0_rtx;
if (! (*insn_data[icode].operand[0].predicate) (op0, mode0))
op0 = copy_to_mode_reg (mode0, op0);
pat = GEN_FCN (icode) (op0);
if (pat)
emit_insn (pat);
return NULL_RTX;
case ALTIVEC_BUILTIN_DSSALL:
emit_insn (gen_altivec_dssall ());
return NULL_RTX;
case ALTIVEC_BUILTIN_DSS:
icode = CODE_FOR_altivec_dss;
arg0 = CALL_EXPR_ARG (exp, 0);
STRIP_NOPS (arg0);
op0 = expand_normal (arg0);
mode0 = insn_data[icode].operand[0].mode;
/* If we got invalid arguments bail out before generating bad rtl. */
if (arg0 == error_mark_node)
return const0_rtx;
if (TREE_CODE (arg0) != INTEGER_CST
|| TREE_INT_CST_LOW (arg0) & ~0x3)
{
error ("argument to %qs must be a 2-bit unsigned literal", "dss");
return const0_rtx;
}
if (! (*insn_data[icode].operand[0].predicate) (op0, mode0))
op0 = copy_to_mode_reg (mode0, op0);
emit_insn (gen_altivec_dss (op0));
return NULL_RTX;
case ALTIVEC_BUILTIN_VEC_INIT_V4SI:
case ALTIVEC_BUILTIN_VEC_INIT_V8HI:
case ALTIVEC_BUILTIN_VEC_INIT_V16QI:
case ALTIVEC_BUILTIN_VEC_INIT_V4SF:
case VSX_BUILTIN_VEC_INIT_V2DF:
case VSX_BUILTIN_VEC_INIT_V2DI:
case VSX_BUILTIN_VEC_INIT_V1TI:
return altivec_expand_vec_init_builtin (TREE_TYPE (exp), exp, target);
case ALTIVEC_BUILTIN_VEC_SET_V4SI:
case ALTIVEC_BUILTIN_VEC_SET_V8HI:
case ALTIVEC_BUILTIN_VEC_SET_V16QI:
case ALTIVEC_BUILTIN_VEC_SET_V4SF:
case VSX_BUILTIN_VEC_SET_V2DF:
case VSX_BUILTIN_VEC_SET_V2DI:
case VSX_BUILTIN_VEC_SET_V1TI:
return altivec_expand_vec_set_builtin (exp);
case ALTIVEC_BUILTIN_VEC_EXT_V4SI:
case ALTIVEC_BUILTIN_VEC_EXT_V8HI:
case ALTIVEC_BUILTIN_VEC_EXT_V16QI:
case ALTIVEC_BUILTIN_VEC_EXT_V4SF:
case VSX_BUILTIN_VEC_EXT_V2DF:
case VSX_BUILTIN_VEC_EXT_V2DI:
case VSX_BUILTIN_VEC_EXT_V1TI:
return altivec_expand_vec_ext_builtin (exp, target);
case P9V_BUILTIN_VEXTRACT4B:
case P9V_BUILTIN_VEC_VEXTRACT4B:
arg1 = CALL_EXPR_ARG (exp, 1);
STRIP_NOPS (arg1);
/* Generate a normal call if it is invalid. */
if (arg1 == error_mark_node)
return expand_call (exp, target, false);
if (TREE_CODE (arg1) != INTEGER_CST || TREE_INT_CST_LOW (arg1) > 12)
{
error ("second argument to %qs must be 0..12", "vec_vextract4b");
return expand_call (exp, target, false);
}
break;
case P9V_BUILTIN_VINSERT4B:
case P9V_BUILTIN_VINSERT4B_DI:
case P9V_BUILTIN_VEC_VINSERT4B:
arg2 = CALL_EXPR_ARG (exp, 2);
STRIP_NOPS (arg2);
/* Generate a normal call if it is invalid. */
if (arg2 == error_mark_node)
return expand_call (exp, target, false);
if (TREE_CODE (arg2) != INTEGER_CST || TREE_INT_CST_LOW (arg2) > 12)
{
error ("third argument to %qs must be 0..12", "vec_vinsert4b");
return expand_call (exp, target, false);
}
break;
default:
break;
/* Fall through. */
}
/* Expand abs* operations. */
d = bdesc_abs;
for (i = 0; i < ARRAY_SIZE (bdesc_abs); i++, d++)
if (d->code == fcode)
return altivec_expand_abs_builtin (d->icode, exp, target);
/* Expand the AltiVec predicates. */
d = bdesc_altivec_preds;
for (i = 0; i < ARRAY_SIZE (bdesc_altivec_preds); i++, d++)
if (d->code == fcode)
return altivec_expand_predicate_builtin (d->icode, exp, target);
/* LV* are funky. We initialized them differently. */
switch (fcode)
{
case ALTIVEC_BUILTIN_LVSL:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvsl,
exp, target, false);
case ALTIVEC_BUILTIN_LVSR:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvsr,
exp, target, false);
case ALTIVEC_BUILTIN_LVEBX:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvebx,
exp, target, false);
case ALTIVEC_BUILTIN_LVEHX:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvehx,
exp, target, false);
case ALTIVEC_BUILTIN_LVEWX:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvewx,
exp, target, false);
case ALTIVEC_BUILTIN_LVXL_V2DF:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvxl_v2df,
exp, target, false);
case ALTIVEC_BUILTIN_LVXL_V2DI:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvxl_v2di,
exp, target, false);
case ALTIVEC_BUILTIN_LVXL_V4SF:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvxl_v4sf,
exp, target, false);
case ALTIVEC_BUILTIN_LVXL:
case ALTIVEC_BUILTIN_LVXL_V4SI:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvxl_v4si,
exp, target, false);
case ALTIVEC_BUILTIN_LVXL_V8HI:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvxl_v8hi,
exp, target, false);
case ALTIVEC_BUILTIN_LVXL_V16QI:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvxl_v16qi,
exp, target, false);
case ALTIVEC_BUILTIN_LVX_V2DF:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvx_v2df_2op,
exp, target, false);
case ALTIVEC_BUILTIN_LVX_V2DI:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvx_v2di_2op,
exp, target, false);
case ALTIVEC_BUILTIN_LVX_V4SF:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvx_v4sf_2op,
exp, target, false);
case ALTIVEC_BUILTIN_LVX:
case ALTIVEC_BUILTIN_LVX_V4SI:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvx_v4si_2op,
exp, target, false);
case ALTIVEC_BUILTIN_LVX_V8HI:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvx_v8hi_2op,
exp, target, false);
case ALTIVEC_BUILTIN_LVX_V16QI:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvx_v16qi_2op,
exp, target, false);
case ALTIVEC_BUILTIN_LVLX:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvlx,
exp, target, true);
case ALTIVEC_BUILTIN_LVLXL:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvlxl,
exp, target, true);
case ALTIVEC_BUILTIN_LVRX:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvrx,
exp, target, true);
case ALTIVEC_BUILTIN_LVRXL:
return altivec_expand_lv_builtin (CODE_FOR_altivec_lvrxl,
exp, target, true);
case VSX_BUILTIN_LXVD2X_V1TI:
return altivec_expand_lv_builtin (CODE_FOR_vsx_load_v1ti,
exp, target, false);
case VSX_BUILTIN_LXVD2X_V2DF:
return altivec_expand_lv_builtin (CODE_FOR_vsx_load_v2df,
exp, target, false);
case VSX_BUILTIN_LXVD2X_V2DI:
return altivec_expand_lv_builtin (CODE_FOR_vsx_load_v2di,
exp, target, false);
case VSX_BUILTIN_LXVW4X_V4SF:
return altivec_expand_lv_builtin (CODE_FOR_vsx_load_v4sf,
exp, target, false);
case VSX_BUILTIN_LXVW4X_V4SI:
return altivec_expand_lv_builtin (CODE_FOR_vsx_load_v4si,
exp, target, false);
case VSX_BUILTIN_LXVW4X_V8HI:
return altivec_expand_lv_builtin (CODE_FOR_vsx_load_v8hi,
exp, target, false);
case VSX_BUILTIN_LXVW4X_V16QI:
return altivec_expand_lv_builtin (CODE_FOR_vsx_load_v16qi,
exp, target, false);
/* For the following on big endian, it's ok to use any appropriate
unaligned-supporting load, so use a generic expander. For
little-endian, the exact element-reversing instruction must
be used. */
case VSX_BUILTIN_LD_ELEMREV_V2DF:
{
enum insn_code code = (BYTES_BIG_ENDIAN ? CODE_FOR_vsx_load_v2df
: CODE_FOR_vsx_ld_elemrev_v2df);
return altivec_expand_lv_builtin (code, exp, target, false);
}
case VSX_BUILTIN_LD_ELEMREV_V2DI:
{
enum insn_code code = (BYTES_BIG_ENDIAN ? CODE_FOR_vsx_load_v2di
: CODE_FOR_vsx_ld_elemrev_v2di);
return altivec_expand_lv_builtin (code, exp, target, false);
}
case VSX_BUILTIN_LD_ELEMREV_V4SF:
{
enum insn_code code = (BYTES_BIG_ENDIAN ? CODE_FOR_vsx_load_v4sf
: CODE_FOR_vsx_ld_elemrev_v4sf);
return altivec_expand_lv_builtin (code, exp, target, false);
}
case VSX_BUILTIN_LD_ELEMREV_V4SI:
{
enum insn_code code = (BYTES_BIG_ENDIAN ? CODE_FOR_vsx_load_v4si
: CODE_FOR_vsx_ld_elemrev_v4si);
return altivec_expand_lv_builtin (code, exp, target, false);
}
case VSX_BUILTIN_LD_ELEMREV_V8HI:
{
enum insn_code code = (BYTES_BIG_ENDIAN ? CODE_FOR_vsx_load_v8hi
: CODE_FOR_vsx_ld_elemrev_v8hi);
return altivec_expand_lv_builtin (code, exp, target, false);
}
case VSX_BUILTIN_LD_ELEMREV_V16QI:
{
enum insn_code code = (BYTES_BIG_ENDIAN ? CODE_FOR_vsx_load_v16qi
: CODE_FOR_vsx_ld_elemrev_v16qi);
return altivec_expand_lv_builtin (code, exp, target, false);
}
break;
default:
break;
/* Fall through. */
}
*expandedp = false;
return NULL_RTX;
}
/* Expand the builtin in EXP and store the result in TARGET. Store
true in *EXPANDEDP if we found a builtin to expand. */
static rtx
paired_expand_builtin (tree exp, rtx target, bool * expandedp)
{
tree fndecl = TREE_OPERAND (CALL_EXPR_FN (exp), 0);
enum rs6000_builtins fcode = (enum rs6000_builtins) DECL_FUNCTION_CODE (fndecl);
const struct builtin_description *d;
size_t i;
*expandedp = true;
switch (fcode)
{
case PAIRED_BUILTIN_STX:
return paired_expand_stv_builtin (CODE_FOR_paired_stx, exp);
case PAIRED_BUILTIN_LX:
return paired_expand_lv_builtin (CODE_FOR_paired_lx, exp, target);
default:
break;
/* Fall through. */
}
/* Expand the paired predicates. */
d = bdesc_paired_preds;
for (i = 0; i < ARRAY_SIZE (bdesc_paired_preds); i++, d++)
if (d->code == fcode)
return paired_expand_predicate_builtin (d->icode, exp, target);
*expandedp = false;
return NULL_RTX;
}
static rtx
paired_expand_predicate_builtin (enum insn_code icode, tree exp, rtx target)
{
rtx pat, scratch, tmp;
tree form = CALL_EXPR_ARG (exp, 0);
tree arg0 = CALL_EXPR_ARG (exp, 1);
tree arg1 = CALL_EXPR_ARG (exp, 2);
rtx op0 = expand_normal (arg0);
rtx op1 = expand_normal (arg1);
machine_mode mode0 = insn_data[icode].operand[1].mode;
machine_mode mode1 = insn_data[icode].operand[2].mode;
int form_int;
enum rtx_code code;
if (TREE_CODE (form) != INTEGER_CST)
{
error ("argument 1 of %s must be a constant",
"__builtin_paired_predicate");
return const0_rtx;
}
else
form_int = TREE_INT_CST_LOW (form);
gcc_assert (mode0 == mode1);
if (arg0 == error_mark_node || arg1 == error_mark_node)
return const0_rtx;
if (target == 0
|| GET_MODE (target) != SImode
|| !(*insn_data[icode].operand[0].predicate) (target, SImode))
target = gen_reg_rtx (SImode);
if (!(*insn_data[icode].operand[1].predicate) (op0, mode0))
op0 = copy_to_mode_reg (mode0, op0);
if (!(*insn_data[icode].operand[2].predicate) (op1, mode1))
op1 = copy_to_mode_reg (mode1, op1);
scratch = gen_reg_rtx (CCFPmode);
pat = GEN_FCN (icode) (scratch, op0, op1);
if (!pat)
return const0_rtx;
emit_insn (pat);
switch (form_int)
{
/* LT bit. */
case 0:
code = LT;
break;
/* GT bit. */
case 1:
code = GT;
break;
/* EQ bit. */
case 2:
code = EQ;
break;
/* UN bit. */
case 3:
emit_insn (gen_move_from_CR_ov_bit (target, scratch));
return target;
default:
error ("argument 1 of %qs is out of range",
"__builtin_paired_predicate");
return const0_rtx;
}
tmp = gen_rtx_fmt_ee (code, SImode, scratch, const0_rtx);
emit_move_insn (target, tmp);
return target;
}
/* Raise an error message for a builtin function that is called without the
appropriate target options being set. */
static void
rs6000_invalid_builtin (enum rs6000_builtins fncode)
{
size_t uns_fncode = (size_t) fncode;
const char *name = rs6000_builtin_info[uns_fncode].name;
HOST_WIDE_INT fnmask = rs6000_builtin_info[uns_fncode].mask;
gcc_assert (name != NULL);
if ((fnmask & RS6000_BTM_CELL) != 0)
error ("builtin function %qs is only valid for the cell processor", name);
else if ((fnmask & RS6000_BTM_VSX) != 0)
error ("builtin function %qs requires the %qs option", name, "-mvsx");
else if ((fnmask & RS6000_BTM_HTM) != 0)
error ("builtin function %qs requires the %qs option", name, "-mhtm");
else if ((fnmask & RS6000_BTM_ALTIVEC) != 0)
error ("builtin function %qs requires the %qs option", name, "-maltivec");
else if ((fnmask & RS6000_BTM_PAIRED) != 0)
error ("builtin function %qs requires the %qs option", name, "-mpaired");
else if ((fnmask & (RS6000_BTM_DFP | RS6000_BTM_P8_VECTOR))
== (RS6000_BTM_DFP | RS6000_BTM_P8_VECTOR))
error ("builtin function %qs requires the %qs and %qs options",
name, "-mhard-dfp", "-mpower8-vector");
else if ((fnmask & RS6000_BTM_DFP) != 0)
error ("builtin function %qs requires the %qs option", name, "-mhard-dfp");
else if ((fnmask & RS6000_BTM_P8_VECTOR) != 0)
error ("builtin function %qs requires the %qs option", name,
"-mpower8-vector");
else if ((fnmask & (RS6000_BTM_P9_VECTOR | RS6000_BTM_64BIT))
== (RS6000_BTM_P9_VECTOR | RS6000_BTM_64BIT))
error ("builtin function %qs requires the %qs and %qs options",
name, "-mcpu=power9", "-m64");
else if ((fnmask & RS6000_BTM_P9_VECTOR) != 0)
error ("builtin function %qs requires the %qs option", name,
"-mcpu=power9");
else if ((fnmask & (RS6000_BTM_P9_MISC | RS6000_BTM_64BIT))
== (RS6000_BTM_P9_MISC | RS6000_BTM_64BIT))
error ("builtin function %qs requires the %qs and %qs options",
name, "-mcpu=power9", "-m64");
else if ((fnmask & RS6000_BTM_P9_MISC) == RS6000_BTM_P9_MISC)
error ("builtin function %qs requires the %qs option", name,
"-mcpu=power9");
else if ((fnmask & (RS6000_BTM_HARD_FLOAT | RS6000_BTM_LDBL128))
== (RS6000_BTM_HARD_FLOAT | RS6000_BTM_LDBL128))
error ("builtin function %qs requires the %qs and %qs options",
name, "-mhard-float", "-mlong-double-128");
else if ((fnmask & RS6000_BTM_HARD_FLOAT) != 0)
error ("builtin function %qs requires the %qs option", name,
"-mhard-float");
else if ((fnmask & RS6000_BTM_FLOAT128_HW) != 0)
error ("builtin function %qs requires ISA 3.0 IEEE 128-bit floating point",
name);
else if ((fnmask & RS6000_BTM_FLOAT128) != 0)
error ("builtin function %qs requires the %qs option", name, "-mfloat128");
else
error ("builtin function %qs is not supported with the current options",
name);
}
/* Target hook for early folding of built-ins, shamelessly stolen
from ia64.c. */
static tree
rs6000_fold_builtin (tree fndecl ATTRIBUTE_UNUSED,
int n_args ATTRIBUTE_UNUSED,
tree *args ATTRIBUTE_UNUSED,
bool ignore ATTRIBUTE_UNUSED)
{
#ifdef SUBTARGET_FOLD_BUILTIN
return SUBTARGET_FOLD_BUILTIN (fndecl, n_args, args, ignore);
#else
return NULL_TREE;
#endif
}
/* Helper function to sort out which built-ins may be valid without having
a LHS. */
static bool
rs6000_builtin_valid_without_lhs (enum rs6000_builtins fn_code)
{
switch (fn_code)
{
case ALTIVEC_BUILTIN_STVX_V16QI:
case ALTIVEC_BUILTIN_STVX_V8HI:
case ALTIVEC_BUILTIN_STVX_V4SI:
case ALTIVEC_BUILTIN_STVX_V4SF:
case ALTIVEC_BUILTIN_STVX_V2DI:
case ALTIVEC_BUILTIN_STVX_V2DF:
return true;
default:
return false;
}
}
/* Helper function to handle the gimple folding of a vector compare
operation. This sets up true/false vectors, and uses the
VEC_COND_EXPR operation.
CODE indicates which comparison is to be made. (EQ, GT, ...).
TYPE indicates the type of the result. */
static tree
fold_build_vec_cmp (tree_code code, tree type,
tree arg0, tree arg1)
{
tree cmp_type = build_same_sized_truth_vector_type (type);
tree zero_vec = build_zero_cst (type);
tree minus_one_vec = build_minus_one_cst (type);
tree cmp = fold_build2 (code, cmp_type, arg0, arg1);
return fold_build3 (VEC_COND_EXPR, type, cmp, minus_one_vec, zero_vec);
}
/* Helper function to handle the in-between steps for the
vector compare built-ins. */
static void
fold_compare_helper (gimple_stmt_iterator *gsi, tree_code code, gimple *stmt)
{
tree arg0 = gimple_call_arg (stmt, 0);
tree arg1 = gimple_call_arg (stmt, 1);
tree lhs = gimple_call_lhs (stmt);
tree cmp = fold_build_vec_cmp (code, TREE_TYPE (lhs), arg0, arg1);
gimple *g = gimple_build_assign (lhs, cmp);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
}
/* Fold a machine-dependent built-in in GIMPLE. (For folding into
a constant, use rs6000_fold_builtin.) */
bool
rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi)
{
gimple *stmt = gsi_stmt (*gsi);
tree fndecl = gimple_call_fndecl (stmt);
gcc_checking_assert (fndecl && DECL_BUILT_IN_CLASS (fndecl) == BUILT_IN_MD);
enum rs6000_builtins fn_code
= (enum rs6000_builtins) DECL_FUNCTION_CODE (fndecl);
tree arg0, arg1, lhs, temp;
gimple *g;
size_t uns_fncode = (size_t) fn_code;
enum insn_code icode = rs6000_builtin_info[uns_fncode].icode;
const char *fn_name1 = rs6000_builtin_info[uns_fncode].name;
const char *fn_name2 = (icode != CODE_FOR_nothing)
? get_insn_name ((int) icode)
: "nothing";
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_gimple_fold_builtin %d %s %s\n",
fn_code, fn_name1, fn_name2);
if (!rs6000_fold_gimple)
return false;
/* Prevent gimple folding for code that does not have a LHS, unless it is
allowed per the rs6000_builtin_valid_without_lhs helper function. */
if (!gimple_call_lhs (stmt) && !rs6000_builtin_valid_without_lhs (fn_code))
return false;
/* Don't fold invalid builtins, let rs6000_expand_builtin diagnose it. */
HOST_WIDE_INT mask = rs6000_builtin_info[uns_fncode].mask;
bool func_valid_p = (rs6000_builtin_mask & mask) == mask;
if (!func_valid_p)
return false;
switch (fn_code)
{
/* Flavors of vec_add. We deliberately don't expand
P8V_BUILTIN_VADDUQM as it gets lowered from V1TImode to
TImode, resulting in much poorer code generation. */
case ALTIVEC_BUILTIN_VADDUBM:
case ALTIVEC_BUILTIN_VADDUHM:
case ALTIVEC_BUILTIN_VADDUWM:
case P8V_BUILTIN_VADDUDM:
case ALTIVEC_BUILTIN_VADDFP:
case VSX_BUILTIN_XVADDDP:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, PLUS_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vec_sub. We deliberately don't expand
P8V_BUILTIN_VSUBUQM. */
case ALTIVEC_BUILTIN_VSUBUBM:
case ALTIVEC_BUILTIN_VSUBUHM:
case ALTIVEC_BUILTIN_VSUBUWM:
case P8V_BUILTIN_VSUBUDM:
case ALTIVEC_BUILTIN_VSUBFP:
case VSX_BUILTIN_XVSUBDP:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, MINUS_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
case VSX_BUILTIN_XVMULSP:
case VSX_BUILTIN_XVMULDP:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, MULT_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Even element flavors of vec_mul (signed). */
case ALTIVEC_BUILTIN_VMULESB:
case ALTIVEC_BUILTIN_VMULESH:
case ALTIVEC_BUILTIN_VMULESW:
/* Even element flavors of vec_mul (unsigned). */
case ALTIVEC_BUILTIN_VMULEUB:
case ALTIVEC_BUILTIN_VMULEUH:
case ALTIVEC_BUILTIN_VMULEUW:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, VEC_WIDEN_MULT_EVEN_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Odd element flavors of vec_mul (signed). */
case ALTIVEC_BUILTIN_VMULOSB:
case ALTIVEC_BUILTIN_VMULOSH:
case ALTIVEC_BUILTIN_VMULOSW:
/* Odd element flavors of vec_mul (unsigned). */
case ALTIVEC_BUILTIN_VMULOUB:
case ALTIVEC_BUILTIN_VMULOUH:
case ALTIVEC_BUILTIN_VMULOUW:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, VEC_WIDEN_MULT_ODD_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vec_div (Integer). */
case VSX_BUILTIN_DIV_V2DI:
case VSX_BUILTIN_UDIV_V2DI:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, TRUNC_DIV_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vec_div (Float). */
case VSX_BUILTIN_XVDIVSP:
case VSX_BUILTIN_XVDIVDP:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, RDIV_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vec_and. */
case ALTIVEC_BUILTIN_VAND:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, BIT_AND_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vec_andc. */
case ALTIVEC_BUILTIN_VANDC:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
temp = create_tmp_reg_or_ssa_name (TREE_TYPE (arg1));
g = gimple_build_assign (temp, BIT_NOT_EXPR, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_insert_before (gsi, g, GSI_SAME_STMT);
g = gimple_build_assign (lhs, BIT_AND_EXPR, arg0, temp);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vec_nand. */
case P8V_BUILTIN_VEC_NAND:
case P8V_BUILTIN_NAND_V16QI:
case P8V_BUILTIN_NAND_V8HI:
case P8V_BUILTIN_NAND_V4SI:
case P8V_BUILTIN_NAND_V4SF:
case P8V_BUILTIN_NAND_V2DF:
case P8V_BUILTIN_NAND_V2DI:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
temp = create_tmp_reg_or_ssa_name (TREE_TYPE (arg1));
g = gimple_build_assign (temp, BIT_AND_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_insert_before (gsi, g, GSI_SAME_STMT);
g = gimple_build_assign (lhs, BIT_NOT_EXPR, temp);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vec_or. */
case ALTIVEC_BUILTIN_VOR:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, BIT_IOR_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* flavors of vec_orc. */
case P8V_BUILTIN_ORC_V16QI:
case P8V_BUILTIN_ORC_V8HI:
case P8V_BUILTIN_ORC_V4SI:
case P8V_BUILTIN_ORC_V4SF:
case P8V_BUILTIN_ORC_V2DF:
case P8V_BUILTIN_ORC_V2DI:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
temp = create_tmp_reg_or_ssa_name (TREE_TYPE (arg1));
g = gimple_build_assign (temp, BIT_NOT_EXPR, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_insert_before (gsi, g, GSI_SAME_STMT);
g = gimple_build_assign (lhs, BIT_IOR_EXPR, arg0, temp);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vec_xor. */
case ALTIVEC_BUILTIN_VXOR:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, BIT_XOR_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vec_nor. */
case ALTIVEC_BUILTIN_VNOR:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
temp = create_tmp_reg_or_ssa_name (TREE_TYPE (arg1));
g = gimple_build_assign (temp, BIT_IOR_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_insert_before (gsi, g, GSI_SAME_STMT);
g = gimple_build_assign (lhs, BIT_NOT_EXPR, temp);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* flavors of vec_abs. */
case ALTIVEC_BUILTIN_ABS_V16QI:
case ALTIVEC_BUILTIN_ABS_V8HI:
case ALTIVEC_BUILTIN_ABS_V4SI:
case ALTIVEC_BUILTIN_ABS_V4SF:
case P8V_BUILTIN_ABS_V2DI:
case VSX_BUILTIN_XVABSDP:
arg0 = gimple_call_arg (stmt, 0);
if (INTEGRAL_TYPE_P (TREE_TYPE (TREE_TYPE (arg0)))
&& !TYPE_OVERFLOW_WRAPS (TREE_TYPE (TREE_TYPE (arg0))))
return false;
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, ABS_EXPR, arg0);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* flavors of vec_min. */
case VSX_BUILTIN_XVMINDP:
case P8V_BUILTIN_VMINSD:
case P8V_BUILTIN_VMINUD:
case ALTIVEC_BUILTIN_VMINSB:
case ALTIVEC_BUILTIN_VMINSH:
case ALTIVEC_BUILTIN_VMINSW:
case ALTIVEC_BUILTIN_VMINUB:
case ALTIVEC_BUILTIN_VMINUH:
case ALTIVEC_BUILTIN_VMINUW:
case ALTIVEC_BUILTIN_VMINFP:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, MIN_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* flavors of vec_max. */
case VSX_BUILTIN_XVMAXDP:
case P8V_BUILTIN_VMAXSD:
case P8V_BUILTIN_VMAXUD:
case ALTIVEC_BUILTIN_VMAXSB:
case ALTIVEC_BUILTIN_VMAXSH:
case ALTIVEC_BUILTIN_VMAXSW:
case ALTIVEC_BUILTIN_VMAXUB:
case ALTIVEC_BUILTIN_VMAXUH:
case ALTIVEC_BUILTIN_VMAXUW:
case ALTIVEC_BUILTIN_VMAXFP:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, MAX_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vec_eqv. */
case P8V_BUILTIN_EQV_V16QI:
case P8V_BUILTIN_EQV_V8HI:
case P8V_BUILTIN_EQV_V4SI:
case P8V_BUILTIN_EQV_V4SF:
case P8V_BUILTIN_EQV_V2DF:
case P8V_BUILTIN_EQV_V2DI:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
temp = create_tmp_reg_or_ssa_name (TREE_TYPE (arg1));
g = gimple_build_assign (temp, BIT_XOR_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_insert_before (gsi, g, GSI_SAME_STMT);
g = gimple_build_assign (lhs, BIT_NOT_EXPR, temp);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vec_rotate_left. */
case ALTIVEC_BUILTIN_VRLB:
case ALTIVEC_BUILTIN_VRLH:
case ALTIVEC_BUILTIN_VRLW:
case P8V_BUILTIN_VRLD:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, LROTATE_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vector shift right algebraic.
vec_sra{b,h,w} -> vsra{b,h,w}. */
case ALTIVEC_BUILTIN_VSRAB:
case ALTIVEC_BUILTIN_VSRAH:
case ALTIVEC_BUILTIN_VSRAW:
case P8V_BUILTIN_VSRAD:
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, RSHIFT_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vector shift left.
builtin_altivec_vsl{b,h,w} -> vsl{b,h,w}. */
case ALTIVEC_BUILTIN_VSLB:
case ALTIVEC_BUILTIN_VSLH:
case ALTIVEC_BUILTIN_VSLW:
case P8V_BUILTIN_VSLD:
arg0 = gimple_call_arg (stmt, 0);
if (INTEGRAL_TYPE_P (TREE_TYPE (TREE_TYPE (arg0)))
&& !TYPE_OVERFLOW_WRAPS (TREE_TYPE (TREE_TYPE (arg0))))
return false;
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
g = gimple_build_assign (lhs, LSHIFT_EXPR, arg0, arg1);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
/* Flavors of vector shift right. */
case ALTIVEC_BUILTIN_VSRB:
case ALTIVEC_BUILTIN_VSRH:
case ALTIVEC_BUILTIN_VSRW:
case P8V_BUILTIN_VSRD:
{
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
lhs = gimple_call_lhs (stmt);
gimple_seq stmts = NULL;
/* Convert arg0 to unsigned. */
tree arg0_unsigned
= gimple_build (&stmts, VIEW_CONVERT_EXPR,
unsigned_type_for (TREE_TYPE (arg0)), arg0);
tree res
= gimple_build (&stmts, RSHIFT_EXPR,
TREE_TYPE (arg0_unsigned), arg0_unsigned, arg1);
/* Convert result back to the lhs type. */
res = gimple_build (&stmts, VIEW_CONVERT_EXPR, TREE_TYPE (lhs), res);
gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT);
update_call_from_tree (gsi, res);
return true;
}
/* Vector loads. */
case ALTIVEC_BUILTIN_LVX_V16QI:
case ALTIVEC_BUILTIN_LVX_V8HI:
case ALTIVEC_BUILTIN_LVX_V4SI:
case ALTIVEC_BUILTIN_LVX_V4SF:
case ALTIVEC_BUILTIN_LVX_V2DI:
case ALTIVEC_BUILTIN_LVX_V2DF:
{
arg0 = gimple_call_arg (stmt, 0); // offset
arg1 = gimple_call_arg (stmt, 1); // address
/* Do not fold for -maltivec=be on LE targets. */
if (VECTOR_ELT_ORDER_BIG && !BYTES_BIG_ENDIAN)
return false;
lhs = gimple_call_lhs (stmt);
location_t loc = gimple_location (stmt);
/* Since arg1 may be cast to a different type, just use ptr_type_node
here instead of trying to enforce TBAA on pointer types. */
tree arg1_type = ptr_type_node;
tree lhs_type = TREE_TYPE (lhs);
/* POINTER_PLUS_EXPR wants the offset to be of type 'sizetype'. Create
the tree using the value from arg0. The resulting type will match
the type of arg1. */
gimple_seq stmts = NULL;
tree temp_offset = gimple_convert (&stmts, loc, sizetype, arg0);
tree temp_addr = gimple_build (&stmts, loc, POINTER_PLUS_EXPR,
arg1_type, arg1, temp_offset);
/* Mask off any lower bits from the address. */
tree aligned_addr = gimple_build (&stmts, loc, BIT_AND_EXPR,
arg1_type, temp_addr,
build_int_cst (arg1_type, -16));
gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT);
/* Use the build2 helper to set up the mem_ref. The MEM_REF could also
take an offset, but since we've already incorporated the offset
above, here we just pass in a zero. */
gimple *g
= gimple_build_assign (lhs, build2 (MEM_REF, lhs_type, aligned_addr,
build_int_cst (arg1_type, 0)));
gimple_set_location (g, loc);
gsi_replace (gsi, g, true);
return true;
}
/* Vector stores. */
case ALTIVEC_BUILTIN_STVX_V16QI:
case ALTIVEC_BUILTIN_STVX_V8HI:
case ALTIVEC_BUILTIN_STVX_V4SI:
case ALTIVEC_BUILTIN_STVX_V4SF:
case ALTIVEC_BUILTIN_STVX_V2DI:
case ALTIVEC_BUILTIN_STVX_V2DF:
{
/* Do not fold for -maltivec=be on LE targets. */
if (VECTOR_ELT_ORDER_BIG && !BYTES_BIG_ENDIAN)
return false;
arg0 = gimple_call_arg (stmt, 0); /* Value to be stored. */
arg1 = gimple_call_arg (stmt, 1); /* Offset. */
tree arg2 = gimple_call_arg (stmt, 2); /* Store-to address. */
location_t loc = gimple_location (stmt);
tree arg0_type = TREE_TYPE (arg0);
/* Use ptr_type_node (no TBAA) for the arg2_type.
FIXME: (Richard) "A proper fix would be to transition this type as
seen from the frontend to GIMPLE, for example in a similar way we
do for MEM_REFs by piggy-backing that on an extra argument, a
constant zero pointer of the alias pointer type to use (which would
also serve as a type indicator of the store itself). I'd use a
target specific internal function for this (not sure if we can have
those target specific, but I guess if it's folded away then that's
fine) and get away with the overload set." */
tree arg2_type = ptr_type_node;
/* POINTER_PLUS_EXPR wants the offset to be of type 'sizetype'. Create
the tree using the value from arg0. The resulting type will match
the type of arg2. */
gimple_seq stmts = NULL;
tree temp_offset = gimple_convert (&stmts, loc, sizetype, arg1);
tree temp_addr = gimple_build (&stmts, loc, POINTER_PLUS_EXPR,
arg2_type, arg2, temp_offset);
/* Mask off any lower bits from the address. */
tree aligned_addr = gimple_build (&stmts, loc, BIT_AND_EXPR,
arg2_type, temp_addr,
build_int_cst (arg2_type, -16));
gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT);
/* The desired gimple result should be similar to:
MEM[(__vector floatD.1407 *)_1] = vf1D.2697; */
gimple *g
= gimple_build_assign (build2 (MEM_REF, arg0_type, aligned_addr,
build_int_cst (arg2_type, 0)), arg0);
gimple_set_location (g, loc);
gsi_replace (gsi, g, true);
return true;
}
/* Vector Fused multiply-add (fma). */
case ALTIVEC_BUILTIN_VMADDFP:
case VSX_BUILTIN_XVMADDDP:
case ALTIVEC_BUILTIN_VMLADDUHM:
{
arg0 = gimple_call_arg (stmt, 0);
arg1 = gimple_call_arg (stmt, 1);
tree arg2 = gimple_call_arg (stmt, 2);
lhs = gimple_call_lhs (stmt);
gimple *g = gimple_build_assign (lhs, FMA_EXPR, arg0, arg1, arg2);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
}
/* Vector compares; EQ, NE, GE, GT, LE. */
case ALTIVEC_BUILTIN_VCMPEQUB:
case ALTIVEC_BUILTIN_VCMPEQUH:
case ALTIVEC_BUILTIN_VCMPEQUW:
case P8V_BUILTIN_VCMPEQUD:
fold_compare_helper (gsi, EQ_EXPR, stmt);
return true;
case P9V_BUILTIN_CMPNEB:
case P9V_BUILTIN_CMPNEH:
case P9V_BUILTIN_CMPNEW:
fold_compare_helper (gsi, NE_EXPR, stmt);
return true;
case VSX_BUILTIN_CMPGE_16QI:
case VSX_BUILTIN_CMPGE_U16QI:
case VSX_BUILTIN_CMPGE_8HI:
case VSX_BUILTIN_CMPGE_U8HI:
case VSX_BUILTIN_CMPGE_4SI:
case VSX_BUILTIN_CMPGE_U4SI:
case VSX_BUILTIN_CMPGE_2DI:
case VSX_BUILTIN_CMPGE_U2DI:
fold_compare_helper (gsi, GE_EXPR, stmt);
return true;
case ALTIVEC_BUILTIN_VCMPGTSB:
case ALTIVEC_BUILTIN_VCMPGTUB:
case ALTIVEC_BUILTIN_VCMPGTSH:
case ALTIVEC_BUILTIN_VCMPGTUH:
case ALTIVEC_BUILTIN_VCMPGTSW:
case ALTIVEC_BUILTIN_VCMPGTUW:
case P8V_BUILTIN_VCMPGTUD:
case P8V_BUILTIN_VCMPGTSD:
fold_compare_helper (gsi, GT_EXPR, stmt);
return true;
case VSX_BUILTIN_CMPLE_16QI:
case VSX_BUILTIN_CMPLE_U16QI:
case VSX_BUILTIN_CMPLE_8HI:
case VSX_BUILTIN_CMPLE_U8HI:
case VSX_BUILTIN_CMPLE_4SI:
case VSX_BUILTIN_CMPLE_U4SI:
case VSX_BUILTIN_CMPLE_2DI:
case VSX_BUILTIN_CMPLE_U2DI:
fold_compare_helper (gsi, LE_EXPR, stmt);
return true;
/* flavors of vec_splat_[us]{8,16,32}. */
case ALTIVEC_BUILTIN_VSPLTISB:
case ALTIVEC_BUILTIN_VSPLTISH:
case ALTIVEC_BUILTIN_VSPLTISW:
{
arg0 = gimple_call_arg (stmt, 0);
lhs = gimple_call_lhs (stmt);
/* Only fold the vec_splat_*() if arg0 is constant. */
if (TREE_CODE (arg0) != INTEGER_CST)
return false;
gimple_seq stmts = NULL;
location_t loc = gimple_location (stmt);
tree splat_value = gimple_convert (&stmts, loc,
TREE_TYPE (TREE_TYPE (lhs)), arg0);
gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT);
tree splat_tree = build_vector_from_val (TREE_TYPE (lhs), splat_value);
g = gimple_build_assign (lhs, splat_tree);
gimple_set_location (g, gimple_location (stmt));
gsi_replace (gsi, g, true);
return true;
}
default:
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "gimple builtin intrinsic not matched:%d %s %s\n",
fn_code, fn_name1, fn_name2);
break;
}
return false;
}
/* Expand an expression EXP that calls a built-in function,
with result going to TARGET if that's convenient
(and in mode MODE if that's convenient).
SUBTARGET may be used as the target for computing one of EXP's operands.
IGNORE is nonzero if the value is to be ignored. */
static rtx
rs6000_expand_builtin (tree exp, rtx target, rtx subtarget ATTRIBUTE_UNUSED,
machine_mode mode ATTRIBUTE_UNUSED,
int ignore ATTRIBUTE_UNUSED)
{
tree fndecl = TREE_OPERAND (CALL_EXPR_FN (exp), 0);
enum rs6000_builtins fcode
= (enum rs6000_builtins)DECL_FUNCTION_CODE (fndecl);
size_t uns_fcode = (size_t)fcode;
const struct builtin_description *d;
size_t i;
rtx ret;
bool success;
HOST_WIDE_INT mask = rs6000_builtin_info[uns_fcode].mask;
bool func_valid_p = ((rs6000_builtin_mask & mask) == mask);
enum insn_code icode = rs6000_builtin_info[uns_fcode].icode;
/* We have two different modes (KFmode, TFmode) that are the IEEE 128-bit
floating point type, depending on whether long double is the IBM extended
double (KFmode) or long double is IEEE 128-bit (TFmode). It is simpler if
we only define one variant of the built-in function, and switch the code
when defining it, rather than defining two built-ins and using the
overload table in rs6000-c.c to switch between the two. If we don't have
the proper assembler, don't do this switch because CODE_FOR_*kf* and
CODE_FOR_*tf* will be CODE_FOR_nothing. */
#ifdef HAVE_AS_POWER9
if (FLOAT128_IEEE_P (TFmode))
switch (icode)
{
default:
break;
case CODE_FOR_sqrtkf2_odd: icode = CODE_FOR_sqrttf2_odd; break;
case CODE_FOR_trunckfdf2_odd: icode = CODE_FOR_trunctfdf2_odd; break;
case CODE_FOR_addkf3_odd: icode = CODE_FOR_addtf3_odd; break;
case CODE_FOR_subkf3_odd: icode = CODE_FOR_subtf3_odd; break;
case CODE_FOR_mulkf3_odd: icode = CODE_FOR_multf3_odd; break;
case CODE_FOR_divkf3_odd: icode = CODE_FOR_divtf3_odd; break;
case CODE_FOR_fmakf4_odd: icode = CODE_FOR_fmatf4_odd; break;
case CODE_FOR_xsxexpqp_kf: icode = CODE_FOR_xsxexpqp_tf; break;
case CODE_FOR_xsxsigqp_kf: icode = CODE_FOR_xsxsigqp_tf; break;
case CODE_FOR_xststdcnegqp_kf: icode = CODE_FOR_xststdcnegqp_tf; break;
case CODE_FOR_xsiexpqp_kf: icode = CODE_FOR_xsiexpqp_tf; break;
case CODE_FOR_xsiexpqpf_kf: icode = CODE_FOR_xsiexpqpf_tf; break;
case CODE_FOR_xststdcqp_kf: icode = CODE_FOR_xststdcqp_tf; break;
}
#endif
if (TARGET_DEBUG_BUILTIN)
{
const char *name1 = rs6000_builtin_info[uns_fcode].name;
const char *name2 = (icode != CODE_FOR_nothing)
? get_insn_name ((int) icode)
: "nothing";
const char *name3;
switch (rs6000_builtin_info[uns_fcode].attr & RS6000_BTC_TYPE_MASK)
{
default: name3 = "unknown"; break;
case RS6000_BTC_SPECIAL: name3 = "special"; break;
case RS6000_BTC_UNARY: name3 = "unary"; break;
case RS6000_BTC_BINARY: name3 = "binary"; break;
case RS6000_BTC_TERNARY: name3 = "ternary"; break;
case RS6000_BTC_PREDICATE: name3 = "predicate"; break;
case RS6000_BTC_ABS: name3 = "abs"; break;
case RS6000_BTC_DST: name3 = "dst"; break;
}
fprintf (stderr,
"rs6000_expand_builtin, %s (%d), insn = %s (%d), type=%s%s\n",
(name1) ? name1 : "---", fcode,
(name2) ? name2 : "---", (int) icode,
name3,
func_valid_p ? "" : ", not valid");
}
if (!func_valid_p)
{
rs6000_invalid_builtin (fcode);
/* Given it is invalid, just generate a normal call. */
return expand_call (exp, target, ignore);
}
switch (fcode)
{
case RS6000_BUILTIN_RECIP:
return rs6000_expand_binop_builtin (CODE_FOR_recipdf3, exp, target);
case RS6000_BUILTIN_RECIPF:
return rs6000_expand_binop_builtin (CODE_FOR_recipsf3, exp, target);
case RS6000_BUILTIN_RSQRTF:
return rs6000_expand_unop_builtin (CODE_FOR_rsqrtsf2, exp, target);
case RS6000_BUILTIN_RSQRT:
return rs6000_expand_unop_builtin (CODE_FOR_rsqrtdf2, exp, target);
case POWER7_BUILTIN_BPERMD:
return rs6000_expand_binop_builtin (((TARGET_64BIT)
? CODE_FOR_bpermd_di
: CODE_FOR_bpermd_si), exp, target);
case RS6000_BUILTIN_GET_TB:
return rs6000_expand_zeroop_builtin (CODE_FOR_rs6000_get_timebase,
target);
case RS6000_BUILTIN_MFTB:
return rs6000_expand_zeroop_builtin (((TARGET_64BIT)
? CODE_FOR_rs6000_mftb_di
: CODE_FOR_rs6000_mftb_si),
target);
case RS6000_BUILTIN_MFFS:
return rs6000_expand_zeroop_builtin (CODE_FOR_rs6000_mffs, target);
case RS6000_BUILTIN_MTFSF:
return rs6000_expand_mtfsf_builtin (CODE_FOR_rs6000_mtfsf, exp);
case RS6000_BUILTIN_CPU_INIT:
case RS6000_BUILTIN_CPU_IS:
case RS6000_BUILTIN_CPU_SUPPORTS:
return cpu_expand_builtin (fcode, exp, target);
case ALTIVEC_BUILTIN_MASK_FOR_LOAD:
case ALTIVEC_BUILTIN_MASK_FOR_STORE:
{
int icode2 = (BYTES_BIG_ENDIAN ? (int) CODE_FOR_altivec_lvsr_direct
: (int) CODE_FOR_altivec_lvsl_direct);
machine_mode tmode = insn_data[icode2].operand[0].mode;
machine_mode mode = insn_data[icode2].operand[1].mode;
tree arg;
rtx op, addr, pat;
gcc_assert (TARGET_ALTIVEC);
arg = CALL_EXPR_ARG (exp, 0);
gcc_assert (POINTER_TYPE_P (TREE_TYPE (arg)));
op = expand_expr (arg, NULL_RTX, Pmode, EXPAND_NORMAL);
addr = memory_address (mode, op);
if (fcode == ALTIVEC_BUILTIN_MASK_FOR_STORE)
op = addr;
else
{
/* For the load case need to negate the address. */
op = gen_reg_rtx (GET_MODE (addr));
emit_insn (gen_rtx_SET (op, gen_rtx_NEG (GET_MODE (addr), addr)));
}
op = gen_rtx_MEM (mode, op);
if (target == 0
|| GET_MODE (target) != tmode
|| ! (*insn_data[icode2].operand[0].predicate) (target, tmode))
target = gen_reg_rtx (tmode);
pat = GEN_FCN (icode2) (target, op);
if (!pat)
return 0;
emit_insn (pat);
return target;
}
case ALTIVEC_BUILTIN_VCFUX:
case ALTIVEC_BUILTIN_VCFSX:
case ALTIVEC_BUILTIN_VCTUXS:
case ALTIVEC_BUILTIN_VCTSXS:
/* FIXME: There's got to be a nicer way to handle this case than
constructing a new CALL_EXPR. */
if (call_expr_nargs (exp) == 1)
{
exp = build_call_nary (TREE_TYPE (exp), CALL_EXPR_FN (exp),
2, CALL_EXPR_ARG (exp, 0), integer_zero_node);
}
break;
default:
break;
}
if (TARGET_ALTIVEC)
{
ret = altivec_expand_builtin (exp, target, &success);
if (success)
return ret;
}
if (TARGET_PAIRED_FLOAT)
{
ret = paired_expand_builtin (exp, target, &success);
if (success)
return ret;
}
if (TARGET_HTM)
{
ret = htm_expand_builtin (exp, target, &success);
if (success)
return ret;
}
unsigned attr = rs6000_builtin_info[uns_fcode].attr & RS6000_BTC_TYPE_MASK;
/* RS6000_BTC_SPECIAL represents no-operand operators. */
gcc_assert (attr == RS6000_BTC_UNARY
|| attr == RS6000_BTC_BINARY
|| attr == RS6000_BTC_TERNARY
|| attr == RS6000_BTC_SPECIAL);
/* Handle simple unary operations. */
d = bdesc_1arg;
for (i = 0; i < ARRAY_SIZE (bdesc_1arg); i++, d++)
if (d->code == fcode)
return rs6000_expand_unop_builtin (icode, exp, target);
/* Handle simple binary operations. */
d = bdesc_2arg;
for (i = 0; i < ARRAY_SIZE (bdesc_2arg); i++, d++)
if (d->code == fcode)
return rs6000_expand_binop_builtin (icode, exp, target);
/* Handle simple ternary operations. */
d = bdesc_3arg;
for (i = 0; i < ARRAY_SIZE (bdesc_3arg); i++, d++)
if (d->code == fcode)
return rs6000_expand_ternop_builtin (icode, exp, target);
/* Handle simple no-argument operations. */
d = bdesc_0arg;
for (i = 0; i < ARRAY_SIZE (bdesc_0arg); i++, d++)
if (d->code == fcode)
return rs6000_expand_zeroop_builtin (icode, target);
gcc_unreachable ();
}
/* Create a builtin vector type with a name. Taking care not to give
the canonical type a name. */
static tree
rs6000_vector_type (const char *name, tree elt_type, unsigned num_elts)
{
tree result = build_vector_type (elt_type, num_elts);
/* Copy so we don't give the canonical type a name. */
result = build_variant_type_copy (result);
add_builtin_type (name, result);
return result;
}
static void
rs6000_init_builtins (void)
{
tree tdecl;
tree ftype;
machine_mode mode;
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_init_builtins%s%s%s\n",
(TARGET_PAIRED_FLOAT) ? ", paired" : "",
(TARGET_ALTIVEC) ? ", altivec" : "",
(TARGET_VSX) ? ", vsx" : "");
V2SI_type_node = build_vector_type (intSI_type_node, 2);
V2SF_type_node = build_vector_type (float_type_node, 2);
V2DI_type_node = rs6000_vector_type (TARGET_POWERPC64 ? "__vector long"
: "__vector long long",
intDI_type_node, 2);
V2DF_type_node = rs6000_vector_type ("__vector double", double_type_node, 2);
V4SI_type_node = rs6000_vector_type ("__vector signed int",
intSI_type_node, 4);
V4SF_type_node = rs6000_vector_type ("__vector float", float_type_node, 4);
V8HI_type_node = rs6000_vector_type ("__vector signed short",
intHI_type_node, 8);
V16QI_type_node = rs6000_vector_type ("__vector signed char",
intQI_type_node, 16);
unsigned_V16QI_type_node = rs6000_vector_type ("__vector unsigned char",
unsigned_intQI_type_node, 16);
unsigned_V8HI_type_node = rs6000_vector_type ("__vector unsigned short",
unsigned_intHI_type_node, 8);
unsigned_V4SI_type_node = rs6000_vector_type ("__vector unsigned int",
unsigned_intSI_type_node, 4);
unsigned_V2DI_type_node = rs6000_vector_type (TARGET_POWERPC64
? "__vector unsigned long"
: "__vector unsigned long long",
unsigned_intDI_type_node, 2);
opaque_V2SF_type_node = build_opaque_vector_type (float_type_node, 2);
opaque_V2SI_type_node = build_opaque_vector_type (intSI_type_node, 2);
opaque_p_V2SI_type_node = build_pointer_type (opaque_V2SI_type_node);
opaque_V4SI_type_node = build_opaque_vector_type (intSI_type_node, 4);
const_str_type_node
= build_pointer_type (build_qualified_type (char_type_node,
TYPE_QUAL_CONST));
/* We use V1TI mode as a special container to hold __int128_t items that
must live in VSX registers. */
if (intTI_type_node)
{
V1TI_type_node = rs6000_vector_type ("__vector __int128",
intTI_type_node, 1);
unsigned_V1TI_type_node
= rs6000_vector_type ("__vector unsigned __int128",
unsigned_intTI_type_node, 1);
}
/* The 'vector bool ...' types must be kept distinct from 'vector unsigned ...'
types, especially in C++ land. Similarly, 'vector pixel' is distinct from
'vector unsigned short'. */
bool_char_type_node = build_distinct_type_copy (unsigned_intQI_type_node);
bool_short_type_node = build_distinct_type_copy (unsigned_intHI_type_node);
bool_int_type_node = build_distinct_type_copy (unsigned_intSI_type_node);
bool_long_type_node = build_distinct_type_copy (unsigned_intDI_type_node);
pixel_type_node = build_distinct_type_copy (unsigned_intHI_type_node);
long_integer_type_internal_node = long_integer_type_node;
long_unsigned_type_internal_node = long_unsigned_type_node;
long_long_integer_type_internal_node = long_long_integer_type_node;
long_long_unsigned_type_internal_node = long_long_unsigned_type_node;
intQI_type_internal_node = intQI_type_node;
uintQI_type_internal_node = unsigned_intQI_type_node;
intHI_type_internal_node = intHI_type_node;
uintHI_type_internal_node = unsigned_intHI_type_node;
intSI_type_internal_node = intSI_type_node;
uintSI_type_internal_node = unsigned_intSI_type_node;
intDI_type_internal_node = intDI_type_node;
uintDI_type_internal_node = unsigned_intDI_type_node;
intTI_type_internal_node = intTI_type_node;
uintTI_type_internal_node = unsigned_intTI_type_node;
float_type_internal_node = float_type_node;
double_type_internal_node = double_type_node;
long_double_type_internal_node = long_double_type_node;
dfloat64_type_internal_node = dfloat64_type_node;
dfloat128_type_internal_node = dfloat128_type_node;
void_type_internal_node = void_type_node;
/* 128-bit floating point support. KFmode is IEEE 128-bit floating point.
IFmode is the IBM extended 128-bit format that is a pair of doubles.
TFmode will be either IEEE 128-bit floating point or the IBM double-double
format that uses a pair of doubles, depending on the switches and
defaults.
If we don't support for either 128-bit IBM double double or IEEE 128-bit
floating point, we need make sure the type is non-zero or else self-test
fails during bootstrap.
We don't register a built-in type for __ibm128 if the type is the same as
long double. Instead we add a #define for __ibm128 in
rs6000_cpu_cpp_builtins to long double.
For IEEE 128-bit floating point, always create the type __ieee128. If the
user used -mfloat128, rs6000-c.c will create a define from __float128 to
__ieee128. */
if (TARGET_LONG_DOUBLE_128 && FLOAT128_IEEE_P (TFmode))
{
ibm128_float_type_node = make_node (REAL_TYPE);
TYPE_PRECISION (ibm128_float_type_node) = 128;
SET_TYPE_MODE (ibm128_float_type_node, IFmode);
layout_type (ibm128_float_type_node);
lang_hooks.types.register_builtin_type (ibm128_float_type_node,
"__ibm128");
}
else
ibm128_float_type_node = long_double_type_node;
if (TARGET_FLOAT128_TYPE)
{
ieee128_float_type_node = float128_type_node;
lang_hooks.types.register_builtin_type (ieee128_float_type_node,
"__ieee128");
}
else
ieee128_float_type_node = long_double_type_node;
/* Initialize the modes for builtin_function_type, mapping a machine mode to
tree type node. */
builtin_mode_to_type[QImode][0] = integer_type_node;
builtin_mode_to_type[HImode][0] = integer_type_node;
builtin_mode_to_type[SImode][0] = intSI_type_node;
builtin_mode_to_type[SImode][1] = unsigned_intSI_type_node;
builtin_mode_to_type[DImode][0] = intDI_type_node;
builtin_mode_to_type[DImode][1] = unsigned_intDI_type_node;
builtin_mode_to_type[TImode][0] = intTI_type_node;
builtin_mode_to_type[TImode][1] = unsigned_intTI_type_node;
builtin_mode_to_type[SFmode][0] = float_type_node;
builtin_mode_to_type[DFmode][0] = double_type_node;
builtin_mode_to_type[IFmode][0] = ibm128_float_type_node;
builtin_mode_to_type[KFmode][0] = ieee128_float_type_node;
builtin_mode_to_type[TFmode][0] = long_double_type_node;
builtin_mode_to_type[DDmode][0] = dfloat64_type_node;
builtin_mode_to_type[TDmode][0] = dfloat128_type_node;
builtin_mode_to_type[V1TImode][0] = V1TI_type_node;
builtin_mode_to_type[V1TImode][1] = unsigned_V1TI_type_node;
builtin_mode_to_type[V2SImode][0] = V2SI_type_node;
builtin_mode_to_type[V2SFmode][0] = V2SF_type_node;
builtin_mode_to_type[V2DImode][0] = V2DI_type_node;
builtin_mode_to_type[V2DImode][1] = unsigned_V2DI_type_node;
builtin_mode_to_type[V2DFmode][0] = V2DF_type_node;
builtin_mode_to_type[V4SImode][0] = V4SI_type_node;
builtin_mode_to_type[V4SImode][1] = unsigned_V4SI_type_node;
builtin_mode_to_type[V4SFmode][0] = V4SF_type_node;
builtin_mode_to_type[V8HImode][0] = V8HI_type_node;
builtin_mode_to_type[V8HImode][1] = unsigned_V8HI_type_node;
builtin_mode_to_type[V16QImode][0] = V16QI_type_node;
builtin_mode_to_type[V16QImode][1] = unsigned_V16QI_type_node;
tdecl = add_builtin_type ("__bool char", bool_char_type_node);
TYPE_NAME (bool_char_type_node) = tdecl;
tdecl = add_builtin_type ("__bool short", bool_short_type_node);
TYPE_NAME (bool_short_type_node) = tdecl;
tdecl = add_builtin_type ("__bool int", bool_int_type_node);
TYPE_NAME (bool_int_type_node) = tdecl;
tdecl = add_builtin_type ("__pixel", pixel_type_node);
TYPE_NAME (pixel_type_node) = tdecl;
bool_V16QI_type_node = rs6000_vector_type ("__vector __bool char",
bool_char_type_node, 16);
bool_V8HI_type_node = rs6000_vector_type ("__vector __bool short",
bool_short_type_node, 8);
bool_V4SI_type_node = rs6000_vector_type ("__vector __bool int",
bool_int_type_node, 4);
bool_V2DI_type_node = rs6000_vector_type (TARGET_POWERPC64
? "__vector __bool long"
: "__vector __bool long long",
bool_long_type_node, 2);
pixel_V8HI_type_node = rs6000_vector_type ("__vector __pixel",
pixel_type_node, 8);
/* Paired builtins are only available if you build a compiler with the
appropriate options, so only create those builtins with the appropriate
compiler option. Create Altivec and VSX builtins on machines with at
least the general purpose extensions (970 and newer) to allow the use of
the target attribute. */
if (TARGET_PAIRED_FLOAT)
paired_init_builtins ();
if (TARGET_EXTRA_BUILTINS)
altivec_init_builtins ();
if (TARGET_HTM)
htm_init_builtins ();
if (TARGET_EXTRA_BUILTINS || TARGET_PAIRED_FLOAT)
rs6000_common_init_builtins ();
ftype = builtin_function_type (DFmode, DFmode, DFmode, VOIDmode,
RS6000_BUILTIN_RECIP, "__builtin_recipdiv");
def_builtin ("__builtin_recipdiv", ftype, RS6000_BUILTIN_RECIP);
ftype = builtin_function_type (SFmode, SFmode, SFmode, VOIDmode,
RS6000_BUILTIN_RECIPF, "__builtin_recipdivf");
def_builtin ("__builtin_recipdivf", ftype, RS6000_BUILTIN_RECIPF);
ftype = builtin_function_type (DFmode, DFmode, VOIDmode, VOIDmode,
RS6000_BUILTIN_RSQRT, "__builtin_rsqrt");
def_builtin ("__builtin_rsqrt", ftype, RS6000_BUILTIN_RSQRT);
ftype = builtin_function_type (SFmode, SFmode, VOIDmode, VOIDmode,
RS6000_BUILTIN_RSQRTF, "__builtin_rsqrtf");
def_builtin ("__builtin_rsqrtf", ftype, RS6000_BUILTIN_RSQRTF);
mode = (TARGET_64BIT) ? DImode : SImode;
ftype = builtin_function_type (mode, mode, mode, VOIDmode,
POWER7_BUILTIN_BPERMD, "__builtin_bpermd");
def_builtin ("__builtin_bpermd", ftype, POWER7_BUILTIN_BPERMD);
ftype = build_function_type_list (unsigned_intDI_type_node,
NULL_TREE);
def_builtin ("__builtin_ppc_get_timebase", ftype, RS6000_BUILTIN_GET_TB);
if (TARGET_64BIT)
ftype = build_function_type_list (unsigned_intDI_type_node,
NULL_TREE);
else
ftype = build_function_type_list (unsigned_intSI_type_node,
NULL_TREE);
def_builtin ("__builtin_ppc_mftb", ftype, RS6000_BUILTIN_MFTB);
ftype = build_function_type_list (double_type_node, NULL_TREE);
def_builtin ("__builtin_mffs", ftype, RS6000_BUILTIN_MFFS);
ftype = build_function_type_list (void_type_node,
intSI_type_node, double_type_node,
NULL_TREE);
def_builtin ("__builtin_mtfsf", ftype, RS6000_BUILTIN_MTFSF);
ftype = build_function_type_list (void_type_node, NULL_TREE);
def_builtin ("__builtin_cpu_init", ftype, RS6000_BUILTIN_CPU_INIT);
ftype = build_function_type_list (bool_int_type_node, const_ptr_type_node,
NULL_TREE);
def_builtin ("__builtin_cpu_is", ftype, RS6000_BUILTIN_CPU_IS);
def_builtin ("__builtin_cpu_supports", ftype, RS6000_BUILTIN_CPU_SUPPORTS);
/* AIX libm provides clog as __clog. */
if (TARGET_XCOFF &&
(tdecl = builtin_decl_explicit (BUILT_IN_CLOG)) != NULL_TREE)
set_user_assembler_name (tdecl, "__clog");
#ifdef SUBTARGET_INIT_BUILTINS
SUBTARGET_INIT_BUILTINS;
#endif
}
/* Returns the rs6000 builtin decl for CODE. */
static tree
rs6000_builtin_decl (unsigned code, bool initialize_p ATTRIBUTE_UNUSED)
{
HOST_WIDE_INT fnmask;
if (code >= RS6000_BUILTIN_COUNT)
return error_mark_node;
fnmask = rs6000_builtin_info[code].mask;
if ((fnmask & rs6000_builtin_mask) != fnmask)
{
rs6000_invalid_builtin ((enum rs6000_builtins)code);
return error_mark_node;
}
return rs6000_builtin_decls[code];
}
static void
paired_init_builtins (void)
{
const struct builtin_description *d;
size_t i;
HOST_WIDE_INT builtin_mask = rs6000_builtin_mask;
tree int_ftype_int_v2sf_v2sf
= build_function_type_list (integer_type_node,
integer_type_node,
V2SF_type_node,
V2SF_type_node,
NULL_TREE);
tree pcfloat_type_node =
build_pointer_type (build_qualified_type
(float_type_node, TYPE_QUAL_CONST));
tree v2sf_ftype_long_pcfloat = build_function_type_list (V2SF_type_node,
long_integer_type_node,
pcfloat_type_node,
NULL_TREE);
tree void_ftype_v2sf_long_pcfloat =
build_function_type_list (void_type_node,
V2SF_type_node,
long_integer_type_node,
pcfloat_type_node,
NULL_TREE);
def_builtin ("__builtin_paired_lx", v2sf_ftype_long_pcfloat,
PAIRED_BUILTIN_LX);
def_builtin ("__builtin_paired_stx", void_ftype_v2sf_long_pcfloat,
PAIRED_BUILTIN_STX);
/* Predicates. */
d = bdesc_paired_preds;
for (i = 0; i < ARRAY_SIZE (bdesc_paired_preds); ++i, d++)
{
tree type;
HOST_WIDE_INT mask = d->mask;
if ((mask & builtin_mask) != mask)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "paired_init_builtins, skip predicate %s\n",
d->name);
continue;
}
/* Cannot define builtin if the instruction is disabled. */
gcc_assert (d->icode != CODE_FOR_nothing);
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "paired pred #%d, insn = %s [%d], mode = %s\n",
(int)i, get_insn_name (d->icode), (int)d->icode,
GET_MODE_NAME (insn_data[d->icode].operand[1].mode));
switch (insn_data[d->icode].operand[1].mode)
{
case E_V2SFmode:
type = int_ftype_int_v2sf_v2sf;
break;
default:
gcc_unreachable ();
}
def_builtin (d->name, type, d->code);
}
}
static void
altivec_init_builtins (void)
{
const struct builtin_description *d;
size_t i;
tree ftype;
tree decl;
HOST_WIDE_INT builtin_mask = rs6000_builtin_mask;
tree pvoid_type_node = build_pointer_type (void_type_node);
tree pcvoid_type_node
= build_pointer_type (build_qualified_type (void_type_node,
TYPE_QUAL_CONST));
tree int_ftype_opaque
= build_function_type_list (integer_type_node,
opaque_V4SI_type_node, NULL_TREE);
tree opaque_ftype_opaque
= build_function_type_list (integer_type_node, NULL_TREE);
tree opaque_ftype_opaque_int
= build_function_type_list (opaque_V4SI_type_node,
opaque_V4SI_type_node, integer_type_node, NULL_TREE);
tree opaque_ftype_opaque_opaque_int
= build_function_type_list (opaque_V4SI_type_node,
opaque_V4SI_type_node, opaque_V4SI_type_node,
integer_type_node, NULL_TREE);
tree opaque_ftype_opaque_opaque_opaque
= build_function_type_list (opaque_V4SI_type_node,
opaque_V4SI_type_node, opaque_V4SI_type_node,
opaque_V4SI_type_node, NULL_TREE);
tree opaque_ftype_opaque_opaque
= build_function_type_list (opaque_V4SI_type_node,
opaque_V4SI_type_node, opaque_V4SI_type_node,
NULL_TREE);
tree int_ftype_int_opaque_opaque
= build_function_type_list (integer_type_node,
integer_type_node, opaque_V4SI_type_node,
opaque_V4SI_type_node, NULL_TREE);
tree int_ftype_int_v4si_v4si
= build_function_type_list (integer_type_node,
integer_type_node, V4SI_type_node,
V4SI_type_node, NULL_TREE);
tree int_ftype_int_v2di_v2di
= build_function_type_list (integer_type_node,
integer_type_node, V2DI_type_node,
V2DI_type_node, NULL_TREE);
tree void_ftype_v4si
= build_function_type_list (void_type_node, V4SI_type_node, NULL_TREE);
tree v8hi_ftype_void
= build_function_type_list (V8HI_type_node, NULL_TREE);
tree void_ftype_void
= build_function_type_list (void_type_node, NULL_TREE);
tree void_ftype_int
= build_function_type_list (void_type_node, integer_type_node, NULL_TREE);
tree opaque_ftype_long_pcvoid
= build_function_type_list (opaque_V4SI_type_node,
long_integer_type_node, pcvoid_type_node,
NULL_TREE);
tree v16qi_ftype_long_pcvoid
= build_function_type_list (V16QI_type_node,
long_integer_type_node, pcvoid_type_node,
NULL_TREE);
tree v8hi_ftype_long_pcvoid
= build_function_type_list (V8HI_type_node,
long_integer_type_node, pcvoid_type_node,
NULL_TREE);
tree v4si_ftype_long_pcvoid
= build_function_type_list (V4SI_type_node,
long_integer_type_node, pcvoid_type_node,
NULL_TREE);
tree v4sf_ftype_long_pcvoid
= build_function_type_list (V4SF_type_node,
long_integer_type_node, pcvoid_type_node,
NULL_TREE);
tree v2df_ftype_long_pcvoid
= build_function_type_list (V2DF_type_node,
long_integer_type_node, pcvoid_type_node,
NULL_TREE);
tree v2di_ftype_long_pcvoid
= build_function_type_list (V2DI_type_node,
long_integer_type_node, pcvoid_type_node,
NULL_TREE);
tree void_ftype_opaque_long_pvoid
= build_function_type_list (void_type_node,
opaque_V4SI_type_node, long_integer_type_node,
pvoid_type_node, NULL_TREE);
tree void_ftype_v4si_long_pvoid
= build_function_type_list (void_type_node,
V4SI_type_node, long_integer_type_node,
pvoid_type_node, NULL_TREE);
tree void_ftype_v16qi_long_pvoid
= build_function_type_list (void_type_node,
V16QI_type_node, long_integer_type_node,
pvoid_type_node, NULL_TREE);
tree void_ftype_v16qi_pvoid_long
= build_function_type_list (void_type_node,
V16QI_type_node, pvoid_type_node,
long_integer_type_node, NULL_TREE);
tree void_ftype_v8hi_long_pvoid
= build_function_type_list (void_type_node,
V8HI_type_node, long_integer_type_node,
pvoid_type_node, NULL_TREE);
tree void_ftype_v4sf_long_pvoid
= build_function_type_list (void_type_node,
V4SF_type_node, long_integer_type_node,
pvoid_type_node, NULL_TREE);
tree void_ftype_v2df_long_pvoid
= build_function_type_list (void_type_node,
V2DF_type_node, long_integer_type_node,
pvoid_type_node, NULL_TREE);
tree void_ftype_v2di_long_pvoid
= build_function_type_list (void_type_node,
V2DI_type_node, long_integer_type_node,
pvoid_type_node, NULL_TREE);
tree int_ftype_int_v8hi_v8hi
= build_function_type_list (integer_type_node,
integer_type_node, V8HI_type_node,
V8HI_type_node, NULL_TREE);
tree int_ftype_int_v16qi_v16qi
= build_function_type_list (integer_type_node,
integer_type_node, V16QI_type_node,
V16QI_type_node, NULL_TREE);
tree int_ftype_int_v4sf_v4sf
= build_function_type_list (integer_type_node,
integer_type_node, V4SF_type_node,
V4SF_type_node, NULL_TREE);
tree int_ftype_int_v2df_v2df
= build_function_type_list (integer_type_node,
integer_type_node, V2DF_type_node,
V2DF_type_node, NULL_TREE);
tree v2di_ftype_v2di
= build_function_type_list (V2DI_type_node, V2DI_type_node, NULL_TREE);
tree v4si_ftype_v4si
= build_function_type_list (V4SI_type_node, V4SI_type_node, NULL_TREE);
tree v8hi_ftype_v8hi
= build_function_type_list (V8HI_type_node, V8HI_type_node, NULL_TREE);
tree v16qi_ftype_v16qi
= build_function_type_list (V16QI_type_node, V16QI_type_node, NULL_TREE);
tree v4sf_ftype_v4sf
= build_function_type_list (V4SF_type_node, V4SF_type_node, NULL_TREE);
tree v2df_ftype_v2df
= build_function_type_list (V2DF_type_node, V2DF_type_node, NULL_TREE);
tree void_ftype_pcvoid_int_int
= build_function_type_list (void_type_node,
pcvoid_type_node, integer_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_altivec_mtvscr", void_ftype_v4si, ALTIVEC_BUILTIN_MTVSCR);
def_builtin ("__builtin_altivec_mfvscr", v8hi_ftype_void, ALTIVEC_BUILTIN_MFVSCR);
def_builtin ("__builtin_altivec_dssall", void_ftype_void, ALTIVEC_BUILTIN_DSSALL);
def_builtin ("__builtin_altivec_dss", void_ftype_int, ALTIVEC_BUILTIN_DSS);
def_builtin ("__builtin_altivec_lvsl", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_LVSL);
def_builtin ("__builtin_altivec_lvsr", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_LVSR);
def_builtin ("__builtin_altivec_lvebx", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_LVEBX);
def_builtin ("__builtin_altivec_lvehx", v8hi_ftype_long_pcvoid, ALTIVEC_BUILTIN_LVEHX);
def_builtin ("__builtin_altivec_lvewx", v4si_ftype_long_pcvoid, ALTIVEC_BUILTIN_LVEWX);
def_builtin ("__builtin_altivec_lvxl", v4si_ftype_long_pcvoid, ALTIVEC_BUILTIN_LVXL);
def_builtin ("__builtin_altivec_lvxl_v2df", v2df_ftype_long_pcvoid,
ALTIVEC_BUILTIN_LVXL_V2DF);
def_builtin ("__builtin_altivec_lvxl_v2di", v2di_ftype_long_pcvoid,
ALTIVEC_BUILTIN_LVXL_V2DI);
def_builtin ("__builtin_altivec_lvxl_v4sf", v4sf_ftype_long_pcvoid,
ALTIVEC_BUILTIN_LVXL_V4SF);
def_builtin ("__builtin_altivec_lvxl_v4si", v4si_ftype_long_pcvoid,
ALTIVEC_BUILTIN_LVXL_V4SI);
def_builtin ("__builtin_altivec_lvxl_v8hi", v8hi_ftype_long_pcvoid,
ALTIVEC_BUILTIN_LVXL_V8HI);
def_builtin ("__builtin_altivec_lvxl_v16qi", v16qi_ftype_long_pcvoid,
ALTIVEC_BUILTIN_LVXL_V16QI);
def_builtin ("__builtin_altivec_lvx", v4si_ftype_long_pcvoid, ALTIVEC_BUILTIN_LVX);
def_builtin ("__builtin_altivec_lvx_v2df", v2df_ftype_long_pcvoid,
ALTIVEC_BUILTIN_LVX_V2DF);
def_builtin ("__builtin_altivec_lvx_v2di", v2di_ftype_long_pcvoid,
ALTIVEC_BUILTIN_LVX_V2DI);
def_builtin ("__builtin_altivec_lvx_v4sf", v4sf_ftype_long_pcvoid,
ALTIVEC_BUILTIN_LVX_V4SF);
def_builtin ("__builtin_altivec_lvx_v4si", v4si_ftype_long_pcvoid,
ALTIVEC_BUILTIN_LVX_V4SI);
def_builtin ("__builtin_altivec_lvx_v8hi", v8hi_ftype_long_pcvoid,
ALTIVEC_BUILTIN_LVX_V8HI);
def_builtin ("__builtin_altivec_lvx_v16qi", v16qi_ftype_long_pcvoid,
ALTIVEC_BUILTIN_LVX_V16QI);
def_builtin ("__builtin_altivec_stvx", void_ftype_v4si_long_pvoid, ALTIVEC_BUILTIN_STVX);
def_builtin ("__builtin_altivec_stvx_v2df", void_ftype_v2df_long_pvoid,
ALTIVEC_BUILTIN_STVX_V2DF);
def_builtin ("__builtin_altivec_stvx_v2di", void_ftype_v2di_long_pvoid,
ALTIVEC_BUILTIN_STVX_V2DI);
def_builtin ("__builtin_altivec_stvx_v4sf", void_ftype_v4sf_long_pvoid,
ALTIVEC_BUILTIN_STVX_V4SF);
def_builtin ("__builtin_altivec_stvx_v4si", void_ftype_v4si_long_pvoid,
ALTIVEC_BUILTIN_STVX_V4SI);
def_builtin ("__builtin_altivec_stvx_v8hi", void_ftype_v8hi_long_pvoid,
ALTIVEC_BUILTIN_STVX_V8HI);
def_builtin ("__builtin_altivec_stvx_v16qi", void_ftype_v16qi_long_pvoid,
ALTIVEC_BUILTIN_STVX_V16QI);
def_builtin ("__builtin_altivec_stvewx", void_ftype_v4si_long_pvoid, ALTIVEC_BUILTIN_STVEWX);
def_builtin ("__builtin_altivec_stvxl", void_ftype_v4si_long_pvoid, ALTIVEC_BUILTIN_STVXL);
def_builtin ("__builtin_altivec_stvxl_v2df", void_ftype_v2df_long_pvoid,
ALTIVEC_BUILTIN_STVXL_V2DF);
def_builtin ("__builtin_altivec_stvxl_v2di", void_ftype_v2di_long_pvoid,
ALTIVEC_BUILTIN_STVXL_V2DI);
def_builtin ("__builtin_altivec_stvxl_v4sf", void_ftype_v4sf_long_pvoid,
ALTIVEC_BUILTIN_STVXL_V4SF);
def_builtin ("__builtin_altivec_stvxl_v4si", void_ftype_v4si_long_pvoid,
ALTIVEC_BUILTIN_STVXL_V4SI);
def_builtin ("__builtin_altivec_stvxl_v8hi", void_ftype_v8hi_long_pvoid,
ALTIVEC_BUILTIN_STVXL_V8HI);
def_builtin ("__builtin_altivec_stvxl_v16qi", void_ftype_v16qi_long_pvoid,
ALTIVEC_BUILTIN_STVXL_V16QI);
def_builtin ("__builtin_altivec_stvebx", void_ftype_v16qi_long_pvoid, ALTIVEC_BUILTIN_STVEBX);
def_builtin ("__builtin_altivec_stvehx", void_ftype_v8hi_long_pvoid, ALTIVEC_BUILTIN_STVEHX);
def_builtin ("__builtin_vec_ld", opaque_ftype_long_pcvoid, ALTIVEC_BUILTIN_VEC_LD);
def_builtin ("__builtin_vec_lde", opaque_ftype_long_pcvoid, ALTIVEC_BUILTIN_VEC_LDE);
def_builtin ("__builtin_vec_ldl", opaque_ftype_long_pcvoid, ALTIVEC_BUILTIN_VEC_LDL);
def_builtin ("__builtin_vec_lvsl", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_VEC_LVSL);
def_builtin ("__builtin_vec_lvsr", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_VEC_LVSR);
def_builtin ("__builtin_vec_lvebx", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_VEC_LVEBX);
def_builtin ("__builtin_vec_lvehx", v8hi_ftype_long_pcvoid, ALTIVEC_BUILTIN_VEC_LVEHX);
def_builtin ("__builtin_vec_lvewx", v4si_ftype_long_pcvoid, ALTIVEC_BUILTIN_VEC_LVEWX);
def_builtin ("__builtin_vec_st", void_ftype_opaque_long_pvoid, ALTIVEC_BUILTIN_VEC_ST);
def_builtin ("__builtin_vec_ste", void_ftype_opaque_long_pvoid, ALTIVEC_BUILTIN_VEC_STE);
def_builtin ("__builtin_vec_stl", void_ftype_opaque_long_pvoid, ALTIVEC_BUILTIN_VEC_STL);
def_builtin ("__builtin_vec_stvewx", void_ftype_opaque_long_pvoid, ALTIVEC_BUILTIN_VEC_STVEWX);
def_builtin ("__builtin_vec_stvebx", void_ftype_opaque_long_pvoid, ALTIVEC_BUILTIN_VEC_STVEBX);
def_builtin ("__builtin_vec_stvehx", void_ftype_opaque_long_pvoid, ALTIVEC_BUILTIN_VEC_STVEHX);
def_builtin ("__builtin_vsx_lxvd2x_v2df", v2df_ftype_long_pcvoid,
VSX_BUILTIN_LXVD2X_V2DF);
def_builtin ("__builtin_vsx_lxvd2x_v2di", v2di_ftype_long_pcvoid,
VSX_BUILTIN_LXVD2X_V2DI);
def_builtin ("__builtin_vsx_lxvw4x_v4sf", v4sf_ftype_long_pcvoid,
VSX_BUILTIN_LXVW4X_V4SF);
def_builtin ("__builtin_vsx_lxvw4x_v4si", v4si_ftype_long_pcvoid,
VSX_BUILTIN_LXVW4X_V4SI);
def_builtin ("__builtin_vsx_lxvw4x_v8hi", v8hi_ftype_long_pcvoid,
VSX_BUILTIN_LXVW4X_V8HI);
def_builtin ("__builtin_vsx_lxvw4x_v16qi", v16qi_ftype_long_pcvoid,
VSX_BUILTIN_LXVW4X_V16QI);
def_builtin ("__builtin_vsx_stxvd2x_v2df", void_ftype_v2df_long_pvoid,
VSX_BUILTIN_STXVD2X_V2DF);
def_builtin ("__builtin_vsx_stxvd2x_v2di", void_ftype_v2di_long_pvoid,
VSX_BUILTIN_STXVD2X_V2DI);
def_builtin ("__builtin_vsx_stxvw4x_v4sf", void_ftype_v4sf_long_pvoid,
VSX_BUILTIN_STXVW4X_V4SF);
def_builtin ("__builtin_vsx_stxvw4x_v4si", void_ftype_v4si_long_pvoid,
VSX_BUILTIN_STXVW4X_V4SI);
def_builtin ("__builtin_vsx_stxvw4x_v8hi", void_ftype_v8hi_long_pvoid,
VSX_BUILTIN_STXVW4X_V8HI);
def_builtin ("__builtin_vsx_stxvw4x_v16qi", void_ftype_v16qi_long_pvoid,
VSX_BUILTIN_STXVW4X_V16QI);
def_builtin ("__builtin_vsx_ld_elemrev_v2df", v2df_ftype_long_pcvoid,
VSX_BUILTIN_LD_ELEMREV_V2DF);
def_builtin ("__builtin_vsx_ld_elemrev_v2di", v2di_ftype_long_pcvoid,
VSX_BUILTIN_LD_ELEMREV_V2DI);
def_builtin ("__builtin_vsx_ld_elemrev_v4sf", v4sf_ftype_long_pcvoid,
VSX_BUILTIN_LD_ELEMREV_V4SF);
def_builtin ("__builtin_vsx_ld_elemrev_v4si", v4si_ftype_long_pcvoid,
VSX_BUILTIN_LD_ELEMREV_V4SI);
def_builtin ("__builtin_vsx_ld_elemrev_v8hi", v8hi_ftype_long_pcvoid,
VSX_BUILTIN_LD_ELEMREV_V8HI);
def_builtin ("__builtin_vsx_ld_elemrev_v16qi", v16qi_ftype_long_pcvoid,
VSX_BUILTIN_LD_ELEMREV_V16QI);
def_builtin ("__builtin_vsx_st_elemrev_v2df", void_ftype_v2df_long_pvoid,
VSX_BUILTIN_ST_ELEMREV_V2DF);
def_builtin ("__builtin_vsx_st_elemrev_v2di", void_ftype_v2di_long_pvoid,
VSX_BUILTIN_ST_ELEMREV_V2DI);
def_builtin ("__builtin_vsx_st_elemrev_v4sf", void_ftype_v4sf_long_pvoid,
VSX_BUILTIN_ST_ELEMREV_V4SF);
def_builtin ("__builtin_vsx_st_elemrev_v4si", void_ftype_v4si_long_pvoid,
VSX_BUILTIN_ST_ELEMREV_V4SI);
def_builtin ("__builtin_vsx_st_elemrev_v8hi", void_ftype_v8hi_long_pvoid,
VSX_BUILTIN_ST_ELEMREV_V8HI);
def_builtin ("__builtin_vsx_st_elemrev_v16qi", void_ftype_v16qi_long_pvoid,
VSX_BUILTIN_ST_ELEMREV_V16QI);
def_builtin ("__builtin_vec_vsx_ld", opaque_ftype_long_pcvoid,
VSX_BUILTIN_VEC_LD);
def_builtin ("__builtin_vec_vsx_st", void_ftype_opaque_long_pvoid,
VSX_BUILTIN_VEC_ST);
def_builtin ("__builtin_vec_xl", opaque_ftype_long_pcvoid,
VSX_BUILTIN_VEC_XL);
def_builtin ("__builtin_vec_xl_be", opaque_ftype_long_pcvoid,
VSX_BUILTIN_VEC_XL_BE);
def_builtin ("__builtin_vec_xst", void_ftype_opaque_long_pvoid,
VSX_BUILTIN_VEC_XST);
def_builtin ("__builtin_vec_xst_be", void_ftype_opaque_long_pvoid,
VSX_BUILTIN_VEC_XST_BE);
def_builtin ("__builtin_vec_step", int_ftype_opaque, ALTIVEC_BUILTIN_VEC_STEP);
def_builtin ("__builtin_vec_splats", opaque_ftype_opaque, ALTIVEC_BUILTIN_VEC_SPLATS);
def_builtin ("__builtin_vec_promote", opaque_ftype_opaque, ALTIVEC_BUILTIN_VEC_PROMOTE);
def_builtin ("__builtin_vec_sld", opaque_ftype_opaque_opaque_int, ALTIVEC_BUILTIN_VEC_SLD);
def_builtin ("__builtin_vec_splat", opaque_ftype_opaque_int, ALTIVEC_BUILTIN_VEC_SPLAT);
def_builtin ("__builtin_vec_extract", opaque_ftype_opaque_int, ALTIVEC_BUILTIN_VEC_EXTRACT);
def_builtin ("__builtin_vec_insert", opaque_ftype_opaque_opaque_int, ALTIVEC_BUILTIN_VEC_INSERT);
def_builtin ("__builtin_vec_vspltw", opaque_ftype_opaque_int, ALTIVEC_BUILTIN_VEC_VSPLTW);
def_builtin ("__builtin_vec_vsplth", opaque_ftype_opaque_int, ALTIVEC_BUILTIN_VEC_VSPLTH);
def_builtin ("__builtin_vec_vspltb", opaque_ftype_opaque_int, ALTIVEC_BUILTIN_VEC_VSPLTB);
def_builtin ("__builtin_vec_ctf", opaque_ftype_opaque_int, ALTIVEC_BUILTIN_VEC_CTF);
def_builtin ("__builtin_vec_vcfsx", opaque_ftype_opaque_int, ALTIVEC_BUILTIN_VEC_VCFSX);
def_builtin ("__builtin_vec_vcfux", opaque_ftype_opaque_int, ALTIVEC_BUILTIN_VEC_VCFUX);
def_builtin ("__builtin_vec_cts", opaque_ftype_opaque_int, ALTIVEC_BUILTIN_VEC_CTS);
def_builtin ("__builtin_vec_ctu", opaque_ftype_opaque_int, ALTIVEC_BUILTIN_VEC_CTU);
def_builtin ("__builtin_vec_adde", opaque_ftype_opaque_opaque_opaque,
ALTIVEC_BUILTIN_VEC_ADDE);
def_builtin ("__builtin_vec_addec", opaque_ftype_opaque_opaque_opaque,
ALTIVEC_BUILTIN_VEC_ADDEC);
def_builtin ("__builtin_vec_cmpne", opaque_ftype_opaque_opaque,
ALTIVEC_BUILTIN_VEC_CMPNE);
def_builtin ("__builtin_vec_mul", opaque_ftype_opaque_opaque,
ALTIVEC_BUILTIN_VEC_MUL);
def_builtin ("__builtin_vec_sube", opaque_ftype_opaque_opaque_opaque,
ALTIVEC_BUILTIN_VEC_SUBE);
def_builtin ("__builtin_vec_subec", opaque_ftype_opaque_opaque_opaque,
ALTIVEC_BUILTIN_VEC_SUBEC);
/* Cell builtins. */
def_builtin ("__builtin_altivec_lvlx", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_LVLX);
def_builtin ("__builtin_altivec_lvlxl", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_LVLXL);
def_builtin ("__builtin_altivec_lvrx", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_LVRX);
def_builtin ("__builtin_altivec_lvrxl", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_LVRXL);
def_builtin ("__builtin_vec_lvlx", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_VEC_LVLX);
def_builtin ("__builtin_vec_lvlxl", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_VEC_LVLXL);
def_builtin ("__builtin_vec_lvrx", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_VEC_LVRX);
def_builtin ("__builtin_vec_lvrxl", v16qi_ftype_long_pcvoid, ALTIVEC_BUILTIN_VEC_LVRXL);
def_builtin ("__builtin_altivec_stvlx", void_ftype_v16qi_long_pvoid, ALTIVEC_BUILTIN_STVLX);
def_builtin ("__builtin_altivec_stvlxl", void_ftype_v16qi_long_pvoid, ALTIVEC_BUILTIN_STVLXL);
def_builtin ("__builtin_altivec_stvrx", void_ftype_v16qi_long_pvoid, ALTIVEC_BUILTIN_STVRX);
def_builtin ("__builtin_altivec_stvrxl", void_ftype_v16qi_long_pvoid, ALTIVEC_BUILTIN_STVRXL);
def_builtin ("__builtin_vec_stvlx", void_ftype_v16qi_long_pvoid, ALTIVEC_BUILTIN_VEC_STVLX);
def_builtin ("__builtin_vec_stvlxl", void_ftype_v16qi_long_pvoid, ALTIVEC_BUILTIN_VEC_STVLXL);
def_builtin ("__builtin_vec_stvrx", void_ftype_v16qi_long_pvoid, ALTIVEC_BUILTIN_VEC_STVRX);
def_builtin ("__builtin_vec_stvrxl", void_ftype_v16qi_long_pvoid, ALTIVEC_BUILTIN_VEC_STVRXL);
if (TARGET_P9_VECTOR)
{
def_builtin ("__builtin_altivec_stxvl", void_ftype_v16qi_pvoid_long,
P9V_BUILTIN_STXVL);
def_builtin ("__builtin_xst_len_r", void_ftype_v16qi_pvoid_long,
P9V_BUILTIN_XST_LEN_R);
}
/* Add the DST variants. */
d = bdesc_dst;
for (i = 0; i < ARRAY_SIZE (bdesc_dst); i++, d++)
{
HOST_WIDE_INT mask = d->mask;
/* It is expected that these dst built-in functions may have
d->icode equal to CODE_FOR_nothing. */
if ((mask & builtin_mask) != mask)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "altivec_init_builtins, skip dst %s\n",
d->name);
continue;
}
def_builtin (d->name, void_ftype_pcvoid_int_int, d->code);
}
/* Initialize the predicates. */
d = bdesc_altivec_preds;
for (i = 0; i < ARRAY_SIZE (bdesc_altivec_preds); i++, d++)
{
machine_mode mode1;
tree type;
HOST_WIDE_INT mask = d->mask;
if ((mask & builtin_mask) != mask)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "altivec_init_builtins, skip predicate %s\n",
d->name);
continue;
}
if (rs6000_overloaded_builtin_p (d->code))
mode1 = VOIDmode;
else
{
/* Cannot define builtin if the instruction is disabled. */
gcc_assert (d->icode != CODE_FOR_nothing);
mode1 = insn_data[d->icode].operand[1].mode;
}
switch (mode1)
{
case E_VOIDmode:
type = int_ftype_int_opaque_opaque;
break;
case E_V2DImode:
type = int_ftype_int_v2di_v2di;
break;
case E_V4SImode:
type = int_ftype_int_v4si_v4si;
break;
case E_V8HImode:
type = int_ftype_int_v8hi_v8hi;
break;
case E_V16QImode:
type = int_ftype_int_v16qi_v16qi;
break;
case E_V4SFmode:
type = int_ftype_int_v4sf_v4sf;
break;
case E_V2DFmode:
type = int_ftype_int_v2df_v2df;
break;
default:
gcc_unreachable ();
}
def_builtin (d->name, type, d->code);
}
/* Initialize the abs* operators. */
d = bdesc_abs;
for (i = 0; i < ARRAY_SIZE (bdesc_abs); i++, d++)
{
machine_mode mode0;
tree type;
HOST_WIDE_INT mask = d->mask;
if ((mask & builtin_mask) != mask)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "altivec_init_builtins, skip abs %s\n",
d->name);
continue;
}
/* Cannot define builtin if the instruction is disabled. */
gcc_assert (d->icode != CODE_FOR_nothing);
mode0 = insn_data[d->icode].operand[0].mode;
switch (mode0)
{
case E_V2DImode:
type = v2di_ftype_v2di;
break;
case E_V4SImode:
type = v4si_ftype_v4si;
break;
case E_V8HImode:
type = v8hi_ftype_v8hi;
break;
case E_V16QImode:
type = v16qi_ftype_v16qi;
break;
case E_V4SFmode:
type = v4sf_ftype_v4sf;
break;
case E_V2DFmode:
type = v2df_ftype_v2df;
break;
default:
gcc_unreachable ();
}
def_builtin (d->name, type, d->code);
}
/* Initialize target builtin that implements
targetm.vectorize.builtin_mask_for_load. */
decl = add_builtin_function ("__builtin_altivec_mask_for_load",
v16qi_ftype_long_pcvoid,
ALTIVEC_BUILTIN_MASK_FOR_LOAD,
BUILT_IN_MD, NULL, NULL_TREE);
TREE_READONLY (decl) = 1;
/* Record the decl. Will be used by rs6000_builtin_mask_for_load. */
altivec_builtin_mask_for_load = decl;
/* Access to the vec_init patterns. */
ftype = build_function_type_list (V4SI_type_node, integer_type_node,
integer_type_node, integer_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_init_v4si", ftype, ALTIVEC_BUILTIN_VEC_INIT_V4SI);
ftype = build_function_type_list (V8HI_type_node, short_integer_type_node,
short_integer_type_node,
short_integer_type_node,
short_integer_type_node,
short_integer_type_node,
short_integer_type_node,
short_integer_type_node,
short_integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_init_v8hi", ftype, ALTIVEC_BUILTIN_VEC_INIT_V8HI);
ftype = build_function_type_list (V16QI_type_node, char_type_node,
char_type_node, char_type_node,
char_type_node, char_type_node,
char_type_node, char_type_node,
char_type_node, char_type_node,
char_type_node, char_type_node,
char_type_node, char_type_node,
char_type_node, char_type_node,
char_type_node, NULL_TREE);
def_builtin ("__builtin_vec_init_v16qi", ftype,
ALTIVEC_BUILTIN_VEC_INIT_V16QI);
ftype = build_function_type_list (V4SF_type_node, float_type_node,
float_type_node, float_type_node,
float_type_node, NULL_TREE);
def_builtin ("__builtin_vec_init_v4sf", ftype, ALTIVEC_BUILTIN_VEC_INIT_V4SF);
/* VSX builtins. */
ftype = build_function_type_list (V2DF_type_node, double_type_node,
double_type_node, NULL_TREE);
def_builtin ("__builtin_vec_init_v2df", ftype, VSX_BUILTIN_VEC_INIT_V2DF);
ftype = build_function_type_list (V2DI_type_node, intDI_type_node,
intDI_type_node, NULL_TREE);
def_builtin ("__builtin_vec_init_v2di", ftype, VSX_BUILTIN_VEC_INIT_V2DI);
/* Access to the vec_set patterns. */
ftype = build_function_type_list (V4SI_type_node, V4SI_type_node,
intSI_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_set_v4si", ftype, ALTIVEC_BUILTIN_VEC_SET_V4SI);
ftype = build_function_type_list (V8HI_type_node, V8HI_type_node,
intHI_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_set_v8hi", ftype, ALTIVEC_BUILTIN_VEC_SET_V8HI);
ftype = build_function_type_list (V16QI_type_node, V16QI_type_node,
intQI_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_set_v16qi", ftype, ALTIVEC_BUILTIN_VEC_SET_V16QI);
ftype = build_function_type_list (V4SF_type_node, V4SF_type_node,
float_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_set_v4sf", ftype, ALTIVEC_BUILTIN_VEC_SET_V4SF);
ftype = build_function_type_list (V2DF_type_node, V2DF_type_node,
double_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_set_v2df", ftype, VSX_BUILTIN_VEC_SET_V2DF);
ftype = build_function_type_list (V2DI_type_node, V2DI_type_node,
intDI_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_set_v2di", ftype, VSX_BUILTIN_VEC_SET_V2DI);
/* Access to the vec_extract patterns. */
ftype = build_function_type_list (intSI_type_node, V4SI_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_ext_v4si", ftype, ALTIVEC_BUILTIN_VEC_EXT_V4SI);
ftype = build_function_type_list (intHI_type_node, V8HI_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_ext_v8hi", ftype, ALTIVEC_BUILTIN_VEC_EXT_V8HI);
ftype = build_function_type_list (intQI_type_node, V16QI_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_ext_v16qi", ftype, ALTIVEC_BUILTIN_VEC_EXT_V16QI);
ftype = build_function_type_list (float_type_node, V4SF_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_ext_v4sf", ftype, ALTIVEC_BUILTIN_VEC_EXT_V4SF);
ftype = build_function_type_list (double_type_node, V2DF_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_ext_v2df", ftype, VSX_BUILTIN_VEC_EXT_V2DF);
ftype = build_function_type_list (intDI_type_node, V2DI_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_ext_v2di", ftype, VSX_BUILTIN_VEC_EXT_V2DI);
if (V1TI_type_node)
{
tree v1ti_ftype_long_pcvoid
= build_function_type_list (V1TI_type_node,
long_integer_type_node, pcvoid_type_node,
NULL_TREE);
tree void_ftype_v1ti_long_pvoid
= build_function_type_list (void_type_node,
V1TI_type_node, long_integer_type_node,
pvoid_type_node, NULL_TREE);
def_builtin ("__builtin_vsx_lxvd2x_v1ti", v1ti_ftype_long_pcvoid,
VSX_BUILTIN_LXVD2X_V1TI);
def_builtin ("__builtin_vsx_stxvd2x_v1ti", void_ftype_v1ti_long_pvoid,
VSX_BUILTIN_STXVD2X_V1TI);
ftype = build_function_type_list (V1TI_type_node, intTI_type_node,
NULL_TREE, NULL_TREE);
def_builtin ("__builtin_vec_init_v1ti", ftype, VSX_BUILTIN_VEC_INIT_V1TI);
ftype = build_function_type_list (V1TI_type_node, V1TI_type_node,
intTI_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_set_v1ti", ftype, VSX_BUILTIN_VEC_SET_V1TI);
ftype = build_function_type_list (intTI_type_node, V1TI_type_node,
integer_type_node, NULL_TREE);
def_builtin ("__builtin_vec_ext_v1ti", ftype, VSX_BUILTIN_VEC_EXT_V1TI);
}
}
static void
htm_init_builtins (void)
{
HOST_WIDE_INT builtin_mask = rs6000_builtin_mask;
const struct builtin_description *d;
size_t i;
d = bdesc_htm;
for (i = 0; i < ARRAY_SIZE (bdesc_htm); i++, d++)
{
tree op[MAX_HTM_OPERANDS], type;
HOST_WIDE_INT mask = d->mask;
unsigned attr = rs6000_builtin_info[d->code].attr;
bool void_func = (attr & RS6000_BTC_VOID);
int attr_args = (attr & RS6000_BTC_TYPE_MASK);
int nopnds = 0;
tree gpr_type_node;
tree rettype;
tree argtype;
/* It is expected that these htm built-in functions may have
d->icode equal to CODE_FOR_nothing. */
if (TARGET_32BIT && TARGET_POWERPC64)
gpr_type_node = long_long_unsigned_type_node;
else
gpr_type_node = long_unsigned_type_node;
if (attr & RS6000_BTC_SPR)
{
rettype = gpr_type_node;
argtype = gpr_type_node;
}
else if (d->code == HTM_BUILTIN_TABORTDC
|| d->code == HTM_BUILTIN_TABORTDCI)
{
rettype = unsigned_type_node;
argtype = gpr_type_node;
}
else
{
rettype = unsigned_type_node;
argtype = unsigned_type_node;
}
if ((mask & builtin_mask) != mask)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "htm_builtin, skip binary %s\n", d->name);
continue;
}
if (d->name == 0)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "htm_builtin, bdesc_htm[%ld] no name\n",
(long unsigned) i);
continue;
}
op[nopnds++] = (void_func) ? void_type_node : rettype;
if (attr_args == RS6000_BTC_UNARY)
op[nopnds++] = argtype;
else if (attr_args == RS6000_BTC_BINARY)
{
op[nopnds++] = argtype;
op[nopnds++] = argtype;
}
else if (attr_args == RS6000_BTC_TERNARY)
{
op[nopnds++] = argtype;
op[nopnds++] = argtype;
op[nopnds++] = argtype;
}
switch (nopnds)
{
case 1:
type = build_function_type_list (op[0], NULL_TREE);
break;
case 2:
type = build_function_type_list (op[0], op[1], NULL_TREE);
break;
case 3:
type = build_function_type_list (op[0], op[1], op[2], NULL_TREE);
break;
case 4:
type = build_function_type_list (op[0], op[1], op[2], op[3],
NULL_TREE);
break;
default:
gcc_unreachable ();
}
def_builtin (d->name, type, d->code);
}
}
/* Hash function for builtin functions with up to 3 arguments and a return
type. */
hashval_t
builtin_hasher::hash (builtin_hash_struct *bh)
{
unsigned ret = 0;
int i;
for (i = 0; i < 4; i++)
{
ret = (ret * (unsigned)MAX_MACHINE_MODE) + ((unsigned)bh->mode[i]);
ret = (ret * 2) + bh->uns_p[i];
}
return ret;
}
/* Compare builtin hash entries H1 and H2 for equivalence. */
bool
builtin_hasher::equal (builtin_hash_struct *p1, builtin_hash_struct *p2)
{
return ((p1->mode[0] == p2->mode[0])
&& (p1->mode[1] == p2->mode[1])
&& (p1->mode[2] == p2->mode[2])
&& (p1->mode[3] == p2->mode[3])
&& (p1->uns_p[0] == p2->uns_p[0])
&& (p1->uns_p[1] == p2->uns_p[1])
&& (p1->uns_p[2] == p2->uns_p[2])
&& (p1->uns_p[3] == p2->uns_p[3]));
}
/* Map types for builtin functions with an explicit return type and up to 3
arguments. Functions with fewer than 3 arguments use VOIDmode as the type
of the argument. */
static tree
builtin_function_type (machine_mode mode_ret, machine_mode mode_arg0,
machine_mode mode_arg1, machine_mode mode_arg2,
enum rs6000_builtins builtin, const char *name)
{
struct builtin_hash_struct h;
struct builtin_hash_struct *h2;
int num_args = 3;
int i;
tree ret_type = NULL_TREE;
tree arg_type[3] = { NULL_TREE, NULL_TREE, NULL_TREE };
/* Create builtin_hash_table. */
if (builtin_hash_table == NULL)
builtin_hash_table = hash_table<builtin_hasher>::create_ggc (1500);
h.type = NULL_TREE;
h.mode[0] = mode_ret;
h.mode[1] = mode_arg0;
h.mode[2] = mode_arg1;
h.mode[3] = mode_arg2;
h.uns_p[0] = 0;
h.uns_p[1] = 0;
h.uns_p[2] = 0;
h.uns_p[3] = 0;
/* If the builtin is a type that produces unsigned results or takes unsigned
arguments, and it is returned as a decl for the vectorizer (such as
widening multiplies, permute), make sure the arguments and return value
are type correct. */
switch (builtin)
{
/* unsigned 1 argument functions. */
case CRYPTO_BUILTIN_VSBOX:
case P8V_BUILTIN_VGBBD:
case MISC_BUILTIN_CDTBCD:
case MISC_BUILTIN_CBCDTD:
h.uns_p[0] = 1;
h.uns_p[1] = 1;
break;
/* unsigned 2 argument functions. */
case ALTIVEC_BUILTIN_VMULEUB:
case ALTIVEC_BUILTIN_VMULEUH:
case ALTIVEC_BUILTIN_VMULEUW:
case ALTIVEC_BUILTIN_VMULOUB:
case ALTIVEC_BUILTIN_VMULOUH:
case ALTIVEC_BUILTIN_VMULOUW:
case CRYPTO_BUILTIN_VCIPHER:
case CRYPTO_BUILTIN_VCIPHERLAST:
case CRYPTO_BUILTIN_VNCIPHER:
case CRYPTO_BUILTIN_VNCIPHERLAST:
case CRYPTO_BUILTIN_VPMSUMB:
case CRYPTO_BUILTIN_VPMSUMH:
case CRYPTO_BUILTIN_VPMSUMW:
case CRYPTO_BUILTIN_VPMSUMD:
case CRYPTO_BUILTIN_VPMSUM:
case MISC_BUILTIN_ADDG6S:
case MISC_BUILTIN_DIVWEU:
case MISC_BUILTIN_DIVWEUO:
case MISC_BUILTIN_DIVDEU:
case MISC_BUILTIN_DIVDEUO:
case VSX_BUILTIN_UDIV_V2DI:
case ALTIVEC_BUILTIN_VMAXUB:
case ALTIVEC_BUILTIN_VMINUB:
case ALTIVEC_BUILTIN_VMAXUH:
case ALTIVEC_BUILTIN_VMINUH:
case ALTIVEC_BUILTIN_VMAXUW:
case ALTIVEC_BUILTIN_VMINUW:
case P8V_BUILTIN_VMAXUD:
case P8V_BUILTIN_VMINUD:
h.uns_p[0] = 1;
h.uns_p[1] = 1;
h.uns_p[2] = 1;
break;
/* unsigned 3 argument functions. */
case ALTIVEC_BUILTIN_VPERM_16QI_UNS:
case ALTIVEC_BUILTIN_VPERM_8HI_UNS:
case ALTIVEC_BUILTIN_VPERM_4SI_UNS:
case ALTIVEC_BUILTIN_VPERM_2DI_UNS:
case ALTIVEC_BUILTIN_VSEL_16QI_UNS:
case ALTIVEC_BUILTIN_VSEL_8HI_UNS:
case ALTIVEC_BUILTIN_VSEL_4SI_UNS:
case ALTIVEC_BUILTIN_VSEL_2DI_UNS:
case VSX_BUILTIN_VPERM_16QI_UNS:
case VSX_BUILTIN_VPERM_8HI_UNS:
case VSX_BUILTIN_VPERM_4SI_UNS:
case VSX_BUILTIN_VPERM_2DI_UNS:
case VSX_BUILTIN_XXSEL_16QI_UNS:
case VSX_BUILTIN_XXSEL_8HI_UNS:
case VSX_BUILTIN_XXSEL_4SI_UNS:
case VSX_BUILTIN_XXSEL_2DI_UNS:
case CRYPTO_BUILTIN_VPERMXOR:
case CRYPTO_BUILTIN_VPERMXOR_V2DI:
case CRYPTO_BUILTIN_VPERMXOR_V4SI:
case CRYPTO_BUILTIN_VPERMXOR_V8HI:
case CRYPTO_BUILTIN_VPERMXOR_V16QI:
case CRYPTO_BUILTIN_VSHASIGMAW:
case CRYPTO_BUILTIN_VSHASIGMAD:
case CRYPTO_BUILTIN_VSHASIGMA:
h.uns_p[0] = 1;
h.uns_p[1] = 1;
h.uns_p[2] = 1;
h.uns_p[3] = 1;
break;
/* signed permute functions with unsigned char mask. */
case ALTIVEC_BUILTIN_VPERM_16QI:
case ALTIVEC_BUILTIN_VPERM_8HI:
case ALTIVEC_BUILTIN_VPERM_4SI:
case ALTIVEC_BUILTIN_VPERM_4SF:
case ALTIVEC_BUILTIN_VPERM_2DI:
case ALTIVEC_BUILTIN_VPERM_2DF:
case VSX_BUILTIN_VPERM_16QI:
case VSX_BUILTIN_VPERM_8HI:
case VSX_BUILTIN_VPERM_4SI:
case VSX_BUILTIN_VPERM_4SF:
case VSX_BUILTIN_VPERM_2DI:
case VSX_BUILTIN_VPERM_2DF:
h.uns_p[3] = 1;
break;
/* unsigned args, signed return. */
case VSX_BUILTIN_XVCVUXDSP:
case VSX_BUILTIN_XVCVUXDDP_UNS:
case ALTIVEC_BUILTIN_UNSFLOAT_V4SI_V4SF:
h.uns_p[1] = 1;
break;
/* signed args, unsigned return. */
case VSX_BUILTIN_XVCVDPUXDS_UNS:
case ALTIVEC_BUILTIN_FIXUNS_V4SF_V4SI:
case MISC_BUILTIN_UNPACK_TD:
case MISC_BUILTIN_UNPACK_V1TI:
h.uns_p[0] = 1;
break;
/* unsigned arguments, bool return (compares). */
case ALTIVEC_BUILTIN_VCMPEQUB:
case ALTIVEC_BUILTIN_VCMPEQUH:
case ALTIVEC_BUILTIN_VCMPEQUW:
case P8V_BUILTIN_VCMPEQUD:
case VSX_BUILTIN_CMPGE_U16QI:
case VSX_BUILTIN_CMPGE_U8HI:
case VSX_BUILTIN_CMPGE_U4SI:
case VSX_BUILTIN_CMPGE_U2DI:
case ALTIVEC_BUILTIN_VCMPGTUB:
case ALTIVEC_BUILTIN_VCMPGTUH:
case ALTIVEC_BUILTIN_VCMPGTUW:
case P8V_BUILTIN_VCMPGTUD:
h.uns_p[1] = 1;
h.uns_p[2] = 1;
break;
/* unsigned arguments for 128-bit pack instructions. */
case MISC_BUILTIN_PACK_TD:
case MISC_BUILTIN_PACK_V1TI:
h.uns_p[1] = 1;
h.uns_p[2] = 1;
break;
/* unsigned second arguments (vector shift right). */
case ALTIVEC_BUILTIN_VSRB:
case ALTIVEC_BUILTIN_VSRH:
case ALTIVEC_BUILTIN_VSRW:
case P8V_BUILTIN_VSRD:
h.uns_p[2] = 1;
break;
default:
break;
}
/* Figure out how many args are present. */
while (num_args > 0 && h.mode[num_args] == VOIDmode)
num_args--;
ret_type = builtin_mode_to_type[h.mode[0]][h.uns_p[0]];
if (!ret_type && h.uns_p[0])
ret_type = builtin_mode_to_type[h.mode[0]][0];
if (!ret_type)
fatal_error (input_location,
"internal error: builtin function %qs had an unexpected "
"return type %qs", name, GET_MODE_NAME (h.mode[0]));
for (i = 0; i < (int) ARRAY_SIZE (arg_type); i++)
arg_type[i] = NULL_TREE;
for (i = 0; i < num_args; i++)
{
int m = (int) h.mode[i+1];
int uns_p = h.uns_p[i+1];
arg_type[i] = builtin_mode_to_type[m][uns_p];
if (!arg_type[i] && uns_p)
arg_type[i] = builtin_mode_to_type[m][0];
if (!arg_type[i])
fatal_error (input_location,
"internal error: builtin function %qs, argument %d "
"had unexpected argument type %qs", name, i,
GET_MODE_NAME (m));
}
builtin_hash_struct **found = builtin_hash_table->find_slot (&h, INSERT);
if (*found == NULL)
{
h2 = ggc_alloc<builtin_hash_struct> ();
*h2 = h;
*found = h2;
h2->type = build_function_type_list (ret_type, arg_type[0], arg_type[1],
arg_type[2], NULL_TREE);
}
return (*found)->type;
}
static void
rs6000_common_init_builtins (void)
{
const struct builtin_description *d;
size_t i;
tree opaque_ftype_opaque = NULL_TREE;
tree opaque_ftype_opaque_opaque = NULL_TREE;
tree opaque_ftype_opaque_opaque_opaque = NULL_TREE;
tree v2si_ftype = NULL_TREE;
tree v2si_ftype_qi = NULL_TREE;
tree v2si_ftype_v2si_qi = NULL_TREE;
tree v2si_ftype_int_qi = NULL_TREE;
HOST_WIDE_INT builtin_mask = rs6000_builtin_mask;
if (!TARGET_PAIRED_FLOAT)
{
builtin_mode_to_type[V2SImode][0] = opaque_V2SI_type_node;
builtin_mode_to_type[V2SFmode][0] = opaque_V2SF_type_node;
}
/* Paired builtins are only available if you build a compiler with the
appropriate options, so only create those builtins with the appropriate
compiler option. Create Altivec and VSX builtins on machines with at
least the general purpose extensions (970 and newer) to allow the use of
the target attribute.. */
if (TARGET_EXTRA_BUILTINS)
builtin_mask |= RS6000_BTM_COMMON;
/* Add the ternary operators. */
d = bdesc_3arg;
for (i = 0; i < ARRAY_SIZE (bdesc_3arg); i++, d++)
{
tree type;
HOST_WIDE_INT mask = d->mask;
if ((mask & builtin_mask) != mask)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin, skip ternary %s\n", d->name);
continue;
}
if (rs6000_overloaded_builtin_p (d->code))
{
if (! (type = opaque_ftype_opaque_opaque_opaque))
type = opaque_ftype_opaque_opaque_opaque
= build_function_type_list (opaque_V4SI_type_node,
opaque_V4SI_type_node,
opaque_V4SI_type_node,
opaque_V4SI_type_node,
NULL_TREE);
}
else
{
enum insn_code icode = d->icode;
if (d->name == 0)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin, bdesc_3arg[%ld] no name\n",
(long unsigned)i);
continue;
}
if (icode == CODE_FOR_nothing)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin, skip ternary %s (no code)\n",
d->name);
continue;
}
type = builtin_function_type (insn_data[icode].operand[0].mode,
insn_data[icode].operand[1].mode,
insn_data[icode].operand[2].mode,
insn_data[icode].operand[3].mode,
d->code, d->name);
}
def_builtin (d->name, type, d->code);
}
/* Add the binary operators. */
d = bdesc_2arg;
for (i = 0; i < ARRAY_SIZE (bdesc_2arg); i++, d++)
{
machine_mode mode0, mode1, mode2;
tree type;
HOST_WIDE_INT mask = d->mask;
if ((mask & builtin_mask) != mask)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin, skip binary %s\n", d->name);
continue;
}
if (rs6000_overloaded_builtin_p (d->code))
{
if (! (type = opaque_ftype_opaque_opaque))
type = opaque_ftype_opaque_opaque
= build_function_type_list (opaque_V4SI_type_node,
opaque_V4SI_type_node,
opaque_V4SI_type_node,
NULL_TREE);
}
else
{
enum insn_code icode = d->icode;
if (d->name == 0)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin, bdesc_2arg[%ld] no name\n",
(long unsigned)i);
continue;
}
if (icode == CODE_FOR_nothing)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin, skip binary %s (no code)\n",
d->name);
continue;
}
mode0 = insn_data[icode].operand[0].mode;
mode1 = insn_data[icode].operand[1].mode;
mode2 = insn_data[icode].operand[2].mode;
if (mode0 == V2SImode && mode1 == V2SImode && mode2 == QImode)
{
if (! (type = v2si_ftype_v2si_qi))
type = v2si_ftype_v2si_qi
= build_function_type_list (opaque_V2SI_type_node,
opaque_V2SI_type_node,
char_type_node,
NULL_TREE);
}
else if (mode0 == V2SImode && GET_MODE_CLASS (mode1) == MODE_INT
&& mode2 == QImode)
{
if (! (type = v2si_ftype_int_qi))
type = v2si_ftype_int_qi
= build_function_type_list (opaque_V2SI_type_node,
integer_type_node,
char_type_node,
NULL_TREE);
}
else
type = builtin_function_type (mode0, mode1, mode2, VOIDmode,
d->code, d->name);
}
def_builtin (d->name, type, d->code);
}
/* Add the simple unary operators. */
d = bdesc_1arg;
for (i = 0; i < ARRAY_SIZE (bdesc_1arg); i++, d++)
{
machine_mode mode0, mode1;
tree type;
HOST_WIDE_INT mask = d->mask;
if ((mask & builtin_mask) != mask)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin, skip unary %s\n", d->name);
continue;
}
if (rs6000_overloaded_builtin_p (d->code))
{
if (! (type = opaque_ftype_opaque))
type = opaque_ftype_opaque
= build_function_type_list (opaque_V4SI_type_node,
opaque_V4SI_type_node,
NULL_TREE);
}
else
{
enum insn_code icode = d->icode;
if (d->name == 0)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin, bdesc_1arg[%ld] no name\n",
(long unsigned)i);
continue;
}
if (icode == CODE_FOR_nothing)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin, skip unary %s (no code)\n",
d->name);
continue;
}
mode0 = insn_data[icode].operand[0].mode;
mode1 = insn_data[icode].operand[1].mode;
if (mode0 == V2SImode && mode1 == QImode)
{
if (! (type = v2si_ftype_qi))
type = v2si_ftype_qi
= build_function_type_list (opaque_V2SI_type_node,
char_type_node,
NULL_TREE);
}
else
type = builtin_function_type (mode0, mode1, VOIDmode, VOIDmode,
d->code, d->name);
}
def_builtin (d->name, type, d->code);
}
/* Add the simple no-argument operators. */
d = bdesc_0arg;
for (i = 0; i < ARRAY_SIZE (bdesc_0arg); i++, d++)
{
machine_mode mode0;
tree type;
HOST_WIDE_INT mask = d->mask;
if ((mask & builtin_mask) != mask)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin, skip no-argument %s\n", d->name);
continue;
}
if (rs6000_overloaded_builtin_p (d->code))
{
if (!opaque_ftype_opaque)
opaque_ftype_opaque
= build_function_type_list (opaque_V4SI_type_node, NULL_TREE);
type = opaque_ftype_opaque;
}
else
{
enum insn_code icode = d->icode;
if (d->name == 0)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr, "rs6000_builtin, bdesc_0arg[%lu] no name\n",
(long unsigned) i);
continue;
}
if (icode == CODE_FOR_nothing)
{
if (TARGET_DEBUG_BUILTIN)
fprintf (stderr,
"rs6000_builtin, skip no-argument %s (no code)\n",
d->name);
continue;
}
mode0 = insn_data[icode].operand[0].mode;
if (mode0 == V2SImode)
{
/* code for paired single */
if (! (type = v2si_ftype))
{
v2si_ftype
= build_function_type_list (opaque_V2SI_type_node,
NULL_TREE);
type = v2si_ftype;
}
}
else
type = builtin_function_type (mode0, VOIDmode, VOIDmode, VOIDmode,
d->code, d->name);
}
def_builtin (d->name, type, d->code);
}
}
/* Set up AIX/Darwin/64-bit Linux quad floating point routines. */
static void
init_float128_ibm (machine_mode mode)
{
if (!TARGET_XL_COMPAT)
{
set_optab_libfunc (add_optab, mode, "__gcc_qadd");
set_optab_libfunc (sub_optab, mode, "__gcc_qsub");
set_optab_libfunc (smul_optab, mode, "__gcc_qmul");
set_optab_libfunc (sdiv_optab, mode, "__gcc_qdiv");
if (!TARGET_HARD_FLOAT)
{
set_optab_libfunc (neg_optab, mode, "__gcc_qneg");
set_optab_libfunc (eq_optab, mode, "__gcc_qeq");
set_optab_libfunc (ne_optab, mode, "__gcc_qne");
set_optab_libfunc (gt_optab, mode, "__gcc_qgt");
set_optab_libfunc (ge_optab, mode, "__gcc_qge");
set_optab_libfunc (lt_optab, mode, "__gcc_qlt");
set_optab_libfunc (le_optab, mode, "__gcc_qle");
set_optab_libfunc (unord_optab, mode, "__gcc_qunord");
set_conv_libfunc (sext_optab, mode, SFmode, "__gcc_stoq");
set_conv_libfunc (sext_optab, mode, DFmode, "__gcc_dtoq");
set_conv_libfunc (trunc_optab, SFmode, mode, "__gcc_qtos");
set_conv_libfunc (trunc_optab, DFmode, mode, "__gcc_qtod");
set_conv_libfunc (sfix_optab, SImode, mode, "__gcc_qtoi");
set_conv_libfunc (ufix_optab, SImode, mode, "__gcc_qtou");
set_conv_libfunc (sfloat_optab, mode, SImode, "__gcc_itoq");
set_conv_libfunc (ufloat_optab, mode, SImode, "__gcc_utoq");
}
}
else
{
set_optab_libfunc (add_optab, mode, "_xlqadd");
set_optab_libfunc (sub_optab, mode, "_xlqsub");
set_optab_libfunc (smul_optab, mode, "_xlqmul");
set_optab_libfunc (sdiv_optab, mode, "_xlqdiv");
}
/* Add various conversions for IFmode to use the traditional TFmode
names. */
if (mode == IFmode)
{
set_conv_libfunc (sext_optab, mode, SDmode, "__dpd_extendsdtf2");
set_conv_libfunc (sext_optab, mode, DDmode, "__dpd_extendddtf2");
set_conv_libfunc (trunc_optab, mode, TDmode, "__dpd_trunctftd2");
set_conv_libfunc (trunc_optab, SDmode, mode, "__dpd_trunctfsd2");
set_conv_libfunc (trunc_optab, DDmode, mode, "__dpd_trunctfdd2");
set_conv_libfunc (sext_optab, TDmode, mode, "__dpd_extendtdtf2");
if (TARGET_POWERPC64)
{
set_conv_libfunc (sfix_optab, TImode, mode, "__fixtfti");
set_conv_libfunc (ufix_optab, TImode, mode, "__fixunstfti");
set_conv_libfunc (sfloat_optab, mode, TImode, "__floattitf");
set_conv_libfunc (ufloat_optab, mode, TImode, "__floatuntitf");
}
}
}
/* Set up IEEE 128-bit floating point routines. Use different names if the
arguments can be passed in a vector register. The historical PowerPC
implementation of IEEE 128-bit floating point used _q_<op> for the names, so
continue to use that if we aren't using vector registers to pass IEEE
128-bit floating point. */
static void
init_float128_ieee (machine_mode mode)
{
if (FLOAT128_VECTOR_P (mode))
{
set_optab_libfunc (add_optab, mode, "__addkf3");
set_optab_libfunc (sub_optab, mode, "__subkf3");
set_optab_libfunc (neg_optab, mode, "__negkf2");
set_optab_libfunc (smul_optab, mode, "__mulkf3");
set_optab_libfunc (sdiv_optab, mode, "__divkf3");
set_optab_libfunc (sqrt_optab, mode, "__sqrtkf2");
set_optab_libfunc (abs_optab, mode, "__abstkf2");
set_optab_libfunc (eq_optab, mode, "__eqkf2");
set_optab_libfunc (ne_optab, mode, "__nekf2");
set_optab_libfunc (gt_optab, mode, "__gtkf2");
set_optab_libfunc (ge_optab, mode, "__gekf2");
set_optab_libfunc (lt_optab, mode, "__ltkf2");
set_optab_libfunc (le_optab, mode, "__lekf2");
set_optab_libfunc (unord_optab, mode, "__unordkf2");
set_conv_libfunc (sext_optab, mode, SFmode, "__extendsfkf2");
set_conv_libfunc (sext_optab, mode, DFmode, "__extenddfkf2");
set_conv_libfunc (trunc_optab, SFmode, mode, "__trunckfsf2");
set_conv_libfunc (trunc_optab, DFmode, mode, "__trunckfdf2");
set_conv_libfunc (sext_optab, mode, IFmode, "__extendtfkf2");
if (mode != TFmode && FLOAT128_IBM_P (TFmode))
set_conv_libfunc (sext_optab, mode, TFmode, "__extendtfkf2");
set_conv_libfunc (trunc_optab, IFmode, mode, "__trunckftf2");
if (mode != TFmode && FLOAT128_IBM_P (TFmode))
set_conv_libfunc (trunc_optab, TFmode, mode, "__trunckftf2");
set_conv_libfunc (sext_optab, mode, SDmode, "__dpd_extendsdkf2");
set_conv_libfunc (sext_optab, mode, DDmode, "__dpd_extendddkf2");
set_conv_libfunc (trunc_optab, mode, TDmode, "__dpd_trunckftd2");
set_conv_libfunc (trunc_optab, SDmode, mode, "__dpd_trunckfsd2");
set_conv_libfunc (trunc_optab, DDmode, mode, "__dpd_trunckfdd2");
set_conv_libfunc (sext_optab, TDmode, mode, "__dpd_extendtdkf2");
set_conv_libfunc (sfix_optab, SImode, mode, "__fixkfsi");
set_conv_libfunc (ufix_optab, SImode, mode, "__fixunskfsi");
set_conv_libfunc (sfix_optab, DImode, mode, "__fixkfdi");
set_conv_libfunc (ufix_optab, DImode, mode, "__fixunskfdi");
set_conv_libfunc (sfloat_optab, mode, SImode, "__floatsikf");
set_conv_libfunc (ufloat_optab, mode, SImode, "__floatunsikf");
set_conv_libfunc (sfloat_optab, mode, DImode, "__floatdikf");
set_conv_libfunc (ufloat_optab, mode, DImode, "__floatundikf");
if (TARGET_POWERPC64)
{
set_conv_libfunc (sfix_optab, TImode, mode, "__fixkfti");
set_conv_libfunc (ufix_optab, TImode, mode, "__fixunskfti");
set_conv_libfunc (sfloat_optab, mode, TImode, "__floattikf");
set_conv_libfunc (ufloat_optab, mode, TImode, "__floatuntikf");
}
}
else
{
set_optab_libfunc (add_optab, mode, "_q_add");
set_optab_libfunc (sub_optab, mode, "_q_sub");
set_optab_libfunc (neg_optab, mode, "_q_neg");
set_optab_libfunc (smul_optab, mode, "_q_mul");
set_optab_libfunc (sdiv_optab, mode, "_q_div");
if (TARGET_PPC_GPOPT)
set_optab_libfunc (sqrt_optab, mode, "_q_sqrt");
set_optab_libfunc (eq_optab, mode, "_q_feq");
set_optab_libfunc (ne_optab, mode, "_q_fne");
set_optab_libfunc (gt_optab, mode, "_q_fgt");
set_optab_libfunc (ge_optab, mode, "_q_fge");
set_optab_libfunc (lt_optab, mode, "_q_flt");
set_optab_libfunc (le_optab, mode, "_q_fle");
set_conv_libfunc (sext_optab, mode, SFmode, "_q_stoq");
set_conv_libfunc (sext_optab, mode, DFmode, "_q_dtoq");
set_conv_libfunc (trunc_optab, SFmode, mode, "_q_qtos");
set_conv_libfunc (trunc_optab, DFmode, mode, "_q_qtod");
set_conv_libfunc (sfix_optab, SImode, mode, "_q_qtoi");
set_conv_libfunc (ufix_optab, SImode, mode, "_q_qtou");
set_conv_libfunc (sfloat_optab, mode, SImode, "_q_itoq");
set_conv_libfunc (ufloat_optab, mode, SImode, "_q_utoq");
}
}
static void
rs6000_init_libfuncs (void)
{
/* __float128 support. */
if (TARGET_FLOAT128_TYPE)
{
init_float128_ibm (IFmode);
init_float128_ieee (KFmode);
}
/* AIX/Darwin/64-bit Linux quad floating point routines. */
if (TARGET_LONG_DOUBLE_128)
{
if (!TARGET_IEEEQUAD)
init_float128_ibm (TFmode);
/* IEEE 128-bit including 32-bit SVR4 quad floating point routines. */
else
init_float128_ieee (TFmode);
}
}
/* Emit a potentially record-form instruction, setting DST from SRC.
If DOT is 0, that is all; otherwise, set CCREG to the result of the
signed comparison of DST with zero. If DOT is 1, the generated RTL
doesn't care about the DST result; if DOT is 2, it does. If CCREG
is CR0 do a single dot insn (as a PARALLEL); otherwise, do a SET and
a separate COMPARE. */
void
rs6000_emit_dot_insn (rtx dst, rtx src, int dot, rtx ccreg)
{
if (dot == 0)
{
emit_move_insn (dst, src);
return;
}
if (cc_reg_not_cr0_operand (ccreg, CCmode))
{
emit_move_insn (dst, src);
emit_move_insn (ccreg, gen_rtx_COMPARE (CCmode, dst, const0_rtx));
return;
}
rtx ccset = gen_rtx_SET (ccreg, gen_rtx_COMPARE (CCmode, src, const0_rtx));
if (dot == 1)
{
rtx clobber = gen_rtx_CLOBBER (VOIDmode, dst);
emit_insn (gen_rtx_PARALLEL (VOIDmode, gen_rtvec (2, ccset, clobber)));
}
else
{
rtx set = gen_rtx_SET (dst, src);
emit_insn (gen_rtx_PARALLEL (VOIDmode, gen_rtvec (2, ccset, set)));
}
}
/* A validation routine: say whether CODE, a condition code, and MODE
match. The other alternatives either don't make sense or should
never be generated. */
void
validate_condition_mode (enum rtx_code code, machine_mode mode)
{
gcc_assert ((GET_RTX_CLASS (code) == RTX_COMPARE
|| GET_RTX_CLASS (code) == RTX_COMM_COMPARE)
&& GET_MODE_CLASS (mode) == MODE_CC);
/* These don't make sense. */
gcc_assert ((code != GT && code != LT && code != GE && code != LE)
|| mode != CCUNSmode);
gcc_assert ((code != GTU && code != LTU && code != GEU && code != LEU)
|| mode == CCUNSmode);
gcc_assert (mode == CCFPmode
|| (code != ORDERED && code != UNORDERED
&& code != UNEQ && code != LTGT
&& code != UNGT && code != UNLT
&& code != UNGE && code != UNLE));
/* These should never be generated except for
flag_finite_math_only. */
gcc_assert (mode != CCFPmode
|| flag_finite_math_only
|| (code != LE && code != GE
&& code != UNEQ && code != LTGT
&& code != UNGT && code != UNLT));
/* These are invalid; the information is not there. */
gcc_assert (mode != CCEQmode || code == EQ || code == NE);
}
/* Return whether MASK (a CONST_INT) is a valid mask for any rlwinm,
rldicl, rldicr, or rldic instruction in mode MODE. If so, if E is
not zero, store there the bit offset (counted from the right) where
the single stretch of 1 bits begins; and similarly for B, the bit
offset where it ends. */
bool
rs6000_is_valid_mask (rtx mask, int *b, int *e, machine_mode mode)
{
unsigned HOST_WIDE_INT val = INTVAL (mask);
unsigned HOST_WIDE_INT bit;
int nb, ne;
int n = GET_MODE_PRECISION (mode);
if (mode != DImode && mode != SImode)
return false;
if (INTVAL (mask) >= 0)
{
bit = val & -val;
ne = exact_log2 (bit);
nb = exact_log2 (val + bit);
}
else if (val + 1 == 0)
{
nb = n;
ne = 0;
}
else if (val & 1)
{
val = ~val;
bit = val & -val;
nb = exact_log2 (bit);
ne = exact_log2 (val + bit);
}
else
{
bit = val & -val;
ne = exact_log2 (bit);
if (val + bit == 0)
nb = n;
else
nb = 0;
}
nb--;
if (nb < 0 || ne < 0 || nb >= n || ne >= n)
return false;
if (b)
*b = nb;
if (e)
*e = ne;
return true;
}
/* Return whether MASK (a CONST_INT) is a valid mask for any rlwinm, rldicl,
or rldicr instruction, to implement an AND with it in mode MODE. */
bool
rs6000_is_valid_and_mask (rtx mask, machine_mode mode)
{
int nb, ne;
if (!rs6000_is_valid_mask (mask, &nb, &ne, mode))
return false;
/* For DImode, we need a rldicl, rldicr, or a rlwinm with mask that
does not wrap. */
if (mode == DImode)
return (ne == 0 || nb == 63 || (nb < 32 && ne <= nb));
/* For SImode, rlwinm can do everything. */
if (mode == SImode)
return (nb < 32 && ne < 32);
return false;
}
/* Return the instruction template for an AND with mask in mode MODE, with
operands OPERANDS. If DOT is true, make it a record-form instruction. */
const char *
rs6000_insn_for_and_mask (machine_mode mode, rtx *operands, bool dot)
{
int nb, ne;
if (!rs6000_is_valid_mask (operands[2], &nb, &ne, mode))
gcc_unreachable ();
if (mode == DImode && ne == 0)
{
operands[3] = GEN_INT (63 - nb);
if (dot)
return "rldicl. %0,%1,0,%3";
return "rldicl %0,%1,0,%3";
}
if (mode == DImode && nb == 63)
{
operands[3] = GEN_INT (63 - ne);
if (dot)
return "rldicr. %0,%1,0,%3";
return "rldicr %0,%1,0,%3";
}
if (nb < 32 && ne < 32)
{
operands[3] = GEN_INT (31 - nb);
operands[4] = GEN_INT (31 - ne);
if (dot)
return "rlwinm. %0,%1,0,%3,%4";
return "rlwinm %0,%1,0,%3,%4";
}
gcc_unreachable ();
}
/* Return whether MASK (a CONST_INT) is a valid mask for any rlw[i]nm,
rld[i]cl, rld[i]cr, or rld[i]c instruction, to implement an AND with
shift SHIFT (a ROTATE, ASHIFT, or LSHIFTRT) in mode MODE. */
bool
rs6000_is_valid_shift_mask (rtx mask, rtx shift, machine_mode mode)
{
int nb, ne;
if (!rs6000_is_valid_mask (mask, &nb, &ne, mode))
return false;
int n = GET_MODE_PRECISION (mode);
int sh = -1;
if (CONST_INT_P (XEXP (shift, 1)))
{
sh = INTVAL (XEXP (shift, 1));
if (sh < 0 || sh >= n)
return false;
}
rtx_code code = GET_CODE (shift);
/* Convert any shift by 0 to a rotate, to simplify below code. */
if (sh == 0)
code = ROTATE;
/* Convert rotate to simple shift if we can, to make analysis simpler. */
if (code == ROTATE && sh >= 0 && nb >= ne && ne >= sh)
code = ASHIFT;
if (code == ROTATE && sh >= 0 && nb >= ne && nb < sh)
{
code = LSHIFTRT;
sh = n - sh;
}
/* DImode rotates need rld*. */
if (mode == DImode && code == ROTATE)
return (nb == 63 || ne == 0 || ne == sh);
/* SImode rotates need rlw*. */
if (mode == SImode && code == ROTATE)
return (nb < 32 && ne < 32 && sh < 32);
/* Wrap-around masks are only okay for rotates. */
if (ne > nb)
return false;
/* Variable shifts are only okay for rotates. */
if (sh < 0)
return false;
/* Don't allow ASHIFT if the mask is wrong for that. */
if (code == ASHIFT && ne < sh)
return false;
/* If we can do it with an rlw*, we can do it. Don't allow LSHIFTRT
if the mask is wrong for that. */
if (nb < 32 && ne < 32 && sh < 32
&& !(code == LSHIFTRT && nb >= 32 - sh))
return true;
/* If we can do it with an rld*, we can do it. Don't allow LSHIFTRT
if the mask is wrong for that. */
if (code == LSHIFTRT)
sh = 64 - sh;
if (nb == 63 || ne == 0 || ne == sh)
return !(code == LSHIFTRT && nb >= sh);
return false;
}
/* Return the instruction template for a shift with mask in mode MODE, with
operands OPERANDS. If DOT is true, make it a record-form instruction. */
const char *
rs6000_insn_for_shift_mask (machine_mode mode, rtx *operands, bool dot)
{
int nb, ne;
if (!rs6000_is_valid_mask (operands[3], &nb, &ne, mode))
gcc_unreachable ();
if (mode == DImode && ne == 0)
{
if (GET_CODE (operands[4]) == LSHIFTRT && INTVAL (operands[2]))
operands[2] = GEN_INT (64 - INTVAL (operands[2]));
operands[3] = GEN_INT (63 - nb);
if (dot)
return "rld%I2cl. %0,%1,%2,%3";
return "rld%I2cl %0,%1,%2,%3";
}
if (mode == DImode && nb == 63)
{
operands[3] = GEN_INT (63 - ne);
if (dot)
return "rld%I2cr. %0,%1,%2,%3";
return "rld%I2cr %0,%1,%2,%3";
}
if (mode == DImode
&& GET_CODE (operands[4]) != LSHIFTRT
&& CONST_INT_P (operands[2])
&& ne == INTVAL (operands[2]))
{
operands[3] = GEN_INT (63 - nb);
if (dot)
return "rld%I2c. %0,%1,%2,%3";
return "rld%I2c %0,%1,%2,%3";
}
if (nb < 32 && ne < 32)
{
if (GET_CODE (operands[4]) == LSHIFTRT && INTVAL (operands[2]))
operands[2] = GEN_INT (32 - INTVAL (operands[2]));
operands[3] = GEN_INT (31 - nb);
operands[4] = GEN_INT (31 - ne);
/* This insn can also be a 64-bit rotate with mask that really makes
it just a shift right (with mask); the %h below are to adjust for
that situation (shift count is >= 32 in that case). */
if (dot)
return "rlw%I2nm. %0,%1,%h2,%3,%4";
return "rlw%I2nm %0,%1,%h2,%3,%4";
}
gcc_unreachable ();
}
/* Return whether MASK (a CONST_INT) is a valid mask for any rlwimi or
rldimi instruction, to implement an insert with shift SHIFT (a ROTATE,
ASHIFT, or LSHIFTRT) in mode MODE. */
bool
rs6000_is_valid_insert_mask (rtx mask, rtx shift, machine_mode mode)
{
int nb, ne;
if (!rs6000_is_valid_mask (mask, &nb, &ne, mode))
return false;
int n = GET_MODE_PRECISION (mode);
int sh = INTVAL (XEXP (shift, 1));
if (sh < 0 || sh >= n)
return false;
rtx_code code = GET_CODE (shift);
/* Convert any shift by 0 to a rotate, to simplify below code. */
if (sh == 0)
code = ROTATE;
/* Convert rotate to simple shift if we can, to make analysis simpler. */
if (code == ROTATE && sh >= 0 && nb >= ne && ne >= sh)
code = ASHIFT;
if (code == ROTATE && sh >= 0 && nb >= ne && nb < sh)
{
code = LSHIFTRT;
sh = n - sh;
}
/* DImode rotates need rldimi. */
if (mode == DImode && code == ROTATE)
return (ne == sh);
/* SImode rotates need rlwimi. */
if (mode == SImode && code == ROTATE)
return (nb < 32 && ne < 32 && sh < 32);
/* Wrap-around masks are only okay for rotates. */
if (ne > nb)
return false;
/* Don't allow ASHIFT if the mask is wrong for that. */
if (code == ASHIFT && ne < sh)
return false;
/* If we can do it with an rlwimi, we can do it. Don't allow LSHIFTRT
if the mask is wrong for that. */
if (nb < 32 && ne < 32 && sh < 32
&& !(code == LSHIFTRT && nb >= 32 - sh))
return true;
/* If we can do it with an rldimi, we can do it. Don't allow LSHIFTRT
if the mask is wrong for that. */
if (code == LSHIFTRT)
sh = 64 - sh;
if (ne == sh)
return !(code == LSHIFTRT && nb >= sh);
return false;
}
/* Return the instruction template for an insert with mask in mode MODE, with
operands OPERANDS. If DOT is true, make it a record-form instruction. */
const char *
rs6000_insn_for_insert_mask (machine_mode mode, rtx *operands, bool dot)
{
int nb, ne;
if (!rs6000_is_valid_mask (operands[3], &nb, &ne, mode))
gcc_unreachable ();
/* Prefer rldimi because rlwimi is cracked. */
if (TARGET_POWERPC64
&& (!dot || mode == DImode)
&& GET_CODE (operands[4]) != LSHIFTRT
&& ne == INTVAL (operands[2]))
{
operands[3] = GEN_INT (63 - nb);
if (dot)
return "rldimi. %0,%1,%2,%3";
return "rldimi %0,%1,%2,%3";
}
if (nb < 32 && ne < 32)
{
if (GET_CODE (operands[4]) == LSHIFTRT && INTVAL (operands[2]))
operands[2] = GEN_INT (32 - INTVAL (operands[2]));
operands[3] = GEN_INT (31 - nb);
operands[4] = GEN_INT (31 - ne);
if (dot)
return "rlwimi. %0,%1,%2,%3,%4";
return "rlwimi %0,%1,%2,%3,%4";
}
gcc_unreachable ();
}
/* Return whether an AND with C (a CONST_INT) in mode MODE can be done
using two machine instructions. */
bool
rs6000_is_valid_2insn_and (rtx c, machine_mode mode)
{
/* There are two kinds of AND we can handle with two insns:
1) those we can do with two rl* insn;
2) ori[s];xori[s].
We do not handle that last case yet. */
/* If there is just one stretch of ones, we can do it. */
if (rs6000_is_valid_mask (c, NULL, NULL, mode))
return true;
/* Otherwise, fill in the lowest "hole"; if we can do the result with
one insn, we can do the whole thing with two. */
unsigned HOST_WIDE_INT val = INTVAL (c);
unsigned HOST_WIDE_INT bit1 = val & -val;
unsigned HOST_WIDE_INT bit2 = (val + bit1) & ~val;
unsigned HOST_WIDE_INT val1 = (val + bit1) & val;
unsigned HOST_WIDE_INT bit3 = val1 & -val1;
return rs6000_is_valid_and_mask (GEN_INT (val + bit3 - bit2), mode);
}
/* Emit the two insns to do an AND in mode MODE, with operands OPERANDS.
If EXPAND is true, split rotate-and-mask instructions we generate to
their constituent parts as well (this is used during expand); if DOT
is 1, make the last insn a record-form instruction clobbering the
destination GPR and setting the CC reg (from operands[3]); if 2, set
that GPR as well as the CC reg. */
void
rs6000_emit_2insn_and (machine_mode mode, rtx *operands, bool expand, int dot)
{
gcc_assert (!(expand && dot));
unsigned HOST_WIDE_INT val = INTVAL (operands[2]);
/* If it is one stretch of ones, it is DImode; shift left, mask, then
shift right. This generates better code than doing the masks without
shifts, or shifting first right and then left. */
int nb, ne;
if (rs6000_is_valid_mask (operands[2], &nb, &ne, mode) && nb >= ne)
{
gcc_assert (mode == DImode);
int shift = 63 - nb;
if (expand)
{
rtx tmp1 = gen_reg_rtx (DImode);
rtx tmp2 = gen_reg_rtx (DImode);
emit_insn (gen_ashldi3 (tmp1, operands[1], GEN_INT (shift)));
emit_insn (gen_anddi3 (tmp2, tmp1, GEN_INT (val << shift)));
emit_insn (gen_lshrdi3 (operands[0], tmp2, GEN_INT (shift)));
}
else
{
rtx tmp = gen_rtx_ASHIFT (mode, operands[1], GEN_INT (shift));
tmp = gen_rtx_AND (mode, tmp, GEN_INT (val << shift));
emit_move_insn (operands[0], tmp);
tmp = gen_rtx_LSHIFTRT (mode, operands[0], GEN_INT (shift));
rs6000_emit_dot_insn (operands[0], tmp, dot, dot ? operands[3] : 0);
}
return;
}
/* Otherwise, make a mask2 that cuts out the lowest "hole", and a mask1
that does the rest. */
unsigned HOST_WIDE_INT bit1 = val & -val;
unsigned HOST_WIDE_INT bit2 = (val + bit1) & ~val;
unsigned HOST_WIDE_INT val1 = (val + bit1) & val;
unsigned HOST_WIDE_INT bit3 = val1 & -val1;
unsigned HOST_WIDE_INT mask1 = -bit3 + bit2 - 1;
unsigned HOST_WIDE_INT mask2 = val + bit3 - bit2;
gcc_assert (rs6000_is_valid_and_mask (GEN_INT (mask2), mode));
/* Two "no-rotate"-and-mask instructions, for SImode. */
if (rs6000_is_valid_and_mask (GEN_INT (mask1), mode))
{
gcc_assert (mode == SImode);
rtx reg = expand ? gen_reg_rtx (mode) : operands[0];
rtx tmp = gen_rtx_AND (mode, operands[1], GEN_INT (mask1));
emit_move_insn (reg, tmp);
tmp = gen_rtx_AND (mode, reg, GEN_INT (mask2));
rs6000_emit_dot_insn (operands[0], tmp, dot, dot ? operands[3] : 0);
return;
}
gcc_assert (mode == DImode);
/* Two "no-rotate"-and-mask instructions, for DImode: both are rlwinm
insns; we have to do the first in SImode, because it wraps. */
if (mask2 <= 0xffffffff
&& rs6000_is_valid_and_mask (GEN_INT (mask1), SImode))
{
rtx reg = expand ? gen_reg_rtx (mode) : operands[0];
rtx tmp = gen_rtx_AND (SImode, gen_lowpart (SImode, operands[1]),
GEN_INT (mask1));
rtx reg_low = gen_lowpart (SImode, reg);
emit_move_insn (reg_low, tmp);
tmp = gen_rtx_AND (mode, reg, GEN_INT (mask2));
rs6000_emit_dot_insn (operands[0], tmp, dot, dot ? operands[3] : 0);
return;
}
/* Two rld* insns: rotate, clear the hole in the middle (which now is
at the top end), rotate back and clear the other hole. */
int right = exact_log2 (bit3);
int left = 64 - right;
/* Rotate the mask too. */
mask1 = (mask1 >> right) | ((bit2 - 1) << left);
if (expand)
{
rtx tmp1 = gen_reg_rtx (DImode);
rtx tmp2 = gen_reg_rtx (DImode);
rtx tmp3 = gen_reg_rtx (DImode);
emit_insn (gen_rotldi3 (tmp1, operands[1], GEN_INT (left)));
emit_insn (gen_anddi3 (tmp2, tmp1, GEN_INT (mask1)));
emit_insn (gen_rotldi3 (tmp3, tmp2, GEN_INT (right)));
emit_insn (gen_anddi3 (operands[0], tmp3, GEN_INT (mask2)));
}
else
{
rtx tmp = gen_rtx_ROTATE (mode, operands[1], GEN_INT (left));
tmp = gen_rtx_AND (mode, tmp, GEN_INT (mask1));
emit_move_insn (operands[0], tmp);
tmp = gen_rtx_ROTATE (mode, operands[0], GEN_INT (right));
tmp = gen_rtx_AND (mode, tmp, GEN_INT (mask2));
rs6000_emit_dot_insn (operands[0], tmp, dot, dot ? operands[3] : 0);
}
}
/* Return 1 if REGNO (reg1) == REGNO (reg2) - 1 making them candidates
for lfq and stfq insns iff the registers are hard registers. */
int
registers_ok_for_quad_peep (rtx reg1, rtx reg2)
{
/* We might have been passed a SUBREG. */
if (GET_CODE (reg1) != REG || GET_CODE (reg2) != REG)
return 0;
/* We might have been passed non floating point registers. */
if (!FP_REGNO_P (REGNO (reg1))
|| !FP_REGNO_P (REGNO (reg2)))
return 0;
return (REGNO (reg1) == REGNO (reg2) - 1);
}
/* Return 1 if addr1 and addr2 are suitable for lfq or stfq insn.
addr1 and addr2 must be in consecutive memory locations
(addr2 == addr1 + 8). */
int
mems_ok_for_quad_peep (rtx mem1, rtx mem2)
{
rtx addr1, addr2;
unsigned int reg1, reg2;
int offset1, offset2;
/* The mems cannot be volatile. */
if (MEM_VOLATILE_P (mem1) || MEM_VOLATILE_P (mem2))
return 0;
addr1 = XEXP (mem1, 0);
addr2 = XEXP (mem2, 0);
/* Extract an offset (if used) from the first addr. */
if (GET_CODE (addr1) == PLUS)
{
/* If not a REG, return zero. */
if (GET_CODE (XEXP (addr1, 0)) != REG)
return 0;
else
{
reg1 = REGNO (XEXP (addr1, 0));
/* The offset must be constant! */
if (GET_CODE (XEXP (addr1, 1)) != CONST_INT)
return 0;
offset1 = INTVAL (XEXP (addr1, 1));
}
}
else if (GET_CODE (addr1) != REG)
return 0;
else
{
reg1 = REGNO (addr1);
/* This was a simple (mem (reg)) expression. Offset is 0. */
offset1 = 0;
}
/* And now for the second addr. */
if (GET_CODE (addr2) == PLUS)
{
/* If not a REG, return zero. */
if (GET_CODE (XEXP (addr2, 0)) != REG)
return 0;
else
{
reg2 = REGNO (XEXP (addr2, 0));
/* The offset must be constant. */
if (GET_CODE (XEXP (addr2, 1)) != CONST_INT)
return 0;
offset2 = INTVAL (XEXP (addr2, 1));
}
}
else if (GET_CODE (addr2) != REG)
return 0;
else
{
reg2 = REGNO (addr2);
/* This was a simple (mem (reg)) expression. Offset is 0. */
offset2 = 0;
}
/* Both of these must have the same base register. */
if (reg1 != reg2)
return 0;
/* The offset for the second addr must be 8 more than the first addr. */
if (offset2 != offset1 + 8)
return 0;
/* All the tests passed. addr1 and addr2 are valid for lfq or stfq
instructions. */
return 1;
}
/* Implement TARGET_SECONDARY_RELOAD_NEEDED_MODE. For SDmode values we
need to use DDmode, in all other cases we can use the same mode. */
static machine_mode
rs6000_secondary_memory_needed_mode (machine_mode mode)
{
if (lra_in_progress && mode == SDmode)
return DDmode;
return mode;
}
/* Classify a register type. Because the FMRGOW/FMRGEW instructions only work
on traditional floating point registers, and the VMRGOW/VMRGEW instructions
only work on the traditional altivec registers, note if an altivec register
was chosen. */
static enum rs6000_reg_type
register_to_reg_type (rtx reg, bool *is_altivec)
{
HOST_WIDE_INT regno;
enum reg_class rclass;
if (GET_CODE (reg) == SUBREG)
reg = SUBREG_REG (reg);
if (!REG_P (reg))
return NO_REG_TYPE;
regno = REGNO (reg);
if (regno >= FIRST_PSEUDO_REGISTER)
{
if (!lra_in_progress && !reload_completed)
return PSEUDO_REG_TYPE;
regno = true_regnum (reg);
if (regno < 0 || regno >= FIRST_PSEUDO_REGISTER)
return PSEUDO_REG_TYPE;
}
gcc_assert (regno >= 0);
if (is_altivec && ALTIVEC_REGNO_P (regno))
*is_altivec = true;
rclass = rs6000_regno_regclass[regno];
return reg_class_to_reg_type[(int)rclass];
}
/* Helper function to return the cost of adding a TOC entry address. */
static inline int
rs6000_secondary_reload_toc_costs (addr_mask_type addr_mask)
{
int ret;
if (TARGET_CMODEL != CMODEL_SMALL)
ret = ((addr_mask & RELOAD_REG_OFFSET) == 0) ? 1 : 2;
else
ret = (TARGET_MINIMAL_TOC) ? 6 : 3;
return ret;
}
/* Helper function for rs6000_secondary_reload to determine whether the memory
address (ADDR) with a given register class (RCLASS) and machine mode (MODE)
needs reloading. Return negative if the memory is not handled by the memory
helper functions and to try a different reload method, 0 if no additional
instructions are need, and positive to give the extra cost for the
memory. */
static int
rs6000_secondary_reload_memory (rtx addr,
enum reg_class rclass,
machine_mode mode)
{
int extra_cost = 0;
rtx reg, and_arg, plus_arg0, plus_arg1;
addr_mask_type addr_mask;
const char *type = NULL;
const char *fail_msg = NULL;
if (GPR_REG_CLASS_P (rclass))
addr_mask = reg_addr[mode].addr_mask[RELOAD_REG_GPR];
else if (rclass == FLOAT_REGS)
addr_mask = reg_addr[mode].addr_mask[RELOAD_REG_FPR];
else if (rclass == ALTIVEC_REGS)
addr_mask = reg_addr[mode].addr_mask[RELOAD_REG_VMX];
/* For the combined VSX_REGS, turn off Altivec AND -16. */
else if (rclass == VSX_REGS)
addr_mask = (reg_addr[mode].addr_mask[RELOAD_REG_VMX]
& ~RELOAD_REG_AND_M16);
/* If the register allocator hasn't made up its mind yet on the register
class to use, settle on defaults to use. */
else if (rclass == NO_REGS)
{
addr_mask = (reg_addr[mode].addr_mask[RELOAD_REG_ANY]
& ~RELOAD_REG_AND_M16);
if ((addr_mask & RELOAD_REG_MULTIPLE) != 0)
addr_mask &= ~(RELOAD_REG_INDEXED
| RELOAD_REG_PRE_INCDEC
| RELOAD_REG_PRE_MODIFY);
}
else
addr_mask = 0;
/* If the register isn't valid in this register class, just return now. */
if ((addr_mask & RELOAD_REG_VALID) == 0)
{
if (TARGET_DEBUG_ADDR)
{
fprintf (stderr,
"rs6000_secondary_reload_memory: mode = %s, class = %s, "
"not valid in class\n",
GET_MODE_NAME (mode), reg_class_names[rclass]);
debug_rtx (addr);
}
return -1;
}
switch (GET_CODE (addr))
{
/* Does the register class supports auto update forms for this mode? We
don't need a scratch register, since the powerpc only supports
PRE_INC, PRE_DEC, and PRE_MODIFY. */
case PRE_INC:
case PRE_DEC:
reg = XEXP (addr, 0);
if (!base_reg_operand (addr, GET_MODE (reg)))
{
fail_msg = "no base register #1";
extra_cost = -1;
}
else if ((addr_mask & RELOAD_REG_PRE_INCDEC) == 0)
{
extra_cost = 1;
type = "update";
}
break;
case PRE_MODIFY:
reg = XEXP (addr, 0);
plus_arg1 = XEXP (addr, 1);
if (!base_reg_operand (reg, GET_MODE (reg))
|| GET_CODE (plus_arg1) != PLUS
|| !rtx_equal_p (reg, XEXP (plus_arg1, 0)))
{
fail_msg = "bad PRE_MODIFY";
extra_cost = -1;
}
else if ((addr_mask & RELOAD_REG_PRE_MODIFY) == 0)
{
extra_cost = 1;
type = "update";
}
break;
/* Do we need to simulate AND -16 to clear the bottom address bits used
in VMX load/stores? Only allow the AND for vector sizes. */
case AND:
and_arg = XEXP (addr, 0);
if (GET_MODE_SIZE (mode) != 16
|| GET_CODE (XEXP (addr, 1)) != CONST_INT
|| INTVAL (XEXP (addr, 1)) != -16)
{
fail_msg = "bad Altivec AND #1";
extra_cost = -1;
}
if (rclass != ALTIVEC_REGS)
{
if (legitimate_indirect_address_p (and_arg, false))
extra_cost = 1;
else if (legitimate_indexed_address_p (and_arg, false))
extra_cost = 2;
else
{
fail_msg = "bad Altivec AND #2";
extra_cost = -1;
}
type = "and";
}
break;
/* If this is an indirect address, make sure it is a base register. */
case REG:
case SUBREG:
if (!legitimate_indirect_address_p (addr, false))
{
extra_cost = 1;
type = "move";
}
break;
/* If this is an indexed address, make sure the register class can handle
indexed addresses for this mode. */
case PLUS:
plus_arg0 = XEXP (addr, 0);
plus_arg1 = XEXP (addr, 1);
/* (plus (plus (reg) (constant)) (constant)) is generated during
push_reload processing, so handle it now. */
if (GET_CODE (plus_arg0) == PLUS && CONST_INT_P (plus_arg1))
{
if ((addr_mask & RELOAD_REG_OFFSET) == 0)
{
extra_cost = 1;
type = "offset";
}
}
/* (plus (plus (reg) (constant)) (reg)) is also generated during
push_reload processing, so handle it now. */
else if (GET_CODE (plus_arg0) == PLUS && REG_P (plus_arg1))
{
if ((addr_mask & RELOAD_REG_INDEXED) == 0)
{
extra_cost = 1;
type = "indexed #2";
}
}
else if (!base_reg_operand (plus_arg0, GET_MODE (plus_arg0)))
{
fail_msg = "no base register #2";
extra_cost = -1;
}
else if (int_reg_operand (plus_arg1, GET_MODE (plus_arg1)))
{
if ((addr_mask & RELOAD_REG_INDEXED) == 0
|| !legitimate_indexed_address_p (addr, false))
{
extra_cost = 1;
type = "indexed";
}
}
else if ((addr_mask & RELOAD_REG_QUAD_OFFSET) != 0
&& CONST_INT_P (plus_arg1))
{
if (!quad_address_offset_p (INTVAL (plus_arg1)))
{
extra_cost = 1;
type = "vector d-form offset";
}
}
/* Make sure the register class can handle offset addresses. */
else if (rs6000_legitimate_offset_address_p (mode, addr, false, true))
{
if ((addr_mask & RELOAD_REG_OFFSET) == 0)
{
extra_cost = 1;
type = "offset #2";
}
}
else
{
fail_msg = "bad PLUS";
extra_cost = -1;
}
break;
case LO_SUM:
/* Quad offsets are restricted and can't handle normal addresses. */
if ((addr_mask & RELOAD_REG_QUAD_OFFSET) != 0)
{
extra_cost = -1;
type = "vector d-form lo_sum";
}
else if (!legitimate_lo_sum_address_p (mode, addr, false))
{
fail_msg = "bad LO_SUM";
extra_cost = -1;
}
if ((addr_mask & RELOAD_REG_OFFSET) == 0)
{
extra_cost = 1;
type = "lo_sum";
}
break;
/* Static addresses need to create a TOC entry. */
case CONST:
case SYMBOL_REF:
case LABEL_REF:
if ((addr_mask & RELOAD_REG_QUAD_OFFSET) != 0)
{
extra_cost = -1;
type = "vector d-form lo_sum #2";
}
else
{
type = "address";
extra_cost = rs6000_secondary_reload_toc_costs (addr_mask);
}
break;
/* TOC references look like offsetable memory. */
case UNSPEC:
if (TARGET_CMODEL == CMODEL_SMALL || XINT (addr, 1) != UNSPEC_TOCREL)
{
fail_msg = "bad UNSPEC";
extra_cost = -1;
}
else if ((addr_mask & RELOAD_REG_QUAD_OFFSET) != 0)
{
extra_cost = -1;
type = "vector d-form lo_sum #3";
}
else if ((addr_mask & RELOAD_REG_OFFSET) == 0)
{
extra_cost = 1;
type = "toc reference";
}
break;
default:
{
fail_msg = "bad address";
extra_cost = -1;
}
}
if (TARGET_DEBUG_ADDR /* && extra_cost != 0 */)
{
if (extra_cost < 0)
fprintf (stderr,
"rs6000_secondary_reload_memory error: mode = %s, "
"class = %s, addr_mask = '%s', %s\n",
GET_MODE_NAME (mode),
reg_class_names[rclass],
rs6000_debug_addr_mask (addr_mask, false),
(fail_msg != NULL) ? fail_msg : "<bad address>");
else
fprintf (stderr,
"rs6000_secondary_reload_memory: mode = %s, class = %s, "
"addr_mask = '%s', extra cost = %d, %s\n",
GET_MODE_NAME (mode),
reg_class_names[rclass],
rs6000_debug_addr_mask (addr_mask, false),
extra_cost,
(type) ? type : "<none>");
debug_rtx (addr);
}
return extra_cost;
}
/* Helper function for rs6000_secondary_reload to return true if a move to a
different register classe is really a simple move. */
static bool
rs6000_secondary_reload_simple_move (enum rs6000_reg_type to_type,
enum rs6000_reg_type from_type,
machine_mode mode)
{
int size = GET_MODE_SIZE (mode);
/* Add support for various direct moves available. In this function, we only
look at cases where we don't need any extra registers, and one or more
simple move insns are issued. Originally small integers are not allowed
in FPR/VSX registers. Single precision binary floating is not a simple
move because we need to convert to the single precision memory layout.
The 4-byte SDmode can be moved. TDmode values are disallowed since they
need special direct move handling, which we do not support yet. */
if (TARGET_DIRECT_MOVE
&& ((to_type == GPR_REG_TYPE && from_type == VSX_REG_TYPE)
|| (to_type == VSX_REG_TYPE && from_type == GPR_REG_TYPE)))
{
if (TARGET_POWERPC64)
{
/* ISA 2.07: MTVSRD or MVFVSRD. */
if (size == 8)
return true;
/* ISA 3.0: MTVSRDD or MFVSRD + MFVSRLD. */
if (size == 16 && TARGET_P9_VECTOR && mode != TDmode)
return true;
}
/* ISA 2.07: MTVSRWZ or MFVSRWZ. */
if (TARGET_P8_VECTOR)
{
if (mode == SImode)
return true;
if (TARGET_P9_VECTOR && (mode == HImode || mode == QImode))
return true;
}
/* ISA 2.07: MTVSRWZ or MFVSRWZ. */
if (mode == SDmode)
return true;
}
/* Power6+: MFTGPR or MFFGPR. */
else if (TARGET_MFPGPR && TARGET_POWERPC64 && size == 8
&& ((to_type == GPR_REG_TYPE && from_type == FPR_REG_TYPE)
|| (to_type == FPR_REG_TYPE && from_type == GPR_REG_TYPE)))
return true;
/* Move to/from SPR. */
else if ((size == 4 || (TARGET_POWERPC64 && size == 8))
&& ((to_type == GPR_REG_TYPE && from_type == SPR_REG_TYPE)
|| (to_type == SPR_REG_TYPE && from_type == GPR_REG_TYPE)))
return true;
return false;
}
/* Direct move helper function for rs6000_secondary_reload, handle all of the
special direct moves that involve allocating an extra register, return the
insn code of the helper function if there is such a function or
CODE_FOR_nothing if not. */
static bool
rs6000_secondary_reload_direct_move (enum rs6000_reg_type to_type,
enum rs6000_reg_type from_type,
machine_mode mode,
secondary_reload_info *sri,
bool altivec_p)
{
bool ret = false;
enum insn_code icode = CODE_FOR_nothing;
int cost = 0;
int size = GET_MODE_SIZE (mode);
if (TARGET_POWERPC64 && size == 16)
{
/* Handle moving 128-bit values from GPRs to VSX point registers on
ISA 2.07 (power8, power9) when running in 64-bit mode using
XXPERMDI to glue the two 64-bit values back together. */
if (to_type == VSX_REG_TYPE && from_type == GPR_REG_TYPE)
{
cost = 3; /* 2 mtvsrd's, 1 xxpermdi. */
icode = reg_addr[mode].reload_vsx_gpr;
}
/* Handle moving 128-bit values from VSX point registers to GPRs on
ISA 2.07 when running in 64-bit mode using XXPERMDI to get access to the
bottom 64-bit value. */
else if (to_type == GPR_REG_TYPE && from_type == VSX_REG_TYPE)
{
cost = 3; /* 2 mfvsrd's, 1 xxpermdi. */
icode = reg_addr[mode].reload_gpr_vsx;
}
}
else if (TARGET_POWERPC64 && mode == SFmode)
{
if (to_type == GPR_REG_TYPE && from_type == VSX_REG_TYPE)
{
cost = 3; /* xscvdpspn, mfvsrd, and. */
icode = reg_addr[mode].reload_gpr_vsx;
}
else if (to_type == VSX_REG_TYPE && from_type == GPR_REG_TYPE)
{
cost = 2; /* mtvsrz, xscvspdpn. */
icode = reg_addr[mode].reload_vsx_gpr;
}
}
else if (!TARGET_POWERPC64 && size == 8)
{
/* Handle moving 64-bit values from GPRs to floating point registers on
ISA 2.07 when running in 32-bit mode using FMRGOW to glue the two
32-bit values back together. Altivec register classes must be handled
specially since a different instruction is used, and the secondary
reload support requires a single instruction class in the scratch
register constraint. However, right now TFmode is not allowed in
Altivec registers, so the pattern will never match. */
if (to_type == VSX_REG_TYPE && from_type == GPR_REG_TYPE && !altivec_p)
{
cost = 3; /* 2 mtvsrwz's, 1 fmrgow. */
icode = reg_addr[mode].reload_fpr_gpr;
}
}
if (icode != CODE_FOR_nothing)
{
ret = true;
if (sri)
{
sri->icode = icode;
sri->extra_cost = cost;
}
}
return ret;
}
/* Return whether a move between two register classes can be done either
directly (simple move) or via a pattern that uses a single extra temporary
(using ISA 2.07's direct move in this case. */
static bool
rs6000_secondary_reload_move (enum rs6000_reg_type to_type,
enum rs6000_reg_type from_type,
machine_mode mode,
secondary_reload_info *sri,
bool altivec_p)
{
/* Fall back to load/store reloads if either type is not a register. */
if (to_type == NO_REG_TYPE || from_type == NO_REG_TYPE)
return false;
/* If we haven't allocated registers yet, assume the move can be done for the
standard register types. */
if ((to_type == PSEUDO_REG_TYPE && from_type == PSEUDO_REG_TYPE)
|| (to_type == PSEUDO_REG_TYPE && IS_STD_REG_TYPE (from_type))
|| (from_type == PSEUDO_REG_TYPE && IS_STD_REG_TYPE (to_type)))
return true;
/* Moves to the same set of registers is a simple move for non-specialized
registers. */
if (to_type == from_type && IS_STD_REG_TYPE (to_type))
return true;
/* Check whether a simple move can be done directly. */
if (rs6000_secondary_reload_simple_move (to_type, from_type, mode))
{
if (sri)
{
sri->icode = CODE_FOR_nothing;
sri->extra_cost = 0;
}
return true;
}
/* Now check if we can do it in a few steps. */
return rs6000_secondary_reload_direct_move (to_type, from_type, mode, sri,
altivec_p);
}
/* Inform reload about cases where moving X with a mode MODE to a register in
RCLASS requires an extra scratch or immediate register. Return the class
needed for the immediate register.
For VSX and Altivec, we may need a register to convert sp+offset into
reg+sp.
For misaligned 64-bit gpr loads and stores we need a register to
convert an offset address to indirect. */
static reg_class_t
rs6000_secondary_reload (bool in_p,
rtx x,
reg_class_t rclass_i,
machine_mode mode,
secondary_reload_info *sri)
{
enum reg_class rclass = (enum reg_class) rclass_i;
reg_class_t ret = ALL_REGS;
enum insn_code icode;
bool default_p = false;
bool done_p = false;
/* Allow subreg of memory before/during reload. */
bool memory_p = (MEM_P (x)
|| (!reload_completed && GET_CODE (x) == SUBREG
&& MEM_P (SUBREG_REG (x))));
sri->icode = CODE_FOR_nothing;
sri->t_icode = CODE_FOR_nothing;
sri->extra_cost = 0;
icode = ((in_p)
? reg_addr[mode].reload_load
: reg_addr[mode].reload_store);
if (REG_P (x) || register_operand (x, mode))
{
enum rs6000_reg_type to_type = reg_class_to_reg_type[(int)rclass];
bool altivec_p = (rclass == ALTIVEC_REGS);
enum rs6000_reg_type from_type = register_to_reg_type (x, &altivec_p);
if (!in_p)
std::swap (to_type, from_type);
/* Can we do a direct move of some sort? */
if (rs6000_secondary_reload_move (to_type, from_type, mode, sri,
altivec_p))
{
icode = (enum insn_code)sri->icode;
default_p = false;
done_p = true;
ret = NO_REGS;
}
}
/* Make sure 0.0 is not reloaded or forced into memory. */
if (x == CONST0_RTX (mode) && VSX_REG_CLASS_P (rclass))
{
ret = NO_REGS;
default_p = false;
done_p = true;
}
/* If this is a scalar floating point value and we want to load it into the
traditional Altivec registers, do it via a move via a traditional floating
point register, unless we have D-form addressing. Also make sure that
non-zero constants use a FPR. */
if (!done_p && reg_addr[mode].scalar_in_vmx_p
&& !mode_supports_vmx_dform (mode)
&& (rclass == VSX_REGS || rclass == ALTIVEC_REGS)
&& (memory_p || (GET_CODE (x) == CONST_DOUBLE)))
{
ret = FLOAT_REGS;
default_p = false;
done_p = true;
}
/* Handle reload of load/stores if we have reload helper functions. */
if (!done_p && icode != CODE_FOR_nothing && memory_p)
{
int extra_cost = rs6000_secondary_reload_memory (XEXP (x, 0), rclass,
mode);
if (extra_cost >= 0)
{
done_p = true;
ret = NO_REGS;
if (extra_cost > 0)
{
sri->extra_cost = extra_cost;
sri->icode = icode;
}
}
}
/* Handle unaligned loads and stores of integer registers. */
if (!done_p && TARGET_POWERPC64
&& reg_class_to_reg_type[(int)rclass] == GPR_REG_TYPE
&& memory_p
&& GET_MODE_SIZE (GET_MODE (x)) >= UNITS_PER_WORD)
{
rtx addr = XEXP (x, 0);
rtx off = address_offset (addr);
if (off != NULL_RTX)
{
unsigned int extra = GET_MODE_SIZE (GET_MODE (x)) - UNITS_PER_WORD;
unsigned HOST_WIDE_INT offset = INTVAL (off);
/* We need a secondary reload when our legitimate_address_p
says the address is good (as otherwise the entire address
will be reloaded), and the offset is not a multiple of
four or we have an address wrap. Address wrap will only
occur for LO_SUMs since legitimate_offset_address_p
rejects addresses for 16-byte mems that will wrap. */
if (GET_CODE (addr) == LO_SUM
? (1 /* legitimate_address_p allows any offset for lo_sum */
&& ((offset & 3) != 0
|| ((offset & 0xffff) ^ 0x8000) >= 0x10000 - extra))
: (offset + 0x8000 < 0x10000 - extra /* legitimate_address_p */
&& (offset & 3) != 0))
{
/* -m32 -mpowerpc64 needs to use a 32-bit scratch register. */
if (in_p)
sri->icode = ((TARGET_32BIT) ? CODE_FOR_reload_si_load
: CODE_FOR_reload_di_load);
else
sri->icode = ((TARGET_32BIT) ? CODE_FOR_reload_si_store
: CODE_FOR_reload_di_store);
sri->extra_cost = 2;
ret = NO_REGS;
done_p = true;
}
else
default_p = true;
}
else
default_p = true;
}
if (!done_p && !TARGET_POWERPC64
&& reg_class_to_reg_type[(int)rclass] == GPR_REG_TYPE
&& memory_p
&& GET_MODE_SIZE (GET_MODE (x)) > UNITS_PER_WORD)
{
rtx addr = XEXP (x, 0);
rtx off = address_offset (addr);
if (off != NULL_RTX)
{
unsigned int extra = GET_MODE_SIZE (GET_MODE (x)) - UNITS_PER_WORD;
unsigned HOST_WIDE_INT offset = INTVAL (off);
/* We need a secondary reload when our legitimate_address_p
says the address is good (as otherwise the entire address
will be reloaded), and we have a wrap.
legitimate_lo_sum_address_p allows LO_SUM addresses to
have any offset so test for wrap in the low 16 bits.
legitimate_offset_address_p checks for the range
[-0x8000,0x7fff] for mode size of 8 and [-0x8000,0x7ff7]
for mode size of 16. We wrap at [0x7ffc,0x7fff] and
[0x7ff4,0x7fff] respectively, so test for the
intersection of these ranges, [0x7ffc,0x7fff] and
[0x7ff4,0x7ff7] respectively.
Note that the address we see here may have been
manipulated by legitimize_reload_address. */
if (GET_CODE (addr) == LO_SUM
? ((offset & 0xffff) ^ 0x8000) >= 0x10000 - extra
: offset - (0x8000 - extra) < UNITS_PER_WORD)
{
if (in_p)
sri->icode = CODE_FOR_reload_si_load;
else
sri->icode = CODE_FOR_reload_si_store;
sri->extra_cost = 2;
ret = NO_REGS;
done_p = true;
}
else
default_p = true;
}
else
default_p = true;
}
if (!done_p)
default_p = true;
if (default_p)
ret = default_secondary_reload (in_p, x, rclass, mode, sri);
gcc_assert (ret != ALL_REGS);
if (TARGET_DEBUG_ADDR)
{
fprintf (stderr,
"\nrs6000_secondary_reload, return %s, in_p = %s, rclass = %s, "
"mode = %s",
reg_class_names[ret],
in_p ? "true" : "false",
reg_class_names[rclass],
GET_MODE_NAME (mode));
if (reload_completed)
fputs (", after reload", stderr);
if (!done_p)
fputs (", done_p not set", stderr);
if (default_p)
fputs (", default secondary reload", stderr);
if (sri->icode != CODE_FOR_nothing)
fprintf (stderr, ", reload func = %s, extra cost = %d",
insn_data[sri->icode].name, sri->extra_cost);
else if (sri->extra_cost > 0)
fprintf (stderr, ", extra cost = %d", sri->extra_cost);
fputs ("\n", stderr);
debug_rtx (x);
}
return ret;
}
/* Better tracing for rs6000_secondary_reload_inner. */
static void
rs6000_secondary_reload_trace (int line, rtx reg, rtx mem, rtx scratch,
bool store_p)
{
rtx set, clobber;
gcc_assert (reg != NULL_RTX && mem != NULL_RTX && scratch != NULL_RTX);
fprintf (stderr, "rs6000_secondary_reload_inner:%d, type = %s\n", line,
store_p ? "store" : "load");
if (store_p)
set = gen_rtx_SET (mem, reg);
else
set = gen_rtx_SET (reg, mem);
clobber = gen_rtx_CLOBBER (VOIDmode, scratch);
debug_rtx (gen_rtx_PARALLEL (VOIDmode, gen_rtvec (2, set, clobber)));
}
static void rs6000_secondary_reload_fail (int, rtx, rtx, rtx, bool)
ATTRIBUTE_NORETURN;
static void
rs6000_secondary_reload_fail (int line, rtx reg, rtx mem, rtx scratch,
bool store_p)
{
rs6000_secondary_reload_trace (line, reg, mem, scratch, store_p);
gcc_unreachable ();
}
/* Fixup reload addresses for values in GPR, FPR, and VMX registers that have
reload helper functions. These were identified in
rs6000_secondary_reload_memory, and if reload decided to use the secondary
reload, it calls the insns:
reload_<RELOAD:mode>_<P:mptrsize>_store
reload_<RELOAD:mode>_<P:mptrsize>_load
which in turn calls this function, to do whatever is necessary to create
valid addresses. */
void
rs6000_secondary_reload_inner (rtx reg, rtx mem, rtx scratch, bool store_p)
{
int regno = true_regnum (reg);
machine_mode mode = GET_MODE (reg);
addr_mask_type addr_mask;
rtx addr;
rtx new_addr;
rtx op_reg, op0, op1;
rtx and_op;
rtx cc_clobber;
rtvec rv;
if (regno < 0 || regno >= FIRST_PSEUDO_REGISTER || !MEM_P (mem)
|| !base_reg_operand (scratch, GET_MODE (scratch)))
rs6000_secondary_reload_fail (__LINE__, reg, mem, scratch, store_p);
if (IN_RANGE (regno, FIRST_GPR_REGNO, LAST_GPR_REGNO))
addr_mask = reg_addr[mode].addr_mask[RELOAD_REG_GPR];
else if (IN_RANGE (regno, FIRST_FPR_REGNO, LAST_FPR_REGNO))
addr_mask = reg_addr[mode].addr_mask[RELOAD_REG_FPR];
else if (IN_RANGE (regno, FIRST_ALTIVEC_REGNO, LAST_ALTIVEC_REGNO))
addr_mask = reg_addr[mode].addr_mask[RELOAD_REG_VMX];
else
rs6000_secondary_reload_fail (__LINE__, reg, mem, scratch, store_p);
/* Make sure the mode is valid in this register class. */
if ((addr_mask & RELOAD_REG_VALID) == 0)
rs6000_secondary_reload_fail (__LINE__, reg, mem, scratch, store_p);
if (TARGET_DEBUG_ADDR)
rs6000_secondary_reload_trace (__LINE__, reg, mem, scratch, store_p);
new_addr = addr = XEXP (mem, 0);
switch (GET_CODE (addr))
{
/* Does the register class support auto update forms for this mode? If
not, do the update now. We don't need a scratch register, since the
powerpc only supports PRE_INC, PRE_DEC, and PRE_MODIFY. */
case PRE_INC:
case PRE_DEC:
op_reg = XEXP (addr, 0);
if (!base_reg_operand (op_reg, Pmode))
rs6000_secondary_reload_fail (__LINE__, reg, mem, scratch, store_p);
if ((addr_mask & RELOAD_REG_PRE_INCDEC) == 0)
{
emit_insn (gen_add2_insn (op_reg, GEN_INT (GET_MODE_SIZE (mode))));
new_addr = op_reg;
}
break;
case PRE_MODIFY:
op0 = XEXP (addr, 0);
op1 = XEXP (addr, 1);
if (!base_reg_operand (op0, Pmode)
|| GET_CODE (op1) != PLUS
|| !rtx_equal_p (op0, XEXP (op1, 0)))
rs6000_secondary_reload_fail (__LINE__, reg, mem, scratch, store_p);
if ((addr_mask & RELOAD_REG_PRE_MODIFY) == 0)
{
emit_insn (gen_rtx_SET (op0, op1));
new_addr = reg;
}
break;
/* Do we need to simulate AND -16 to clear the bottom address bits used
in VMX load/stores? */
case AND:
op0 = XEXP (addr, 0);
op1 = XEXP (addr, 1);
if ((addr_mask & RELOAD_REG_AND_M16) == 0)
{
if (REG_P (op0) || GET_CODE (op0) == SUBREG)
op_reg = op0;
else if (GET_CODE (op1) == PLUS)
{
emit_insn (gen_rtx_SET (scratch, op1));
op_reg = scratch;
}
else
rs6000_secondary_reload_fail (__LINE__, reg, mem, scratch, store_p);
and_op = gen_rtx_AND (GET_MODE (scratch), op_reg, op1);
cc_clobber = gen_rtx_CLOBBER (VOIDmode, gen_rtx_SCRATCH (CCmode));
rv = gen_rtvec (2, gen_rtx_SET (scratch, and_op), cc_clobber);
emit_insn (gen_rtx_PARALLEL (VOIDmode, rv));
new_addr = scratch;
}
break;
/* If this is an indirect address, make sure it is a base register. */
case REG:
case SUBREG:
if (!base_reg_operand (addr, GET_MODE (addr)))
{
emit_insn (gen_rtx_SET (scratch, addr));
new_addr = scratch;
}
break;
/* If this is an indexed address, make sure the register class can handle
indexed addresses for this mode. */
case PLUS:
op0 = XEXP (addr, 0);
op1 = XEXP (addr, 1);
if (!base_reg_operand (op0, Pmode))
rs6000_secondary_reload_fail (__LINE__, reg, mem, scratch, store_p);
else if (int_reg_operand (op1, Pmode))
{
if ((addr_mask & RELOAD_REG_INDEXED) == 0)
{
emit_insn (gen_rtx_SET (scratch, addr));
new_addr = scratch;
}
}
else if (mode_supports_vsx_dform_quad (mode) && CONST_INT_P (op1))
{
if (((addr_mask & RELOAD_REG_QUAD_OFFSET) == 0)
|| !quad_address_p (addr, mode, false))
{
emit_insn (gen_rtx_SET (scratch, addr));
new_addr = scratch;
}
}
/* Make sure the register class can handle offset addresses. */
else if (rs6000_legitimate_offset_address_p (mode, addr, false, true))
{
if ((addr_mask & RELOAD_REG_OFFSET) == 0)
{
emit_insn (gen_rtx_SET (scratch, addr));
new_addr = scratch;
}
}
else
rs6000_secondary_reload_fail (__LINE__, reg, mem, scratch, store_p);
break;
case LO_SUM:
op0 = XEXP (addr, 0);
op1 = XEXP (addr, 1);
if (!base_reg_operand (op0, Pmode))
rs6000_secondary_reload_fail (__LINE__, reg, mem, scratch, store_p);
else if (int_reg_operand (op1, Pmode))
{
if ((addr_mask & RELOAD_REG_INDEXED) == 0)
{
emit_insn (gen_rtx_SET (scratch, addr));
new_addr = scratch;
}
}
/* Quad offsets are restricted and can't handle normal addresses. */
else if (mode_supports_vsx_dform_quad (mode))
{
emit_insn (gen_rtx_SET (scratch, addr));
new_addr = scratch;
}
/* Make sure the register class can handle offset addresses. */
else if (legitimate_lo_sum_address_p (mode, addr, false))
{
if ((addr_mask & RELOAD_REG_OFFSET) == 0)
{
emit_insn (gen_rtx_SET (scratch, addr));
new_addr = scratch;
}
}
else
rs6000_secondary_reload_fail (__LINE__, reg, mem, scratch, store_p);
break;
case SYMBOL_REF:
case CONST:
case LABEL_REF:
rs6000_emit_move (scratch, addr, Pmode);
new_addr = scratch;
break;
default:
rs6000_secondary_reload_fail (__LINE__, reg, mem, scratch, store_p);
}
/* Adjust the address if it changed. */
if (addr != new_addr)
{
mem = replace_equiv_address_nv (mem, new_addr);
if (TARGET_DEBUG_ADDR)
fprintf (stderr, "\nrs6000_secondary_reload_inner, mem adjusted.\n");
}
/* Now create the move. */
if (store_p)
emit_insn (gen_rtx_SET (mem, reg));
else
emit_insn (gen_rtx_SET (reg, mem));
return;
}
/* Convert reloads involving 64-bit gprs and misaligned offset
addressing, or multiple 32-bit gprs and offsets that are too large,
to use indirect addressing. */
void
rs6000_secondary_reload_gpr (rtx reg, rtx mem, rtx scratch, bool store_p)
{
int regno = true_regnum (reg);
enum reg_class rclass;
rtx addr;
rtx scratch_or_premodify = scratch;
if (TARGET_DEBUG_ADDR)
{
fprintf (stderr, "\nrs6000_secondary_reload_gpr, type = %s\n",
store_p ? "store" : "load");
fprintf (stderr, "reg:\n");
debug_rtx (reg);
fprintf (stderr, "mem:\n");
debug_rtx (mem);
fprintf (stderr, "scratch:\n");
debug_rtx (scratch);
}
gcc_assert (regno >= 0 && regno < FIRST_PSEUDO_REGISTER);
gcc_assert (GET_CODE (mem) == MEM);
rclass = REGNO_REG_CLASS (regno);
gcc_assert (rclass == GENERAL_REGS || rclass == BASE_REGS);
addr = XEXP (mem, 0);
if (GET_CODE (addr) == PRE_MODIFY)
{
gcc_assert (REG_P (XEXP (addr, 0))
&& GET_CODE (XEXP (addr, 1)) == PLUS
&& XEXP (XEXP (addr, 1), 0) == XEXP (addr, 0));
scratch_or_premodify = XEXP (addr, 0);
if (!HARD_REGISTER_P (scratch_or_premodify))
/* If we have a pseudo here then reload will have arranged
to have it replaced, but only in the original insn.
Use the replacement here too. */
scratch_or_premodify = find_replacement (&XEXP (addr, 0));
/* RTL emitted by rs6000_secondary_reload_gpr uses RTL
expressions from the original insn, without unsharing them.
Any RTL that points into the original insn will of course
have register replacements applied. That is why we don't
need to look for replacements under the PLUS. */
addr = XEXP (addr, 1);
}
gcc_assert (GET_CODE (addr) == PLUS || GET_CODE (addr) == LO_SUM);
rs6000_emit_move (scratch_or_premodify, addr, Pmode);
mem = replace_equiv_address_nv (mem, scratch_or_premodify);
/* Now create the move. */
if (store_p)
emit_insn (gen_rtx_SET (mem, reg));
else
emit_insn (gen_rtx_SET (reg, mem));
return;
}
/* Given an rtx X being reloaded into a reg required to be
in class CLASS, return the class of reg to actually use.
In general this is just CLASS; but on some machines
in some cases it is preferable to use a more restrictive class.
On the RS/6000, we have to return NO_REGS when we want to reload a
floating-point CONST_DOUBLE to force it to be copied to memory.
We also don't want to reload integer values into floating-point
registers if we can at all help it. In fact, this can
cause reload to die, if it tries to generate a reload of CTR
into a FP register and discovers it doesn't have the memory location
required.
??? Would it be a good idea to have reload do the converse, that is
try to reload floating modes into FP registers if possible?
*/
static enum reg_class
rs6000_preferred_reload_class (rtx x, enum reg_class rclass)
{
machine_mode mode = GET_MODE (x);
bool is_constant = CONSTANT_P (x);
/* If a mode can't go in FPR/ALTIVEC/VSX registers, don't return a preferred
reload class for it. */
if ((rclass == ALTIVEC_REGS || rclass == VSX_REGS)
&& (reg_addr[mode].addr_mask[RELOAD_REG_VMX] & RELOAD_REG_VALID) == 0)
return NO_REGS;
if ((rclass == FLOAT_REGS || rclass == VSX_REGS)
&& (reg_addr[mode].addr_mask[RELOAD_REG_FPR] & RELOAD_REG_VALID) == 0)
return NO_REGS;
/* For VSX, see if we should prefer FLOAT_REGS or ALTIVEC_REGS. Do not allow
the reloading of address expressions using PLUS into floating point
registers. */
if (TARGET_VSX && VSX_REG_CLASS_P (rclass) && GET_CODE (x) != PLUS)
{
if (is_constant)
{
/* Zero is always allowed in all VSX registers. */
if (x == CONST0_RTX (mode))
return rclass;
/* If this is a vector constant that can be formed with a few Altivec
instructions, we want altivec registers. */
if (GET_CODE (x) == CONST_VECTOR && easy_vector_constant (x, mode))
return ALTIVEC_REGS;
/* If this is an integer constant that can easily be loaded into
vector registers, allow it. */
if (CONST_INT_P (x))
{
HOST_WIDE_INT value = INTVAL (x);
/* ISA 2.07 can generate -1 in all registers with XXLORC. ISA
2.06 can generate it in the Altivec registers with
VSPLTI<x>. */
if (value == -1)
{
if (TARGET_P8_VECTOR)
return rclass;
else if (rclass == ALTIVEC_REGS || rclass == VSX_REGS)
return ALTIVEC_REGS;
else
return NO_REGS;
}
/* ISA 3.0 can load -128..127 using the XXSPLTIB instruction and
a sign extend in the Altivec registers. */
if (IN_RANGE (value, -128, 127) && TARGET_P9_VECTOR
&& (rclass == ALTIVEC_REGS || rclass == VSX_REGS))
return ALTIVEC_REGS;
}
/* Force constant to memory. */
return NO_REGS;
}
/* D-form addressing can easily reload the value. */
if (mode_supports_vmx_dform (mode)
|| mode_supports_vsx_dform_quad (mode))
return rclass;
/* If this is a scalar floating point value and we don't have D-form
addressing, prefer the traditional floating point registers so that we
can use D-form (register+offset) addressing. */
if (rclass == VSX_REGS
&& (mode == SFmode || GET_MODE_SIZE (mode) == 8))
return FLOAT_REGS;
/* Prefer the Altivec registers if Altivec is handling the vector
operations (i.e. V16QI, V8HI, and V4SI), or if we prefer Altivec
loads. */
if (VECTOR_UNIT_ALTIVEC_P (mode) || VECTOR_MEM_ALTIVEC_P (mode)
|| mode == V1TImode)
return ALTIVEC_REGS;
return rclass;
}
if (is_constant || GET_CODE (x) == PLUS)
{
if (reg_class_subset_p (GENERAL_REGS, rclass))
return GENERAL_REGS;
if (reg_class_subset_p (BASE_REGS, rclass))
return BASE_REGS;
return NO_REGS;
}
if (GET_MODE_CLASS (mode) == MODE_INT && rclass == NON_SPECIAL_REGS)
return GENERAL_REGS;
return rclass;
}
/* Debug version of rs6000_preferred_reload_class. */
static enum reg_class
rs6000_debug_preferred_reload_class (rtx x, enum reg_class rclass)
{
enum reg_class ret = rs6000_preferred_reload_class (x, rclass);
fprintf (stderr,
"\nrs6000_preferred_reload_class, return %s, rclass = %s, "
"mode = %s, x:\n",
reg_class_names[ret], reg_class_names[rclass],
GET_MODE_NAME (GET_MODE (x)));
debug_rtx (x);
return ret;
}
/* If we are copying between FP or AltiVec registers and anything else, we need
a memory location. The exception is when we are targeting ppc64 and the
move to/from fpr to gpr instructions are available. Also, under VSX, you
can copy vector registers from the FP register set to the Altivec register
set and vice versa. */
static bool
rs6000_secondary_memory_needed (machine_mode mode,
reg_class_t from_class,
reg_class_t to_class)
{
enum rs6000_reg_type from_type, to_type;
bool altivec_p = ((from_class == ALTIVEC_REGS)
|| (to_class == ALTIVEC_REGS));
/* If a simple/direct move is available, we don't need secondary memory */
from_type = reg_class_to_reg_type[(int)from_class];
to_type = reg_class_to_reg_type[(int)to_class];
if (rs6000_secondary_reload_move (to_type, from_type, mode,
(secondary_reload_info *)0, altivec_p))
return false;
/* If we have a floating point or vector register class, we need to use
memory to transfer the data. */
if (IS_FP_VECT_REG_TYPE (from_type) || IS_FP_VECT_REG_TYPE (to_type))
return true;
return false;
}
/* Debug version of rs6000_secondary_memory_needed. */
static bool
rs6000_debug_secondary_memory_needed (machine_mode mode,
reg_class_t from_class,
reg_class_t to_class)
{
bool ret = rs6000_secondary_memory_needed (mode, from_class, to_class);
fprintf (stderr,
"rs6000_secondary_memory_needed, return: %s, from_class = %s, "
"to_class = %s, mode = %s\n",
ret ? "true" : "false",
reg_class_names[from_class],
reg_class_names[to_class],
GET_MODE_NAME (mode));
return ret;
}
/* Return the register class of a scratch register needed to copy IN into
or out of a register in RCLASS in MODE. If it can be done directly,
NO_REGS is returned. */
static enum reg_class
rs6000_secondary_reload_class (enum reg_class rclass, machine_mode mode,
rtx in)
{
int regno;
if (TARGET_ELF || (DEFAULT_ABI == ABI_DARWIN
#if TARGET_MACHO
&& MACHOPIC_INDIRECT
#endif
))
{
/* We cannot copy a symbolic operand directly into anything
other than BASE_REGS for TARGET_ELF. So indicate that a
register from BASE_REGS is needed as an intermediate
register.
On Darwin, pic addresses require a load from memory, which
needs a base register. */
if (rclass != BASE_REGS
&& (GET_CODE (in) == SYMBOL_REF
|| GET_CODE (in) == HIGH
|| GET_CODE (in) == LABEL_REF
|| GET_CODE (in) == CONST))
return BASE_REGS;
}
if (GET_CODE (in) == REG)
{
regno = REGNO (in);
if (regno >= FIRST_PSEUDO_REGISTER)
{
regno = true_regnum (in);
if (regno >= FIRST_PSEUDO_REGISTER)
regno = -1;
}
}
else if (GET_CODE (in) == SUBREG)
{
regno = true_regnum (in);
if (regno >= FIRST_PSEUDO_REGISTER)
regno = -1;
}
else
regno = -1;
/* If we have VSX register moves, prefer moving scalar values between
Altivec registers and GPR by going via an FPR (and then via memory)
instead of reloading the secondary memory address for Altivec moves. */
if (TARGET_VSX
&& GET_MODE_SIZE (mode) < 16
&& !mode_supports_vmx_dform (mode)
&& (((rclass == GENERAL_REGS || rclass == BASE_REGS)
&& (regno >= 0 && ALTIVEC_REGNO_P (regno)))
|| ((rclass == VSX_REGS || rclass == ALTIVEC_REGS)
&& (regno >= 0 && INT_REGNO_P (regno)))))
return FLOAT_REGS;
/* We can place anything into GENERAL_REGS and can put GENERAL_REGS
into anything. */
if (rclass == GENERAL_REGS || rclass == BASE_REGS
|| (regno >= 0 && INT_REGNO_P (regno)))
return NO_REGS;
/* Constants, memory, and VSX registers can go into VSX registers (both the
traditional floating point and the altivec registers). */
if (rclass == VSX_REGS
&& (regno == -1 || VSX_REGNO_P (regno)))
return NO_REGS;
/* Constants, memory, and FP registers can go into FP registers. */
if ((regno == -1 || FP_REGNO_P (regno))
&& (rclass == FLOAT_REGS || rclass == NON_SPECIAL_REGS))
return (mode != SDmode || lra_in_progress) ? NO_REGS : GENERAL_REGS;
/* Memory, and AltiVec registers can go into AltiVec registers. */
if ((regno == -1 || ALTIVEC_REGNO_P (regno))
&& rclass == ALTIVEC_REGS)
return NO_REGS;
/* We can copy among the CR registers. */
if ((rclass == CR_REGS || rclass == CR0_REGS)
&& regno >= 0 && CR_REGNO_P (regno))
return NO_REGS;
/* Otherwise, we need GENERAL_REGS. */
return GENERAL_REGS;
}
/* Debug version of rs6000_secondary_reload_class. */
static enum reg_class
rs6000_debug_secondary_reload_class (enum reg_class rclass,
machine_mode mode, rtx in)
{
enum reg_class ret = rs6000_secondary_reload_class (rclass, mode, in);
fprintf (stderr,
"\nrs6000_secondary_reload_class, return %s, rclass = %s, "
"mode = %s, input rtx:\n",
reg_class_names[ret], reg_class_names[rclass],
GET_MODE_NAME (mode));
debug_rtx (in);
return ret;
}
/* Implement TARGET_CAN_CHANGE_MODE_CLASS. */
static bool
rs6000_can_change_mode_class (machine_mode from,
machine_mode to,
reg_class_t rclass)
{
unsigned from_size = GET_MODE_SIZE (from);
unsigned to_size = GET_MODE_SIZE (to);
if (from_size != to_size)
{
enum reg_class xclass = (TARGET_VSX) ? VSX_REGS : FLOAT_REGS;
if (reg_classes_intersect_p (xclass, rclass))
{
unsigned to_nregs = hard_regno_nregs (FIRST_FPR_REGNO, to);
unsigned from_nregs = hard_regno_nregs (FIRST_FPR_REGNO, from);
bool to_float128_vector_p = FLOAT128_VECTOR_P (to);
bool from_float128_vector_p = FLOAT128_VECTOR_P (from);
/* Don't allow 64-bit types to overlap with 128-bit types that take a
single register under VSX because the scalar part of the register
is in the upper 64-bits, and not the lower 64-bits. Types like
TFmode/TDmode that take 2 scalar register can overlap. 128-bit
IEEE floating point can't overlap, and neither can small
values. */
if (to_float128_vector_p && from_float128_vector_p)
return true;
else if (to_float128_vector_p || from_float128_vector_p)
return false;
/* TDmode in floating-mode registers must always go into a register
pair with the most significant word in the even-numbered register
to match ISA requirements. In little-endian mode, this does not
match subreg numbering, so we cannot allow subregs. */
if (!BYTES_BIG_ENDIAN && (to == TDmode || from == TDmode))
return false;
if (from_size < 8 || to_size < 8)
return false;
if (from_size == 8 && (8 * to_nregs) != to_size)
return false;
if (to_size == 8 && (8 * from_nregs) != from_size)
return false;
return true;
}
else
return true;
}
/* Since the VSX register set includes traditional floating point registers
and altivec registers, just check for the size being different instead of
trying to check whether the modes are vector modes. Otherwise it won't
allow say DF and DI to change classes. For types like TFmode and TDmode
that take 2 64-bit registers, rather than a single 128-bit register, don't
allow subregs of those types to other 128 bit types. */
if (TARGET_VSX && VSX_REG_CLASS_P (rclass))
{
unsigned num_regs = (from_size + 15) / 16;
if (hard_regno_nregs (FIRST_FPR_REGNO, to) > num_regs
|| hard_regno_nregs (FIRST_FPR_REGNO, from) > num_regs)
return false;
return (from_size == 8 || from_size == 16);
}
if (TARGET_ALTIVEC && rclass == ALTIVEC_REGS
&& (ALTIVEC_VECTOR_MODE (from) + ALTIVEC_VECTOR_MODE (to)) == 1)
return false;
return true;
}
/* Debug version of rs6000_can_change_mode_class. */
static bool
rs6000_debug_can_change_mode_class (machine_mode from,
machine_mode to,
reg_class_t rclass)
{
bool ret = rs6000_can_change_mode_class (from, to, rclass);
fprintf (stderr,
"rs6000_can_change_mode_class, return %s, from = %s, "
"to = %s, rclass = %s\n",
ret ? "true" : "false",
GET_MODE_NAME (from), GET_MODE_NAME (to),
reg_class_names[rclass]);
return ret;
}
/* Return a string to do a move operation of 128 bits of data. */
const char *
rs6000_output_move_128bit (rtx operands[])
{
rtx dest = operands[0];
rtx src = operands[1];
machine_mode mode = GET_MODE (dest);
int dest_regno;
int src_regno;
bool dest_gpr_p, dest_fp_p, dest_vmx_p, dest_vsx_p;
bool src_gpr_p, src_fp_p, src_vmx_p, src_vsx_p;
if (REG_P (dest))
{
dest_regno = REGNO (dest);
dest_gpr_p = INT_REGNO_P (dest_regno);
dest_fp_p = FP_REGNO_P (dest_regno);
dest_vmx_p = ALTIVEC_REGNO_P (dest_regno);
dest_vsx_p = dest_fp_p | dest_vmx_p;
}
else
{
dest_regno = -1;
dest_gpr_p = dest_fp_p = dest_vmx_p = dest_vsx_p = false;
}
if (REG_P (src))
{
src_regno = REGNO (src);
src_gpr_p = INT_REGNO_P (src_regno);
src_fp_p = FP_REGNO_P (src_regno);
src_vmx_p = ALTIVEC_REGNO_P (src_regno);
src_vsx_p = src_fp_p | src_vmx_p;
}
else
{
src_regno = -1;
src_gpr_p = src_fp_p = src_vmx_p = src_vsx_p = false;
}
/* Register moves. */
if (dest_regno >= 0 && src_regno >= 0)
{
if (dest_gpr_p)
{
if (src_gpr_p)
return "#";
if (TARGET_DIRECT_MOVE_128 && src_vsx_p)
return (WORDS_BIG_ENDIAN
? "mfvsrd %0,%x1\n\tmfvsrld %L0,%x1"
: "mfvsrd %L0,%x1\n\tmfvsrld %0,%x1");
else if (TARGET_VSX && TARGET_DIRECT_MOVE && src_vsx_p)
return "#";
}
else if (TARGET_VSX && dest_vsx_p)
{
if (src_vsx_p)
return "xxlor %x0,%x1,%x1";
else if (TARGET_DIRECT_MOVE_128 && src_gpr_p)
return (WORDS_BIG_ENDIAN
? "mtvsrdd %x0,%1,%L1"
: "mtvsrdd %x0,%L1,%1");
else if (TARGET_DIRECT_MOVE && src_gpr_p)
return "#";
}
else if (TARGET_ALTIVEC && dest_vmx_p && src_vmx_p)
return "vor %0,%1,%1";
else if (dest_fp_p && src_fp_p)
return "#";
}
/* Loads. */
else if (dest_regno >= 0 && MEM_P (src))
{
if (dest_gpr_p)
{
if (TARGET_QUAD_MEMORY && quad_load_store_p (dest, src))
return "lq %0,%1";
else
return "#";
}
else if (TARGET_ALTIVEC && dest_vmx_p
&& altivec_indexed_or_indirect_operand (src, mode))
return "lvx %0,%y1";
else if (TARGET_VSX && dest_vsx_p)
{
if (mode_supports_vsx_dform_quad (mode)
&& quad_address_p (XEXP (src, 0), mode, true))
return "lxv %x0,%1";
else if (TARGET_P9_VECTOR)
return "lxvx %x0,%y1";
else if (mode == V16QImode || mode == V8HImode || mode == V4SImode)
return "lxvw4x %x0,%y1";
else
return "lxvd2x %x0,%y1";
}
else if (TARGET_ALTIVEC && dest_vmx_p)
return "lvx %0,%y1";
else if (dest_fp_p)
return "#";
}
/* Stores. */
else if (src_regno >= 0 && MEM_P (dest))
{
if (src_gpr_p)
{
if (TARGET_QUAD_MEMORY && quad_load_store_p (dest, src))
return "stq %1,%0";
else
return "#";
}
else if (TARGET_ALTIVEC && src_vmx_p
&& altivec_indexed_or_indirect_operand (src, mode))
return "stvx %1,%y0";
else if (TARGET_VSX && src_vsx_p)
{
if (mode_supports_vsx_dform_quad (mode)
&& quad_address_p (XEXP (dest, 0), mode, true))
return "stxv %x1,%0";
else if (TARGET_P9_VECTOR)
return "stxvx %x1,%y0";
else if (mode == V16QImode || mode == V8HImode || mode == V4SImode)
return "stxvw4x %x1,%y0";
else
return "stxvd2x %x1,%y0";
}
else if (TARGET_ALTIVEC && src_vmx_p)
return "stvx %1,%y0";
else if (src_fp_p)
return "#";
}
/* Constants. */
else if (dest_regno >= 0
&& (GET_CODE (src) == CONST_INT
|| GET_CODE (src) == CONST_WIDE_INT
|| GET_CODE (src) == CONST_DOUBLE
|| GET_CODE (src) == CONST_VECTOR))
{
if (dest_gpr_p)
return "#";
else if ((dest_vmx_p && TARGET_ALTIVEC)
|| (dest_vsx_p && TARGET_VSX))
return output_vec_const_move (operands);
}
fatal_insn ("Bad 128-bit move", gen_rtx_SET (dest, src));
}
/* Validate a 128-bit move. */
bool
rs6000_move_128bit_ok_p (rtx operands[])
{
machine_mode mode = GET_MODE (operands[0]);
return (gpc_reg_operand (operands[0], mode)
|| gpc_reg_operand (operands[1], mode));
}
/* Return true if a 128-bit move needs to be split. */
bool
rs6000_split_128bit_ok_p (rtx operands[])
{
if (!reload_completed)
return false;
if (!gpr_or_gpr_p (operands[0], operands[1]))
return false;
if (quad_load_store_p (operands[0], operands[1]))
return false;
return true;
}
/* Given a comparison operation, return the bit number in CCR to test. We
know this is a valid comparison.
SCC_P is 1 if this is for an scc. That means that %D will have been
used instead of %C, so the bits will be in different places.
Return -1 if OP isn't a valid comparison for some reason. */
int
ccr_bit (rtx op, int scc_p)
{
enum rtx_code code = GET_CODE (op);
machine_mode cc_mode;
int cc_regnum;
int base_bit;
rtx reg;
if (!COMPARISON_P (op))
return -1;
reg = XEXP (op, 0);
gcc_assert (GET_CODE (reg) == REG && CR_REGNO_P (REGNO (reg)));
cc_mode = GET_MODE (reg);
cc_regnum = REGNO (reg);
base_bit = 4 * (cc_regnum - CR0_REGNO);
validate_condition_mode (code, cc_mode);
/* When generating a sCOND operation, only positive conditions are
allowed. */
gcc_assert (!scc_p
|| code == EQ || code == GT || code == LT || code == UNORDERED
|| code == GTU || code == LTU);
switch (code)
{
case NE:
return scc_p ? base_bit + 3 : base_bit + 2;
case EQ:
return base_bit + 2;
case GT: case GTU: case UNLE:
return base_bit + 1;
case LT: case LTU: case UNGE:
return base_bit;
case ORDERED: case UNORDERED:
return base_bit + 3;
case GE: case GEU:
/* If scc, we will have done a cror to put the bit in the
unordered position. So test that bit. For integer, this is ! LT
unless this is an scc insn. */
return scc_p ? base_bit + 3 : base_bit;
case LE: case LEU:
return scc_p ? base_bit + 3 : base_bit + 1;
default:
gcc_unreachable ();
}
}
/* Return the GOT register. */
rtx
rs6000_got_register (rtx value ATTRIBUTE_UNUSED)
{
/* The second flow pass currently (June 1999) can't update
regs_ever_live without disturbing other parts of the compiler, so
update it here to make the prolog/epilogue code happy. */
if (!can_create_pseudo_p ()
&& !df_regs_ever_live_p (RS6000_PIC_OFFSET_TABLE_REGNUM))
df_set_regs_ever_live (RS6000_PIC_OFFSET_TABLE_REGNUM, true);
crtl->uses_pic_offset_table = 1;
return pic_offset_table_rtx;
}
static rs6000_stack_t stack_info;
/* Function to init struct machine_function.
This will be called, via a pointer variable,
from push_function_context. */
static struct machine_function *
rs6000_init_machine_status (void)
{
stack_info.reload_completed = 0;
return ggc_cleared_alloc<machine_function> ();
}
#define INT_P(X) (GET_CODE (X) == CONST_INT && GET_MODE (X) == VOIDmode)
/* Write out a function code label. */
void
rs6000_output_function_entry (FILE *file, const char *fname)
{
if (fname[0] != '.')
{
switch (DEFAULT_ABI)
{
default:
gcc_unreachable ();
case ABI_AIX:
if (DOT_SYMBOLS)
putc ('.', file);
else
ASM_OUTPUT_INTERNAL_LABEL_PREFIX (file, "L.");
break;
case ABI_ELFv2:
case ABI_V4:
case ABI_DARWIN:
break;
}
}
RS6000_OUTPUT_BASENAME (file, fname);
}
/* Print an operand. Recognize special options, documented below. */
#if TARGET_ELF
#define SMALL_DATA_RELOC ((rs6000_sdata == SDATA_EABI) ? "sda21" : "sdarel")
#define SMALL_DATA_REG ((rs6000_sdata == SDATA_EABI) ? 0 : 13)
#else
#define SMALL_DATA_RELOC "sda21"
#define SMALL_DATA_REG 0
#endif
void
print_operand (FILE *file, rtx x, int code)
{
int i;
unsigned HOST_WIDE_INT uval;
switch (code)
{
/* %a is output_address. */
/* %c is output_addr_const if a CONSTANT_ADDRESS_P, otherwise
output_operand. */
case 'D':
/* Like 'J' but get to the GT bit only. */
gcc_assert (REG_P (x));
/* Bit 1 is GT bit. */
i = 4 * (REGNO (x) - CR0_REGNO) + 1;
/* Add one for shift count in rlinm for scc. */
fprintf (file, "%d", i + 1);
return;
case 'e':
/* If the low 16 bits are 0, but some other bit is set, write 's'. */
if (! INT_P (x))
{
output_operand_lossage ("invalid %%e value");
return;
}
uval = INTVAL (x);
if ((uval & 0xffff) == 0 && uval != 0)
putc ('s', file);
return;
case 'E':
/* X is a CR register. Print the number of the EQ bit of the CR */
if (GET_CODE (x) != REG || ! CR_REGNO_P (REGNO (x)))
output_operand_lossage ("invalid %%E value");
else
fprintf (file, "%d", 4 * (REGNO (x) - CR0_REGNO) + 2);
return;
case 'f':
/* X is a CR register. Print the shift count needed to move it
to the high-order four bits. */
if (GET_CODE (x) != REG || ! CR_REGNO_P (REGNO (x)))
output_operand_lossage ("invalid %%f value");
else
fprintf (file, "%d", 4 * (REGNO (x) - CR0_REGNO));
return;
case 'F':
/* Similar, but print the count for the rotate in the opposite
direction. */
if (GET_CODE (x) != REG || ! CR_REGNO_P (REGNO (x)))
output_operand_lossage ("invalid %%F value");
else
fprintf (file, "%d", 32 - 4 * (REGNO (x) - CR0_REGNO));
return;
case 'G':
/* X is a constant integer. If it is negative, print "m",
otherwise print "z". This is to make an aze or ame insn. */
if (GET_CODE (x) != CONST_INT)
output_operand_lossage ("invalid %%G value");
else if (INTVAL (x) >= 0)
putc ('z', file);
else
putc ('m', file);
return;
case 'h':
/* If constant, output low-order five bits. Otherwise, write
normally. */
if (INT_P (x))
fprintf (file, HOST_WIDE_INT_PRINT_DEC, INTVAL (x) & 31);
else
print_operand (file, x, 0);
return;
case 'H':
/* If constant, output low-order six bits. Otherwise, write
normally. */
if (INT_P (x))
fprintf (file, HOST_WIDE_INT_PRINT_DEC, INTVAL (x) & 63);
else
print_operand (file, x, 0);
return;
case 'I':
/* Print `i' if this is a constant, else nothing. */
if (INT_P (x))
putc ('i', file);
return;
case 'j':
/* Write the bit number in CCR for jump. */
i = ccr_bit (x, 0);
if (i == -1)
output_operand_lossage ("invalid %%j code");
else
fprintf (file, "%d", i);
return;
case 'J':
/* Similar, but add one for shift count in rlinm for scc and pass
scc flag to `ccr_bit'. */
i = ccr_bit (x, 1);
if (i == -1)
output_operand_lossage ("invalid %%J code");
else
/* If we want bit 31, write a shift count of zero, not 32. */
fprintf (file, "%d", i == 31 ? 0 : i + 1);
return;
case 'k':
/* X must be a constant. Write the 1's complement of the
constant. */
if (! INT_P (x))
output_operand_lossage ("invalid %%k value");
else
fprintf (file, HOST_WIDE_INT_PRINT_DEC, ~ INTVAL (x));
return;
case 'K':
/* X must be a symbolic constant on ELF. Write an
expression suitable for an 'addi' that adds in the low 16
bits of the MEM. */
if (GET_CODE (x) == CONST)
{
if (GET_CODE (XEXP (x, 0)) != PLUS
|| (GET_CODE (XEXP (XEXP (x, 0), 0)) != SYMBOL_REF
&& GET_CODE (XEXP (XEXP (x, 0), 0)) != LABEL_REF)
|| GET_CODE (XEXP (XEXP (x, 0), 1)) != CONST_INT)
output_operand_lossage ("invalid %%K value");
}
print_operand_address (file, x);
fputs ("@l", file);
return;
/* %l is output_asm_label. */
case 'L':
/* Write second word of DImode or DFmode reference. Works on register
or non-indexed memory only. */
if (REG_P (x))
fputs (reg_names[REGNO (x) + 1], file);
else if (MEM_P (x))
{
machine_mode mode = GET_MODE (x);
/* Handle possible auto-increment. Since it is pre-increment and
we have already done it, we can just use an offset of word. */
if (GET_CODE (XEXP (x, 0)) == PRE_INC
|| GET_CODE (XEXP (x, 0)) == PRE_DEC)
output_address (mode, plus_constant (Pmode, XEXP (XEXP (x, 0), 0),
UNITS_PER_WORD));
else if (GET_CODE (XEXP (x, 0)) == PRE_MODIFY)
output_address (mode, plus_constant (Pmode, XEXP (XEXP (x, 0), 0),
UNITS_PER_WORD));
else
output_address (mode, XEXP (adjust_address_nv (x, SImode,
UNITS_PER_WORD),
0));
if (small_data_operand (x, GET_MODE (x)))
fprintf (file, "@%s(%s)", SMALL_DATA_RELOC,
reg_names[SMALL_DATA_REG]);
}
return;
case 'N':
/* Write the number of elements in the vector times 4. */
if (GET_CODE (x) != PARALLEL)
output_operand_lossage ("invalid %%N value");
else
fprintf (file, "%d", XVECLEN (x, 0) * 4);
return;
case 'O':
/* Similar, but subtract 1 first. */
if (GET_CODE (x) != PARALLEL)
output_operand_lossage ("invalid %%O value");
else
fprintf (file, "%d", (XVECLEN (x, 0) - 1) * 4);
return;
case 'p':
/* X is a CONST_INT that is a power of two. Output the logarithm. */
if (! INT_P (x)
|| INTVAL (x) < 0
|| (i = exact_log2 (INTVAL (x))) < 0)
output_operand_lossage ("invalid %%p value");
else
fprintf (file, "%d", i);
return;
case 'P':
/* The operand must be an indirect memory reference. The result
is the register name. */
if (GET_CODE (x) != MEM || GET_CODE (XEXP (x, 0)) != REG
|| REGNO (XEXP (x, 0)) >= 32)
output_operand_lossage ("invalid %%P value");
else
fputs (reg_names[REGNO (XEXP (x, 0))], file);
return;
case 'q':
/* This outputs the logical code corresponding to a boolean
expression. The expression may have one or both operands
negated (if one, only the first one). For condition register
logical operations, it will also treat the negated
CR codes as NOTs, but not handle NOTs of them. */
{
const char *const *t = 0;
const char *s;
enum rtx_code code = GET_CODE (x);
static const char * const tbl[3][3] = {
{ "and", "andc", "nor" },
{ "or", "orc", "nand" },
{ "xor", "eqv", "xor" } };
if (code == AND)
t = tbl[0];
else if (code == IOR)
t = tbl[1];
else if (code == XOR)
t = tbl[2];
else
output_operand_lossage ("invalid %%q value");
if (GET_CODE (XEXP (x, 0)) != NOT)
s = t[0];
else
{
if (GET_CODE (XEXP (x, 1)) == NOT)
s = t[2];
else
s = t[1];
}
fputs (s, file);
}
return;
case 'Q':
if (! TARGET_MFCRF)
return;
fputc (',', file);
/* FALLTHRU */
case 'R':
/* X is a CR register. Print the mask for `mtcrf'. */
if (GET_CODE (x) != REG || ! CR_REGNO_P (REGNO (x)))
output_operand_lossage ("invalid %%R value");
else
fprintf (file, "%d", 128 >> (REGNO (x) - CR0_REGNO));
return;
case 's':
/* Low 5 bits of 32 - value */
if (! INT_P (x))
output_operand_lossage ("invalid %%s value");
else
fprintf (file, HOST_WIDE_INT_PRINT_DEC, (32 - INTVAL (x)) & 31);
return;
case 't':
/* Like 'J' but get to the OVERFLOW/UNORDERED bit. */
gcc_assert (REG_P (x) && GET_MODE (x) == CCmode);
/* Bit 3 is OV bit. */
i = 4 * (REGNO (x) - CR0_REGNO) + 3;
/* If we want bit 31, write a shift count of zero, not 32. */
fprintf (file, "%d", i == 31 ? 0 : i + 1);
return;
case 'T':
/* Print the symbolic name of a branch target register. */
if (GET_CODE (x) != REG || (REGNO (x) != LR_REGNO
&& REGNO (x) != CTR_REGNO))
output_operand_lossage ("invalid %%T value");
else if (REGNO (x) == LR_REGNO)
fputs ("lr", file);
else
fputs ("ctr", file);
return;
case 'u':
/* High-order or low-order 16 bits of constant, whichever is non-zero,
for use in unsigned operand. */
if (! INT_P (x))
{
output_operand_lossage ("invalid %%u value");
return;
}
uval = INTVAL (x);
if ((uval & 0xffff) == 0)
uval >>= 16;
fprintf (file, HOST_WIDE_INT_PRINT_HEX, uval & 0xffff);
return;
case 'v':
/* High-order 16 bits of constant for use in signed operand. */
if (! INT_P (x))
output_operand_lossage ("invalid %%v value");
else
fprintf (file, HOST_WIDE_INT_PRINT_HEX,
(INTVAL (x) >> 16) & 0xffff);
return;
case 'U':
/* Print `u' if this has an auto-increment or auto-decrement. */
if (MEM_P (x)
&& (GET_CODE (XEXP (x, 0)) == PRE_INC
|| GET_CODE (XEXP (x, 0)) == PRE_DEC
|| GET_CODE (XEXP (x, 0)) == PRE_MODIFY))
putc ('u', file);
return;
case 'V':
/* Print the trap code for this operand. */
switch (GET_CODE (x))
{
case EQ:
fputs ("eq", file); /* 4 */
break;
case NE:
fputs ("ne", file); /* 24 */
break;
case LT:
fputs ("lt", file); /* 16 */
break;
case LE:
fputs ("le", file); /* 20 */
break;
case GT:
fputs ("gt", file); /* 8 */
break;
case GE:
fputs ("ge", file); /* 12 */
break;
case LTU:
fputs ("llt", file); /* 2 */
break;
case LEU:
fputs ("lle", file); /* 6 */
break;
case GTU:
fputs ("lgt", file); /* 1 */
break;
case GEU:
fputs ("lge", file); /* 5 */
break;
default:
gcc_unreachable ();
}
break;
case 'w':
/* If constant, low-order 16 bits of constant, signed. Otherwise, write
normally. */
if (INT_P (x))
fprintf (file, HOST_WIDE_INT_PRINT_DEC,
((INTVAL (x) & 0xffff) ^ 0x8000) - 0x8000);
else
print_operand (file, x, 0);
return;
case 'x':
/* X is a FPR or Altivec register used in a VSX context. */
if (GET_CODE (x) != REG || !VSX_REGNO_P (REGNO (x)))
output_operand_lossage ("invalid %%x value");
else
{
int reg = REGNO (x);
int vsx_reg = (FP_REGNO_P (reg)
? reg - 32
: reg - FIRST_ALTIVEC_REGNO + 32);
#ifdef TARGET_REGNAMES
if (TARGET_REGNAMES)
fprintf (file, "%%vs%d", vsx_reg);
else
#endif
fprintf (file, "%d", vsx_reg);
}
return;
case 'X':
if (MEM_P (x)
&& (legitimate_indexed_address_p (XEXP (x, 0), 0)
|| (GET_CODE (XEXP (x, 0)) == PRE_MODIFY
&& legitimate_indexed_address_p (XEXP (XEXP (x, 0), 1), 0))))
putc ('x', file);
return;
case 'Y':
/* Like 'L', for third word of TImode/PTImode */
if (REG_P (x))
fputs (reg_names[REGNO (x) + 2], file);
else if (MEM_P (x))
{
machine_mode mode = GET_MODE (x);
if (GET_CODE (XEXP (x, 0)) == PRE_INC
|| GET_CODE (XEXP (x, 0)) == PRE_DEC)
output_address (mode, plus_constant (Pmode,
XEXP (XEXP (x, 0), 0), 8));
else if (GET_CODE (XEXP (x, 0)) == PRE_MODIFY)
output_address (mode, plus_constant (Pmode,
XEXP (XEXP (x, 0), 0), 8));
else
output_address (mode, XEXP (adjust_address_nv (x, SImode, 8), 0));
if (small_data_operand (x, GET_MODE (x)))
fprintf (file, "@%s(%s)", SMALL_DATA_RELOC,
reg_names[SMALL_DATA_REG]);
}
return;
case 'z':
/* X is a SYMBOL_REF. Write out the name preceded by a
period and without any trailing data in brackets. Used for function
names. If we are configured for System V (or the embedded ABI) on
the PowerPC, do not emit the period, since those systems do not use
TOCs and the like. */
gcc_assert (GET_CODE (x) == SYMBOL_REF);
/* For macho, check to see if we need a stub. */
if (TARGET_MACHO)
{
const char *name = XSTR (x, 0);
#if TARGET_MACHO
if (darwin_emit_branch_islands
&& MACHOPIC_INDIRECT
&& machopic_classify_symbol (x) == MACHOPIC_UNDEFINED_FUNCTION)
name = machopic_indirection_name (x, /*stub_p=*/true);
#endif
assemble_name (file, name);
}
else if (!DOT_SYMBOLS)
assemble_name (file, XSTR (x, 0));
else
rs6000_output_function_entry (file, XSTR (x, 0));
return;
case 'Z':
/* Like 'L', for last word of TImode/PTImode. */
if (REG_P (x))
fputs (reg_names[REGNO (x) + 3], file);
else if (MEM_P (x))
{
machine_mode mode = GET_MODE (x);
if (GET_CODE (XEXP (x, 0)) == PRE_INC
|| GET_CODE (XEXP (x, 0)) == PRE_DEC)
output_address (mode, plus_constant (Pmode,
XEXP (XEXP (x, 0), 0), 12));
else if (GET_CODE (XEXP (x, 0)) == PRE_MODIFY)
output_address (mode, plus_constant (Pmode,
XEXP (XEXP (x, 0), 0), 12));
else
output_address (mode, XEXP (adjust_address_nv (x, SImode, 12), 0));
if (small_data_operand (x, GET_MODE (x)))
fprintf (file, "@%s(%s)", SMALL_DATA_RELOC,
reg_names[SMALL_DATA_REG]);
}
return;
/* Print AltiVec memory operand. */
case 'y':
{
rtx tmp;
gcc_assert (MEM_P (x));
tmp = XEXP (x, 0);
if (VECTOR_MEM_ALTIVEC_P (GET_MODE (x))
&& GET_CODE (tmp) == AND
&& GET_CODE (XEXP (tmp, 1)) == CONST_INT
&& INTVAL (XEXP (tmp, 1)) == -16)
tmp = XEXP (tmp, 0);
else if (VECTOR_MEM_VSX_P (GET_MODE (x))
&& GET_CODE (tmp) == PRE_MODIFY)
tmp = XEXP (tmp, 1);
if (REG_P (tmp))
fprintf (file, "0,%s", reg_names[REGNO (tmp)]);
else
{
if (GET_CODE (tmp) != PLUS
|| !REG_P (XEXP (tmp, 0))
|| !REG_P (XEXP (tmp, 1)))
{
output_operand_lossage ("invalid %%y value, try using the 'Z' constraint");
break;
}
if (REGNO (XEXP (tmp, 0)) == 0)
fprintf (file, "%s,%s", reg_names[ REGNO (XEXP (tmp, 1)) ],
reg_names[ REGNO (XEXP (tmp, 0)) ]);
else
fprintf (file, "%s,%s", reg_names[ REGNO (XEXP (tmp, 0)) ],
reg_names[ REGNO (XEXP (tmp, 1)) ]);
}
break;
}
case 0:
if (REG_P (x))
fprintf (file, "%s", reg_names[REGNO (x)]);
else if (MEM_P (x))
{
/* We need to handle PRE_INC and PRE_DEC here, since we need to
know the width from the mode. */
if (GET_CODE (XEXP (x, 0)) == PRE_INC)
fprintf (file, "%d(%s)", GET_MODE_SIZE (GET_MODE (x)),
reg_names[REGNO (XEXP (XEXP (x, 0), 0))]);
else if (GET_CODE (XEXP (x, 0)) == PRE_DEC)
fprintf (file, "%d(%s)", - GET_MODE_SIZE (GET_MODE (x)),
reg_names[REGNO (XEXP (XEXP (x, 0), 0))]);
else if (GET_CODE (XEXP (x, 0)) == PRE_MODIFY)
output_address (GET_MODE (x), XEXP (XEXP (x, 0), 1));
else
output_address (GET_MODE (x), XEXP (x, 0));
}
else
{
if (toc_relative_expr_p (x, false, &tocrel_base_oac, &tocrel_offset_oac))
/* This hack along with a corresponding hack in
rs6000_output_addr_const_extra arranges to output addends
where the assembler expects to find them. eg.
(plus (unspec [(symbol_ref ("x")) (reg 2)] tocrel) 4)
without this hack would be output as "x@toc+4". We
want "x+4@toc". */
output_addr_const (file, CONST_CAST_RTX (tocrel_base_oac));
else
output_addr_const (file, x);
}
return;
case '&':
if (const char *name = get_some_local_dynamic_name ())
assemble_name (file, name);
else
output_operand_lossage ("'%%&' used without any "
"local dynamic TLS references");
return;
default:
output_operand_lossage ("invalid %%xn code");
}
}
/* Print the address of an operand. */
void
print_operand_address (FILE *file, rtx x)
{
if (REG_P (x))
fprintf (file, "0(%s)", reg_names[ REGNO (x) ]);
else if (GET_CODE (x) == SYMBOL_REF || GET_CODE (x) == CONST
|| GET_CODE (x) == LABEL_REF)
{
output_addr_const (file, x);
if (small_data_operand (x, GET_MODE (x)))
fprintf (file, "@%s(%s)", SMALL_DATA_RELOC,
reg_names[SMALL_DATA_REG]);
else
gcc_assert (!TARGET_TOC);
}
else if (GET_CODE (x) == PLUS && REG_P (XEXP (x, 0))
&& REG_P (XEXP (x, 1)))
{
if (REGNO (XEXP (x, 0)) == 0)
fprintf (file, "%s,%s", reg_names[ REGNO (XEXP (x, 1)) ],
reg_names[ REGNO (XEXP (x, 0)) ]);
else
fprintf (file, "%s,%s", reg_names[ REGNO (XEXP (x, 0)) ],
reg_names[ REGNO (XEXP (x, 1)) ]);
}
else if (GET_CODE (x) == PLUS && REG_P (XEXP (x, 0))
&& GET_CODE (XEXP (x, 1)) == CONST_INT)
fprintf (file, HOST_WIDE_INT_PRINT_DEC "(%s)",
INTVAL (XEXP (x, 1)), reg_names[ REGNO (XEXP (x, 0)) ]);
#if TARGET_MACHO
else if (GET_CODE (x) == LO_SUM && REG_P (XEXP (x, 0))
&& CONSTANT_P (XEXP (x, 1)))
{
fprintf (file, "lo16(");
output_addr_const (file, XEXP (x, 1));
fprintf (file, ")(%s)", reg_names[ REGNO (XEXP (x, 0)) ]);
}
#endif
#if TARGET_ELF
else if (GET_CODE (x) == LO_SUM && REG_P (XEXP (x, 0))
&& CONSTANT_P (XEXP (x, 1)))
{
output_addr_const (file, XEXP (x, 1));
fprintf (file, "@l(%s)", reg_names[ REGNO (XEXP (x, 0)) ]);
}
#endif
else if (toc_relative_expr_p (x, false, &tocrel_base_oac, &tocrel_offset_oac))
{
/* This hack along with a corresponding hack in
rs6000_output_addr_const_extra arranges to output addends
where the assembler expects to find them. eg.
(lo_sum (reg 9)
. (plus (unspec [(symbol_ref ("x")) (reg 2)] tocrel) 8))
without this hack would be output as "x@toc+8@l(9)". We
want "x+8@toc@l(9)". */
output_addr_const (file, CONST_CAST_RTX (tocrel_base_oac));
if (GET_CODE (x) == LO_SUM)
fprintf (file, "@l(%s)", reg_names[REGNO (XEXP (x, 0))]);
else
fprintf (file, "(%s)", reg_names[REGNO (XVECEXP (tocrel_base_oac, 0, 1))]);
}
else
gcc_unreachable ();
}
/* Implement TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA. */
static bool
rs6000_output_addr_const_extra (FILE *file, rtx x)
{
if (GET_CODE (x) == UNSPEC)
switch (XINT (x, 1))
{
case UNSPEC_TOCREL:
gcc_checking_assert (GET_CODE (XVECEXP (x, 0, 0)) == SYMBOL_REF
&& REG_P (XVECEXP (x, 0, 1))
&& REGNO (XVECEXP (x, 0, 1)) == TOC_REGISTER);
output_addr_const (file, XVECEXP (x, 0, 0));
if (x == tocrel_base_oac && tocrel_offset_oac != const0_rtx)
{
if (INTVAL (tocrel_offset_oac) >= 0)
fprintf (file, "+");
output_addr_const (file, CONST_CAST_RTX (tocrel_offset_oac));
}
if (!TARGET_AIX || (TARGET_ELF && TARGET_MINIMAL_TOC))
{
putc ('-', file);
assemble_name (file, toc_label_name);
need_toc_init = 1;
}
else if (TARGET_ELF)
fputs ("@toc", file);
return true;
#if TARGET_MACHO
case UNSPEC_MACHOPIC_OFFSET:
output_addr_const (file, XVECEXP (x, 0, 0));
putc ('-', file);
machopic_output_function_base_name (file);
return true;
#endif
}
return false;
}
/* Target hook for assembling integer objects. The PowerPC version has
to handle fixup entries for relocatable code if RELOCATABLE_NEEDS_FIXUP
is defined. It also needs to handle DI-mode objects on 64-bit
targets. */
static bool
rs6000_assemble_integer (rtx x, unsigned int size, int aligned_p)
{
#ifdef RELOCATABLE_NEEDS_FIXUP
/* Special handling for SI values. */
if (RELOCATABLE_NEEDS_FIXUP && size == 4 && aligned_p)
{
static int recurse = 0;
/* For -mrelocatable, we mark all addresses that need to be fixed up in
the .fixup section. Since the TOC section is already relocated, we
don't need to mark it here. We used to skip the text section, but it
should never be valid for relocated addresses to be placed in the text
section. */
if (DEFAULT_ABI == ABI_V4
&& (TARGET_RELOCATABLE || flag_pic > 1)
&& in_section != toc_section
&& !recurse
&& !CONST_SCALAR_INT_P (x)
&& CONSTANT_P (x))
{
char buf[256];
recurse = 1;
ASM_GENERATE_INTERNAL_LABEL (buf, "LCP", fixuplabelno);
fixuplabelno++;
ASM_OUTPUT_LABEL (asm_out_file, buf);
fprintf (asm_out_file, "\t.long\t(");
output_addr_const (asm_out_file, x);
fprintf (asm_out_file, ")@fixup\n");
fprintf (asm_out_file, "\t.section\t\".fixup\",\"aw\"\n");
ASM_OUTPUT_ALIGN (asm_out_file, 2);
fprintf (asm_out_file, "\t.long\t");
assemble_name (asm_out_file, buf);
fprintf (asm_out_file, "\n\t.previous\n");
recurse = 0;
return true;
}
/* Remove initial .'s to turn a -mcall-aixdesc function
address into the address of the descriptor, not the function
itself. */
else if (GET_CODE (x) == SYMBOL_REF
&& XSTR (x, 0)[0] == '.'
&& DEFAULT_ABI == ABI_AIX)
{
const char *name = XSTR (x, 0);
while (*name == '.')
name++;
fprintf (asm_out_file, "\t.long\t%s\n", name);
return true;
}
}
#endif /* RELOCATABLE_NEEDS_FIXUP */
return default_assemble_integer (x, size, aligned_p);
}
#if defined (HAVE_GAS_HIDDEN) && !TARGET_MACHO
/* Emit an assembler directive to set symbol visibility for DECL to
VISIBILITY_TYPE. */
static void
rs6000_assemble_visibility (tree decl, int vis)
{
if (TARGET_XCOFF)
return;
/* Functions need to have their entry point symbol visibility set as
well as their descriptor symbol visibility. */
if (DEFAULT_ABI == ABI_AIX
&& DOT_SYMBOLS
&& TREE_CODE (decl) == FUNCTION_DECL)
{
static const char * const visibility_types[] = {
NULL, "protected", "hidden", "internal"
};
const char *name, *type;
name = ((* targetm.strip_name_encoding)
(IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (decl))));
type = visibility_types[vis];
fprintf (asm_out_file, "\t.%s\t%s\n", type, name);
fprintf (asm_out_file, "\t.%s\t.%s\n", type, name);
}
else
default_assemble_visibility (decl, vis);
}
#endif
enum rtx_code
rs6000_reverse_condition (machine_mode mode, enum rtx_code code)
{
/* Reversal of FP compares takes care -- an ordered compare
becomes an unordered compare and vice versa. */
if (mode == CCFPmode
&& (!flag_finite_math_only
|| code == UNLT || code == UNLE || code == UNGT || code == UNGE
|| code == UNEQ || code == LTGT))
return reverse_condition_maybe_unordered (code);
else
return reverse_condition (code);
}
/* Generate a compare for CODE. Return a brand-new rtx that
represents the result of the compare. */
static rtx
rs6000_generate_compare (rtx cmp, machine_mode mode)
{
machine_mode comp_mode;
rtx compare_result;
enum rtx_code code = GET_CODE (cmp);
rtx op0 = XEXP (cmp, 0);
rtx op1 = XEXP (cmp, 1);
if (!TARGET_FLOAT128_HW && FLOAT128_VECTOR_P (mode))
comp_mode = CCmode;
else if (FLOAT_MODE_P (mode))
comp_mode = CCFPmode;
else if (code == GTU || code == LTU
|| code == GEU || code == LEU)
comp_mode = CCUNSmode;
else if ((code == EQ || code == NE)
&& unsigned_reg_p (op0)
&& (unsigned_reg_p (op1)
|| (CONST_INT_P (op1) && INTVAL (op1) != 0)))
/* These are unsigned values, perhaps there will be a later
ordering compare that can be shared with this one. */
comp_mode = CCUNSmode;
else
comp_mode = CCmode;
/* If we have an unsigned compare, make sure we don't have a signed value as
an immediate. */
if (comp_mode == CCUNSmode && GET_CODE (op1) == CONST_INT
&& INTVAL (op1) < 0)
{
op0 = copy_rtx_if_shared (op0);
op1 = force_reg (GET_MODE (op0), op1);
cmp = gen_rtx_fmt_ee (code, GET_MODE (cmp), op0, op1);
}
/* First, the compare. */
compare_result = gen_reg_rtx (comp_mode);
/* IEEE 128-bit support in VSX registers when we do not have hardware
support. */
if (!TARGET_FLOAT128_HW && FLOAT128_VECTOR_P (mode))
{
rtx libfunc = NULL_RTX;
bool check_nan = false;
rtx dest;
switch (code)
{
case EQ:
case NE:
libfunc = optab_libfunc (eq_optab, mode);
break;
case GT:
case GE:
libfunc = optab_libfunc (ge_optab, mode);
break;
case LT:
case LE:
libfunc = optab_libfunc (le_optab, mode);
break;
case UNORDERED:
case ORDERED:
libfunc = optab_libfunc (unord_optab, mode);
code = (code == UNORDERED) ? NE : EQ;
break;
case UNGE:
case UNGT:
check_nan = true;
libfunc = optab_libfunc (ge_optab, mode);
code = (code == UNGE) ? GE : GT;
break;
case UNLE:
case UNLT:
check_nan = true;
libfunc = optab_libfunc (le_optab, mode);
code = (code == UNLE) ? LE : LT;
break;
case UNEQ:
case LTGT:
check_nan = true;
libfunc = optab_libfunc (eq_optab, mode);
code = (code = UNEQ) ? EQ : NE;
break;
default:
gcc_unreachable ();
}
gcc_assert (libfunc);
if (!check_nan)
dest = emit_library_call_value (libfunc, NULL_RTX, LCT_CONST,
SImode, op0, mode, op1, mode);
/* The library signals an exception for signalling NaNs, so we need to
handle isgreater, etc. by first checking isordered. */
else
{
rtx ne_rtx, normal_dest, unord_dest;
rtx unord_func = optab_libfunc (unord_optab, mode);
rtx join_label = gen_label_rtx ();
rtx join_ref = gen_rtx_LABEL_REF (VOIDmode, join_label);
rtx unord_cmp = gen_reg_rtx (comp_mode);
/* Test for either value being a NaN. */
gcc_assert (unord_func);
unord_dest = emit_library_call_value (unord_func, NULL_RTX, LCT_CONST,
SImode, op0, mode, op1, mode);
/* Set value (0) if either value is a NaN, and jump to the join
label. */
dest = gen_reg_rtx (SImode);
emit_move_insn (dest, const1_rtx);
emit_insn (gen_rtx_SET (unord_cmp,
gen_rtx_COMPARE (comp_mode, unord_dest,
const0_rtx)));
ne_rtx = gen_rtx_NE (comp_mode, unord_cmp, const0_rtx);
emit_jump_insn (gen_rtx_SET (pc_rtx,
gen_rtx_IF_THEN_ELSE (VOIDmode, ne_rtx,
join_ref,
pc_rtx)));
/* Do the normal comparison, knowing that the values are not
NaNs. */
normal_dest = emit_library_call_value (libfunc, NULL_RTX, LCT_CONST,
SImode, op0, mode, op1, mode);
emit_insn (gen_cstoresi4 (dest,
gen_rtx_fmt_ee (code, SImode, normal_dest,
const0_rtx),
normal_dest, const0_rtx));
/* Join NaN and non-Nan paths. Compare dest against 0. */
emit_label (join_label);
code = NE;
}
emit_insn (gen_rtx_SET (compare_result,
gen_rtx_COMPARE (comp_mode, dest, const0_rtx)));
}
else
{
/* Generate XLC-compatible TFmode compare as PARALLEL with extra
CLOBBERs to match cmptf_internal2 pattern. */
if (comp_mode == CCFPmode && TARGET_XL_COMPAT
&& FLOAT128_IBM_P (GET_MODE (op0))
&& TARGET_HARD_FLOAT)
emit_insn (gen_rtx_PARALLEL (VOIDmode,
gen_rtvec (10,
gen_rtx_SET (compare_result,
gen_rtx_COMPARE (comp_mode, op0, op1)),
gen_rtx_CLOBBER (VOIDmode, gen_rtx_SCRATCH (DFmode)),
gen_rtx_CLOBBER (VOIDmode, gen_rtx_SCRATCH (DFmode)),
gen_rtx_CLOBBER (VOIDmode, gen_rtx_SCRATCH (DFmode)),
gen_rtx_CLOBBER (VOIDmode, gen_rtx_SCRATCH (DFmode)),
gen_rtx_CLOBBER (VOIDmode, gen_rtx_SCRATCH (DFmode)),
gen_rtx_CLOBBER (VOIDmode, gen_rtx_SCRATCH (DFmode)),
gen_rtx_CLOBBER (VOIDmode, gen_rtx_SCRATCH (DFmode)),
gen_rtx_CLOBBER (VOIDmode, gen_rtx_SCRATCH (DFmode)),
gen_rtx_CLOBBER (VOIDmode, gen_rtx_SCRATCH (Pmode)))));
else if (GET_CODE (op1) == UNSPEC
&& XINT (op1, 1) == UNSPEC_SP_TEST)
{
rtx op1b = XVECEXP (op1, 0, 0);
comp_mode = CCEQmode;
compare_result = gen_reg_rtx (CCEQmode);
if (TARGET_64BIT)
emit_insn (gen_stack_protect_testdi (compare_result, op0, op1b));
else
emit_insn (gen_stack_protect_testsi (compare_result, op0, op1b));
}
else
emit_insn (gen_rtx_SET (compare_result,
gen_rtx_COMPARE (comp_mode, op0, op1)));
}
/* Some kinds of FP comparisons need an OR operation;
under flag_finite_math_only we don't bother. */
if (FLOAT_MODE_P (mode)
&& (!FLOAT128_IEEE_P (mode) || TARGET_FLOAT128_HW)
&& !flag_finite_math_only
&& (code == LE || code == GE
|| code == UNEQ || code == LTGT
|| code == UNGT || code == UNLT))
{
enum rtx_code or1, or2;
rtx or1_rtx, or2_rtx, compare2_rtx;
rtx or_result = gen_reg_rtx (CCEQmode);
switch (code)
{
case LE: or1 = LT; or2 = EQ; break;
case GE: or1 = GT; or2 = EQ; break;
case UNEQ: or1 = UNORDERED; or2 = EQ; break;
case LTGT: or1 = LT; or2 = GT; break;
case UNGT: or1 = UNORDERED; or2 = GT; break;
case UNLT: or1 = UNORDERED; or2 = LT; break;
default: gcc_unreachable ();
}
validate_condition_mode (or1, comp_mode);
validate_condition_mode (or2, comp_mode);
or1_rtx = gen_rtx_fmt_ee (or1, SImode, compare_result, const0_rtx);
or2_rtx = gen_rtx_fmt_ee (or2, SImode, compare_result, const0_rtx);
compare2_rtx = gen_rtx_COMPARE (CCEQmode,
gen_rtx_IOR (SImode, or1_rtx, or2_rtx),
const_true_rtx);
emit_insn (gen_rtx_SET (or_result, compare2_rtx));
compare_result = or_result;
code = EQ;
}
validate_condition_mode (code, GET_MODE (compare_result));
return gen_rtx_fmt_ee (code, VOIDmode, compare_result, const0_rtx);
}
/* Return the diagnostic message string if the binary operation OP is
not permitted on TYPE1 and TYPE2, NULL otherwise. */
static const char*
rs6000_invalid_binary_op (int op ATTRIBUTE_UNUSED,
const_tree type1,
const_tree type2)
{
machine_mode mode1 = TYPE_MODE (type1);
machine_mode mode2 = TYPE_MODE (type2);
/* For complex modes, use the inner type. */
if (COMPLEX_MODE_P (mode1))
mode1 = GET_MODE_INNER (mode1);
if (COMPLEX_MODE_P (mode2))
mode2 = GET_MODE_INNER (mode2);
/* Don't allow IEEE 754R 128-bit binary floating point and IBM extended
double to intermix unless -mfloat128-convert. */
if (mode1 == mode2)
return NULL;
if (!TARGET_FLOAT128_CVT)
{
if ((mode1 == KFmode && mode2 == IFmode)
|| (mode1 == IFmode && mode2 == KFmode))
return N_("__float128 and __ibm128 cannot be used in the same "
"expression");
if (TARGET_IEEEQUAD
&& ((mode1 == IFmode && mode2 == TFmode)
|| (mode1 == TFmode && mode2 == IFmode)))
return N_("__ibm128 and long double cannot be used in the same "
"expression");
if (!TARGET_IEEEQUAD
&& ((mode1 == KFmode && mode2 == TFmode)
|| (mode1 == TFmode && mode2 == KFmode)))
return N_("__float128 and long double cannot be used in the same "
"expression");
}
return NULL;
}
/* Expand floating point conversion to/from __float128 and __ibm128. */
void
rs6000_expand_float128_convert (rtx dest, rtx src, bool unsigned_p)
{
machine_mode dest_mode = GET_MODE (dest);
machine_mode src_mode = GET_MODE (src);
convert_optab cvt = unknown_optab;
bool do_move = false;
rtx libfunc = NULL_RTX;
rtx dest2;
typedef rtx (*rtx_2func_t) (rtx, rtx);
rtx_2func_t hw_convert = (rtx_2func_t)0;
size_t kf_or_tf;
struct hw_conv_t {
rtx_2func_t from_df;
rtx_2func_t from_sf;
rtx_2func_t from_si_sign;
rtx_2func_t from_si_uns;
rtx_2func_t from_di_sign;
rtx_2func_t from_di_uns;
rtx_2func_t to_df;
rtx_2func_t to_sf;
rtx_2func_t to_si_sign;
rtx_2func_t to_si_uns;
rtx_2func_t to_di_sign;
rtx_2func_t to_di_uns;
} hw_conversions[2] = {
/* convertions to/from KFmode */
{
gen_extenddfkf2_hw, /* KFmode <- DFmode. */
gen_extendsfkf2_hw, /* KFmode <- SFmode. */
gen_float_kfsi2_hw, /* KFmode <- SImode (signed). */
gen_floatuns_kfsi2_hw, /* KFmode <- SImode (unsigned). */
gen_float_kfdi2_hw, /* KFmode <- DImode (signed). */
gen_floatuns_kfdi2_hw, /* KFmode <- DImode (unsigned). */
gen_trunckfdf2_hw, /* DFmode <- KFmode. */
gen_trunckfsf2_hw, /* SFmode <- KFmode. */
gen_fix_kfsi2_hw, /* SImode <- KFmode (signed). */
gen_fixuns_kfsi2_hw, /* SImode <- KFmode (unsigned). */
gen_fix_kfdi2_hw, /* DImode <- KFmode (signed). */
gen_fixuns_kfdi2_hw, /* DImode <- KFmode (unsigned). */
},
/* convertions to/from TFmode */
{
gen_extenddftf2_hw, /* TFmode <- DFmode. */
gen_extendsftf2_hw, /* TFmode <- SFmode. */
gen_float_tfsi2_hw, /* TFmode <- SImode (signed). */
gen_floatuns_tfsi2_hw, /* TFmode <- SImode (unsigned). */
gen_float_tfdi2_hw, /* TFmode <- DImode (signed). */
gen_floatuns_tfdi2_hw, /* TFmode <- DImode (unsigned). */
gen_trunctfdf2_hw, /* DFmode <- TFmode. */
gen_trunctfsf2_hw, /* SFmode <- TFmode. */
gen_fix_tfsi2_hw, /* SImode <- TFmode (signed). */
gen_fixuns_tfsi2_hw, /* SImode <- TFmode (unsigned). */
gen_fix_tfdi2_hw, /* DImode <- TFmode (signed). */
gen_fixuns_tfdi2_hw, /* DImode <- TFmode (unsigned). */
},
};
if (dest_mode == src_mode)
gcc_unreachable ();
/* Eliminate memory operations. */
if (MEM_P (src))
src = force_reg (src_mode, src);
if (MEM_P (dest))
{
rtx tmp = gen_reg_rtx (dest_mode);
rs6000_expand_float128_convert (tmp, src, unsigned_p);
rs6000_emit_move (dest, tmp, dest_mode);
return;
}
/* Convert to IEEE 128-bit floating point. */
if (FLOAT128_IEEE_P (dest_mode))
{
if (dest_mode == KFmode)
kf_or_tf = 0;
else if (dest_mode == TFmode)
kf_or_tf = 1;
else
gcc_unreachable ();
switch (src_mode)
{
case E_DFmode:
cvt = sext_optab;
hw_convert = hw_conversions[kf_or_tf].from_df;
break;
case E_SFmode:
cvt = sext_optab;
hw_convert = hw_conversions[kf_or_tf].from_sf;
break;
case E_KFmode:
case E_IFmode:
case E_TFmode:
if (FLOAT128_IBM_P (src_mode))
cvt = sext_optab;
else
do_move = true;
break;
case E_SImode:
if (unsigned_p)
{
cvt = ufloat_optab;
hw_convert = hw_conversions[kf_or_tf].from_si_uns;
}
else
{
cvt = sfloat_optab;
hw_convert = hw_conversions[kf_or_tf].from_si_sign;
}
break;
case E_DImode:
if (unsigned_p)
{
cvt = ufloat_optab;
hw_convert = hw_conversions[kf_or_tf].from_di_uns;
}
else
{
cvt = sfloat_optab;
hw_convert = hw_conversions[kf_or_tf].from_di_sign;
}
break;
default:
gcc_unreachable ();
}
}
/* Convert from IEEE 128-bit floating point. */
else if (FLOAT128_IEEE_P (src_mode))
{
if (src_mode == KFmode)
kf_or_tf = 0;
else if (src_mode == TFmode)
kf_or_tf = 1;
else
gcc_unreachable ();
switch (dest_mode)
{
case E_DFmode:
cvt = trunc_optab;
hw_convert = hw_conversions[kf_or_tf].to_df;
break;
case E_SFmode:
cvt = trunc_optab;
hw_convert = hw_conversions[kf_or_tf].to_sf;
break;
case E_KFmode:
case E_IFmode:
case E_TFmode:
if (FLOAT128_IBM_P (dest_mode))
cvt = trunc_optab;
else
do_move = true;
break;
case E_SImode:
if (unsigned_p)
{
cvt = ufix_optab;
hw_convert = hw_conversions[kf_or_tf].to_si_uns;
}
else
{
cvt = sfix_optab;
hw_convert = hw_conversions[kf_or_tf].to_si_sign;
}
break;
case E_DImode:
if (unsigned_p)
{
cvt = ufix_optab;
hw_convert = hw_conversions[kf_or_tf].to_di_uns;
}
else
{
cvt = sfix_optab;
hw_convert = hw_conversions[kf_or_tf].to_di_sign;
}
break;
default:
gcc_unreachable ();
}
}
/* Both IBM format. */
else if (FLOAT128_IBM_P (dest_mode) && FLOAT128_IBM_P (src_mode))
do_move = true;
else
gcc_unreachable ();
/* Handle conversion between TFmode/KFmode. */
if (do_move)
emit_move_insn (dest, gen_lowpart (dest_mode, src));
/* Handle conversion if we have hardware support. */
else if (TARGET_FLOAT128_HW && hw_convert)
emit_insn ((hw_convert) (dest, src));
/* Call an external function to do the conversion. */
else if (cvt != unknown_optab)
{
libfunc = convert_optab_libfunc (cvt, dest_mode, src_mode);
gcc_assert (libfunc != NULL_RTX);
dest2 = emit_library_call_value (libfunc, dest, LCT_CONST, dest_mode,
src, src_mode);
gcc_assert (dest2 != NULL_RTX);
if (!rtx_equal_p (dest, dest2))
emit_move_insn (dest, dest2);
}
else
gcc_unreachable ();
return;
}
/* Emit RTL that sets a register to zero if OP1 and OP2 are equal. SCRATCH
can be used as that dest register. Return the dest register. */
rtx
rs6000_emit_eqne (machine_mode mode, rtx op1, rtx op2, rtx scratch)
{
if (op2 == const0_rtx)
return op1;
if (GET_CODE (scratch) == SCRATCH)
scratch = gen_reg_rtx (mode);
if (logical_operand (op2, mode))
emit_insn (gen_rtx_SET (scratch, gen_rtx_XOR (mode, op1, op2)));
else
emit_insn (gen_rtx_SET (scratch,
gen_rtx_PLUS (mode, op1, negate_rtx (mode, op2))));
return scratch;
}
void
rs6000_emit_sCOND (machine_mode mode, rtx operands[])
{
rtx condition_rtx;
machine_mode op_mode;
enum rtx_code cond_code;
rtx result = operands[0];
condition_rtx = rs6000_generate_compare (operands[1], mode);
cond_code = GET_CODE (condition_rtx);
if (cond_code == NE
|| cond_code == GE || cond_code == LE
|| cond_code == GEU || cond_code == LEU
|| cond_code == ORDERED || cond_code == UNGE || cond_code == UNLE)
{
rtx not_result = gen_reg_rtx (CCEQmode);
rtx not_op, rev_cond_rtx;
machine_mode cc_mode;
cc_mode = GET_MODE (XEXP (condition_rtx, 0));
rev_cond_rtx = gen_rtx_fmt_ee (rs6000_reverse_condition (cc_mode, cond_code),
SImode, XEXP (condition_rtx, 0), const0_rtx);
not_op = gen_rtx_COMPARE (CCEQmode, rev_cond_rtx, const0_rtx);
emit_insn (gen_rtx_SET (not_result, not_op));
condition_rtx = gen_rtx_EQ (VOIDmode, not_result, const0_rtx);
}
op_mode = GET_MODE (XEXP (operands[1], 0));
if (op_mode == VOIDmode)
op_mode = GET_MODE (XEXP (operands[1], 1));
if (TARGET_POWERPC64 && (op_mode == DImode || FLOAT_MODE_P (mode)))
{
PUT_MODE (condition_rtx, DImode);
convert_move (result, condition_rtx, 0);
}
else
{
PUT_MODE (condition_rtx, SImode);
emit_insn (gen_rtx_SET (result, condition_rtx));
}
}
/* Emit a branch of kind CODE to location LOC. */
void
rs6000_emit_cbranch (machine_mode mode, rtx operands[])
{
rtx condition_rtx, loc_ref;
condition_rtx = rs6000_generate_compare (operands[0], mode);
loc_ref = gen_rtx_LABEL_REF (VOIDmode, operands[3]);
emit_jump_insn (gen_rtx_SET (pc_rtx,
gen_rtx_IF_THEN_ELSE (VOIDmode, condition_rtx,
loc_ref, pc_rtx)));
}
/* Return the string to output a conditional branch to LABEL, which is
the operand template of the label, or NULL if the branch is really a
conditional return.
OP is the conditional expression. XEXP (OP, 0) is assumed to be a
condition code register and its mode specifies what kind of
comparison we made.
REVERSED is nonzero if we should reverse the sense of the comparison.
INSN is the insn. */
char *
output_cbranch (rtx op, const char *label, int reversed, rtx_insn *insn)
{
static char string[64];
enum rtx_code code = GET_CODE (op);
rtx cc_reg = XEXP (op, 0);
machine_mode mode = GET_MODE (cc_reg);
int cc_regno = REGNO (cc_reg) - CR0_REGNO;
int need_longbranch = label != NULL && get_attr_length (insn) == 8;
int really_reversed = reversed ^ need_longbranch;
char *s = string;
const char *ccode;
const char *pred;
rtx note;
validate_condition_mode (code, mode);
/* Work out which way this really branches. We could use
reverse_condition_maybe_unordered here always but this
makes the resulting assembler clearer. */
if (really_reversed)
{
/* Reversal of FP compares takes care -- an ordered compare
becomes an unordered compare and vice versa. */
if (mode == CCFPmode)
code = reverse_condition_maybe_unordered (code);
else
code = reverse_condition (code);
}
switch (code)
{
/* Not all of these are actually distinct opcodes, but
we distinguish them for clarity of the resulting assembler. */
case NE: case LTGT:
ccode = "ne"; break;
case EQ: case UNEQ:
ccode = "eq"; break;
case GE: case GEU:
ccode = "ge"; break;
case GT: case GTU: case UNGT:
ccode = "gt"; break;
case LE: case LEU:
ccode = "le"; break;
case LT: case LTU: case UNLT:
ccode = "lt"; break;
case UNORDERED: ccode = "un"; break;
case ORDERED: ccode = "nu"; break;
case UNGE: ccode = "nl"; break;
case UNLE: ccode = "ng"; break;
default:
gcc_unreachable ();
}
/* Maybe we have a guess as to how likely the branch is. */
pred = "";
note = find_reg_note (insn, REG_BR_PROB, NULL_RTX);
if (note != NULL_RTX)
{
/* PROB is the difference from 50%. */
int prob = profile_probability::from_reg_br_prob_note (XINT (note, 0))
.to_reg_br_prob_base () - REG_BR_PROB_BASE / 2;
/* Only hint for highly probable/improbable branches on newer cpus when
we have real profile data, as static prediction overrides processor
dynamic prediction. For older cpus we may as well always hint, but
assume not taken for branches that are very close to 50% as a
mispredicted taken branch is more expensive than a
mispredicted not-taken branch. */
if (rs6000_always_hint
|| (abs (prob) > REG_BR_PROB_BASE / 100 * 48
&& (profile_status_for_fn (cfun) != PROFILE_GUESSED)
&& br_prob_note_reliable_p (note)))
{
if (abs (prob) > REG_BR_PROB_BASE / 20
&& ((prob > 0) ^ need_longbranch))
pred = "+";
else
pred = "-";
}
}
if (label == NULL)
s += sprintf (s, "b%slr%s ", ccode, pred);
else
s += sprintf (s, "b%s%s ", ccode, pred);
/* We need to escape any '%' characters in the reg_names string.
Assume they'd only be the first character.... */
if (reg_names[cc_regno + CR0_REGNO][0] == '%')
*s++ = '%';
s += sprintf (s, "%s", reg_names[cc_regno + CR0_REGNO]);
if (label != NULL)
{
/* If the branch distance was too far, we may have to use an
unconditional branch to go the distance. */
if (need_longbranch)
s += sprintf (s, ",$+8\n\tb %s", label);
else
s += sprintf (s, ",%s", label);
}
return string;
}
/* Return insn for VSX or Altivec comparisons. */
static rtx
rs6000_emit_vector_compare_inner (enum rtx_code code, rtx op0, rtx op1)
{
rtx mask;
machine_mode mode = GET_MODE (op0);
switch (code)
{
default:
break;
case GE:
if (GET_MODE_CLASS (mode) == MODE_VECTOR_INT)
return NULL_RTX;
/* FALLTHRU */
case EQ:
case GT:
case GTU:
case ORDERED:
case UNORDERED:
case UNEQ:
case LTGT:
mask = gen_reg_rtx (mode);
emit_insn (gen_rtx_SET (mask, gen_rtx_fmt_ee (code, mode, op0, op1)));
return mask;
}
return NULL_RTX;
}
/* Emit vector compare for operands OP0 and OP1 using code RCODE.
DMODE is expected destination mode. This is a recursive function. */
static rtx
rs6000_emit_vector_compare (enum rtx_code rcode,
rtx op0, rtx op1,
machine_mode dmode)
{
rtx mask;
bool swap_operands = false;
bool try_again = false;
gcc_assert (VECTOR_UNIT_ALTIVEC_OR_VSX_P (dmode));
gcc_assert (GET_MODE (op0) == GET_MODE (op1));
/* See if the comparison works as is. */
mask = rs6000_emit_vector_compare_inner (rcode, op0, op1);
if (mask)
return mask;
switch (rcode)
{
case LT:
rcode = GT;
swap_operands = true;
try_again = true;
break;
case LTU:
rcode = GTU;
swap_operands = true;
try_again = true;
break;
case NE:
case UNLE:
case UNLT:
case UNGE:
case UNGT:
/* Invert condition and try again.
e.g., A != B becomes ~(A==B). */
{
enum rtx_code rev_code;
enum insn_code nor_code;
rtx mask2;
rev_code = reverse_condition_maybe_unordered (rcode);
if (rev_code == UNKNOWN)
return NULL_RTX;
nor_code = optab_handler (one_cmpl_optab, dmode);
if (nor_code == CODE_FOR_nothing)
return NULL_RTX;
mask2 = rs6000_emit_vector_compare (rev_code, op0, op1, dmode);
if (!mask2)
return NULL_RTX;
mask = gen_reg_rtx (dmode);
emit_insn (GEN_FCN (nor_code) (mask, mask2));
return mask;
}
break;
case GE:
case GEU:
case LE:
case LEU:
/* Try GT/GTU/LT/LTU OR EQ */
{
rtx c_rtx, eq_rtx;
enum insn_code ior_code;
enum rtx_code new_code;
switch (rcode)
{
case GE:
new_code = GT;
break;
case GEU:
new_code = GTU;
break;
case LE:
new_code = LT;
break;
case LEU:
new_code = LTU;
break;
default:
gcc_unreachable ();
}
ior_code = optab_handler (ior_optab, dmode);
if (ior_code == CODE_FOR_nothing)
return NULL_RTX;
c_rtx = rs6000_emit_vector_compare (new_code, op0, op1, dmode);
if (!c_rtx)
return NULL_RTX;
eq_rtx = rs6000_emit_vector_compare (EQ, op0, op1, dmode);
if (!eq_rtx)
return NULL_RTX;
mask = gen_reg_rtx (dmode);
emit_insn (GEN_FCN (ior_code) (mask, c_rtx, eq_rtx));
return mask;
}
break;
default:
return NULL_RTX;
}
if (try_again)
{
if (swap_operands)
std::swap (op0, op1);
mask = rs6000_emit_vector_compare_inner (rcode, op0, op1);
if (mask)
return mask;
}
/* You only get two chances. */
return NULL_RTX;
}
/* Emit vector conditional expression. DEST is destination. OP_TRUE and
OP_FALSE are two VEC_COND_EXPR operands. CC_OP0 and CC_OP1 are the two
operands for the relation operation COND. */
int
rs6000_emit_vector_cond_expr (rtx dest, rtx op_true, rtx op_false,
rtx cond, rtx cc_op0, rtx cc_op1)
{
machine_mode dest_mode = GET_MODE (dest);
machine_mode mask_mode = GET_MODE (cc_op0);
enum rtx_code rcode = GET_CODE (cond);
machine_mode cc_mode = CCmode;
rtx mask;
rtx cond2;
bool invert_move = false;
if (VECTOR_UNIT_NONE_P (dest_mode))
return 0;
gcc_assert (GET_MODE_SIZE (dest_mode) == GET_MODE_SIZE (mask_mode)
&& GET_MODE_NUNITS (dest_mode) == GET_MODE_NUNITS (mask_mode));
switch (rcode)
{
/* Swap operands if we can, and fall back to doing the operation as
specified, and doing a NOR to invert the test. */
case NE:
case UNLE:
case UNLT:
case UNGE:
case UNGT:
/* Invert condition and try again.
e.g., A = (B != C) ? D : E becomes A = (B == C) ? E : D. */
invert_move = true;
rcode = reverse_condition_maybe_unordered (rcode);
if (rcode == UNKNOWN)
return 0;
break;
case GE:
case LE:
if (GET_MODE_CLASS (mask_mode) == MODE_VECTOR_INT)
{
/* Invert condition to avoid compound test. */
invert_move = true;
rcode = reverse_condition (rcode);
}
break;
case GTU:
case GEU:
case LTU:
case LEU:
/* Mark unsigned tests with CCUNSmode. */
cc_mode = CCUNSmode;
/* Invert condition to avoid compound test if necessary. */
if (rcode == GEU || rcode == LEU)
{
invert_move = true;
rcode = reverse_condition (rcode);
}
break;
default:
break;
}
/* Get the vector mask for the given relational operations. */
mask = rs6000_emit_vector_compare (rcode, cc_op0, cc_op1, mask_mode);
if (!mask)
return 0;
if (invert_move)
std::swap (op_true, op_false);
/* Optimize vec1 == vec2, to know the mask generates -1/0. */
if (GET_MODE_CLASS (dest_mode) == MODE_VECTOR_INT
&& (GET_CODE (op_true) == CONST_VECTOR
|| GET_CODE (op_false) == CONST_VECTOR))
{
rtx constant_0 = CONST0_RTX (dest_mode);
rtx constant_m1 = CONSTM1_RTX (dest_mode);
if (op_true == constant_m1 && op_false == constant_0)
{
emit_move_insn (dest, mask);
return 1;
}
else if (op_true == constant_0 && op_false == constant_m1)
{
emit_insn (gen_rtx_SET (dest, gen_rtx_NOT (dest_mode, mask)));
return 1;
}
/* If we can't use the vector comparison directly, perhaps we can use
the mask for the true or false fields, instead of loading up a
constant. */
if (op_true == constant_m1)
op_true = mask;
if (op_false == constant_0)
op_false = mask;
}
if (!REG_P (op_true) && !SUBREG_P (op_true))
op_true = force_reg (dest_mode, op_true);
if (!REG_P (op_false) && !SUBREG_P (op_false))
op_false = force_reg (dest_mode, op_false);
cond2 = gen_rtx_fmt_ee (NE, cc_mode, gen_lowpart (dest_mode, mask),
CONST0_RTX (dest_mode));
emit_insn (gen_rtx_SET (dest,
gen_rtx_IF_THEN_ELSE (dest_mode,
cond2,
op_true,
op_false)));
return 1;
}
/* ISA 3.0 (power9) minmax subcase to emit a XSMAXCDP or XSMINCDP instruction
for SF/DF scalars. Move TRUE_COND to DEST if OP of the operands of the last
comparison is nonzero/true, FALSE_COND if it is zero/false. Return 0 if the
hardware has no such operation. */
static int
rs6000_emit_p9_fp_minmax (rtx dest, rtx op, rtx true_cond, rtx false_cond)
{
enum rtx_code code = GET_CODE (op);
rtx op0 = XEXP (op, 0);
rtx op1 = XEXP (op, 1);
machine_mode compare_mode = GET_MODE (op0);
machine_mode result_mode = GET_MODE (dest);
bool max_p = false;
if (result_mode != compare_mode)
return 0;
if (code == GE || code == GT)
max_p = true;
else if (code == LE || code == LT)
max_p = false;
else
return 0;
if (rtx_equal_p (op0, true_cond) && rtx_equal_p (op1, false_cond))
;
else if (rtx_equal_p (op1, true_cond) && rtx_equal_p (op0, false_cond))
max_p = !max_p;
else
return 0;
rs6000_emit_minmax (dest, max_p ? SMAX : SMIN, op0, op1);
return 1;
}
/* ISA 3.0 (power9) conditional move subcase to emit XSCMP{EQ,GE,GT,NE}DP and
XXSEL instructions for SF/DF scalars. Move TRUE_COND to DEST if OP of the
operands of the last comparison is nonzero/true, FALSE_COND if it is
zero/false. Return 0 if the hardware has no such operation. */
static int
rs6000_emit_p9_fp_cmove (rtx dest, rtx op, rtx true_cond, rtx false_cond)
{
enum rtx_code code = GET_CODE (op);
rtx op0 = XEXP (op, 0);
rtx op1 = XEXP (op, 1);
machine_mode result_mode = GET_MODE (dest);
rtx compare_rtx;
rtx cmove_rtx;
rtx clobber_rtx;
if (!can_create_pseudo_p ())
return 0;
switch (code)
{
case EQ:
case GE:
case GT:
break;
case NE:
case LT:
case LE:
code = swap_condition (code);
std::swap (op0, op1);
break;
default:
return 0;
}
/* Generate: [(parallel [(set (dest)
(if_then_else (op (cmp1) (cmp2))
(true)
(false)))
(clobber (scratch))])]. */
compare_rtx = gen_rtx_fmt_ee (code, CCFPmode, op0, op1);
cmove_rtx = gen_rtx_SET (dest,
gen_rtx_IF_THEN_ELSE (result_mode,
compare_rtx,
true_cond,
false_cond));
clobber_rtx = gen_rtx_CLOBBER (VOIDmode, gen_rtx_SCRATCH (V2DImode));
emit_insn (gen_rtx_PARALLEL (VOIDmode,
gen_rtvec (2, cmove_rtx, clobber_rtx)));
return 1;
}
/* Emit a conditional move: move TRUE_COND to DEST if OP of the
operands of the last comparison is nonzero/true, FALSE_COND if it
is zero/false. Return 0 if the hardware has no such operation. */
int
rs6000_emit_cmove (rtx dest, rtx op, rtx true_cond, rtx false_cond)
{
enum rtx_code code = GET_CODE (op);
rtx op0 = XEXP (op, 0);
rtx op1 = XEXP (op, 1);
machine_mode compare_mode = GET_MODE (op0);
machine_mode result_mode = GET_MODE (dest);
rtx temp;
bool is_against_zero;
/* These modes should always match. */
if (GET_MODE (op1) != compare_mode
/* In the isel case however, we can use a compare immediate, so
op1 may be a small constant. */
&& (!TARGET_ISEL || !short_cint_operand (op1, VOIDmode)))
return 0;
if (GET_MODE (true_cond) != result_mode)
return 0;
if (GET_MODE (false_cond) != result_mode)
return 0;
/* See if we can use the ISA 3.0 (power9) min/max/compare functions. */
if (TARGET_P9_MINMAX
&& (compare_mode == SFmode || compare_mode == DFmode)
&& (result_mode == SFmode || result_mode == DFmode))
{
if (rs6000_emit_p9_fp_minmax (dest, op, true_cond, false_cond))
return 1;
if (rs6000_emit_p9_fp_cmove (dest, op, true_cond, false_cond))
return 1;
}
/* Don't allow using floating point comparisons for integer results for
now. */
if (FLOAT_MODE_P (compare_mode) && !FLOAT_MODE_P (result_mode))
return 0;
/* First, work out if the hardware can do this at all, or
if it's too slow.... */
if (!FLOAT_MODE_P (compare_mode))
{
if (TARGET_ISEL)
return rs6000_emit_int_cmove (dest, op, true_cond, false_cond);
return 0;
}
is_against_zero = op1 == CONST0_RTX (compare_mode);
/* A floating-point subtract might overflow, underflow, or produce
an inexact result, thus changing the floating-point flags, so it
can't be generated if we care about that. It's safe if one side
of the construct is zero, since then no subtract will be
generated. */
if (SCALAR_FLOAT_MODE_P (compare_mode)
&& flag_trapping_math && ! is_against_zero)
return 0;
/* Eliminate half of the comparisons by switching operands, this
makes the remaining code simpler. */
if (code == UNLT || code == UNGT || code == UNORDERED || code == NE
|| code == LTGT || code == LT || code == UNLE)
{
code = reverse_condition_maybe_unordered (code);
temp = true_cond;
true_cond = false_cond;
false_cond = temp;
}
/* UNEQ and LTGT take four instructions for a comparison with zero,
it'll probably be faster to use a branch here too. */
if (code == UNEQ && HONOR_NANS (compare_mode))
return 0;
/* We're going to try to implement comparisons by performing
a subtract, then comparing against zero. Unfortunately,
Inf - Inf is NaN which is not zero, and so if we don't
know that the operand is finite and the comparison
would treat EQ different to UNORDERED, we can't do it. */
if (HONOR_INFINITIES (compare_mode)
&& code != GT && code != UNGE
&& (GET_CODE (op1) != CONST_DOUBLE
|| real_isinf (CONST_DOUBLE_REAL_VALUE (op1)))
/* Constructs of the form (a OP b ? a : b) are safe. */
&& ((! rtx_equal_p (op0, false_cond) && ! rtx_equal_p (op1, false_cond))
|| (! rtx_equal_p (op0, true_cond)
&& ! rtx_equal_p (op1, true_cond))))
return 0;
/* At this point we know we can use fsel. */
/* Reduce the comparison to a comparison against zero. */
if (! is_against_zero)
{
temp = gen_reg_rtx (compare_mode);
emit_insn (gen_rtx_SET (temp, gen_rtx_MINUS (compare_mode, op0, op1)));
op0 = temp;
op1 = CONST0_RTX (compare_mode);
}
/* If we don't care about NaNs we can reduce some of the comparisons
down to faster ones. */
if (! HONOR_NANS (compare_mode))
switch (code)
{
case GT:
code = LE;
temp = true_cond;
true_cond = false_cond;
false_cond = temp;
break;
case UNGE:
code = GE;
break;
case UNEQ:
code = EQ;
break;
default:
break;
}
/* Now, reduce everything down to a GE. */
switch (code)
{
case GE:
break;
case LE:
temp = gen_reg_rtx (compare_mode);
emit_insn (gen_rtx_SET (temp, gen_rtx_NEG (compare_mode, op0)));
op0 = temp;
break;
case ORDERED:
temp = gen_reg_rtx (compare_mode);
emit_insn (gen_rtx_SET (temp, gen_rtx_ABS (compare_mode, op0)));
op0 = temp;
break;
case EQ:
temp = gen_reg_rtx (compare_mode);
emit_insn (gen_rtx_SET (temp,
gen_rtx_NEG (compare_mode,
gen_rtx_ABS (compare_mode, op0))));
op0 = temp;
break;
case UNGE:
/* a UNGE 0 <-> (a GE 0 || -a UNLT 0) */
temp = gen_reg_rtx (result_mode);
emit_insn (gen_rtx_SET (temp,
gen_rtx_IF_THEN_ELSE (result_mode,
gen_rtx_GE (VOIDmode,
op0, op1),
true_cond, false_cond)));
false_cond = true_cond;
true_cond = temp;
temp = gen_reg_rtx (compare_mode);
emit_insn (gen_rtx_SET (temp, gen_rtx_NEG (compare_mode, op0)));
op0 = temp;
break;
case GT:
/* a GT 0 <-> (a GE 0 && -a UNLT 0) */
temp = gen_reg_rtx (result_mode);
emit_insn (gen_rtx_SET (temp,
gen_rtx_IF_THEN_ELSE (result_mode,
gen_rtx_GE (VOIDmode,
op0, op1),
true_cond, false_cond)));
true_cond = false_cond;
false_cond = temp;
temp = gen_reg_rtx (compare_mode);
emit_insn (gen_rtx_SET (temp, gen_rtx_NEG (compare_mode, op0)));
op0 = temp;
break;
default:
gcc_unreachable ();
}
emit_insn (gen_rtx_SET (dest,
gen_rtx_IF_THEN_ELSE (result_mode,
gen_rtx_GE (VOIDmode,
op0, op1),
true_cond, false_cond)));
return 1;
}
/* Same as above, but for ints (isel). */
int
rs6000_emit_int_cmove (rtx dest, rtx op, rtx true_cond, rtx false_cond)
{
rtx condition_rtx, cr;
machine_mode mode = GET_MODE (dest);
enum rtx_code cond_code;
rtx (*isel_func) (rtx, rtx, rtx, rtx, rtx);
bool signedp;
if (mode != SImode && (!TARGET_POWERPC64 || mode != DImode))
return 0;
/* We still have to do the compare, because isel doesn't do a
compare, it just looks at the CRx bits set by a previous compare
instruction. */
condition_rtx = rs6000_generate_compare (op, mode);
cond_code = GET_CODE (condition_rtx);
cr = XEXP (condition_rtx, 0);
signedp = GET_MODE (cr) == CCmode;
isel_func = (mode == SImode
? (signedp ? gen_isel_signed_si : gen_isel_unsigned_si)
: (signedp ? gen_isel_signed_di : gen_isel_unsigned_di));
switch (cond_code)
{
case LT: case GT: case LTU: case GTU: case EQ:
/* isel handles these directly. */
break;
default:
/* We need to swap the sense of the comparison. */
{
std::swap (false_cond, true_cond);
PUT_CODE (condition_rtx, reverse_condition (cond_code));
}
break;
}
false_cond = force_reg (mode, false_cond);
if (true_cond != const0_rtx)
true_cond = force_reg (mode, true_cond);
emit_insn (isel_func (dest, condition_rtx, true_cond, false_cond, cr));
return 1;
}
void
rs6000_emit_minmax (rtx dest, enum rtx_code code, rtx op0, rtx op1)
{
machine_mode mode = GET_MODE (op0);
enum rtx_code c;
rtx target;
/* VSX/altivec have direct min/max insns. */
if ((code == SMAX || code == SMIN)
&& (VECTOR_UNIT_ALTIVEC_OR_VSX_P (mode)
|| (mode == SFmode && VECTOR_UNIT_VSX_P (DFmode))))
{
emit_insn (gen_rtx_SET (dest, gen_rtx_fmt_ee (code, mode, op0, op1)));
return;
}
if (code == SMAX || code == SMIN)
c = GE;
else
c = GEU;
if (code == SMAX || code == UMAX)
target = emit_conditional_move (dest, c, op0, op1, mode,
op0, op1, mode, 0);
else
target = emit_conditional_move (dest, c, op0, op1, mode,
op1, op0, mode, 0);
gcc_assert (target);
if (target != dest)
emit_move_insn (dest, target);
}
/* Split a signbit operation on 64-bit machines with direct move. Also allow
for the value to come from memory or if it is already loaded into a GPR. */
void
rs6000_split_signbit (rtx dest, rtx src)
{
machine_mode d_mode = GET_MODE (dest);
machine_mode s_mode = GET_MODE (src);
rtx dest_di = (d_mode == DImode) ? dest : gen_lowpart (DImode, dest);
rtx shift_reg = dest_di;
gcc_assert (FLOAT128_IEEE_P (s_mode) && TARGET_POWERPC64);
if (MEM_P (src))
{
rtx mem = (WORDS_BIG_ENDIAN
? adjust_address (src, DImode, 0)
: adjust_address (src, DImode, 8));
emit_insn (gen_rtx_SET (dest_di, mem));
}
else
{
unsigned int r = reg_or_subregno (src);
if (INT_REGNO_P (r))
shift_reg = gen_rtx_REG (DImode, r + (BYTES_BIG_ENDIAN == 0));
else
{
/* Generate the special mfvsrd instruction to get it in a GPR. */
gcc_assert (VSX_REGNO_P (r));
if (s_mode == KFmode)
emit_insn (gen_signbitkf2_dm2 (dest_di, src));
else
emit_insn (gen_signbittf2_dm2 (dest_di, src));
}
}
emit_insn (gen_lshrdi3 (dest_di, shift_reg, GEN_INT (63)));
return;
}
/* A subroutine of the atomic operation splitters. Jump to LABEL if
COND is true. Mark the jump as unlikely to be taken. */
static void
emit_unlikely_jump (rtx cond, rtx label)
{
rtx x = gen_rtx_IF_THEN_ELSE (VOIDmode, cond, label, pc_rtx);
rtx_insn *insn = emit_jump_insn (gen_rtx_SET (pc_rtx, x));
add_reg_br_prob_note (insn, profile_probability::very_unlikely ());
}
/* A subroutine of the atomic operation splitters. Emit a load-locked
instruction in MODE. For QI/HImode, possibly use a pattern than includes
the zero_extend operation. */
static void
emit_load_locked (machine_mode mode, rtx reg, rtx mem)
{
rtx (*fn) (rtx, rtx) = NULL;
switch (mode)
{
case E_QImode:
fn = gen_load_lockedqi;
break;
case E_HImode:
fn = gen_load_lockedhi;
break;
case E_SImode:
if (GET_MODE (mem) == QImode)
fn = gen_load_lockedqi_si;
else if (GET_MODE (mem) == HImode)
fn = gen_load_lockedhi_si;
else
fn = gen_load_lockedsi;
break;
case E_DImode:
fn = gen_load_lockeddi;
break;
case E_TImode:
fn = gen_load_lockedti;
break;
default:
gcc_unreachable ();
}
emit_insn (fn (reg, mem));
}
/* A subroutine of the atomic operation splitters. Emit a store-conditional
instruction in MODE. */
static void
emit_store_conditional (machine_mode mode, rtx res, rtx mem, rtx val)
{
rtx (*fn) (rtx, rtx, rtx) = NULL;
switch (mode)
{
case E_QImode:
fn = gen_store_conditionalqi;
break;
case E_HImode:
fn = gen_store_conditionalhi;
break;
case E_SImode:
fn = gen_store_conditionalsi;
break;
case E_DImode:
fn = gen_store_conditionaldi;
break;
case E_TImode:
fn = gen_store_conditionalti;
break;
default:
gcc_unreachable ();
}
/* Emit sync before stwcx. to address PPC405 Erratum. */
if (PPC405_ERRATUM77)
emit_insn (gen_hwsync ());
emit_insn (fn (res, mem, val));
}
/* Expand barriers before and after a load_locked/store_cond sequence. */
static rtx
rs6000_pre_atomic_barrier (rtx mem, enum memmodel model)
{
rtx addr = XEXP (mem, 0);
if (!legitimate_indirect_address_p (addr, reload_completed)
&& !legitimate_indexed_address_p (addr, reload_completed))
{
addr = force_reg (Pmode, addr);
mem = replace_equiv_address_nv (mem, addr);
}
switch (model)
{
case MEMMODEL_RELAXED:
case MEMMODEL_CONSUME:
case MEMMODEL_ACQUIRE:
break;
case MEMMODEL_RELEASE:
case MEMMODEL_ACQ_REL:
emit_insn (gen_lwsync ());
break;
case MEMMODEL_SEQ_CST:
emit_insn (gen_hwsync ());
break;
default:
gcc_unreachable ();
}
return mem;
}
static void
rs6000_post_atomic_barrier (enum memmodel model)
{
switch (model)
{
case MEMMODEL_RELAXED:
case MEMMODEL_CONSUME:
case MEMMODEL_RELEASE:
break;
case MEMMODEL_ACQUIRE:
case MEMMODEL_ACQ_REL:
case MEMMODEL_SEQ_CST:
emit_insn (gen_isync ());
break;
default:
gcc_unreachable ();
}
}
/* A subroutine of the various atomic expanders. For sub-word operations,
we must adjust things to operate on SImode. Given the original MEM,
return a new aligned memory. Also build and return the quantities by
which to shift and mask. */
static rtx
rs6000_adjust_atomic_subword (rtx orig_mem, rtx *pshift, rtx *pmask)
{
rtx addr, align, shift, mask, mem;
HOST_WIDE_INT shift_mask;
machine_mode mode = GET_MODE (orig_mem);
/* For smaller modes, we have to implement this via SImode. */
shift_mask = (mode == QImode ? 0x18 : 0x10);
addr = XEXP (orig_mem, 0);
addr = force_reg (GET_MODE (addr), addr);
/* Aligned memory containing subword. Generate a new memory. We
do not want any of the existing MEM_ATTR data, as we're now
accessing memory outside the original object. */
align = expand_simple_binop (Pmode, AND, addr, GEN_INT (-4),
NULL_RTX, 1, OPTAB_LIB_WIDEN);
mem = gen_rtx_MEM (SImode, align);
MEM_VOLATILE_P (mem) = MEM_VOLATILE_P (orig_mem);
if (MEM_ALIAS_SET (orig_mem) == ALIAS_SET_MEMORY_BARRIER)
set_mem_alias_set (mem, ALIAS_SET_MEMORY_BARRIER);
/* Shift amount for subword relative to aligned word. */
shift = gen_reg_rtx (SImode);
addr = gen_lowpart (SImode, addr);
rtx tmp = gen_reg_rtx (SImode);
emit_insn (gen_ashlsi3 (tmp, addr, GEN_INT (3)));
emit_insn (gen_andsi3 (shift, tmp, GEN_INT (shift_mask)));
if (BYTES_BIG_ENDIAN)
shift = expand_simple_binop (SImode, XOR, shift, GEN_INT (shift_mask),
shift, 1, OPTAB_LIB_WIDEN);
*pshift = shift;
/* Mask for insertion. */
mask = expand_simple_binop (SImode, ASHIFT, GEN_INT (GET_MODE_MASK (mode)),
shift, NULL_RTX, 1, OPTAB_LIB_WIDEN);
*pmask = mask;
return mem;
}
/* A subroutine of the various atomic expanders. For sub-word operands,
combine OLDVAL and NEWVAL via MASK. Returns a new pseduo. */
static rtx
rs6000_mask_atomic_subword (rtx oldval, rtx newval, rtx mask)
{
rtx x;
x = gen_reg_rtx (SImode);
emit_insn (gen_rtx_SET (x, gen_rtx_AND (SImode,
gen_rtx_NOT (SImode, mask),
oldval)));
x = expand_simple_binop (SImode, IOR, newval, x, x, 1, OPTAB_LIB_WIDEN);
return x;
}
/* A subroutine of the various atomic expanders. For sub-word operands,
extract WIDE to NARROW via SHIFT. */
static void
rs6000_finish_atomic_subword (rtx narrow, rtx wide, rtx shift)
{
wide = expand_simple_binop (SImode, LSHIFTRT, wide, shift,
wide, 1, OPTAB_LIB_WIDEN);
emit_move_insn (narrow, gen_lowpart (GET_MODE (narrow), wide));
}
/* Expand an atomic compare and swap operation. */
void
rs6000_expand_atomic_compare_and_swap (rtx operands[])
{
rtx boolval, retval, mem, oldval, newval, cond;
rtx label1, label2, x, mask, shift;
machine_mode mode, orig_mode;
enum memmodel mod_s, mod_f;
bool is_weak;
boolval = operands[0];
retval = operands[1];
mem = operands[2];
oldval = operands[3];
newval = operands[4];
is_weak = (INTVAL (operands[5]) != 0);
mod_s = memmodel_base (INTVAL (operands[6]));
mod_f = memmodel_base (INTVAL (operands[7]));
orig_mode = mode = GET_MODE (mem);
mask = shift = NULL_RTX;
if (mode == QImode || mode == HImode)
{
/* Before power8, we didn't have access to lbarx/lharx, so generate a
lwarx and shift/mask operations. With power8, we need to do the
comparison in SImode, but the store is still done in QI/HImode. */
oldval = convert_modes (SImode, mode, oldval, 1);
if (!TARGET_SYNC_HI_QI)
{
mem = rs6000_adjust_atomic_subword (mem, &shift, &mask);
/* Shift and mask OLDVAL into position with the word. */
oldval = expand_simple_binop (SImode, ASHIFT, oldval, shift,
NULL_RTX, 1, OPTAB_LIB_WIDEN);
/* Shift and mask NEWVAL into position within the word. */
newval = convert_modes (SImode, mode, newval, 1);
newval = expand_simple_binop (SImode, ASHIFT, newval, shift,
NULL_RTX, 1, OPTAB_LIB_WIDEN);
}
/* Prepare to adjust the return value. */
retval = gen_reg_rtx (SImode);
mode = SImode;
}
else if (reg_overlap_mentioned_p (retval, oldval))
oldval = copy_to_reg (oldval);
if (mode != TImode && !reg_or_short_operand (oldval, mode))
oldval = copy_to_mode_reg (mode, oldval);
if (reg_overlap_mentioned_p (retval, newval))
newval = copy_to_reg (newval);
mem = rs6000_pre_atomic_barrier (mem, mod_s);
label1 = NULL_RTX;
if (!is_weak)
{
label1 = gen_rtx_LABEL_REF (VOIDmode, gen_label_rtx ());
emit_label (XEXP (label1, 0));
}
label2 = gen_rtx_LABEL_REF (VOIDmode, gen_label_rtx ());
emit_load_locked (mode, retval, mem);
x = retval;
if (mask)
x = expand_simple_binop (SImode, AND, retval, mask,
NULL_RTX, 1, OPTAB_LIB_WIDEN);
cond = gen_reg_rtx (CCmode);
/* If we have TImode, synthesize a comparison. */
if (mode != TImode)
x = gen_rtx_COMPARE (CCmode, x, oldval);
else
{
rtx xor1_result = gen_reg_rtx (DImode);
rtx xor2_result = gen_reg_rtx (DImode);
rtx or_result = gen_reg_rtx (DImode);
rtx new_word0 = simplify_gen_subreg (DImode, x, TImode, 0);
rtx new_word1 = simplify_gen_subreg (DImode, x, TImode, 8);
rtx old_word0 = simplify_gen_subreg (DImode, oldval, TImode, 0);
rtx old_word1 = simplify_gen_subreg (DImode, oldval, TImode, 8);
emit_insn (gen_xordi3 (xor1_result, new_word0, old_word0));
emit_insn (gen_xordi3 (xor2_result, new_word1, old_word1));
emit_insn (gen_iordi3 (or_result, xor1_result, xor2_result));
x = gen_rtx_COMPARE (CCmode, or_result, const0_rtx);
}
emit_insn (gen_rtx_SET (cond, x));
x = gen_rtx_NE (VOIDmode, cond, const0_rtx);
emit_unlikely_jump (x, label2);
x = newval;
if (mask)
x = rs6000_mask_atomic_subword (retval, newval, mask);
emit_store_conditional (orig_mode, cond, mem, x);
if (!is_weak)
{
x = gen_rtx_NE (VOIDmode, cond, const0_rtx);
emit_unlikely_jump (x, label1);
}
if (!is_mm_relaxed (mod_f))
emit_label (XEXP (label2, 0));
rs6000_post_atomic_barrier (mod_s);
if (is_mm_relaxed (mod_f))
emit_label (XEXP (label2, 0));
if (shift)
rs6000_finish_atomic_subword (operands[1], retval, shift);
else if (mode != GET_MODE (operands[1]))
convert_move (operands[1], retval, 1);
/* In all cases, CR0 contains EQ on success, and NE on failure. */
x = gen_rtx_EQ (SImode, cond, const0_rtx);
emit_insn (gen_rtx_SET (boolval, x));
}
/* Expand an atomic exchange operation. */
void
rs6000_expand_atomic_exchange (rtx operands[])
{
rtx retval, mem, val, cond;
machine_mode mode;
enum memmodel model;
rtx label, x, mask, shift;
retval = operands[0];
mem = operands[1];
val = operands[2];
model = memmodel_base (INTVAL (operands[3]));
mode = GET_MODE (mem);
mask = shift = NULL_RTX;
if (!TARGET_SYNC_HI_QI && (mode == QImode || mode == HImode))
{
mem = rs6000_adjust_atomic_subword (mem, &shift, &mask);
/* Shift and mask VAL into position with the word. */
val = convert_modes (SImode, mode, val, 1);
val = expand_simple_binop (SImode, ASHIFT, val, shift,
NULL_RTX, 1, OPTAB_LIB_WIDEN);
/* Prepare to adjust the return value. */
retval = gen_reg_rtx (SImode);
mode = SImode;
}
mem = rs6000_pre_atomic_barrier (mem, model);
label = gen_rtx_LABEL_REF (VOIDmode, gen_label_rtx ());
emit_label (XEXP (label, 0));
emit_load_locked (mode, retval, mem);
x = val;
if (mask)
x = rs6000_mask_atomic_subword (retval, val, mask);
cond = gen_reg_rtx (CCmode);
emit_store_conditional (mode, cond, mem, x);
x = gen_rtx_NE (VOIDmode, cond, const0_rtx);
emit_unlikely_jump (x, label);
rs6000_post_atomic_barrier (model);
if (shift)
rs6000_finish_atomic_subword (operands[0], retval, shift);
}
/* Expand an atomic fetch-and-operate pattern. CODE is the binary operation
to perform. MEM is the memory on which to operate. VAL is the second
operand of the binary operator. BEFORE and AFTER are optional locations to
return the value of MEM either before of after the operation. MODEL_RTX
is a CONST_INT containing the memory model to use. */
void
rs6000_expand_atomic_op (enum rtx_code code, rtx mem, rtx val,
rtx orig_before, rtx orig_after, rtx model_rtx)
{
enum memmodel model = memmodel_base (INTVAL (model_rtx));
machine_mode mode = GET_MODE (mem);
machine_mode store_mode = mode;
rtx label, x, cond, mask, shift;
rtx before = orig_before, after = orig_after;
mask = shift = NULL_RTX;
/* On power8, we want to use SImode for the operation. On previous systems,
use the operation in a subword and shift/mask to get the proper byte or
halfword. */
if (mode == QImode || mode == HImode)
{
if (TARGET_SYNC_HI_QI)
{
val = convert_modes (SImode, mode, val, 1);
/* Prepare to adjust the return value. */
before = gen_reg_rtx (SImode);
if (after)
after = gen_reg_rtx (SImode);
mode = SImode;
}
else
{
mem = rs6000_adjust_atomic_subword (mem, &shift, &mask);
/* Shift and mask VAL into position with the word. */
val = convert_modes (SImode, mode, val, 1);
val = expand_simple_binop (SImode, ASHIFT, val, shift,
NULL_RTX, 1, OPTAB_LIB_WIDEN);
switch (code)
{
case IOR:
case XOR:
/* We've already zero-extended VAL. That is sufficient to
make certain that it does not affect other bits. */
mask = NULL;
break;
case AND:
/* If we make certain that all of the other bits in VAL are
set, that will be sufficient to not affect other bits. */
x = gen_rtx_NOT (SImode, mask);
x = gen_rtx_IOR (SImode, x, val);
emit_insn (gen_rtx_SET (val, x));
mask = NULL;
break;
case NOT:
case PLUS:
case MINUS:
/* These will all affect bits outside the field and need
adjustment via MASK within the loop. */
break;
default:
gcc_unreachable ();
}
/* Prepare to adjust the return value. */
before = gen_reg_rtx (SImode);
if (after)
after = gen_reg_rtx (SImode);
store_mode = mode = SImode;
}
}
mem = rs6000_pre_atomic_barrier (mem, model);
label = gen_label_rtx ();
emit_label (label);
label = gen_rtx_LABEL_REF (VOIDmode, label);
if (before == NULL_RTX)
before = gen_reg_rtx (mode);
emit_load_locked (mode, before, mem);
if (code == NOT)
{
x = expand_simple_binop (mode, AND, before, val,
NULL_RTX, 1, OPTAB_LIB_WIDEN);
after = expand_simple_unop (mode, NOT, x, after, 1);
}
else
{
after = expand_simple_binop (mode, code, before, val,
after, 1, OPTAB_LIB_WIDEN);
}
x = after;
if (mask)
{
x = expand_simple_binop (SImode, AND, after, mask,
NULL_RTX, 1, OPTAB_LIB_WIDEN);
x = rs6000_mask_atomic_subword (before, x, mask);
}
else if (store_mode != mode)
x = convert_modes (store_mode, mode, x, 1);
cond = gen_reg_rtx (CCmode);
emit_store_conditional (store_mode, cond, mem, x);
x = gen_rtx_NE (VOIDmode, cond, const0_rtx);
emit_unlikely_jump (x, label);
rs6000_post_atomic_barrier (model);
if (shift)
{
/* QImode/HImode on machines without lbarx/lharx where we do a lwarx and
then do the calcuations in a SImode register. */
if (orig_before)
rs6000_finish_atomic_subword (orig_before, before, shift);
if (orig_after)
rs6000_finish_atomic_subword (orig_after, after, shift);
}
else if (store_mode != mode)
{
/* QImode/HImode on machines with lbarx/lharx where we do the native
operation and then do the calcuations in a SImode register. */
if (orig_before)
convert_move (orig_before, before, 1);
if (orig_after)
convert_move (orig_after, after, 1);
}
else if (orig_after && after != orig_after)
emit_move_insn (orig_after, after);
}
/* Emit instructions to move SRC to DST. Called by splitters for
multi-register moves. It will emit at most one instruction for
each register that is accessed; that is, it won't emit li/lis pairs
(or equivalent for 64-bit code). One of SRC or DST must be a hard
register. */
void
rs6000_split_multireg_move (rtx dst, rtx src)
{
/* The register number of the first register being moved. */
int reg;
/* The mode that is to be moved. */
machine_mode mode;
/* The mode that the move is being done in, and its size. */
machine_mode reg_mode;
int reg_mode_size;
/* The number of registers that will be moved. */
int nregs;
reg = REG_P (dst) ? REGNO (dst) : REGNO (src);
mode = GET_MODE (dst);
nregs = hard_regno_nregs (reg, mode);
if (FP_REGNO_P (reg))
reg_mode = DECIMAL_FLOAT_MODE_P (mode) ? DDmode :
((TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT) ? DFmode : SFmode);
else if (ALTIVEC_REGNO_P (reg))
reg_mode = V16QImode;
else
reg_mode = word_mode;
reg_mode_size = GET_MODE_SIZE (reg_mode);
gcc_assert (reg_mode_size * nregs == GET_MODE_SIZE (mode));
/* TDmode residing in FP registers is special, since the ISA requires that
the lower-numbered word of a register pair is always the most significant
word, even in little-endian mode. This does not match the usual subreg
semantics, so we cannnot use simplify_gen_subreg in those cases. Access
the appropriate constituent registers "by hand" in little-endian mode.
Note we do not need to check for destructive overlap here since TDmode
can only reside in even/odd register pairs. */
if (FP_REGNO_P (reg) && DECIMAL_FLOAT_MODE_P (mode) && !BYTES_BIG_ENDIAN)
{
rtx p_src, p_dst;
int i;
for (i = 0; i < nregs; i++)
{
if (REG_P (src) && FP_REGNO_P (REGNO (src)))
p_src = gen_rtx_REG (reg_mode, REGNO (src) + nregs - 1 - i);
else
p_src = simplify_gen_subreg (reg_mode, src, mode,
i * reg_mode_size);
if (REG_P (dst) && FP_REGNO_P (REGNO (dst)))
p_dst = gen_rtx_REG (reg_mode, REGNO (dst) + nregs - 1 - i);
else
p_dst = simplify_gen_subreg (reg_mode, dst, mode,
i * reg_mode_size);
emit_insn (gen_rtx_SET (p_dst, p_src));
}
return;
}
if (REG_P (src) && REG_P (dst) && (REGNO (src) < REGNO (dst)))
{
/* Move register range backwards, if we might have destructive
overlap. */
int i;
for (i = nregs - 1; i >= 0; i--)
emit_insn (gen_rtx_SET (simplify_gen_subreg (reg_mode, dst, mode,
i * reg_mode_size),
simplify_gen_subreg (reg_mode, src, mode,
i * reg_mode_size)));
}
else
{
int i;
int j = -1;
bool used_update = false;
rtx restore_basereg = NULL_RTX;
if (MEM_P (src) && INT_REGNO_P (reg))
{
rtx breg;
if (GET_CODE (XEXP (src, 0)) == PRE_INC
|| GET_CODE (XEXP (src, 0)) == PRE_DEC)
{
rtx delta_rtx;
breg = XEXP (XEXP (src, 0), 0);
delta_rtx = (GET_CODE (XEXP (src, 0)) == PRE_INC
? GEN_INT (GET_MODE_SIZE (GET_MODE (src)))
: GEN_INT (-GET_MODE_SIZE (GET_MODE (src))));
emit_insn (gen_add3_insn (breg, breg, delta_rtx));
src = replace_equiv_address (src, breg);
}
else if (! rs6000_offsettable_memref_p (src, reg_mode))
{
if (GET_CODE (XEXP (src, 0)) == PRE_MODIFY)
{
rtx basereg = XEXP (XEXP (src, 0), 0);
if (TARGET_UPDATE)
{
rtx ndst = simplify_gen_subreg (reg_mode, dst, mode, 0);
emit_insn (gen_rtx_SET (ndst,
gen_rtx_MEM (reg_mode,
XEXP (src, 0))));
used_update = true;
}
else
emit_insn (gen_rtx_SET (basereg,
XEXP (XEXP (src, 0), 1)));
src = replace_equiv_address (src, basereg);
}
else
{
rtx basereg = gen_rtx_REG (Pmode, reg);
emit_insn (gen_rtx_SET (basereg, XEXP (src, 0)));
src = replace_equiv_address (src, basereg);
}
}
breg = XEXP (src, 0);
if (GET_CODE (breg) == PLUS || GET_CODE (breg) == LO_SUM)
breg = XEXP (breg, 0);
/* If the base register we are using to address memory is
also a destination reg, then change that register last. */
if (REG_P (breg)
&& REGNO (breg) >= REGNO (dst)
&& REGNO (breg) < REGNO (dst) + nregs)
j = REGNO (breg) - REGNO (dst);
}
else if (MEM_P (dst) && INT_REGNO_P (reg))
{
rtx breg;
if (GET_CODE (XEXP (dst, 0)) == PRE_INC
|| GET_CODE (XEXP (dst, 0)) == PRE_DEC)
{
rtx delta_rtx;
breg = XEXP (XEXP (dst, 0), 0);
delta_rtx = (GET_CODE (XEXP (dst, 0)) == PRE_INC
? GEN_INT (GET_MODE_SIZE (GET_MODE (dst)))
: GEN_INT (-GET_MODE_SIZE (GET_MODE (dst))));
/* We have to update the breg before doing the store.
Use store with update, if available. */
if (TARGET_UPDATE)
{
rtx nsrc = simplify_gen_subreg (reg_mode, src, mode, 0);
emit_insn (TARGET_32BIT
? (TARGET_POWERPC64
? gen_movdi_si_update (breg, breg, delta_rtx, nsrc)
: gen_movsi_update (breg, breg, delta_rtx, nsrc))
: gen_movdi_di_update (breg, breg, delta_rtx, nsrc));
used_update = true;
}
else
emit_insn (gen_add3_insn (breg, breg, delta_rtx));
dst = replace_equiv_address (dst, breg);
}
else if (!rs6000_offsettable_memref_p (dst, reg_mode)
&& GET_CODE (XEXP (dst, 0)) != LO_SUM)
{
if (GET_CODE (XEXP (dst, 0)) == PRE_MODIFY)
{
rtx basereg = XEXP (XEXP (dst, 0), 0);
if (TARGET_UPDATE)
{
rtx nsrc = simplify_gen_subreg (reg_mode, src, mode, 0);
emit_insn (gen_rtx_SET (gen_rtx_MEM (reg_mode,
XEXP (dst, 0)),
nsrc));
used_update = true;
}
else
emit_insn (gen_rtx_SET (basereg,
XEXP (XEXP (dst, 0), 1)));
dst = replace_equiv_address (dst, basereg);
}
else
{
rtx basereg = XEXP (XEXP (dst, 0), 0);
rtx offsetreg = XEXP (XEXP (dst, 0), 1);
gcc_assert (GET_CODE (XEXP (dst, 0)) == PLUS
&& REG_P (basereg)
&& REG_P (offsetreg)
&& REGNO (basereg) != REGNO (offsetreg));
if (REGNO (basereg) == 0)
{
rtx tmp = offsetreg;
offsetreg = basereg;
basereg = tmp;
}
emit_insn (gen_add3_insn (basereg, basereg, offsetreg));
restore_basereg = gen_sub3_insn (basereg, basereg, offsetreg);
dst = replace_equiv_address (dst, basereg);
}
}
else if (GET_CODE (XEXP (dst, 0)) != LO_SUM)
gcc_assert (rs6000_offsettable_memref_p (dst, reg_mode));
}
for (i = 0; i < nregs; i++)
{
/* Calculate index to next subword. */
++j;
if (j == nregs)
j = 0;
/* If compiler already emitted move of first word by
store with update, no need to do anything. */
if (j == 0 && used_update)
continue;
emit_insn (gen_rtx_SET (simplify_gen_subreg (reg_mode, dst, mode,
j * reg_mode_size),
simplify_gen_subreg (reg_mode, src, mode,
j * reg_mode_size)));
}
if (restore_basereg != NULL_RTX)
emit_insn (restore_basereg);
}
}
/* This page contains routines that are used to determine what the
function prologue and epilogue code will do and write them out. */
/* Determine whether the REG is really used. */
static bool
save_reg_p (int reg)
{
/* We need to mark the PIC offset register live for the same conditions
as it is set up, or otherwise it won't be saved before we clobber it. */
if (reg == RS6000_PIC_OFFSET_TABLE_REGNUM && !TARGET_SINGLE_PIC_BASE)
{
/* When calling eh_return, we must return true for all the cases
where conditional_register_usage marks the PIC offset reg
call used. */
if (TARGET_TOC && TARGET_MINIMAL_TOC
&& (crtl->calls_eh_return
|| df_regs_ever_live_p (reg)
|| !constant_pool_empty_p ()))
return true;
if ((DEFAULT_ABI == ABI_V4 || DEFAULT_ABI == ABI_DARWIN)
&& flag_pic)
return true;
}
return !call_used_regs[reg] && df_regs_ever_live_p (reg);
}
/* Return the first fixed-point register that is required to be
saved. 32 if none. */
int
first_reg_to_save (void)
{
int first_reg;
/* Find lowest numbered live register. */
for (first_reg = 13; first_reg <= 31; first_reg++)
if (save_reg_p (first_reg))
break;
#if TARGET_MACHO
if (flag_pic
&& crtl->uses_pic_offset_table
&& first_reg > RS6000_PIC_OFFSET_TABLE_REGNUM)
return RS6000_PIC_OFFSET_TABLE_REGNUM;
#endif
return first_reg;
}
/* Similar, for FP regs. */
int
first_fp_reg_to_save (void)
{
int first_reg;
/* Find lowest numbered live register. */
for (first_reg = 14 + 32; first_reg <= 63; first_reg++)
if (save_reg_p (first_reg))
break;
return first_reg;
}
/* Similar, for AltiVec regs. */
static int
first_altivec_reg_to_save (void)
{
int i;
/* Stack frame remains as is unless we are in AltiVec ABI. */
if (! TARGET_ALTIVEC_ABI)
return LAST_ALTIVEC_REGNO + 1;
/* On Darwin, the unwind routines are compiled without
TARGET_ALTIVEC, and use save_world to save/restore the
altivec registers when necessary. */
if (DEFAULT_ABI == ABI_DARWIN && crtl->calls_eh_return
&& ! TARGET_ALTIVEC)
return FIRST_ALTIVEC_REGNO + 20;
/* Find lowest numbered live register. */
for (i = FIRST_ALTIVEC_REGNO + 20; i <= LAST_ALTIVEC_REGNO; ++i)
if (save_reg_p (i))
break;
return i;
}
/* Return a 32-bit mask of the AltiVec registers we need to set in
VRSAVE. Bit n of the return value is 1 if Vn is live. The MSB in
the 32-bit word is 0. */
static unsigned int
compute_vrsave_mask (void)
{
unsigned int i, mask = 0;
/* On Darwin, the unwind routines are compiled without
TARGET_ALTIVEC, and use save_world to save/restore the
call-saved altivec registers when necessary. */
if (DEFAULT_ABI == ABI_DARWIN && crtl->calls_eh_return
&& ! TARGET_ALTIVEC)
mask |= 0xFFF;
/* First, find out if we use _any_ altivec registers. */
for (i = FIRST_ALTIVEC_REGNO; i <= LAST_ALTIVEC_REGNO; ++i)
if (df_regs_ever_live_p (i))
mask |= ALTIVEC_REG_BIT (i);
if (mask == 0)
return mask;
/* Next, remove the argument registers from the set. These must
be in the VRSAVE mask set by the caller, so we don't need to add
them in again. More importantly, the mask we compute here is
used to generate CLOBBERs in the set_vrsave insn, and we do not
wish the argument registers to die. */
for (i = ALTIVEC_ARG_MIN_REG; i < (unsigned) crtl->args.info.vregno; i++)
mask &= ~ALTIVEC_REG_BIT (i);
/* Similarly, remove the return value from the set. */
{
bool yes = false;
diddle_return_value (is_altivec_return_reg, &yes);
if (yes)
mask &= ~ALTIVEC_REG_BIT (ALTIVEC_ARG_RETURN);
}
return mask;
}
/* For a very restricted set of circumstances, we can cut down the
size of prologues/epilogues by calling our own save/restore-the-world
routines. */
static void
compute_save_world_info (rs6000_stack_t *info)
{
info->world_save_p = 1;
info->world_save_p
= (WORLD_SAVE_P (info)
&& DEFAULT_ABI == ABI_DARWIN
&& !cfun->has_nonlocal_label
&& info->first_fp_reg_save == FIRST_SAVED_FP_REGNO
&& info->first_gp_reg_save == FIRST_SAVED_GP_REGNO
&& info->first_altivec_reg_save == FIRST_SAVED_ALTIVEC_REGNO
&& info->cr_save_p);
/* This will not work in conjunction with sibcalls. Make sure there
are none. (This check is expensive, but seldom executed.) */
if (WORLD_SAVE_P (info))
{
rtx_insn *insn;
for (insn = get_last_insn_anywhere (); insn; insn = PREV_INSN (insn))
if (CALL_P (insn) && SIBLING_CALL_P (insn))
{
info->world_save_p = 0;
break;
}
}
if (WORLD_SAVE_P (info))
{
/* Even if we're not touching VRsave, make sure there's room on the
stack for it, if it looks like we're calling SAVE_WORLD, which
will attempt to save it. */
info->vrsave_size = 4;
/* If we are going to save the world, we need to save the link register too. */
info->lr_save_p = 1;
/* "Save" the VRsave register too if we're saving the world. */
if (info->vrsave_mask == 0)
info->vrsave_mask = compute_vrsave_mask ();
/* Because the Darwin register save/restore routines only handle
F14 .. F31 and V20 .. V31 as per the ABI, perform a consistency
check. */
gcc_assert (info->first_fp_reg_save >= FIRST_SAVED_FP_REGNO
&& (info->first_altivec_reg_save
>= FIRST_SAVED_ALTIVEC_REGNO));
}
return;
}
static void
is_altivec_return_reg (rtx reg, void *xyes)
{
bool *yes = (bool *) xyes;
if (REGNO (reg) == ALTIVEC_ARG_RETURN)
*yes = true;
}
/* Return whether REG is a global user reg or has been specifed by
-ffixed-REG. We should not restore these, and so cannot use
lmw or out-of-line restore functions if there are any. We also
can't save them (well, emit frame notes for them), because frame
unwinding during exception handling will restore saved registers. */
static bool
fixed_reg_p (int reg)
{
/* Ignore fixed_regs[RS6000_PIC_OFFSET_TABLE_REGNUM] when the
backend sets it, overriding anything the user might have given. */
if (reg == RS6000_PIC_OFFSET_TABLE_REGNUM
&& ((DEFAULT_ABI == ABI_V4 && flag_pic)
|| (DEFAULT_ABI == ABI_DARWIN && flag_pic)
|| (TARGET_TOC && TARGET_MINIMAL_TOC)))
return false;
return fixed_regs[reg];
}
/* Determine the strategy for savings/restoring registers. */
enum {
SAVE_MULTIPLE = 0x1,
SAVE_INLINE_GPRS = 0x2,
SAVE_INLINE_FPRS = 0x4,
SAVE_NOINLINE_GPRS_SAVES_LR = 0x8,
SAVE_NOINLINE_FPRS_SAVES_LR = 0x10,
SAVE_INLINE_VRS = 0x20,
REST_MULTIPLE = 0x100,
REST_INLINE_GPRS = 0x200,
REST_INLINE_FPRS = 0x400,
REST_NOINLINE_FPRS_DOESNT_RESTORE_LR = 0x800,
REST_INLINE_VRS = 0x1000
};
static int
rs6000_savres_strategy (rs6000_stack_t *info,
bool using_static_chain_p)
{
int strategy = 0;
/* Select between in-line and out-of-line save and restore of regs.
First, all the obvious cases where we don't use out-of-line. */
if (crtl->calls_eh_return
|| cfun->machine->ra_need_lr)
strategy |= (SAVE_INLINE_FPRS | REST_INLINE_FPRS
| SAVE_INLINE_GPRS | REST_INLINE_GPRS
| SAVE_INLINE_VRS | REST_INLINE_VRS);
if (info->first_gp_reg_save == 32)
strategy |= SAVE_INLINE_GPRS | REST_INLINE_GPRS;
if (info->first_fp_reg_save == 64
/* The out-of-line FP routines use double-precision stores;
we can't use those routines if we don't have such stores. */
|| (TARGET_HARD_FLOAT && !TARGET_DOUBLE_FLOAT))
strategy |= SAVE_INLINE_FPRS | REST_INLINE_FPRS;
if (info->first_altivec_reg_save == LAST_ALTIVEC_REGNO + 1)
strategy |= SAVE_INLINE_VRS | REST_INLINE_VRS;
/* Define cutoff for using out-of-line functions to save registers. */
if (DEFAULT_ABI == ABI_V4 || TARGET_ELF)
{
if (!optimize_size)
{
strategy |= SAVE_INLINE_FPRS | REST_INLINE_FPRS;
strategy |= SAVE_INLINE_GPRS | REST_INLINE_GPRS;
strategy |= SAVE_INLINE_VRS | REST_INLINE_VRS;
}
else
{
/* Prefer out-of-line restore if it will exit. */
if (info->first_fp_reg_save > 61)
strategy |= SAVE_INLINE_FPRS;
if (info->first_gp_reg_save > 29)
{
if (info->first_fp_reg_save == 64)
strategy |= SAVE_INLINE_GPRS;
else
strategy |= SAVE_INLINE_GPRS | REST_INLINE_GPRS;
}
if (info->first_altivec_reg_save == LAST_ALTIVEC_REGNO)
strategy |= SAVE_INLINE_VRS | REST_INLINE_VRS;
}
}
else if (DEFAULT_ABI == ABI_DARWIN)
{
if (info->first_fp_reg_save > 60)
strategy |= SAVE_INLINE_FPRS | REST_INLINE_FPRS;
if (info->first_gp_reg_save > 29)
strategy |= SAVE_INLINE_GPRS | REST_INLINE_GPRS;
strategy |= SAVE_INLINE_VRS | REST_INLINE_VRS;
}
else
{
gcc_checking_assert (DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2);
if ((flag_shrink_wrap_separate && optimize_function_for_speed_p (cfun))
|| info->first_fp_reg_save > 61)
strategy |= SAVE_INLINE_FPRS | REST_INLINE_FPRS;
strategy |= SAVE_INLINE_GPRS | REST_INLINE_GPRS;
strategy |= SAVE_INLINE_VRS | REST_INLINE_VRS;
}
/* Don't bother to try to save things out-of-line if r11 is occupied
by the static chain. It would require too much fiddling and the
static chain is rarely used anyway. FPRs are saved w.r.t the stack
pointer on Darwin, and AIX uses r1 or r12. */
if (using_static_chain_p
&& (DEFAULT_ABI == ABI_V4 || DEFAULT_ABI == ABI_DARWIN))
strategy |= ((DEFAULT_ABI == ABI_DARWIN ? 0 : SAVE_INLINE_FPRS)
| SAVE_INLINE_GPRS
| SAVE_INLINE_VRS);
/* Don't ever restore fixed regs. That means we can't use the
out-of-line register restore functions if a fixed reg is in the
range of regs restored. */
if (!(strategy & REST_INLINE_FPRS))
for (int i = info->first_fp_reg_save; i < 64; i++)
if (fixed_regs[i])
{
strategy |= REST_INLINE_FPRS;
break;
}
/* We can only use the out-of-line routines to restore fprs if we've
saved all the registers from first_fp_reg_save in the prologue.
Otherwise, we risk loading garbage. Of course, if we have saved
out-of-line then we know we haven't skipped any fprs. */
if ((strategy & SAVE_INLINE_FPRS)
&& !(strategy & REST_INLINE_FPRS))
for (int i = info->first_fp_reg_save; i < 64; i++)
if (!save_reg_p (i))
{
strategy |= REST_INLINE_FPRS;
break;
}
/* Similarly, for altivec regs. */
if (!(strategy & REST_INLINE_VRS))
for (int i = info->first_altivec_reg_save; i < LAST_ALTIVEC_REGNO + 1; i++)
if (fixed_regs[i])
{
strategy |= REST_INLINE_VRS;
break;
}
if ((strategy & SAVE_INLINE_VRS)
&& !(strategy & REST_INLINE_VRS))
for (int i = info->first_altivec_reg_save; i < LAST_ALTIVEC_REGNO + 1; i++)
if (!save_reg_p (i))
{
strategy |= REST_INLINE_VRS;
break;
}
/* info->lr_save_p isn't yet set if the only reason lr needs to be
saved is an out-of-line save or restore. Set up the value for
the next test (excluding out-of-line gprs). */
bool lr_save_p = (info->lr_save_p
|| !(strategy & SAVE_INLINE_FPRS)
|| !(strategy & SAVE_INLINE_VRS)
|| !(strategy & REST_INLINE_FPRS)
|| !(strategy & REST_INLINE_VRS));
if (TARGET_MULTIPLE
&& !TARGET_POWERPC64
&& info->first_gp_reg_save < 31
&& !(flag_shrink_wrap
&& flag_shrink_wrap_separate
&& optimize_function_for_speed_p (cfun)))
{
int count = 0;
for (int i = info->first_gp_reg_save; i < 32; i++)
if (save_reg_p (i))
count++;
if (count <= 1)
/* Don't use store multiple if only one reg needs to be
saved. This can occur for example when the ABI_V4 pic reg
(r30) needs to be saved to make calls, but r31 is not
used. */
strategy |= SAVE_INLINE_GPRS | REST_INLINE_GPRS;
else
{
/* Prefer store multiple for saves over out-of-line
routines, since the store-multiple instruction will
always be smaller. */
strategy |= SAVE_INLINE_GPRS | SAVE_MULTIPLE;
/* The situation is more complicated with load multiple.
We'd prefer to use the out-of-line routines for restores,
since the "exit" out-of-line routines can handle the
restore of LR and the frame teardown. However if doesn't
make sense to use the out-of-line routine if that is the
only reason we'd need to save LR, and we can't use the
"exit" out-of-line gpr restore if we have saved some
fprs; In those cases it is advantageous to use load
multiple when available. */
if (info->first_fp_reg_save != 64 || !lr_save_p)
strategy |= REST_INLINE_GPRS | REST_MULTIPLE;
}
}
/* Using the "exit" out-of-line routine does not improve code size
if using it would require lr to be saved and if only saving one
or two gprs. */
else if (!lr_save_p && info->first_gp_reg_save > 29)
strategy |= SAVE_INLINE_GPRS | REST_INLINE_GPRS;
/* Don't ever restore fixed regs. */
if ((strategy & (REST_INLINE_GPRS | REST_MULTIPLE)) != REST_INLINE_GPRS)
for (int i = info->first_gp_reg_save; i < 32; i++)
if (fixed_reg_p (i))
{
strategy |= REST_INLINE_GPRS;
strategy &= ~REST_MULTIPLE;
break;
}
/* We can only use load multiple or the out-of-line routines to
restore gprs if we've saved all the registers from
first_gp_reg_save. Otherwise, we risk loading garbage.
Of course, if we have saved out-of-line or used stmw then we know
we haven't skipped any gprs. */
if ((strategy & (SAVE_INLINE_GPRS | SAVE_MULTIPLE)) == SAVE_INLINE_GPRS
&& (strategy & (REST_INLINE_GPRS | REST_MULTIPLE)) != REST_INLINE_GPRS)
for (int i = info->first_gp_reg_save; i < 32; i++)
if (!save_reg_p (i))
{
strategy |= REST_INLINE_GPRS;
strategy &= ~REST_MULTIPLE;
break;
}
if (TARGET_ELF && TARGET_64BIT)
{
if (!(strategy & SAVE_INLINE_FPRS))
strategy |= SAVE_NOINLINE_FPRS_SAVES_LR;
else if (!(strategy & SAVE_INLINE_GPRS)
&& info->first_fp_reg_save == 64)
strategy |= SAVE_NOINLINE_GPRS_SAVES_LR;
}
else if (TARGET_AIX && !(strategy & REST_INLINE_FPRS))
strategy |= REST_NOINLINE_FPRS_DOESNT_RESTORE_LR;
if (TARGET_MACHO && !(strategy & SAVE_INLINE_FPRS))
strategy |= SAVE_NOINLINE_FPRS_SAVES_LR;
return strategy;
}
/* Calculate the stack information for the current function. This is
complicated by having two separate calling sequences, the AIX calling
sequence and the V.4 calling sequence.
AIX (and Darwin/Mac OS X) stack frames look like:
32-bit 64-bit
SP----> +---------------------------------------+
| back chain to caller | 0 0
+---------------------------------------+
| saved CR | 4 8 (8-11)
+---------------------------------------+
| saved LR | 8 16
+---------------------------------------+
| reserved for compilers | 12 24
+---------------------------------------+
| reserved for binders | 16 32
+---------------------------------------+
| saved TOC pointer | 20 40
+---------------------------------------+
| Parameter save area (+padding*) (P) | 24 48
+---------------------------------------+
| Alloca space (A) | 24+P etc.
+---------------------------------------+
| Local variable space (L) | 24+P+A
+---------------------------------------+
| Float/int conversion temporary (X) | 24+P+A+L
+---------------------------------------+
| Save area for AltiVec registers (W) | 24+P+A+L+X
+---------------------------------------+
| AltiVec alignment padding (Y) | 24+P+A+L+X+W
+---------------------------------------+
| Save area for VRSAVE register (Z) | 24+P+A+L+X+W+Y
+---------------------------------------+
| Save area for GP registers (G) | 24+P+A+X+L+X+W+Y+Z
+---------------------------------------+
| Save area for FP registers (F) | 24+P+A+X+L+X+W+Y+Z+G
+---------------------------------------+
old SP->| back chain to caller's caller |
+---------------------------------------+
* If the alloca area is present, the parameter save area is
padded so that the former starts 16-byte aligned.
The required alignment for AIX configurations is two words (i.e., 8
or 16 bytes).
The ELFv2 ABI is a variant of the AIX ABI. Stack frames look like:
SP----> +---------------------------------------+
| Back chain to caller | 0
+---------------------------------------+
| Save area for CR | 8
+---------------------------------------+
| Saved LR | 16
+---------------------------------------+
| Saved TOC pointer | 24
+---------------------------------------+
| Parameter save area (+padding*) (P) | 32
+---------------------------------------+
| Alloca space (A) | 32+P
+---------------------------------------+
| Local variable space (L) | 32+P+A
+---------------------------------------+
| Save area for AltiVec registers (W) | 32+P+A+L
+---------------------------------------+
| AltiVec alignment padding (Y) | 32+P+A+L+W
+---------------------------------------+
| Save area for GP registers (G) | 32+P+A+L+W+Y
+---------------------------------------+
| Save area for FP registers (F) | 32+P+A+L+W+Y+G
+---------------------------------------+
old SP->| back chain to caller's caller | 32+P+A+L+W+Y+G+F
+---------------------------------------+
* If the alloca area is present, the parameter save area is
padded so that the former starts 16-byte aligned.
V.4 stack frames look like:
SP----> +---------------------------------------+
| back chain to caller | 0
+---------------------------------------+
| caller's saved LR | 4
+---------------------------------------+
| Parameter save area (+padding*) (P) | 8
+---------------------------------------+
| Alloca space (A) | 8+P
+---------------------------------------+
| Varargs save area (V) | 8+P+A
+---------------------------------------+
| Local variable space (L) | 8+P+A+V
+---------------------------------------+
| Float/int conversion temporary (X) | 8+P+A+V+L
+---------------------------------------+
| Save area for AltiVec registers (W) | 8+P+A+V+L+X
+---------------------------------------+
| AltiVec alignment padding (Y) | 8+P+A+V+L+X+W
+---------------------------------------+
| Save area for VRSAVE register (Z) | 8+P+A+V+L+X+W+Y
+---------------------------------------+
| saved CR (C) | 8+P+A+V+L+X+W+Y+Z
+---------------------------------------+
| Save area for GP registers (G) | 8+P+A+V+L+X+W+Y+Z+C
+---------------------------------------+
| Save area for FP registers (F) | 8+P+A+V+L+X+W+Y+Z+C+G
+---------------------------------------+
old SP->| back chain to caller's caller |
+---------------------------------------+
* If the alloca area is present and the required alignment is
16 bytes, the parameter save area is padded so that the
alloca area starts 16-byte aligned.
The required alignment for V.4 is 16 bytes, or 8 bytes if -meabi is
given. (But note below and in sysv4.h that we require only 8 and
may round up the size of our stack frame anyways. The historical
reason is early versions of powerpc-linux which didn't properly
align the stack at program startup. A happy side-effect is that
-mno-eabi libraries can be used with -meabi programs.)
The EABI configuration defaults to the V.4 layout. However,
the stack alignment requirements may differ. If -mno-eabi is not
given, the required stack alignment is 8 bytes; if -mno-eabi is
given, the required alignment is 16 bytes. (But see V.4 comment
above.) */
#ifndef ABI_STACK_BOUNDARY
#define ABI_STACK_BOUNDARY STACK_BOUNDARY
#endif
static rs6000_stack_t *
rs6000_stack_info (void)
{
/* We should never be called for thunks, we are not set up for that. */
gcc_assert (!cfun->is_thunk);
rs6000_stack_t *info = &stack_info;
int reg_size = TARGET_32BIT ? 4 : 8;
int ehrd_size;
int ehcr_size;
int save_align;
int first_gp;
HOST_WIDE_INT non_fixed_size;
bool using_static_chain_p;
if (reload_completed && info->reload_completed)
return info;
memset (info, 0, sizeof (*info));
info->reload_completed = reload_completed;
/* Select which calling sequence. */
info->abi = DEFAULT_ABI;
/* Calculate which registers need to be saved & save area size. */
info->first_gp_reg_save = first_reg_to_save ();
/* Assume that we will have to save RS6000_PIC_OFFSET_TABLE_REGNUM,
even if it currently looks like we won't. Reload may need it to
get at a constant; if so, it will have already created a constant
pool entry for it. */
if (((TARGET_TOC && TARGET_MINIMAL_TOC)
|| (flag_pic == 1 && DEFAULT_ABI == ABI_V4)
|| (flag_pic && DEFAULT_ABI == ABI_DARWIN))
&& crtl->uses_const_pool
&& info->first_gp_reg_save > RS6000_PIC_OFFSET_TABLE_REGNUM)
first_gp = RS6000_PIC_OFFSET_TABLE_REGNUM;
else
first_gp = info->first_gp_reg_save;
info->gp_size = reg_size * (32 - first_gp);
info->first_fp_reg_save = first_fp_reg_to_save ();
info->fp_size = 8 * (64 - info->first_fp_reg_save);
info->first_altivec_reg_save = first_altivec_reg_to_save ();
info->altivec_size = 16 * (LAST_ALTIVEC_REGNO + 1
- info->first_altivec_reg_save);
/* Does this function call anything? */
info->calls_p = (!crtl->is_leaf || cfun->machine->ra_needs_full_frame);
/* Determine if we need to save the condition code registers. */
if (save_reg_p (CR2_REGNO)
|| save_reg_p (CR3_REGNO)
|| save_reg_p (CR4_REGNO))
{
info->cr_save_p = 1;
if (DEFAULT_ABI == ABI_V4)
info->cr_size = reg_size;
}
/* If the current function calls __builtin_eh_return, then we need
to allocate stack space for registers that will hold data for
the exception handler. */
if (crtl->calls_eh_return)
{
unsigned int i;
for (i = 0; EH_RETURN_DATA_REGNO (i) != INVALID_REGNUM; ++i)
continue;
ehrd_size = i * UNITS_PER_WORD;
}
else
ehrd_size = 0;
/* In the ELFv2 ABI, we also need to allocate space for separate
CR field save areas if the function calls __builtin_eh_return. */
if (DEFAULT_ABI == ABI_ELFv2 && crtl->calls_eh_return)
{
/* This hard-codes that we have three call-saved CR fields. */
ehcr_size = 3 * reg_size;
/* We do *not* use the regular CR save mechanism. */
info->cr_save_p = 0;
}
else
ehcr_size = 0;
/* Determine various sizes. */
info->reg_size = reg_size;
info->fixed_size = RS6000_SAVE_AREA;
info->vars_size = RS6000_ALIGN (get_frame_size (), 8);
if (cfun->calls_alloca)
info->parm_size =
RS6000_ALIGN (crtl->outgoing_args_size + info->fixed_size,
STACK_BOUNDARY / BITS_PER_UNIT) - info->fixed_size;
else
info->parm_size = RS6000_ALIGN (crtl->outgoing_args_size,
TARGET_ALTIVEC ? 16 : 8);
if (FRAME_GROWS_DOWNWARD)
info->vars_size
+= RS6000_ALIGN (info->fixed_size + info->vars_size + info->parm_size,
ABI_STACK_BOUNDARY / BITS_PER_UNIT)
- (info->fixed_size + info->vars_size + info->parm_size);
if (TARGET_ALTIVEC_ABI)
info->vrsave_mask = compute_vrsave_mask ();
if (TARGET_ALTIVEC_VRSAVE && info->vrsave_mask)
info->vrsave_size = 4;
compute_save_world_info (info);
/* Calculate the offsets. */
switch (DEFAULT_ABI)
{
case ABI_NONE:
default:
gcc_unreachable ();
case ABI_AIX:
case ABI_ELFv2:
case ABI_DARWIN:
info->fp_save_offset = -info->fp_size;
info->gp_save_offset = info->fp_save_offset - info->gp_size;
if (TARGET_ALTIVEC_ABI)
{
info->vrsave_save_offset = info->gp_save_offset - info->vrsave_size;
/* Align stack so vector save area is on a quadword boundary.
The padding goes above the vectors. */
if (info->altivec_size != 0)
info->altivec_padding_size = info->vrsave_save_offset & 0xF;
info->altivec_save_offset = info->vrsave_save_offset
- info->altivec_padding_size
- info->altivec_size;
gcc_assert (info->altivec_size == 0
|| info->altivec_save_offset % 16 == 0);
/* Adjust for AltiVec case. */
info->ehrd_offset = info->altivec_save_offset - ehrd_size;
}
else
info->ehrd_offset = info->gp_save_offset - ehrd_size;
info->ehcr_offset = info->ehrd_offset - ehcr_size;
info->cr_save_offset = reg_size; /* first word when 64-bit. */
info->lr_save_offset = 2*reg_size;
break;
case ABI_V4:
info->fp_save_offset = -info->fp_size;
info->gp_save_offset = info->fp_save_offset - info->gp_size;
info->cr_save_offset = info->gp_save_offset - info->cr_size;
if (TARGET_ALTIVEC_ABI)
{
info->vrsave_save_offset = info->cr_save_offset - info->vrsave_size;
/* Align stack so vector save area is on a quadword boundary. */
if (info->altivec_size != 0)
info->altivec_padding_size = 16 - (-info->vrsave_save_offset % 16);
info->altivec_save_offset = info->vrsave_save_offset
- info->altivec_padding_size
- info->altivec_size;
/* Adjust for AltiVec case. */
info->ehrd_offset = info->altivec_save_offset;
}
else
info->ehrd_offset = info->cr_save_offset;
info->ehrd_offset -= ehrd_size;
info->lr_save_offset = reg_size;
}
save_align = (TARGET_ALTIVEC_ABI || DEFAULT_ABI == ABI_DARWIN) ? 16 : 8;
info->save_size = RS6000_ALIGN (info->fp_size
+ info->gp_size
+ info->altivec_size
+ info->altivec_padding_size
+ ehrd_size
+ ehcr_size
+ info->cr_size
+ info->vrsave_size,
save_align);
non_fixed_size = info->vars_size + info->parm_size + info->save_size;
info->total_size = RS6000_ALIGN (non_fixed_size + info->fixed_size,
ABI_STACK_BOUNDARY / BITS_PER_UNIT);
/* Determine if we need to save the link register. */
if (info->calls_p
|| ((DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
&& crtl->profile
&& !TARGET_PROFILE_KERNEL)
|| (DEFAULT_ABI == ABI_V4 && cfun->calls_alloca)
#ifdef TARGET_RELOCATABLE
|| (DEFAULT_ABI == ABI_V4
&& (TARGET_RELOCATABLE || flag_pic > 1)
&& !constant_pool_empty_p ())
#endif
|| rs6000_ra_ever_killed ())
info->lr_save_p = 1;
using_static_chain_p = (cfun->static_chain_decl != NULL_TREE
&& df_regs_ever_live_p (STATIC_CHAIN_REGNUM)
&& call_used_regs[STATIC_CHAIN_REGNUM]);
info->savres_strategy = rs6000_savres_strategy (info, using_static_chain_p);
if (!(info->savres_strategy & SAVE_INLINE_GPRS)
|| !(info->savres_strategy & SAVE_INLINE_FPRS)
|| !(info->savres_strategy & SAVE_INLINE_VRS)
|| !(info->savres_strategy & REST_INLINE_GPRS)
|| !(info->savres_strategy & REST_INLINE_FPRS)
|| !(info->savres_strategy & REST_INLINE_VRS))
info->lr_save_p = 1;
if (info->lr_save_p)
df_set_regs_ever_live (LR_REGNO, true);
/* Determine if we need to allocate any stack frame:
For AIX we need to push the stack if a frame pointer is needed
(because the stack might be dynamically adjusted), if we are
debugging, if we make calls, or if the sum of fp_save, gp_save,
and local variables are more than the space needed to save all
non-volatile registers: 32-bit: 18*8 + 19*4 = 220 or 64-bit: 18*8
+ 18*8 = 288 (GPR13 reserved).
For V.4 we don't have the stack cushion that AIX uses, but assume
that the debugger can handle stackless frames. */
if (info->calls_p)
info->push_p = 1;
else if (DEFAULT_ABI == ABI_V4)
info->push_p = non_fixed_size != 0;
else if (frame_pointer_needed)
info->push_p = 1;
else if (TARGET_XCOFF && write_symbols != NO_DEBUG)
info->push_p = 1;
else
info->push_p = non_fixed_size > (TARGET_32BIT ? 220 : 288);
return info;
}
static void
debug_stack_info (rs6000_stack_t *info)
{
const char *abi_string;
if (! info)
info = rs6000_stack_info ();
fprintf (stderr, "\nStack information for function %s:\n",
((current_function_decl && DECL_NAME (current_function_decl))
? IDENTIFIER_POINTER (DECL_NAME (current_function_decl))
: "<unknown>"));
switch (info->abi)
{
default: abi_string = "Unknown"; break;
case ABI_NONE: abi_string = "NONE"; break;
case ABI_AIX: abi_string = "AIX"; break;
case ABI_ELFv2: abi_string = "ELFv2"; break;
case ABI_DARWIN: abi_string = "Darwin"; break;
case ABI_V4: abi_string = "V.4"; break;
}
fprintf (stderr, "\tABI = %5s\n", abi_string);
if (TARGET_ALTIVEC_ABI)
fprintf (stderr, "\tALTIVEC ABI extensions enabled.\n");
if (info->first_gp_reg_save != 32)
fprintf (stderr, "\tfirst_gp_reg_save = %5d\n", info->first_gp_reg_save);
if (info->first_fp_reg_save != 64)
fprintf (stderr, "\tfirst_fp_reg_save = %5d\n", info->first_fp_reg_save);
if (info->first_altivec_reg_save <= LAST_ALTIVEC_REGNO)
fprintf (stderr, "\tfirst_altivec_reg_save = %5d\n",
info->first_altivec_reg_save);
if (info->lr_save_p)
fprintf (stderr, "\tlr_save_p = %5d\n", info->lr_save_p);
if (info->cr_save_p)
fprintf (stderr, "\tcr_save_p = %5d\n", info->cr_save_p);
if (info->vrsave_mask)
fprintf (stderr, "\tvrsave_mask = 0x%x\n", info->vrsave_mask);
if (info->push_p)
fprintf (stderr, "\tpush_p = %5d\n", info->push_p);
if (info->calls_p)
fprintf (stderr, "\tcalls_p = %5d\n", info->calls_p);
if (info->gp_size)
fprintf (stderr, "\tgp_save_offset = %5d\n", info->gp_save_offset);
if (info->fp_size)
fprintf (stderr, "\tfp_save_offset = %5d\n", info->fp_save_offset);
if (info->altivec_size)
fprintf (stderr, "\taltivec_save_offset = %5d\n",
info->altivec_save_offset);
if (info->vrsave_size)
fprintf (stderr, "\tvrsave_save_offset = %5d\n",
info->vrsave_save_offset);
if (info->lr_save_p)
fprintf (stderr, "\tlr_save_offset = %5d\n", info->lr_save_offset);
if (info->cr_save_p)
fprintf (stderr, "\tcr_save_offset = %5d\n", info->cr_save_offset);
if (info->varargs_save_offset)
fprintf (stderr, "\tvarargs_save_offset = %5d\n", info->varargs_save_offset);
if (info->total_size)
fprintf (stderr, "\ttotal_size = " HOST_WIDE_INT_PRINT_DEC"\n",
info->total_size);
if (info->vars_size)
fprintf (stderr, "\tvars_size = " HOST_WIDE_INT_PRINT_DEC"\n",
info->vars_size);
if (info->parm_size)
fprintf (stderr, "\tparm_size = %5d\n", info->parm_size);
if (info->fixed_size)
fprintf (stderr, "\tfixed_size = %5d\n", info->fixed_size);
if (info->gp_size)
fprintf (stderr, "\tgp_size = %5d\n", info->gp_size);
if (info->fp_size)
fprintf (stderr, "\tfp_size = %5d\n", info->fp_size);
if (info->altivec_size)
fprintf (stderr, "\taltivec_size = %5d\n", info->altivec_size);
if (info->vrsave_size)
fprintf (stderr, "\tvrsave_size = %5d\n", info->vrsave_size);
if (info->altivec_padding_size)
fprintf (stderr, "\taltivec_padding_size= %5d\n",
info->altivec_padding_size);
if (info->cr_size)
fprintf (stderr, "\tcr_size = %5d\n", info->cr_size);
if (info->save_size)
fprintf (stderr, "\tsave_size = %5d\n", info->save_size);
if (info->reg_size != 4)
fprintf (stderr, "\treg_size = %5d\n", info->reg_size);
fprintf (stderr, "\tsave-strategy = %04x\n", info->savres_strategy);
fprintf (stderr, "\n");
}
rtx
rs6000_return_addr (int count, rtx frame)
{
/* We can't use get_hard_reg_initial_val for LR when count == 0 if LR
is trashed by the prologue, as it is for PIC on ABI_V4 and Darwin. */
if (count != 0
|| ((DEFAULT_ABI == ABI_V4 || DEFAULT_ABI == ABI_DARWIN) && flag_pic))
{
cfun->machine->ra_needs_full_frame = 1;
if (count == 0)
/* FRAME is set to frame_pointer_rtx by the generic code, but that
is good for loading 0(r1) only when !FRAME_GROWS_DOWNWARD. */
frame = stack_pointer_rtx;
rtx prev_frame_addr = memory_address (Pmode, frame);
rtx prev_frame = copy_to_reg (gen_rtx_MEM (Pmode, prev_frame_addr));
rtx lr_save_off = plus_constant (Pmode,
prev_frame, RETURN_ADDRESS_OFFSET);
rtx lr_save_addr = memory_address (Pmode, lr_save_off);
return gen_rtx_MEM (Pmode, lr_save_addr);
}
cfun->machine->ra_need_lr = 1;
return get_hard_reg_initial_val (Pmode, LR_REGNO);
}
/* Say whether a function is a candidate for sibcall handling or not. */
static bool
rs6000_function_ok_for_sibcall (tree decl, tree exp)
{
tree fntype;
if (decl)
fntype = TREE_TYPE (decl);
else
fntype = TREE_TYPE (TREE_TYPE (CALL_EXPR_FN (exp)));
/* We can't do it if the called function has more vector parameters
than the current function; there's nowhere to put the VRsave code. */
if (TARGET_ALTIVEC_ABI
&& TARGET_ALTIVEC_VRSAVE
&& !(decl && decl == current_function_decl))
{
function_args_iterator args_iter;
tree type;
int nvreg = 0;
/* Functions with vector parameters are required to have a
prototype, so the argument type info must be available
here. */
FOREACH_FUNCTION_ARGS(fntype, type, args_iter)
if (TREE_CODE (type) == VECTOR_TYPE
&& ALTIVEC_OR_VSX_VECTOR_MODE (TYPE_MODE (type)))
nvreg++;
FOREACH_FUNCTION_ARGS(TREE_TYPE (current_function_decl), type, args_iter)
if (TREE_CODE (type) == VECTOR_TYPE
&& ALTIVEC_OR_VSX_VECTOR_MODE (TYPE_MODE (type)))
nvreg--;
if (nvreg > 0)
return false;
}
/* Under the AIX or ELFv2 ABIs we can't allow calls to non-local
functions, because the callee may have a different TOC pointer to
the caller and there's no way to ensure we restore the TOC when
we return. With the secure-plt SYSV ABI we can't make non-local
calls when -fpic/PIC because the plt call stubs use r30. */
if (DEFAULT_ABI == ABI_DARWIN
|| ((DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
&& decl
&& !DECL_EXTERNAL (decl)
&& !DECL_WEAK (decl)
&& (*targetm.binds_local_p) (decl))
|| (DEFAULT_ABI == ABI_V4
&& (!TARGET_SECURE_PLT
|| !flag_pic
|| (decl
&& (*targetm.binds_local_p) (decl)))))
{
tree attr_list = TYPE_ATTRIBUTES (fntype);
if (!lookup_attribute ("longcall", attr_list)
|| lookup_attribute ("shortcall", attr_list))
return true;
}
return false;
}
static int
rs6000_ra_ever_killed (void)
{
rtx_insn *top;
rtx reg;
rtx_insn *insn;
if (cfun->is_thunk)
return 0;
if (cfun->machine->lr_save_state)
return cfun->machine->lr_save_state - 1;
/* regs_ever_live has LR marked as used if any sibcalls are present,
but this should not force saving and restoring in the
pro/epilogue. Likewise, reg_set_between_p thinks a sibcall
clobbers LR, so that is inappropriate. */
/* Also, the prologue can generate a store into LR that
doesn't really count, like this:
move LR->R0
bcl to set PIC register
move LR->R31
move R0->LR
When we're called from the epilogue, we need to avoid counting
this as a store. */
push_topmost_sequence ();
top = get_insns ();
pop_topmost_sequence ();
reg = gen_rtx_REG (Pmode, LR_REGNO);
for (insn = NEXT_INSN (top); insn != NULL_RTX; insn = NEXT_INSN (insn))
{
if (INSN_P (insn))
{
if (CALL_P (insn))
{
if (!SIBLING_CALL_P (insn))
return 1;
}
else if (find_regno_note (insn, REG_INC, LR_REGNO))
return 1;
else if (set_of (reg, insn) != NULL_RTX
&& !prologue_epilogue_contains (insn))
return 1;
}
}
return 0;
}
/* Emit instructions needed to load the TOC register.
This is only needed when TARGET_TOC, TARGET_MINIMAL_TOC, and there is
a constant pool; or for SVR4 -fpic. */
void
rs6000_emit_load_toc_table (int fromprolog)
{
rtx dest;
dest = gen_rtx_REG (Pmode, RS6000_PIC_OFFSET_TABLE_REGNUM);
if (TARGET_ELF && TARGET_SECURE_PLT && DEFAULT_ABI == ABI_V4 && flag_pic)
{
char buf[30];
rtx lab, tmp1, tmp2, got;
lab = gen_label_rtx ();
ASM_GENERATE_INTERNAL_LABEL (buf, "L", CODE_LABEL_NUMBER (lab));
lab = gen_rtx_SYMBOL_REF (Pmode, ggc_strdup (buf));
if (flag_pic == 2)
{
got = gen_rtx_SYMBOL_REF (Pmode, ggc_strdup (toc_label_name));
need_toc_init = 1;
}
else
got = rs6000_got_sym ();
tmp1 = tmp2 = dest;
if (!fromprolog)
{
tmp1 = gen_reg_rtx (Pmode);
tmp2 = gen_reg_rtx (Pmode);
}
emit_insn (gen_load_toc_v4_PIC_1 (lab));
emit_move_insn (tmp1, gen_rtx_REG (Pmode, LR_REGNO));
emit_insn (gen_load_toc_v4_PIC_3b (tmp2, tmp1, got, lab));
emit_insn (gen_load_toc_v4_PIC_3c (dest, tmp2, got, lab));
}
else if (TARGET_ELF && DEFAULT_ABI == ABI_V4 && flag_pic == 1)
{
emit_insn (gen_load_toc_v4_pic_si ());
emit_move_insn (dest, gen_rtx_REG (Pmode, LR_REGNO));
}
else if (TARGET_ELF && DEFAULT_ABI == ABI_V4 && flag_pic == 2)
{
char buf[30];
rtx temp0 = (fromprolog
? gen_rtx_REG (Pmode, 0)
: gen_reg_rtx (Pmode));
if (fromprolog)
{
rtx symF, symL;
ASM_GENERATE_INTERNAL_LABEL (buf, "LCF", rs6000_pic_labelno);
symF = gen_rtx_SYMBOL_REF (Pmode, ggc_strdup (buf));
ASM_GENERATE_INTERNAL_LABEL (buf, "LCL", rs6000_pic_labelno);
symL = gen_rtx_SYMBOL_REF (Pmode, ggc_strdup (buf));
emit_insn (gen_load_toc_v4_PIC_1 (symF));
emit_move_insn (dest, gen_rtx_REG (Pmode, LR_REGNO));
emit_insn (gen_load_toc_v4_PIC_2 (temp0, dest, symL, symF));
}
else
{
rtx tocsym, lab;
tocsym = gen_rtx_SYMBOL_REF (Pmode, ggc_strdup (toc_label_name));
need_toc_init = 1;
lab = gen_label_rtx ();
emit_insn (gen_load_toc_v4_PIC_1b (tocsym, lab));
emit_move_insn (dest, gen_rtx_REG (Pmode, LR_REGNO));
if (TARGET_LINK_STACK)
emit_insn (gen_addsi3 (dest, dest, GEN_INT (4)));
emit_move_insn (temp0, gen_rtx_MEM (Pmode, dest));
}
emit_insn (gen_addsi3 (dest, temp0, dest));
}
else if (TARGET_ELF && !TARGET_AIX && flag_pic == 0 && TARGET_MINIMAL_TOC)
{
/* This is for AIX code running in non-PIC ELF32. */
rtx realsym = gen_rtx_SYMBOL_REF (Pmode, ggc_strdup (toc_label_name));
need_toc_init = 1;
emit_insn (gen_elf_high (dest, realsym));
emit_insn (gen_elf_low (dest, dest, realsym));
}
else
{
gcc_assert (DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2);
if (TARGET_32BIT)
emit_insn (gen_load_toc_aix_si (dest));
else
emit_insn (gen_load_toc_aix_di (dest));
}
}
/* Emit instructions to restore the link register after determining where
its value has been stored. */
void
rs6000_emit_eh_reg_restore (rtx source, rtx scratch)
{
rs6000_stack_t *info = rs6000_stack_info ();
rtx operands[2];
operands[0] = source;
operands[1] = scratch;
if (info->lr_save_p)
{
rtx frame_rtx = stack_pointer_rtx;
HOST_WIDE_INT sp_offset = 0;
rtx tmp;
if (frame_pointer_needed
|| cfun->calls_alloca
|| info->total_size > 32767)
{
tmp = gen_frame_mem (Pmode, frame_rtx);
emit_move_insn (operands[1], tmp);
frame_rtx = operands[1];
}
else if (info->push_p)
sp_offset = info->total_size;
tmp = plus_constant (Pmode, frame_rtx,
info->lr_save_offset + sp_offset);
tmp = gen_frame_mem (Pmode, tmp);
emit_move_insn (tmp, operands[0]);
}
else
emit_move_insn (gen_rtx_REG (Pmode, LR_REGNO), operands[0]);
/* Freeze lr_save_p. We've just emitted rtl that depends on the
state of lr_save_p so any change from here on would be a bug. In
particular, stop rs6000_ra_ever_killed from considering the SET
of lr we may have added just above. */
cfun->machine->lr_save_state = info->lr_save_p + 1;
}
static GTY(()) alias_set_type set = -1;
alias_set_type
get_TOC_alias_set (void)
{
if (set == -1)
set = new_alias_set ();
return set;
}
/* This returns nonzero if the current function uses the TOC. This is
determined by the presence of (use (unspec ... UNSPEC_TOC)), which
is generated by the ABI_V4 load_toc_* patterns.
Return 2 instead of 1 if the load_toc_* pattern is in the function
partition that doesn't start the function. */
#if TARGET_ELF
static int
uses_TOC (void)
{
rtx_insn *insn;
int ret = 1;
for (insn = get_insns (); insn; insn = NEXT_INSN (insn))
{
if (INSN_P (insn))
{
rtx pat = PATTERN (insn);
int i;
if (GET_CODE (pat) == PARALLEL)
for (i = 0; i < XVECLEN (pat, 0); i++)
{
rtx sub = XVECEXP (pat, 0, i);
if (GET_CODE (sub) == USE)
{
sub = XEXP (sub, 0);
if (GET_CODE (sub) == UNSPEC
&& XINT (sub, 1) == UNSPEC_TOC)
return ret;
}
}
}
else if (crtl->has_bb_partition
&& NOTE_P (insn)
&& NOTE_KIND (insn) == NOTE_INSN_SWITCH_TEXT_SECTIONS)
ret = 2;
}
return 0;
}
#endif
rtx
create_TOC_reference (rtx symbol, rtx largetoc_reg)
{
rtx tocrel, tocreg, hi;
if (TARGET_DEBUG_ADDR)
{
if (GET_CODE (symbol) == SYMBOL_REF)
fprintf (stderr, "\ncreate_TOC_reference, (symbol_ref %s)\n",
XSTR (symbol, 0));
else
{
fprintf (stderr, "\ncreate_TOC_reference, code %s:\n",
GET_RTX_NAME (GET_CODE (symbol)));
debug_rtx (symbol);
}
}
if (!can_create_pseudo_p ())
df_set_regs_ever_live (TOC_REGISTER, true);
tocreg = gen_rtx_REG (Pmode, TOC_REGISTER);
tocrel = gen_rtx_UNSPEC (Pmode, gen_rtvec (2, symbol, tocreg), UNSPEC_TOCREL);
if (TARGET_CMODEL == CMODEL_SMALL || can_create_pseudo_p ())
return tocrel;
hi = gen_rtx_HIGH (Pmode, copy_rtx (tocrel));
if (largetoc_reg != NULL)
{
emit_move_insn (largetoc_reg, hi);
hi = largetoc_reg;
}
return gen_rtx_LO_SUM (Pmode, hi, tocrel);
}
/* Issue assembly directives that create a reference to the given DWARF
FRAME_TABLE_LABEL from the current function section. */
void
rs6000_aix_asm_output_dwarf_table_ref (char * frame_table_label)
{
fprintf (asm_out_file, "\t.ref %s\n",
(* targetm.strip_name_encoding) (frame_table_label));
}
/* This ties together stack memory (MEM with an alias set of frame_alias_set)
and the change to the stack pointer. */
static void
rs6000_emit_stack_tie (rtx fp, bool hard_frame_needed)
{
rtvec p;
int i;
rtx regs[3];
i = 0;
regs[i++] = gen_rtx_REG (Pmode, STACK_POINTER_REGNUM);
if (hard_frame_needed)
regs[i++] = gen_rtx_REG (Pmode, HARD_FRAME_POINTER_REGNUM);
if (!(REGNO (fp) == STACK_POINTER_REGNUM
|| (hard_frame_needed
&& REGNO (fp) == HARD_FRAME_POINTER_REGNUM)))
regs[i++] = fp;
p = rtvec_alloc (i);
while (--i >= 0)
{
rtx mem = gen_frame_mem (BLKmode, regs[i]);
RTVEC_ELT (p, i) = gen_rtx_SET (mem, const0_rtx);
}
emit_insn (gen_stack_tie (gen_rtx_PARALLEL (VOIDmode, p)));
}
/* Allocate SIZE_INT bytes on the stack using a store with update style insn
and set the appropriate attributes for the generated insn. Return the
first insn which adjusts the stack pointer or the last insn before
the stack adjustment loop.
SIZE_INT is used to create the CFI note for the allocation.
SIZE_RTX is an rtx containing the size of the adjustment. Note that
since stacks grow to lower addresses its runtime value is -SIZE_INT.
ORIG_SP contains the backchain value that must be stored at *sp. */
static rtx_insn *
rs6000_emit_allocate_stack_1 (HOST_WIDE_INT size_int, rtx orig_sp)
{
rtx_insn *insn;
rtx size_rtx = GEN_INT (-size_int);
if (size_int > 32767)
{
rtx tmp_reg = gen_rtx_REG (Pmode, 0);
/* Need a note here so that try_split doesn't get confused. */
if (get_last_insn () == NULL_RTX)
emit_note (NOTE_INSN_DELETED);
insn = emit_move_insn (tmp_reg, size_rtx);
try_split (PATTERN (insn), insn, 0);
size_rtx = tmp_reg;
}
if (Pmode == SImode)
insn = emit_insn (gen_movsi_update_stack (stack_pointer_rtx,
stack_pointer_rtx,
size_rtx,
orig_sp));
else
insn = emit_insn (gen_movdi_di_update_stack (stack_pointer_rtx,
stack_pointer_rtx,
size_rtx,
orig_sp));
rtx par = PATTERN (insn);
gcc_assert (GET_CODE (par) == PARALLEL);
rtx set = XVECEXP (par, 0, 0);
gcc_assert (GET_CODE (set) == SET);
rtx mem = SET_DEST (set);
gcc_assert (MEM_P (mem));
MEM_NOTRAP_P (mem) = 1;
set_mem_alias_set (mem, get_frame_alias_set ());
RTX_FRAME_RELATED_P (insn) = 1;
add_reg_note (insn, REG_FRAME_RELATED_EXPR,
gen_rtx_SET (stack_pointer_rtx,
gen_rtx_PLUS (Pmode,
stack_pointer_rtx,
GEN_INT (-size_int))));
/* Emit a blockage to ensure the allocation/probing insns are
not optimized, combined, removed, etc. Add REG_STACK_CHECK
note for similar reasons. */
if (flag_stack_clash_protection)
{
add_reg_note (insn, REG_STACK_CHECK, const0_rtx);
emit_insn (gen_blockage ());
}
return insn;
}
static HOST_WIDE_INT
get_stack_clash_protection_probe_interval (void)
{
return (HOST_WIDE_INT_1U
<< PARAM_VALUE (PARAM_STACK_CLASH_PROTECTION_PROBE_INTERVAL));
}
static HOST_WIDE_INT
get_stack_clash_protection_guard_size (void)
{
return (HOST_WIDE_INT_1U
<< PARAM_VALUE (PARAM_STACK_CLASH_PROTECTION_GUARD_SIZE));
}
/* Allocate ORIG_SIZE bytes on the stack and probe the newly
allocated space every STACK_CLASH_PROTECTION_PROBE_INTERVAL bytes.
COPY_REG, if non-null, should contain a copy of the original
stack pointer at exit from this function.
This is subtly different than the Ada probing in that it tries hard to
prevent attacks that jump the stack guard. Thus it is never allowed to
allocate more than STACK_CLASH_PROTECTION_PROBE_INTERVAL bytes of stack
space without a suitable probe. */
static rtx_insn *
rs6000_emit_probe_stack_range_stack_clash (HOST_WIDE_INT orig_size,
rtx copy_reg)
{
rtx orig_sp = copy_reg;
HOST_WIDE_INT probe_interval = get_stack_clash_protection_probe_interval ();
/* Round the size down to a multiple of PROBE_INTERVAL. */
HOST_WIDE_INT rounded_size = ROUND_DOWN (orig_size, probe_interval);
/* If explicitly requested,
or the rounded size is not the same as the original size
or the the rounded size is greater than a page,
then we will need a copy of the original stack pointer. */
if (rounded_size != orig_size
|| rounded_size > probe_interval
|| copy_reg)
{
/* If the caller did not request a copy of the incoming stack
pointer, then we use r0 to hold the copy. */
if (!copy_reg)
orig_sp = gen_rtx_REG (Pmode, 0);
emit_move_insn (orig_sp, stack_pointer_rtx);
}
/* There's three cases here.
One is a single probe which is the most common and most efficiently
implemented as it does not have to have a copy of the original
stack pointer if there are no residuals.
Second is unrolled allocation/probes which we use if there's just
a few of them. It needs to save the original stack pointer into a
temporary for use as a source register in the allocation/probe.
Last is a loop. This is the most uncommon case and least efficient. */
rtx_insn *retval = NULL;
if (rounded_size == probe_interval)
{
retval = rs6000_emit_allocate_stack_1 (probe_interval, stack_pointer_rtx);
dump_stack_clash_frame_info (PROBE_INLINE, rounded_size != orig_size);
}
else if (rounded_size <= 8 * probe_interval)
{
/* The ABI requires using the store with update insns to allocate
space and store the backchain into the stack
So we save the current stack pointer into a temporary, then
emit the store-with-update insns to store the saved stack pointer
into the right location in each new page. */
for (int i = 0; i < rounded_size; i += probe_interval)
{
rtx_insn *insn
= rs6000_emit_allocate_stack_1 (probe_interval, orig_sp);
/* Save the first stack adjustment in RETVAL. */
if (i == 0)
retval = insn;
}
dump_stack_clash_frame_info (PROBE_INLINE, rounded_size != orig_size);
}
else
{
/* Compute the ending address. */
rtx end_addr
= copy_reg ? gen_rtx_REG (Pmode, 0) : gen_rtx_REG (Pmode, 12);
rtx rs = GEN_INT (-rounded_size);
rtx_insn *insn;
if (add_operand (rs, Pmode))
insn = emit_insn (gen_add3_insn (end_addr, stack_pointer_rtx, rs));
else
{
emit_move_insn (end_addr, GEN_INT (-rounded_size));
insn = emit_insn (gen_add3_insn (end_addr, end_addr,
stack_pointer_rtx));
/* Describe the effect of INSN to the CFI engine. */
add_reg_note (insn, REG_FRAME_RELATED_EXPR,
gen_rtx_SET (end_addr,
gen_rtx_PLUS (Pmode, stack_pointer_rtx,
rs)));
}
RTX_FRAME_RELATED_P (insn) = 1;
/* Emit the loop. */
if (TARGET_64BIT)
retval = emit_insn (gen_probe_stack_rangedi (stack_pointer_rtx,
stack_pointer_rtx, orig_sp,
end_addr));
else
retval = emit_insn (gen_probe_stack_rangesi (stack_pointer_rtx,
stack_pointer_rtx, orig_sp,
end_addr));
RTX_FRAME_RELATED_P (retval) = 1;
/* Describe the effect of INSN to the CFI engine. */
add_reg_note (retval, REG_FRAME_RELATED_EXPR,
gen_rtx_SET (stack_pointer_rtx, end_addr));
/* Emit a blockage to ensure the allocation/probing insns are
not optimized, combined, removed, etc. Other cases handle this
within their call to rs6000_emit_allocate_stack_1. */
emit_insn (gen_blockage ());
dump_stack_clash_frame_info (PROBE_LOOP, rounded_size != orig_size);
}
if (orig_size != rounded_size)
{
/* Allocate (and implicitly probe) any residual space. */
HOST_WIDE_INT residual = orig_size - rounded_size;
rtx_insn *insn = rs6000_emit_allocate_stack_1 (residual, orig_sp);
/* If the residual was the only allocation, then we can return the
allocating insn. */
if (!retval)
retval = insn;
}
return retval;
}
/* Emit the correct code for allocating stack space, as insns.
If COPY_REG, make sure a copy of the old frame is left there.
The generated code may use hard register 0 as a temporary. */
static rtx_insn *
rs6000_emit_allocate_stack (HOST_WIDE_INT size, rtx copy_reg, int copy_off)
{
rtx_insn *insn;
rtx stack_reg = gen_rtx_REG (Pmode, STACK_POINTER_REGNUM);
rtx tmp_reg = gen_rtx_REG (Pmode, 0);
rtx todec = gen_int_mode (-size, Pmode);
if (INTVAL (todec) != -size)
{
warning (0, "stack frame too large");
emit_insn (gen_trap ());
return 0;
}
if (crtl->limit_stack)
{
if (REG_P (stack_limit_rtx)
&& REGNO (stack_limit_rtx) > 1
&& REGNO (stack_limit_rtx) <= 31)
{
rtx_insn *insn
= gen_add3_insn (tmp_reg, stack_limit_rtx, GEN_INT (size));
gcc_assert (insn);
emit_insn (insn);
emit_insn (gen_cond_trap (LTU, stack_reg, tmp_reg, const0_rtx));
}
else if (GET_CODE (stack_limit_rtx) == SYMBOL_REF
&& TARGET_32BIT
&& DEFAULT_ABI == ABI_V4
&& !flag_pic)
{
rtx toload = gen_rtx_CONST (VOIDmode,
gen_rtx_PLUS (Pmode,
stack_limit_rtx,
GEN_INT (size)));
emit_insn (gen_elf_high (tmp_reg, toload));
emit_insn (gen_elf_low (tmp_reg, tmp_reg, toload));
emit_insn (gen_cond_trap (LTU, stack_reg, tmp_reg,
const0_rtx));
}
else
warning (0, "stack limit expression is not supported");
}
if (flag_stack_clash_protection)
{
if (size < get_stack_clash_protection_guard_size ())
dump_stack_clash_frame_info (NO_PROBE_SMALL_FRAME, true);
else
{
rtx_insn *insn = rs6000_emit_probe_stack_range_stack_clash (size,
copy_reg);
/* If we asked for a copy with an offset, then we still need add in
the offset. */
if (copy_reg && copy_off)
emit_insn (gen_add3_insn (copy_reg, copy_reg, GEN_INT (copy_off)));
return insn;
}
}
if (copy_reg)
{
if (copy_off != 0)
emit_insn (gen_add3_insn (copy_reg, stack_reg, GEN_INT (copy_off)));
else
emit_move_insn (copy_reg, stack_reg);
}
/* Since we didn't use gen_frame_mem to generate the MEM, grab
it now and set the alias set/attributes. The above gen_*_update
calls will generate a PARALLEL with the MEM set being the first
operation. */
insn = rs6000_emit_allocate_stack_1 (size, stack_reg);
return insn;
}
#define PROBE_INTERVAL (1 << STACK_CHECK_PROBE_INTERVAL_EXP)
#if PROBE_INTERVAL > 32768
#error Cannot use indexed addressing mode for stack probing
#endif
/* Emit code to probe a range of stack addresses from FIRST to FIRST+SIZE,
inclusive. These are offsets from the current stack pointer. */
static void
rs6000_emit_probe_stack_range (HOST_WIDE_INT first, HOST_WIDE_INT size)
{
/* See if we have a constant small number of probes to generate. If so,
that's the easy case. */
if (first + size <= 32768)
{
HOST_WIDE_INT i;
/* Probe at FIRST + N * PROBE_INTERVAL for values of N from 1 until
it exceeds SIZE. If only one probe is needed, this will not
generate any code. Then probe at FIRST + SIZE. */
for (i = PROBE_INTERVAL; i < size; i += PROBE_INTERVAL)
emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx,
-(first + i)));
emit_stack_probe (plus_constant (Pmode, stack_pointer_rtx,
-(first + size)));
}
/* Otherwise, do the same as above, but in a loop. Note that we must be
extra careful with variables wrapping around because we might be at
the very top (or the very bottom) of the address space and we have
to be able to handle this case properly; in particular, we use an
equality test for the loop condition. */
else
{
HOST_WIDE_INT rounded_size;
rtx r12 = gen_rtx_REG (Pmode, 12);
rtx r0 = gen_rtx_REG (Pmode, 0);
/* Sanity check for the addressing mode we're going to use. */
gcc_assert (first <= 32768);
/* Step 1: round SIZE to the previous multiple of the interval. */
rounded_size = ROUND_DOWN (size, PROBE_INTERVAL);
/* Step 2: compute initial and final value of the loop counter. */
/* TEST_ADDR = SP + FIRST. */
emit_insn (gen_rtx_SET (r12, plus_constant (Pmode, stack_pointer_rtx,
-first)));
/* LAST_ADDR = SP + FIRST + ROUNDED_SIZE. */
if (rounded_size > 32768)
{
emit_move_insn (r0, GEN_INT (-rounded_size));
emit_insn (gen_rtx_SET (r0, gen_rtx_PLUS (Pmode, r12, r0)));
}
else
emit_insn (gen_rtx_SET (r0, plus_constant (Pmode, r12,
-rounded_size)));
/* Step 3: the loop
do
{
TEST_ADDR = TEST_ADDR + PROBE_INTERVAL
probe at TEST_ADDR
}
while (TEST_ADDR != LAST_ADDR)
probes at FIRST + N * PROBE_INTERVAL for values of N from 1
until it is equal to ROUNDED_SIZE. */
if (TARGET_64BIT)
emit_insn (gen_probe_stack_rangedi (r12, r12, stack_pointer_rtx, r0));
else
emit_insn (gen_probe_stack_rangesi (r12, r12, stack_pointer_rtx, r0));
/* Step 4: probe at FIRST + SIZE if we cannot assert at compile-time
that SIZE is equal to ROUNDED_SIZE. */
if (size != rounded_size)
emit_stack_probe (plus_constant (Pmode, r12, rounded_size - size));
}
}
/* Probe a range of stack addresses from REG1 to REG2 inclusive. These are
addresses, not offsets. */
static const char *
output_probe_stack_range_1 (rtx reg1, rtx reg2)
{
static int labelno = 0;
char loop_lab[32];
rtx xops[2];
ASM_GENERATE_INTERNAL_LABEL (loop_lab, "LPSRL", labelno++);
/* Loop. */
ASM_OUTPUT_INTERNAL_LABEL (asm_out_file, loop_lab);
/* TEST_ADDR = TEST_ADDR + PROBE_INTERVAL. */
xops[0] = reg1;
xops[1] = GEN_INT (-PROBE_INTERVAL);
output_asm_insn ("addi %0,%0,%1", xops);
/* Probe at TEST_ADDR. */
xops[1] = gen_rtx_REG (Pmode, 0);
output_asm_insn ("stw %1,0(%0)", xops);
/* Test if TEST_ADDR == LAST_ADDR. */
xops[1] = reg2;
if (TARGET_64BIT)
output_asm_insn ("cmpd 0,%0,%1", xops);
else
output_asm_insn ("cmpw 0,%0,%1", xops);
/* Branch. */
fputs ("\tbne 0,", asm_out_file);
assemble_name_raw (asm_out_file, loop_lab);
fputc ('\n', asm_out_file);
return "";
}
/* This function is called when rs6000_frame_related is processing
SETs within a PARALLEL, and returns whether the REGNO save ought to
be marked RTX_FRAME_RELATED_P. The PARALLELs involved are those
for out-of-line register save functions, store multiple, and the
Darwin world_save. They may contain registers that don't really
need saving. */
static bool
interesting_frame_related_regno (unsigned int regno)
{
/* Saves apparently of r0 are actually saving LR. It doesn't make
sense to substitute the regno here to test save_reg_p (LR_REGNO).
We *know* LR needs saving, and dwarf2cfi.c is able to deduce that
(set (mem) (r0)) is saving LR from a prior (set (r0) (lr)) marked
as frame related. */
if (regno == 0)
return true;
/* If we see CR2 then we are here on a Darwin world save. Saves of
CR2 signify the whole CR is being saved. This is a long-standing
ABI wart fixed by ELFv2. As for r0/lr there is no need to check
that CR needs to be saved. */
if (regno == CR2_REGNO)
return true;
/* Omit frame info for any user-defined global regs. If frame info
is supplied for them, frame unwinding will restore a user reg.
Also omit frame info for any reg we don't need to save, as that
bloats frame info and can cause problems with shrink wrapping.
Since global regs won't be seen as needing to be saved, both of
these conditions are covered by save_reg_p. */
return save_reg_p (regno);
}
/* Probe a range of stack addresses from REG1 to REG3 inclusive. These are
addresses, not offsets.
REG2 contains the backchain that must be stored into *sp at each allocation.
This is subtly different than the Ada probing above in that it tries hard
to prevent attacks that jump the stack guard. Thus, it is never allowed
to allocate more than PROBE_INTERVAL bytes of stack space without a
suitable probe. */
static const char *
output_probe_stack_range_stack_clash (rtx reg1, rtx reg2, rtx reg3)
{
static int labelno = 0;
char loop_lab[32];
rtx xops[3];
HOST_WIDE_INT probe_interval = get_stack_clash_protection_probe_interval ();
ASM_GENERATE_INTERNAL_LABEL (loop_lab, "LPSRL", labelno++);
ASM_OUTPUT_INTERNAL_LABEL (asm_out_file, loop_lab);
/* This allocates and probes. */
xops[0] = reg1;
xops[1] = reg2;
xops[2] = GEN_INT (-probe_interval);
if (TARGET_64BIT)
output_asm_insn ("stdu %1,%2(%0)", xops);
else
output_asm_insn ("stwu %1,%2(%0)", xops);
/* Jump to LOOP_LAB if TEST_ADDR != LAST_ADDR. */
xops[0] = reg1;
xops[1] = reg3;
if (TARGET_64BIT)
output_asm_insn ("cmpd 0,%0,%1", xops);
else
output_asm_insn ("cmpw 0,%0,%1", xops);
fputs ("\tbne 0,", asm_out_file);
assemble_name_raw (asm_out_file, loop_lab);
fputc ('\n', asm_out_file);
return "";
}
/* Wrapper around the output_probe_stack_range routines. */
const char *
output_probe_stack_range (rtx reg1, rtx reg2, rtx reg3)
{
if (flag_stack_clash_protection)
return output_probe_stack_range_stack_clash (reg1, reg2, reg3);
else
return output_probe_stack_range_1 (reg1, reg3);
}
/* Add to 'insn' a note which is PATTERN (INSN) but with REG replaced
with (plus:P (reg 1) VAL), and with REG2 replaced with REPL2 if REG2
is not NULL. It would be nice if dwarf2out_frame_debug_expr could
deduce these equivalences by itself so it wasn't necessary to hold
its hand so much. Don't be tempted to always supply d2_f_d_e with
the actual cfa register, ie. r31 when we are using a hard frame
pointer. That fails when saving regs off r1, and sched moves the
r31 setup past the reg saves. */
static rtx_insn *
rs6000_frame_related (rtx_insn *insn, rtx reg, HOST_WIDE_INT val,
rtx reg2, rtx repl2)
{
rtx repl;
if (REGNO (reg) == STACK_POINTER_REGNUM)
{
gcc_checking_assert (val == 0);
repl = NULL_RTX;
}
else
repl = gen_rtx_PLUS (Pmode, gen_rtx_REG (Pmode, STACK_POINTER_REGNUM),
GEN_INT (val));
rtx pat = PATTERN (insn);
if (!repl && !reg2)
{
/* No need for any replacement. Just set RTX_FRAME_RELATED_P. */
if (GET_CODE (pat) == PARALLEL)
for (int i = 0; i < XVECLEN (pat, 0); i++)
if (GET_CODE (XVECEXP (pat, 0, i)) == SET)
{
rtx set = XVECEXP (pat, 0, i);
if (!REG_P (SET_SRC (set))
|| interesting_frame_related_regno (REGNO (SET_SRC (set))))
RTX_FRAME_RELATED_P (set) = 1;
}
RTX_FRAME_RELATED_P (insn) = 1;
return insn;
}
/* We expect that 'pat' is either a SET or a PARALLEL containing
SETs (and possibly other stuff). In a PARALLEL, all the SETs
are important so they all have to be marked RTX_FRAME_RELATED_P.
Call simplify_replace_rtx on the SETs rather than the whole insn
so as to leave the other stuff alone (for example USE of r12). */
set_used_flags (pat);
if (GET_CODE (pat) == SET)
{
if (repl)
pat = simplify_replace_rtx (pat, reg, repl);
if (reg2)
pat = simplify_replace_rtx (pat, reg2, repl2);
}
else if (GET_CODE (pat) == PARALLEL)
{
pat = shallow_copy_rtx (pat);
XVEC (pat, 0) = shallow_copy_rtvec (XVEC (pat, 0));
for (int i = 0; i < XVECLEN (pat, 0); i++)
if (GET_CODE (XVECEXP (pat, 0, i)) == SET)
{
rtx set = XVECEXP (pat, 0, i);
if (repl)
set = simplify_replace_rtx (set, reg, repl);
if (reg2)
set = simplify_replace_rtx (set, reg2, repl2);
XVECEXP (pat, 0, i) = set;
if (!REG_P (SET_SRC (set))
|| interesting_frame_related_regno (REGNO (SET_SRC (set))))
RTX_FRAME_RELATED_P (set) = 1;
}
}
else
gcc_unreachable ();
RTX_FRAME_RELATED_P (insn) = 1;
add_reg_note (insn, REG_FRAME_RELATED_EXPR, copy_rtx_if_shared (pat));
return insn;
}
/* Returns an insn that has a vrsave set operation with the
appropriate CLOBBERs. */
static rtx
generate_set_vrsave (rtx reg, rs6000_stack_t *info, int epiloguep)
{
int nclobs, i;
rtx insn, clobs[TOTAL_ALTIVEC_REGS + 1];
rtx vrsave = gen_rtx_REG (SImode, VRSAVE_REGNO);
clobs[0]
= gen_rtx_SET (vrsave,
gen_rtx_UNSPEC_VOLATILE (SImode,
gen_rtvec (2, reg, vrsave),
UNSPECV_SET_VRSAVE));
nclobs = 1;
/* We need to clobber the registers in the mask so the scheduler
does not move sets to VRSAVE before sets of AltiVec registers.
However, if the function receives nonlocal gotos, reload will set
all call saved registers live. We will end up with:
(set (reg 999) (mem))
(parallel [ (set (reg vrsave) (unspec blah))
(clobber (reg 999))])
The clobber will cause the store into reg 999 to be dead, and
flow will attempt to delete an epilogue insn. In this case, we
need an unspec use/set of the register. */
for (i = FIRST_ALTIVEC_REGNO; i <= LAST_ALTIVEC_REGNO; ++i)
if (info->vrsave_mask & ALTIVEC_REG_BIT (i))
{
if (!epiloguep || call_used_regs [i])
clobs[nclobs++] = gen_rtx_CLOBBER (VOIDmode,
gen_rtx_REG (V4SImode, i));
else
{
rtx reg = gen_rtx_REG (V4SImode, i);
clobs[nclobs++]
= gen_rtx_SET (reg,
gen_rtx_UNSPEC (V4SImode,
gen_rtvec (1, reg), 27));
}
}
insn = gen_rtx_PARALLEL (VOIDmode, rtvec_alloc (nclobs));
for (i = 0; i < nclobs; ++i)
XVECEXP (insn, 0, i) = clobs[i];
return insn;
}
static rtx
gen_frame_set (rtx reg, rtx frame_reg, int offset, bool store)
{
rtx addr, mem;
addr = gen_rtx_PLUS (Pmode, frame_reg, GEN_INT (offset));
mem = gen_frame_mem (GET_MODE (reg), addr);
return gen_rtx_SET (store ? mem : reg, store ? reg : mem);
}
static rtx
gen_frame_load (rtx reg, rtx frame_reg, int offset)
{
return gen_frame_set (reg, frame_reg, offset, false);
}
static rtx
gen_frame_store (rtx reg, rtx frame_reg, int offset)
{
return gen_frame_set (reg, frame_reg, offset, true);
}
/* Save a register into the frame, and emit RTX_FRAME_RELATED_P notes.
Save REGNO into [FRAME_REG + OFFSET] in mode MODE. */
static rtx_insn *
emit_frame_save (rtx frame_reg, machine_mode mode,
unsigned int regno, int offset, HOST_WIDE_INT frame_reg_to_sp)
{
rtx reg;
/* Some cases that need register indexed addressing. */
gcc_checking_assert (!(TARGET_ALTIVEC_ABI && ALTIVEC_VECTOR_MODE (mode))
|| (TARGET_VSX && ALTIVEC_OR_VSX_VECTOR_MODE (mode)));
reg = gen_rtx_REG (mode, regno);
rtx_insn *insn = emit_insn (gen_frame_store (reg, frame_reg, offset));
return rs6000_frame_related (insn, frame_reg, frame_reg_to_sp,
NULL_RTX, NULL_RTX);
}
/* Emit an offset memory reference suitable for a frame store, while
converting to a valid addressing mode. */
static rtx
gen_frame_mem_offset (machine_mode mode, rtx reg, int offset)
{
return gen_frame_mem (mode, gen_rtx_PLUS (Pmode, reg, GEN_INT (offset)));
}
#ifndef TARGET_FIX_AND_CONTINUE
#define TARGET_FIX_AND_CONTINUE 0
#endif
/* It's really GPR 13 or 14, FPR 14 and VR 20. We need the smallest. */
#define FIRST_SAVRES_REGISTER FIRST_SAVED_GP_REGNO
#define LAST_SAVRES_REGISTER 31
#define N_SAVRES_REGISTERS (LAST_SAVRES_REGISTER - FIRST_SAVRES_REGISTER + 1)
enum {
SAVRES_LR = 0x1,
SAVRES_SAVE = 0x2,
SAVRES_REG = 0x0c,
SAVRES_GPR = 0,
SAVRES_FPR = 4,
SAVRES_VR = 8
};
static GTY(()) rtx savres_routine_syms[N_SAVRES_REGISTERS][12];
/* Temporary holding space for an out-of-line register save/restore
routine name. */
static char savres_routine_name[30];
/* Return the name for an out-of-line register save/restore routine.
We are saving/restoring GPRs if GPR is true. */
static char *
rs6000_savres_routine_name (int regno, int sel)
{
const char *prefix = "";
const char *suffix = "";
/* Different targets are supposed to define
{SAVE,RESTORE}_FP_{PREFIX,SUFFIX} with the idea that the needed
routine name could be defined with:
sprintf (name, "%s%d%s", SAVE_FP_PREFIX, regno, SAVE_FP_SUFFIX)
This is a nice idea in practice, but in reality, things are
complicated in several ways:
- ELF targets have save/restore routines for GPRs.
- PPC64 ELF targets have routines for save/restore of GPRs that
differ in what they do with the link register, so having a set
prefix doesn't work. (We only use one of the save routines at
the moment, though.)
- PPC32 elf targets have "exit" versions of the restore routines
that restore the link register and can save some extra space.
These require an extra suffix. (There are also "tail" versions
of the restore routines and "GOT" versions of the save routines,
but we don't generate those at present. Same problems apply,
though.)
We deal with all this by synthesizing our own prefix/suffix and
using that for the simple sprintf call shown above. */
if (DEFAULT_ABI == ABI_V4)
{
if (TARGET_64BIT)
goto aix_names;
if ((sel & SAVRES_REG) == SAVRES_GPR)
prefix = (sel & SAVRES_SAVE) ? "_savegpr_" : "_restgpr_";
else if ((sel & SAVRES_REG) == SAVRES_FPR)
prefix = (sel & SAVRES_SAVE) ? "_savefpr_" : "_restfpr_";
else if ((sel & SAVRES_REG) == SAVRES_VR)
prefix = (sel & SAVRES_SAVE) ? "_savevr_" : "_restvr_";
else
abort ();
if ((sel & SAVRES_LR))
suffix = "_x";
}
else if (DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
{
#if !defined (POWERPC_LINUX) && !defined (POWERPC_FREEBSD)
/* No out-of-line save/restore routines for GPRs on AIX. */
gcc_assert (!TARGET_AIX || (sel & SAVRES_REG) != SAVRES_GPR);
#endif
aix_names:
if ((sel & SAVRES_REG) == SAVRES_GPR)
prefix = ((sel & SAVRES_SAVE)
? ((sel & SAVRES_LR) ? "_savegpr0_" : "_savegpr1_")
: ((sel & SAVRES_LR) ? "_restgpr0_" : "_restgpr1_"));
else if ((sel & SAVRES_REG) == SAVRES_FPR)
{
#if defined (POWERPC_LINUX) || defined (POWERPC_FREEBSD)
if ((sel & SAVRES_LR))
prefix = ((sel & SAVRES_SAVE) ? "_savefpr_" : "_restfpr_");
else
#endif
{
prefix = (sel & SAVRES_SAVE) ? SAVE_FP_PREFIX : RESTORE_FP_PREFIX;
suffix = (sel & SAVRES_SAVE) ? SAVE_FP_SUFFIX : RESTORE_FP_SUFFIX;
}
}
else if ((sel & SAVRES_REG) == SAVRES_VR)
prefix = (sel & SAVRES_SAVE) ? "_savevr_" : "_restvr_";
else
abort ();
}
if (DEFAULT_ABI == ABI_DARWIN)
{
/* The Darwin approach is (slightly) different, in order to be
compatible with code generated by the system toolchain. There is a
single symbol for the start of save sequence, and the code here
embeds an offset into that code on the basis of the first register
to be saved. */
prefix = (sel & SAVRES_SAVE) ? "save" : "rest" ;
if ((sel & SAVRES_REG) == SAVRES_GPR)
sprintf (savres_routine_name, "*%sGPR%s%s%.0d ; %s r%d-r31", prefix,
((sel & SAVRES_LR) ? "x" : ""), (regno == 13 ? "" : "+"),
(regno - 13) * 4, prefix, regno);
else if ((sel & SAVRES_REG) == SAVRES_FPR)
sprintf (savres_routine_name, "*%sFP%s%.0d ; %s f%d-f31", prefix,
(regno == 14 ? "" : "+"), (regno - 14) * 4, prefix, regno);
else if ((sel & SAVRES_REG) == SAVRES_VR)
sprintf (savres_routine_name, "*%sVEC%s%.0d ; %s v%d-v31", prefix,
(regno == 20 ? "" : "+"), (regno - 20) * 8, prefix, regno);
else
abort ();
}
else
sprintf (savres_routine_name, "%s%d%s", prefix, regno, suffix);
return savres_routine_name;
}
/* Return an RTL SYMBOL_REF for an out-of-line register save/restore routine.
We are saving/restoring GPRs if GPR is true. */
static rtx
rs6000_savres_routine_sym (rs6000_stack_t *info, int sel)
{
int regno = ((sel & SAVRES_REG) == SAVRES_GPR
? info->first_gp_reg_save
: (sel & SAVRES_REG) == SAVRES_FPR
? info->first_fp_reg_save - 32
: (sel & SAVRES_REG) == SAVRES_VR
? info->first_altivec_reg_save - FIRST_ALTIVEC_REGNO
: -1);
rtx sym;
int select = sel;
/* Don't generate bogus routine names. */
gcc_assert (FIRST_SAVRES_REGISTER <= regno
&& regno <= LAST_SAVRES_REGISTER
&& select >= 0 && select <= 12);
sym = savres_routine_syms[regno-FIRST_SAVRES_REGISTER][select];
if (sym == NULL)
{
char *name;
name = rs6000_savres_routine_name (regno, sel);
sym = savres_routine_syms[regno-FIRST_SAVRES_REGISTER][select]
= gen_rtx_SYMBOL_REF (Pmode, ggc_strdup (name));
SYMBOL_REF_FLAGS (sym) |= SYMBOL_FLAG_FUNCTION;
}
return sym;
}
/* Emit a sequence of insns, including a stack tie if needed, for
resetting the stack pointer. If UPDT_REGNO is not 1, then don't
reset the stack pointer, but move the base of the frame into
reg UPDT_REGNO for use by out-of-line register restore routines. */
static rtx
rs6000_emit_stack_reset (rtx frame_reg_rtx, HOST_WIDE_INT frame_off,
unsigned updt_regno)
{
/* If there is nothing to do, don't do anything. */
if (frame_off == 0 && REGNO (frame_reg_rtx) == updt_regno)
return NULL_RTX;
rtx updt_reg_rtx = gen_rtx_REG (Pmode, updt_regno);
/* This blockage is needed so that sched doesn't decide to move
the sp change before the register restores. */
if (DEFAULT_ABI == ABI_V4)
return emit_insn (gen_stack_restore_tie (updt_reg_rtx, frame_reg_rtx,
GEN_INT (frame_off)));
/* If we are restoring registers out-of-line, we will be using the
"exit" variants of the restore routines, which will reset the
stack for us. But we do need to point updt_reg into the
right place for those routines. */
if (frame_off != 0)
return emit_insn (gen_add3_insn (updt_reg_rtx,
frame_reg_rtx, GEN_INT (frame_off)));
else
return emit_move_insn (updt_reg_rtx, frame_reg_rtx);
return NULL_RTX;
}
/* Return the register number used as a pointer by out-of-line
save/restore functions. */
static inline unsigned
ptr_regno_for_savres (int sel)
{
if (DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
return (sel & SAVRES_REG) == SAVRES_FPR || (sel & SAVRES_LR) ? 1 : 12;
return DEFAULT_ABI == ABI_DARWIN && (sel & SAVRES_REG) == SAVRES_FPR ? 1 : 11;
}
/* Construct a parallel rtx describing the effect of a call to an
out-of-line register save/restore routine, and emit the insn
or jump_insn as appropriate. */
static rtx_insn *
rs6000_emit_savres_rtx (rs6000_stack_t *info,
rtx frame_reg_rtx, int save_area_offset, int lr_offset,
machine_mode reg_mode, int sel)
{
int i;
int offset, start_reg, end_reg, n_regs, use_reg;
int reg_size = GET_MODE_SIZE (reg_mode);
rtx sym;
rtvec p;
rtx par;
rtx_insn *insn;
offset = 0;
start_reg = ((sel & SAVRES_REG) == SAVRES_GPR
? info->first_gp_reg_save
: (sel & SAVRES_REG) == SAVRES_FPR
? info->first_fp_reg_save
: (sel & SAVRES_REG) == SAVRES_VR
? info->first_altivec_reg_save
: -1);
end_reg = ((sel & SAVRES_REG) == SAVRES_GPR
? 32
: (sel & SAVRES_REG) == SAVRES_FPR
? 64
: (sel & SAVRES_REG) == SAVRES_VR
? LAST_ALTIVEC_REGNO + 1
: -1);
n_regs = end_reg - start_reg;
p = rtvec_alloc (3 + ((sel & SAVRES_LR) ? 1 : 0)
+ ((sel & SAVRES_REG) == SAVRES_VR ? 1 : 0)
+ n_regs);
if (!(sel & SAVRES_SAVE) && (sel & SAVRES_LR))
RTVEC_ELT (p, offset++) = ret_rtx;
RTVEC_ELT (p, offset++)
= gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (Pmode, LR_REGNO));
sym = rs6000_savres_routine_sym (info, sel);
RTVEC_ELT (p, offset++) = gen_rtx_USE (VOIDmode, sym);
use_reg = ptr_regno_for_savres (sel);
if ((sel & SAVRES_REG) == SAVRES_VR)
{
/* Vector regs are saved/restored using [reg+reg] addressing. */
RTVEC_ELT (p, offset++)
= gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (Pmode, use_reg));
RTVEC_ELT (p, offset++)
= gen_rtx_USE (VOIDmode, gen_rtx_REG (Pmode, 0));
}
else
RTVEC_ELT (p, offset++)
= gen_rtx_USE (VOIDmode, gen_rtx_REG (Pmode, use_reg));
for (i = 0; i < end_reg - start_reg; i++)
RTVEC_ELT (p, i + offset)
= gen_frame_set (gen_rtx_REG (reg_mode, start_reg + i),
frame_reg_rtx, save_area_offset + reg_size * i,
(sel & SAVRES_SAVE) != 0);
if ((sel & SAVRES_SAVE) && (sel & SAVRES_LR))
RTVEC_ELT (p, i + offset)
= gen_frame_store (gen_rtx_REG (Pmode, 0), frame_reg_rtx, lr_offset);
par = gen_rtx_PARALLEL (VOIDmode, p);
if (!(sel & SAVRES_SAVE) && (sel & SAVRES_LR))
{
insn = emit_jump_insn (par);
JUMP_LABEL (insn) = ret_rtx;
}
else
insn = emit_insn (par);
return insn;
}
/* Emit prologue code to store CR fields that need to be saved into REG. This
function should only be called when moving the non-volatile CRs to REG, it
is not a general purpose routine to move the entire set of CRs to REG.
Specifically, gen_prologue_movesi_from_cr() does not contain uses of the
volatile CRs. */
static void
rs6000_emit_prologue_move_from_cr (rtx reg)
{
/* Only the ELFv2 ABI allows storing only selected fields. */
if (DEFAULT_ABI == ABI_ELFv2 && TARGET_MFCRF)
{
int i, cr_reg[8], count = 0;
/* Collect CR fields that must be saved. */
for (i = 0; i < 8; i++)
if (save_reg_p (CR0_REGNO + i))
cr_reg[count++] = i;
/* If it's just a single one, use mfcrf. */
if (count == 1)
{
rtvec p = rtvec_alloc (1);
rtvec r = rtvec_alloc (2);
RTVEC_ELT (r, 0) = gen_rtx_REG (CCmode, CR0_REGNO + cr_reg[0]);
RTVEC_ELT (r, 1) = GEN_INT (1 << (7 - cr_reg[0]));
RTVEC_ELT (p, 0)
= gen_rtx_SET (reg,
gen_rtx_UNSPEC (SImode, r, UNSPEC_MOVESI_FROM_CR));
emit_insn (gen_rtx_PARALLEL (VOIDmode, p));
return;
}
/* ??? It might be better to handle count == 2 / 3 cases here
as well, using logical operations to combine the values. */
}
emit_insn (gen_prologue_movesi_from_cr (reg));
}
/* Return whether the split-stack arg pointer (r12) is used. */
static bool
split_stack_arg_pointer_used_p (void)
{
/* If the pseudo holding the arg pointer is no longer a pseudo,
then the arg pointer is used. */
if (cfun->machine->split_stack_arg_pointer != NULL_RTX
&& (!REG_P (cfun->machine->split_stack_arg_pointer)
|| (REGNO (cfun->machine->split_stack_arg_pointer)
< FIRST_PSEUDO_REGISTER)))
return true;
/* Unfortunately we also need to do some code scanning, since
r12 may have been substituted for the pseudo. */
rtx_insn *insn;
basic_block bb = ENTRY_BLOCK_PTR_FOR_FN (cfun)->next_bb;
FOR_BB_INSNS (bb, insn)
if (NONDEBUG_INSN_P (insn))
{
/* A call destroys r12. */
if (CALL_P (insn))
return false;
df_ref use;
FOR_EACH_INSN_USE (use, insn)
{
rtx x = DF_REF_REG (use);
if (REG_P (x) && REGNO (x) == 12)
return true;
}
df_ref def;
FOR_EACH_INSN_DEF (def, insn)
{
rtx x = DF_REF_REG (def);
if (REG_P (x) && REGNO (x) == 12)
return false;
}
}
return bitmap_bit_p (DF_LR_OUT (bb), 12);
}
/* Return whether we need to emit an ELFv2 global entry point prologue. */
static bool
rs6000_global_entry_point_needed_p (void)
{
/* Only needed for the ELFv2 ABI. */
if (DEFAULT_ABI != ABI_ELFv2)
return false;
/* With -msingle-pic-base, we assume the whole program shares the same
TOC, so no global entry point prologues are needed anywhere. */
if (TARGET_SINGLE_PIC_BASE)
return false;
/* Ensure we have a global entry point for thunks. ??? We could
avoid that if the target routine doesn't need a global entry point,
but we do not know whether this is the case at this point. */
if (cfun->is_thunk)
return true;
/* For regular functions, rs6000_emit_prologue sets this flag if the
routine ever uses the TOC pointer. */
return cfun->machine->r2_setup_needed;
}
/* Implement TARGET_SHRINK_WRAP_GET_SEPARATE_COMPONENTS. */
static sbitmap
rs6000_get_separate_components (void)
{
rs6000_stack_t *info = rs6000_stack_info ();
if (WORLD_SAVE_P (info))
return NULL;
gcc_assert (!(info->savres_strategy & SAVE_MULTIPLE)
&& !(info->savres_strategy & REST_MULTIPLE));
/* Component 0 is the save/restore of LR (done via GPR0).
Component 2 is the save of the TOC (GPR2).
Components 13..31 are the save/restore of GPR13..GPR31.
Components 46..63 are the save/restore of FPR14..FPR31. */
cfun->machine->n_components = 64;
sbitmap components = sbitmap_alloc (cfun->machine->n_components);
bitmap_clear (components);
int reg_size = TARGET_32BIT ? 4 : 8;
int fp_reg_size = 8;
/* The GPRs we need saved to the frame. */
if ((info->savres_strategy & SAVE_INLINE_GPRS)
&& (info->savres_strategy & REST_INLINE_GPRS))
{
int offset = info->gp_save_offset;
if (info->push_p)
offset += info->total_size;
for (unsigned regno = info->first_gp_reg_save; regno < 32; regno++)
{
if (IN_RANGE (offset, -0x8000, 0x7fff)
&& save_reg_p (regno))
bitmap_set_bit (components, regno);
offset += reg_size;
}
}
/* Don't mess with the hard frame pointer. */
if (frame_pointer_needed)
bitmap_clear_bit (components, HARD_FRAME_POINTER_REGNUM);
/* Don't mess with the fixed TOC register. */
if ((TARGET_TOC && TARGET_MINIMAL_TOC)
|| (flag_pic == 1 && DEFAULT_ABI == ABI_V4)
|| (flag_pic && DEFAULT_ABI == ABI_DARWIN))
bitmap_clear_bit (components, RS6000_PIC_OFFSET_TABLE_REGNUM);
/* The FPRs we need saved to the frame. */
if ((info->savres_strategy & SAVE_INLINE_FPRS)
&& (info->savres_strategy & REST_INLINE_FPRS))
{
int offset = info->fp_save_offset;
if (info->push_p)
offset += info->total_size;
for (unsigned regno = info->first_fp_reg_save; regno < 64; regno++)
{
if (IN_RANGE (offset, -0x8000, 0x7fff) && save_reg_p (regno))
bitmap_set_bit (components, regno);
offset += fp_reg_size;
}
}
/* Optimize LR save and restore if we can. This is component 0. Any
out-of-line register save/restore routines need LR. */
if (info->lr_save_p
&& !(flag_pic && (DEFAULT_ABI == ABI_V4 || DEFAULT_ABI == ABI_DARWIN))
&& (info->savres_strategy & SAVE_INLINE_GPRS)
&& (info->savres_strategy & REST_INLINE_GPRS)
&& (info->savres_strategy & SAVE_INLINE_FPRS)
&& (info->savres_strategy & REST_INLINE_FPRS)
&& (info->savres_strategy & SAVE_INLINE_VRS)
&& (info->savres_strategy & REST_INLINE_VRS))
{
int offset = info->lr_save_offset;
if (info->push_p)
offset += info->total_size;
if (IN_RANGE (offset, -0x8000, 0x7fff))
bitmap_set_bit (components, 0);
}
/* Optimize saving the TOC. This is component 2. */
if (cfun->machine->save_toc_in_prologue)
bitmap_set_bit (components, 2);
return components;
}
/* Implement TARGET_SHRINK_WRAP_COMPONENTS_FOR_BB. */
static sbitmap
rs6000_components_for_bb (basic_block bb)
{
rs6000_stack_t *info = rs6000_stack_info ();
bitmap in = DF_LIVE_IN (bb);
bitmap gen = &DF_LIVE_BB_INFO (bb)->gen;
bitmap kill = &DF_LIVE_BB_INFO (bb)->kill;
sbitmap components = sbitmap_alloc (cfun->machine->n_components);
bitmap_clear (components);
/* A register is used in a bb if it is in the IN, GEN, or KILL sets. */
/* GPRs. */
for (unsigned regno = info->first_gp_reg_save; regno < 32; regno++)
if (bitmap_bit_p (in, regno)
|| bitmap_bit_p (gen, regno)
|| bitmap_bit_p (kill, regno))
bitmap_set_bit (components, regno);
/* FPRs. */
for (unsigned regno = info->first_fp_reg_save; regno < 64; regno++)
if (bitmap_bit_p (in, regno)
|| bitmap_bit_p (gen, regno)
|| bitmap_bit_p (kill, regno))
bitmap_set_bit (components, regno);
/* The link register. */
if (bitmap_bit_p (in, LR_REGNO)
|| bitmap_bit_p (gen, LR_REGNO)
|| bitmap_bit_p (kill, LR_REGNO))
bitmap_set_bit (components, 0);
/* The TOC save. */
if (bitmap_bit_p (in, TOC_REGNUM)
|| bitmap_bit_p (gen, TOC_REGNUM)
|| bitmap_bit_p (kill, TOC_REGNUM))
bitmap_set_bit (components, 2);
return components;
}
/* Implement TARGET_SHRINK_WRAP_DISQUALIFY_COMPONENTS. */
static void
rs6000_disqualify_components (sbitmap components, edge e,
sbitmap edge_components, bool /*is_prologue*/)
{
/* Our LR pro/epilogue code moves LR via R0, so R0 had better not be
live where we want to place that code. */
if (bitmap_bit_p (edge_components, 0)
&& bitmap_bit_p (DF_LIVE_IN (e->dest), 0))
{
if (dump_file)
fprintf (dump_file, "Disqualifying LR because GPR0 is live "
"on entry to bb %d\n", e->dest->index);
bitmap_clear_bit (components, 0);
}
}
/* Implement TARGET_SHRINK_WRAP_EMIT_PROLOGUE_COMPONENTS. */
static void
rs6000_emit_prologue_components (sbitmap components)
{
rs6000_stack_t *info = rs6000_stack_info ();
rtx ptr_reg = gen_rtx_REG (Pmode, frame_pointer_needed
? HARD_FRAME_POINTER_REGNUM
: STACK_POINTER_REGNUM);
machine_mode reg_mode = Pmode;
int reg_size = TARGET_32BIT ? 4 : 8;
machine_mode fp_reg_mode = (TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT)
? DFmode : SFmode;
int fp_reg_size = 8;
/* Prologue for LR. */
if (bitmap_bit_p (components, 0))
{
rtx reg = gen_rtx_REG (reg_mode, 0);
rtx_insn *insn = emit_move_insn (reg, gen_rtx_REG (reg_mode, LR_REGNO));
RTX_FRAME_RELATED_P (insn) = 1;
add_reg_note (insn, REG_CFA_REGISTER, NULL);
int offset = info->lr_save_offset;
if (info->push_p)
offset += info->total_size;
insn = emit_insn (gen_frame_store (reg, ptr_reg, offset));
RTX_FRAME_RELATED_P (insn) = 1;
rtx lr = gen_rtx_REG (reg_mode, LR_REGNO);
rtx mem = copy_rtx (SET_DEST (single_set (insn)));
add_reg_note (insn, REG_CFA_OFFSET, gen_rtx_SET (mem, lr));
}
/* Prologue for TOC. */
if (bitmap_bit_p (components, 2))
{
rtx reg = gen_rtx_REG (reg_mode, TOC_REGNUM);
rtx sp_reg = gen_rtx_REG (Pmode, STACK_POINTER_REGNUM);
emit_insn (gen_frame_store (reg, sp_reg, RS6000_TOC_SAVE_SLOT));
}
/* Prologue for the GPRs. */
int offset = info->gp_save_offset;
if (info->push_p)
offset += info->total_size;
for (int i = info->first_gp_reg_save; i < 32; i++)
{
if (bitmap_bit_p (components, i))
{
rtx reg = gen_rtx_REG (reg_mode, i);
rtx_insn *insn = emit_insn (gen_frame_store (reg, ptr_reg, offset));
RTX_FRAME_RELATED_P (insn) = 1;
rtx set = copy_rtx (single_set (insn));
add_reg_note (insn, REG_CFA_OFFSET, set);
}
offset += reg_size;
}
/* Prologue for the FPRs. */
offset = info->fp_save_offset;
if (info->push_p)
offset += info->total_size;
for (int i = info->first_fp_reg_save; i < 64; i++)
{
if (bitmap_bit_p (components, i))
{
rtx reg = gen_rtx_REG (fp_reg_mode, i);
rtx_insn *insn = emit_insn (gen_frame_store (reg, ptr_reg, offset));
RTX_FRAME_RELATED_P (insn) = 1;
rtx set = copy_rtx (single_set (insn));
add_reg_note (insn, REG_CFA_OFFSET, set);
}
offset += fp_reg_size;
}
}
/* Implement TARGET_SHRINK_WRAP_EMIT_EPILOGUE_COMPONENTS. */
static void
rs6000_emit_epilogue_components (sbitmap components)
{
rs6000_stack_t *info = rs6000_stack_info ();
rtx ptr_reg = gen_rtx_REG (Pmode, frame_pointer_needed
? HARD_FRAME_POINTER_REGNUM
: STACK_POINTER_REGNUM);
machine_mode reg_mode = Pmode;
int reg_size = TARGET_32BIT ? 4 : 8;
machine_mode fp_reg_mode = (TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT)
? DFmode : SFmode;
int fp_reg_size = 8;
/* Epilogue for the FPRs. */
int offset = info->fp_save_offset;
if (info->push_p)
offset += info->total_size;
for (int i = info->first_fp_reg_save; i < 64; i++)
{
if (bitmap_bit_p (components, i))
{
rtx reg = gen_rtx_REG (fp_reg_mode, i);
rtx_insn *insn = emit_insn (gen_frame_load (reg, ptr_reg, offset));
RTX_FRAME_RELATED_P (insn) = 1;
add_reg_note (insn, REG_CFA_RESTORE, reg);
}
offset += fp_reg_size;
}
/* Epilogue for the GPRs. */
offset = info->gp_save_offset;
if (info->push_p)
offset += info->total_size;
for (int i = info->first_gp_reg_save; i < 32; i++)
{
if (bitmap_bit_p (components, i))
{
rtx reg = gen_rtx_REG (reg_mode, i);
rtx_insn *insn = emit_insn (gen_frame_load (reg, ptr_reg, offset));
RTX_FRAME_RELATED_P (insn) = 1;
add_reg_note (insn, REG_CFA_RESTORE, reg);
}
offset += reg_size;
}
/* Epilogue for LR. */
if (bitmap_bit_p (components, 0))
{
int offset = info->lr_save_offset;
if (info->push_p)
offset += info->total_size;
rtx reg = gen_rtx_REG (reg_mode, 0);
rtx_insn *insn = emit_insn (gen_frame_load (reg, ptr_reg, offset));
rtx lr = gen_rtx_REG (Pmode, LR_REGNO);
insn = emit_move_insn (lr, reg);
RTX_FRAME_RELATED_P (insn) = 1;
add_reg_note (insn, REG_CFA_RESTORE, lr);
}
}
/* Implement TARGET_SHRINK_WRAP_SET_HANDLED_COMPONENTS. */
static void
rs6000_set_handled_components (sbitmap components)
{
rs6000_stack_t *info = rs6000_stack_info ();
for (int i = info->first_gp_reg_save; i < 32; i++)
if (bitmap_bit_p (components, i))
cfun->machine->gpr_is_wrapped_separately[i] = true;
for (int i = info->first_fp_reg_save; i < 64; i++)
if (bitmap_bit_p (components, i))
cfun->machine->fpr_is_wrapped_separately[i - 32] = true;
if (bitmap_bit_p (components, 0))
cfun->machine->lr_is_wrapped_separately = true;
if (bitmap_bit_p (components, 2))
cfun->machine->toc_is_wrapped_separately = true;
}
/* VRSAVE is a bit vector representing which AltiVec registers
are used. The OS uses this to determine which vector
registers to save on a context switch. We need to save
VRSAVE on the stack frame, add whatever AltiVec registers we
used in this function, and do the corresponding magic in the
epilogue. */
static void
emit_vrsave_prologue (rs6000_stack_t *info, int save_regno,
HOST_WIDE_INT frame_off, rtx frame_reg_rtx)
{
/* Get VRSAVE into a GPR. */
rtx reg = gen_rtx_REG (SImode, save_regno);
rtx vrsave = gen_rtx_REG (SImode, VRSAVE_REGNO);
if (TARGET_MACHO)
emit_insn (gen_get_vrsave_internal (reg));
else
emit_insn (gen_rtx_SET (reg, vrsave));
/* Save VRSAVE. */
int offset = info->vrsave_save_offset + frame_off;
emit_insn (gen_frame_store (reg, frame_reg_rtx, offset));
/* Include the registers in the mask. */
emit_insn (gen_iorsi3 (reg, reg, GEN_INT (info->vrsave_mask)));
emit_insn (generate_set_vrsave (reg, info, 0));
}
/* Set up the arg pointer (r12) for -fsplit-stack code. If __morestack was
called, it left the arg pointer to the old stack in r29. Otherwise, the
arg pointer is the top of the current frame. */
static void
emit_split_stack_prologue (rs6000_stack_t *info, rtx_insn *sp_adjust,
HOST_WIDE_INT frame_off, rtx frame_reg_rtx)
{
cfun->machine->split_stack_argp_used = true;
if (sp_adjust)
{
rtx r12 = gen_rtx_REG (Pmode, 12);
rtx sp_reg_rtx = gen_rtx_REG (Pmode, STACK_POINTER_REGNUM);
rtx set_r12 = gen_rtx_SET (r12, sp_reg_rtx);
emit_insn_before (set_r12, sp_adjust);
}
else if (frame_off != 0 || REGNO (frame_reg_rtx) != 12)
{
rtx r12 = gen_rtx_REG (Pmode, 12);
if (frame_off == 0)
emit_move_insn (r12, frame_reg_rtx);
else
emit_insn (gen_add3_insn (r12, frame_reg_rtx, GEN_INT (frame_off)));
}
if (info->push_p)
{
rtx r12 = gen_rtx_REG (Pmode, 12);
rtx r29 = gen_rtx_REG (Pmode, 29);
rtx cr7 = gen_rtx_REG (CCUNSmode, CR7_REGNO);
rtx not_more = gen_label_rtx ();
rtx jump;
jump = gen_rtx_IF_THEN_ELSE (VOIDmode,
gen_rtx_GEU (VOIDmode, cr7, const0_rtx),
gen_rtx_LABEL_REF (VOIDmode, not_more),
pc_rtx);
jump = emit_jump_insn (gen_rtx_SET (pc_rtx, jump));
JUMP_LABEL (jump) = not_more;
LABEL_NUSES (not_more) += 1;
emit_move_insn (r12, r29);
emit_label (not_more);
}
}
/* Emit function prologue as insns. */
void
rs6000_emit_prologue (void)
{
rs6000_stack_t *info = rs6000_stack_info ();
machine_mode reg_mode = Pmode;
int reg_size = TARGET_32BIT ? 4 : 8;
machine_mode fp_reg_mode = (TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT)
? DFmode : SFmode;
int fp_reg_size = 8;
rtx sp_reg_rtx = gen_rtx_REG (Pmode, STACK_POINTER_REGNUM);
rtx frame_reg_rtx = sp_reg_rtx;
unsigned int cr_save_regno;
rtx cr_save_rtx = NULL_RTX;
rtx_insn *insn;
int strategy;
int using_static_chain_p = (cfun->static_chain_decl != NULL_TREE
&& df_regs_ever_live_p (STATIC_CHAIN_REGNUM)
&& call_used_regs[STATIC_CHAIN_REGNUM]);
int using_split_stack = (flag_split_stack
&& (lookup_attribute ("no_split_stack",
DECL_ATTRIBUTES (cfun->decl))
== NULL));
/* Offset to top of frame for frame_reg and sp respectively. */
HOST_WIDE_INT frame_off = 0;
HOST_WIDE_INT sp_off = 0;
/* sp_adjust is the stack adjusting instruction, tracked so that the
insn setting up the split-stack arg pointer can be emitted just
prior to it, when r12 is not used here for other purposes. */
rtx_insn *sp_adjust = 0;
#if CHECKING_P
/* Track and check usage of r0, r11, r12. */
int reg_inuse = using_static_chain_p ? 1 << 11 : 0;
#define START_USE(R) do \
{ \
gcc_assert ((reg_inuse & (1 << (R))) == 0); \
reg_inuse |= 1 << (R); \
} while (0)
#define END_USE(R) do \
{ \
gcc_assert ((reg_inuse & (1 << (R))) != 0); \
reg_inuse &= ~(1 << (R)); \
} while (0)
#define NOT_INUSE(R) do \
{ \
gcc_assert ((reg_inuse & (1 << (R))) == 0); \
} while (0)
#else
#define START_USE(R) do {} while (0)
#define END_USE(R) do {} while (0)
#define NOT_INUSE(R) do {} while (0)
#endif
if (DEFAULT_ABI == ABI_ELFv2
&& !TARGET_SINGLE_PIC_BASE)
{
cfun->machine->r2_setup_needed = df_regs_ever_live_p (TOC_REGNUM);
/* With -mminimal-toc we may generate an extra use of r2 below. */
if (TARGET_TOC && TARGET_MINIMAL_TOC
&& !constant_pool_empty_p ())
cfun->machine->r2_setup_needed = true;
}
if (flag_stack_usage_info)
current_function_static_stack_size = info->total_size;
if (flag_stack_check == STATIC_BUILTIN_STACK_CHECK)
{
HOST_WIDE_INT size = info->total_size;
if (crtl->is_leaf && !cfun->calls_alloca)
{
if (size > PROBE_INTERVAL && size > get_stack_check_protect ())
rs6000_emit_probe_stack_range (get_stack_check_protect (),
size - get_stack_check_protect ());
}
else if (size > 0)
rs6000_emit_probe_stack_range (get_stack_check_protect (), size);
}
if (TARGET_FIX_AND_CONTINUE)
{
/* gdb on darwin arranges to forward a function from the old
address by modifying the first 5 instructions of the function
to branch to the overriding function. This is necessary to
permit function pointers that point to the old function to
actually forward to the new function. */
emit_insn (gen_nop ());
emit_insn (gen_nop ());
emit_insn (gen_nop ());
emit_insn (gen_nop ());
emit_insn (gen_nop ());
}
/* Handle world saves specially here. */
if (WORLD_SAVE_P (info))
{
int i, j, sz;
rtx treg;
rtvec p;
rtx reg0;
/* save_world expects lr in r0. */
reg0 = gen_rtx_REG (Pmode, 0);
if (info->lr_save_p)
{
insn = emit_move_insn (reg0,
gen_rtx_REG (Pmode, LR_REGNO));
RTX_FRAME_RELATED_P (insn) = 1;
}
/* The SAVE_WORLD and RESTORE_WORLD routines make a number of
assumptions about the offsets of various bits of the stack
frame. */
gcc_assert (info->gp_save_offset == -220
&& info->fp_save_offset == -144
&& info->lr_save_offset == 8
&& info->cr_save_offset == 4
&& info->push_p
&& info->lr_save_p
&& (!crtl->calls_eh_return
|| info->ehrd_offset == -432)
&& info->vrsave_save_offset == -224
&& info->altivec_save_offset == -416);
treg = gen_rtx_REG (SImode, 11);
emit_move_insn (treg, GEN_INT (-info->total_size));
/* SAVE_WORLD takes the caller's LR in R0 and the frame size
in R11. It also clobbers R12, so beware! */
/* Preserve CR2 for save_world prologues */
sz = 5;
sz += 32 - info->first_gp_reg_save;
sz += 64 - info->first_fp_reg_save;
sz += LAST_ALTIVEC_REGNO - info->first_altivec_reg_save + 1;
p = rtvec_alloc (sz);
j = 0;
RTVEC_ELT (p, j++) = gen_rtx_CLOBBER (VOIDmode,
gen_rtx_REG (SImode,
LR_REGNO));
RTVEC_ELT (p, j++) = gen_rtx_USE (VOIDmode,
gen_rtx_SYMBOL_REF (Pmode,
"*save_world"));
/* We do floats first so that the instruction pattern matches
properly. */
for (i = 0; i < 64 - info->first_fp_reg_save; i++)
RTVEC_ELT (p, j++)
= gen_frame_store (gen_rtx_REG (TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT
? DFmode : SFmode,
info->first_fp_reg_save + i),
frame_reg_rtx,
info->fp_save_offset + frame_off + 8 * i);
for (i = 0; info->first_altivec_reg_save + i <= LAST_ALTIVEC_REGNO; i++)
RTVEC_ELT (p, j++)
= gen_frame_store (gen_rtx_REG (V4SImode,
info->first_altivec_reg_save + i),
frame_reg_rtx,
info->altivec_save_offset + frame_off + 16 * i);
for (i = 0; i < 32 - info->first_gp_reg_save; i++)
RTVEC_ELT (p, j++)
= gen_frame_store (gen_rtx_REG (reg_mode, info->first_gp_reg_save + i),
frame_reg_rtx,
info->gp_save_offset + frame_off + reg_size * i);
/* CR register traditionally saved as CR2. */
RTVEC_ELT (p, j++)
= gen_frame_store (gen_rtx_REG (SImode, CR2_REGNO),
frame_reg_rtx, info->cr_save_offset + frame_off);
/* Explain about use of R0. */
if (info->lr_save_p)
RTVEC_ELT (p, j++)
= gen_frame_store (reg0,
frame_reg_rtx, info->lr_save_offset + frame_off);
/* Explain what happens to the stack pointer. */
{
rtx newval = gen_rtx_PLUS (Pmode, sp_reg_rtx, treg);
RTVEC_ELT (p, j++) = gen_rtx_SET (sp_reg_rtx, newval);
}
insn = emit_insn (gen_rtx_PARALLEL (VOIDmode, p));
rs6000_frame_related (insn, frame_reg_rtx, sp_off - frame_off,
treg, GEN_INT (-info->total_size));
sp_off = frame_off = info->total_size;
}
strategy = info->savres_strategy;
/* For V.4, update stack before we do any saving and set back pointer. */
if (! WORLD_SAVE_P (info)
&& info->push_p
&& (DEFAULT_ABI == ABI_V4
|| crtl->calls_eh_return))
{
bool need_r11 = (!(strategy & SAVE_INLINE_FPRS)
|| !(strategy & SAVE_INLINE_GPRS)
|| !(strategy & SAVE_INLINE_VRS));
int ptr_regno = -1;
rtx ptr_reg = NULL_RTX;
int ptr_off = 0;
if (info->total_size < 32767)
frame_off = info->total_size;
else if (need_r11)
ptr_regno = 11;
else if (info->cr_save_p
|| info->lr_save_p
|| info->first_fp_reg_save < 64
|| info->first_gp_reg_save < 32
|| info->altivec_size != 0
|| info->vrsave_size != 0
|| crtl->calls_eh_return)
ptr_regno = 12;
else
{
/* The prologue won't be saving any regs so there is no need
to set up a frame register to access any frame save area.
We also won't be using frame_off anywhere below, but set
the correct value anyway to protect against future
changes to this function. */
frame_off = info->total_size;
}
if (ptr_regno != -1)
{
/* Set up the frame offset to that needed by the first
out-of-line save function. */
START_USE (ptr_regno);
ptr_reg = gen_rtx_REG (Pmode, ptr_regno);
frame_reg_rtx = ptr_reg;
if (!(strategy & SAVE_INLINE_FPRS) && info->fp_size != 0)
gcc_checking_assert (info->fp_save_offset + info->fp_size == 0);
else if (!(strategy & SAVE_INLINE_GPRS) && info->first_gp_reg_save < 32)
ptr_off = info->gp_save_offset + info->gp_size;
else if (!(strategy & SAVE_INLINE_VRS) && info->altivec_size != 0)
ptr_off = info->altivec_save_offset + info->altivec_size;
frame_off = -ptr_off;
}
sp_adjust = rs6000_emit_allocate_stack (info->total_size,
ptr_reg, ptr_off);
if (REGNO (frame_reg_rtx) == 12)
sp_adjust = 0;
sp_off = info->total_size;
if (frame_reg_rtx != sp_reg_rtx)
rs6000_emit_stack_tie (frame_reg_rtx, false);
}
/* If we use the link register, get it into r0. */
if (!WORLD_SAVE_P (info) && info->lr_save_p
&& !cfun->machine->lr_is_wrapped_separately)
{
rtx addr, reg, mem;
reg = gen_rtx_REG (Pmode, 0);
START_USE (0);
insn = emit_move_insn (reg, gen_rtx_REG (Pmode, LR_REGNO));
RTX_FRAME_RELATED_P (insn) = 1;
if (!(strategy & (SAVE_NOINLINE_GPRS_SAVES_LR
| SAVE_NOINLINE_FPRS_SAVES_LR)))
{
addr = gen_rtx_PLUS (Pmode, frame_reg_rtx,
GEN_INT (info->lr_save_offset + frame_off));
mem = gen_rtx_MEM (Pmode, addr);
/* This should not be of rs6000_sr_alias_set, because of
__builtin_return_address. */
insn = emit_move_insn (mem, reg);
rs6000_frame_related (insn, frame_reg_rtx, sp_off - frame_off,
NULL_RTX, NULL_RTX);
END_USE (0);
}
}
/* If we need to save CR, put it into r12 or r11. Choose r12 except when
r12 will be needed by out-of-line gpr restore. */
cr_save_regno = ((DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
&& !(strategy & (SAVE_INLINE_GPRS
| SAVE_NOINLINE_GPRS_SAVES_LR))
? 11 : 12);
if (!WORLD_SAVE_P (info)
&& info->cr_save_p
&& REGNO (frame_reg_rtx) != cr_save_regno
&& !(using_static_chain_p && cr_save_regno == 11)
&& !(using_split_stack && cr_save_regno == 12 && sp_adjust))
{
cr_save_rtx = gen_rtx_REG (SImode, cr_save_regno);
START_USE (cr_save_regno);
rs6000_emit_prologue_move_from_cr (cr_save_rtx);
}
/* Do any required saving of fpr's. If only one or two to save, do
it ourselves. Otherwise, call function. */
if (!WORLD_SAVE_P (info) && (strategy & SAVE_INLINE_FPRS))
{
int offset = info->fp_save_offset + frame_off;
for (int i = info->first_fp_reg_save; i < 64; i++)
{
if (save_reg_p (i)
&& !cfun->machine->fpr_is_wrapped_separately[i - 32])
emit_frame_save (frame_reg_rtx, fp_reg_mode, i, offset,
sp_off - frame_off);
offset += fp_reg_size;
}
}
else if (!WORLD_SAVE_P (info) && info->first_fp_reg_save != 64)
{
bool lr = (strategy & SAVE_NOINLINE_FPRS_SAVES_LR) != 0;
int sel = SAVRES_SAVE | SAVRES_FPR | (lr ? SAVRES_LR : 0);
unsigned ptr_regno = ptr_regno_for_savres (sel);
rtx ptr_reg = frame_reg_rtx;
if (REGNO (frame_reg_rtx) == ptr_regno)
gcc_checking_assert (frame_off == 0);
else
{
ptr_reg = gen_rtx_REG (Pmode, ptr_regno);
NOT_INUSE (ptr_regno);
emit_insn (gen_add3_insn (ptr_reg,
frame_reg_rtx, GEN_INT (frame_off)));
}
insn = rs6000_emit_savres_rtx (info, ptr_reg,
info->fp_save_offset,
info->lr_save_offset,
DFmode, sel);
rs6000_frame_related (insn, ptr_reg, sp_off,
NULL_RTX, NULL_RTX);
if (lr)
END_USE (0);
}
/* Save GPRs. This is done as a PARALLEL if we are using
the store-multiple instructions. */
if (!WORLD_SAVE_P (info) && !(strategy & SAVE_INLINE_GPRS))
{
bool lr = (strategy & SAVE_NOINLINE_GPRS_SAVES_LR) != 0;
int sel = SAVRES_SAVE | SAVRES_GPR | (lr ? SAVRES_LR : 0);
unsigned ptr_regno = ptr_regno_for_savres (sel);
rtx ptr_reg = frame_reg_rtx;
bool ptr_set_up = REGNO (ptr_reg) == ptr_regno;
int end_save = info->gp_save_offset + info->gp_size;
int ptr_off;
if (ptr_regno == 12)
sp_adjust = 0;
if (!ptr_set_up)
ptr_reg = gen_rtx_REG (Pmode, ptr_regno);
/* Need to adjust r11 (r12) if we saved any FPRs. */
if (end_save + frame_off != 0)
{
rtx offset = GEN_INT (end_save + frame_off);
if (ptr_set_up)
frame_off = -end_save;
else
NOT_INUSE (ptr_regno);
emit_insn (gen_add3_insn (ptr_reg, frame_reg_rtx, offset));
}
else if (!ptr_set_up)
{
NOT_INUSE (ptr_regno);
emit_move_insn (ptr_reg, frame_reg_rtx);
}
ptr_off = -end_save;
insn = rs6000_emit_savres_rtx (info, ptr_reg,
info->gp_save_offset + ptr_off,
info->lr_save_offset + ptr_off,
reg_mode, sel);
rs6000_frame_related (insn, ptr_reg, sp_off - ptr_off,
NULL_RTX, NULL_RTX);
if (lr)
END_USE (0);
}
else if (!WORLD_SAVE_P (info) && (strategy & SAVE_MULTIPLE))
{
rtvec p;
int i;
p = rtvec_alloc (32 - info->first_gp_reg_save);
for (i = 0; i < 32 - info->first_gp_reg_save; i++)
RTVEC_ELT (p, i)
= gen_frame_store (gen_rtx_REG (reg_mode, info->first_gp_reg_save + i),
frame_reg_rtx,
info->gp_save_offset + frame_off + reg_size * i);
insn = emit_insn (gen_rtx_PARALLEL (VOIDmode, p));
rs6000_frame_related (insn, frame_reg_rtx, sp_off - frame_off,
NULL_RTX, NULL_RTX);
}
else if (!WORLD_SAVE_P (info))
{
int offset = info->gp_save_offset + frame_off;
for (int i = info->first_gp_reg_save; i < 32; i++)
{
if (save_reg_p (i)
&& !cfun->machine->gpr_is_wrapped_separately[i])
emit_frame_save (frame_reg_rtx, reg_mode, i, offset,
sp_off - frame_off);
offset += reg_size;
}
}
if (crtl->calls_eh_return)
{
unsigned int i;
rtvec p;
for (i = 0; ; ++i)
{
unsigned int regno = EH_RETURN_DATA_REGNO (i);
if (regno == INVALID_REGNUM)
break;
}
p = rtvec_alloc (i);
for (i = 0; ; ++i)
{
unsigned int regno = EH_RETURN_DATA_REGNO (i);
if (regno == INVALID_REGNUM)
break;
rtx set
= gen_frame_store (gen_rtx_REG (reg_mode, regno),
sp_reg_rtx,
info->ehrd_offset + sp_off + reg_size * (int) i);
RTVEC_ELT (p, i) = set;
RTX_FRAME_RELATED_P (set) = 1;
}
insn = emit_insn (gen_blockage ());
RTX_FRAME_RELATED_P (insn) = 1;
add_reg_note (insn, REG_FRAME_RELATED_EXPR, gen_rtx_PARALLEL (VOIDmode, p));
}
/* In AIX ABI we need to make sure r2 is really saved. */
if (TARGET_AIX && crtl->calls_eh_return)
{
rtx tmp_reg, tmp_reg_si, hi, lo, compare_result, toc_save_done, jump;
rtx join_insn, note;
rtx_insn *save_insn;
long toc_restore_insn;
tmp_reg = gen_rtx_REG (Pmode, 11);
tmp_reg_si = gen_rtx_REG (SImode, 11);
if (using_static_chain_p)
{
START_USE (0);
emit_move_insn (gen_rtx_REG (Pmode, 0), tmp_reg);
}
else
START_USE (11);
emit_move_insn (tmp_reg, gen_rtx_REG (Pmode, LR_REGNO));
/* Peek at instruction to which this function returns. If it's
restoring r2, then we know we've already saved r2. We can't
unconditionally save r2 because the value we have will already
be updated if we arrived at this function via a plt call or
toc adjusting stub. */
emit_move_insn (tmp_reg_si, gen_rtx_MEM (SImode, tmp_reg));
toc_restore_insn = ((TARGET_32BIT ? 0x80410000 : 0xE8410000)
+ RS6000_TOC_SAVE_SLOT);
hi = gen_int_mode (toc_restore_insn & ~0xffff, SImode);
emit_insn (gen_xorsi3 (tmp_reg_si, tmp_reg_si, hi));
compare_result = gen_rtx_REG (CCUNSmode, CR0_REGNO);
validate_condition_mode (EQ, CCUNSmode);
lo = gen_int_mode (toc_restore_insn & 0xffff, SImode);
emit_insn (gen_rtx_SET (compare_result,
gen_rtx_COMPARE (CCUNSmode, tmp_reg_si, lo)));
toc_save_done = gen_label_rtx ();
jump = gen_rtx_IF_THEN_ELSE (VOIDmode,
gen_rtx_EQ (VOIDmode, compare_result,
const0_rtx),
gen_rtx_LABEL_REF (VOIDmode, toc_save_done),
pc_rtx);
jump = emit_jump_insn (gen_rtx_SET (pc_rtx, jump));
JUMP_LABEL (jump) = toc_save_done;
LABEL_NUSES (toc_save_done) += 1;
save_insn = emit_frame_save (frame_reg_rtx, reg_mode,
TOC_REGNUM, frame_off + RS6000_TOC_SAVE_SLOT,
sp_off - frame_off);
emit_label (toc_save_done);
/* ??? If we leave SAVE_INSN as marked as saving R2, then we'll
have a CFG that has different saves along different paths.
Move the note to a dummy blockage insn, which describes that
R2 is unconditionally saved after the label. */
/* ??? An alternate representation might be a special insn pattern
containing both the branch and the store. That might let the
code that minimizes the number of DW_CFA_advance opcodes better
freedom in placing the annotations. */
note = find_reg_note (save_insn, REG_FRAME_RELATED_EXPR, NULL);
if (note)
remove_note (save_insn, note);
else
note = alloc_reg_note (REG_FRAME_RELATED_EXPR,
copy_rtx (PATTERN (save_insn)), NULL_RTX);
RTX_FRAME_RELATED_P (save_insn) = 0;
join_insn = emit_insn (gen_blockage ());
REG_NOTES (join_insn) = note;
RTX_FRAME_RELATED_P (join_insn) = 1;
if (using_static_chain_p)
{
emit_move_insn (tmp_reg, gen_rtx_REG (Pmode, 0));
END_USE (0);
}
else
END_USE (11);
}
/* Save CR if we use any that must be preserved. */
if (!WORLD_SAVE_P (info) && info->cr_save_p)
{
rtx addr = gen_rtx_PLUS (Pmode, frame_reg_rtx,
GEN_INT (info->cr_save_offset + frame_off));
rtx mem = gen_frame_mem (SImode, addr);
/* If we didn't copy cr before, do so now using r0. */
if (cr_save_rtx == NULL_RTX)
{
START_USE (0);
cr_save_rtx = gen_rtx_REG (SImode, 0);
rs6000_emit_prologue_move_from_cr (cr_save_rtx);
}
/* Saving CR requires a two-instruction sequence: one instruction
to move the CR to a general-purpose register, and a second
instruction that stores the GPR to memory.
We do not emit any DWARF CFI records for the first of these,
because we cannot properly represent the fact that CR is saved in
a register. One reason is that we cannot express that multiple
CR fields are saved; another reason is that on 64-bit, the size
of the CR register in DWARF (4 bytes) differs from the size of
a general-purpose register.
This means if any intervening instruction were to clobber one of
the call-saved CR fields, we'd have incorrect CFI. To prevent
this from happening, we mark the store to memory as a use of
those CR fields, which prevents any such instruction from being
scheduled in between the two instructions. */
rtx crsave_v[9];
int n_crsave = 0;
int i;
crsave_v[n_crsave++] = gen_rtx_SET (mem, cr_save_rtx);
for (i = 0; i < 8; i++)
if (save_reg_p (CR0_REGNO + i))
crsave_v[n_crsave++]
= gen_rtx_USE (VOIDmode, gen_rtx_REG (CCmode, CR0_REGNO + i));
insn = emit_insn (gen_rtx_PARALLEL (VOIDmode,
gen_rtvec_v (n_crsave, crsave_v)));
END_USE (REGNO (cr_save_rtx));
/* Now, there's no way that dwarf2out_frame_debug_expr is going to
understand '(unspec:SI [(reg:CC 68) ...] UNSPEC_MOVESI_FROM_CR)',
so we need to construct a frame expression manually. */
RTX_FRAME_RELATED_P (insn) = 1;
/* Update address to be stack-pointer relative, like
rs6000_frame_related would do. */
addr = gen_rtx_PLUS (Pmode, gen_rtx_REG (Pmode, STACK_POINTER_REGNUM),
GEN_INT (info->cr_save_offset + sp_off));
mem = gen_frame_mem (SImode, addr);
if (DEFAULT_ABI == ABI_ELFv2)
{
/* In the ELFv2 ABI we generate separate CFI records for each
CR field that was actually saved. They all point to the
same 32-bit stack slot. */
rtx crframe[8];
int n_crframe = 0;
for (i = 0; i < 8; i++)
if (save_reg_p (CR0_REGNO + i))
{
crframe[n_crframe]
= gen_rtx_SET (mem, gen_rtx_REG (SImode, CR0_REGNO + i));
RTX_FRAME_RELATED_P (crframe[n_crframe]) = 1;
n_crframe++;
}
add_reg_note (insn, REG_FRAME_RELATED_EXPR,
gen_rtx_PARALLEL (VOIDmode,
gen_rtvec_v (n_crframe, crframe)));
}
else
{
/* In other ABIs, by convention, we use a single CR regnum to
represent the fact that all call-saved CR fields are saved.
We use CR2_REGNO to be compatible with gcc-2.95 on Linux. */
rtx set = gen_rtx_SET (mem, gen_rtx_REG (SImode, CR2_REGNO));
add_reg_note (insn, REG_FRAME_RELATED_EXPR, set);
}
}
/* In the ELFv2 ABI we need to save all call-saved CR fields into
*separate* slots if the routine calls __builtin_eh_return, so
that they can be independently restored by the unwinder. */
if (DEFAULT_ABI == ABI_ELFv2 && crtl->calls_eh_return)
{
int i, cr_off = info->ehcr_offset;
rtx crsave;
/* ??? We might get better performance by using multiple mfocrf
instructions. */
crsave = gen_rtx_REG (SImode, 0);
emit_insn (gen_prologue_movesi_from_cr (crsave));
for (i = 0; i < 8; i++)
if (!call_used_regs[CR0_REGNO + i])
{
rtvec p = rtvec_alloc (2);
RTVEC_ELT (p, 0)
= gen_frame_store (crsave, frame_reg_rtx, cr_off + frame_off);
RTVEC_ELT (p, 1)
= gen_rtx_USE (VOIDmode, gen_rtx_REG (CCmode, CR0_REGNO + i));
insn = emit_insn (gen_rtx_PARALLEL (VOIDmode, p));
RTX_FRAME_RELATED_P (insn) = 1;
add_reg_note (insn, REG_FRAME_RELATED_EXPR,
gen_frame_store (gen_rtx_REG (SImode, CR0_REGNO + i),
sp_reg_rtx, cr_off + sp_off));
cr_off += reg_size;
}
}
/* If we are emitting stack probes, but allocate no stack, then
just note that in the dump file. */
if (flag_stack_clash_protection
&& dump_file
&& !info->push_p)
dump_stack_clash_frame_info (NO_PROBE_NO_FRAME, false);
/* Update stack and set back pointer unless this is V.4,
for which it was done previously. */
if (!WORLD_SAVE_P (info) && info->push_p
&& !(DEFAULT_ABI == ABI_V4 || crtl->calls_eh_return))
{
rtx ptr_reg = NULL;
int ptr_off = 0;
/* If saving altivec regs we need to be able to address all save
locations using a 16-bit offset. */
if ((strategy & SAVE_INLINE_VRS) == 0
|| (info->altivec_size != 0
&& (info->altivec_save_offset + info->altivec_size - 16
+ info->total_size - frame_off) > 32767)
|| (info->vrsave_size != 0
&& (info->vrsave_save_offset
+ info->total_size - frame_off) > 32767))
{
int sel = SAVRES_SAVE | SAVRES_VR;
unsigned ptr_regno = ptr_regno_for_savres (sel);
if (using_static_chain_p
&& ptr_regno == STATIC_CHAIN_REGNUM)
ptr_regno = 12;
if (REGNO (frame_reg_rtx) != ptr_regno)
START_USE (ptr_regno);
ptr_reg = gen_rtx_REG (Pmode, ptr_regno);
frame_reg_rtx = ptr_reg;
ptr_off = info->altivec_save_offset + info->altivec_size;
frame_off = -ptr_off;
}
else if (REGNO (frame_reg_rtx) == 1)
frame_off = info->total_size;
sp_adjust = rs6000_emit_allocate_stack (info->total_size,
ptr_reg, ptr_off);
if (REGNO (frame_reg_rtx) == 12)
sp_adjust = 0;
sp_off = info->total_size;
if (frame_reg_rtx != sp_reg_rtx)
rs6000_emit_stack_tie (frame_reg_rtx, false);
}
/* Set frame pointer, if needed. */
if (frame_pointer_needed)
{
insn = emit_move_insn (gen_rtx_REG (Pmode, HARD_FRAME_POINTER_REGNUM),
sp_reg_rtx);
RTX_FRAME_RELATED_P (insn) = 1;
}
/* Save AltiVec registers if needed. Save here because the red zone does
not always include AltiVec registers. */
if (!WORLD_SAVE_P (info)
&& info->altivec_size != 0 && (strategy & SAVE_INLINE_VRS) == 0)
{
int end_save = info->altivec_save_offset + info->altivec_size;
int ptr_off;
/* Oddly, the vector save/restore functions point r0 at the end
of the save area, then use r11 or r12 to load offsets for
[reg+reg] addressing. */
rtx ptr_reg = gen_rtx_REG (Pmode, 0);
int scratch_regno = ptr_regno_for_savres (SAVRES_SAVE | SAVRES_VR);
rtx scratch_reg = gen_rtx_REG (Pmode, scratch_regno);
gcc_checking_assert (scratch_regno == 11 || scratch_regno == 12);
NOT_INUSE (0);
if (scratch_regno == 12)
sp_adjust = 0;
if (end_save + frame_off != 0)
{
rtx offset = GEN_INT (end_save + frame_off);
emit_insn (gen_add3_insn (ptr_reg, frame_reg_rtx, offset));
}
else
emit_move_insn (ptr_reg, frame_reg_rtx);
ptr_off = -end_save;
insn = rs6000_emit_savres_rtx (info, scratch_reg,
info->altivec_save_offset + ptr_off,
0, V4SImode, SAVRES_SAVE | SAVRES_VR);
rs6000_frame_related (insn, scratch_reg, sp_off - ptr_off,
NULL_RTX, NULL_RTX);
if (REGNO (frame_reg_rtx) == REGNO (scratch_reg))
{
/* The oddity mentioned above clobbered our frame reg. */
emit_move_insn (frame_reg_rtx, ptr_reg);
frame_off = ptr_off;
}
}
else if (!WORLD_SAVE_P (info)
&& info->altivec_size != 0)
{
int i;
for (i = info->first_altivec_reg_save; i <= LAST_ALTIVEC_REGNO; ++i)
if (info->vrsave_mask & ALTIVEC_REG_BIT (i))
{
rtx areg, savereg, mem;
HOST_WIDE_INT offset;
offset = (info->altivec_save_offset + frame_off
+ 16 * (i - info->first_altivec_reg_save));
savereg = gen_rtx_REG (V4SImode, i);
if (TARGET_P9_VECTOR && quad_address_offset_p (offset))
{
mem = gen_frame_mem (V4SImode,
gen_rtx_PLUS (Pmode, frame_reg_rtx,
GEN_INT (offset)));
insn = emit_insn (gen_rtx_SET (mem, savereg));
areg = NULL_RTX;
}
else
{
NOT_INUSE (0);
areg = gen_rtx_REG (Pmode, 0);
emit_move_insn (areg, GEN_INT (offset));
/* AltiVec addressing mode is [reg+reg]. */
mem = gen_frame_mem (V4SImode,
gen_rtx_PLUS (Pmode, frame_reg_rtx, areg));
/* Rather than emitting a generic move, force use of the stvx
instruction, which we always want on ISA 2.07 (power8) systems.
In particular we don't want xxpermdi/stxvd2x for little
endian. */
insn = emit_insn (gen_altivec_stvx_v4si_internal (mem, savereg));
}
rs6000_frame_related (insn, frame_reg_rtx, sp_off - frame_off,
areg, GEN_INT (offset));
}
}
/* VRSAVE is a bit vector representing which AltiVec registers
are used. The OS uses this to determine which vector
registers to save on a context switch. We need to save
VRSAVE on the stack frame, add whatever AltiVec registers we
used in this function, and do the corresponding magic in the
epilogue. */
if (!WORLD_SAVE_P (info) && info->vrsave_size != 0)
{
/* Get VRSAVE into a GPR. Note that ABI_V4 and ABI_DARWIN might
be using r12 as frame_reg_rtx and r11 as the static chain
pointer for nested functions. */
int save_regno = 12;
if ((DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
&& !using_static_chain_p)
save_regno = 11;
else if (using_split_stack || REGNO (frame_reg_rtx) == 12)
{
save_regno = 11;
if (using_static_chain_p)
save_regno = 0;
}
NOT_INUSE (save_regno);
emit_vrsave_prologue (info, save_regno, frame_off, frame_reg_rtx);
}
/* If we are using RS6000_PIC_OFFSET_TABLE_REGNUM, we need to set it up. */
if (!TARGET_SINGLE_PIC_BASE
&& ((TARGET_TOC && TARGET_MINIMAL_TOC
&& !constant_pool_empty_p ())
|| (DEFAULT_ABI == ABI_V4
&& (flag_pic == 1 || (flag_pic && TARGET_SECURE_PLT))
&& df_regs_ever_live_p (RS6000_PIC_OFFSET_TABLE_REGNUM))))
{
/* If emit_load_toc_table will use the link register, we need to save
it. We use R12 for this purpose because emit_load_toc_table
can use register 0. This allows us to use a plain 'blr' to return
from the procedure more often. */
int save_LR_around_toc_setup = (TARGET_ELF
&& DEFAULT_ABI == ABI_V4
&& flag_pic
&& ! info->lr_save_p
&& EDGE_COUNT (EXIT_BLOCK_PTR_FOR_FN (cfun)->preds) > 0);
if (save_LR_around_toc_setup)
{
rtx lr = gen_rtx_REG (Pmode, LR_REGNO);
rtx tmp = gen_rtx_REG (Pmode, 12);
sp_adjust = 0;
insn = emit_move_insn (tmp, lr);
RTX_FRAME_RELATED_P (insn) = 1;
rs6000_emit_load_toc_table (TRUE);
insn = emit_move_insn (lr, tmp);
add_reg_note (insn, REG_CFA_RESTORE, lr);
RTX_FRAME_RELATED_P (insn) = 1;
}
else
rs6000_emit_load_toc_table (TRUE);
}
#if TARGET_MACHO
if (!TARGET_SINGLE_PIC_BASE
&& DEFAULT_ABI == ABI_DARWIN
&& flag_pic && crtl->uses_pic_offset_table)
{
rtx lr = gen_rtx_REG (Pmode, LR_REGNO);
rtx src = gen_rtx_SYMBOL_REF (Pmode, MACHOPIC_FUNCTION_BASE_NAME);
/* Save and restore LR locally around this call (in R0). */
if (!info->lr_save_p)
emit_move_insn (gen_rtx_REG (Pmode, 0), lr);
emit_insn (gen_load_macho_picbase (src));
emit_move_insn (gen_rtx_REG (Pmode,
RS6000_PIC_OFFSET_TABLE_REGNUM),
lr);
if (!info->lr_save_p)
emit_move_insn (lr, gen_rtx_REG (Pmode, 0));
}
#endif
/* If we need to, save the TOC register after doing the stack setup.
Do not emit eh frame info for this save. The unwinder wants info,
conceptually attached to instructions in this function, about
register values in the caller of this function. This R2 may have
already been changed from the value in the caller.
We don't attempt to write accurate DWARF EH frame info for R2
because code emitted by gcc for a (non-pointer) function call
doesn't save and restore R2. Instead, R2 is managed out-of-line
by a linker generated plt call stub when the function resides in
a shared library. This behavior is costly to describe in DWARF,
both in terms of the size of DWARF info and the time taken in the
unwinder to interpret it. R2 changes, apart from the
calls_eh_return case earlier in this function, are handled by
linux-unwind.h frob_update_context. */
if (rs6000_save_toc_in_prologue_p ()
&& !cfun->machine->toc_is_wrapped_separately)
{
rtx reg = gen_rtx_REG (reg_mode, TOC_REGNUM);
emit_insn (gen_frame_store (reg, sp_reg_rtx, RS6000_TOC_SAVE_SLOT));
}
/* Set up the arg pointer (r12) for -fsplit-stack code. */
if (using_split_stack && split_stack_arg_pointer_used_p ())
emit_split_stack_prologue (info, sp_adjust, frame_off, frame_reg_rtx);
}
/* Output .extern statements for the save/restore routines we use. */
static void
rs6000_output_savres_externs (FILE *file)
{
rs6000_stack_t *info = rs6000_stack_info ();
if (TARGET_DEBUG_STACK)
debug_stack_info (info);
/* Write .extern for any function we will call to save and restore
fp values. */
if (info->first_fp_reg_save < 64
&& !TARGET_MACHO
&& !TARGET_ELF)
{
char *name;
int regno = info->first_fp_reg_save - 32;
if ((info->savres_strategy & SAVE_INLINE_FPRS) == 0)
{
bool lr = (info->savres_strategy & SAVE_NOINLINE_FPRS_SAVES_LR) != 0;
int sel = SAVRES_SAVE | SAVRES_FPR | (lr ? SAVRES_LR : 0);
name = rs6000_savres_routine_name (regno, sel);
fprintf (file, "\t.extern %s\n", name);
}
if ((info->savres_strategy & REST_INLINE_FPRS) == 0)
{
bool lr = (info->savres_strategy
& REST_NOINLINE_FPRS_DOESNT_RESTORE_LR) == 0;
int sel = SAVRES_FPR | (lr ? SAVRES_LR : 0);
name = rs6000_savres_routine_name (regno, sel);
fprintf (file, "\t.extern %s\n", name);
}
}
}
/* Write function prologue. */
static void
rs6000_output_function_prologue (FILE *file)
{
if (!cfun->is_thunk)
rs6000_output_savres_externs (file);
/* ELFv2 ABI r2 setup code and local entry point. This must follow
immediately after the global entry point label. */
if (rs6000_global_entry_point_needed_p ())
{
const char *name = XSTR (XEXP (DECL_RTL (current_function_decl), 0), 0);
(*targetm.asm_out.internal_label) (file, "LCF", rs6000_pic_labelno);
if (TARGET_CMODEL != CMODEL_LARGE)
{
/* In the small and medium code models, we assume the TOC is less
2 GB away from the text section, so it can be computed via the
following two-instruction sequence. */
char buf[256];
ASM_GENERATE_INTERNAL_LABEL (buf, "LCF", rs6000_pic_labelno);
fprintf (file, "0:\taddis 2,12,.TOC.-");
assemble_name (file, buf);
fprintf (file, "@ha\n");
fprintf (file, "\taddi 2,2,.TOC.-");
assemble_name (file, buf);
fprintf (file, "@l\n");
}
else
{
/* In the large code model, we allow arbitrary offsets between the
TOC and the text section, so we have to load the offset from
memory. The data field is emitted directly before the global
entry point in rs6000_elf_declare_function_name. */
char buf[256];
#ifdef HAVE_AS_ENTRY_MARKERS
/* If supported by the linker, emit a marker relocation. If the
total code size of the final executable or shared library
happens to fit into 2 GB after all, the linker will replace
this code sequence with the sequence for the small or medium
code model. */
fprintf (file, "\t.reloc .,R_PPC64_ENTRY\n");
#endif
fprintf (file, "\tld 2,");
ASM_GENERATE_INTERNAL_LABEL (buf, "LCL", rs6000_pic_labelno);
assemble_name (file, buf);
fprintf (file, "-");
ASM_GENERATE_INTERNAL_LABEL (buf, "LCF", rs6000_pic_labelno);
assemble_name (file, buf);
fprintf (file, "(12)\n");
fprintf (file, "\tadd 2,2,12\n");
}
fputs ("\t.localentry\t", file);
assemble_name (file, name);
fputs (",.-", file);
assemble_name (file, name);
fputs ("\n", file);
}
/* Output -mprofile-kernel code. This needs to be done here instead of
in output_function_profile since it must go after the ELFv2 ABI
local entry point. */
if (TARGET_PROFILE_KERNEL && crtl->profile)
{
gcc_assert (DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2);
gcc_assert (!TARGET_32BIT);
asm_fprintf (file, "\tmflr %s\n", reg_names[0]);
/* In the ELFv2 ABI we have no compiler stack word. It must be
the resposibility of _mcount to preserve the static chain
register if required. */
if (DEFAULT_ABI != ABI_ELFv2
&& cfun->static_chain_decl != NULL)
{
asm_fprintf (file, "\tstd %s,24(%s)\n",
reg_names[STATIC_CHAIN_REGNUM], reg_names[1]);
fprintf (file, "\tbl %s\n", RS6000_MCOUNT);
asm_fprintf (file, "\tld %s,24(%s)\n",
reg_names[STATIC_CHAIN_REGNUM], reg_names[1]);
}
else
fprintf (file, "\tbl %s\n", RS6000_MCOUNT);
}
rs6000_pic_labelno++;
}
/* -mprofile-kernel code calls mcount before the function prolog,
so a profiled leaf function should stay a leaf function. */
static bool
rs6000_keep_leaf_when_profiled ()
{
return TARGET_PROFILE_KERNEL;
}
/* Non-zero if vmx regs are restored before the frame pop, zero if
we restore after the pop when possible. */
#define ALWAYS_RESTORE_ALTIVEC_BEFORE_POP 0
/* Restoring cr is a two step process: loading a reg from the frame
save, then moving the reg to cr. For ABI_V4 we must let the
unwinder know that the stack location is no longer valid at or
before the stack deallocation, but we can't emit a cfa_restore for
cr at the stack deallocation like we do for other registers.
The trouble is that it is possible for the move to cr to be
scheduled after the stack deallocation. So say exactly where cr
is located on each of the two insns. */
static rtx
load_cr_save (int regno, rtx frame_reg_rtx, int offset, bool exit_func)
{
rtx mem = gen_frame_mem_offset (SImode, frame_reg_rtx, offset);
rtx reg = gen_rtx_REG (SImode, regno);
rtx_insn *insn = emit_move_insn (reg, mem);
if (!exit_func && DEFAULT_ABI == ABI_V4)
{
rtx cr = gen_rtx_REG (SImode, CR2_REGNO);
rtx set = gen_rtx_SET (reg, cr);
add_reg_note (insn, REG_CFA_REGISTER, set);
RTX_FRAME_RELATED_P (insn) = 1;
}
return reg;
}
/* Reload CR from REG. */
static void
restore_saved_cr (rtx reg, int using_mfcr_multiple, bool exit_func)
{
int count = 0;
int i;
if (using_mfcr_multiple)
{
for (i = 0; i < 8; i++)
if (save_reg_p (CR0_REGNO + i))
count++;
gcc_assert (count);
}
if (using_mfcr_multiple && count > 1)
{
rtx_insn *insn;
rtvec p;
int ndx;
p = rtvec_alloc (count);
ndx = 0;
for (i = 0; i < 8; i++)
if (save_reg_p (CR0_REGNO + i))
{
rtvec r = rtvec_alloc (2);
RTVEC_ELT (r, 0) = reg;
RTVEC_ELT (r, 1) = GEN_INT (1 << (7-i));
RTVEC_ELT (p, ndx) =
gen_rtx_SET (gen_rtx_REG (CCmode, CR0_REGNO + i),
gen_rtx_UNSPEC (CCmode, r, UNSPEC_MOVESI_TO_CR));
ndx++;
}
insn = emit_insn (gen_rtx_PARALLEL (VOIDmode, p));
gcc_assert (ndx == count);
/* For the ELFv2 ABI we generate a CFA_RESTORE for each
CR field separately. */
if (!exit_func && DEFAULT_ABI == ABI_ELFv2 && flag_shrink_wrap)
{
for (i = 0; i < 8; i++)
if (save_reg_p (CR0_REGNO + i))
add_reg_note (insn, REG_CFA_RESTORE,
gen_rtx_REG (SImode, CR0_REGNO + i));
RTX_FRAME_RELATED_P (insn) = 1;
}
}
else
for (i = 0; i < 8; i++)
if (save_reg_p (CR0_REGNO + i))
{
rtx insn = emit_insn (gen_movsi_to_cr_one
(gen_rtx_REG (CCmode, CR0_REGNO + i), reg));
/* For the ELFv2 ABI we generate a CFA_RESTORE for each
CR field separately, attached to the insn that in fact
restores this particular CR field. */
if (!exit_func && DEFAULT_ABI == ABI_ELFv2 && flag_shrink_wrap)
{
add_reg_note (insn, REG_CFA_RESTORE,
gen_rtx_REG (SImode, CR0_REGNO + i));
RTX_FRAME_RELATED_P (insn) = 1;
}
}
/* For other ABIs, we just generate a single CFA_RESTORE for CR2. */
if (!exit_func && DEFAULT_ABI != ABI_ELFv2
&& (DEFAULT_ABI == ABI_V4 || flag_shrink_wrap))
{
rtx_insn *insn = get_last_insn ();
rtx cr = gen_rtx_REG (SImode, CR2_REGNO);
add_reg_note (insn, REG_CFA_RESTORE, cr);
RTX_FRAME_RELATED_P (insn) = 1;
}
}
/* Like cr, the move to lr instruction can be scheduled after the
stack deallocation, but unlike cr, its stack frame save is still
valid. So we only need to emit the cfa_restore on the correct
instruction. */
static void
load_lr_save (int regno, rtx frame_reg_rtx, int offset)
{
rtx mem = gen_frame_mem_offset (Pmode, frame_reg_rtx, offset);
rtx reg = gen_rtx_REG (Pmode, regno);
emit_move_insn (reg, mem);
}
static void
restore_saved_lr (int regno, bool exit_func)
{
rtx reg = gen_rtx_REG (Pmode, regno);
rtx lr = gen_rtx_REG (Pmode, LR_REGNO);
rtx_insn *insn = emit_move_insn (lr, reg);
if (!exit_func && flag_shrink_wrap)
{
add_reg_note (insn, REG_CFA_RESTORE, lr);
RTX_FRAME_RELATED_P (insn) = 1;
}
}
static rtx
add_crlr_cfa_restore (const rs6000_stack_t *info, rtx cfa_restores)
{
if (DEFAULT_ABI == ABI_ELFv2)
{
int i;
for (i = 0; i < 8; i++)
if (save_reg_p (CR0_REGNO + i))
{
rtx cr = gen_rtx_REG (SImode, CR0_REGNO + i);
cfa_restores = alloc_reg_note (REG_CFA_RESTORE, cr,
cfa_restores);
}
}
else if (info->cr_save_p)
cfa_restores = alloc_reg_note (REG_CFA_RESTORE,
gen_rtx_REG (SImode, CR2_REGNO),
cfa_restores);
if (info->lr_save_p)
cfa_restores = alloc_reg_note (REG_CFA_RESTORE,
gen_rtx_REG (Pmode, LR_REGNO),
cfa_restores);
return cfa_restores;
}
/* Return true if OFFSET from stack pointer can be clobbered by signals.
V.4 doesn't have any stack cushion, AIX ABIs have 220 or 288 bytes
below stack pointer not cloberred by signals. */
static inline bool
offset_below_red_zone_p (HOST_WIDE_INT offset)
{
return offset < (DEFAULT_ABI == ABI_V4
? 0
: TARGET_32BIT ? -220 : -288);
}
/* Append CFA_RESTORES to any existing REG_NOTES on the last insn. */
static void
emit_cfa_restores (rtx cfa_restores)
{
rtx_insn *insn = get_last_insn ();
rtx *loc = ®_NOTES (insn);
while (*loc)
loc = &XEXP (*loc, 1);
*loc = cfa_restores;
RTX_FRAME_RELATED_P (insn) = 1;
}
/* Emit function epilogue as insns. */
void
rs6000_emit_epilogue (int sibcall)
{
rs6000_stack_t *info;
int restoring_GPRs_inline;
int restoring_FPRs_inline;
int using_load_multiple;
int using_mtcr_multiple;
int use_backchain_to_restore_sp;
int restore_lr;
int strategy;
HOST_WIDE_INT frame_off = 0;
rtx sp_reg_rtx = gen_rtx_REG (Pmode, 1);
rtx frame_reg_rtx = sp_reg_rtx;
rtx cfa_restores = NULL_RTX;
rtx insn;
rtx cr_save_reg = NULL_RTX;
machine_mode reg_mode = Pmode;
int reg_size = TARGET_32BIT ? 4 : 8;
machine_mode fp_reg_mode = (TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT)
? DFmode : SFmode;
int fp_reg_size = 8;
int i;
bool exit_func;
unsigned ptr_regno;
info = rs6000_stack_info ();
strategy = info->savres_strategy;
using_load_multiple = strategy & REST_MULTIPLE;
restoring_FPRs_inline = sibcall || (strategy & REST_INLINE_FPRS);
restoring_GPRs_inline = sibcall || (strategy & REST_INLINE_GPRS);
using_mtcr_multiple = (rs6000_tune == PROCESSOR_PPC601
|| rs6000_tune == PROCESSOR_PPC603
|| rs6000_tune == PROCESSOR_PPC750
|| optimize_size);
/* Restore via the backchain when we have a large frame, since this
is more efficient than an addis, addi pair. The second condition
here will not trigger at the moment; We don't actually need a
frame pointer for alloca, but the generic parts of the compiler
give us one anyway. */
use_backchain_to_restore_sp = (info->total_size + (info->lr_save_p
? info->lr_save_offset
: 0) > 32767
|| (cfun->calls_alloca
&& !frame_pointer_needed));
restore_lr = (info->lr_save_p
&& (restoring_FPRs_inline
|| (strategy & REST_NOINLINE_FPRS_DOESNT_RESTORE_LR))
&& (restoring_GPRs_inline
|| info->first_fp_reg_save < 64)
&& !cfun->machine->lr_is_wrapped_separately);
if (WORLD_SAVE_P (info))
{
int i, j;
char rname[30];
const char *alloc_rname;
rtvec p;
/* eh_rest_world_r10 will return to the location saved in the LR
stack slot (which is not likely to be our caller.)
Input: R10 -- stack adjustment. Clobbers R0, R11, R12, R7, R8.
rest_world is similar, except any R10 parameter is ignored.
The exception-handling stuff that was here in 2.95 is no
longer necessary. */
p = rtvec_alloc (9
+ 32 - info->first_gp_reg_save
+ LAST_ALTIVEC_REGNO + 1 - info->first_altivec_reg_save
+ 63 + 1 - info->first_fp_reg_save);
strcpy (rname, ((crtl->calls_eh_return) ?
"*eh_rest_world_r10" : "*rest_world"));
alloc_rname = ggc_strdup (rname);
j = 0;
RTVEC_ELT (p, j++) = ret_rtx;
RTVEC_ELT (p, j++)
= gen_rtx_USE (VOIDmode, gen_rtx_SYMBOL_REF (Pmode, alloc_rname));
/* The instruction pattern requires a clobber here;
it is shared with the restVEC helper. */
RTVEC_ELT (p, j++)
= gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (Pmode, 11));
{
/* CR register traditionally saved as CR2. */
rtx reg = gen_rtx_REG (SImode, CR2_REGNO);
RTVEC_ELT (p, j++)
= gen_frame_load (reg, frame_reg_rtx, info->cr_save_offset);
if (flag_shrink_wrap)
{
cfa_restores = alloc_reg_note (REG_CFA_RESTORE,
gen_rtx_REG (Pmode, LR_REGNO),
cfa_restores);
cfa_restores = alloc_reg_note (REG_CFA_RESTORE, reg, cfa_restores);
}
}
for (i = 0; i < 32 - info->first_gp_reg_save; i++)
{
rtx reg = gen_rtx_REG (reg_mode, info->first_gp_reg_save + i);
RTVEC_ELT (p, j++)
= gen_frame_load (reg,
frame_reg_rtx, info->gp_save_offset + reg_size * i);
if (flag_shrink_wrap
&& save_reg_p (info->first_gp_reg_save + i))
cfa_restores = alloc_reg_note (REG_CFA_RESTORE, reg, cfa_restores);
}
for (i = 0; info->first_altivec_reg_save + i <= LAST_ALTIVEC_REGNO; i++)
{
rtx reg = gen_rtx_REG (V4SImode, info->first_altivec_reg_save + i);
RTVEC_ELT (p, j++)
= gen_frame_load (reg,
frame_reg_rtx, info->altivec_save_offset + 16 * i);
if (flag_shrink_wrap
&& save_reg_p (info->first_altivec_reg_save + i))
cfa_restores = alloc_reg_note (REG_CFA_RESTORE, reg, cfa_restores);
}
for (i = 0; info->first_fp_reg_save + i <= 63; i++)
{
rtx reg = gen_rtx_REG ((TARGET_HARD_FLOAT && TARGET_DOUBLE_FLOAT
? DFmode : SFmode),
info->first_fp_reg_save + i);
RTVEC_ELT (p, j++)
= gen_frame_load (reg, frame_reg_rtx, info->fp_save_offset + 8 * i);
if (flag_shrink_wrap
&& save_reg_p (info->first_fp_reg_save + i))
cfa_restores = alloc_reg_note (REG_CFA_RESTORE, reg, cfa_restores);
}
RTVEC_ELT (p, j++)
= gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (Pmode, 0));
RTVEC_ELT (p, j++)
= gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (SImode, 12));
RTVEC_ELT (p, j++)
= gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (SImode, 7));
RTVEC_ELT (p, j++)
= gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (SImode, 8));
RTVEC_ELT (p, j++)
= gen_rtx_USE (VOIDmode, gen_rtx_REG (SImode, 10));
insn = emit_jump_insn (gen_rtx_PARALLEL (VOIDmode, p));
if (flag_shrink_wrap)
{
REG_NOTES (insn) = cfa_restores;
add_reg_note (insn, REG_CFA_DEF_CFA, sp_reg_rtx);
RTX_FRAME_RELATED_P (insn) = 1;
}
return;
}
/* frame_reg_rtx + frame_off points to the top of this stack frame. */
if (info->push_p)
frame_off = info->total_size;
/* Restore AltiVec registers if we must do so before adjusting the
stack. */
if (info->altivec_size != 0
&& (ALWAYS_RESTORE_ALTIVEC_BEFORE_POP
|| (DEFAULT_ABI != ABI_V4
&& offset_below_red_zone_p (info->altivec_save_offset))))
{
int i;
int scratch_regno = ptr_regno_for_savres (SAVRES_VR);
gcc_checking_assert (scratch_regno == 11 || scratch_regno == 12);
if (use_backchain_to_restore_sp)
{
int frame_regno = 11;
if ((strategy & REST_INLINE_VRS) == 0)
{
/* Of r11 and r12, select the one not clobbered by an
out-of-line restore function for the frame register. */
frame_regno = 11 + 12 - scratch_regno;
}
frame_reg_rtx = gen_rtx_REG (Pmode, frame_regno);
emit_move_insn (frame_reg_rtx,
gen_rtx_MEM (Pmode, sp_reg_rtx));
frame_off = 0;
}
else if (frame_pointer_needed)
frame_reg_rtx = hard_frame_pointer_rtx;
if ((strategy & REST_INLINE_VRS) == 0)
{
int end_save = info->altivec_save_offset + info->altivec_size;
int ptr_off;
rtx ptr_reg = gen_rtx_REG (Pmode, 0);
rtx scratch_reg = gen_rtx_REG (Pmode, scratch_regno);
if (end_save + frame_off != 0)
{
rtx offset = GEN_INT (end_save + frame_off);
emit_insn (gen_add3_insn (ptr_reg, frame_reg_rtx, offset));
}
else
emit_move_insn (ptr_reg, frame_reg_rtx);
ptr_off = -end_save;
insn = rs6000_emit_savres_rtx (info, scratch_reg,
info->altivec_save_offset + ptr_off,
0, V4SImode, SAVRES_VR);
}
else
{
for (i = info->first_altivec_reg_save; i <= LAST_ALTIVEC_REGNO; ++i)
if (info->vrsave_mask & ALTIVEC_REG_BIT (i))
{
rtx addr, areg, mem, insn;
rtx reg = gen_rtx_REG (V4SImode, i);
HOST_WIDE_INT offset
= (info->altivec_save_offset + frame_off
+ 16 * (i - info->first_altivec_reg_save));
if (TARGET_P9_VECTOR && quad_address_offset_p (offset))
{
mem = gen_frame_mem (V4SImode,
gen_rtx_PLUS (Pmode, frame_reg_rtx,
GEN_INT (offset)));
insn = gen_rtx_SET (reg, mem);
}
else
{
areg = gen_rtx_REG (Pmode, 0);
emit_move_insn (areg, GEN_INT (offset));
/* AltiVec addressing mode is [reg+reg]. */
addr = gen_rtx_PLUS (Pmode, frame_reg_rtx, areg);
mem = gen_frame_mem (V4SImode, addr);
/* Rather than emitting a generic move, force use of the
lvx instruction, which we always want. In particular we
don't want lxvd2x/xxpermdi for little endian. */
insn = gen_altivec_lvx_v4si_internal (reg, mem);
}
(void) emit_insn (insn);
}
}
for (i = info->first_altivec_reg_save; i <= LAST_ALTIVEC_REGNO; ++i)
if (((strategy & REST_INLINE_VRS) == 0
|| (info->vrsave_mask & ALTIVEC_REG_BIT (i)) != 0)
&& (flag_shrink_wrap
|| (offset_below_red_zone_p
(info->altivec_save_offset
+ 16 * (i - info->first_altivec_reg_save))))
&& save_reg_p (i))
{
rtx reg = gen_rtx_REG (V4SImode, i);
cfa_restores = alloc_reg_note (REG_CFA_RESTORE, reg, cfa_restores);
}
}
/* Restore VRSAVE if we must do so before adjusting the stack. */
if (info->vrsave_size != 0
&& (ALWAYS_RESTORE_ALTIVEC_BEFORE_POP
|| (DEFAULT_ABI != ABI_V4
&& offset_below_red_zone_p (info->vrsave_save_offset))))
{
rtx reg;
if (frame_reg_rtx == sp_reg_rtx)
{
if (use_backchain_to_restore_sp)
{
frame_reg_rtx = gen_rtx_REG (Pmode, 11);
emit_move_insn (frame_reg_rtx,
gen_rtx_MEM (Pmode, sp_reg_rtx));
frame_off = 0;
}
else if (frame_pointer_needed)
frame_reg_rtx = hard_frame_pointer_rtx;
}
reg = gen_rtx_REG (SImode, 12);
emit_insn (gen_frame_load (reg, frame_reg_rtx,
info->vrsave_save_offset + frame_off));
emit_insn (generate_set_vrsave (reg, info, 1));
}
insn = NULL_RTX;
/* If we have a large stack frame, restore the old stack pointer
using the backchain. */
if (use_backchain_to_restore_sp)
{
if (frame_reg_rtx == sp_reg_rtx)
{
/* Under V.4, don't reset the stack pointer until after we're done
loading the saved registers. */
if (DEFAULT_ABI == ABI_V4)
frame_reg_rtx = gen_rtx_REG (Pmode, 11);
insn = emit_move_insn (frame_reg_rtx,
gen_rtx_MEM (Pmode, sp_reg_rtx));
frame_off = 0;
}
else if (ALWAYS_RESTORE_ALTIVEC_BEFORE_POP
&& DEFAULT_ABI == ABI_V4)
/* frame_reg_rtx has been set up by the altivec restore. */
;
else
{
insn = emit_move_insn (sp_reg_rtx, frame_reg_rtx);
frame_reg_rtx = sp_reg_rtx;
}
}
/* If we have a frame pointer, we can restore the old stack pointer
from it. */
else if (frame_pointer_needed)
{
frame_reg_rtx = sp_reg_rtx;
if (DEFAULT_ABI == ABI_V4)
frame_reg_rtx = gen_rtx_REG (Pmode, 11);
/* Prevent reordering memory accesses against stack pointer restore. */
else if (cfun->calls_alloca
|| offset_below_red_zone_p (-info->total_size))
rs6000_emit_stack_tie (frame_reg_rtx, true);
insn = emit_insn (gen_add3_insn (frame_reg_rtx, hard_frame_pointer_rtx,
GEN_INT (info->total_size)));
frame_off = 0;
}
else if (info->push_p
&& DEFAULT_ABI != ABI_V4
&& !crtl->calls_eh_return)
{
/* Prevent reordering memory accesses against stack pointer restore. */
if (cfun->calls_alloca
|| offset_below_red_zone_p (-info->total_size))
rs6000_emit_stack_tie (frame_reg_rtx, false);
insn = emit_insn (gen_add3_insn (sp_reg_rtx, sp_reg_rtx,
GEN_INT (info->total_size)));
frame_off = 0;
}
if (insn && frame_reg_rtx == sp_reg_rtx)
{
if (cfa_restores)
{
REG_NOTES (insn) = cfa_restores;
cfa_restores = NULL_RTX;
}
add_reg_note (insn, REG_CFA_DEF_CFA, sp_reg_rtx);
RTX_FRAME_RELATED_P (insn) = 1;
}
/* Restore AltiVec registers if we have not done so already. */
if (!ALWAYS_RESTORE_ALTIVEC_BEFORE_POP
&& info->altivec_size != 0
&& (DEFAULT_ABI == ABI_V4
|| !offset_below_red_zone_p (info->altivec_save_offset)))
{
int i;
if ((strategy & REST_INLINE_VRS) == 0)
{
int end_save = info->altivec_save_offset + info->altivec_size;
int ptr_off;
rtx ptr_reg = gen_rtx_REG (Pmode, 0);
int scratch_regno = ptr_regno_for_savres (SAVRES_VR);
rtx scratch_reg = gen_rtx_REG (Pmode, scratch_regno);
if (end_save + frame_off != 0)
{
rtx offset = GEN_INT (end_save + frame_off);
emit_insn (gen_add3_insn (ptr_reg, frame_reg_rtx, offset));
}
else
emit_move_insn (ptr_reg, frame_reg_rtx);
ptr_off = -end_save;
insn = rs6000_emit_savres_rtx (info, scratch_reg,
info->altivec_save_offset + ptr_off,
0, V4SImode, SAVRES_VR);
if (REGNO (frame_reg_rtx) == REGNO (scratch_reg))
{
/* Frame reg was clobbered by out-of-line save. Restore it
from ptr_reg, and if we are calling out-of-line gpr or
fpr restore set up the correct pointer and offset. */
unsigned newptr_regno = 1;
if (!restoring_GPRs_inline)
{
bool lr = info->gp_save_offset + info->gp_size == 0;
int sel = SAVRES_GPR | (lr ? SAVRES_LR : 0);
newptr_regno = ptr_regno_for_savres (sel);
end_save = info->gp_save_offset + info->gp_size;
}
else if (!restoring_FPRs_inline)
{
bool lr = !(strategy & REST_NOINLINE_FPRS_DOESNT_RESTORE_LR);
int sel = SAVRES_FPR | (lr ? SAVRES_LR : 0);
newptr_regno = ptr_regno_for_savres (sel);
end_save = info->fp_save_offset + info->fp_size;
}
if (newptr_regno != 1 && REGNO (frame_reg_rtx) != newptr_regno)
frame_reg_rtx = gen_rtx_REG (Pmode, newptr_regno);
if (end_save + ptr_off != 0)
{
rtx offset = GEN_INT (end_save + ptr_off);
frame_off = -end_save;
if (TARGET_32BIT)
emit_insn (gen_addsi3_carry (frame_reg_rtx,
ptr_reg, offset));
else
emit_insn (gen_adddi3_carry (frame_reg_rtx,
ptr_reg, offset));
}
else
{
frame_off = ptr_off;
emit_move_insn (frame_reg_rtx, ptr_reg);
}
}
}
else
{
for (i = info->first_altivec_reg_save; i <= LAST_ALTIVEC_REGNO; ++i)
if (info->vrsave_mask & ALTIVEC_REG_BIT (i))
{
rtx addr, areg, mem, insn;
rtx reg = gen_rtx_REG (V4SImode, i);
HOST_WIDE_INT offset
= (info->altivec_save_offset + frame_off
+ 16 * (i - info->first_altivec_reg_save));
if (TARGET_P9_VECTOR && quad_address_offset_p (offset))
{
mem = gen_frame_mem (V4SImode,
gen_rtx_PLUS (Pmode, frame_reg_rtx,
GEN_INT (offset)));
insn = gen_rtx_SET (reg, mem);
}
else
{
areg = gen_rtx_REG (Pmode, 0);
emit_move_insn (areg, GEN_INT (offset));
/* AltiVec addressing mode is [reg+reg]. */
addr = gen_rtx_PLUS (Pmode, frame_reg_rtx, areg);
mem = gen_frame_mem (V4SImode, addr);
/* Rather than emitting a generic move, force use of the
lvx instruction, which we always want. In particular we
don't want lxvd2x/xxpermdi for little endian. */
insn = gen_altivec_lvx_v4si_internal (reg, mem);
}
(void) emit_insn (insn);
}
}
for (i = info->first_altivec_reg_save; i <= LAST_ALTIVEC_REGNO; ++i)
if (((strategy & REST_INLINE_VRS) == 0
|| (info->vrsave_mask & ALTIVEC_REG_BIT (i)) != 0)
&& (DEFAULT_ABI == ABI_V4 || flag_shrink_wrap)
&& save_reg_p (i))
{
rtx reg = gen_rtx_REG (V4SImode, i);
cfa_restores = alloc_reg_note (REG_CFA_RESTORE, reg, cfa_restores);
}
}
/* Restore VRSAVE if we have not done so already. */
if (!ALWAYS_RESTORE_ALTIVEC_BEFORE_POP
&& info->vrsave_size != 0
&& (DEFAULT_ABI == ABI_V4
|| !offset_below_red_zone_p (info->vrsave_save_offset)))
{
rtx reg;
reg = gen_rtx_REG (SImode, 12);
emit_insn (gen_frame_load (reg, frame_reg_rtx,
info->vrsave_save_offset + frame_off));
emit_insn (generate_set_vrsave (reg, info, 1));
}
/* If we exit by an out-of-line restore function on ABI_V4 then that
function will deallocate the stack, so we don't need to worry
about the unwinder restoring cr from an invalid stack frame
location. */
exit_func = (!restoring_FPRs_inline
|| (!restoring_GPRs_inline
&& info->first_fp_reg_save == 64));
/* In the ELFv2 ABI we need to restore all call-saved CR fields from
*separate* slots if the routine calls __builtin_eh_return, so
that they can be independently restored by the unwinder. */
if (DEFAULT_ABI == ABI_ELFv2 && crtl->calls_eh_return)
{
int i, cr_off = info->ehcr_offset;
for (i = 0; i < 8; i++)
if (!call_used_regs[CR0_REGNO + i])
{
rtx reg = gen_rtx_REG (SImode, 0);
emit_insn (gen_frame_load (reg, frame_reg_rtx,
cr_off + frame_off));
insn = emit_insn (gen_movsi_to_cr_one
(gen_rtx_REG (CCmode, CR0_REGNO + i), reg));
if (!exit_func && flag_shrink_wrap)
{
add_reg_note (insn, REG_CFA_RESTORE,
gen_rtx_REG (SImode, CR0_REGNO + i));
RTX_FRAME_RELATED_P (insn) = 1;
}
cr_off += reg_size;
}
}
/* Get the old lr if we saved it. If we are restoring registers
out-of-line, then the out-of-line routines can do this for us. */
if (restore_lr && restoring_GPRs_inline)
load_lr_save (0, frame_reg_rtx, info->lr_save_offset + frame_off);
/* Get the old cr if we saved it. */
if (info->cr_save_p)
{
unsigned cr_save_regno = 12;
if (!restoring_GPRs_inline)
{
/* Ensure we don't use the register used by the out-of-line
gpr register restore below. */
bool lr = info->gp_save_offset + info->gp_size == 0;
int sel = SAVRES_GPR | (lr ? SAVRES_LR : 0);
int gpr_ptr_regno = ptr_regno_for_savres (sel);
if (gpr_ptr_regno == 12)
cr_save_regno = 11;
gcc_checking_assert (REGNO (frame_reg_rtx) != cr_save_regno);
}
else if (REGNO (frame_reg_rtx) == 12)
cr_save_regno = 11;
cr_save_reg = load_cr_save (cr_save_regno, frame_reg_rtx,
info->cr_save_offset + frame_off,
exit_func);
}
/* Set LR here to try to overlap restores below. */
if (restore_lr && restoring_GPRs_inline)
restore_saved_lr (0, exit_func);
/* Load exception handler data registers, if needed. */
if (crtl->calls_eh_return)
{
unsigned int i, regno;
if (TARGET_AIX)
{
rtx reg = gen_rtx_REG (reg_mode, 2);
emit_insn (gen_frame_load (reg, frame_reg_rtx,
frame_off + RS6000_TOC_SAVE_SLOT));
}
for (i = 0; ; ++i)
{
rtx mem;
regno = EH_RETURN_DATA_REGNO (i);
if (regno == INVALID_REGNUM)
break;
mem = gen_frame_mem_offset (reg_mode, frame_reg_rtx,
info->ehrd_offset + frame_off
+ reg_size * (int) i);
emit_move_insn (gen_rtx_REG (reg_mode, regno), mem);
}
}
/* Restore GPRs. This is done as a PARALLEL if we are using
the load-multiple instructions. */
if (!restoring_GPRs_inline)
{
/* We are jumping to an out-of-line function. */
rtx ptr_reg;
int end_save = info->gp_save_offset + info->gp_size;
bool can_use_exit = end_save == 0;
int sel = SAVRES_GPR | (can_use_exit ? SAVRES_LR : 0);
int ptr_off;
/* Emit stack reset code if we need it. */
ptr_regno = ptr_regno_for_savres (sel);
ptr_reg = gen_rtx_REG (Pmode, ptr_regno);
if (can_use_exit)
rs6000_emit_stack_reset (frame_reg_rtx, frame_off, ptr_regno);
else if (end_save + frame_off != 0)
emit_insn (gen_add3_insn (ptr_reg, frame_reg_rtx,
GEN_INT (end_save + frame_off)));
else if (REGNO (frame_reg_rtx) != ptr_regno)
emit_move_insn (ptr_reg, frame_reg_rtx);
if (REGNO (frame_reg_rtx) == ptr_regno)
frame_off = -end_save;
if (can_use_exit && info->cr_save_p)
restore_saved_cr (cr_save_reg, using_mtcr_multiple, true);
ptr_off = -end_save;
rs6000_emit_savres_rtx (info, ptr_reg,
info->gp_save_offset + ptr_off,
info->lr_save_offset + ptr_off,
reg_mode, sel);
}
else if (using_load_multiple)
{
rtvec p;
p = rtvec_alloc (32 - info->first_gp_reg_save);
for (i = 0; i < 32 - info->first_gp_reg_save; i++)
RTVEC_ELT (p, i)
= gen_frame_load (gen_rtx_REG (reg_mode, info->first_gp_reg_save + i),
frame_reg_rtx,
info->gp_save_offset + frame_off + reg_size * i);
emit_insn (gen_rtx_PARALLEL (VOIDmode, p));
}
else
{
int offset = info->gp_save_offset + frame_off;
for (i = info->first_gp_reg_save; i < 32; i++)
{
if (save_reg_p (i)
&& !cfun->machine->gpr_is_wrapped_separately[i])
{
rtx reg = gen_rtx_REG (reg_mode, i);
emit_insn (gen_frame_load (reg, frame_reg_rtx, offset));
}
offset += reg_size;
}
}
if (DEFAULT_ABI == ABI_V4 || flag_shrink_wrap)
{
/* If the frame pointer was used then we can't delay emitting
a REG_CFA_DEF_CFA note. This must happen on the insn that
restores the frame pointer, r31. We may have already emitted
a REG_CFA_DEF_CFA note, but that's OK; A duplicate is
discarded by dwarf2cfi.c/dwarf2out.c, and in any case would
be harmless if emitted. */
if (frame_pointer_needed)
{
insn = get_last_insn ();
add_reg_note (insn, REG_CFA_DEF_CFA,
plus_constant (Pmode, frame_reg_rtx, frame_off));
RTX_FRAME_RELATED_P (insn) = 1;
}
/* Set up cfa_restores. We always need these when
shrink-wrapping. If not shrink-wrapping then we only need
the cfa_restore when the stack location is no longer valid.
The cfa_restores must be emitted on or before the insn that
invalidates the stack, and of course must not be emitted
before the insn that actually does the restore. The latter
is why it is a bad idea to emit the cfa_restores as a group
on the last instruction here that actually does a restore:
That insn may be reordered with respect to others doing
restores. */
if (flag_shrink_wrap
&& !restoring_GPRs_inline
&& info->first_fp_reg_save == 64)
cfa_restores = add_crlr_cfa_restore (info, cfa_restores);
for (i = info->first_gp_reg_save; i < 32; i++)
if (save_reg_p (i)
&& !cfun->machine->gpr_is_wrapped_separately[i])
{
rtx reg = gen_rtx_REG (reg_mode, i);
cfa_restores = alloc_reg_note (REG_CFA_RESTORE, reg, cfa_restores);
}
}
if (!restoring_GPRs_inline
&& info->first_fp_reg_save == 64)
{
/* We are jumping to an out-of-line function. */
if (cfa_restores)
emit_cfa_restores (cfa_restores);
return;
}
if (restore_lr && !restoring_GPRs_inline)
{
load_lr_save (0, frame_reg_rtx, info->lr_save_offset + frame_off);
restore_saved_lr (0, exit_func);
}
/* Restore fpr's if we need to do it without calling a function. */
if (restoring_FPRs_inline)
{
int offset = info->fp_save_offset + frame_off;
for (i = info->first_fp_reg_save; i < 64; i++)
{
if (save_reg_p (i)
&& !cfun->machine->fpr_is_wrapped_separately[i - 32])
{
rtx reg = gen_rtx_REG (fp_reg_mode, i);
emit_insn (gen_frame_load (reg, frame_reg_rtx, offset));
if (DEFAULT_ABI == ABI_V4 || flag_shrink_wrap)
cfa_restores = alloc_reg_note (REG_CFA_RESTORE, reg,
cfa_restores);
}
offset += fp_reg_size;
}
}
/* If we saved cr, restore it here. Just those that were used. */
if (info->cr_save_p)
restore_saved_cr (cr_save_reg, using_mtcr_multiple, exit_func);
/* If this is V.4, unwind the stack pointer after all of the loads
have been done, or set up r11 if we are restoring fp out of line. */
ptr_regno = 1;
if (!restoring_FPRs_inline)
{
bool lr = (strategy & REST_NOINLINE_FPRS_DOESNT_RESTORE_LR) == 0;
int sel = SAVRES_FPR | (lr ? SAVRES_LR : 0);
ptr_regno = ptr_regno_for_savres (sel);
}
insn = rs6000_emit_stack_reset (frame_reg_rtx, frame_off, ptr_regno);
if (REGNO (frame_reg_rtx) == ptr_regno)
frame_off = 0;
if (insn && restoring_FPRs_inline)
{
if (cfa_restores)
{
REG_NOTES (insn) = cfa_restores;
cfa_restores = NULL_RTX;
}
add_reg_note (insn, REG_CFA_DEF_CFA, sp_reg_rtx);
RTX_FRAME_RELATED_P (insn) = 1;
}
if (crtl->calls_eh_return)
{
rtx sa = EH_RETURN_STACKADJ_RTX;
emit_insn (gen_add3_insn (sp_reg_rtx, sp_reg_rtx, sa));
}
if (!sibcall && restoring_FPRs_inline)
{
if (cfa_restores)
{
/* We can't hang the cfa_restores off a simple return,
since the shrink-wrap code sometimes uses an existing
return. This means there might be a path from
pre-prologue code to this return, and dwarf2cfi code
wants the eh_frame unwinder state to be the same on
all paths to any point. So we need to emit the
cfa_restores before the return. For -m64 we really
don't need epilogue cfa_restores at all, except for
this irritating dwarf2cfi with shrink-wrap
requirement; The stack red-zone means eh_frame info
from the prologue telling the unwinder to restore
from the stack is perfectly good right to the end of
the function. */
emit_insn (gen_blockage ());
emit_cfa_restores (cfa_restores);
cfa_restores = NULL_RTX;
}
emit_jump_insn (targetm.gen_simple_return ());
}
if (!sibcall && !restoring_FPRs_inline)
{
bool lr = (strategy & REST_NOINLINE_FPRS_DOESNT_RESTORE_LR) == 0;
rtvec p = rtvec_alloc (3 + !!lr + 64 - info->first_fp_reg_save);
int elt = 0;
RTVEC_ELT (p, elt++) = ret_rtx;
if (lr)
RTVEC_ELT (p, elt++)
= gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (Pmode, LR_REGNO));
/* We have to restore more than two FP registers, so branch to the
restore function. It will return to our caller. */
int i;
int reg;
rtx sym;
if (flag_shrink_wrap)
cfa_restores = add_crlr_cfa_restore (info, cfa_restores);
sym = rs6000_savres_routine_sym (info, SAVRES_FPR | (lr ? SAVRES_LR : 0));
RTVEC_ELT (p, elt++) = gen_rtx_USE (VOIDmode, sym);
reg = (DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)? 1 : 11;
RTVEC_ELT (p, elt++) = gen_rtx_USE (VOIDmode, gen_rtx_REG (Pmode, reg));
for (i = 0; i < 64 - info->first_fp_reg_save; i++)
{
rtx reg = gen_rtx_REG (DFmode, info->first_fp_reg_save + i);
RTVEC_ELT (p, elt++)
= gen_frame_load (reg, sp_reg_rtx, info->fp_save_offset + 8 * i);
if (flag_shrink_wrap
&& save_reg_p (info->first_fp_reg_save + i))
cfa_restores = alloc_reg_note (REG_CFA_RESTORE, reg, cfa_restores);
}
emit_jump_insn (gen_rtx_PARALLEL (VOIDmode, p));
}
if (cfa_restores)
{
if (sibcall)
/* Ensure the cfa_restores are hung off an insn that won't
be reordered above other restores. */
emit_insn (gen_blockage ());
emit_cfa_restores (cfa_restores);
}
}
/* Write function epilogue. */
static void
rs6000_output_function_epilogue (FILE *file)
{
#if TARGET_MACHO
macho_branch_islands ();
{
rtx_insn *insn = get_last_insn ();
rtx_insn *deleted_debug_label = NULL;
/* Mach-O doesn't support labels at the end of objects, so if
it looks like we might want one, take special action.
First, collect any sequence of deleted debug labels. */
while (insn
&& NOTE_P (insn)
&& NOTE_KIND (insn) != NOTE_INSN_DELETED_LABEL)
{
/* Don't insert a nop for NOTE_INSN_DELETED_DEBUG_LABEL
notes only, instead set their CODE_LABEL_NUMBER to -1,
otherwise there would be code generation differences
in between -g and -g0. */
if (NOTE_P (insn) && NOTE_KIND (insn) == NOTE_INSN_DELETED_DEBUG_LABEL)
deleted_debug_label = insn;
insn = PREV_INSN (insn);
}
/* Second, if we have:
label:
barrier
then this needs to be detected, so skip past the barrier. */
if (insn && BARRIER_P (insn))
insn = PREV_INSN (insn);
/* Up to now we've only seen notes or barriers. */
if (insn)
{
if (LABEL_P (insn)
|| (NOTE_P (insn)
&& NOTE_KIND (insn) == NOTE_INSN_DELETED_LABEL))
/* Trailing label: <barrier>. */
fputs ("\tnop\n", file);
else
{
/* Lastly, see if we have a completely empty function body. */
while (insn && ! INSN_P (insn))
insn = PREV_INSN (insn);
/* If we don't find any insns, we've got an empty function body;
I.e. completely empty - without a return or branch. This is
taken as the case where a function body has been removed
because it contains an inline __builtin_unreachable(). GCC
states that reaching __builtin_unreachable() means UB so we're
not obliged to do anything special; however, we want
non-zero-sized function bodies. To meet this, and help the
user out, let's trap the case. */
if (insn == NULL)
fputs ("\ttrap\n", file);
}
}
else if (deleted_debug_label)
for (insn = deleted_debug_label; insn; insn = NEXT_INSN (insn))
if (NOTE_KIND (insn) == NOTE_INSN_DELETED_DEBUG_LABEL)
CODE_LABEL_NUMBER (insn) = -1;
}
#endif
/* Output a traceback table here. See /usr/include/sys/debug.h for info
on its format.
We don't output a traceback table if -finhibit-size-directive was
used. The documentation for -finhibit-size-directive reads
``don't output a @code{.size} assembler directive, or anything
else that would cause trouble if the function is split in the
middle, and the two halves are placed at locations far apart in
memory.'' The traceback table has this property, since it
includes the offset from the start of the function to the
traceback table itself.
System V.4 Powerpc's (and the embedded ABI derived from it) use a
different traceback table. */
if ((DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
&& ! flag_inhibit_size_directive
&& rs6000_traceback != traceback_none && !cfun->is_thunk)
{
const char *fname = NULL;
const char *language_string = lang_hooks.name;
int fixed_parms = 0, float_parms = 0, parm_info = 0;
int i;
int optional_tbtab;
rs6000_stack_t *info = rs6000_stack_info ();
if (rs6000_traceback == traceback_full)
optional_tbtab = 1;
else if (rs6000_traceback == traceback_part)
optional_tbtab = 0;
else
optional_tbtab = !optimize_size && !TARGET_ELF;
if (optional_tbtab)
{
fname = XSTR (XEXP (DECL_RTL (current_function_decl), 0), 0);
while (*fname == '.') /* V.4 encodes . in the name */
fname++;
/* Need label immediately before tbtab, so we can compute
its offset from the function start. */
ASM_OUTPUT_INTERNAL_LABEL_PREFIX (file, "LT");
ASM_OUTPUT_LABEL (file, fname);
}
/* The .tbtab pseudo-op can only be used for the first eight
expressions, since it can't handle the possibly variable
length fields that follow. However, if you omit the optional
fields, the assembler outputs zeros for all optional fields
anyways, giving each variable length field is minimum length
(as defined in sys/debug.h). Thus we can not use the .tbtab
pseudo-op at all. */
/* An all-zero word flags the start of the tbtab, for debuggers
that have to find it by searching forward from the entry
point or from the current pc. */
fputs ("\t.long 0\n", file);
/* Tbtab format type. Use format type 0. */
fputs ("\t.byte 0,", file);
/* Language type. Unfortunately, there does not seem to be any
official way to discover the language being compiled, so we
use language_string.
C is 0. Fortran is 1. Pascal is 2. Ada is 3. C++ is 9.
Java is 13. Objective-C is 14. Objective-C++ isn't assigned
a number, so for now use 9. LTO, Go and JIT aren't assigned numbers
either, so for now use 0. */
if (lang_GNU_C ()
|| ! strcmp (language_string, "GNU GIMPLE")
|| ! strcmp (language_string, "GNU Go")
|| ! strcmp (language_string, "libgccjit"))
i = 0;
else if (! strcmp (language_string, "GNU F77")
|| lang_GNU_Fortran ())
i = 1;
else if (! strcmp (language_string, "GNU Pascal"))
i = 2;
else if (! strcmp (language_string, "GNU Ada"))
i = 3;
else if (lang_GNU_CXX ()
|| ! strcmp (language_string, "GNU Objective-C++"))
i = 9;
else if (! strcmp (language_string, "GNU Java"))
i = 13;
else if (! strcmp (language_string, "GNU Objective-C"))
i = 14;
else
gcc_unreachable ();
fprintf (file, "%d,", i);
/* 8 single bit fields: global linkage (not set for C extern linkage,
apparently a PL/I convention?), out-of-line epilogue/prologue, offset
from start of procedure stored in tbtab, internal function, function
has controlled storage, function has no toc, function uses fp,
function logs/aborts fp operations. */
/* Assume that fp operations are used if any fp reg must be saved. */
fprintf (file, "%d,",
(optional_tbtab << 5) | ((info->first_fp_reg_save != 64) << 1));
/* 6 bitfields: function is interrupt handler, name present in
proc table, function calls alloca, on condition directives
(controls stack walks, 3 bits), saves condition reg, saves
link reg. */
/* The `function calls alloca' bit seems to be set whenever reg 31 is
set up as a frame pointer, even when there is no alloca call. */
fprintf (file, "%d,",
((optional_tbtab << 6)
| ((optional_tbtab & frame_pointer_needed) << 5)
| (info->cr_save_p << 1)
| (info->lr_save_p)));
/* 3 bitfields: saves backchain, fixup code, number of fpr saved
(6 bits). */
fprintf (file, "%d,",
(info->push_p << 7) | (64 - info->first_fp_reg_save));
/* 2 bitfields: spare bits (2 bits), number of gpr saved (6 bits). */
fprintf (file, "%d,", (32 - first_reg_to_save ()));
if (optional_tbtab)
{
/* Compute the parameter info from the function decl argument
list. */
tree decl;
int next_parm_info_bit = 31;
for (decl = DECL_ARGUMENTS (current_function_decl);
decl; decl = DECL_CHAIN (decl))
{
rtx parameter = DECL_INCOMING_RTL (decl);
machine_mode mode = GET_MODE (parameter);
if (GET_CODE (parameter) == REG)
{
if (SCALAR_FLOAT_MODE_P (mode))
{
int bits;
float_parms++;
switch (mode)
{
case E_SFmode:
case E_SDmode:
bits = 0x2;
break;
case E_DFmode:
case E_DDmode:
case E_TFmode:
case E_TDmode:
case E_IFmode:
case E_KFmode:
bits = 0x3;
break;
default:
gcc_unreachable ();
}
/* If only one bit will fit, don't or in this entry. */
if (next_parm_info_bit > 0)
parm_info |= (bits << (next_parm_info_bit - 1));
next_parm_info_bit -= 2;
}
else
{
fixed_parms += ((GET_MODE_SIZE (mode)
+ (UNITS_PER_WORD - 1))
/ UNITS_PER_WORD);
next_parm_info_bit -= 1;
}
}
}
}
/* Number of fixed point parameters. */
/* This is actually the number of words of fixed point parameters; thus
an 8 byte struct counts as 2; and thus the maximum value is 8. */
fprintf (file, "%d,", fixed_parms);
/* 2 bitfields: number of floating point parameters (7 bits), parameters
all on stack. */
/* This is actually the number of fp registers that hold parameters;
and thus the maximum value is 13. */
/* Set parameters on stack bit if parameters are not in their original
registers, regardless of whether they are on the stack? Xlc
seems to set the bit when not optimizing. */
fprintf (file, "%d\n", ((float_parms << 1) | (! optimize)));
if (optional_tbtab)
{
/* Optional fields follow. Some are variable length. */
/* Parameter types, left adjusted bit fields: 0 fixed, 10 single
float, 11 double float. */
/* There is an entry for each parameter in a register, in the order
that they occur in the parameter list. Any intervening arguments
on the stack are ignored. If the list overflows a long (max
possible length 34 bits) then completely leave off all elements
that don't fit. */
/* Only emit this long if there was at least one parameter. */
if (fixed_parms || float_parms)
fprintf (file, "\t.long %d\n", parm_info);
/* Offset from start of code to tb table. */
fputs ("\t.long ", file);
ASM_OUTPUT_INTERNAL_LABEL_PREFIX (file, "LT");
RS6000_OUTPUT_BASENAME (file, fname);
putc ('-', file);
rs6000_output_function_entry (file, fname);
putc ('\n', file);
/* Interrupt handler mask. */
/* Omit this long, since we never set the interrupt handler bit
above. */
/* Number of CTL (controlled storage) anchors. */
/* Omit this long, since the has_ctl bit is never set above. */
/* Displacement into stack of each CTL anchor. */
/* Omit this list of longs, because there are no CTL anchors. */
/* Length of function name. */
if (*fname == '*')
++fname;
fprintf (file, "\t.short %d\n", (int) strlen (fname));
/* Function name. */
assemble_string (fname, strlen (fname));
/* Register for alloca automatic storage; this is always reg 31.
Only emit this if the alloca bit was set above. */
if (frame_pointer_needed)
fputs ("\t.byte 31\n", file);
fputs ("\t.align 2\n", file);
}
}
/* Arrange to define .LCTOC1 label, if not already done. */
if (need_toc_init)
{
need_toc_init = 0;
if (!toc_initialized)
{
switch_to_section (toc_section);
switch_to_section (current_function_section ());
}
}
}
/* -fsplit-stack support. */
/* A SYMBOL_REF for __morestack. */
static GTY(()) rtx morestack_ref;
static rtx
gen_add3_const (rtx rt, rtx ra, long c)
{
if (TARGET_64BIT)
return gen_adddi3 (rt, ra, GEN_INT (c));
else
return gen_addsi3 (rt, ra, GEN_INT (c));
}
/* Emit -fsplit-stack prologue, which goes before the regular function
prologue (at local entry point in the case of ELFv2). */
void
rs6000_expand_split_stack_prologue (void)
{
rs6000_stack_t *info = rs6000_stack_info ();
unsigned HOST_WIDE_INT allocate;
long alloc_hi, alloc_lo;
rtx r0, r1, r12, lr, ok_label, compare, jump, call_fusage;
rtx_insn *insn;
gcc_assert (flag_split_stack && reload_completed);
if (!info->push_p)
return;
if (global_regs[29])
{
error ("%qs uses register r29", "-fsplit-stack");
inform (DECL_SOURCE_LOCATION (global_regs_decl[29]),
"conflicts with %qD", global_regs_decl[29]);
}
allocate = info->total_size;
if (allocate > (unsigned HOST_WIDE_INT) 1 << 31)
{
sorry ("Stack frame larger than 2G is not supported for -fsplit-stack");
return;
}
if (morestack_ref == NULL_RTX)
{
morestack_ref = gen_rtx_SYMBOL_REF (Pmode, "__morestack");
SYMBOL_REF_FLAGS (morestack_ref) |= (SYMBOL_FLAG_LOCAL
| SYMBOL_FLAG_FUNCTION);
}
r0 = gen_rtx_REG (Pmode, 0);
r1 = gen_rtx_REG (Pmode, STACK_POINTER_REGNUM);
r12 = gen_rtx_REG (Pmode, 12);
emit_insn (gen_load_split_stack_limit (r0));
/* Always emit two insns here to calculate the requested stack,
so that the linker can edit them when adjusting size for calling
non-split-stack code. */
alloc_hi = (-allocate + 0x8000) & ~0xffffL;
alloc_lo = -allocate - alloc_hi;
if (alloc_hi != 0)
{
emit_insn (gen_add3_const (r12, r1, alloc_hi));
if (alloc_lo != 0)
emit_insn (gen_add3_const (r12, r12, alloc_lo));
else
emit_insn (gen_nop ());
}
else
{
emit_insn (gen_add3_const (r12, r1, alloc_lo));
emit_insn (gen_nop ());
}
compare = gen_rtx_REG (CCUNSmode, CR7_REGNO);
emit_insn (gen_rtx_SET (compare, gen_rtx_COMPARE (CCUNSmode, r12, r0)));
ok_label = gen_label_rtx ();
jump = gen_rtx_IF_THEN_ELSE (VOIDmode,
gen_rtx_GEU (VOIDmode, compare, const0_rtx),
gen_rtx_LABEL_REF (VOIDmode, ok_label),
pc_rtx);
insn = emit_jump_insn (gen_rtx_SET (pc_rtx, jump));
JUMP_LABEL (insn) = ok_label;
/* Mark the jump as very likely to be taken. */
add_reg_br_prob_note (insn, profile_probability::very_likely ());
lr = gen_rtx_REG (Pmode, LR_REGNO);
insn = emit_move_insn (r0, lr);
RTX_FRAME_RELATED_P (insn) = 1;
insn = emit_insn (gen_frame_store (r0, r1, info->lr_save_offset));
RTX_FRAME_RELATED_P (insn) = 1;
insn = emit_call_insn (gen_call (gen_rtx_MEM (SImode, morestack_ref),
const0_rtx, const0_rtx));
call_fusage = NULL_RTX;
use_reg (&call_fusage, r12);
/* Say the call uses r0, even though it doesn't, to stop regrename
from twiddling with the insns saving lr, trashing args for cfun.
The insns restoring lr are similarly protected by making
split_stack_return use r0. */
use_reg (&call_fusage, r0);
add_function_usage_to (insn, call_fusage);
/* Indicate that this function can't jump to non-local gotos. */
make_reg_eh_region_note_nothrow_nononlocal (insn);
emit_insn (gen_frame_load (r0, r1, info->lr_save_offset));
insn = emit_move_insn (lr, r0);
add_reg_note (insn, REG_CFA_RESTORE, lr);
RTX_FRAME_RELATED_P (insn) = 1;
emit_insn (gen_split_stack_return ());
emit_label (ok_label);
LABEL_NUSES (ok_label) = 1;
}
/* Return the internal arg pointer used for function incoming
arguments. When -fsplit-stack, the arg pointer is r12 so we need
to copy it to a pseudo in order for it to be preserved over calls
and suchlike. We'd really like to use a pseudo here for the
internal arg pointer but data-flow analysis is not prepared to
accept pseudos as live at the beginning of a function. */
static rtx
rs6000_internal_arg_pointer (void)
{
if (flag_split_stack
&& (lookup_attribute ("no_split_stack", DECL_ATTRIBUTES (cfun->decl))
== NULL))
{
if (cfun->machine->split_stack_arg_pointer == NULL_RTX)
{
rtx pat;
cfun->machine->split_stack_arg_pointer = gen_reg_rtx (Pmode);
REG_POINTER (cfun->machine->split_stack_arg_pointer) = 1;
/* Put the pseudo initialization right after the note at the
beginning of the function. */
pat = gen_rtx_SET (cfun->machine->split_stack_arg_pointer,
gen_rtx_REG (Pmode, 12));
push_topmost_sequence ();
emit_insn_after (pat, get_insns ());
pop_topmost_sequence ();
}
return plus_constant (Pmode, cfun->machine->split_stack_arg_pointer,
FIRST_PARM_OFFSET (current_function_decl));
}
return virtual_incoming_args_rtx;
}
/* We may have to tell the dataflow pass that the split stack prologue
is initializing a register. */
static void
rs6000_live_on_entry (bitmap regs)
{
if (flag_split_stack)
bitmap_set_bit (regs, 12);
}
/* Emit -fsplit-stack dynamic stack allocation space check. */
void
rs6000_split_stack_space_check (rtx size, rtx label)
{
rtx sp = gen_rtx_REG (Pmode, STACK_POINTER_REGNUM);
rtx limit = gen_reg_rtx (Pmode);
rtx requested = gen_reg_rtx (Pmode);
rtx cmp = gen_reg_rtx (CCUNSmode);
rtx jump;
emit_insn (gen_load_split_stack_limit (limit));
if (CONST_INT_P (size))
emit_insn (gen_add3_insn (requested, sp, GEN_INT (-INTVAL (size))));
else
{
size = force_reg (Pmode, size);
emit_move_insn (requested, gen_rtx_MINUS (Pmode, sp, size));
}
emit_insn (gen_rtx_SET (cmp, gen_rtx_COMPARE (CCUNSmode, requested, limit)));
jump = gen_rtx_IF_THEN_ELSE (VOIDmode,
gen_rtx_GEU (VOIDmode, cmp, const0_rtx),
gen_rtx_LABEL_REF (VOIDmode, label),
pc_rtx);
jump = emit_jump_insn (gen_rtx_SET (pc_rtx, jump));
JUMP_LABEL (jump) = label;
}
/* A C compound statement that outputs the assembler code for a thunk
function, used to implement C++ virtual function calls with
multiple inheritance. The thunk acts as a wrapper around a virtual
function, adjusting the implicit object parameter before handing
control off to the real function.
First, emit code to add the integer DELTA to the location that
contains the incoming first argument. Assume that this argument
contains a pointer, and is the one used to pass the `this' pointer
in C++. This is the incoming argument *before* the function
prologue, e.g. `%o0' on a sparc. The addition must preserve the
values of all other incoming arguments.
After the addition, emit code to jump to FUNCTION, which is a
`FUNCTION_DECL'. This is a direct pure jump, not a call, and does
not touch the return address. Hence returning from FUNCTION will
return to whoever called the current `thunk'.
The effect must be as if FUNCTION had been called directly with the
adjusted first argument. This macro is responsible for emitting
all of the code for a thunk function; output_function_prologue()
and output_function_epilogue() are not invoked.
The THUNK_FNDECL is redundant. (DELTA and FUNCTION have already
been extracted from it.) It might possibly be useful on some
targets, but probably not.
If you do not define this macro, the target-independent code in the
C++ frontend will generate a less efficient heavyweight thunk that
calls FUNCTION instead of jumping to it. The generic approach does
not support varargs. */
static void
rs6000_output_mi_thunk (FILE *file, tree thunk_fndecl ATTRIBUTE_UNUSED,
HOST_WIDE_INT delta, HOST_WIDE_INT vcall_offset,
tree function)
{
rtx this_rtx, funexp;
rtx_insn *insn;
reload_completed = 1;
epilogue_completed = 1;
/* Mark the end of the (empty) prologue. */
emit_note (NOTE_INSN_PROLOGUE_END);
/* Find the "this" pointer. If the function returns a structure,
the structure return pointer is in r3. */
if (aggregate_value_p (TREE_TYPE (TREE_TYPE (function)), function))
this_rtx = gen_rtx_REG (Pmode, 4);
else
this_rtx = gen_rtx_REG (Pmode, 3);
/* Apply the constant offset, if required. */
if (delta)
emit_insn (gen_add3_insn (this_rtx, this_rtx, GEN_INT (delta)));
/* Apply the offset from the vtable, if required. */
if (vcall_offset)
{
rtx vcall_offset_rtx = GEN_INT (vcall_offset);
rtx tmp = gen_rtx_REG (Pmode, 12);
emit_move_insn (tmp, gen_rtx_MEM (Pmode, this_rtx));
if (((unsigned HOST_WIDE_INT) vcall_offset) + 0x8000 >= 0x10000)
{
emit_insn (gen_add3_insn (tmp, tmp, vcall_offset_rtx));
emit_move_insn (tmp, gen_rtx_MEM (Pmode, tmp));
}
else
{
rtx loc = gen_rtx_PLUS (Pmode, tmp, vcall_offset_rtx);
emit_move_insn (tmp, gen_rtx_MEM (Pmode, loc));
}
emit_insn (gen_add3_insn (this_rtx, this_rtx, tmp));
}
/* Generate a tail call to the target function. */
if (!TREE_USED (function))
{
assemble_external (function);
TREE_USED (function) = 1;
}
funexp = XEXP (DECL_RTL (function), 0);
funexp = gen_rtx_MEM (FUNCTION_MODE, funexp);
#if TARGET_MACHO
if (MACHOPIC_INDIRECT)
funexp = machopic_indirect_call_target (funexp);
#endif
/* gen_sibcall expects reload to convert scratch pseudo to LR so we must
generate sibcall RTL explicitly. */
insn = emit_call_insn (
gen_rtx_PARALLEL (VOIDmode,
gen_rtvec (3,
gen_rtx_CALL (VOIDmode,
funexp, const0_rtx),
gen_rtx_USE (VOIDmode, const0_rtx),
simple_return_rtx)));
SIBLING_CALL_P (insn) = 1;
emit_barrier ();
/* Run just enough of rest_of_compilation to get the insns emitted.
There's not really enough bulk here to make other passes such as
instruction scheduling worth while. Note that use_thunk calls
assemble_start_function and assemble_end_function. */
insn = get_insns ();
shorten_branches (insn);
final_start_function (insn, file, 1);
final (insn, file, 1);
final_end_function ();
reload_completed = 0;
epilogue_completed = 0;
}
/* A quick summary of the various types of 'constant-pool tables'
under PowerPC:
Target Flags Name One table per
AIX (none) AIX TOC object file
AIX -mfull-toc AIX TOC object file
AIX -mminimal-toc AIX minimal TOC translation unit
SVR4/EABI (none) SVR4 SDATA object file
SVR4/EABI -fpic SVR4 pic object file
SVR4/EABI -fPIC SVR4 PIC translation unit
SVR4/EABI -mrelocatable EABI TOC function
SVR4/EABI -maix AIX TOC object file
SVR4/EABI -maix -mminimal-toc
AIX minimal TOC translation unit
Name Reg. Set by entries contains:
made by addrs? fp? sum?
AIX TOC 2 crt0 as Y option option
AIX minimal TOC 30 prolog gcc Y Y option
SVR4 SDATA 13 crt0 gcc N Y N
SVR4 pic 30 prolog ld Y not yet N
SVR4 PIC 30 prolog gcc Y option option
EABI TOC 30 prolog gcc Y option option
*/
/* Hash functions for the hash table. */
static unsigned
rs6000_hash_constant (rtx k)
{
enum rtx_code code = GET_CODE (k);
machine_mode mode = GET_MODE (k);
unsigned result = (code << 3) ^ mode;
const char *format;
int flen, fidx;
format = GET_RTX_FORMAT (code);
flen = strlen (format);
fidx = 0;
switch (code)
{
case LABEL_REF:
return result * 1231 + (unsigned) INSN_UID (XEXP (k, 0));
case CONST_WIDE_INT:
{
int i;
flen = CONST_WIDE_INT_NUNITS (k);
for (i = 0; i < flen; i++)
result = result * 613 + CONST_WIDE_INT_ELT (k, i);
return result;
}
case CONST_DOUBLE:
if (mode != VOIDmode)
return real_hash (CONST_DOUBLE_REAL_VALUE (k)) * result;
flen = 2;
break;
case CODE_LABEL:
fidx = 3;
break;
default:
break;
}
for (; fidx < flen; fidx++)
switch (format[fidx])
{
case 's':
{
unsigned i, len;
const char *str = XSTR (k, fidx);
len = strlen (str);
result = result * 613 + len;
for (i = 0; i < len; i++)
result = result * 613 + (unsigned) str[i];
break;
}
case 'u':
case 'e':
result = result * 1231 + rs6000_hash_constant (XEXP (k, fidx));
break;
case 'i':
case 'n':
result = result * 613 + (unsigned) XINT (k, fidx);
break;
case 'w':
if (sizeof (unsigned) >= sizeof (HOST_WIDE_INT))
result = result * 613 + (unsigned) XWINT (k, fidx);
else
{
size_t i;
for (i = 0; i < sizeof (HOST_WIDE_INT) / sizeof (unsigned); i++)
result = result * 613 + (unsigned) (XWINT (k, fidx)
>> CHAR_BIT * i);
}
break;
case '0':
break;
default:
gcc_unreachable ();
}
return result;
}
hashval_t
toc_hasher::hash (toc_hash_struct *thc)
{
return rs6000_hash_constant (thc->key) ^ thc->key_mode;
}
/* Compare H1 and H2 for equivalence. */
bool
toc_hasher::equal (toc_hash_struct *h1, toc_hash_struct *h2)
{
rtx r1 = h1->key;
rtx r2 = h2->key;
if (h1->key_mode != h2->key_mode)
return 0;
return rtx_equal_p (r1, r2);
}
/* These are the names given by the C++ front-end to vtables, and
vtable-like objects. Ideally, this logic should not be here;
instead, there should be some programmatic way of inquiring as
to whether or not an object is a vtable. */
#define VTABLE_NAME_P(NAME) \
(strncmp ("_vt.", name, strlen ("_vt.")) == 0 \
|| strncmp ("_ZTV", name, strlen ("_ZTV")) == 0 \
|| strncmp ("_ZTT", name, strlen ("_ZTT")) == 0 \
|| strncmp ("_ZTI", name, strlen ("_ZTI")) == 0 \
|| strncmp ("_ZTC", name, strlen ("_ZTC")) == 0)
#ifdef NO_DOLLAR_IN_LABEL
/* Return a GGC-allocated character string translating dollar signs in
input NAME to underscores. Used by XCOFF ASM_OUTPUT_LABELREF. */
const char *
rs6000_xcoff_strip_dollar (const char *name)
{
char *strip, *p;
const char *q;
size_t len;
q = (const char *) strchr (name, '$');
if (q == 0 || q == name)
return name;
len = strlen (name);
strip = XALLOCAVEC (char, len + 1);
strcpy (strip, name);
p = strip + (q - name);
while (p)
{
*p = '_';
p = strchr (p + 1, '$');
}
return ggc_alloc_string (strip, len);
}
#endif
void
rs6000_output_symbol_ref (FILE *file, rtx x)
{
const char *name = XSTR (x, 0);
/* Currently C++ toc references to vtables can be emitted before it
is decided whether the vtable is public or private. If this is
the case, then the linker will eventually complain that there is
a reference to an unknown section. Thus, for vtables only,
we emit the TOC reference to reference the identifier and not the
symbol. */
if (VTABLE_NAME_P (name))
{
RS6000_OUTPUT_BASENAME (file, name);
}
else
assemble_name (file, name);
}
/* Output a TOC entry. We derive the entry name from what is being
written. */
void
output_toc (FILE *file, rtx x, int labelno, machine_mode mode)
{
char buf[256];
const char *name = buf;
rtx base = x;
HOST_WIDE_INT offset = 0;
gcc_assert (!TARGET_NO_TOC);
/* When the linker won't eliminate them, don't output duplicate
TOC entries (this happens on AIX if there is any kind of TOC,
and on SVR4 under -fPIC or -mrelocatable). Don't do this for
CODE_LABELs. */
if (TARGET_TOC && GET_CODE (x) != LABEL_REF)
{
struct toc_hash_struct *h;
/* Create toc_hash_table. This can't be done at TARGET_OPTION_OVERRIDE
time because GGC is not initialized at that point. */
if (toc_hash_table == NULL)
toc_hash_table = hash_table<toc_hasher>::create_ggc (1021);
h = ggc_alloc<toc_hash_struct> ();
h->key = x;
h->key_mode = mode;
h->labelno = labelno;
toc_hash_struct **found = toc_hash_table->find_slot (h, INSERT);
if (*found == NULL)
*found = h;
else /* This is indeed a duplicate.
Set this label equal to that label. */
{
fputs ("\t.set ", file);
ASM_OUTPUT_INTERNAL_LABEL_PREFIX (file, "LC");
fprintf (file, "%d,", labelno);
ASM_OUTPUT_INTERNAL_LABEL_PREFIX (file, "LC");
fprintf (file, "%d\n", ((*found)->labelno));
#ifdef HAVE_AS_TLS
if (TARGET_XCOFF && GET_CODE (x) == SYMBOL_REF
&& (SYMBOL_REF_TLS_MODEL (x) == TLS_MODEL_GLOBAL_DYNAMIC
|| SYMBOL_REF_TLS_MODEL (x) == TLS_MODEL_LOCAL_DYNAMIC))
{
fputs ("\t.set ", file);
ASM_OUTPUT_INTERNAL_LABEL_PREFIX (file, "LCM");
fprintf (file, "%d,", labelno);
ASM_OUTPUT_INTERNAL_LABEL_PREFIX (file, "LCM");
fprintf (file, "%d\n", ((*found)->labelno));
}
#endif
return;
}
}
/* If we're going to put a double constant in the TOC, make sure it's
aligned properly when strict alignment is on. */
if ((CONST_DOUBLE_P (x) || CONST_WIDE_INT_P (x))
&& STRICT_ALIGNMENT
&& GET_MODE_BITSIZE (mode) >= 64
&& ! (TARGET_NO_FP_IN_TOC && ! TARGET_MINIMAL_TOC)) {
ASM_OUTPUT_ALIGN (file, 3);
}
(*targetm.asm_out.internal_label) (file, "LC", labelno);
/* Handle FP constants specially. Note that if we have a minimal
TOC, things we put here aren't actually in the TOC, so we can allow
FP constants. */
if (GET_CODE (x) == CONST_DOUBLE &&
(GET_MODE (x) == TFmode || GET_MODE (x) == TDmode
|| GET_MODE (x) == IFmode || GET_MODE (x) == KFmode))
{
long k[4];
if (DECIMAL_FLOAT_MODE_P (GET_MODE (x)))
REAL_VALUE_TO_TARGET_DECIMAL128 (*CONST_DOUBLE_REAL_VALUE (x), k);
else
REAL_VALUE_TO_TARGET_LONG_DOUBLE (*CONST_DOUBLE_REAL_VALUE (x), k);
if (TARGET_64BIT)
{
if (TARGET_ELF || TARGET_MINIMAL_TOC)
fputs (DOUBLE_INT_ASM_OP, file);
else
fprintf (file, "\t.tc FT_%lx_%lx_%lx_%lx[TC],",
k[0] & 0xffffffff, k[1] & 0xffffffff,
k[2] & 0xffffffff, k[3] & 0xffffffff);
fprintf (file, "0x%lx%08lx,0x%lx%08lx\n",
k[WORDS_BIG_ENDIAN ? 0 : 1] & 0xffffffff,
k[WORDS_BIG_ENDIAN ? 1 : 0] & 0xffffffff,
k[WORDS_BIG_ENDIAN ? 2 : 3] & 0xffffffff,
k[WORDS_BIG_ENDIAN ? 3 : 2] & 0xffffffff);
return;
}
else
{
if (TARGET_ELF || TARGET_MINIMAL_TOC)
fputs ("\t.long ", file);
else
fprintf (file, "\t.tc FT_%lx_%lx_%lx_%lx[TC],",
k[0] & 0xffffffff, k[1] & 0xffffffff,
k[2] & 0xffffffff, k[3] & 0xffffffff);
fprintf (file, "0x%lx,0x%lx,0x%lx,0x%lx\n",
k[0] & 0xffffffff, k[1] & 0xffffffff,
k[2] & 0xffffffff, k[3] & 0xffffffff);
return;
}
}
else if (GET_CODE (x) == CONST_DOUBLE &&
(GET_MODE (x) == DFmode || GET_MODE (x) == DDmode))
{
long k[2];
if (DECIMAL_FLOAT_MODE_P (GET_MODE (x)))
REAL_VALUE_TO_TARGET_DECIMAL64 (*CONST_DOUBLE_REAL_VALUE (x), k);
else
REAL_VALUE_TO_TARGET_DOUBLE (*CONST_DOUBLE_REAL_VALUE (x), k);
if (TARGET_64BIT)
{
if (TARGET_ELF || TARGET_MINIMAL_TOC)
fputs (DOUBLE_INT_ASM_OP, file);
else
fprintf (file, "\t.tc FD_%lx_%lx[TC],",
k[0] & 0xffffffff, k[1] & 0xffffffff);
fprintf (file, "0x%lx%08lx\n",
k[WORDS_BIG_ENDIAN ? 0 : 1] & 0xffffffff,
k[WORDS_BIG_ENDIAN ? 1 : 0] & 0xffffffff);
return;
}
else
{
if (TARGET_ELF || TARGET_MINIMAL_TOC)
fputs ("\t.long ", file);
else
fprintf (file, "\t.tc FD_%lx_%lx[TC],",
k[0] & 0xffffffff, k[1] & 0xffffffff);
fprintf (file, "0x%lx,0x%lx\n",
k[0] & 0xffffffff, k[1] & 0xffffffff);
return;
}
}
else if (GET_CODE (x) == CONST_DOUBLE &&
(GET_MODE (x) == SFmode || GET_MODE (x) == SDmode))
{
long l;
if (DECIMAL_FLOAT_MODE_P (GET_MODE (x)))
REAL_VALUE_TO_TARGET_DECIMAL32 (*CONST_DOUBLE_REAL_VALUE (x), l);
else
REAL_VALUE_TO_TARGET_SINGLE (*CONST_DOUBLE_REAL_VALUE (x), l);
if (TARGET_64BIT)
{
if (TARGET_ELF || TARGET_MINIMAL_TOC)
fputs (DOUBLE_INT_ASM_OP, file);
else
fprintf (file, "\t.tc FS_%lx[TC],", l & 0xffffffff);
if (WORDS_BIG_ENDIAN)
fprintf (file, "0x%lx00000000\n", l & 0xffffffff);
else
fprintf (file, "0x%lx\n", l & 0xffffffff);
return;
}
else
{
if (TARGET_ELF || TARGET_MINIMAL_TOC)
fputs ("\t.long ", file);
else
fprintf (file, "\t.tc FS_%lx[TC],", l & 0xffffffff);
fprintf (file, "0x%lx\n", l & 0xffffffff);
return;
}
}
else if (GET_MODE (x) == VOIDmode && GET_CODE (x) == CONST_INT)
{
unsigned HOST_WIDE_INT low;
HOST_WIDE_INT high;
low = INTVAL (x) & 0xffffffff;
high = (HOST_WIDE_INT) INTVAL (x) >> 32;
/* TOC entries are always Pmode-sized, so when big-endian
smaller integer constants in the TOC need to be padded.
(This is still a win over putting the constants in
a separate constant pool, because then we'd have
to have both a TOC entry _and_ the actual constant.)
For a 32-bit target, CONST_INT values are loaded and shifted
entirely within `low' and can be stored in one TOC entry. */
/* It would be easy to make this work, but it doesn't now. */
gcc_assert (!TARGET_64BIT || POINTER_SIZE >= GET_MODE_BITSIZE (mode));
if (WORDS_BIG_ENDIAN && POINTER_SIZE > GET_MODE_BITSIZE (mode))
{
low |= high << 32;
low <<= POINTER_SIZE - GET_MODE_BITSIZE (mode);
high = (HOST_WIDE_INT) low >> 32;
low &= 0xffffffff;
}
if (TARGET_64BIT)
{
if (TARGET_ELF || TARGET_MINIMAL_TOC)
fputs (DOUBLE_INT_ASM_OP, file);
else
fprintf (file, "\t.tc ID_%lx_%lx[TC],",
(long) high & 0xffffffff, (long) low & 0xffffffff);
fprintf (file, "0x%lx%08lx\n",
(long) high & 0xffffffff, (long) low & 0xffffffff);
return;
}
else
{
if (POINTER_SIZE < GET_MODE_BITSIZE (mode))
{
if (TARGET_ELF || TARGET_MINIMAL_TOC)
fputs ("\t.long ", file);
else
fprintf (file, "\t.tc ID_%lx_%lx[TC],",
(long) high & 0xffffffff, (long) low & 0xffffffff);
fprintf (file, "0x%lx,0x%lx\n",
(long) high & 0xffffffff, (long) low & 0xffffffff);
}
else
{
if (TARGET_ELF || TARGET_MINIMAL_TOC)
fputs ("\t.long ", file);
else
fprintf (file, "\t.tc IS_%lx[TC],", (long) low & 0xffffffff);
fprintf (file, "0x%lx\n", (long) low & 0xffffffff);
}
return;
}
}
if (GET_CODE (x) == CONST)
{
gcc_assert (GET_CODE (XEXP (x, 0)) == PLUS
&& GET_CODE (XEXP (XEXP (x, 0), 1)) == CONST_INT);
base = XEXP (XEXP (x, 0), 0);
offset = INTVAL (XEXP (XEXP (x, 0), 1));
}
switch (GET_CODE (base))
{
case SYMBOL_REF:
name = XSTR (base, 0);
break;
case LABEL_REF:
ASM_GENERATE_INTERNAL_LABEL (buf, "L",
CODE_LABEL_NUMBER (XEXP (base, 0)));
break;
case CODE_LABEL:
ASM_GENERATE_INTERNAL_LABEL (buf, "L", CODE_LABEL_NUMBER (base));
break;
default:
gcc_unreachable ();
}
if (TARGET_ELF || TARGET_MINIMAL_TOC)
fputs (TARGET_32BIT ? "\t.long " : DOUBLE_INT_ASM_OP, file);
else
{
fputs ("\t.tc ", file);
RS6000_OUTPUT_BASENAME (file, name);
if (offset < 0)
fprintf (file, ".N" HOST_WIDE_INT_PRINT_UNSIGNED, - offset);
else if (offset)
fprintf (file, ".P" HOST_WIDE_INT_PRINT_UNSIGNED, offset);
/* Mark large TOC symbols on AIX with [TE] so they are mapped
after other TOC symbols, reducing overflow of small TOC access
to [TC] symbols. */
fputs (TARGET_XCOFF && TARGET_CMODEL != CMODEL_SMALL
? "[TE]," : "[TC],", file);
}
/* Currently C++ toc references to vtables can be emitted before it
is decided whether the vtable is public or private. If this is
the case, then the linker will eventually complain that there is
a TOC reference to an unknown section. Thus, for vtables only,
we emit the TOC reference to reference the symbol and not the
section. */
if (VTABLE_NAME_P (name))
{
RS6000_OUTPUT_BASENAME (file, name);
if (offset < 0)
fprintf (file, HOST_WIDE_INT_PRINT_DEC, offset);
else if (offset > 0)
fprintf (file, "+" HOST_WIDE_INT_PRINT_DEC, offset);
}
else
output_addr_const (file, x);
#if HAVE_AS_TLS
if (TARGET_XCOFF && GET_CODE (base) == SYMBOL_REF)
{
switch (SYMBOL_REF_TLS_MODEL (base))
{
case 0:
break;
case TLS_MODEL_LOCAL_EXEC:
fputs ("@le", file);
break;
case TLS_MODEL_INITIAL_EXEC:
fputs ("@ie", file);
break;
/* Use global-dynamic for local-dynamic. */
case TLS_MODEL_GLOBAL_DYNAMIC:
case TLS_MODEL_LOCAL_DYNAMIC:
putc ('\n', file);
(*targetm.asm_out.internal_label) (file, "LCM", labelno);
fputs ("\t.tc .", file);
RS6000_OUTPUT_BASENAME (file, name);
fputs ("[TC],", file);
output_addr_const (file, x);
fputs ("@m", file);
break;
default:
gcc_unreachable ();
}
}
#endif
putc ('\n', file);
}
/* Output an assembler pseudo-op to write an ASCII string of N characters
starting at P to FILE.
On the RS/6000, we have to do this using the .byte operation and
write out special characters outside the quoted string.
Also, the assembler is broken; very long strings are truncated,
so we must artificially break them up early. */
void
output_ascii (FILE *file, const char *p, int n)
{
char c;
int i, count_string;
const char *for_string = "\t.byte \"";
const char *for_decimal = "\t.byte ";
const char *to_close = NULL;
count_string = 0;
for (i = 0; i < n; i++)
{
c = *p++;
if (c >= ' ' && c < 0177)
{
if (for_string)
fputs (for_string, file);
putc (c, file);
/* Write two quotes to get one. */
if (c == '"')
{
putc (c, file);
++count_string;
}
for_string = NULL;
for_decimal = "\"\n\t.byte ";
to_close = "\"\n";
++count_string;
if (count_string >= 512)
{
fputs (to_close, file);
for_string = "\t.byte \"";
for_decimal = "\t.byte ";
to_close = NULL;
count_string = 0;
}
}
else
{
if (for_decimal)
fputs (for_decimal, file);
fprintf (file, "%d", c);
for_string = "\n\t.byte \"";
for_decimal = ", ";
to_close = "\n";
count_string = 0;
}
}
/* Now close the string if we have written one. Then end the line. */
if (to_close)
fputs (to_close, file);
}
/* Generate a unique section name for FILENAME for a section type
represented by SECTION_DESC. Output goes into BUF.
SECTION_DESC can be any string, as long as it is different for each
possible section type.
We name the section in the same manner as xlc. The name begins with an
underscore followed by the filename (after stripping any leading directory
names) with the last period replaced by the string SECTION_DESC. If
FILENAME does not contain a period, SECTION_DESC is appended to the end of
the name. */
void
rs6000_gen_section_name (char **buf, const char *filename,
const char *section_desc)
{
const char *q, *after_last_slash, *last_period = 0;
char *p;
int len;
after_last_slash = filename;
for (q = filename; *q; q++)
{
if (*q == '/')
after_last_slash = q + 1;
else if (*q == '.')
last_period = q;
}
len = strlen (after_last_slash) + strlen (section_desc) + 2;
*buf = (char *) xmalloc (len);
p = *buf;
*p++ = '_';
for (q = after_last_slash; *q; q++)
{
if (q == last_period)
{
strcpy (p, section_desc);
p += strlen (section_desc);
break;
}
else if (ISALNUM (*q))
*p++ = *q;
}
if (last_period == 0)
strcpy (p, section_desc);
else
*p = '\0';
}
/* Emit profile function. */
void
output_profile_hook (int labelno ATTRIBUTE_UNUSED)
{
/* Non-standard profiling for kernels, which just saves LR then calls
_mcount without worrying about arg saves. The idea is to change
the function prologue as little as possible as it isn't easy to
account for arg save/restore code added just for _mcount. */
if (TARGET_PROFILE_KERNEL)
return;
if (DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
{
#ifndef NO_PROFILE_COUNTERS
# define NO_PROFILE_COUNTERS 0
#endif
if (NO_PROFILE_COUNTERS)
emit_library_call (init_one_libfunc (RS6000_MCOUNT),
LCT_NORMAL, VOIDmode);
else
{
char buf[30];
const char *label_name;
rtx fun;
ASM_GENERATE_INTERNAL_LABEL (buf, "LP", labelno);
label_name = ggc_strdup ((*targetm.strip_name_encoding) (buf));
fun = gen_rtx_SYMBOL_REF (Pmode, label_name);
emit_library_call (init_one_libfunc (RS6000_MCOUNT),
LCT_NORMAL, VOIDmode, fun, Pmode);
}
}
else if (DEFAULT_ABI == ABI_DARWIN)
{
const char *mcount_name = RS6000_MCOUNT;
int caller_addr_regno = LR_REGNO;
/* Be conservative and always set this, at least for now. */
crtl->uses_pic_offset_table = 1;
#if TARGET_MACHO
/* For PIC code, set up a stub and collect the caller's address
from r0, which is where the prologue puts it. */
if (MACHOPIC_INDIRECT
&& crtl->uses_pic_offset_table)
caller_addr_regno = 0;
#endif
emit_library_call (gen_rtx_SYMBOL_REF (Pmode, mcount_name),
LCT_NORMAL, VOIDmode,
gen_rtx_REG (Pmode, caller_addr_regno), Pmode);
}
}
/* Write function profiler code. */
void
output_function_profiler (FILE *file, int labelno)
{
char buf[100];
switch (DEFAULT_ABI)
{
default:
gcc_unreachable ();
case ABI_V4:
if (!TARGET_32BIT)
{
warning (0, "no profiling of 64-bit code for this ABI");
return;
}
ASM_GENERATE_INTERNAL_LABEL (buf, "LP", labelno);
fprintf (file, "\tmflr %s\n", reg_names[0]);
if (NO_PROFILE_COUNTERS)
{
asm_fprintf (file, "\tstw %s,4(%s)\n",
reg_names[0], reg_names[1]);
}
else if (TARGET_SECURE_PLT && flag_pic)
{
if (TARGET_LINK_STACK)
{
char name[32];
get_ppc476_thunk_name (name);
asm_fprintf (file, "\tbl %s\n", name);
}
else
asm_fprintf (file, "\tbcl 20,31,1f\n1:\n");
asm_fprintf (file, "\tstw %s,4(%s)\n",
reg_names[0], reg_names[1]);
asm_fprintf (file, "\tmflr %s\n", reg_names[12]);
asm_fprintf (file, "\taddis %s,%s,",
reg_names[12], reg_names[12]);
assemble_name (file, buf);
asm_fprintf (file, "-1b@ha\n\tla %s,", reg_names[0]);
assemble_name (file, buf);
asm_fprintf (file, "-1b@l(%s)\n", reg_names[12]);
}
else if (flag_pic == 1)
{
fputs ("\tbl _GLOBAL_OFFSET_TABLE_@local-4\n", file);
asm_fprintf (file, "\tstw %s,4(%s)\n",
reg_names[0], reg_names[1]);
asm_fprintf (file, "\tmflr %s\n", reg_names[12]);
asm_fprintf (file, "\tlwz %s,", reg_names[0]);
assemble_name (file, buf);
asm_fprintf (file, "@got(%s)\n", reg_names[12]);
}
else if (flag_pic > 1)
{
asm_fprintf (file, "\tstw %s,4(%s)\n",
reg_names[0], reg_names[1]);
/* Now, we need to get the address of the label. */
if (TARGET_LINK_STACK)
{
char name[32];
get_ppc476_thunk_name (name);
asm_fprintf (file, "\tbl %s\n\tb 1f\n\t.long ", name);
assemble_name (file, buf);
fputs ("-.\n1:", file);
asm_fprintf (file, "\tmflr %s\n", reg_names[11]);
asm_fprintf (file, "\taddi %s,%s,4\n",
reg_names[11], reg_names[11]);
}
else
{
fputs ("\tbcl 20,31,1f\n\t.long ", file);
assemble_name (file, buf);
fputs ("-.\n1:", file);
asm_fprintf (file, "\tmflr %s\n", reg_names[11]);
}
asm_fprintf (file, "\tlwz %s,0(%s)\n",
reg_names[0], reg_names[11]);
asm_fprintf (file, "\tadd %s,%s,%s\n",
reg_names[0], reg_names[0], reg_names[11]);
}
else
{
asm_fprintf (file, "\tlis %s,", reg_names[12]);
assemble_name (file, buf);
fputs ("@ha\n", file);
asm_fprintf (file, "\tstw %s,4(%s)\n",
reg_names[0], reg_names[1]);
asm_fprintf (file, "\tla %s,", reg_names[0]);
assemble_name (file, buf);
asm_fprintf (file, "@l(%s)\n", reg_names[12]);
}
/* ABI_V4 saves the static chain reg with ASM_OUTPUT_REG_PUSH. */
fprintf (file, "\tbl %s%s\n",
RS6000_MCOUNT, flag_pic ? "@plt" : "");
break;
case ABI_AIX:
case ABI_ELFv2:
case ABI_DARWIN:
/* Don't do anything, done in output_profile_hook (). */
break;
}
}
/* The following variable value is the last issued insn. */
static rtx_insn *last_scheduled_insn;
/* The following variable helps to balance issuing of load and
store instructions */
static int load_store_pendulum;
/* The following variable helps pair divide insns during scheduling. */
static int divide_cnt;
/* The following variable helps pair and alternate vector and vector load
insns during scheduling. */
static int vec_pairing;
/* Power4 load update and store update instructions are cracked into a
load or store and an integer insn which are executed in the same cycle.
Branches have their own dispatch slot which does not count against the
GCC issue rate, but it changes the program flow so there are no other
instructions to issue in this cycle. */
static int
rs6000_variable_issue_1 (rtx_insn *insn, int more)
{
last_scheduled_insn = insn;
if (GET_CODE (PATTERN (insn)) == USE
|| GET_CODE (PATTERN (insn)) == CLOBBER)
{
cached_can_issue_more = more;
return cached_can_issue_more;
}
if (insn_terminates_group_p (insn, current_group))
{
cached_can_issue_more = 0;
return cached_can_issue_more;
}
/* If no reservation, but reach here */
if (recog_memoized (insn) < 0)
return more;
if (rs6000_sched_groups)
{
if (is_microcoded_insn (insn))
cached_can_issue_more = 0;
else if (is_cracked_insn (insn))
cached_can_issue_more = more > 2 ? more - 2 : 0;
else
cached_can_issue_more = more - 1;
return cached_can_issue_more;
}
if (rs6000_tune == PROCESSOR_CELL && is_nonpipeline_insn (insn))
return 0;
cached_can_issue_more = more - 1;
return cached_can_issue_more;
}
static int
rs6000_variable_issue (FILE *stream, int verbose, rtx_insn *insn, int more)
{
int r = rs6000_variable_issue_1 (insn, more);
if (verbose)
fprintf (stream, "// rs6000_variable_issue (more = %d) = %d\n", more, r);
return r;
}
/* Adjust the cost of a scheduling dependency. Return the new cost of
a dependency LINK or INSN on DEP_INSN. COST is the current cost. */
static int
rs6000_adjust_cost (rtx_insn *insn, int dep_type, rtx_insn *dep_insn, int cost,
unsigned int)
{
enum attr_type attr_type;
if (recog_memoized (insn) < 0 || recog_memoized (dep_insn) < 0)
return cost;
switch (dep_type)
{
case REG_DEP_TRUE:
{
/* Data dependency; DEP_INSN writes a register that INSN reads
some cycles later. */
/* Separate a load from a narrower, dependent store. */
if ((rs6000_sched_groups || rs6000_tune == PROCESSOR_POWER9)
&& GET_CODE (PATTERN (insn)) == SET
&& GET_CODE (PATTERN (dep_insn)) == SET
&& GET_CODE (XEXP (PATTERN (insn), 1)) == MEM
&& GET_CODE (XEXP (PATTERN (dep_insn), 0)) == MEM
&& (GET_MODE_SIZE (GET_MODE (XEXP (PATTERN (insn), 1)))
> GET_MODE_SIZE (GET_MODE (XEXP (PATTERN (dep_insn), 0)))))
return cost + 14;
attr_type = get_attr_type (insn);
switch (attr_type)
{
case TYPE_JMPREG:
/* Tell the first scheduling pass about the latency between
a mtctr and bctr (and mtlr and br/blr). The first
scheduling pass will not know about this latency since
the mtctr instruction, which has the latency associated
to it, will be generated by reload. */
return 4;
case TYPE_BRANCH:
/* Leave some extra cycles between a compare and its
dependent branch, to inhibit expensive mispredicts. */
if ((rs6000_tune == PROCESSOR_PPC603
|| rs6000_tune == PROCESSOR_PPC604
|| rs6000_tune == PROCESSOR_PPC604e
|| rs6000_tune == PROCESSOR_PPC620
|| rs6000_tune == PROCESSOR_PPC630
|| rs6000_tune == PROCESSOR_PPC750
|| rs6000_tune == PROCESSOR_PPC7400
|| rs6000_tune == PROCESSOR_PPC7450
|| rs6000_tune == PROCESSOR_PPCE5500
|| rs6000_tune == PROCESSOR_PPCE6500
|| rs6000_tune == PROCESSOR_POWER4
|| rs6000_tune == PROCESSOR_POWER5
|| rs6000_tune == PROCESSOR_POWER7
|| rs6000_tune == PROCESSOR_POWER8
|| rs6000_tune == PROCESSOR_POWER9
|| rs6000_tune == PROCESSOR_CELL)
&& recog_memoized (dep_insn)
&& (INSN_CODE (dep_insn) >= 0))
switch (get_attr_type (dep_insn))
{
case TYPE_CMP:
case TYPE_FPCOMPARE:
case TYPE_CR_LOGICAL:
case TYPE_DELAYED_CR:
return cost + 2;
case TYPE_EXTS:
case TYPE_MUL:
if (get_attr_dot (dep_insn) == DOT_YES)
return cost + 2;
else
break;
case TYPE_SHIFT:
if (get_attr_dot (dep_insn) == DOT_YES
&& get_attr_var_shift (dep_insn) == VAR_SHIFT_NO)
return cost + 2;
else
break;
default:
break;
}
break;
case TYPE_STORE:
case TYPE_FPSTORE:
if ((rs6000_tune == PROCESSOR_POWER6)
&& recog_memoized (dep_insn)
&& (INSN_CODE (dep_insn) >= 0))
{
if (GET_CODE (PATTERN (insn)) != SET)
/* If this happens, we have to extend this to schedule
optimally. Return default for now. */
return cost;
/* Adjust the cost for the case where the value written
by a fixed point operation is used as the address
gen value on a store. */
switch (get_attr_type (dep_insn))
{
case TYPE_LOAD:
case TYPE_CNTLZ:
{
if (! rs6000_store_data_bypass_p (dep_insn, insn))
return get_attr_sign_extend (dep_insn)
== SIGN_EXTEND_YES ? 6 : 4;
break;
}
case TYPE_SHIFT:
{
if (! rs6000_store_data_bypass_p (dep_insn, insn))
return get_attr_var_shift (dep_insn) == VAR_SHIFT_YES ?
6 : 3;
break;
}
case TYPE_INTEGER:
case TYPE_ADD:
case TYPE_LOGICAL:
case TYPE_EXTS:
case TYPE_INSERT:
{
if (! rs6000_store_data_bypass_p (dep_insn, insn))
return 3;
break;
}
case TYPE_STORE:
case TYPE_FPLOAD:
case TYPE_FPSTORE:
{
if (get_attr_update (dep_insn) == UPDATE_YES
&& ! rs6000_store_data_bypass_p (dep_insn, insn))
return 3;
break;
}
case TYPE_MUL:
{
if (! rs6000_store_data_bypass_p (dep_insn, insn))
return 17;
break;
}
case TYPE_DIV:
{
if (! rs6000_store_data_bypass_p (dep_insn, insn))
return get_attr_size (dep_insn) == SIZE_32 ? 45 : 57;
break;
}
default:
break;
}
}
break;
case TYPE_LOAD:
if ((rs6000_tune == PROCESSOR_POWER6)
&& recog_memoized (dep_insn)
&& (INSN_CODE (dep_insn) >= 0))
{
/* Adjust the cost for the case where the value written
by a fixed point instruction is used within the address
gen portion of a subsequent load(u)(x) */
switch (get_attr_type (dep_insn))
{
case TYPE_LOAD:
case TYPE_CNTLZ:
{
if (set_to_load_agen (dep_insn, insn))
return get_attr_sign_extend (dep_insn)
== SIGN_EXTEND_YES ? 6 : 4;
break;
}
case TYPE_SHIFT:
{
if (set_to_load_agen (dep_insn, insn))
return get_attr_var_shift (dep_insn) == VAR_SHIFT_YES ?
6 : 3;
break;
}
case TYPE_INTEGER:
case TYPE_ADD:
case TYPE_LOGICAL:
case TYPE_EXTS:
case TYPE_INSERT:
{
if (set_to_load_agen (dep_insn, insn))
return 3;
break;
}
case TYPE_STORE:
case TYPE_FPLOAD:
case TYPE_FPSTORE:
{
if (get_attr_update (dep_insn) == UPDATE_YES
&& set_to_load_agen (dep_insn, insn))
return 3;
break;
}
case TYPE_MUL:
{
if (set_to_load_agen (dep_insn, insn))
return 17;
break;
}
case TYPE_DIV:
{
if (set_to_load_agen (dep_insn, insn))
return get_attr_size (dep_insn) == SIZE_32 ? 45 : 57;
break;
}
default:
break;
}
}
break;
case TYPE_FPLOAD:
if ((rs6000_tune == PROCESSOR_POWER6)
&& get_attr_update (insn) == UPDATE_NO
&& recog_memoized (dep_insn)
&& (INSN_CODE (dep_insn) >= 0)
&& (get_attr_type (dep_insn) == TYPE_MFFGPR))
return 2;
default:
break;
}
/* Fall out to return default cost. */
}
break;
case REG_DEP_OUTPUT:
/* Output dependency; DEP_INSN writes a register that INSN writes some
cycles later. */
if ((rs6000_tune == PROCESSOR_POWER6)
&& recog_memoized (dep_insn)
&& (INSN_CODE (dep_insn) >= 0))
{
attr_type = get_attr_type (insn);
switch (attr_type)
{
case TYPE_FP:
case TYPE_FPSIMPLE:
if (get_attr_type (dep_insn) == TYPE_FP
|| get_attr_type (dep_insn) == TYPE_FPSIMPLE)
return 1;
break;
case TYPE_FPLOAD:
if (get_attr_update (insn) == UPDATE_NO
&& get_attr_type (dep_insn) == TYPE_MFFGPR)
return 2;
break;
default:
break;
}
}
/* Fall through, no cost for output dependency. */
/* FALLTHRU */
case REG_DEP_ANTI:
/* Anti dependency; DEP_INSN reads a register that INSN writes some
cycles later. */
return 0;
default:
gcc_unreachable ();
}
return cost;
}
/* Debug version of rs6000_adjust_cost. */
static int
rs6000_debug_adjust_cost (rtx_insn *insn, int dep_type, rtx_insn *dep_insn,
int cost, unsigned int dw)
{
int ret = rs6000_adjust_cost (insn, dep_type, dep_insn, cost, dw);
if (ret != cost)
{
const char *dep;
switch (dep_type)
{
default: dep = "unknown depencency"; break;
case REG_DEP_TRUE: dep = "data dependency"; break;
case REG_DEP_OUTPUT: dep = "output dependency"; break;
case REG_DEP_ANTI: dep = "anti depencency"; break;
}
fprintf (stderr,
"\nrs6000_adjust_cost, final cost = %d, orig cost = %d, "
"%s, insn:\n", ret, cost, dep);
debug_rtx (insn);
}
return ret;
}
/* The function returns a true if INSN is microcoded.
Return false otherwise. */
static bool
is_microcoded_insn (rtx_insn *insn)
{
if (!insn || !NONDEBUG_INSN_P (insn)
|| GET_CODE (PATTERN (insn)) == USE
|| GET_CODE (PATTERN (insn)) == CLOBBER)
return false;
if (rs6000_tune == PROCESSOR_CELL)
return get_attr_cell_micro (insn) == CELL_MICRO_ALWAYS;
if (rs6000_sched_groups
&& (rs6000_tune == PROCESSOR_POWER4 || rs6000_tune == PROCESSOR_POWER5))
{
enum attr_type type = get_attr_type (insn);
if ((type == TYPE_LOAD
&& get_attr_update (insn) == UPDATE_YES
&& get_attr_sign_extend (insn) == SIGN_EXTEND_YES)
|| ((type == TYPE_LOAD || type == TYPE_STORE)
&& get_attr_update (insn) == UPDATE_YES
&& get_attr_indexed (insn) == INDEXED_YES)
|| type == TYPE_MFCR)
return true;
}
return false;
}
/* The function returns true if INSN is cracked into 2 instructions
by the processor (and therefore occupies 2 issue slots). */
static bool
is_cracked_insn (rtx_insn *insn)
{
if (!insn || !NONDEBUG_INSN_P (insn)
|| GET_CODE (PATTERN (insn)) == USE
|| GET_CODE (PATTERN (insn)) == CLOBBER)
return false;
if (rs6000_sched_groups
&& (rs6000_tune == PROCESSOR_POWER4 || rs6000_tune == PROCESSOR_POWER5))
{
enum attr_type type = get_attr_type (insn);
if ((type == TYPE_LOAD
&& get_attr_sign_extend (insn) == SIGN_EXTEND_YES
&& get_attr_update (insn) == UPDATE_NO)
|| (type == TYPE_LOAD
&& get_attr_sign_extend (insn) == SIGN_EXTEND_NO
&& get_attr_update (insn) == UPDATE_YES
&& get_attr_indexed (insn) == INDEXED_NO)
|| (type == TYPE_STORE
&& get_attr_update (insn) == UPDATE_YES
&& get_attr_indexed (insn) == INDEXED_NO)
|| ((type == TYPE_FPLOAD || type == TYPE_FPSTORE)
&& get_attr_update (insn) == UPDATE_YES)
|| type == TYPE_DELAYED_CR
|| (type == TYPE_EXTS
&& get_attr_dot (insn) == DOT_YES)
|| (type == TYPE_SHIFT
&& get_attr_dot (insn) == DOT_YES
&& get_attr_var_shift (insn) == VAR_SHIFT_NO)
|| (type == TYPE_MUL
&& get_attr_dot (insn) == DOT_YES)
|| type == TYPE_DIV
|| (type == TYPE_INSERT
&& get_attr_size (insn) == SIZE_32))
return true;
}
return false;
}
/* The function returns true if INSN can be issued only from
the branch slot. */
static bool
is_branch_slot_insn (rtx_insn *insn)
{
if (!insn || !NONDEBUG_INSN_P (insn)
|| GET_CODE (PATTERN (insn)) == USE
|| GET_CODE (PATTERN (insn)) == CLOBBER)
return false;
if (rs6000_sched_groups)
{
enum attr_type type = get_attr_type (insn);
if (type == TYPE_BRANCH || type == TYPE_JMPREG)
return true;
return false;
}
return false;
}
/* The function returns true if out_inst sets a value that is
used in the address generation computation of in_insn */
static bool
set_to_load_agen (rtx_insn *out_insn, rtx_insn *in_insn)
{
rtx out_set, in_set;
/* For performance reasons, only handle the simple case where
both loads are a single_set. */
out_set = single_set (out_insn);
if (out_set)
{
in_set = single_set (in_insn);
if (in_set)
return reg_mentioned_p (SET_DEST (out_set), SET_SRC (in_set));
}
return false;
}
/* Try to determine base/offset/size parts of the given MEM.
Return true if successful, false if all the values couldn't
be determined.
This function only looks for REG or REG+CONST address forms.
REG+REG address form will return false. */
static bool
get_memref_parts (rtx mem, rtx *base, HOST_WIDE_INT *offset,
HOST_WIDE_INT *size)
{
rtx addr_rtx;
if MEM_SIZE_KNOWN_P (mem)
*size = MEM_SIZE (mem);
else
return false;
addr_rtx = (XEXP (mem, 0));
if (GET_CODE (addr_rtx) == PRE_MODIFY)
addr_rtx = XEXP (addr_rtx, 1);
*offset = 0;
while (GET_CODE (addr_rtx) == PLUS
&& CONST_INT_P (XEXP (addr_rtx, 1)))
{
*offset += INTVAL (XEXP (addr_rtx, 1));
addr_rtx = XEXP (addr_rtx, 0);
}
if (!REG_P (addr_rtx))
return false;
*base = addr_rtx;
return true;
}
/* The function returns true if the target storage location of
mem1 is adjacent to the target storage location of mem2 */
/* Return 1 if memory locations are adjacent. */
static bool
adjacent_mem_locations (rtx mem1, rtx mem2)
{
rtx reg1, reg2;
HOST_WIDE_INT off1, size1, off2, size2;
if (get_memref_parts (mem1, ®1, &off1, &size1)
&& get_memref_parts (mem2, ®2, &off2, &size2))
return ((REGNO (reg1) == REGNO (reg2))
&& ((off1 + size1 == off2)
|| (off2 + size2 == off1)));
return false;
}
/* This function returns true if it can be determined that the two MEM
locations overlap by at least 1 byte based on base reg/offset/size. */
static bool
mem_locations_overlap (rtx mem1, rtx mem2)
{
rtx reg1, reg2;
HOST_WIDE_INT off1, size1, off2, size2;
if (get_memref_parts (mem1, ®1, &off1, &size1)
&& get_memref_parts (mem2, ®2, &off2, &size2))
return ((REGNO (reg1) == REGNO (reg2))
&& (((off1 <= off2) && (off1 + size1 > off2))
|| ((off2 <= off1) && (off2 + size2 > off1))));
return false;
}
/* A C statement (sans semicolon) to update the integer scheduling
priority INSN_PRIORITY (INSN). Increase the priority to execute the
INSN earlier, reduce the priority to execute INSN later. Do not
define this macro if you do not need to adjust the scheduling
priorities of insns. */
static int
rs6000_adjust_priority (rtx_insn *insn ATTRIBUTE_UNUSED, int priority)
{
rtx load_mem, str_mem;
/* On machines (like the 750) which have asymmetric integer units,
where one integer unit can do multiply and divides and the other
can't, reduce the priority of multiply/divide so it is scheduled
before other integer operations. */
#if 0
if (! INSN_P (insn))
return priority;
if (GET_CODE (PATTERN (insn)) == USE)
return priority;
switch (rs6000_tune) {
case PROCESSOR_PPC750:
switch (get_attr_type (insn))
{
default:
break;
case TYPE_MUL:
case TYPE_DIV:
fprintf (stderr, "priority was %#x (%d) before adjustment\n",
priority, priority);
if (priority >= 0 && priority < 0x01000000)
priority >>= 3;
break;
}
}
#endif
if (insn_must_be_first_in_group (insn)
&& reload_completed
&& current_sched_info->sched_max_insns_priority
&& rs6000_sched_restricted_insns_priority)
{
/* Prioritize insns that can be dispatched only in the first
dispatch slot. */
if (rs6000_sched_restricted_insns_priority == 1)
/* Attach highest priority to insn. This means that in
haifa-sched.c:ready_sort(), dispatch-slot restriction considerations
precede 'priority' (critical path) considerations. */
return current_sched_info->sched_max_insns_priority;
else if (rs6000_sched_restricted_insns_priority == 2)
/* Increase priority of insn by a minimal amount. This means that in
haifa-sched.c:ready_sort(), only 'priority' (critical path)
considerations precede dispatch-slot restriction considerations. */
return (priority + 1);
}
if (rs6000_tune == PROCESSOR_POWER6
&& ((load_store_pendulum == -2 && is_load_insn (insn, &load_mem))
|| (load_store_pendulum == 2 && is_store_insn (insn, &str_mem))))
/* Attach highest priority to insn if the scheduler has just issued two
stores and this instruction is a load, or two loads and this instruction
is a store. Power6 wants loads and stores scheduled alternately
when possible */
return current_sched_info->sched_max_insns_priority;
return priority;
}
/* Return true if the instruction is nonpipelined on the Cell. */
static bool
is_nonpipeline_insn (rtx_insn *insn)
{
enum attr_type type;
if (!insn || !NONDEBUG_INSN_P (insn)
|| GET_CODE (PATTERN (insn)) == USE
|| GET_CODE (PATTERN (insn)) == CLOBBER)
return false;
type = get_attr_type (insn);
if (type == TYPE_MUL
|| type == TYPE_DIV
|| type == TYPE_SDIV
|| type == TYPE_DDIV
|| type == TYPE_SSQRT
|| type == TYPE_DSQRT
|| type == TYPE_MFCR
|| type == TYPE_MFCRF
|| type == TYPE_MFJMPR)
{
return true;
}
return false;
}
/* Return how many instructions the machine can issue per cycle. */
static int
rs6000_issue_rate (void)
{
/* Unless scheduling for register pressure, use issue rate of 1 for
first scheduling pass to decrease degradation. */
if (!reload_completed && !flag_sched_pressure)
return 1;
switch (rs6000_tune) {
case PROCESSOR_RS64A:
case PROCESSOR_PPC601: /* ? */
case PROCESSOR_PPC7450:
return 3;
case PROCESSOR_PPC440:
case PROCESSOR_PPC603:
case PROCESSOR_PPC750:
case PROCESSOR_PPC7400:
case PROCESSOR_PPC8540:
case PROCESSOR_PPC8548:
case PROCESSOR_CELL:
case PROCESSOR_PPCE300C2:
case PROCESSOR_PPCE300C3:
case PROCESSOR_PPCE500MC:
case PROCESSOR_PPCE500MC64:
case PROCESSOR_PPCE5500:
case PROCESSOR_PPCE6500:
case PROCESSOR_TITAN:
return 2;
case PROCESSOR_PPC476:
case PROCESSOR_PPC604:
case PROCESSOR_PPC604e:
case PROCESSOR_PPC620:
case PROCESSOR_PPC630:
return 4;
case PROCESSOR_POWER4:
case PROCESSOR_POWER5:
case PROCESSOR_POWER6:
case PROCESSOR_POWER7:
return 5;
case PROCESSOR_POWER8:
return 7;
case PROCESSOR_POWER9:
return 6;
default:
return 1;
}
}
/* Return how many instructions to look ahead for better insn
scheduling. */
static int
rs6000_use_sched_lookahead (void)
{
switch (rs6000_tune)
{
case PROCESSOR_PPC8540:
case PROCESSOR_PPC8548:
return 4;
case PROCESSOR_CELL:
return (reload_completed ? 8 : 0);
default:
return 0;
}
}
/* We are choosing insn from the ready queue. Return zero if INSN can be
chosen. */
static int
rs6000_use_sched_lookahead_guard (rtx_insn *insn, int ready_index)
{
if (ready_index == 0)
return 0;
if (rs6000_tune != PROCESSOR_CELL)
return 0;
gcc_assert (insn != NULL_RTX && INSN_P (insn));
if (!reload_completed
|| is_nonpipeline_insn (insn)
|| is_microcoded_insn (insn))
return 1;
return 0;
}
/* Determine if PAT refers to memory. If so, set MEM_REF to the MEM rtx
and return true. */
static bool
find_mem_ref (rtx pat, rtx *mem_ref)
{
const char * fmt;
int i, j;
/* stack_tie does not produce any real memory traffic. */
if (tie_operand (pat, VOIDmode))
return false;
if (GET_CODE (pat) == MEM)
{
*mem_ref = pat;
return true;
}
/* Recursively process the pattern. */
fmt = GET_RTX_FORMAT (GET_CODE (pat));
for (i = GET_RTX_LENGTH (GET_CODE (pat)) - 1; i >= 0; i--)
{
if (fmt[i] == 'e')
{
if (find_mem_ref (XEXP (pat, i), mem_ref))
return true;
}
else if (fmt[i] == 'E')
for (j = XVECLEN (pat, i) - 1; j >= 0; j--)
{
if (find_mem_ref (XVECEXP (pat, i, j), mem_ref))
return true;
}
}
return false;
}
/* Determine if PAT is a PATTERN of a load insn. */
static bool
is_load_insn1 (rtx pat, rtx *load_mem)
{
if (!pat || pat == NULL_RTX)
return false;
if (GET_CODE (pat) == SET)
return find_mem_ref (SET_SRC (pat), load_mem);
if (GET_CODE (pat) == PARALLEL)
{
int i;
for (i = 0; i < XVECLEN (pat, 0); i++)
if (is_load_insn1 (XVECEXP (pat, 0, i), load_mem))
return true;
}
return false;
}
/* Determine if INSN loads from memory. */
static bool
is_load_insn (rtx insn, rtx *load_mem)
{
if (!insn || !INSN_P (insn))
return false;
if (CALL_P (insn))
return false;
return is_load_insn1 (PATTERN (insn), load_mem);
}
/* Determine if PAT is a PATTERN of a store insn. */
static bool
is_store_insn1 (rtx pat, rtx *str_mem)
{
if (!pat || pat == NULL_RTX)
return false;
if (GET_CODE (pat) == SET)
return find_mem_ref (SET_DEST (pat), str_mem);
if (GET_CODE (pat) == PARALLEL)
{
int i;
for (i = 0; i < XVECLEN (pat, 0); i++)
if (is_store_insn1 (XVECEXP (pat, 0, i), str_mem))
return true;
}
return false;
}
/* Determine if INSN stores to memory. */
static bool
is_store_insn (rtx insn, rtx *str_mem)
{
if (!insn || !INSN_P (insn))
return false;
return is_store_insn1 (PATTERN (insn), str_mem);
}
/* Return whether TYPE is a Power9 pairable vector instruction type. */
static bool
is_power9_pairable_vec_type (enum attr_type type)
{
switch (type)
{
case TYPE_VECSIMPLE:
case TYPE_VECCOMPLEX:
case TYPE_VECDIV:
case TYPE_VECCMP:
case TYPE_VECPERM:
case TYPE_VECFLOAT:
case TYPE_VECFDIV:
case TYPE_VECDOUBLE:
return true;
default:
break;
}
return false;
}
/* Returns whether the dependence between INSN and NEXT is considered
costly by the given target. */
static bool
rs6000_is_costly_dependence (dep_t dep, int cost, int distance)
{
rtx insn;
rtx next;
rtx load_mem, str_mem;
/* If the flag is not enabled - no dependence is considered costly;
allow all dependent insns in the same group.
This is the most aggressive option. */
if (rs6000_sched_costly_dep == no_dep_costly)
return false;
/* If the flag is set to 1 - a dependence is always considered costly;
do not allow dependent instructions in the same group.
This is the most conservative option. */
if (rs6000_sched_costly_dep == all_deps_costly)
return true;
insn = DEP_PRO (dep);
next = DEP_CON (dep);
if (rs6000_sched_costly_dep == store_to_load_dep_costly
&& is_load_insn (next, &load_mem)
&& is_store_insn (insn, &str_mem))
/* Prevent load after store in the same group. */
return true;
if (rs6000_sched_costly_dep == true_store_to_load_dep_costly
&& is_load_insn (next, &load_mem)
&& is_store_insn (insn, &str_mem)
&& DEP_TYPE (dep) == REG_DEP_TRUE
&& mem_locations_overlap(str_mem, load_mem))
/* Prevent load after store in the same group if it is a true
dependence. */
return true;
/* The flag is set to X; dependences with latency >= X are considered costly,
and will not be scheduled in the same group. */
if (rs6000_sched_costly_dep <= max_dep_latency
&& ((cost - distance) >= (int)rs6000_sched_costly_dep))
return true;
return false;
}
/* Return the next insn after INSN that is found before TAIL is reached,
skipping any "non-active" insns - insns that will not actually occupy
an issue slot. Return NULL_RTX if such an insn is not found. */
static rtx_insn *
get_next_active_insn (rtx_insn *insn, rtx_insn *tail)
{
if (insn == NULL_RTX || insn == tail)
return NULL;
while (1)
{
insn = NEXT_INSN (insn);
if (insn == NULL_RTX || insn == tail)
return NULL;
if (CALL_P (insn)
|| JUMP_P (insn) || JUMP_TABLE_DATA_P (insn)
|| (NONJUMP_INSN_P (insn)
&& GET_CODE (PATTERN (insn)) != USE
&& GET_CODE (PATTERN (insn)) != CLOBBER
&& INSN_CODE (insn) != CODE_FOR_stack_tie))
break;
}
return insn;
}
/* Do Power9 specific sched_reorder2 reordering of ready list. */
static int
power9_sched_reorder2 (rtx_insn **ready, int lastpos)
{
int pos;
int i;
rtx_insn *tmp;
enum attr_type type, type2;
type = get_attr_type (last_scheduled_insn);
/* Try to issue fixed point divides back-to-back in pairs so they will be
routed to separate execution units and execute in parallel. */
if (type == TYPE_DIV && divide_cnt == 0)
{
/* First divide has been scheduled. */
divide_cnt = 1;
/* Scan the ready list looking for another divide, if found move it
to the end of the list so it is chosen next. */
pos = lastpos;
while (pos >= 0)
{
if (recog_memoized (ready[pos]) >= 0
&& get_attr_type (ready[pos]) == TYPE_DIV)
{
tmp = ready[pos];
for (i = pos; i < lastpos; i++)
ready[i] = ready[i + 1];
ready[lastpos] = tmp;
break;
}
pos--;
}
}
else
{
/* Last insn was the 2nd divide or not a divide, reset the counter. */
divide_cnt = 0;
/* The best dispatch throughput for vector and vector load insns can be
achieved by interleaving a vector and vector load such that they'll
dispatch to the same superslice. If this pairing cannot be achieved
then it is best to pair vector insns together and vector load insns
together.
To aid in this pairing, vec_pairing maintains the current state with
the following values:
0 : Initial state, no vecload/vector pairing has been started.
1 : A vecload or vector insn has been issued and a candidate for
pairing has been found and moved to the end of the ready
list. */
if (type == TYPE_VECLOAD)
{
/* Issued a vecload. */
if (vec_pairing == 0)
{
int vecload_pos = -1;
/* We issued a single vecload, look for a vector insn to pair it
with. If one isn't found, try to pair another vecload. */
pos = lastpos;
while (pos >= 0)
{
if (recog_memoized (ready[pos]) >= 0)
{
type2 = get_attr_type (ready[pos]);
if (is_power9_pairable_vec_type (type2))
{
/* Found a vector insn to pair with, move it to the
end of the ready list so it is scheduled next. */
tmp = ready[pos];
for (i = pos; i < lastpos; i++)
ready[i] = ready[i + 1];
ready[lastpos] = tmp;
vec_pairing = 1;
return cached_can_issue_more;
}
else if (type2 == TYPE_VECLOAD && vecload_pos == -1)
/* Remember position of first vecload seen. */
vecload_pos = pos;
}
pos--;
}
if (vecload_pos >= 0)
{
/* Didn't find a vector to pair with but did find a vecload,
move it to the end of the ready list. */
tmp = ready[vecload_pos];
for (i = vecload_pos; i < lastpos; i++)
ready[i] = ready[i + 1];
ready[lastpos] = tmp;
vec_pairing = 1;
return cached_can_issue_more;
}
}
}
else if (is_power9_pairable_vec_type (type))
{
/* Issued a vector operation. */
if (vec_pairing == 0)
{
int vec_pos = -1;
/* We issued a single vector insn, look for a vecload to pair it
with. If one isn't found, try to pair another vector. */
pos = lastpos;
while (pos >= 0)
{
if (recog_memoized (ready[pos]) >= 0)
{
type2 = get_attr_type (ready[pos]);
if (type2 == TYPE_VECLOAD)
{
/* Found a vecload insn to pair with, move it to the
end of the ready list so it is scheduled next. */
tmp = ready[pos];
for (i = pos; i < lastpos; i++)
ready[i] = ready[i + 1];
ready[lastpos] = tmp;
vec_pairing = 1;
return cached_can_issue_more;
}
else if (is_power9_pairable_vec_type (type2)
&& vec_pos == -1)
/* Remember position of first vector insn seen. */
vec_pos = pos;
}
pos--;
}
if (vec_pos >= 0)
{
/* Didn't find a vecload to pair with but did find a vector
insn, move it to the end of the ready list. */
tmp = ready[vec_pos];
for (i = vec_pos; i < lastpos; i++)
ready[i] = ready[i + 1];
ready[lastpos] = tmp;
vec_pairing = 1;
return cached_can_issue_more;
}
}
}
/* We've either finished a vec/vecload pair, couldn't find an insn to
continue the current pair, or the last insn had nothing to do with
with pairing. In any case, reset the state. */
vec_pairing = 0;
}
return cached_can_issue_more;
}
/* We are about to begin issuing insns for this clock cycle. */
static int
rs6000_sched_reorder (FILE *dump ATTRIBUTE_UNUSED, int sched_verbose,
rtx_insn **ready ATTRIBUTE_UNUSED,
int *pn_ready ATTRIBUTE_UNUSED,
int clock_var ATTRIBUTE_UNUSED)
{
int n_ready = *pn_ready;
if (sched_verbose)
fprintf (dump, "// rs6000_sched_reorder :\n");
/* Reorder the ready list, if the second to last ready insn
is a nonepipeline insn. */
if (rs6000_tune == PROCESSOR_CELL && n_ready > 1)
{
if (is_nonpipeline_insn (ready[n_ready - 1])
&& (recog_memoized (ready[n_ready - 2]) > 0))
/* Simply swap first two insns. */
std::swap (ready[n_ready - 1], ready[n_ready - 2]);
}
if (rs6000_tune == PROCESSOR_POWER6)
load_store_pendulum = 0;
return rs6000_issue_rate ();
}
/* Like rs6000_sched_reorder, but called after issuing each insn. */
static int
rs6000_sched_reorder2 (FILE *dump, int sched_verbose, rtx_insn **ready,
int *pn_ready, int clock_var ATTRIBUTE_UNUSED)
{
if (sched_verbose)
fprintf (dump, "// rs6000_sched_reorder2 :\n");
/* For Power6, we need to handle some special cases to try and keep the
store queue from overflowing and triggering expensive flushes.
This code monitors how load and store instructions are being issued
and skews the ready list one way or the other to increase the likelihood
that a desired instruction is issued at the proper time.
A couple of things are done. First, we maintain a "load_store_pendulum"
to track the current state of load/store issue.
- If the pendulum is at zero, then no loads or stores have been
issued in the current cycle so we do nothing.
- If the pendulum is 1, then a single load has been issued in this
cycle and we attempt to locate another load in the ready list to
issue with it.
- If the pendulum is -2, then two stores have already been
issued in this cycle, so we increase the priority of the first load
in the ready list to increase it's likelihood of being chosen first
in the next cycle.
- If the pendulum is -1, then a single store has been issued in this
cycle and we attempt to locate another store in the ready list to
issue with it, preferring a store to an adjacent memory location to
facilitate store pairing in the store queue.
- If the pendulum is 2, then two loads have already been
issued in this cycle, so we increase the priority of the first store
in the ready list to increase it's likelihood of being chosen first
in the next cycle.
- If the pendulum < -2 or > 2, then do nothing.
Note: This code covers the most common scenarios. There exist non
load/store instructions which make use of the LSU and which
would need to be accounted for to strictly model the behavior
of the machine. Those instructions are currently unaccounted
for to help minimize compile time overhead of this code.
*/
if (rs6000_tune == PROCESSOR_POWER6 && last_scheduled_insn)
{
int pos;
int i;
rtx_insn *tmp;
rtx load_mem, str_mem;
if (is_store_insn (last_scheduled_insn, &str_mem))
/* Issuing a store, swing the load_store_pendulum to the left */
load_store_pendulum--;
else if (is_load_insn (last_scheduled_insn, &load_mem))
/* Issuing a load, swing the load_store_pendulum to the right */
load_store_pendulum++;
else
return cached_can_issue_more;
/* If the pendulum is balanced, or there is only one instruction on
the ready list, then all is well, so return. */
if ((load_store_pendulum == 0) || (*pn_ready <= 1))
return cached_can_issue_more;
if (load_store_pendulum == 1)
{
/* A load has been issued in this cycle. Scan the ready list
for another load to issue with it */
pos = *pn_ready-1;
while (pos >= 0)
{
if (is_load_insn (ready[pos], &load_mem))
{
/* Found a load. Move it to the head of the ready list,
and adjust it's priority so that it is more likely to
stay there */
tmp = ready[pos];
for (i=pos; i<*pn_ready-1; i++)
ready[i] = ready[i + 1];
ready[*pn_ready-1] = tmp;
if (!sel_sched_p () && INSN_PRIORITY_KNOWN (tmp))
INSN_PRIORITY (tmp)++;
break;
}
pos--;
}
}
else if (load_store_pendulum == -2)
{
/* Two stores have been issued in this cycle. Increase the
priority of the first load in the ready list to favor it for
issuing in the next cycle. */
pos = *pn_ready-1;
while (pos >= 0)
{
if (is_load_insn (ready[pos], &load_mem)
&& !sel_sched_p ()
&& INSN_PRIORITY_KNOWN (ready[pos]))
{
INSN_PRIORITY (ready[pos])++;
/* Adjust the pendulum to account for the fact that a load
was found and increased in priority. This is to prevent
increasing the priority of multiple loads */
load_store_pendulum--;
break;
}
pos--;
}
}
else if (load_store_pendulum == -1)
{
/* A store has been issued in this cycle. Scan the ready list for
another store to issue with it, preferring a store to an adjacent
memory location */
int first_store_pos = -1;
pos = *pn_ready-1;
while (pos >= 0)
{
if (is_store_insn (ready[pos], &str_mem))
{
rtx str_mem2;
/* Maintain the index of the first store found on the
list */
if (first_store_pos == -1)
first_store_pos = pos;
if (is_store_insn (last_scheduled_insn, &str_mem2)
&& adjacent_mem_locations (str_mem, str_mem2))
{
/* Found an adjacent store. Move it to the head of the
ready list, and adjust it's priority so that it is
more likely to stay there */
tmp = ready[pos];
for (i=pos; i<*pn_ready-1; i++)
ready[i] = ready[i + 1];
ready[*pn_ready-1] = tmp;
if (!sel_sched_p () && INSN_PRIORITY_KNOWN (tmp))
INSN_PRIORITY (tmp)++;
first_store_pos = -1;
break;
};
}
pos--;
}
if (first_store_pos >= 0)
{
/* An adjacent store wasn't found, but a non-adjacent store was,
so move the non-adjacent store to the front of the ready
list, and adjust its priority so that it is more likely to
stay there. */
tmp = ready[first_store_pos];
for (i=first_store_pos; i<*pn_ready-1; i++)
ready[i] = ready[i + 1];
ready[*pn_ready-1] = tmp;
if (!sel_sched_p () && INSN_PRIORITY_KNOWN (tmp))
INSN_PRIORITY (tmp)++;
}
}
else if (load_store_pendulum == 2)
{
/* Two loads have been issued in this cycle. Increase the priority
of the first store in the ready list to favor it for issuing in
the next cycle. */
pos = *pn_ready-1;
while (pos >= 0)
{
if (is_store_insn (ready[pos], &str_mem)
&& !sel_sched_p ()
&& INSN_PRIORITY_KNOWN (ready[pos]))
{
INSN_PRIORITY (ready[pos])++;
/* Adjust the pendulum to account for the fact that a store
was found and increased in priority. This is to prevent
increasing the priority of multiple stores */
load_store_pendulum++;
break;
}
pos--;
}
}
}
/* Do Power9 dependent reordering if necessary. */
if (rs6000_tune == PROCESSOR_POWER9 && last_scheduled_insn
&& recog_memoized (last_scheduled_insn) >= 0)
return power9_sched_reorder2 (ready, *pn_ready - 1);
return cached_can_issue_more;
}
/* Return whether the presence of INSN causes a dispatch group termination
of group WHICH_GROUP.
If WHICH_GROUP == current_group, this function will return true if INSN
causes the termination of the current group (i.e, the dispatch group to
which INSN belongs). This means that INSN will be the last insn in the
group it belongs to.
If WHICH_GROUP == previous_group, this function will return true if INSN
causes the termination of the previous group (i.e, the dispatch group that
precedes the group to which INSN belongs). This means that INSN will be
the first insn in the group it belongs to). */
static bool
insn_terminates_group_p (rtx_insn *insn, enum group_termination which_group)
{
bool first, last;
if (! insn)
return false;
first = insn_must_be_first_in_group (insn);
last = insn_must_be_last_in_group (insn);
if (first && last)
return true;
if (which_group == current_group)
return last;
else if (which_group == previous_group)
return first;
return false;
}
static bool
insn_must_be_first_in_group (rtx_insn *insn)
{
enum attr_type type;
if (!insn
|| NOTE_P (insn)
|| DEBUG_INSN_P (insn)
|| GET_CODE (PATTERN (insn)) == USE
|| GET_CODE (PATTERN (insn)) == CLOBBER)
return false;
switch (rs6000_tune)
{
case PROCESSOR_POWER5:
if (is_cracked_insn (insn))
return true;
/* FALLTHRU */
case PROCESSOR_POWER4:
if (is_microcoded_insn (insn))
return true;
if (!rs6000_sched_groups)
return false;
type = get_attr_type (insn);
switch (type)
{
case TYPE_MFCR:
case TYPE_MFCRF:
case TYPE_MTCR:
case TYPE_DELAYED_CR:
case TYPE_CR_LOGICAL:
case TYPE_MTJMPR:
case TYPE_MFJMPR:
case TYPE_DIV:
case TYPE_LOAD_L:
case TYPE_STORE_C:
case TYPE_ISYNC:
case TYPE_SYNC:
return true;
default:
break;
}
break;
case PROCESSOR_POWER6:
type = get_attr_type (insn);
switch (type)
{
case TYPE_EXTS:
case TYPE_CNTLZ:
case TYPE_TRAP:
case TYPE_MUL:
case TYPE_INSERT:
case TYPE_FPCOMPARE:
case TYPE_MFCR:
case TYPE_MTCR:
case TYPE_MFJMPR:
case TYPE_MTJMPR:
case TYPE_ISYNC:
case TYPE_SYNC:
case TYPE_LOAD_L:
case TYPE_STORE_C:
return true;
case TYPE_SHIFT:
if (get_attr_dot (insn) == DOT_NO
|| get_attr_var_shift (insn) == VAR_SHIFT_NO)
return true;
else
break;
case TYPE_DIV:
if (get_attr_size (insn) == SIZE_32)
return true;
else
break;
case TYPE_LOAD:
case TYPE_STORE:
case TYPE_FPLOAD:
case TYPE_FPSTORE:
if (get_attr_update (insn) == UPDATE_YES)
return true;
else
break;
default:
break;
}
break;
case PROCESSOR_POWER7:
type = get_attr_type (insn);
switch (type)
{
case TYPE_CR_LOGICAL:
case TYPE_MFCR:
case TYPE_MFCRF:
case TYPE_MTCR:
case TYPE_DIV:
case TYPE_ISYNC:
case TYPE_LOAD_L:
case TYPE_STORE_C:
case TYPE_MFJMPR:
case TYPE_MTJMPR:
return true;
case TYPE_MUL:
case TYPE_SHIFT:
case TYPE_EXTS:
if (get_attr_dot (insn) == DOT_YES)
return true;
else
break;
case TYPE_LOAD:
if (get_attr_sign_extend (insn) == SIGN_EXTEND_YES
|| get_attr_update (insn) == UPDATE_YES)
return true;
else
break;
case TYPE_STORE:
case TYPE_FPLOAD:
case TYPE_FPSTORE:
if (get_attr_update (insn) == UPDATE_YES)
return true;
else
break;
default:
break;
}
break;
case PROCESSOR_POWER8:
type = get_attr_type (insn);
switch (type)
{
case TYPE_CR_LOGICAL:
case TYPE_DELAYED_CR:
case TYPE_MFCR:
case TYPE_MFCRF:
case TYPE_MTCR:
case TYPE_SYNC:
case TYPE_ISYNC:
case TYPE_LOAD_L:
case TYPE_STORE_C:
case TYPE_VECSTORE:
case TYPE_MFJMPR:
case TYPE_MTJMPR:
return true;
case TYPE_SHIFT:
case TYPE_EXTS:
case TYPE_MUL:
if (get_attr_dot (insn) == DOT_YES)
return true;
else
break;
case TYPE_LOAD:
if (get_attr_sign_extend (insn) == SIGN_EXTEND_YES
|| get_attr_update (insn) == UPDATE_YES)
return true;
else
break;
case TYPE_STORE:
if (get_attr_update (insn) == UPDATE_YES
&& get_attr_indexed (insn) == INDEXED_YES)
return true;
else
break;
default:
break;
}
break;
default:
break;
}
return false;
}
static bool
insn_must_be_last_in_group (rtx_insn *insn)
{
enum attr_type type;
if (!insn
|| NOTE_P (insn)
|| DEBUG_INSN_P (insn)
|| GET_CODE (PATTERN (insn)) == USE
|| GET_CODE (PATTERN (insn)) == CLOBBER)
return false;
switch (rs6000_tune) {
case PROCESSOR_POWER4:
case PROCESSOR_POWER5:
if (is_microcoded_insn (insn))
return true;
if (is_branch_slot_insn (insn))
return true;
break;
case PROCESSOR_POWER6:
type = get_attr_type (insn);
switch (type)
{
case TYPE_EXTS:
case TYPE_CNTLZ:
case TYPE_TRAP:
case TYPE_MUL:
case TYPE_FPCOMPARE:
case TYPE_MFCR:
case TYPE_MTCR:
case TYPE_MFJMPR:
case TYPE_MTJMPR:
case TYPE_ISYNC:
case TYPE_SYNC:
case TYPE_LOAD_L:
case TYPE_STORE_C:
return true;
case TYPE_SHIFT:
if (get_attr_dot (insn) == DOT_NO
|| get_attr_var_shift (insn) == VAR_SHIFT_NO)
return true;
else
break;
case TYPE_DIV:
if (get_attr_size (insn) == SIZE_32)
return true;
else
break;
default:
break;
}
break;
case PROCESSOR_POWER7:
type = get_attr_type (insn);
switch (type)
{
case TYPE_ISYNC:
case TYPE_SYNC:
case TYPE_LOAD_L:
case TYPE_STORE_C:
return true;
case TYPE_LOAD:
if (get_attr_sign_extend (insn) == SIGN_EXTEND_YES
&& get_attr_update (insn) == UPDATE_YES)
return true;
else
break;
case TYPE_STORE:
if (get_attr_update (insn) == UPDATE_YES
&& get_attr_indexed (insn) == INDEXED_YES)
return true;
else
break;
default:
break;
}
break;
case PROCESSOR_POWER8:
type = get_attr_type (insn);
switch (type)
{
case TYPE_MFCR:
case TYPE_MTCR:
case TYPE_ISYNC:
case TYPE_SYNC:
case TYPE_LOAD_L:
case TYPE_STORE_C:
return true;
case TYPE_LOAD:
if (get_attr_sign_extend (insn) == SIGN_EXTEND_YES
&& get_attr_update (insn) == UPDATE_YES)
return true;
else
break;
case TYPE_STORE:
if (get_attr_update (insn) == UPDATE_YES
&& get_attr_indexed (insn) == INDEXED_YES)
return true;
else
break;
default:
break;
}
break;
default:
break;
}
return false;
}
/* Return true if it is recommended to keep NEXT_INSN "far" (in a separate
dispatch group) from the insns in GROUP_INSNS. Return false otherwise. */
static bool
is_costly_group (rtx *group_insns, rtx next_insn)
{
int i;
int issue_rate = rs6000_issue_rate ();
for (i = 0; i < issue_rate; i++)
{
sd_iterator_def sd_it;
dep_t dep;
rtx insn = group_insns[i];
if (!insn)
continue;
FOR_EACH_DEP (insn, SD_LIST_RES_FORW, sd_it, dep)
{
rtx next = DEP_CON (dep);
if (next == next_insn
&& rs6000_is_costly_dependence (dep, dep_cost (dep), 0))
return true;
}
}
return false;
}
/* Utility of the function redefine_groups.
Check if it is too costly to schedule NEXT_INSN together with GROUP_INSNS
in the same dispatch group. If so, insert nops before NEXT_INSN, in order
to keep it "far" (in a separate group) from GROUP_INSNS, following
one of the following schemes, depending on the value of the flag
-minsert_sched_nops = X:
(1) X == sched_finish_regroup_exact: insert exactly as many nops as needed
in order to force NEXT_INSN into a separate group.
(2) X < sched_finish_regroup_exact: insert exactly X nops.
GROUP_END, CAN_ISSUE_MORE and GROUP_COUNT record the state after nop
insertion (has a group just ended, how many vacant issue slots remain in the
last group, and how many dispatch groups were encountered so far). */
static int
force_new_group (int sched_verbose, FILE *dump, rtx *group_insns,
rtx_insn *next_insn, bool *group_end, int can_issue_more,
int *group_count)
{
rtx nop;
bool force;
int issue_rate = rs6000_issue_rate ();
bool end = *group_end;
int i;
if (next_insn == NULL_RTX || DEBUG_INSN_P (next_insn))
return can_issue_more;
if (rs6000_sched_insert_nops > sched_finish_regroup_exact)
return can_issue_more;
force = is_costly_group (group_insns, next_insn);
if (!force)
return can_issue_more;
if (sched_verbose > 6)
fprintf (dump,"force: group count = %d, can_issue_more = %d\n",
*group_count ,can_issue_more);
if (rs6000_sched_insert_nops == sched_finish_regroup_exact)
{
if (*group_end)
can_issue_more = 0;
/* Since only a branch can be issued in the last issue_slot, it is
sufficient to insert 'can_issue_more - 1' nops if next_insn is not
a branch. If next_insn is a branch, we insert 'can_issue_more' nops;
in this case the last nop will start a new group and the branch
will be forced to the new group. */
if (can_issue_more && !is_branch_slot_insn (next_insn))
can_issue_more--;
/* Do we have a special group ending nop? */
if (rs6000_tune == PROCESSOR_POWER6 || rs6000_tune == PROCESSOR_POWER7
|| rs6000_tune == PROCESSOR_POWER8)
{
nop = gen_group_ending_nop ();
emit_insn_before (nop, next_insn);
can_issue_more = 0;
}
else
while (can_issue_more > 0)
{
nop = gen_nop ();
emit_insn_before (nop, next_insn);
can_issue_more--;
}
*group_end = true;
return 0;
}
if (rs6000_sched_insert_nops < sched_finish_regroup_exact)
{
int n_nops = rs6000_sched_insert_nops;
/* Nops can't be issued from the branch slot, so the effective
issue_rate for nops is 'issue_rate - 1'. */
if (can_issue_more == 0)
can_issue_more = issue_rate;
can_issue_more--;
if (can_issue_more == 0)
{
can_issue_more = issue_rate - 1;
(*group_count)++;
end = true;
for (i = 0; i < issue_rate; i++)
{
group_insns[i] = 0;
}
}
while (n_nops > 0)
{
nop = gen_nop ();
emit_insn_before (nop, next_insn);
if (can_issue_more == issue_rate - 1) /* new group begins */
end = false;
can_issue_more--;
if (can_issue_more == 0)
{
can_issue_more = issue_rate - 1;
(*group_count)++;
end = true;
for (i = 0; i < issue_rate; i++)
{
group_insns[i] = 0;
}
}
n_nops--;
}
/* Scale back relative to 'issue_rate' (instead of 'issue_rate - 1'). */
can_issue_more++;
/* Is next_insn going to start a new group? */
*group_end
= (end
|| (can_issue_more == 1 && !is_branch_slot_insn (next_insn))
|| (can_issue_more <= 2 && is_cracked_insn (next_insn))
|| (can_issue_more < issue_rate &&
insn_terminates_group_p (next_insn, previous_group)));
if (*group_end && end)
(*group_count)--;
if (sched_verbose > 6)
fprintf (dump, "done force: group count = %d, can_issue_more = %d\n",
*group_count, can_issue_more);
return can_issue_more;
}
return can_issue_more;
}
/* This function tries to synch the dispatch groups that the compiler "sees"
with the dispatch groups that the processor dispatcher is expected to
form in practice. It tries to achieve this synchronization by forcing the
estimated processor grouping on the compiler (as opposed to the function
'pad_goups' which tries to force the scheduler's grouping on the processor).
The function scans the insn sequence between PREV_HEAD_INSN and TAIL and
examines the (estimated) dispatch groups that will be formed by the processor
dispatcher. It marks these group boundaries to reflect the estimated
processor grouping, overriding the grouping that the scheduler had marked.
Depending on the value of the flag '-minsert-sched-nops' this function can
force certain insns into separate groups or force a certain distance between
them by inserting nops, for example, if there exists a "costly dependence"
between the insns.
The function estimates the group boundaries that the processor will form as
follows: It keeps track of how many vacant issue slots are available after
each insn. A subsequent insn will start a new group if one of the following
4 cases applies:
- no more vacant issue slots remain in the current dispatch group.
- only the last issue slot, which is the branch slot, is vacant, but the next
insn is not a branch.
- only the last 2 or less issue slots, including the branch slot, are vacant,
which means that a cracked insn (which occupies two issue slots) can't be
issued in this group.
- less than 'issue_rate' slots are vacant, and the next insn always needs to
start a new group. */
static int
redefine_groups (FILE *dump, int sched_verbose, rtx_insn *prev_head_insn,
rtx_insn *tail)
{
rtx_insn *insn, *next_insn;
int issue_rate;
int can_issue_more;
int slot, i;
bool group_end;
int group_count = 0;
rtx *group_insns;
/* Initialize. */
issue_rate = rs6000_issue_rate ();
group_insns = XALLOCAVEC (rtx, issue_rate);
for (i = 0; i < issue_rate; i++)
{
group_insns[i] = 0;
}
can_issue_more = issue_rate;
slot = 0;
insn = get_next_active_insn (prev_head_insn, tail);
group_end = false;
while (insn != NULL_RTX)
{
slot = (issue_rate - can_issue_more);
group_insns[slot] = insn;
can_issue_more =
rs6000_variable_issue (dump, sched_verbose, insn, can_issue_more);
if (insn_terminates_group_p (insn, current_group))
can_issue_more = 0;
next_insn = get_next_active_insn (insn, tail);
if (next_insn == NULL_RTX)
return group_count + 1;
/* Is next_insn going to start a new group? */
group_end
= (can_issue_more == 0
|| (can_issue_more == 1 && !is_branch_slot_insn (next_insn))
|| (can_issue_more <= 2 && is_cracked_insn (next_insn))
|| (can_issue_more < issue_rate &&
insn_terminates_group_p (next_insn, previous_group)));
can_issue_more = force_new_group (sched_verbose, dump, group_insns,
next_insn, &group_end, can_issue_more,
&group_count);
if (group_end)
{
group_count++;
can_issue_more = 0;
for (i = 0; i < issue_rate; i++)
{
group_insns[i] = 0;
}
}
if (GET_MODE (next_insn) == TImode && can_issue_more)
PUT_MODE (next_insn, VOIDmode);
else if (!can_issue_more && GET_MODE (next_insn) != TImode)
PUT_MODE (next_insn, TImode);
insn = next_insn;
if (can_issue_more == 0)
can_issue_more = issue_rate;
} /* while */
return group_count;
}
/* Scan the insn sequence between PREV_HEAD_INSN and TAIL and examine the
dispatch group boundaries that the scheduler had marked. Pad with nops
any dispatch groups which have vacant issue slots, in order to force the
scheduler's grouping on the processor dispatcher. The function
returns the number of dispatch groups found. */
static int
pad_groups (FILE *dump, int sched_verbose, rtx_insn *prev_head_insn,
rtx_insn *tail)
{
rtx_insn *insn, *next_insn;
rtx nop;
int issue_rate;
int can_issue_more;
int group_end;
int group_count = 0;
/* Initialize issue_rate. */
issue_rate = rs6000_issue_rate ();
can_issue_more = issue_rate;
insn = get_next_active_insn (prev_head_insn, tail);
next_insn = get_next_active_insn (insn, tail);
while (insn != NULL_RTX)
{
can_issue_more =
rs6000_variable_issue (dump, sched_verbose, insn, can_issue_more);
group_end = (next_insn == NULL_RTX || GET_MODE (next_insn) == TImode);
if (next_insn == NULL_RTX)
break;
if (group_end)
{
/* If the scheduler had marked group termination at this location
(between insn and next_insn), and neither insn nor next_insn will
force group termination, pad the group with nops to force group
termination. */
if (can_issue_more
&& (rs6000_sched_insert_nops == sched_finish_pad_groups)
&& !insn_terminates_group_p (insn, current_group)
&& !insn_terminates_group_p (next_insn, previous_group))
{
if (!is_branch_slot_insn (next_insn))
can_issue_more--;
while (can_issue_more)
{
nop = gen_nop ();
emit_insn_before (nop, next_insn);
can_issue_more--;
}
}
can_issue_more = issue_rate;
group_count++;
}
insn = next_insn;
next_insn = get_next_active_insn (insn, tail);
}
return group_count;
}
/* We're beginning a new block. Initialize data structures as necessary. */
static void
rs6000_sched_init (FILE *dump ATTRIBUTE_UNUSED,
int sched_verbose ATTRIBUTE_UNUSED,
int max_ready ATTRIBUTE_UNUSED)
{
last_scheduled_insn = NULL;
load_store_pendulum = 0;
divide_cnt = 0;
vec_pairing = 0;
}
/* The following function is called at the end of scheduling BB.
After reload, it inserts nops at insn group bundling. */
static void
rs6000_sched_finish (FILE *dump, int sched_verbose)
{
int n_groups;
if (sched_verbose)
fprintf (dump, "=== Finishing schedule.\n");
if (reload_completed && rs6000_sched_groups)
{
/* Do not run sched_finish hook when selective scheduling enabled. */
if (sel_sched_p ())
return;
if (rs6000_sched_insert_nops == sched_finish_none)
return;
if (rs6000_sched_insert_nops == sched_finish_pad_groups)
n_groups = pad_groups (dump, sched_verbose,
current_sched_info->prev_head,
current_sched_info->next_tail);
else
n_groups = redefine_groups (dump, sched_verbose,
current_sched_info->prev_head,
current_sched_info->next_tail);
if (sched_verbose >= 6)
{
fprintf (dump, "ngroups = %d\n", n_groups);
print_rtl (dump, current_sched_info->prev_head);
fprintf (dump, "Done finish_sched\n");
}
}
}
struct rs6000_sched_context
{
short cached_can_issue_more;
rtx_insn *last_scheduled_insn;
int load_store_pendulum;
int divide_cnt;
int vec_pairing;
};
typedef struct rs6000_sched_context rs6000_sched_context_def;
typedef rs6000_sched_context_def *rs6000_sched_context_t;
/* Allocate store for new scheduling context. */
static void *
rs6000_alloc_sched_context (void)
{
return xmalloc (sizeof (rs6000_sched_context_def));
}
/* If CLEAN_P is true then initializes _SC with clean data,
and from the global context otherwise. */
static void
rs6000_init_sched_context (void *_sc, bool clean_p)
{
rs6000_sched_context_t sc = (rs6000_sched_context_t) _sc;
if (clean_p)
{
sc->cached_can_issue_more = 0;
sc->last_scheduled_insn = NULL;
sc->load_store_pendulum = 0;
sc->divide_cnt = 0;
sc->vec_pairing = 0;
}
else
{
sc->cached_can_issue_more = cached_can_issue_more;
sc->last_scheduled_insn = last_scheduled_insn;
sc->load_store_pendulum = load_store_pendulum;
sc->divide_cnt = divide_cnt;
sc->vec_pairing = vec_pairing;
}
}
/* Sets the global scheduling context to the one pointed to by _SC. */
static void
rs6000_set_sched_context (void *_sc)
{
rs6000_sched_context_t sc = (rs6000_sched_context_t) _sc;
gcc_assert (sc != NULL);
cached_can_issue_more = sc->cached_can_issue_more;
last_scheduled_insn = sc->last_scheduled_insn;
load_store_pendulum = sc->load_store_pendulum;
divide_cnt = sc->divide_cnt;
vec_pairing = sc->vec_pairing;
}
/* Free _SC. */
static void
rs6000_free_sched_context (void *_sc)
{
gcc_assert (_sc != NULL);
free (_sc);
}
static bool
rs6000_sched_can_speculate_insn (rtx_insn *insn)
{
switch (get_attr_type (insn))
{
case TYPE_DIV:
case TYPE_SDIV:
case TYPE_DDIV:
case TYPE_VECDIV:
case TYPE_SSQRT:
case TYPE_DSQRT:
return false;
default:
return true;
}
}
/* Length in units of the trampoline for entering a nested function. */
int
rs6000_trampoline_size (void)
{
int ret = 0;
switch (DEFAULT_ABI)
{
default:
gcc_unreachable ();
case ABI_AIX:
ret = (TARGET_32BIT) ? 12 : 24;
break;
case ABI_ELFv2:
gcc_assert (!TARGET_32BIT);
ret = 32;
break;
case ABI_DARWIN:
case ABI_V4:
ret = (TARGET_32BIT) ? 40 : 48;
break;
}
return ret;
}
/* Emit RTL insns to initialize the variable parts of a trampoline.
FNADDR is an RTX for the address of the function's pure code.
CXT is an RTX for the static chain value for the function. */
static void
rs6000_trampoline_init (rtx m_tramp, tree fndecl, rtx cxt)
{
int regsize = (TARGET_32BIT) ? 4 : 8;
rtx fnaddr = XEXP (DECL_RTL (fndecl), 0);
rtx ctx_reg = force_reg (Pmode, cxt);
rtx addr = force_reg (Pmode, XEXP (m_tramp, 0));
switch (DEFAULT_ABI)
{
default:
gcc_unreachable ();
/* Under AIX, just build the 3 word function descriptor */
case ABI_AIX:
{
rtx fnmem, fn_reg, toc_reg;
if (!TARGET_POINTERS_TO_NESTED_FUNCTIONS)
error ("you cannot take the address of a nested function if you use "
"the %qs option", "-mno-pointers-to-nested-functions");
fnmem = gen_const_mem (Pmode, force_reg (Pmode, fnaddr));
fn_reg = gen_reg_rtx (Pmode);
toc_reg = gen_reg_rtx (Pmode);
/* Macro to shorten the code expansions below. */
# define MEM_PLUS(MEM, OFFSET) adjust_address (MEM, Pmode, OFFSET)
m_tramp = replace_equiv_address (m_tramp, addr);
emit_move_insn (fn_reg, MEM_PLUS (fnmem, 0));
emit_move_insn (toc_reg, MEM_PLUS (fnmem, regsize));
emit_move_insn (MEM_PLUS (m_tramp, 0), fn_reg);
emit_move_insn (MEM_PLUS (m_tramp, regsize), toc_reg);
emit_move_insn (MEM_PLUS (m_tramp, 2*regsize), ctx_reg);
# undef MEM_PLUS
}
break;
/* Under V.4/eabi/darwin, __trampoline_setup does the real work. */
case ABI_ELFv2:
case ABI_DARWIN:
case ABI_V4:
emit_library_call (gen_rtx_SYMBOL_REF (Pmode, "__trampoline_setup"),
LCT_NORMAL, VOIDmode,
addr, Pmode,
GEN_INT (rs6000_trampoline_size ()), SImode,
fnaddr, Pmode,
ctx_reg, Pmode);
break;
}
}
/* Returns TRUE iff the target attribute indicated by ATTR_ID takes a plain
identifier as an argument, so the front end shouldn't look it up. */
static bool
rs6000_attribute_takes_identifier_p (const_tree attr_id)
{
return is_attribute_p ("altivec", attr_id);
}
/* Handle the "altivec" attribute. The attribute may have
arguments as follows:
__attribute__((altivec(vector__)))
__attribute__((altivec(pixel__))) (always followed by 'unsigned short')
__attribute__((altivec(bool__))) (always followed by 'unsigned')
and may appear more than once (e.g., 'vector bool char') in a
given declaration. */
static tree
rs6000_handle_altivec_attribute (tree *node,
tree name ATTRIBUTE_UNUSED,
tree args,
int flags ATTRIBUTE_UNUSED,
bool *no_add_attrs)
{
tree type = *node, result = NULL_TREE;
machine_mode mode;
int unsigned_p;
char altivec_type
= ((args && TREE_CODE (args) == TREE_LIST && TREE_VALUE (args)
&& TREE_CODE (TREE_VALUE (args)) == IDENTIFIER_NODE)
? *IDENTIFIER_POINTER (TREE_VALUE (args))
: '?');
while (POINTER_TYPE_P (type)
|| TREE_CODE (type) == FUNCTION_TYPE
|| TREE_CODE (type) == METHOD_TYPE
|| TREE_CODE (type) == ARRAY_TYPE)
type = TREE_TYPE (type);
mode = TYPE_MODE (type);
/* Check for invalid AltiVec type qualifiers. */
if (type == long_double_type_node)
error ("use of %<long double%> in AltiVec types is invalid");
else if (type == boolean_type_node)
error ("use of boolean types in AltiVec types is invalid");
else if (TREE_CODE (type) == COMPLEX_TYPE)
error ("use of %<complex%> in AltiVec types is invalid");
else if (DECIMAL_FLOAT_MODE_P (mode))
error ("use of decimal floating point types in AltiVec types is invalid");
else if (!TARGET_VSX)
{
if (type == long_unsigned_type_node || type == long_integer_type_node)
{
if (TARGET_64BIT)
error ("use of %<long%> in AltiVec types is invalid for "
"64-bit code without %qs", "-mvsx");
else if (rs6000_warn_altivec_long)
warning (0, "use of %<long%> in AltiVec types is deprecated; "
"use %<int%>");
}
else if (type == long_long_unsigned_type_node
|| type == long_long_integer_type_node)
error ("use of %<long long%> in AltiVec types is invalid without %qs",
"-mvsx");
else if (type == double_type_node)
error ("use of %<double%> in AltiVec types is invalid without %qs",
"-mvsx");
}
switch (altivec_type)
{
case 'v':
unsigned_p = TYPE_UNSIGNED (type);
switch (mode)
{
case E_TImode:
result = (unsigned_p ? unsigned_V1TI_type_node : V1TI_type_node);
break;
case E_DImode:
result = (unsigned_p ? unsigned_V2DI_type_node : V2DI_type_node);
break;
case E_SImode:
result = (unsigned_p ? unsigned_V4SI_type_node : V4SI_type_node);
break;
case E_HImode:
result = (unsigned_p ? unsigned_V8HI_type_node : V8HI_type_node);
break;
case E_QImode:
result = (unsigned_p ? unsigned_V16QI_type_node : V16QI_type_node);
break;
case E_SFmode: result = V4SF_type_node; break;
case E_DFmode: result = V2DF_type_node; break;
/* If the user says 'vector int bool', we may be handed the 'bool'
attribute _before_ the 'vector' attribute, and so select the
proper type in the 'b' case below. */
case E_V4SImode: case E_V8HImode: case E_V16QImode: case E_V4SFmode:
case E_V2DImode: case E_V2DFmode:
result = type;
default: break;
}
break;
case 'b':
switch (mode)
{
case E_DImode: case E_V2DImode: result = bool_V2DI_type_node; break;
case E_SImode: case E_V4SImode: result = bool_V4SI_type_node; break;
case E_HImode: case E_V8HImode: result = bool_V8HI_type_node; break;
case E_QImode: case E_V16QImode: result = bool_V16QI_type_node;
default: break;
}
break;
case 'p':
switch (mode)
{
case E_V8HImode: result = pixel_V8HI_type_node;
default: break;
}
default: break;
}
/* Propagate qualifiers attached to the element type
onto the vector type. */
if (result && result != type && TYPE_QUALS (type))
result = build_qualified_type (result, TYPE_QUALS (type));
*no_add_attrs = true; /* No need to hang on to the attribute. */
if (result)
*node = lang_hooks.types.reconstruct_complex_type (*node, result);
return NULL_TREE;
}
/* AltiVec defines four built-in scalar types that serve as vector
elements; we must teach the compiler how to mangle them. */
static const char *
rs6000_mangle_type (const_tree type)
{
type = TYPE_MAIN_VARIANT (type);
if (TREE_CODE (type) != VOID_TYPE && TREE_CODE (type) != BOOLEAN_TYPE
&& TREE_CODE (type) != INTEGER_TYPE && TREE_CODE (type) != REAL_TYPE)
return NULL;
if (type == bool_char_type_node) return "U6__boolc";
if (type == bool_short_type_node) return "U6__bools";
if (type == pixel_type_node) return "u7__pixel";
if (type == bool_int_type_node) return "U6__booli";
if (type == bool_long_type_node) return "U6__booll";
/* Use a unique name for __float128 rather than trying to use "e" or "g". Use
"g" for IBM extended double, no matter whether it is long double (using
-mabi=ibmlongdouble) or the distinct __ibm128 type. */
if (TARGET_FLOAT128_TYPE)
{
if (type == ieee128_float_type_node)
return "U10__float128";
if (TARGET_LONG_DOUBLE_128)
{
if (type == long_double_type_node)
return (TARGET_IEEEQUAD) ? "U10__float128" : "g";
if (type == ibm128_float_type_node)
return "g";
}
}
/* Mangle IBM extended float long double as `g' (__float128) on
powerpc*-linux where long-double-64 previously was the default. */
if (TYPE_MAIN_VARIANT (type) == long_double_type_node
&& TARGET_ELF
&& TARGET_LONG_DOUBLE_128
&& !TARGET_IEEEQUAD)
return "g";
/* For all other types, use normal C++ mangling. */
return NULL;
}
/* Handle a "longcall" or "shortcall" attribute; arguments as in
struct attribute_spec.handler. */
static tree
rs6000_handle_longcall_attribute (tree *node, tree name,
tree args ATTRIBUTE_UNUSED,
int flags ATTRIBUTE_UNUSED,
bool *no_add_attrs)
{
if (TREE_CODE (*node) != FUNCTION_TYPE
&& TREE_CODE (*node) != FIELD_DECL
&& TREE_CODE (*node) != TYPE_DECL)
{
warning (OPT_Wattributes, "%qE attribute only applies to functions",
name);
*no_add_attrs = true;
}
return NULL_TREE;
}
/* Set longcall attributes on all functions declared when
rs6000_default_long_calls is true. */
static void
rs6000_set_default_type_attributes (tree type)
{
if (rs6000_default_long_calls
&& (TREE_CODE (type) == FUNCTION_TYPE
|| TREE_CODE (type) == METHOD_TYPE))
TYPE_ATTRIBUTES (type) = tree_cons (get_identifier ("longcall"),
NULL_TREE,
TYPE_ATTRIBUTES (type));
#if TARGET_MACHO
darwin_set_default_type_attributes (type);
#endif
}
/* Return a reference suitable for calling a function with the
longcall attribute. */
rtx
rs6000_longcall_ref (rtx call_ref)
{
const char *call_name;
tree node;
if (GET_CODE (call_ref) != SYMBOL_REF)
return call_ref;
/* System V adds '.' to the internal name, so skip them. */
call_name = XSTR (call_ref, 0);
if (*call_name == '.')
{
while (*call_name == '.')
call_name++;
node = get_identifier (call_name);
call_ref = gen_rtx_SYMBOL_REF (VOIDmode, IDENTIFIER_POINTER (node));
}
return force_reg (Pmode, call_ref);
}
#ifndef TARGET_USE_MS_BITFIELD_LAYOUT
#define TARGET_USE_MS_BITFIELD_LAYOUT 0
#endif
/* Handle a "ms_struct" or "gcc_struct" attribute; arguments as in
struct attribute_spec.handler. */
static tree
rs6000_handle_struct_attribute (tree *node, tree name,
tree args ATTRIBUTE_UNUSED,
int flags ATTRIBUTE_UNUSED, bool *no_add_attrs)
{
tree *type = NULL;
if (DECL_P (*node))
{
if (TREE_CODE (*node) == TYPE_DECL)
type = &TREE_TYPE (*node);
}
else
type = node;
if (!(type && (TREE_CODE (*type) == RECORD_TYPE
|| TREE_CODE (*type) == UNION_TYPE)))
{
warning (OPT_Wattributes, "%qE attribute ignored", name);
*no_add_attrs = true;
}
else if ((is_attribute_p ("ms_struct", name)
&& lookup_attribute ("gcc_struct", TYPE_ATTRIBUTES (*type)))
|| ((is_attribute_p ("gcc_struct", name)
&& lookup_attribute ("ms_struct", TYPE_ATTRIBUTES (*type)))))
{
warning (OPT_Wattributes, "%qE incompatible attribute ignored",
name);
*no_add_attrs = true;
}
return NULL_TREE;
}
static bool
rs6000_ms_bitfield_layout_p (const_tree record_type)
{
return (TARGET_USE_MS_BITFIELD_LAYOUT &&
!lookup_attribute ("gcc_struct", TYPE_ATTRIBUTES (record_type)))
|| lookup_attribute ("ms_struct", TYPE_ATTRIBUTES (record_type));
}
#ifdef USING_ELFOS_H
/* A get_unnamed_section callback, used for switching to toc_section. */
static void
rs6000_elf_output_toc_section_asm_op (const void *data ATTRIBUTE_UNUSED)
{
if ((DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
&& TARGET_MINIMAL_TOC)
{
if (!toc_initialized)
{
fprintf (asm_out_file, "%s\n", TOC_SECTION_ASM_OP);
ASM_OUTPUT_ALIGN (asm_out_file, TARGET_64BIT ? 3 : 2);
(*targetm.asm_out.internal_label) (asm_out_file, "LCTOC", 0);
fprintf (asm_out_file, "\t.tc ");
ASM_OUTPUT_INTERNAL_LABEL_PREFIX (asm_out_file, "LCTOC1[TC],");
ASM_OUTPUT_INTERNAL_LABEL_PREFIX (asm_out_file, "LCTOC1");
fprintf (asm_out_file, "\n");
fprintf (asm_out_file, "%s\n", MINIMAL_TOC_SECTION_ASM_OP);
ASM_OUTPUT_ALIGN (asm_out_file, TARGET_64BIT ? 3 : 2);
ASM_OUTPUT_INTERNAL_LABEL_PREFIX (asm_out_file, "LCTOC1");
fprintf (asm_out_file, " = .+32768\n");
toc_initialized = 1;
}
else
fprintf (asm_out_file, "%s\n", MINIMAL_TOC_SECTION_ASM_OP);
}
else if (DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
{
fprintf (asm_out_file, "%s\n", TOC_SECTION_ASM_OP);
if (!toc_initialized)
{
ASM_OUTPUT_ALIGN (asm_out_file, TARGET_64BIT ? 3 : 2);
toc_initialized = 1;
}
}
else
{
fprintf (asm_out_file, "%s\n", MINIMAL_TOC_SECTION_ASM_OP);
if (!toc_initialized)
{
ASM_OUTPUT_ALIGN (asm_out_file, TARGET_64BIT ? 3 : 2);
ASM_OUTPUT_INTERNAL_LABEL_PREFIX (asm_out_file, "LCTOC1");
fprintf (asm_out_file, " = .+32768\n");
toc_initialized = 1;
}
}
}
/* Implement TARGET_ASM_INIT_SECTIONS. */
static void
rs6000_elf_asm_init_sections (void)
{
toc_section
= get_unnamed_section (0, rs6000_elf_output_toc_section_asm_op, NULL);
sdata2_section
= get_unnamed_section (SECTION_WRITE, output_section_asm_op,
SDATA2_SECTION_ASM_OP);
}
/* Implement TARGET_SELECT_RTX_SECTION. */
static section *
rs6000_elf_select_rtx_section (machine_mode mode, rtx x,
unsigned HOST_WIDE_INT align)
{
if (ASM_OUTPUT_SPECIAL_POOL_ENTRY_P (x, mode))
return toc_section;
else
return default_elf_select_rtx_section (mode, x, align);
}
/* For a SYMBOL_REF, set generic flags and then perform some
target-specific processing.
When the AIX ABI is requested on a non-AIX system, replace the
function name with the real name (with a leading .) rather than the
function descriptor name. This saves a lot of overriding code to
read the prefixes. */
static void rs6000_elf_encode_section_info (tree, rtx, int) ATTRIBUTE_UNUSED;
static void
rs6000_elf_encode_section_info (tree decl, rtx rtl, int first)
{
default_encode_section_info (decl, rtl, first);
if (first
&& TREE_CODE (decl) == FUNCTION_DECL
&& !TARGET_AIX
&& DEFAULT_ABI == ABI_AIX)
{
rtx sym_ref = XEXP (rtl, 0);
size_t len = strlen (XSTR (sym_ref, 0));
char *str = XALLOCAVEC (char, len + 2);
str[0] = '.';
memcpy (str + 1, XSTR (sym_ref, 0), len + 1);
XSTR (sym_ref, 0) = ggc_alloc_string (str, len + 1);
}
}
static inline bool
compare_section_name (const char *section, const char *templ)
{
int len;
len = strlen (templ);
return (strncmp (section, templ, len) == 0
&& (section[len] == 0 || section[len] == '.'));
}
bool
rs6000_elf_in_small_data_p (const_tree decl)
{
if (rs6000_sdata == SDATA_NONE)
return false;
/* We want to merge strings, so we never consider them small data. */
if (TREE_CODE (decl) == STRING_CST)
return false;
/* Functions are never in the small data area. */
if (TREE_CODE (decl) == FUNCTION_DECL)
return false;
if (TREE_CODE (decl) == VAR_DECL && DECL_SECTION_NAME (decl))
{
const char *section = DECL_SECTION_NAME (decl);
if (compare_section_name (section, ".sdata")
|| compare_section_name (section, ".sdata2")
|| compare_section_name (section, ".gnu.linkonce.s")
|| compare_section_name (section, ".sbss")
|| compare_section_name (section, ".sbss2")
|| compare_section_name (section, ".gnu.linkonce.sb")
|| strcmp (section, ".PPC.EMB.sdata0") == 0
|| strcmp (section, ".PPC.EMB.sbss0") == 0)
return true;
}
else
{
HOST_WIDE_INT size = int_size_in_bytes (TREE_TYPE (decl));
if (size > 0
&& size <= g_switch_value
/* If it's not public, and we're not going to reference it there,
there's no need to put it in the small data section. */
&& (rs6000_sdata != SDATA_DATA || TREE_PUBLIC (decl)))
return true;
}
return false;
}
#endif /* USING_ELFOS_H */
/* Implement TARGET_USE_BLOCKS_FOR_CONSTANT_P. */
static bool
rs6000_use_blocks_for_constant_p (machine_mode mode, const_rtx x)
{
return !ASM_OUTPUT_SPECIAL_POOL_ENTRY_P (x, mode);
}
/* Do not place thread-local symbols refs in the object blocks. */
static bool
rs6000_use_blocks_for_decl_p (const_tree decl)
{
return !DECL_THREAD_LOCAL_P (decl);
}
/* Return a REG that occurs in ADDR with coefficient 1.
ADDR can be effectively incremented by incrementing REG.
r0 is special and we must not select it as an address
register by this routine since our caller will try to
increment the returned register via an "la" instruction. */
rtx
find_addr_reg (rtx addr)
{
while (GET_CODE (addr) == PLUS)
{
if (GET_CODE (XEXP (addr, 0)) == REG
&& REGNO (XEXP (addr, 0)) != 0)
addr = XEXP (addr, 0);
else if (GET_CODE (XEXP (addr, 1)) == REG
&& REGNO (XEXP (addr, 1)) != 0)
addr = XEXP (addr, 1);
else if (CONSTANT_P (XEXP (addr, 0)))
addr = XEXP (addr, 1);
else if (CONSTANT_P (XEXP (addr, 1)))
addr = XEXP (addr, 0);
else
gcc_unreachable ();
}
gcc_assert (GET_CODE (addr) == REG && REGNO (addr) != 0);
return addr;
}
void
rs6000_fatal_bad_address (rtx op)
{
fatal_insn ("bad address", op);
}
#if TARGET_MACHO
typedef struct branch_island_d {
tree function_name;
tree label_name;
int line_number;
} branch_island;
static vec<branch_island, va_gc> *branch_islands;
/* Remember to generate a branch island for far calls to the given
function. */
static void
add_compiler_branch_island (tree label_name, tree function_name,
int line_number)
{
branch_island bi = {function_name, label_name, line_number};
vec_safe_push (branch_islands, bi);
}
/* Generate far-jump branch islands for everything recorded in
branch_islands. Invoked immediately after the last instruction of
the epilogue has been emitted; the branch islands must be appended
to, and contiguous with, the function body. Mach-O stubs are
generated in machopic_output_stub(). */
static void
macho_branch_islands (void)
{
char tmp_buf[512];
while (!vec_safe_is_empty (branch_islands))
{
branch_island *bi = &branch_islands->last ();
const char *label = IDENTIFIER_POINTER (bi->label_name);
const char *name = IDENTIFIER_POINTER (bi->function_name);
char name_buf[512];
/* Cheap copy of the details from the Darwin ASM_OUTPUT_LABELREF(). */
if (name[0] == '*' || name[0] == '&')
strcpy (name_buf, name+1);
else
{
name_buf[0] = '_';
strcpy (name_buf+1, name);
}
strcpy (tmp_buf, "\n");
strcat (tmp_buf, label);
#if defined (DBX_DEBUGGING_INFO) || defined (XCOFF_DEBUGGING_INFO)
if (write_symbols == DBX_DEBUG || write_symbols == XCOFF_DEBUG)
dbxout_stabd (N_SLINE, bi->line_number);
#endif /* DBX_DEBUGGING_INFO || XCOFF_DEBUGGING_INFO */
if (flag_pic)
{
if (TARGET_LINK_STACK)
{
char name[32];
get_ppc476_thunk_name (name);
strcat (tmp_buf, ":\n\tmflr r0\n\tbl ");
strcat (tmp_buf, name);
strcat (tmp_buf, "\n");
strcat (tmp_buf, label);
strcat (tmp_buf, "_pic:\n\tmflr r11\n");
}
else
{
strcat (tmp_buf, ":\n\tmflr r0\n\tbcl 20,31,");
strcat (tmp_buf, label);
strcat (tmp_buf, "_pic\n");
strcat (tmp_buf, label);
strcat (tmp_buf, "_pic:\n\tmflr r11\n");
}
strcat (tmp_buf, "\taddis r11,r11,ha16(");
strcat (tmp_buf, name_buf);
strcat (tmp_buf, " - ");
strcat (tmp_buf, label);
strcat (tmp_buf, "_pic)\n");
strcat (tmp_buf, "\tmtlr r0\n");
strcat (tmp_buf, "\taddi r12,r11,lo16(");
strcat (tmp_buf, name_buf);
strcat (tmp_buf, " - ");
strcat (tmp_buf, label);
strcat (tmp_buf, "_pic)\n");
strcat (tmp_buf, "\tmtctr r12\n\tbctr\n");
}
else
{
strcat (tmp_buf, ":\nlis r12,hi16(");
strcat (tmp_buf, name_buf);
strcat (tmp_buf, ")\n\tori r12,r12,lo16(");
strcat (tmp_buf, name_buf);
strcat (tmp_buf, ")\n\tmtctr r12\n\tbctr");
}
output_asm_insn (tmp_buf, 0);
#if defined (DBX_DEBUGGING_INFO) || defined (XCOFF_DEBUGGING_INFO)
if (write_symbols == DBX_DEBUG || write_symbols == XCOFF_DEBUG)
dbxout_stabd (N_SLINE, bi->line_number);
#endif /* DBX_DEBUGGING_INFO || XCOFF_DEBUGGING_INFO */
branch_islands->pop ();
}
}
/* NO_PREVIOUS_DEF checks in the link list whether the function name is
already there or not. */
static int
no_previous_def (tree function_name)
{
branch_island *bi;
unsigned ix;
FOR_EACH_VEC_SAFE_ELT (branch_islands, ix, bi)
if (function_name == bi->function_name)
return 0;
return 1;
}
/* GET_PREV_LABEL gets the label name from the previous definition of
the function. */
static tree
get_prev_label (tree function_name)
{
branch_island *bi;
unsigned ix;
FOR_EACH_VEC_SAFE_ELT (branch_islands, ix, bi)
if (function_name == bi->function_name)
return bi->label_name;
return NULL_TREE;
}
/* INSN is either a function call or a millicode call. It may have an
unconditional jump in its delay slot.
CALL_DEST is the routine we are calling. */
char *
output_call (rtx_insn *insn, rtx *operands, int dest_operand_number,
int cookie_operand_number)
{
static char buf[256];
if (darwin_emit_branch_islands
&& GET_CODE (operands[dest_operand_number]) == SYMBOL_REF
&& (INTVAL (operands[cookie_operand_number]) & CALL_LONG))
{
tree labelname;
tree funname = get_identifier (XSTR (operands[dest_operand_number], 0));
if (no_previous_def (funname))
{
rtx label_rtx = gen_label_rtx ();
char *label_buf, temp_buf[256];
ASM_GENERATE_INTERNAL_LABEL (temp_buf, "L",
CODE_LABEL_NUMBER (label_rtx));
label_buf = temp_buf[0] == '*' ? temp_buf + 1 : temp_buf;
labelname = get_identifier (label_buf);
add_compiler_branch_island (labelname, funname, insn_line (insn));
}
else
labelname = get_prev_label (funname);
/* "jbsr foo, L42" is Mach-O for "Link as 'bl foo' if a 'bl'
instruction will reach 'foo', otherwise link as 'bl L42'".
"L42" should be a 'branch island', that will do a far jump to
'foo'. Branch islands are generated in
macho_branch_islands(). */
sprintf (buf, "jbsr %%z%d,%.246s",
dest_operand_number, IDENTIFIER_POINTER (labelname));
}
else
sprintf (buf, "bl %%z%d", dest_operand_number);
return buf;
}
/* Generate PIC and indirect symbol stubs. */
void
machopic_output_stub (FILE *file, const char *symb, const char *stub)
{
unsigned int length;
char *symbol_name, *lazy_ptr_name;
char *local_label_0;
static int label = 0;
/* Lose our funky encoding stuff so it doesn't contaminate the stub. */
symb = (*targetm.strip_name_encoding) (symb);
length = strlen (symb);
symbol_name = XALLOCAVEC (char, length + 32);
GEN_SYMBOL_NAME_FOR_SYMBOL (symbol_name, symb, length);
lazy_ptr_name = XALLOCAVEC (char, length + 32);
GEN_LAZY_PTR_NAME_FOR_SYMBOL (lazy_ptr_name, symb, length);
if (flag_pic == 2)
switch_to_section (darwin_sections[machopic_picsymbol_stub1_section]);
else
switch_to_section (darwin_sections[machopic_symbol_stub1_section]);
if (flag_pic == 2)
{
fprintf (file, "\t.align 5\n");
fprintf (file, "%s:\n", stub);
fprintf (file, "\t.indirect_symbol %s\n", symbol_name);
label++;
local_label_0 = XALLOCAVEC (char, sizeof ("\"L00000000000$spb\""));
sprintf (local_label_0, "\"L%011d$spb\"", label);
fprintf (file, "\tmflr r0\n");
if (TARGET_LINK_STACK)
{
char name[32];
get_ppc476_thunk_name (name);
fprintf (file, "\tbl %s\n", name);
fprintf (file, "%s:\n\tmflr r11\n", local_label_0);
}
else
{
fprintf (file, "\tbcl 20,31,%s\n", local_label_0);
fprintf (file, "%s:\n\tmflr r11\n", local_label_0);
}
fprintf (file, "\taddis r11,r11,ha16(%s-%s)\n",
lazy_ptr_name, local_label_0);
fprintf (file, "\tmtlr r0\n");
fprintf (file, "\t%s r12,lo16(%s-%s)(r11)\n",
(TARGET_64BIT ? "ldu" : "lwzu"),
lazy_ptr_name, local_label_0);
fprintf (file, "\tmtctr r12\n");
fprintf (file, "\tbctr\n");
}
else
{
fprintf (file, "\t.align 4\n");
fprintf (file, "%s:\n", stub);
fprintf (file, "\t.indirect_symbol %s\n", symbol_name);
fprintf (file, "\tlis r11,ha16(%s)\n", lazy_ptr_name);
fprintf (file, "\t%s r12,lo16(%s)(r11)\n",
(TARGET_64BIT ? "ldu" : "lwzu"),
lazy_ptr_name);
fprintf (file, "\tmtctr r12\n");
fprintf (file, "\tbctr\n");
}
switch_to_section (darwin_sections[machopic_lazy_symbol_ptr_section]);
fprintf (file, "%s:\n", lazy_ptr_name);
fprintf (file, "\t.indirect_symbol %s\n", symbol_name);
fprintf (file, "%sdyld_stub_binding_helper\n",
(TARGET_64BIT ? DOUBLE_INT_ASM_OP : "\t.long\t"));
}
/* Legitimize PIC addresses. If the address is already
position-independent, we return ORIG. Newly generated
position-independent addresses go into a reg. This is REG if non
zero, otherwise we allocate register(s) as necessary. */
#define SMALL_INT(X) ((UINTVAL (X) + 0x8000) < 0x10000)
rtx
rs6000_machopic_legitimize_pic_address (rtx orig, machine_mode mode,
rtx reg)
{
rtx base, offset;
if (reg == NULL && !reload_completed)
reg = gen_reg_rtx (Pmode);
if (GET_CODE (orig) == CONST)
{
rtx reg_temp;
if (GET_CODE (XEXP (orig, 0)) == PLUS
&& XEXP (XEXP (orig, 0), 0) == pic_offset_table_rtx)
return orig;
gcc_assert (GET_CODE (XEXP (orig, 0)) == PLUS);
/* Use a different reg for the intermediate value, as
it will be marked UNCHANGING. */
reg_temp = !can_create_pseudo_p () ? reg : gen_reg_rtx (Pmode);
base = rs6000_machopic_legitimize_pic_address (XEXP (XEXP (orig, 0), 0),
Pmode, reg_temp);
offset =
rs6000_machopic_legitimize_pic_address (XEXP (XEXP (orig, 0), 1),
Pmode, reg);
if (GET_CODE (offset) == CONST_INT)
{
if (SMALL_INT (offset))
return plus_constant (Pmode, base, INTVAL (offset));
else if (!reload_completed)
offset = force_reg (Pmode, offset);
else
{
rtx mem = force_const_mem (Pmode, orig);
return machopic_legitimize_pic_address (mem, Pmode, reg);
}
}
return gen_rtx_PLUS (Pmode, base, offset);
}
/* Fall back on generic machopic code. */
return machopic_legitimize_pic_address (orig, mode, reg);
}
/* Output a .machine directive for the Darwin assembler, and call
the generic start_file routine. */
static void
rs6000_darwin_file_start (void)
{
static const struct
{
const char *arg;
const char *name;
HOST_WIDE_INT if_set;
} mapping[] = {
{ "ppc64", "ppc64", MASK_64BIT },
{ "970", "ppc970", MASK_PPC_GPOPT | MASK_MFCRF | MASK_POWERPC64 },
{ "power4", "ppc970", 0 },
{ "G5", "ppc970", 0 },
{ "7450", "ppc7450", 0 },
{ "7400", "ppc7400", MASK_ALTIVEC },
{ "G4", "ppc7400", 0 },
{ "750", "ppc750", 0 },
{ "740", "ppc750", 0 },
{ "G3", "ppc750", 0 },
{ "604e", "ppc604e", 0 },
{ "604", "ppc604", 0 },
{ "603e", "ppc603", 0 },
{ "603", "ppc603", 0 },
{ "601", "ppc601", 0 },
{ NULL, "ppc", 0 } };
const char *cpu_id = "";
size_t i;
rs6000_file_start ();
darwin_file_start ();
/* Determine the argument to -mcpu=. Default to G3 if not specified. */
if (rs6000_default_cpu != 0 && rs6000_default_cpu[0] != '\0')
cpu_id = rs6000_default_cpu;
if (global_options_set.x_rs6000_cpu_index)
cpu_id = processor_target_table[rs6000_cpu_index].name;
/* Look through the mapping array. Pick the first name that either
matches the argument, has a bit set in IF_SET that is also set
in the target flags, or has a NULL name. */
i = 0;
while (mapping[i].arg != NULL
&& strcmp (mapping[i].arg, cpu_id) != 0
&& (mapping[i].if_set & rs6000_isa_flags) == 0)
i++;
fprintf (asm_out_file, "\t.machine %s\n", mapping[i].name);
}
#endif /* TARGET_MACHO */
#if TARGET_ELF
static int
rs6000_elf_reloc_rw_mask (void)
{
if (flag_pic)
return 3;
else if (DEFAULT_ABI == ABI_AIX || DEFAULT_ABI == ABI_ELFv2)
return 2;
else
return 0;
}
/* Record an element in the table of global constructors. SYMBOL is
a SYMBOL_REF of the function to be called; PRIORITY is a number
between 0 and MAX_INIT_PRIORITY.
This differs from default_named_section_asm_out_constructor in
that we have special handling for -mrelocatable. */
static void rs6000_elf_asm_out_constructor (rtx, int) ATTRIBUTE_UNUSED;
static void
rs6000_elf_asm_out_constructor (rtx symbol, int priority)
{
const char *section = ".ctors";
char buf[18];
if (priority != DEFAULT_INIT_PRIORITY)
{
sprintf (buf, ".ctors.%.5u",
/* Invert the numbering so the linker puts us in the proper
order; constructors are run from right to left, and the
linker sorts in increasing order. */
MAX_INIT_PRIORITY - priority);
section = buf;
}
switch_to_section (get_section (section, SECTION_WRITE, NULL));
assemble_align (POINTER_SIZE);
if (DEFAULT_ABI == ABI_V4
&& (TARGET_RELOCATABLE || flag_pic > 1))
{
fputs ("\t.long (", asm_out_file);
output_addr_const (asm_out_file, symbol);
fputs (")@fixup\n", asm_out_file);
}
else
assemble_integer (symbol, POINTER_SIZE / BITS_PER_UNIT, POINTER_SIZE, 1);
}
static void rs6000_elf_asm_out_destructor (rtx, int) ATTRIBUTE_UNUSED;
static void
rs6000_elf_asm_out_destructor (rtx symbol, int priority)
{
const char *section = ".dtors";
char buf[18];
if (priority != DEFAULT_INIT_PRIORITY)
{
sprintf (buf, ".dtors.%.5u",
/* Invert the numbering so the linker puts us in the proper
order; constructors are run from right to left, and the
linker sorts in increasing order. */
MAX_INIT_PRIORITY - priority);
section = buf;
}
switch_to_section (get_section (section, SECTION_WRITE, NULL));
assemble_align (POINTER_SIZE);
if (DEFAULT_ABI == ABI_V4
&& (TARGET_RELOCATABLE || flag_pic > 1))
{
fputs ("\t.long (", asm_out_file);
output_addr_const (asm_out_file, symbol);
fputs (")@fixup\n", asm_out_file);
}
else
assemble_integer (symbol, POINTER_SIZE / BITS_PER_UNIT, POINTER_SIZE, 1);
}
void
rs6000_elf_declare_function_name (FILE *file, const char *name, tree decl)
{
if (TARGET_64BIT && DEFAULT_ABI != ABI_ELFv2)
{
fputs ("\t.section\t\".opd\",\"aw\"\n\t.align 3\n", file);
ASM_OUTPUT_LABEL (file, name);
fputs (DOUBLE_INT_ASM_OP, file);
rs6000_output_function_entry (file, name);
fputs (",.TOC.@tocbase,0\n\t.previous\n", file);
if (DOT_SYMBOLS)
{
fputs ("\t.size\t", file);
assemble_name (file, name);
fputs (",24\n\t.type\t.", file);
assemble_name (file, name);
fputs (",@function\n", file);
if (TREE_PUBLIC (decl) && ! DECL_WEAK (decl))
{
fputs ("\t.globl\t.", file);
assemble_name (file, name);
putc ('\n', file);
}
}
else
ASM_OUTPUT_TYPE_DIRECTIVE (file, name, "function");
ASM_DECLARE_RESULT (file, DECL_RESULT (decl));
rs6000_output_function_entry (file, name);
fputs (":\n", file);
return;
}
int uses_toc;
if (DEFAULT_ABI == ABI_V4
&& (TARGET_RELOCATABLE || flag_pic > 1)
&& !TARGET_SECURE_PLT
&& (!constant_pool_empty_p () || crtl->profile)
&& (uses_toc = uses_TOC ()))
{
char buf[256];
if (uses_toc == 2)
switch_to_other_text_partition ();
(*targetm.asm_out.internal_label) (file, "LCL", rs6000_pic_labelno);
fprintf (file, "\t.long ");
assemble_name (file, toc_label_name);
need_toc_init = 1;
putc ('-', file);
ASM_GENERATE_INTERNAL_LABEL (buf, "LCF", rs6000_pic_labelno);
assemble_name (file, buf);
putc ('\n', file);
if (uses_toc == 2)
switch_to_other_text_partition ();
}
ASM_OUTPUT_TYPE_DIRECTIVE (file, name, "function");
ASM_DECLARE_RESULT (file, DECL_RESULT (decl));
if (TARGET_CMODEL == CMODEL_LARGE && rs6000_global_entry_point_needed_p ())
{
char buf[256];
(*targetm.asm_out.internal_label) (file, "LCL", rs6000_pic_labelno);
fprintf (file, "\t.quad .TOC.-");
ASM_GENERATE_INTERNAL_LABEL (buf, "LCF", rs6000_pic_labelno);
assemble_name (file, buf);
putc ('\n', file);
}
if (DEFAULT_ABI == ABI_AIX)
{
const char *desc_name, *orig_name;
orig_name = (*targetm.strip_name_encoding) (name);
desc_name = orig_name;
while (*desc_name == '.')
desc_name++;
if (TREE_PUBLIC (decl))
fprintf (file, "\t.globl %s\n", desc_name);
fprintf (file, "%s\n", MINIMAL_TOC_SECTION_ASM_OP);
fprintf (file, "%s:\n", desc_name);
fprintf (file, "\t.long %s\n", orig_name);
fputs ("\t.long _GLOBAL_OFFSET_TABLE_\n", file);
fputs ("\t.long 0\n", file);
fprintf (file, "\t.previous\n");
}
ASM_OUTPUT_LABEL (file, name);
}
static void rs6000_elf_file_end (void) ATTRIBUTE_UNUSED;
static void
rs6000_elf_file_end (void)
{
#ifdef HAVE_AS_GNU_ATTRIBUTE
/* ??? The value emitted depends on options active at file end.
Assume anyone using #pragma or attributes that might change
options knows what they are doing. */
if ((TARGET_64BIT || DEFAULT_ABI == ABI_V4)
&& rs6000_passes_float)
{
int fp;
if (TARGET_DF_FPR)
fp = 1;
else if (TARGET_SF_FPR)
fp = 3;
else
fp = 2;
if (rs6000_passes_long_double)
{
if (!TARGET_LONG_DOUBLE_128)
fp |= 2 * 4;
else if (TARGET_IEEEQUAD)
fp |= 3 * 4;
else
fp |= 1 * 4;
}
fprintf (asm_out_file, "\t.gnu_attribute 4, %d\n", fp);
}
if (TARGET_32BIT && DEFAULT_ABI == ABI_V4)
{
if (rs6000_passes_vector)
fprintf (asm_out_file, "\t.gnu_attribute 8, %d\n",
(TARGET_ALTIVEC_ABI ? 2 : 1));
if (rs6000_returns_struct)
fprintf (asm_out_file, "\t.gnu_attribute 12, %d\n",
aix_struct_return ? 2 : 1);
}
#endif
#if defined (POWERPC_LINUX) || defined (POWERPC_FREEBSD)
if (TARGET_32BIT || DEFAULT_ABI == ABI_ELFv2)
file_end_indicate_exec_stack ();
#endif
if (flag_split_stack)
file_end_indicate_split_stack ();
if (cpu_builtin_p)
{
/* We have expanded a CPU builtin, so we need to emit a reference to
the special symbol that LIBC uses to declare it supports the
AT_PLATFORM and AT_HWCAP/AT_HWCAP2 in the TCB feature. */
switch_to_section (data_section);
fprintf (asm_out_file, "\t.align %u\n", TARGET_32BIT ? 2 : 3);
fprintf (asm_out_file, "\t%s %s\n",
TARGET_32BIT ? ".long" : ".quad", tcb_verification_symbol);
}
}
#endif
#if TARGET_XCOFF
#ifndef HAVE_XCOFF_DWARF_EXTRAS
#define HAVE_XCOFF_DWARF_EXTRAS 0
#endif
static enum unwind_info_type
rs6000_xcoff_debug_unwind_info (void)
{
return UI_NONE;
}
static void
rs6000_xcoff_asm_output_anchor (rtx symbol)
{
char buffer[100];
sprintf (buffer, "$ + " HOST_WIDE_INT_PRINT_DEC,
SYMBOL_REF_BLOCK_OFFSET (symbol));
fprintf (asm_out_file, "%s", SET_ASM_OP);
RS6000_OUTPUT_BASENAME (asm_out_file, XSTR (symbol, 0));
fprintf (asm_out_file, ",");
RS6000_OUTPUT_BASENAME (asm_out_file, buffer);
fprintf (asm_out_file, "\n");
}
static void
rs6000_xcoff_asm_globalize_label (FILE *stream, const char *name)
{
fputs (GLOBAL_ASM_OP, stream);
RS6000_OUTPUT_BASENAME (stream, name);
putc ('\n', stream);
}
/* A get_unnamed_decl callback, used for read-only sections. PTR
points to the section string variable. */
static void
rs6000_xcoff_output_readonly_section_asm_op (const void *directive)
{
fprintf (asm_out_file, "\t.csect %s[RO],%s\n",
*(const char *const *) directive,
XCOFF_CSECT_DEFAULT_ALIGNMENT_STR);
}
/* Likewise for read-write sections. */
static void
rs6000_xcoff_output_readwrite_section_asm_op (const void *directive)
{
fprintf (asm_out_file, "\t.csect %s[RW],%s\n",
*(const char *const *) directive,
XCOFF_CSECT_DEFAULT_ALIGNMENT_STR);
}
static void
rs6000_xcoff_output_tls_section_asm_op (const void *directive)
{
fprintf (asm_out_file, "\t.csect %s[TL],%s\n",
*(const char *const *) directive,
XCOFF_CSECT_DEFAULT_ALIGNMENT_STR);
}
/* A get_unnamed_section callback, used for switching to toc_section. */
static void
rs6000_xcoff_output_toc_section_asm_op (const void *data ATTRIBUTE_UNUSED)
{
if (TARGET_MINIMAL_TOC)
{
/* toc_section is always selected at least once from
rs6000_xcoff_file_start, so this is guaranteed to
always be defined once and only once in each file. */
if (!toc_initialized)
{
fputs ("\t.toc\nLCTOC..1:\n", asm_out_file);
fputs ("\t.tc toc_table[TC],toc_table[RW]\n", asm_out_file);
toc_initialized = 1;
}
fprintf (asm_out_file, "\t.csect toc_table[RW]%s\n",
(TARGET_32BIT ? "" : ",3"));
}
else
fputs ("\t.toc\n", asm_out_file);
}
/* Implement TARGET_ASM_INIT_SECTIONS. */
static void
rs6000_xcoff_asm_init_sections (void)
{
read_only_data_section
= get_unnamed_section (0, rs6000_xcoff_output_readonly_section_asm_op,
&xcoff_read_only_section_name);
private_data_section
= get_unnamed_section (SECTION_WRITE,
rs6000_xcoff_output_readwrite_section_asm_op,
&xcoff_private_data_section_name);
tls_data_section
= get_unnamed_section (SECTION_TLS,
rs6000_xcoff_output_tls_section_asm_op,
&xcoff_tls_data_section_name);
tls_private_data_section
= get_unnamed_section (SECTION_TLS,
rs6000_xcoff_output_tls_section_asm_op,
&xcoff_private_data_section_name);
read_only_private_data_section
= get_unnamed_section (0, rs6000_xcoff_output_readonly_section_asm_op,
&xcoff_private_data_section_name);
toc_section
= get_unnamed_section (0, rs6000_xcoff_output_toc_section_asm_op, NULL);
readonly_data_section = read_only_data_section;
}
static int
rs6000_xcoff_reloc_rw_mask (void)
{
return 3;
}
static void
rs6000_xcoff_asm_named_section (const char *name, unsigned int flags,
tree decl ATTRIBUTE_UNUSED)
{
int smclass;
static const char * const suffix[5] = { "PR", "RO", "RW", "TL", "XO" };
if (flags & SECTION_EXCLUDE)
smclass = 4;
else if (flags & SECTION_DEBUG)
{
fprintf (asm_out_file, "\t.dwsect %s\n", name);
return;
}
else if (flags & SECTION_CODE)
smclass = 0;
else if (flags & SECTION_TLS)
smclass = 3;
else if (flags & SECTION_WRITE)
smclass = 2;
else
smclass = 1;
fprintf (asm_out_file, "\t.csect %s%s[%s],%u\n",
(flags & SECTION_CODE) ? "." : "",
name, suffix[smclass], flags & SECTION_ENTSIZE);
}
#define IN_NAMED_SECTION(DECL) \
((TREE_CODE (DECL) == FUNCTION_DECL || TREE_CODE (DECL) == VAR_DECL) \
&& DECL_SECTION_NAME (DECL) != NULL)
static section *
rs6000_xcoff_select_section (tree decl, int reloc,
unsigned HOST_WIDE_INT align)
{
/* Place variables with alignment stricter than BIGGEST_ALIGNMENT into
named section. */
if (align > BIGGEST_ALIGNMENT)
{
resolve_unique_section (decl, reloc, true);
if (IN_NAMED_SECTION (decl))
return get_named_section (decl, NULL, reloc);
}
if (decl_readonly_section (decl, reloc))
{
if (TREE_PUBLIC (decl))
return read_only_data_section;
else
return read_only_private_data_section;
}
else
{
#if HAVE_AS_TLS
if (TREE_CODE (decl) == VAR_DECL && DECL_THREAD_LOCAL_P (decl))
{
if (TREE_PUBLIC (decl))
return tls_data_section;
else if (bss_initializer_p (decl))
{
/* Convert to COMMON to emit in BSS. */
DECL_COMMON (decl) = 1;
return tls_comm_section;
}
else
return tls_private_data_section;
}
else
#endif
if (TREE_PUBLIC (decl))
return data_section;
else
return private_data_section;
}
}
static void
rs6000_xcoff_unique_section (tree decl, int reloc ATTRIBUTE_UNUSED)
{
const char *name;
/* Use select_section for private data and uninitialized data with
alignment <= BIGGEST_ALIGNMENT. */
if (!TREE_PUBLIC (decl)
|| DECL_COMMON (decl)
|| (DECL_INITIAL (decl) == NULL_TREE
&& DECL_ALIGN (decl) <= BIGGEST_ALIGNMENT)
|| DECL_INITIAL (decl) == error_mark_node
|| (flag_zero_initialized_in_bss
&& initializer_zerop (DECL_INITIAL (decl))))
return;
name = IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (decl));
name = (*targetm.strip_name_encoding) (name);
set_decl_section_name (decl, name);
}
/* Select section for constant in constant pool.
On RS/6000, all constants are in the private read-only data area.
However, if this is being placed in the TOC it must be output as a
toc entry. */
static section *
rs6000_xcoff_select_rtx_section (machine_mode mode, rtx x,
unsigned HOST_WIDE_INT align ATTRIBUTE_UNUSED)
{
if (ASM_OUTPUT_SPECIAL_POOL_ENTRY_P (x, mode))
return toc_section;
else
return read_only_private_data_section;
}
/* Remove any trailing [DS] or the like from the symbol name. */
static const char *
rs6000_xcoff_strip_name_encoding (const char *name)
{
size_t len;
if (*name == '*')
name++;
len = strlen (name);
if (name[len - 1] == ']')
return ggc_alloc_string (name, len - 4);
else
return name;
}
/* Section attributes. AIX is always PIC. */
static unsigned int
rs6000_xcoff_section_type_flags (tree decl, const char *name, int reloc)
{
unsigned int align;
unsigned int flags = default_section_type_flags (decl, name, reloc);
/* Align to at least UNIT size. */
if ((flags & SECTION_CODE) != 0 || !decl || !DECL_P (decl))
align = MIN_UNITS_PER_WORD;
else
/* Increase alignment of large objects if not already stricter. */
align = MAX ((DECL_ALIGN (decl) / BITS_PER_UNIT),
int_size_in_bytes (TREE_TYPE (decl)) > MIN_UNITS_PER_WORD
? UNITS_PER_FP_WORD : MIN_UNITS_PER_WORD);
return flags | (exact_log2 (align) & SECTION_ENTSIZE);
}
/* Output at beginning of assembler file.
Initialize the section names for the RS/6000 at this point.
Specify filename, including full path, to assembler.
We want to go into the TOC section so at least one .toc will be emitted.
Also, in order to output proper .bs/.es pairs, we need at least one static
[RW] section emitted.
Finally, declare mcount when profiling to make the assembler happy. */
static void
rs6000_xcoff_file_start (void)
{
rs6000_gen_section_name (&xcoff_bss_section_name,
main_input_filename, ".bss_");
rs6000_gen_section_name (&xcoff_private_data_section_name,
main_input_filename, ".rw_");
rs6000_gen_section_name (&xcoff_read_only_section_name,
main_input_filename, ".ro_");
rs6000_gen_section_name (&xcoff_tls_data_section_name,
main_input_filename, ".tls_");
rs6000_gen_section_name (&xcoff_tbss_section_name,
main_input_filename, ".tbss_[UL]");
fputs ("\t.file\t", asm_out_file);
output_quoted_string (asm_out_file, main_input_filename);
fputc ('\n', asm_out_file);
if (write_symbols != NO_DEBUG)
switch_to_section (private_data_section);
switch_to_section (toc_section);
switch_to_section (text_section);
if (profile_flag)
fprintf (asm_out_file, "\t.extern %s\n", RS6000_MCOUNT);
rs6000_file_start ();
}
/* Output at end of assembler file.
On the RS/6000, referencing data should automatically pull in text. */
static void
rs6000_xcoff_file_end (void)
{
switch_to_section (text_section);
fputs ("_section_.text:\n", asm_out_file);
switch_to_section (data_section);
fputs (TARGET_32BIT
? "\t.long _section_.text\n" : "\t.llong _section_.text\n",
asm_out_file);
}
struct declare_alias_data
{
FILE *file;
bool function_descriptor;
};
/* Declare alias N. A helper function for for_node_and_aliases. */
static bool
rs6000_declare_alias (struct symtab_node *n, void *d)
{
struct declare_alias_data *data = (struct declare_alias_data *)d;
/* Main symbol is output specially, because varasm machinery does part of
the job for us - we do not need to declare .globl/lglobs and such. */
if (!n->alias || n->weakref)
return false;
if (lookup_attribute ("ifunc", DECL_ATTRIBUTES (n->decl)))
return false;
/* Prevent assemble_alias from trying to use .set pseudo operation
that does not behave as expected by the middle-end. */
TREE_ASM_WRITTEN (n->decl) = true;
const char *name = IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (n->decl));
char *buffer = (char *) alloca (strlen (name) + 2);
char *p;
int dollar_inside = 0;
strcpy (buffer, name);
p = strchr (buffer, '$');
while (p) {
*p = '_';
dollar_inside++;
p = strchr (p + 1, '$');
}
if (TREE_PUBLIC (n->decl))
{
if (!RS6000_WEAK || !DECL_WEAK (n->decl))
{
if (dollar_inside) {
if (data->function_descriptor)
fprintf(data->file, "\t.rename .%s,\".%s\"\n", buffer, name);
fprintf(data->file, "\t.rename %s,\"%s\"\n", buffer, name);
}
if (data->function_descriptor)
{
fputs ("\t.globl .", data->file);
RS6000_OUTPUT_BASENAME (data->file, buffer);
putc ('\n', data->file);
}
fputs ("\t.globl ", data->file);
RS6000_OUTPUT_BASENAME (data->file, buffer);
putc ('\n', data->file);
}
#ifdef ASM_WEAKEN_DECL
else if (DECL_WEAK (n->decl) && !data->function_descriptor)
ASM_WEAKEN_DECL (data->file, n->decl, name, NULL);
#endif
}
else
{
if (dollar_inside)
{
if (data->function_descriptor)
fprintf(data->file, "\t.rename .%s,\".%s\"\n", buffer, name);
fprintf(data->file, "\t.rename %s,\"%s\"\n", buffer, name);
}
if (data->function_descriptor)
{
fputs ("\t.lglobl .", data->file);
RS6000_OUTPUT_BASENAME (data->file, buffer);
putc ('\n', data->file);
}
fputs ("\t.lglobl ", data->file);
RS6000_OUTPUT_BASENAME (data->file, buffer);
putc ('\n', data->file);
}
if (data->function_descriptor)
fputs (".", data->file);
RS6000_OUTPUT_BASENAME (data->file, buffer);
fputs (":\n", data->file);
return false;
}
#ifdef HAVE_GAS_HIDDEN
/* Helper function to calculate visibility of a DECL
and return the value as a const string. */
static const char *
rs6000_xcoff_visibility (tree decl)
{
static const char * const visibility_types[] = {
"", ",protected", ",hidden", ",internal"
};
enum symbol_visibility vis = DECL_VISIBILITY (decl);
if (TREE_CODE (decl) == FUNCTION_DECL
&& cgraph_node::get (decl)
&& cgraph_node::get (decl)->instrumentation_clone
&& cgraph_node::get (decl)->instrumented_version)
vis = DECL_VISIBILITY (cgraph_node::get (decl)->instrumented_version->decl);
return visibility_types[vis];
}
#endif
/* This macro produces the initial definition of a function name.
On the RS/6000, we need to place an extra '.' in the function name and
output the function descriptor.
Dollar signs are converted to underscores.
The csect for the function will have already been created when
text_section was selected. We do have to go back to that csect, however.
The third and fourth parameters to the .function pseudo-op (16 and 044)
are placeholders which no longer have any use.
Because AIX assembler's .set command has unexpected semantics, we output
all aliases as alternative labels in front of the definition. */
void
rs6000_xcoff_declare_function_name (FILE *file, const char *name, tree decl)
{
char *buffer = (char *) alloca (strlen (name) + 1);
char *p;
int dollar_inside = 0;
struct declare_alias_data data = {file, false};
strcpy (buffer, name);
p = strchr (buffer, '$');
while (p) {
*p = '_';
dollar_inside++;
p = strchr (p + 1, '$');
}
if (TREE_PUBLIC (decl))
{
if (!RS6000_WEAK || !DECL_WEAK (decl))
{
if (dollar_inside) {
fprintf(file, "\t.rename .%s,\".%s\"\n", buffer, name);
fprintf(file, "\t.rename %s,\"%s\"\n", buffer, name);
}
fputs ("\t.globl .", file);
RS6000_OUTPUT_BASENAME (file, buffer);
#ifdef HAVE_GAS_HIDDEN
fputs (rs6000_xcoff_visibility (decl), file);
#endif
putc ('\n', file);
}
}
else
{
if (dollar_inside) {
fprintf(file, "\t.rename .%s,\".%s\"\n", buffer, name);
fprintf(file, "\t.rename %s,\"%s\"\n", buffer, name);
}
fputs ("\t.lglobl .", file);
RS6000_OUTPUT_BASENAME (file, buffer);
putc ('\n', file);
}
fputs ("\t.csect ", file);
RS6000_OUTPUT_BASENAME (file, buffer);
fputs (TARGET_32BIT ? "[DS]\n" : "[DS],3\n", file);
RS6000_OUTPUT_BASENAME (file, buffer);
fputs (":\n", file);
symtab_node::get (decl)->call_for_symbol_and_aliases (rs6000_declare_alias,
&data, true);
fputs (TARGET_32BIT ? "\t.long ." : "\t.llong .", file);
RS6000_OUTPUT_BASENAME (file, buffer);
fputs (", TOC[tc0], 0\n", file);
in_section = NULL;
switch_to_section (function_section (decl));
putc ('.', file);
RS6000_OUTPUT_BASENAME (file, buffer);
fputs (":\n", file);
data.function_descriptor = true;
symtab_node::get (decl)->call_for_symbol_and_aliases (rs6000_declare_alias,
&data, true);
if (!DECL_IGNORED_P (decl))
{
if (write_symbols == DBX_DEBUG || write_symbols == XCOFF_DEBUG)
xcoffout_declare_function (file, decl, buffer);
else if (write_symbols == DWARF2_DEBUG)
{
name = (*targetm.strip_name_encoding) (name);
fprintf (file, "\t.function .%s,.%s,2,0\n", name, name);
}
}
return;
}
/* Output assembly language to globalize a symbol from a DECL,
possibly with visibility. */
void
rs6000_xcoff_asm_globalize_decl_name (FILE *stream, tree decl)
{
const char *name = XSTR (XEXP (DECL_RTL (decl), 0), 0);
fputs (GLOBAL_ASM_OP, stream);
RS6000_OUTPUT_BASENAME (stream, name);
#ifdef HAVE_GAS_HIDDEN
fputs (rs6000_xcoff_visibility (decl), stream);
#endif
putc ('\n', stream);
}
/* Output assembly language to define a symbol as COMMON from a DECL,
possibly with visibility. */
void
rs6000_xcoff_asm_output_aligned_decl_common (FILE *stream,
tree decl ATTRIBUTE_UNUSED,
const char *name,
unsigned HOST_WIDE_INT size,
unsigned HOST_WIDE_INT align)
{
unsigned HOST_WIDE_INT align2 = 2;
if (align > 32)
align2 = floor_log2 (align / BITS_PER_UNIT);
else if (size > 4)
align2 = 3;
fputs (COMMON_ASM_OP, stream);
RS6000_OUTPUT_BASENAME (stream, name);
fprintf (stream,
"," HOST_WIDE_INT_PRINT_UNSIGNED "," HOST_WIDE_INT_PRINT_UNSIGNED,
size, align2);
#ifdef HAVE_GAS_HIDDEN
if (decl != NULL)
fputs (rs6000_xcoff_visibility (decl), stream);
#endif
putc ('\n', stream);
}
/* This macro produces the initial definition of a object (variable) name.
Because AIX assembler's .set command has unexpected semantics, we output
all aliases as alternative labels in front of the definition. */
void
rs6000_xcoff_declare_object_name (FILE *file, const char *name, tree decl)
{
struct declare_alias_data data = {file, false};
RS6000_OUTPUT_BASENAME (file, name);
fputs (":\n", file);
symtab_node::get_create (decl)->call_for_symbol_and_aliases (rs6000_declare_alias,
&data, true);
}
/* Overide the default 'SYMBOL-.' syntax with AIX compatible 'SYMBOL-$'. */
void
rs6000_asm_output_dwarf_pcrel (FILE *file, int size, const char *label)
{
fputs (integer_asm_op (size, FALSE), file);
assemble_name (file, label);
fputs ("-$", file);
}
/* Output a symbol offset relative to the dbase for the current object.
We use __gcc_unwind_dbase as an arbitrary base for dbase and assume
signed offsets.
__gcc_unwind_dbase is embedded in all executables/libraries through
libgcc/config/rs6000/crtdbase.S. */
void
rs6000_asm_output_dwarf_datarel (FILE *file, int size, const char *label)
{
fputs (integer_asm_op (size, FALSE), file);
assemble_name (file, label);
fputs("-__gcc_unwind_dbase", file);
}
#ifdef HAVE_AS_TLS
static void
rs6000_xcoff_encode_section_info (tree decl, rtx rtl, int first)
{
rtx symbol;
int flags;
const char *symname;
default_encode_section_info (decl, rtl, first);
/* Careful not to prod global register variables. */
if (!MEM_P (rtl))
return;
symbol = XEXP (rtl, 0);
if (GET_CODE (symbol) != SYMBOL_REF)
return;
flags = SYMBOL_REF_FLAGS (symbol);
if (TREE_CODE (decl) == VAR_DECL && DECL_THREAD_LOCAL_P (decl))
flags &= ~SYMBOL_FLAG_HAS_BLOCK_INFO;
SYMBOL_REF_FLAGS (symbol) = flags;
/* Append mapping class to extern decls. */
symname = XSTR (symbol, 0);
if (decl /* sync condition with assemble_external () */
&& DECL_P (decl) && DECL_EXTERNAL (decl) && TREE_PUBLIC (decl)
&& ((TREE_CODE (decl) == VAR_DECL && !DECL_THREAD_LOCAL_P (decl))
|| TREE_CODE (decl) == FUNCTION_DECL)
&& symname[strlen (symname) - 1] != ']')
{
char *newname = (char *) alloca (strlen (symname) + 5);
strcpy (newname, symname);
strcat (newname, (TREE_CODE (decl) == FUNCTION_DECL
? "[DS]" : "[UA]"));
XSTR (symbol, 0) = ggc_strdup (newname);
}
}
#endif /* HAVE_AS_TLS */
#endif /* TARGET_XCOFF */
void
rs6000_asm_weaken_decl (FILE *stream, tree decl,
const char *name, const char *val)
{
fputs ("\t.weak\t", stream);
RS6000_OUTPUT_BASENAME (stream, name);
if (decl && TREE_CODE (decl) == FUNCTION_DECL
&& DEFAULT_ABI == ABI_AIX && DOT_SYMBOLS)
{
if (TARGET_XCOFF)
fputs ("[DS]", stream);
#if TARGET_XCOFF && HAVE_GAS_HIDDEN
if (TARGET_XCOFF)
fputs (rs6000_xcoff_visibility (decl), stream);
#endif
fputs ("\n\t.weak\t.", stream);
RS6000_OUTPUT_BASENAME (stream, name);
}
#if TARGET_XCOFF && HAVE_GAS_HIDDEN
if (TARGET_XCOFF)
fputs (rs6000_xcoff_visibility (decl), stream);
#endif
fputc ('\n', stream);
if (val)
{
#ifdef ASM_OUTPUT_DEF
ASM_OUTPUT_DEF (stream, name, val);
#endif
if (decl && TREE_CODE (decl) == FUNCTION_DECL
&& DEFAULT_ABI == ABI_AIX && DOT_SYMBOLS)
{
fputs ("\t.set\t.", stream);
RS6000_OUTPUT_BASENAME (stream, name);
fputs (",.", stream);
RS6000_OUTPUT_BASENAME (stream, val);
fputc ('\n', stream);
}
}
}
/* Return true if INSN should not be copied. */
static bool
rs6000_cannot_copy_insn_p (rtx_insn *insn)
{
return recog_memoized (insn) >= 0
&& get_attr_cannot_copy (insn);
}
/* Compute a (partial) cost for rtx X. Return true if the complete
cost has been computed, and false if subexpressions should be
scanned. In either case, *TOTAL contains the cost result. */
static bool
rs6000_rtx_costs (rtx x, machine_mode mode, int outer_code,
int opno ATTRIBUTE_UNUSED, int *total, bool speed)
{
int code = GET_CODE (x);
switch (code)
{
/* On the RS/6000, if it is valid in the insn, it is free. */
case CONST_INT:
if (((outer_code == SET
|| outer_code == PLUS
|| outer_code == MINUS)
&& (satisfies_constraint_I (x)
|| satisfies_constraint_L (x)))
|| (outer_code == AND
&& (satisfies_constraint_K (x)
|| (mode == SImode
? satisfies_constraint_L (x)
: satisfies_constraint_J (x))))
|| ((outer_code == IOR || outer_code == XOR)
&& (satisfies_constraint_K (x)
|| (mode == SImode
? satisfies_constraint_L (x)
: satisfies_constraint_J (x))))
|| outer_code == ASHIFT
|| outer_code == ASHIFTRT
|| outer_code == LSHIFTRT
|| outer_code == ROTATE
|| outer_code == ROTATERT
|| outer_code == ZERO_EXTRACT
|| (outer_code == MULT
&& satisfies_constraint_I (x))
|| ((outer_code == DIV || outer_code == UDIV
|| outer_code == MOD || outer_code == UMOD)
&& exact_log2 (INTVAL (x)) >= 0)
|| (outer_code == COMPARE
&& (satisfies_constraint_I (x)
|| satisfies_constraint_K (x)))
|| ((outer_code == EQ || outer_code == NE)
&& (satisfies_constraint_I (x)
|| satisfies_constraint_K (x)
|| (mode == SImode
? satisfies_constraint_L (x)
: satisfies_constraint_J (x))))
|| (outer_code == GTU
&& satisfies_constraint_I (x))
|| (outer_code == LTU
&& satisfies_constraint_P (x)))
{
*total = 0;
return true;
}
else if ((outer_code == PLUS
&& reg_or_add_cint_operand (x, VOIDmode))
|| (outer_code == MINUS
&& reg_or_sub_cint_operand (x, VOIDmode))
|| ((outer_code == SET
|| outer_code == IOR
|| outer_code == XOR)
&& (INTVAL (x)
& ~ (unsigned HOST_WIDE_INT) 0xffffffff) == 0))
{
*total = COSTS_N_INSNS (1);
return true;
}
/* FALLTHRU */
case CONST_DOUBLE:
case CONST_WIDE_INT:
case CONST:
case HIGH:
case SYMBOL_REF:
*total = !speed ? COSTS_N_INSNS (1) + 1 : COSTS_N_INSNS (2);
return true;
case MEM:
/* When optimizing for size, MEM should be slightly more expensive
than generating address, e.g., (plus (reg) (const)).
L1 cache latency is about two instructions. */
*total = !speed ? COSTS_N_INSNS (1) + 1 : COSTS_N_INSNS (2);
if (rs6000_slow_unaligned_access (mode, MEM_ALIGN (x)))
*total += COSTS_N_INSNS (100);
return true;
case LABEL_REF:
*total = 0;
return true;
case PLUS:
case MINUS:
if (FLOAT_MODE_P (mode))
*total = rs6000_cost->fp;
else
*total = COSTS_N_INSNS (1);
return false;
case MULT:
if (GET_CODE (XEXP (x, 1)) == CONST_INT
&& satisfies_constraint_I (XEXP (x, 1)))
{
if (INTVAL (XEXP (x, 1)) >= -256
&& INTVAL (XEXP (x, 1)) <= 255)
*total = rs6000_cost->mulsi_const9;
else
*total = rs6000_cost->mulsi_const;
}
else if (mode == SFmode)
*total = rs6000_cost->fp;
else if (FLOAT_MODE_P (mode))
*total = rs6000_cost->dmul;
else if (mode == DImode)
*total = rs6000_cost->muldi;
else
*total = rs6000_cost->mulsi;
return false;
case FMA:
if (mode == SFmode)
*total = rs6000_cost->fp;
else
*total = rs6000_cost->dmul;
break;
case DIV:
case MOD:
if (FLOAT_MODE_P (mode))
{
*total = mode == DFmode ? rs6000_cost->ddiv
: rs6000_cost->sdiv;
return false;
}
/* FALLTHRU */
case UDIV:
case UMOD:
if (GET_CODE (XEXP (x, 1)) == CONST_INT
&& exact_log2 (INTVAL (XEXP (x, 1))) >= 0)
{
if (code == DIV || code == MOD)
/* Shift, addze */
*total = COSTS_N_INSNS (2);
else
/* Shift */
*total = COSTS_N_INSNS (1);
}
else
{
if (GET_MODE (XEXP (x, 1)) == DImode)
*total = rs6000_cost->divdi;
else
*total = rs6000_cost->divsi;
}
/* Add in shift and subtract for MOD unless we have a mod instruction. */
if (!TARGET_MODULO && (code == MOD || code == UMOD))
*total += COSTS_N_INSNS (2);
return false;
case CTZ:
*total = COSTS_N_INSNS (TARGET_CTZ ? 1 : 4);
return false;
case FFS:
*total = COSTS_N_INSNS (4);
return false;
case POPCOUNT:
*total = COSTS_N_INSNS (TARGET_POPCNTD ? 1 : 6);
return false;
case PARITY:
*total = COSTS_N_INSNS (TARGET_CMPB ? 2 : 6);
return false;
case NOT:
if (outer_code == AND || outer_code == IOR || outer_code == XOR)
*total = 0;
else
*total = COSTS_N_INSNS (1);
return false;
case AND:
if (CONST_INT_P (XEXP (x, 1)))
{
rtx left = XEXP (x, 0);
rtx_code left_code = GET_CODE (left);
/* rotate-and-mask: 1 insn. */
if ((left_code == ROTATE
|| left_code == ASHIFT
|| left_code == LSHIFTRT)
&& rs6000_is_valid_shift_mask (XEXP (x, 1), left, mode))
{
*total = rtx_cost (XEXP (left, 0), mode, left_code, 0, speed);
if (!CONST_INT_P (XEXP (left, 1)))
*total += rtx_cost (XEXP (left, 1), SImode, left_code, 1, speed);
*total += COSTS_N_INSNS (1);
return true;
}
/* rotate-and-mask (no rotate), andi., andis.: 1 insn. */
HOST_WIDE_INT val = INTVAL (XEXP (x, 1));
if (rs6000_is_valid_and_mask (XEXP (x, 1), mode)
|| (val & 0xffff) == val
|| (val & 0xffff0000) == val
|| ((val & 0xffff) == 0 && mode == SImode))
{
*total = rtx_cost (left, mode, AND, 0, speed);
*total += COSTS_N_INSNS (1);
return true;
}
/* 2 insns. */
if (rs6000_is_valid_2insn_and (XEXP (x, 1), mode))
{
*total = rtx_cost (left, mode, AND, 0, speed);
*total += COSTS_N_INSNS (2);
return true;
}
}
*total = COSTS_N_INSNS (1);
return false;
case IOR:
/* FIXME */
*total = COSTS_N_INSNS (1);
return true;
case CLZ:
case XOR:
case ZERO_EXTRACT:
*total = COSTS_N_INSNS (1);
return false;
case ASHIFT:
/* The EXTSWSLI instruction is a combined instruction. Don't count both
the sign extend and shift separately within the insn. */
if (TARGET_EXTSWSLI && mode == DImode
&& GET_CODE (XEXP (x, 0)) == SIGN_EXTEND
&& GET_MODE (XEXP (XEXP (x, 0), 0)) == SImode)
{
*total = 0;
return false;
}
/* fall through */
case ASHIFTRT:
case LSHIFTRT:
case ROTATE:
case ROTATERT:
/* Handle mul_highpart. */
if (outer_code == TRUNCATE
&& GET_CODE (XEXP (x, 0)) == MULT)
{
if (mode == DImode)
*total = rs6000_cost->muldi;
else
*total = rs6000_cost->mulsi;
return true;
}
else if (outer_code == AND)
*total = 0;
else
*total = COSTS_N_INSNS (1);
return false;
case SIGN_EXTEND:
case ZERO_EXTEND:
if (GET_CODE (XEXP (x, 0)) == MEM)
*total = 0;
else
*total = COSTS_N_INSNS (1);
return false;
case COMPARE:
case NEG:
case ABS:
if (!FLOAT_MODE_P (mode))
{
*total = COSTS_N_INSNS (1);
return false;
}
/* FALLTHRU */
case FLOAT:
case UNSIGNED_FLOAT:
case FIX:
case UNSIGNED_FIX:
case FLOAT_TRUNCATE:
*total = rs6000_cost->fp;
return false;
case FLOAT_EXTEND:
if (mode == DFmode)
*total = rs6000_cost->sfdf_convert;
else
*total = rs6000_cost->fp;
return false;
case UNSPEC:
switch (XINT (x, 1))
{
case UNSPEC_FRSP:
*total = rs6000_cost->fp;
return true;
default:
break;
}
break;
case CALL:
case IF_THEN_ELSE:
if (!speed)
{
*total = COSTS_N_INSNS (1);
return true;
}
else if (FLOAT_MODE_P (mode) && TARGET_PPC_GFXOPT && TARGET_HARD_FLOAT)
{
*total = rs6000_cost->fp;
return false;
}
break;
case NE:
case EQ:
case GTU:
case LTU:
/* Carry bit requires mode == Pmode.
NEG or PLUS already counted so only add one. */
if (mode == Pmode
&& (outer_code == NEG || outer_code == PLUS))
{
*total = COSTS_N_INSNS (1);
return true;
}
/* FALLTHRU */
case GT:
case LT:
case UNORDERED:
if (outer_code == SET)
{
if (XEXP (x, 1) == const0_rtx)
{
*total = COSTS_N_INSNS (2);
return true;
}
else
{
*total = COSTS_N_INSNS (3);
return false;
}
}
/* CC COMPARE. */
if (outer_code == COMPARE)
{
*total = 0;
return true;
}
break;
default:
break;
}
return false;
}
/* Debug form of r6000_rtx_costs that is selected if -mdebug=cost. */
static bool
rs6000_debug_rtx_costs (rtx x, machine_mode mode, int outer_code,
int opno, int *total, bool speed)
{
bool ret = rs6000_rtx_costs (x, mode, outer_code, opno, total, speed);
fprintf (stderr,
"\nrs6000_rtx_costs, return = %s, mode = %s, outer_code = %s, "
"opno = %d, total = %d, speed = %s, x:\n",
ret ? "complete" : "scan inner",
GET_MODE_NAME (mode),
GET_RTX_NAME (outer_code),
opno,
*total,
speed ? "true" : "false");
debug_rtx (x);
return ret;
}
static int
rs6000_insn_cost (rtx_insn *insn, bool speed)
{
if (recog_memoized (insn) < 0)
return 0;
if (!speed)
return get_attr_length (insn);
int cost = get_attr_cost (insn);
if (cost > 0)
return cost;
int n = get_attr_length (insn) / 4;
enum attr_type type = get_attr_type (insn);
switch (type)
{
case TYPE_LOAD:
case TYPE_FPLOAD:
case TYPE_VECLOAD:
cost = COSTS_N_INSNS (n + 1);
break;
case TYPE_MUL:
switch (get_attr_size (insn))
{
case SIZE_8:
cost = COSTS_N_INSNS (n - 1) + rs6000_cost->mulsi_const9;
break;
case SIZE_16:
cost = COSTS_N_INSNS (n - 1) + rs6000_cost->mulsi_const;
break;
case SIZE_32:
cost = COSTS_N_INSNS (n - 1) + rs6000_cost->mulsi;
break;
case SIZE_64:
cost = COSTS_N_INSNS (n - 1) + rs6000_cost->muldi;
break;
default:
gcc_unreachable ();
}
break;
case TYPE_DIV:
switch (get_attr_size (insn))
{
case SIZE_32:
cost = COSTS_N_INSNS (n - 1) + rs6000_cost->divsi;
break;
case SIZE_64:
cost = COSTS_N_INSNS (n - 1) + rs6000_cost->divdi;
break;
default:
gcc_unreachable ();
}
break;
case TYPE_FP:
cost = n * rs6000_cost->fp;
break;
case TYPE_DMUL:
cost = n * rs6000_cost->dmul;
break;
case TYPE_SDIV:
cost = n * rs6000_cost->sdiv;
break;
case TYPE_DDIV:
cost = n * rs6000_cost->ddiv;
break;
case TYPE_SYNC:
case TYPE_LOAD_L:
case TYPE_MFCR:
case TYPE_MFCRF:
cost = COSTS_N_INSNS (n + 2);
break;
default:
cost = COSTS_N_INSNS (n);
}
return cost;
}
/* Debug form of ADDRESS_COST that is selected if -mdebug=cost. */
static int
rs6000_debug_address_cost (rtx x, machine_mode mode,
addr_space_t as, bool speed)
{
int ret = TARGET_ADDRESS_COST (x, mode, as, speed);
fprintf (stderr, "\nrs6000_address_cost, return = %d, speed = %s, x:\n",
ret, speed ? "true" : "false");
debug_rtx (x);
return ret;
}
/* A C expression returning the cost of moving data from a register of class
CLASS1 to one of CLASS2. */
static int
rs6000_register_move_cost (machine_mode mode,
reg_class_t from, reg_class_t to)
{
int ret;
if (TARGET_DEBUG_COST)
dbg_cost_ctrl++;
/* Moves from/to GENERAL_REGS. */
if (reg_classes_intersect_p (to, GENERAL_REGS)
|| reg_classes_intersect_p (from, GENERAL_REGS))
{
reg_class_t rclass = from;
if (! reg_classes_intersect_p (to, GENERAL_REGS))
rclass = to;
if (rclass == FLOAT_REGS || rclass == ALTIVEC_REGS || rclass == VSX_REGS)
ret = (rs6000_memory_move_cost (mode, rclass, false)
+ rs6000_memory_move_cost (mode, GENERAL_REGS, false));
/* It's more expensive to move CR_REGS than CR0_REGS because of the
shift. */
else if (rclass == CR_REGS)
ret = 4;
/* For those processors that have slow LR/CTR moves, make them more
expensive than memory in order to bias spills to memory .*/
else if ((rs6000_tune == PROCESSOR_POWER6
|| rs6000_tune == PROCESSOR_POWER7
|| rs6000_tune == PROCESSOR_POWER8
|| rs6000_tune == PROCESSOR_POWER9)
&& reg_classes_intersect_p (rclass, LINK_OR_CTR_REGS))
ret = 6 * hard_regno_nregs (0, mode);
else
/* A move will cost one instruction per GPR moved. */
ret = 2 * hard_regno_nregs (0, mode);
}
/* If we have VSX, we can easily move between FPR or Altivec registers. */
else if (VECTOR_MEM_VSX_P (mode)
&& reg_classes_intersect_p (to, VSX_REGS)
&& reg_classes_intersect_p (from, VSX_REGS))
ret = 2 * hard_regno_nregs (FIRST_FPR_REGNO, mode);
/* Moving between two similar registers is just one instruction. */
else if (reg_classes_intersect_p (to, from))
ret = (FLOAT128_2REG_P (mode)) ? 4 : 2;
/* Everything else has to go through GENERAL_REGS. */
else
ret = (rs6000_register_move_cost (mode, GENERAL_REGS, to)
+ rs6000_register_move_cost (mode, from, GENERAL_REGS));
if (TARGET_DEBUG_COST)
{
if (dbg_cost_ctrl == 1)
fprintf (stderr,
"rs6000_register_move_cost:, ret=%d, mode=%s, from=%s, to=%s\n",
ret, GET_MODE_NAME (mode), reg_class_names[from],
reg_class_names[to]);
dbg_cost_ctrl--;
}
return ret;
}
/* A C expressions returning the cost of moving data of MODE from a register to
or from memory. */
static int
rs6000_memory_move_cost (machine_mode mode, reg_class_t rclass,
bool in ATTRIBUTE_UNUSED)
{
int ret;
if (TARGET_DEBUG_COST)
dbg_cost_ctrl++;
if (reg_classes_intersect_p (rclass, GENERAL_REGS))
ret = 4 * hard_regno_nregs (0, mode);
else if ((reg_classes_intersect_p (rclass, FLOAT_REGS)
|| reg_classes_intersect_p (rclass, VSX_REGS)))
ret = 4 * hard_regno_nregs (32, mode);
else if (reg_classes_intersect_p (rclass, ALTIVEC_REGS))
ret = 4 * hard_regno_nregs (FIRST_ALTIVEC_REGNO, mode);
else
ret = 4 + rs6000_register_move_cost (mode, rclass, GENERAL_REGS);
if (TARGET_DEBUG_COST)
{
if (dbg_cost_ctrl == 1)
fprintf (stderr,
"rs6000_memory_move_cost: ret=%d, mode=%s, rclass=%s, in=%d\n",
ret, GET_MODE_NAME (mode), reg_class_names[rclass], in);
dbg_cost_ctrl--;
}
return ret;
}
/* Returns a code for a target-specific builtin that implements
reciprocal of the function, or NULL_TREE if not available. */
static tree
rs6000_builtin_reciprocal (tree fndecl)
{
switch (DECL_FUNCTION_CODE (fndecl))
{
case VSX_BUILTIN_XVSQRTDP:
if (!RS6000_RECIP_AUTO_RSQRTE_P (V2DFmode))
return NULL_TREE;
return rs6000_builtin_decls[VSX_BUILTIN_RSQRT_2DF];
case VSX_BUILTIN_XVSQRTSP:
if (!RS6000_RECIP_AUTO_RSQRTE_P (V4SFmode))
return NULL_TREE;
return rs6000_builtin_decls[VSX_BUILTIN_RSQRT_4SF];
default:
return NULL_TREE;
}
}
/* Load up a constant. If the mode is a vector mode, splat the value across
all of the vector elements. */
static rtx
rs6000_load_constant_and_splat (machine_mode mode, REAL_VALUE_TYPE dconst)
{
rtx reg;
if (mode == SFmode || mode == DFmode)
{
rtx d = const_double_from_real_value (dconst, mode);
reg = force_reg (mode, d);
}
else if (mode == V4SFmode)
{
rtx d = const_double_from_real_value (dconst, SFmode);
rtvec v = gen_rtvec (4, d, d, d, d);
reg = gen_reg_rtx (mode);
rs6000_expand_vector_init (reg, gen_rtx_PARALLEL (mode, v));
}
else if (mode == V2DFmode)
{
rtx d = const_double_from_real_value (dconst, DFmode);
rtvec v = gen_rtvec (2, d, d);
reg = gen_reg_rtx (mode);
rs6000_expand_vector_init (reg, gen_rtx_PARALLEL (mode, v));
}
else
gcc_unreachable ();
return reg;
}
/* Generate an FMA instruction. */
static void
rs6000_emit_madd (rtx target, rtx m1, rtx m2, rtx a)
{
machine_mode mode = GET_MODE (target);
rtx dst;
dst = expand_ternary_op (mode, fma_optab, m1, m2, a, target, 0);
gcc_assert (dst != NULL);
if (dst != target)
emit_move_insn (target, dst);
}
/* Generate a FNMSUB instruction: dst = -fma(m1, m2, -a). */
static void
rs6000_emit_nmsub (rtx dst, rtx m1, rtx m2, rtx a)
{
machine_mode mode = GET_MODE (dst);
rtx r;
/* This is a tad more complicated, since the fnma_optab is for
a different expression: fma(-m1, m2, a), which is the same
thing except in the case of signed zeros.
Fortunately we know that if FMA is supported that FNMSUB is
also supported in the ISA. Just expand it directly. */
gcc_assert (optab_handler (fma_optab, mode) != CODE_FOR_nothing);
r = gen_rtx_NEG (mode, a);
r = gen_rtx_FMA (mode, m1, m2, r);
r = gen_rtx_NEG (mode, r);
emit_insn (gen_rtx_SET (dst, r));
}
/* Newton-Raphson approximation of floating point divide DST = N/D. If NOTE_P,
add a reg_note saying that this was a division. Support both scalar and
vector divide. Assumes no trapping math and finite arguments. */
void
rs6000_emit_swdiv (rtx dst, rtx n, rtx d, bool note_p)
{
machine_mode mode = GET_MODE (dst);
rtx one, x0, e0, x1, xprev, eprev, xnext, enext, u, v;
int i;
/* Low precision estimates guarantee 5 bits of accuracy. High
precision estimates guarantee 14 bits of accuracy. SFmode
requires 23 bits of accuracy. DFmode requires 52 bits of
accuracy. Each pass at least doubles the accuracy, leading
to the following. */
int passes = (TARGET_RECIP_PRECISION) ? 1 : 3;
if (mode == DFmode || mode == V2DFmode)
passes++;
enum insn_code code = optab_handler (smul_optab, mode);
insn_gen_fn gen_mul = GEN_FCN (code);
gcc_assert (code != CODE_FOR_nothing);
one = rs6000_load_constant_and_splat (mode, dconst1);
/* x0 = 1./d estimate */
x0 = gen_reg_rtx (mode);
emit_insn (gen_rtx_SET (x0, gen_rtx_UNSPEC (mode, gen_rtvec (1, d),
UNSPEC_FRES)));
/* Each iteration but the last calculates x_(i+1) = x_i * (2 - d * x_i). */
if (passes > 1) {
/* e0 = 1. - d * x0 */
e0 = gen_reg_rtx (mode);
rs6000_emit_nmsub (e0, d, x0, one);
/* x1 = x0 + e0 * x0 */
x1 = gen_reg_rtx (mode);
rs6000_emit_madd (x1, e0, x0, x0);
for (i = 0, xprev = x1, eprev = e0; i < passes - 2;
++i, xprev = xnext, eprev = enext) {
/* enext = eprev * eprev */
enext = gen_reg_rtx (mode);
emit_insn (gen_mul (enext, eprev, eprev));
/* xnext = xprev + enext * xprev */
xnext = gen_reg_rtx (mode);
rs6000_emit_madd (xnext, enext, xprev, xprev);
}
} else
xprev = x0;
/* The last iteration calculates x_(i+1) = n * x_i * (2 - d * x_i). */
/* u = n * xprev */
u = gen_reg_rtx (mode);
emit_insn (gen_mul (u, n, xprev));
/* v = n - (d * u) */
v = gen_reg_rtx (mode);
rs6000_emit_nmsub (v, d, u, n);
/* dst = (v * xprev) + u */
rs6000_emit_madd (dst, v, xprev, u);
if (note_p)
add_reg_note (get_last_insn (), REG_EQUAL, gen_rtx_DIV (mode, n, d));
}
/* Goldschmidt's Algorithm for single/double-precision floating point
sqrt and rsqrt. Assumes no trapping math and finite arguments. */
void
rs6000_emit_swsqrt (rtx dst, rtx src, bool recip)
{
machine_mode mode = GET_MODE (src);
rtx e = gen_reg_rtx (mode);
rtx g = gen_reg_rtx (mode);
rtx h = gen_reg_rtx (mode);
/* Low precision estimates guarantee 5 bits of accuracy. High
precision estimates guarantee 14 bits of accuracy. SFmode
requires 23 bits of accuracy. DFmode requires 52 bits of
accuracy. Each pass at least doubles the accuracy, leading
to the following. */
int passes = (TARGET_RECIP_PRECISION) ? 1 : 3;
if (mode == DFmode || mode == V2DFmode)
passes++;
int i;
rtx mhalf;
enum insn_code code = optab_handler (smul_optab, mode);
insn_gen_fn gen_mul = GEN_FCN (code);
gcc_assert (code != CODE_FOR_nothing);
mhalf = rs6000_load_constant_and_splat (mode, dconsthalf);
/* e = rsqrt estimate */
emit_insn (gen_rtx_SET (e, gen_rtx_UNSPEC (mode, gen_rtvec (1, src),
UNSPEC_RSQRT)));
/* If (src == 0.0) filter infinity to prevent NaN for sqrt(0.0). */
if (!recip)
{
rtx zero = force_reg (mode, CONST0_RTX (mode));
if (mode == SFmode)
{
rtx target = emit_conditional_move (e, GT, src, zero, mode,
e, zero, mode, 0);
if (target != e)
emit_move_insn (e, target);
}
else
{
rtx cond = gen_rtx_GT (VOIDmode, e, zero);
rs6000_emit_vector_cond_expr (e, e, zero, cond, src, zero);
}
}
/* g = sqrt estimate. */
emit_insn (gen_mul (g, e, src));
/* h = 1/(2*sqrt) estimate. */
emit_insn (gen_mul (h, e, mhalf));
if (recip)
{
if (passes == 1)
{
rtx t = gen_reg_rtx (mode);
rs6000_emit_nmsub (t, g, h, mhalf);
/* Apply correction directly to 1/rsqrt estimate. */
rs6000_emit_madd (dst, e, t, e);
}
else
{
for (i = 0; i < passes; i++)
{
rtx t1 = gen_reg_rtx (mode);
rtx g1 = gen_reg_rtx (mode);
rtx h1 = gen_reg_rtx (mode);
rs6000_emit_nmsub (t1, g, h, mhalf);
rs6000_emit_madd (g1, g, t1, g);
rs6000_emit_madd (h1, h, t1, h);
g = g1;
h = h1;
}
/* Multiply by 2 for 1/rsqrt. */
emit_insn (gen_add3_insn (dst, h, h));
}
}
else
{
rtx t = gen_reg_rtx (mode);
rs6000_emit_nmsub (t, g, h, mhalf);
rs6000_emit_madd (dst, g, t, g);
}
return;
}
/* Emit popcount intrinsic on TARGET_POPCNTB (Power5) and TARGET_POPCNTD
(Power7) targets. DST is the target, and SRC is the argument operand. */
void
rs6000_emit_popcount (rtx dst, rtx src)
{
machine_mode mode = GET_MODE (dst);
rtx tmp1, tmp2;
/* Use the PPC ISA 2.06 popcnt{w,d} instruction if we can. */
if (TARGET_POPCNTD)
{
if (mode == SImode)
emit_insn (gen_popcntdsi2 (dst, src));
else
emit_insn (gen_popcntddi2 (dst, src));
return;
}
tmp1 = gen_reg_rtx (mode);
if (mode == SImode)
{
emit_insn (gen_popcntbsi2 (tmp1, src));
tmp2 = expand_mult (SImode, tmp1, GEN_INT (0x01010101),
NULL_RTX, 0);
tmp2 = force_reg (SImode, tmp2);
emit_insn (gen_lshrsi3 (dst, tmp2, GEN_INT (24)));
}
else
{
emit_insn (gen_popcntbdi2 (tmp1, src));
tmp2 = expand_mult (DImode, tmp1,
GEN_INT ((HOST_WIDE_INT)
0x01010101 << 32 | 0x01010101),
NULL_RTX, 0);
tmp2 = force_reg (DImode, tmp2);
emit_insn (gen_lshrdi3 (dst, tmp2, GEN_INT (56)));
}
}
/* Emit parity intrinsic on TARGET_POPCNTB targets. DST is the
target, and SRC is the argument operand. */
void
rs6000_emit_parity (rtx dst, rtx src)
{
machine_mode mode = GET_MODE (dst);
rtx tmp;
tmp = gen_reg_rtx (mode);
/* Use the PPC ISA 2.05 prtyw/prtyd instruction if we can. */
if (TARGET_CMPB)
{
if (mode == SImode)
{
emit_insn (gen_popcntbsi2 (tmp, src));
emit_insn (gen_paritysi2_cmpb (dst, tmp));
}
else
{
emit_insn (gen_popcntbdi2 (tmp, src));
emit_insn (gen_paritydi2_cmpb (dst, tmp));
}
return;
}
if (mode == SImode)
{
/* Is mult+shift >= shift+xor+shift+xor? */
if (rs6000_cost->mulsi_const >= COSTS_N_INSNS (3))
{
rtx tmp1, tmp2, tmp3, tmp4;
tmp1 = gen_reg_rtx (SImode);
emit_insn (gen_popcntbsi2 (tmp1, src));
tmp2 = gen_reg_rtx (SImode);
emit_insn (gen_lshrsi3 (tmp2, tmp1, GEN_INT (16)));
tmp3 = gen_reg_rtx (SImode);
emit_insn (gen_xorsi3 (tmp3, tmp1, tmp2));
tmp4 = gen_reg_rtx (SImode);
emit_insn (gen_lshrsi3 (tmp4, tmp3, GEN_INT (8)));
emit_insn (gen_xorsi3 (tmp, tmp3, tmp4));
}
else
rs6000_emit_popcount (tmp, src);
emit_insn (gen_andsi3 (dst, tmp, const1_rtx));
}
else
{
/* Is mult+shift >= shift+xor+shift+xor+shift+xor? */
if (rs6000_cost->muldi >= COSTS_N_INSNS (5))
{
rtx tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;
tmp1 = gen_reg_rtx (DImode);
emit_insn (gen_popcntbdi2 (tmp1, src));
tmp2 = gen_reg_rtx (DImode);
emit_insn (gen_lshrdi3 (tmp2, tmp1, GEN_INT (32)));
tmp3 = gen_reg_rtx (DImode);
emit_insn (gen_xordi3 (tmp3, tmp1, tmp2));
tmp4 = gen_reg_rtx (DImode);
emit_insn (gen_lshrdi3 (tmp4, tmp3, GEN_INT (16)));
tmp5 = gen_reg_rtx (DImode);
emit_insn (gen_xordi3 (tmp5, tmp3, tmp4));
tmp6 = gen_reg_rtx (DImode);
emit_insn (gen_lshrdi3 (tmp6, tmp5, GEN_INT (8)));
emit_insn (gen_xordi3 (tmp, tmp5, tmp6));
}
else
rs6000_emit_popcount (tmp, src);
emit_insn (gen_anddi3 (dst, tmp, const1_rtx));
}
}
/* Expand an Altivec constant permutation for little endian mode.
There are two issues: First, the two input operands must be
swapped so that together they form a double-wide array in LE
order. Second, the vperm instruction has surprising behavior
in LE mode: it interprets the elements of the source vectors
in BE mode ("left to right") and interprets the elements of
the destination vector in LE mode ("right to left"). To
correct for this, we must subtract each element of the permute
control vector from 31.
For example, suppose we want to concatenate vr10 = {0, 1, 2, 3}
with vr11 = {4, 5, 6, 7} and extract {0, 2, 4, 6} using a vperm.
We place {0,1,2,3,8,9,10,11,16,17,18,19,24,25,26,27} in vr12 to
serve as the permute control vector. Then, in BE mode,
vperm 9,10,11,12
places the desired result in vr9. However, in LE mode the
vector contents will be
vr10 = 00000003 00000002 00000001 00000000
vr11 = 00000007 00000006 00000005 00000004
The result of the vperm using the same permute control vector is
vr9 = 05000000 07000000 01000000 03000000
That is, the leftmost 4 bytes of vr10 are interpreted as the
source for the rightmost 4 bytes of vr9, and so on.
If we change the permute control vector to
vr12 = {31,20,29,28,23,22,21,20,15,14,13,12,7,6,5,4}
and issue
vperm 9,11,10,12
we get the desired
vr9 = 00000006 00000004 00000002 00000000. */
void
altivec_expand_vec_perm_const_le (rtx operands[4])
{
unsigned int i;
rtx perm[16];
rtx constv, unspec;
rtx target = operands[0];
rtx op0 = operands[1];
rtx op1 = operands[2];
rtx sel = operands[3];
/* Unpack and adjust the constant selector. */
for (i = 0; i < 16; ++i)
{
rtx e = XVECEXP (sel, 0, i);
unsigned int elt = 31 - (INTVAL (e) & 31);
perm[i] = GEN_INT (elt);
}
/* Expand to a permute, swapping the inputs and using the
adjusted selector. */
if (!REG_P (op0))
op0 = force_reg (V16QImode, op0);
if (!REG_P (op1))
op1 = force_reg (V16QImode, op1);
constv = gen_rtx_CONST_VECTOR (V16QImode, gen_rtvec_v (16, perm));
constv = force_reg (V16QImode, constv);
unspec = gen_rtx_UNSPEC (V16QImode, gen_rtvec (3, op1, op0, constv),
UNSPEC_VPERM);
if (!REG_P (target))
{
rtx tmp = gen_reg_rtx (V16QImode);
emit_move_insn (tmp, unspec);
unspec = tmp;
}
emit_move_insn (target, unspec);
}
/* Similarly to altivec_expand_vec_perm_const_le, we must adjust the
permute control vector. But here it's not a constant, so we must
generate a vector NAND or NOR to do the adjustment. */
void
altivec_expand_vec_perm_le (rtx operands[4])
{
rtx notx, iorx, unspec;
rtx target = operands[0];
rtx op0 = operands[1];
rtx op1 = operands[2];
rtx sel = operands[3];
rtx tmp = target;
rtx norreg = gen_reg_rtx (V16QImode);
machine_mode mode = GET_MODE (target);
/* Get everything in regs so the pattern matches. */
if (!REG_P (op0))
op0 = force_reg (mode, op0);
if (!REG_P (op1))
op1 = force_reg (mode, op1);
if (!REG_P (sel))
sel = force_reg (V16QImode, sel);
if (!REG_P (target))
tmp = gen_reg_rtx (mode);
if (TARGET_P9_VECTOR)
{
unspec = gen_rtx_UNSPEC (mode, gen_rtvec (3, op0, op1, sel),
UNSPEC_VPERMR);
}
else
{
/* Invert the selector with a VNAND if available, else a VNOR.
The VNAND is preferred for future fusion opportunities. */
notx = gen_rtx_NOT (V16QImode, sel);
iorx = (TARGET_P8_VECTOR
? gen_rtx_IOR (V16QImode, notx, notx)
: gen_rtx_AND (V16QImode, notx, notx));
emit_insn (gen_rtx_SET (norreg, iorx));
/* Permute with operands reversed and adjusted selector. */
unspec = gen_rtx_UNSPEC (mode, gen_rtvec (3, op1, op0, norreg),
UNSPEC_VPERM);
}
/* Copy into target, possibly by way of a register. */
if (!REG_P (target))
{
emit_move_insn (tmp, unspec);
unspec = tmp;
}
emit_move_insn (target, unspec);
}
/* Expand an Altivec constant permutation. Return true if we match
an efficient implementation; false to fall back to VPERM. */
bool
altivec_expand_vec_perm_const (rtx operands[4])
{
struct altivec_perm_insn {
HOST_WIDE_INT mask;
enum insn_code impl;
unsigned char perm[16];
};
static const struct altivec_perm_insn patterns[] = {
{ OPTION_MASK_ALTIVEC, CODE_FOR_altivec_vpkuhum_direct,
{ 1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31 } },
{ OPTION_MASK_ALTIVEC, CODE_FOR_altivec_vpkuwum_direct,
{ 2, 3, 6, 7, 10, 11, 14, 15, 18, 19, 22, 23, 26, 27, 30, 31 } },
{ OPTION_MASK_ALTIVEC,
(BYTES_BIG_ENDIAN ? CODE_FOR_altivec_vmrghb_direct
: CODE_FOR_altivec_vmrglb_direct),
{ 0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23 } },
{ OPTION_MASK_ALTIVEC,
(BYTES_BIG_ENDIAN ? CODE_FOR_altivec_vmrghh_direct
: CODE_FOR_altivec_vmrglh_direct),
{ 0, 1, 16, 17, 2, 3, 18, 19, 4, 5, 20, 21, 6, 7, 22, 23 } },
{ OPTION_MASK_ALTIVEC,
(BYTES_BIG_ENDIAN ? CODE_FOR_altivec_vmrghw_direct
: CODE_FOR_altivec_vmrglw_direct),
{ 0, 1, 2, 3, 16, 17, 18, 19, 4, 5, 6, 7, 20, 21, 22, 23 } },
{ OPTION_MASK_ALTIVEC,
(BYTES_BIG_ENDIAN ? CODE_FOR_altivec_vmrglb_direct
: CODE_FOR_altivec_vmrghb_direct),
{ 8, 24, 9, 25, 10, 26, 11, 27, 12, 28, 13, 29, 14, 30, 15, 31 } },
{ OPTION_MASK_ALTIVEC,
(BYTES_BIG_ENDIAN ? CODE_FOR_altivec_vmrglh_direct
: CODE_FOR_altivec_vmrghh_direct),
{ 8, 9, 24, 25, 10, 11, 26, 27, 12, 13, 28, 29, 14, 15, 30, 31 } },
{ OPTION_MASK_ALTIVEC,
(BYTES_BIG_ENDIAN ? CODE_FOR_altivec_vmrglw_direct
: CODE_FOR_altivec_vmrghw_direct),
{ 8, 9, 10, 11, 24, 25, 26, 27, 12, 13, 14, 15, 28, 29, 30, 31 } },
{ OPTION_MASK_P8_VECTOR,
(BYTES_BIG_ENDIAN ? CODE_FOR_p8_vmrgew_v4sf_direct
: CODE_FOR_p8_vmrgow_v4sf_direct),
{ 0, 1, 2, 3, 16, 17, 18, 19, 8, 9, 10, 11, 24, 25, 26, 27 } },
{ OPTION_MASK_P8_VECTOR,
(BYTES_BIG_ENDIAN ? CODE_FOR_p8_vmrgow_v4sf_direct
: CODE_FOR_p8_vmrgew_v4sf_direct),
{ 4, 5, 6, 7, 20, 21, 22, 23, 12, 13, 14, 15, 28, 29, 30, 31 } }
};
unsigned int i, j, elt, which;
unsigned char perm[16];
rtx target, op0, op1, sel, x;
bool one_vec;
target = operands[0];
op0 = operands[1];
op1 = operands[2];
sel = operands[3];
/* Unpack the constant selector. */
for (i = which = 0; i < 16; ++i)
{
rtx e = XVECEXP (sel, 0, i);
elt = INTVAL (e) & 31;
which |= (elt < 16 ? 1 : 2);
perm[i] = elt;
}
/* Simplify the constant selector based on operands. */
switch (which)
{
default:
gcc_unreachable ();
case 3:
one_vec = false;
if (!rtx_equal_p (op0, op1))
break;
/* FALLTHRU */
case 2:
for (i = 0; i < 16; ++i)
perm[i] &= 15;
op0 = op1;
one_vec = true;
break;
case 1:
op1 = op0;
one_vec = true;
break;
}
/* Look for splat patterns. */
if (one_vec)
{
elt = perm[0];
for (i = 0; i < 16; ++i)
if (perm[i] != elt)
break;
if (i == 16)
{
if (!BYTES_BIG_ENDIAN)
elt = 15 - elt;
emit_insn (gen_altivec_vspltb_direct (target, op0, GEN_INT (elt)));
return true;
}
if (elt % 2 == 0)
{
for (i = 0; i < 16; i += 2)
if (perm[i] != elt || perm[i + 1] != elt + 1)
break;
if (i == 16)
{
int field = BYTES_BIG_ENDIAN ? elt / 2 : 7 - elt / 2;
x = gen_reg_rtx (V8HImode);
emit_insn (gen_altivec_vsplth_direct (x, gen_lowpart (V8HImode, op0),
GEN_INT (field)));
emit_move_insn (target, gen_lowpart (V16QImode, x));
return true;
}
}
if (elt % 4 == 0)
{
for (i = 0; i < 16; i += 4)
if (perm[i] != elt
|| perm[i + 1] != elt + 1
|| perm[i + 2] != elt + 2
|| perm[i + 3] != elt + 3)
break;
if (i == 16)
{
int field = BYTES_BIG_ENDIAN ? elt / 4 : 3 - elt / 4;
x = gen_reg_rtx (V4SImode);
emit_insn (gen_altivec_vspltw_direct (x, gen_lowpart (V4SImode, op0),
GEN_INT (field)));
emit_move_insn (target, gen_lowpart (V16QImode, x));
return true;
}
}
}
/* Look for merge and pack patterns. */
for (j = 0; j < ARRAY_SIZE (patterns); ++j)
{
bool swapped;
if ((patterns[j].mask & rs6000_isa_flags) == 0)
continue;
elt = patterns[j].perm[0];
if (perm[0] == elt)
swapped = false;
else if (perm[0] == elt + 16)
swapped = true;
else
continue;
for (i = 1; i < 16; ++i)
{
elt = patterns[j].perm[i];
if (swapped)
elt = (elt >= 16 ? elt - 16 : elt + 16);
else if (one_vec && elt >= 16)
elt -= 16;
if (perm[i] != elt)
break;
}
if (i == 16)
{
enum insn_code icode = patterns[j].impl;
machine_mode omode = insn_data[icode].operand[0].mode;
machine_mode imode = insn_data[icode].operand[1].mode;
/* For little-endian, don't use vpkuwum and vpkuhum if the
underlying vector type is not V4SI and V8HI, respectively.
For example, using vpkuwum with a V8HI picks up the even
halfwords (BE numbering) when the even halfwords (LE
numbering) are what we need. */
if (!BYTES_BIG_ENDIAN
&& icode == CODE_FOR_altivec_vpkuwum_direct
&& ((GET_CODE (op0) == REG
&& GET_MODE (op0) != V4SImode)
|| (GET_CODE (op0) == SUBREG
&& GET_MODE (XEXP (op0, 0)) != V4SImode)))
continue;
if (!BYTES_BIG_ENDIAN
&& icode == CODE_FOR_altivec_vpkuhum_direct
&& ((GET_CODE (op0) == REG
&& GET_MODE (op0) != V8HImode)
|| (GET_CODE (op0) == SUBREG
&& GET_MODE (XEXP (op0, 0)) != V8HImode)))
continue;
/* For little-endian, the two input operands must be swapped
(or swapped back) to ensure proper right-to-left numbering
from 0 to 2N-1. */
if (swapped ^ !BYTES_BIG_ENDIAN)
std::swap (op0, op1);
if (imode != V16QImode)
{
op0 = gen_lowpart (imode, op0);
op1 = gen_lowpart (imode, op1);
}
if (omode == V16QImode)
x = target;
else
x = gen_reg_rtx (omode);
emit_insn (GEN_FCN (icode) (x, op0, op1));
if (omode != V16QImode)
emit_move_insn (target, gen_lowpart (V16QImode, x));
return true;
}
}
if (!BYTES_BIG_ENDIAN)
{
altivec_expand_vec_perm_const_le (operands);
return true;
}
return false;
}
/* Expand a Paired Single or VSX Permute Doubleword constant permutation.
Return true if we match an efficient implementation. */
static bool
rs6000_expand_vec_perm_const_1 (rtx target, rtx op0, rtx op1,
unsigned char perm0, unsigned char perm1)
{
rtx x;
/* If both selectors come from the same operand, fold to single op. */
if ((perm0 & 2) == (perm1 & 2))
{
if (perm0 & 2)
op0 = op1;
else
op1 = op0;
}
/* If both operands are equal, fold to simpler permutation. */
if (rtx_equal_p (op0, op1))
{
perm0 = perm0 & 1;
perm1 = (perm1 & 1) + 2;
}
/* If the first selector comes from the second operand, swap. */
else if (perm0 & 2)
{
if (perm1 & 2)
return false;
perm0 -= 2;
perm1 += 2;
std::swap (op0, op1);
}
/* If the second selector does not come from the second operand, fail. */
else if ((perm1 & 2) == 0)
return false;
/* Success! */
if (target != NULL)
{
machine_mode vmode, dmode;
rtvec v;
vmode = GET_MODE (target);
gcc_assert (GET_MODE_NUNITS (vmode) == 2);
dmode = mode_for_vector (GET_MODE_INNER (vmode), 4).require ();
x = gen_rtx_VEC_CONCAT (dmode, op0, op1);
v = gen_rtvec (2, GEN_INT (perm0), GEN_INT (perm1));
x = gen_rtx_VEC_SELECT (vmode, x, gen_rtx_PARALLEL (VOIDmode, v));
emit_insn (gen_rtx_SET (target, x));
}
return true;
}
bool
rs6000_expand_vec_perm_const (rtx operands[4])
{
rtx target, op0, op1, sel;
unsigned char perm0, perm1;
target = operands[0];
op0 = operands[1];
op1 = operands[2];
sel = operands[3];
/* Unpack the constant selector. */
perm0 = INTVAL (XVECEXP (sel, 0, 0)) & 3;
perm1 = INTVAL (XVECEXP (sel, 0, 1)) & 3;
return rs6000_expand_vec_perm_const_1 (target, op0, op1, perm0, perm1);
}
/* Test whether a constant permutation is supported. */
static bool
rs6000_vectorize_vec_perm_const_ok (machine_mode vmode, vec_perm_indices sel)
{
/* AltiVec (and thus VSX) can handle arbitrary permutations. */
if (TARGET_ALTIVEC)
return true;
/* Check for ps_merge* or evmerge* insns. */
if (TARGET_PAIRED_FLOAT && vmode == V2SFmode)
{
rtx op0 = gen_raw_REG (vmode, LAST_VIRTUAL_REGISTER + 1);
rtx op1 = gen_raw_REG (vmode, LAST_VIRTUAL_REGISTER + 2);
return rs6000_expand_vec_perm_const_1 (NULL, op0, op1, sel[0], sel[1]);
}
return false;
}
/* A subroutine for rs6000_expand_extract_even & rs6000_expand_interleave. */
static void
rs6000_do_expand_vec_perm (rtx target, rtx op0, rtx op1,
machine_mode vmode, unsigned nelt, rtx perm[])
{
machine_mode imode;
rtx x;
imode = vmode;
if (GET_MODE_CLASS (vmode) != MODE_VECTOR_INT)
imode = mode_for_int_vector (vmode).require ();
x = gen_rtx_CONST_VECTOR (imode, gen_rtvec_v (nelt, perm));
x = expand_vec_perm (vmode, op0, op1, x, target);
if (x != target)
emit_move_insn (target, x);
}
/* Expand an extract even operation. */
void
rs6000_expand_extract_even (rtx target, rtx op0, rtx op1)
{
machine_mode vmode = GET_MODE (target);
unsigned i, nelt = GET_MODE_NUNITS (vmode);
rtx perm[16];
for (i = 0; i < nelt; i++)
perm[i] = GEN_INT (i * 2);
rs6000_do_expand_vec_perm (target, op0, op1, vmode, nelt, perm);
}
/* Expand a vector interleave operation. */
void
rs6000_expand_interleave (rtx target, rtx op0, rtx op1, bool highp)
{
machine_mode vmode = GET_MODE (target);
unsigned i, high, nelt = GET_MODE_NUNITS (vmode);
rtx perm[16];
high = (highp ? 0 : nelt / 2);
for (i = 0; i < nelt / 2; i++)
{
perm[i * 2] = GEN_INT (i + high);
perm[i * 2 + 1] = GEN_INT (i + nelt + high);
}
rs6000_do_expand_vec_perm (target, op0, op1, vmode, nelt, perm);
}
/* Scale a V2DF vector SRC by two to the SCALE and place in TGT. */
void
rs6000_scale_v2df (rtx tgt, rtx src, int scale)
{
HOST_WIDE_INT hwi_scale (scale);
REAL_VALUE_TYPE r_pow;
rtvec v = rtvec_alloc (2);
rtx elt;
rtx scale_vec = gen_reg_rtx (V2DFmode);
(void)real_powi (&r_pow, DFmode, &dconst2, hwi_scale);
elt = const_double_from_real_value (r_pow, DFmode);
RTVEC_ELT (v, 0) = elt;
RTVEC_ELT (v, 1) = elt;
rs6000_expand_vector_init (scale_vec, gen_rtx_PARALLEL (V2DFmode, v));
emit_insn (gen_mulv2df3 (tgt, src, scale_vec));
}
/* Return an RTX representing where to find the function value of a
function returning MODE. */
static rtx
rs6000_complex_function_value (machine_mode mode)
{
unsigned int regno;
rtx r1, r2;
machine_mode inner = GET_MODE_INNER (mode);
unsigned int inner_bytes = GET_MODE_UNIT_SIZE (mode);
if (TARGET_FLOAT128_TYPE
&& (mode == KCmode
|| (mode == TCmode && TARGET_IEEEQUAD)))
regno = ALTIVEC_ARG_RETURN;
else if (FLOAT_MODE_P (mode) && TARGET_HARD_FLOAT)
regno = FP_ARG_RETURN;
else
{
regno = GP_ARG_RETURN;
/* 32-bit is OK since it'll go in r3/r4. */
if (TARGET_32BIT && inner_bytes >= 4)
return gen_rtx_REG (mode, regno);
}
if (inner_bytes >= 8)
return gen_rtx_REG (mode, regno);
r1 = gen_rtx_EXPR_LIST (inner, gen_rtx_REG (inner, regno),
const0_rtx);
r2 = gen_rtx_EXPR_LIST (inner, gen_rtx_REG (inner, regno + 1),
GEN_INT (inner_bytes));
return gen_rtx_PARALLEL (mode, gen_rtvec (2, r1, r2));
}
/* Return an rtx describing a return value of MODE as a PARALLEL
in N_ELTS registers, each of mode ELT_MODE, starting at REGNO,
stride REG_STRIDE. */
static rtx
rs6000_parallel_return (machine_mode mode,
int n_elts, machine_mode elt_mode,
unsigned int regno, unsigned int reg_stride)
{
rtx par = gen_rtx_PARALLEL (mode, rtvec_alloc (n_elts));
int i;
for (i = 0; i < n_elts; i++)
{
rtx r = gen_rtx_REG (elt_mode, regno);
rtx off = GEN_INT (i * GET_MODE_SIZE (elt_mode));
XVECEXP (par, 0, i) = gen_rtx_EXPR_LIST (VOIDmode, r, off);
regno += reg_stride;
}
return par;
}
/* Target hook for TARGET_FUNCTION_VALUE.
An integer value is in r3 and a floating-point value is in fp1,
unless -msoft-float. */
static rtx
rs6000_function_value (const_tree valtype,
const_tree fn_decl_or_type ATTRIBUTE_UNUSED,
bool outgoing ATTRIBUTE_UNUSED)
{
machine_mode mode;
unsigned int regno;
machine_mode elt_mode;
int n_elts;
/* Special handling for structs in darwin64. */
if (TARGET_MACHO
&& rs6000_darwin64_struct_check_p (TYPE_MODE (valtype), valtype))
{
CUMULATIVE_ARGS valcum;
rtx valret;
valcum.words = 0;
valcum.fregno = FP_ARG_MIN_REG;
valcum.vregno = ALTIVEC_ARG_MIN_REG;
/* Do a trial code generation as if this were going to be passed as
an argument; if any part goes in memory, we return NULL. */
valret = rs6000_darwin64_record_arg (&valcum, valtype, true, /* retval= */ true);
if (valret)
return valret;
/* Otherwise fall through to standard ABI rules. */
}
mode = TYPE_MODE (valtype);
/* The ELFv2 ABI returns homogeneous VFP aggregates in registers. */
if (rs6000_discover_homogeneous_aggregate (mode, valtype, &elt_mode, &n_elts))
{
int first_reg, n_regs;
if (SCALAR_FLOAT_MODE_NOT_VECTOR_P (elt_mode))
{
/* _Decimal128 must use even/odd register pairs. */
first_reg = (elt_mode == TDmode) ? FP_ARG_RETURN + 1 : FP_ARG_RETURN;
n_regs = (GET_MODE_SIZE (elt_mode) + 7) >> 3;
}
else
{
first_reg = ALTIVEC_ARG_RETURN;
n_regs = 1;
}
return rs6000_parallel_return (mode, n_elts, elt_mode, first_reg, n_regs);
}
/* Some return value types need be split in -mpowerpc64, 32bit ABI. */
if (TARGET_32BIT && TARGET_POWERPC64)
switch (mode)
{
default:
break;
case E_DImode:
case E_SCmode:
case E_DCmode:
case E_TCmode:
int count = GET_MODE_SIZE (mode) / 4;
return rs6000_parallel_return (mode, count, SImode, GP_ARG_RETURN, 1);
}
if ((INTEGRAL_TYPE_P (valtype)
&& GET_MODE_BITSIZE (mode) < (TARGET_32BIT ? 32 : 64))
|| POINTER_TYPE_P (valtype))
mode = TARGET_32BIT ? SImode : DImode;
if (DECIMAL_FLOAT_MODE_P (mode) && TARGET_HARD_FLOAT)
/* _Decimal128 must use an even/odd register pair. */
regno = (mode == TDmode) ? FP_ARG_RETURN + 1 : FP_ARG_RETURN;
else if (SCALAR_FLOAT_TYPE_P (valtype) && TARGET_HARD_FLOAT
&& !FLOAT128_VECTOR_P (mode)
&& ((TARGET_SINGLE_FLOAT && (mode == SFmode)) || TARGET_DOUBLE_FLOAT))
regno = FP_ARG_RETURN;
else if (TREE_CODE (valtype) == COMPLEX_TYPE
&& targetm.calls.split_complex_arg)
return rs6000_complex_function_value (mode);
/* VSX is a superset of Altivec and adds V2DImode/V2DFmode. Since the same
return register is used in both cases, and we won't see V2DImode/V2DFmode
for pure altivec, combine the two cases. */
else if ((TREE_CODE (valtype) == VECTOR_TYPE || FLOAT128_VECTOR_P (mode))
&& TARGET_ALTIVEC && TARGET_ALTIVEC_ABI
&& ALTIVEC_OR_VSX_VECTOR_MODE (mode))
regno = ALTIVEC_ARG_RETURN;
else
regno = GP_ARG_RETURN;
return gen_rtx_REG (mode, regno);
}
/* Define how to find the value returned by a library function
assuming the value has mode MODE. */
rtx
rs6000_libcall_value (machine_mode mode)
{
unsigned int regno;
/* Long long return value need be split in -mpowerpc64, 32bit ABI. */
if (TARGET_32BIT && TARGET_POWERPC64 && mode == DImode)
return rs6000_parallel_return (mode, 2, SImode, GP_ARG_RETURN, 1);
if (DECIMAL_FLOAT_MODE_P (mode) && TARGET_HARD_FLOAT)
/* _Decimal128 must use an even/odd register pair. */
regno = (mode == TDmode) ? FP_ARG_RETURN + 1 : FP_ARG_RETURN;
else if (SCALAR_FLOAT_MODE_NOT_VECTOR_P (mode)
&& TARGET_HARD_FLOAT
&& ((TARGET_SINGLE_FLOAT && mode == SFmode) || TARGET_DOUBLE_FLOAT))
regno = FP_ARG_RETURN;
/* VSX is a superset of Altivec and adds V2DImode/V2DFmode. Since the same
return register is used in both cases, and we won't see V2DImode/V2DFmode
for pure altivec, combine the two cases. */
else if (ALTIVEC_OR_VSX_VECTOR_MODE (mode)
&& TARGET_ALTIVEC && TARGET_ALTIVEC_ABI)
regno = ALTIVEC_ARG_RETURN;
else if (COMPLEX_MODE_P (mode) && targetm.calls.split_complex_arg)
return rs6000_complex_function_value (mode);
else
regno = GP_ARG_RETURN;
return gen_rtx_REG (mode, regno);
}
/* Compute register pressure classes. We implement the target hook to avoid
IRA picking something like NON_SPECIAL_REGS as a pressure class, which can
lead to incorrect estimates of number of available registers and therefor
increased register pressure/spill. */
static int
rs6000_compute_pressure_classes (enum reg_class *pressure_classes)
{
int n;
n = 0;
pressure_classes[n++] = GENERAL_REGS;
if (TARGET_VSX)
pressure_classes[n++] = VSX_REGS;
else
{
if (TARGET_ALTIVEC)
pressure_classes[n++] = ALTIVEC_REGS;
if (TARGET_HARD_FLOAT)
pressure_classes[n++] = FLOAT_REGS;
}
pressure_classes[n++] = CR_REGS;
pressure_classes[n++] = SPECIAL_REGS;
return n;
}
/* Given FROM and TO register numbers, say whether this elimination is allowed.
Frame pointer elimination is automatically handled.
For the RS/6000, if frame pointer elimination is being done, we would like
to convert ap into fp, not sp.
We need r30 if -mminimal-toc was specified, and there are constant pool
references. */
static bool
rs6000_can_eliminate (const int from, const int to)
{
return (from == ARG_POINTER_REGNUM && to == STACK_POINTER_REGNUM
? ! frame_pointer_needed
: from == RS6000_PIC_OFFSET_TABLE_REGNUM
? ! TARGET_MINIMAL_TOC || TARGET_NO_TOC
|| constant_pool_empty_p ()
: true);
}
/* Define the offset between two registers, FROM to be eliminated and its
replacement TO, at the start of a routine. */
HOST_WIDE_INT
rs6000_initial_elimination_offset (int from, int to)
{
rs6000_stack_t *info = rs6000_stack_info ();
HOST_WIDE_INT offset;
if (from == HARD_FRAME_POINTER_REGNUM && to == STACK_POINTER_REGNUM)
offset = info->push_p ? 0 : -info->total_size;
else if (from == FRAME_POINTER_REGNUM && to == STACK_POINTER_REGNUM)
{
offset = info->push_p ? 0 : -info->total_size;
if (FRAME_GROWS_DOWNWARD)
offset += info->fixed_size + info->vars_size + info->parm_size;
}
else if (from == FRAME_POINTER_REGNUM && to == HARD_FRAME_POINTER_REGNUM)
offset = FRAME_GROWS_DOWNWARD
? info->fixed_size + info->vars_size + info->parm_size
: 0;
else if (from == ARG_POINTER_REGNUM && to == HARD_FRAME_POINTER_REGNUM)
offset = info->total_size;
else if (from == ARG_POINTER_REGNUM && to == STACK_POINTER_REGNUM)
offset = info->push_p ? info->total_size : 0;
else if (from == RS6000_PIC_OFFSET_TABLE_REGNUM)
offset = 0;
else
gcc_unreachable ();
return offset;
}
/* Fill in sizes of registers used by unwinder. */
static void
rs6000_init_dwarf_reg_sizes_extra (tree address)
{
if (TARGET_MACHO && ! TARGET_ALTIVEC)
{
int i;
machine_mode mode = TYPE_MODE (char_type_node);
rtx addr = expand_expr (address, NULL_RTX, VOIDmode, EXPAND_NORMAL);
rtx mem = gen_rtx_MEM (BLKmode, addr);
rtx value = gen_int_mode (16, mode);
/* On Darwin, libgcc may be built to run on both G3 and G4/5.
The unwinder still needs to know the size of Altivec registers. */
for (i = FIRST_ALTIVEC_REGNO; i < LAST_ALTIVEC_REGNO+1; i++)
{
int column = DWARF_REG_TO_UNWIND_COLUMN
(DWARF2_FRAME_REG_OUT (DWARF_FRAME_REGNUM (i), true));
HOST_WIDE_INT offset = column * GET_MODE_SIZE (mode);
emit_move_insn (adjust_address (mem, mode, offset), value);
}
}
}
/* Map internal gcc register numbers to debug format register numbers.
FORMAT specifies the type of debug register number to use:
0 -- debug information, except for frame-related sections
1 -- DWARF .debug_frame section
2 -- DWARF .eh_frame section */
unsigned int
rs6000_dbx_register_number (unsigned int regno, unsigned int format)
{
/* Except for the above, we use the internal number for non-DWARF
debug information, and also for .eh_frame. */
if ((format == 0 && write_symbols != DWARF2_DEBUG) || format == 2)
return regno;
/* On some platforms, we use the standard DWARF register
numbering for .debug_info and .debug_frame. */
#ifdef RS6000_USE_DWARF_NUMBERING
if (regno <= 63)
return regno;
if (regno == LR_REGNO)
return 108;
if (regno == CTR_REGNO)
return 109;
/* Special handling for CR for .debug_frame: rs6000_emit_prologue has
translated any combination of CR2, CR3, CR4 saves to a save of CR2.
The actual code emitted saves the whole of CR, so we map CR2_REGNO
to the DWARF reg for CR. */
if (format == 1 && regno == CR2_REGNO)
return 64;
if (CR_REGNO_P (regno))
return regno - CR0_REGNO + 86;
if (regno == CA_REGNO)
return 101; /* XER */
if (ALTIVEC_REGNO_P (regno))
return regno - FIRST_ALTIVEC_REGNO + 1124;
if (regno == VRSAVE_REGNO)
return 356;
if (regno == VSCR_REGNO)
return 67;
#endif
return regno;
}
/* target hook eh_return_filter_mode */
static scalar_int_mode
rs6000_eh_return_filter_mode (void)
{
return TARGET_32BIT ? SImode : word_mode;
}
/* Target hook for scalar_mode_supported_p. */
static bool
rs6000_scalar_mode_supported_p (scalar_mode mode)
{
/* -m32 does not support TImode. This is the default, from
default_scalar_mode_supported_p. For -m32 -mpowerpc64 we want the
same ABI as for -m32. But default_scalar_mode_supported_p allows
integer modes of precision 2 * BITS_PER_WORD, which matches TImode
for -mpowerpc64. */
if (TARGET_32BIT && mode == TImode)
return false;
if (DECIMAL_FLOAT_MODE_P (mode))
return default_decimal_float_supported_p ();
else if (TARGET_FLOAT128_TYPE && (mode == KFmode || mode == IFmode))
return true;
else
return default_scalar_mode_supported_p (mode);
}
/* Target hook for vector_mode_supported_p. */
static bool
rs6000_vector_mode_supported_p (machine_mode mode)
{
if (TARGET_PAIRED_FLOAT && PAIRED_VECTOR_MODE (mode))
return true;
/* There is no vector form for IEEE 128-bit. If we return true for IEEE
128-bit, the compiler might try to widen IEEE 128-bit to IBM
double-double. */
else if (VECTOR_MEM_ALTIVEC_OR_VSX_P (mode) && !FLOAT128_IEEE_P (mode))
return true;
else
return false;
}
/* Target hook for floatn_mode. */
static opt_scalar_float_mode
rs6000_floatn_mode (int n, bool extended)
{
if (extended)
{
switch (n)
{
case 32:
return DFmode;
case 64:
if (TARGET_FLOAT128_TYPE)
return (FLOAT128_IEEE_P (TFmode)) ? TFmode : KFmode;
else
return opt_scalar_float_mode ();
case 128:
return opt_scalar_float_mode ();
default:
/* Those are the only valid _FloatNx types. */
gcc_unreachable ();
}
}
else
{
switch (n)
{
case 32:
return SFmode;
case 64:
return DFmode;
case 128:
if (TARGET_FLOAT128_TYPE)
return (FLOAT128_IEEE_P (TFmode)) ? TFmode : KFmode;
else
return opt_scalar_float_mode ();
default:
return opt_scalar_float_mode ();
}
}
}
/* Target hook for c_mode_for_suffix. */
static machine_mode
rs6000_c_mode_for_suffix (char suffix)
{
if (TARGET_FLOAT128_TYPE)
{
if (suffix == 'q' || suffix == 'Q')
return (FLOAT128_IEEE_P (TFmode)) ? TFmode : KFmode;
/* At the moment, we are not defining a suffix for IBM extended double.
If/when the default for -mabi=ieeelongdouble is changed, and we want
to support __ibm128 constants in legacy library code, we may need to
re-evalaute this decision. Currently, c-lex.c only supports 'w' and
'q' as machine dependent suffixes. The x86_64 port uses 'w' for
__float80 constants. */
}
return VOIDmode;
}
/* Target hook for invalid_arg_for_unprototyped_fn. */
static const char *
invalid_arg_for_unprototyped_fn (const_tree typelist, const_tree funcdecl, const_tree val)
{
return (!rs6000_darwin64_abi
&& typelist == 0
&& TREE_CODE (TREE_TYPE (val)) == VECTOR_TYPE
&& (funcdecl == NULL_TREE
|| (TREE_CODE (funcdecl) == FUNCTION_DECL
&& DECL_BUILT_IN_CLASS (funcdecl) != BUILT_IN_MD)))
? N_("AltiVec argument passed to unprototyped function")
: NULL;
}
/* For TARGET_SECURE_PLT 32-bit PIC code we can save PIC register
setup by using __stack_chk_fail_local hidden function instead of
calling __stack_chk_fail directly. Otherwise it is better to call
__stack_chk_fail directly. */
static tree ATTRIBUTE_UNUSED
rs6000_stack_protect_fail (void)
{
return (DEFAULT_ABI == ABI_V4 && TARGET_SECURE_PLT && flag_pic)
? default_hidden_stack_protect_fail ()
: default_external_stack_protect_fail ();
}
/* Implement the TARGET_ASAN_SHADOW_OFFSET hook. */
#if TARGET_ELF
static unsigned HOST_WIDE_INT
rs6000_asan_shadow_offset (void)
{
return (unsigned HOST_WIDE_INT) 1 << (TARGET_64BIT ? 41 : 29);
}
#endif
/* Mask options that we want to support inside of attribute((target)) and
#pragma GCC target operations. Note, we do not include things like
64/32-bit, endianness, hard/soft floating point, etc. that would have
different calling sequences. */
struct rs6000_opt_mask {
const char *name; /* option name */
HOST_WIDE_INT mask; /* mask to set */
bool invert; /* invert sense of mask */
bool valid_target; /* option is a target option */
};
static struct rs6000_opt_mask const rs6000_opt_masks[] =
{
{ "altivec", OPTION_MASK_ALTIVEC, false, true },
{ "cmpb", OPTION_MASK_CMPB, false, true },
{ "crypto", OPTION_MASK_CRYPTO, false, true },
{ "direct-move", OPTION_MASK_DIRECT_MOVE, false, true },
{ "dlmzb", OPTION_MASK_DLMZB, false, true },
{ "efficient-unaligned-vsx", OPTION_MASK_EFFICIENT_UNALIGNED_VSX,
false, true },
{ "float128", OPTION_MASK_FLOAT128_KEYWORD, false, true },
{ "float128-hardware", OPTION_MASK_FLOAT128_HW, false, true },
{ "fprnd", OPTION_MASK_FPRND, false, true },
{ "hard-dfp", OPTION_MASK_DFP, false, true },
{ "htm", OPTION_MASK_HTM, false, true },
{ "isel", OPTION_MASK_ISEL, false, true },
{ "mfcrf", OPTION_MASK_MFCRF, false, true },
{ "mfpgpr", OPTION_MASK_MFPGPR, false, true },
{ "modulo", OPTION_MASK_MODULO, false, true },
{ "mulhw", OPTION_MASK_MULHW, false, true },
{ "multiple", OPTION_MASK_MULTIPLE, false, true },
{ "popcntb", OPTION_MASK_POPCNTB, false, true },
{ "popcntd", OPTION_MASK_POPCNTD, false, true },
{ "power8-fusion", OPTION_MASK_P8_FUSION, false, true },
{ "power8-fusion-sign", OPTION_MASK_P8_FUSION_SIGN, false, true },
{ "power8-vector", OPTION_MASK_P8_VECTOR, false, true },
{ "power9-fusion", OPTION_MASK_P9_FUSION, false, true },
{ "power9-minmax", OPTION_MASK_P9_MINMAX, false, true },
{ "power9-misc", OPTION_MASK_P9_MISC, false, true },
{ "power9-vector", OPTION_MASK_P9_VECTOR, false, true },
{ "powerpc-gfxopt", OPTION_MASK_PPC_GFXOPT, false, true },
{ "powerpc-gpopt", OPTION_MASK_PPC_GPOPT, false, true },
{ "quad-memory", OPTION_MASK_QUAD_MEMORY, false, true },
{ "quad-memory-atomic", OPTION_MASK_QUAD_MEMORY_ATOMIC, false, true },
{ "recip-precision", OPTION_MASK_RECIP_PRECISION, false, true },
{ "save-toc-indirect", OPTION_MASK_SAVE_TOC_INDIRECT, false, true },
{ "string", OPTION_MASK_STRING, false, true },
{ "toc-fusion", OPTION_MASK_TOC_FUSION, false, true },
{ "update", OPTION_MASK_NO_UPDATE, true , true },
{ "vsx", OPTION_MASK_VSX, false, true },
#ifdef OPTION_MASK_64BIT
#if TARGET_AIX_OS
{ "aix64", OPTION_MASK_64BIT, false, false },
{ "aix32", OPTION_MASK_64BIT, true, false },
#else
{ "64", OPTION_MASK_64BIT, false, false },
{ "32", OPTION_MASK_64BIT, true, false },
#endif
#endif
#ifdef OPTION_MASK_EABI
{ "eabi", OPTION_MASK_EABI, false, false },
#endif
#ifdef OPTION_MASK_LITTLE_ENDIAN
{ "little", OPTION_MASK_LITTLE_ENDIAN, false, false },
{ "big", OPTION_MASK_LITTLE_ENDIAN, true, false },
#endif
#ifdef OPTION_MASK_RELOCATABLE
{ "relocatable", OPTION_MASK_RELOCATABLE, false, false },
#endif
#ifdef OPTION_MASK_STRICT_ALIGN
{ "strict-align", OPTION_MASK_STRICT_ALIGN, false, false },
#endif
{ "soft-float", OPTION_MASK_SOFT_FLOAT, false, false },
{ "string", OPTION_MASK_STRING, false, false },
};
/* Builtin mask mapping for printing the flags. */
static struct rs6000_opt_mask const rs6000_builtin_mask_names[] =
{
{ "altivec", RS6000_BTM_ALTIVEC, false, false },
{ "vsx", RS6000_BTM_VSX, false, false },
{ "paired", RS6000_BTM_PAIRED, false, false },
{ "fre", RS6000_BTM_FRE, false, false },
{ "fres", RS6000_BTM_FRES, false, false },
{ "frsqrte", RS6000_BTM_FRSQRTE, false, false },
{ "frsqrtes", RS6000_BTM_FRSQRTES, false, false },
{ "popcntd", RS6000_BTM_POPCNTD, false, false },
{ "cell", RS6000_BTM_CELL, false, false },
{ "power8-vector", RS6000_BTM_P8_VECTOR, false, false },
{ "power9-vector", RS6000_BTM_P9_VECTOR, false, false },
{ "power9-misc", RS6000_BTM_P9_MISC, false, false },
{ "crypto", RS6000_BTM_CRYPTO, false, false },
{ "htm", RS6000_BTM_HTM, false, false },
{ "hard-dfp", RS6000_BTM_DFP, false, false },
{ "hard-float", RS6000_BTM_HARD_FLOAT, false, false },
{ "long-double-128", RS6000_BTM_LDBL128, false, false },
{ "float128", RS6000_BTM_FLOAT128, false, false },
{ "float128-hw", RS6000_BTM_FLOAT128_HW,false, false },
};
/* Option variables that we want to support inside attribute((target)) and
#pragma GCC target operations. */
struct rs6000_opt_var {
const char *name; /* option name */
size_t global_offset; /* offset of the option in global_options. */
size_t target_offset; /* offset of the option in target options. */
};
static struct rs6000_opt_var const rs6000_opt_vars[] =
{
{ "friz",
offsetof (struct gcc_options, x_TARGET_FRIZ),
offsetof (struct cl_target_option, x_TARGET_FRIZ), },
{ "avoid-indexed-addresses",
offsetof (struct gcc_options, x_TARGET_AVOID_XFORM),
offsetof (struct cl_target_option, x_TARGET_AVOID_XFORM) },
{ "paired",
offsetof (struct gcc_options, x_rs6000_paired_float),
offsetof (struct cl_target_option, x_rs6000_paired_float), },
{ "longcall",
offsetof (struct gcc_options, x_rs6000_default_long_calls),
offsetof (struct cl_target_option, x_rs6000_default_long_calls), },
{ "optimize-swaps",
offsetof (struct gcc_options, x_rs6000_optimize_swaps),
offsetof (struct cl_target_option, x_rs6000_optimize_swaps), },
{ "allow-movmisalign",
offsetof (struct gcc_options, x_TARGET_ALLOW_MOVMISALIGN),
offsetof (struct cl_target_option, x_TARGET_ALLOW_MOVMISALIGN), },
{ "sched-groups",
offsetof (struct gcc_options, x_TARGET_SCHED_GROUPS),
offsetof (struct cl_target_option, x_TARGET_SCHED_GROUPS), },
{ "always-hint",
offsetof (struct gcc_options, x_TARGET_ALWAYS_HINT),
offsetof (struct cl_target_option, x_TARGET_ALWAYS_HINT), },
{ "align-branch-targets",
offsetof (struct gcc_options, x_TARGET_ALIGN_BRANCH_TARGETS),
offsetof (struct cl_target_option, x_TARGET_ALIGN_BRANCH_TARGETS), },
{ "tls-markers",
offsetof (struct gcc_options, x_tls_markers),
offsetof (struct cl_target_option, x_tls_markers), },
{ "sched-prolog",
offsetof (struct gcc_options, x_TARGET_SCHED_PROLOG),
offsetof (struct cl_target_option, x_TARGET_SCHED_PROLOG), },
{ "sched-epilog",
offsetof (struct gcc_options, x_TARGET_SCHED_PROLOG),
offsetof (struct cl_target_option, x_TARGET_SCHED_PROLOG), },
};
/* Inner function to handle attribute((target("..."))) and #pragma GCC target
parsing. Return true if there were no errors. */
static bool
rs6000_inner_target_options (tree args, bool attr_p)
{
bool ret = true;
if (args == NULL_TREE)
;
else if (TREE_CODE (args) == STRING_CST)
{
char *p = ASTRDUP (TREE_STRING_POINTER (args));
char *q;
while ((q = strtok (p, ",")) != NULL)
{
bool error_p = false;
bool not_valid_p = false;
const char *cpu_opt = NULL;
p = NULL;
if (strncmp (q, "cpu=", 4) == 0)
{
int cpu_index = rs6000_cpu_name_lookup (q+4);
if (cpu_index >= 0)
rs6000_cpu_index = cpu_index;
else
{
error_p = true;
cpu_opt = q+4;
}
}
else if (strncmp (q, "tune=", 5) == 0)
{
int tune_index = rs6000_cpu_name_lookup (q+5);
if (tune_index >= 0)
rs6000_tune_index = tune_index;
else
{
error_p = true;
cpu_opt = q+5;
}
}
else
{
size_t i;
bool invert = false;
char *r = q;
error_p = true;
if (strncmp (r, "no-", 3) == 0)
{
invert = true;
r += 3;
}
for (i = 0; i < ARRAY_SIZE (rs6000_opt_masks); i++)
if (strcmp (r, rs6000_opt_masks[i].name) == 0)
{
HOST_WIDE_INT mask = rs6000_opt_masks[i].mask;
if (!rs6000_opt_masks[i].valid_target)
not_valid_p = true;
else
{
error_p = false;
rs6000_isa_flags_explicit |= mask;
/* VSX needs altivec, so -mvsx automagically sets
altivec and disables -mavoid-indexed-addresses. */
if (!invert)
{
if (mask == OPTION_MASK_VSX)
{
mask |= OPTION_MASK_ALTIVEC;
TARGET_AVOID_XFORM = 0;
}
}
if (rs6000_opt_masks[i].invert)
invert = !invert;
if (invert)
rs6000_isa_flags &= ~mask;
else
rs6000_isa_flags |= mask;
}
break;
}
if (error_p && !not_valid_p)
{
for (i = 0; i < ARRAY_SIZE (rs6000_opt_vars); i++)
if (strcmp (r, rs6000_opt_vars[i].name) == 0)
{
size_t j = rs6000_opt_vars[i].global_offset;
*((int *) ((char *)&global_options + j)) = !invert;
error_p = false;
not_valid_p = false;
break;
}
}
}
if (error_p)
{
const char *eprefix, *esuffix;
ret = false;
if (attr_p)
{
eprefix = "__attribute__((__target__(";
esuffix = ")))";
}
else
{
eprefix = "#pragma GCC target ";
esuffix = "";
}
if (cpu_opt)
error ("invalid cpu %qs for %s%qs%s", cpu_opt, eprefix,
q, esuffix);
else if (not_valid_p)
error ("%s%qs%s is not allowed", eprefix, q, esuffix);
else
error ("%s%qs%s is invalid", eprefix, q, esuffix);
}
}
}
else if (TREE_CODE (args) == TREE_LIST)
{
do
{
tree value = TREE_VALUE (args);
if (value)
{
bool ret2 = rs6000_inner_target_options (value, attr_p);
if (!ret2)
ret = false;
}
args = TREE_CHAIN (args);
}
while (args != NULL_TREE);
}
else
{
error ("attribute %<target%> argument not a string");
return false;
}
return ret;
}
/* Print out the target options as a list for -mdebug=target. */
static void
rs6000_debug_target_options (tree args, const char *prefix)
{
if (args == NULL_TREE)
fprintf (stderr, "%s<NULL>", prefix);
else if (TREE_CODE (args) == STRING_CST)
{
char *p = ASTRDUP (TREE_STRING_POINTER (args));
char *q;
while ((q = strtok (p, ",")) != NULL)
{
p = NULL;
fprintf (stderr, "%s\"%s\"", prefix, q);
prefix = ", ";
}
}
else if (TREE_CODE (args) == TREE_LIST)
{
do
{
tree value = TREE_VALUE (args);
if (value)
{
rs6000_debug_target_options (value, prefix);
prefix = ", ";
}
args = TREE_CHAIN (args);
}
while (args != NULL_TREE);
}
else
gcc_unreachable ();
return;
}
/* Hook to validate attribute((target("..."))). */
static bool
rs6000_valid_attribute_p (tree fndecl,
tree ARG_UNUSED (name),
tree args,
int flags)
{
struct cl_target_option cur_target;
bool ret;
tree old_optimize;
tree new_target, new_optimize;
tree func_optimize;
gcc_assert ((fndecl != NULL_TREE) && (args != NULL_TREE));
if (TARGET_DEBUG_TARGET)
{
tree tname = DECL_NAME (fndecl);
fprintf (stderr, "\n==================== rs6000_valid_attribute_p:\n");
if (tname)
fprintf (stderr, "function: %.*s\n",
(int) IDENTIFIER_LENGTH (tname),
IDENTIFIER_POINTER (tname));
else
fprintf (stderr, "function: unknown\n");
fprintf (stderr, "args:");
rs6000_debug_target_options (args, " ");
fprintf (stderr, "\n");
if (flags)
fprintf (stderr, "flags: 0x%x\n", flags);
fprintf (stderr, "--------------------\n");
}
/* attribute((target("default"))) does nothing, beyond
affecting multi-versioning. */
if (TREE_VALUE (args)
&& TREE_CODE (TREE_VALUE (args)) == STRING_CST
&& TREE_CHAIN (args) == NULL_TREE
&& strcmp (TREE_STRING_POINTER (TREE_VALUE (args)), "default") == 0)
return true;
old_optimize = build_optimization_node (&global_options);
func_optimize = DECL_FUNCTION_SPECIFIC_OPTIMIZATION (fndecl);
/* If the function changed the optimization levels as well as setting target
options, start with the optimizations specified. */
if (func_optimize && func_optimize != old_optimize)
cl_optimization_restore (&global_options,
TREE_OPTIMIZATION (func_optimize));
/* The target attributes may also change some optimization flags, so update
the optimization options if necessary. */
cl_target_option_save (&cur_target, &global_options);
rs6000_cpu_index = rs6000_tune_index = -1;
ret = rs6000_inner_target_options (args, true);
/* Set up any additional state. */
if (ret)
{
ret = rs6000_option_override_internal (false);
new_target = build_target_option_node (&global_options);
}
else
new_target = NULL;
new_optimize = build_optimization_node (&global_options);
if (!new_target)
ret = false;
else if (fndecl)
{
DECL_FUNCTION_SPECIFIC_TARGET (fndecl) = new_target;
if (old_optimize != new_optimize)
DECL_FUNCTION_SPECIFIC_OPTIMIZATION (fndecl) = new_optimize;
}
cl_target_option_restore (&global_options, &cur_target);
if (old_optimize != new_optimize)
cl_optimization_restore (&global_options,
TREE_OPTIMIZATION (old_optimize));
return ret;
}
/* Hook to validate the current #pragma GCC target and set the state, and
update the macros based on what was changed. If ARGS is NULL, then
POP_TARGET is used to reset the options. */
bool
rs6000_pragma_target_parse (tree args, tree pop_target)
{
tree prev_tree = build_target_option_node (&global_options);
tree cur_tree;
struct cl_target_option *prev_opt, *cur_opt;
HOST_WIDE_INT prev_flags, cur_flags, diff_flags;
HOST_WIDE_INT prev_bumask, cur_bumask, diff_bumask;
if (TARGET_DEBUG_TARGET)
{
fprintf (stderr, "\n==================== rs6000_pragma_target_parse\n");
fprintf (stderr, "args:");
rs6000_debug_target_options (args, " ");
fprintf (stderr, "\n");
if (pop_target)
{
fprintf (stderr, "pop_target:\n");
debug_tree (pop_target);
}
else
fprintf (stderr, "pop_target: <NULL>\n");
fprintf (stderr, "--------------------\n");
}
if (! args)
{
cur_tree = ((pop_target)
? pop_target
: target_option_default_node);
cl_target_option_restore (&global_options,
TREE_TARGET_OPTION (cur_tree));
}
else
{
rs6000_cpu_index = rs6000_tune_index = -1;
if (!rs6000_inner_target_options (args, false)
|| !rs6000_option_override_internal (false)
|| (cur_tree = build_target_option_node (&global_options))
== NULL_TREE)
{
if (TARGET_DEBUG_BUILTIN || TARGET_DEBUG_TARGET)
fprintf (stderr, "invalid pragma\n");
return false;
}
}
target_option_current_node = cur_tree;
rs6000_activate_target_options (target_option_current_node);
/* If we have the preprocessor linked in (i.e. C or C++ languages), possibly
change the macros that are defined. */
if (rs6000_target_modify_macros_ptr)
{
prev_opt = TREE_TARGET_OPTION (prev_tree);
prev_bumask = prev_opt->x_rs6000_builtin_mask;
prev_flags = prev_opt->x_rs6000_isa_flags;
cur_opt = TREE_TARGET_OPTION (cur_tree);
cur_flags = cur_opt->x_rs6000_isa_flags;
cur_bumask = cur_opt->x_rs6000_builtin_mask;
diff_bumask = (prev_bumask ^ cur_bumask);
diff_flags = (prev_flags ^ cur_flags);
if ((diff_flags != 0) || (diff_bumask != 0))
{
/* Delete old macros. */
rs6000_target_modify_macros_ptr (false,
prev_flags & diff_flags,
prev_bumask & diff_bumask);
/* Define new macros. */
rs6000_target_modify_macros_ptr (true,
cur_flags & diff_flags,
cur_bumask & diff_bumask);
}
}
return true;
}
/* Remember the last target of rs6000_set_current_function. */
static GTY(()) tree rs6000_previous_fndecl;
/* Restore target's globals from NEW_TREE and invalidate the
rs6000_previous_fndecl cache. */
void
rs6000_activate_target_options (tree new_tree)
{
cl_target_option_restore (&global_options, TREE_TARGET_OPTION (new_tree));
if (TREE_TARGET_GLOBALS (new_tree))
restore_target_globals (TREE_TARGET_GLOBALS (new_tree));
else if (new_tree == target_option_default_node)
restore_target_globals (&default_target_globals);
else
TREE_TARGET_GLOBALS (new_tree) = save_target_globals_default_opts ();
rs6000_previous_fndecl = NULL_TREE;
}
/* Establish appropriate back-end context for processing the function
FNDECL. The argument might be NULL to indicate processing at top
level, outside of any function scope. */
static void
rs6000_set_current_function (tree fndecl)
{
if (TARGET_DEBUG_TARGET)
{
fprintf (stderr, "\n==================== rs6000_set_current_function");
if (fndecl)
fprintf (stderr, ", fndecl %s (%p)",
(DECL_NAME (fndecl)
? IDENTIFIER_POINTER (DECL_NAME (fndecl))
: "<unknown>"), (void *)fndecl);
if (rs6000_previous_fndecl)
fprintf (stderr, ", prev_fndecl (%p)", (void *)rs6000_previous_fndecl);
fprintf (stderr, "\n");
}
/* Only change the context if the function changes. This hook is called
several times in the course of compiling a function, and we don't want to
slow things down too much or call target_reinit when it isn't safe. */
if (fndecl == rs6000_previous_fndecl)
return;
tree old_tree;
if (rs6000_previous_fndecl == NULL_TREE)
old_tree = target_option_current_node;
else if (DECL_FUNCTION_SPECIFIC_TARGET (rs6000_previous_fndecl))
old_tree = DECL_FUNCTION_SPECIFIC_TARGET (rs6000_previous_fndecl);
else
old_tree = target_option_default_node;
tree new_tree;
if (fndecl == NULL_TREE)
{
if (old_tree != target_option_current_node)
new_tree = target_option_current_node;
else
new_tree = NULL_TREE;
}
else
{
new_tree = DECL_FUNCTION_SPECIFIC_TARGET (fndecl);
if (new_tree == NULL_TREE)
new_tree = target_option_default_node;
}
if (TARGET_DEBUG_TARGET)
{
if (new_tree)
{
fprintf (stderr, "\nnew fndecl target specific options:\n");
debug_tree (new_tree);
}
if (old_tree)
{
fprintf (stderr, "\nold fndecl target specific options:\n");
debug_tree (old_tree);
}
if (old_tree != NULL_TREE || new_tree != NULL_TREE)
fprintf (stderr, "--------------------\n");
}
if (new_tree && old_tree != new_tree)
rs6000_activate_target_options (new_tree);
if (fndecl)
rs6000_previous_fndecl = fndecl;
}
/* Save the current options */
static void
rs6000_function_specific_save (struct cl_target_option *ptr,
struct gcc_options *opts)
{
ptr->x_rs6000_isa_flags = opts->x_rs6000_isa_flags;
ptr->x_rs6000_isa_flags_explicit = opts->x_rs6000_isa_flags_explicit;
}
/* Restore the current options */
static void
rs6000_function_specific_restore (struct gcc_options *opts,
struct cl_target_option *ptr)
{
opts->x_rs6000_isa_flags = ptr->x_rs6000_isa_flags;
opts->x_rs6000_isa_flags_explicit = ptr->x_rs6000_isa_flags_explicit;
(void) rs6000_option_override_internal (false);
}
/* Print the current options */
static void
rs6000_function_specific_print (FILE *file, int indent,
struct cl_target_option *ptr)
{
rs6000_print_isa_options (file, indent, "Isa options set",
ptr->x_rs6000_isa_flags);
rs6000_print_isa_options (file, indent, "Isa options explicit",
ptr->x_rs6000_isa_flags_explicit);
}
/* Helper function to print the current isa or misc options on a line. */
static void
rs6000_print_options_internal (FILE *file,
int indent,
const char *string,
HOST_WIDE_INT flags,
const char *prefix,
const struct rs6000_opt_mask *opts,
size_t num_elements)
{
size_t i;
size_t start_column = 0;
size_t cur_column;
size_t max_column = 120;
size_t prefix_len = strlen (prefix);
size_t comma_len = 0;
const char *comma = "";
if (indent)
start_column += fprintf (file, "%*s", indent, "");
if (!flags)
{
fprintf (stderr, DEBUG_FMT_S, string, "<none>");
return;
}
start_column += fprintf (stderr, DEBUG_FMT_WX, string, flags);
/* Print the various mask options. */
cur_column = start_column;
for (i = 0; i < num_elements; i++)
{
bool invert = opts[i].invert;
const char *name = opts[i].name;
const char *no_str = "";
HOST_WIDE_INT mask = opts[i].mask;
size_t len = comma_len + prefix_len + strlen (name);
if (!invert)
{
if ((flags & mask) == 0)
{
no_str = "no-";
len += sizeof ("no-") - 1;
}
flags &= ~mask;
}
else
{
if ((flags & mask) != 0)
{
no_str = "no-";
len += sizeof ("no-") - 1;
}
flags |= mask;
}
cur_column += len;
if (cur_column > max_column)
{
fprintf (stderr, ", \\\n%*s", (int)start_column, "");
cur_column = start_column + len;
comma = "";
}
fprintf (file, "%s%s%s%s", comma, prefix, no_str, name);
comma = ", ";
comma_len = sizeof (", ") - 1;
}
fputs ("\n", file);
}
/* Helper function to print the current isa options on a line. */
static void
rs6000_print_isa_options (FILE *file, int indent, const char *string,
HOST_WIDE_INT flags)
{
rs6000_print_options_internal (file, indent, string, flags, "-m",
&rs6000_opt_masks[0],
ARRAY_SIZE (rs6000_opt_masks));
}
static void
rs6000_print_builtin_options (FILE *file, int indent, const char *string,
HOST_WIDE_INT flags)
{
rs6000_print_options_internal (file, indent, string, flags, "",
&rs6000_builtin_mask_names[0],
ARRAY_SIZE (rs6000_builtin_mask_names));
}
/* If the user used -mno-vsx, we need turn off all of the implicit ISA 2.06,
2.07, and 3.0 options that relate to the vector unit (-mdirect-move,
-mupper-regs-df, etc.).
If the user used -mno-power8-vector, we need to turn off all of the implicit
ISA 2.07 and 3.0 options that relate to the vector unit.
If the user used -mno-power9-vector, we need to turn off all of the implicit
ISA 3.0 options that relate to the vector unit.
This function does not handle explicit options such as the user specifying
-mdirect-move. These are handled in rs6000_option_override_internal, and
the appropriate error is given if needed.
We return a mask of all of the implicit options that should not be enabled
by default. */
static HOST_WIDE_INT
rs6000_disable_incompatible_switches (void)
{
HOST_WIDE_INT ignore_masks = rs6000_isa_flags_explicit;
size_t i, j;
static const struct {
const HOST_WIDE_INT no_flag; /* flag explicitly turned off. */
const HOST_WIDE_INT dep_flags; /* flags that depend on this option. */
const char *const name; /* name of the switch. */
} flags[] = {
{ OPTION_MASK_P9_VECTOR, OTHER_P9_VECTOR_MASKS, "power9-vector" },
{ OPTION_MASK_P8_VECTOR, OTHER_P8_VECTOR_MASKS, "power8-vector" },
{ OPTION_MASK_VSX, OTHER_VSX_VECTOR_MASKS, "vsx" },
};
for (i = 0; i < ARRAY_SIZE (flags); i++)
{
HOST_WIDE_INT no_flag = flags[i].no_flag;
if ((rs6000_isa_flags & no_flag) == 0
&& (rs6000_isa_flags_explicit & no_flag) != 0)
{
HOST_WIDE_INT dep_flags = flags[i].dep_flags;
HOST_WIDE_INT set_flags = (rs6000_isa_flags_explicit
& rs6000_isa_flags
& dep_flags);
if (set_flags)
{
for (j = 0; j < ARRAY_SIZE (rs6000_opt_masks); j++)
if ((set_flags & rs6000_opt_masks[j].mask) != 0)
{
set_flags &= ~rs6000_opt_masks[j].mask;
error ("%<-mno-%s%> turns off %<-m%s%>",
flags[i].name,
rs6000_opt_masks[j].name);
}
gcc_assert (!set_flags);
}
rs6000_isa_flags &= ~dep_flags;
ignore_masks |= no_flag | dep_flags;
}
}
return ignore_masks;
}
/* Helper function for printing the function name when debugging. */
static const char *
get_decl_name (tree fn)
{
tree name;
if (!fn)
return "<null>";
name = DECL_NAME (fn);
if (!name)
return "<no-name>";
return IDENTIFIER_POINTER (name);
}
/* Return the clone id of the target we are compiling code for in a target
clone. The clone id is ordered from 0 (default) to CLONE_MAX-1 and gives
the priority list for the target clones (ordered from lowest to
highest). */
static int
rs6000_clone_priority (tree fndecl)
{
tree fn_opts = DECL_FUNCTION_SPECIFIC_TARGET (fndecl);
HOST_WIDE_INT isa_masks;
int ret = CLONE_DEFAULT;
tree attrs = lookup_attribute ("target", DECL_ATTRIBUTES (fndecl));
const char *attrs_str = NULL;
attrs = TREE_VALUE (TREE_VALUE (attrs));
attrs_str = TREE_STRING_POINTER (attrs);
/* Return priority zero for default function. Return the ISA needed for the
function if it is not the default. */
if (strcmp (attrs_str, "default") != 0)
{
if (fn_opts == NULL_TREE)
fn_opts = target_option_default_node;
if (!fn_opts || !TREE_TARGET_OPTION (fn_opts))
isa_masks = rs6000_isa_flags;
else
isa_masks = TREE_TARGET_OPTION (fn_opts)->x_rs6000_isa_flags;
for (ret = CLONE_MAX - 1; ret != 0; ret--)
if ((rs6000_clone_map[ret].isa_mask & isa_masks) != 0)
break;
}
if (TARGET_DEBUG_TARGET)
fprintf (stderr, "rs6000_get_function_version_priority (%s) => %d\n",
get_decl_name (fndecl), ret);
return ret;
}
/* This compares the priority of target features in function DECL1 and DECL2.
It returns positive value if DECL1 is higher priority, negative value if
DECL2 is higher priority and 0 if they are the same. Note, priorities are
ordered from lowest (CLONE_DEFAULT) to highest (currently CLONE_ISA_3_0). */
static int
rs6000_compare_version_priority (tree decl1, tree decl2)
{
int priority1 = rs6000_clone_priority (decl1);
int priority2 = rs6000_clone_priority (decl2);
int ret = priority1 - priority2;
if (TARGET_DEBUG_TARGET)
fprintf (stderr, "rs6000_compare_version_priority (%s, %s) => %d\n",
get_decl_name (decl1), get_decl_name (decl2), ret);
return ret;
}
/* Make a dispatcher declaration for the multi-versioned function DECL.
Calls to DECL function will be replaced with calls to the dispatcher
by the front-end. Returns the decl of the dispatcher function. */
static tree
rs6000_get_function_versions_dispatcher (void *decl)
{
tree fn = (tree) decl;
struct cgraph_node *node = NULL;
struct cgraph_node *default_node = NULL;
struct cgraph_function_version_info *node_v = NULL;
struct cgraph_function_version_info *first_v = NULL;
tree dispatch_decl = NULL;
struct cgraph_function_version_info *default_version_info = NULL;
gcc_assert (fn != NULL && DECL_FUNCTION_VERSIONED (fn));
if (TARGET_DEBUG_TARGET)
fprintf (stderr, "rs6000_get_function_versions_dispatcher (%s)\n",
get_decl_name (fn));
node = cgraph_node::get (fn);
gcc_assert (node != NULL);
node_v = node->function_version ();
gcc_assert (node_v != NULL);
if (node_v->dispatcher_resolver != NULL)
return node_v->dispatcher_resolver;
/* Find the default version and make it the first node. */
first_v = node_v;
/* Go to the beginning of the chain. */
while (first_v->prev != NULL)
first_v = first_v->prev;
default_version_info = first_v;
while (default_version_info != NULL)
{
const tree decl2 = default_version_info->this_node->decl;
if (is_function_default_version (decl2))
break;
default_version_info = default_version_info->next;
}
/* If there is no default node, just return NULL. */
if (default_version_info == NULL)
return NULL;
/* Make default info the first node. */
if (first_v != default_version_info)
{
default_version_info->prev->next = default_version_info->next;
if (default_version_info->next)
default_version_info->next->prev = default_version_info->prev;
first_v->prev = default_version_info;
default_version_info->next = first_v;
default_version_info->prev = NULL;
}
default_node = default_version_info->this_node;
#ifndef TARGET_LIBC_PROVIDES_HWCAP_IN_TCB
error_at (DECL_SOURCE_LOCATION (default_node->decl),
"target_clones attribute needs GLIBC (2.23 and newer) that "
"exports hardware capability bits");
#else
if (targetm.has_ifunc_p ())
{
struct cgraph_function_version_info *it_v = NULL;
struct cgraph_node *dispatcher_node = NULL;
struct cgraph_function_version_info *dispatcher_version_info = NULL;
/* Right now, the dispatching is done via ifunc. */
dispatch_decl = make_dispatcher_decl (default_node->decl);
dispatcher_node = cgraph_node::get_create (dispatch_decl);
gcc_assert (dispatcher_node != NULL);
dispatcher_node->dispatcher_function = 1;
dispatcher_version_info
= dispatcher_node->insert_new_function_version ();
dispatcher_version_info->next = default_version_info;
dispatcher_node->definition = 1;
/* Set the dispatcher for all the versions. */
it_v = default_version_info;
while (it_v != NULL)
{
it_v->dispatcher_resolver = dispatch_decl;
it_v = it_v->next;
}
}
else
{
error_at (DECL_SOURCE_LOCATION (default_node->decl),
"multiversioning needs ifunc which is not supported "
"on this target");
}
#endif
return dispatch_decl;
}
/* Make the resolver function decl to dispatch the versions of a multi-
versioned function, DEFAULT_DECL. Create an empty basic block in the
resolver and store the pointer in EMPTY_BB. Return the decl of the resolver
function. */
static tree
make_resolver_func (const tree default_decl,
const tree dispatch_decl,
basic_block *empty_bb)
{
/* Make the resolver function static. The resolver function returns
void *. */
tree decl_name = clone_function_name (default_decl, "resolver");
const char *resolver_name = IDENTIFIER_POINTER (decl_name);
tree type = build_function_type_list (ptr_type_node, NULL_TREE);
tree decl = build_fn_decl (resolver_name, type);
SET_DECL_ASSEMBLER_NAME (decl, decl_name);
DECL_NAME (decl) = decl_name;
TREE_USED (decl) = 1;
DECL_ARTIFICIAL (decl) = 1;
DECL_IGNORED_P (decl) = 0;
TREE_PUBLIC (decl) = 0;
DECL_UNINLINABLE (decl) = 1;
/* Resolver is not external, body is generated. */
DECL_EXTERNAL (decl) = 0;
DECL_EXTERNAL (dispatch_decl) = 0;
DECL_CONTEXT (decl) = NULL_TREE;
DECL_INITIAL (decl) = make_node (BLOCK);
DECL_STATIC_CONSTRUCTOR (decl) = 0;
/* Build result decl and add to function_decl. */
tree t = build_decl (UNKNOWN_LOCATION, RESULT_DECL, NULL_TREE, ptr_type_node);
DECL_ARTIFICIAL (t) = 1;
DECL_IGNORED_P (t) = 1;
DECL_RESULT (decl) = t;
gimplify_function_tree (decl);
push_cfun (DECL_STRUCT_FUNCTION (decl));
*empty_bb = init_lowered_empty_function (decl, false,
profile_count::uninitialized ());
cgraph_node::add_new_function (decl, true);
symtab->call_cgraph_insertion_hooks (cgraph_node::get_create (decl));
pop_cfun ();
/* Mark dispatch_decl as "ifunc" with resolver as resolver_name. */
DECL_ATTRIBUTES (dispatch_decl)
= make_attribute ("ifunc", resolver_name, DECL_ATTRIBUTES (dispatch_decl));
cgraph_node::create_same_body_alias (dispatch_decl, decl);
return decl;
}
/* This adds a condition to the basic_block NEW_BB in function FUNCTION_DECL to
return a pointer to VERSION_DECL if we are running on a machine that
supports the index CLONE_ISA hardware architecture bits. This function will
be called during version dispatch to decide which function version to
execute. It returns the basic block at the end, to which more conditions
can be added. */
static basic_block
add_condition_to_bb (tree function_decl, tree version_decl,
int clone_isa, basic_block new_bb)
{
push_cfun (DECL_STRUCT_FUNCTION (function_decl));
gcc_assert (new_bb != NULL);
gimple_seq gseq = bb_seq (new_bb);
tree convert_expr = build1 (CONVERT_EXPR, ptr_type_node,
build_fold_addr_expr (version_decl));
tree result_var = create_tmp_var (ptr_type_node);
gimple *convert_stmt = gimple_build_assign (result_var, convert_expr);
gimple *return_stmt = gimple_build_return (result_var);
if (clone_isa == CLONE_DEFAULT)
{
gimple_seq_add_stmt (&gseq, convert_stmt);
gimple_seq_add_stmt (&gseq, return_stmt);
set_bb_seq (new_bb, gseq);
gimple_set_bb (convert_stmt, new_bb);
gimple_set_bb (return_stmt, new_bb);
pop_cfun ();
return new_bb;
}
tree bool_zero = build_int_cst (bool_int_type_node, 0);
tree cond_var = create_tmp_var (bool_int_type_node);
tree predicate_decl = rs6000_builtin_decls [(int) RS6000_BUILTIN_CPU_SUPPORTS];
const char *arg_str = rs6000_clone_map[clone_isa].name;
tree predicate_arg = build_string_literal (strlen (arg_str) + 1, arg_str);
gimple *call_cond_stmt = gimple_build_call (predicate_decl, 1, predicate_arg);
gimple_call_set_lhs (call_cond_stmt, cond_var);
gimple_set_block (call_cond_stmt, DECL_INITIAL (function_decl));
gimple_set_bb (call_cond_stmt, new_bb);
gimple_seq_add_stmt (&gseq, call_cond_stmt);
gimple *if_else_stmt = gimple_build_cond (NE_EXPR, cond_var, bool_zero,
NULL_TREE, NULL_TREE);
gimple_set_block (if_else_stmt, DECL_INITIAL (function_decl));
gimple_set_bb (if_else_stmt, new_bb);
gimple_seq_add_stmt (&gseq, if_else_stmt);
gimple_seq_add_stmt (&gseq, convert_stmt);
gimple_seq_add_stmt (&gseq, return_stmt);
set_bb_seq (new_bb, gseq);
basic_block bb1 = new_bb;
edge e12 = split_block (bb1, if_else_stmt);
basic_block bb2 = e12->dest;
e12->flags &= ~EDGE_FALLTHRU;
e12->flags |= EDGE_TRUE_VALUE;
edge e23 = split_block (bb2, return_stmt);
gimple_set_bb (convert_stmt, bb2);
gimple_set_bb (return_stmt, bb2);
basic_block bb3 = e23->dest;
make_edge (bb1, bb3, EDGE_FALSE_VALUE);
remove_edge (e23);
make_edge (bb2, EXIT_BLOCK_PTR_FOR_FN (cfun), 0);
pop_cfun ();
return bb3;
}
/* This function generates the dispatch function for multi-versioned functions.
DISPATCH_DECL is the function which will contain the dispatch logic.
FNDECLS are the function choices for dispatch, and is a tree chain.
EMPTY_BB is the basic block pointer in DISPATCH_DECL in which the dispatch
code is generated. */
static int
dispatch_function_versions (tree dispatch_decl,
void *fndecls_p,
basic_block *empty_bb)
{
int ix;
tree ele;
vec<tree> *fndecls;
tree clones[CLONE_MAX];
if (TARGET_DEBUG_TARGET)
fputs ("dispatch_function_versions, top\n", stderr);
gcc_assert (dispatch_decl != NULL
&& fndecls_p != NULL
&& empty_bb != NULL);
/* fndecls_p is actually a vector. */
fndecls = static_cast<vec<tree> *> (fndecls_p);
/* At least one more version other than the default. */
gcc_assert (fndecls->length () >= 2);
/* The first version in the vector is the default decl. */
memset ((void *) clones, '\0', sizeof (clones));
clones[CLONE_DEFAULT] = (*fndecls)[0];
/* On the PowerPC, we do not need to call __builtin_cpu_init, which is a NOP
on the PowerPC (on the x86_64, it is not a NOP). The builtin function
__builtin_cpu_support ensures that the TOC fields are setup by requiring a
recent glibc. If we ever need to call __builtin_cpu_init, we would need
to insert the code here to do the call. */
for (ix = 1; fndecls->iterate (ix, &ele); ++ix)
{
int priority = rs6000_clone_priority (ele);
if (!clones[priority])
clones[priority] = ele;
}
for (ix = CLONE_MAX - 1; ix >= 0; ix--)
if (clones[ix])
{
if (TARGET_DEBUG_TARGET)
fprintf (stderr, "dispatch_function_versions, clone %d, %s\n",
ix, get_decl_name (clones[ix]));
*empty_bb = add_condition_to_bb (dispatch_decl, clones[ix], ix,
*empty_bb);
}
return 0;
}
/* Generate the dispatching code body to dispatch multi-versioned function
DECL. The target hook is called to process the "target" attributes and
provide the code to dispatch the right function at run-time. NODE points
to the dispatcher decl whose body will be created. */
static tree
rs6000_generate_version_dispatcher_body (void *node_p)
{
tree resolver;
basic_block empty_bb;
struct cgraph_node *node = (cgraph_node *) node_p;
struct cgraph_function_version_info *ninfo = node->function_version ();
if (ninfo->dispatcher_resolver)
return ninfo->dispatcher_resolver;
/* node is going to be an alias, so remove the finalized bit. */
node->definition = false;
/* The first version in the chain corresponds to the default version. */
ninfo->dispatcher_resolver = resolver
= make_resolver_func (ninfo->next->this_node->decl, node->decl, &empty_bb);
if (TARGET_DEBUG_TARGET)
fprintf (stderr, "rs6000_get_function_versions_dispatcher, %s\n",
get_decl_name (resolver));
push_cfun (DECL_STRUCT_FUNCTION (resolver));
auto_vec<tree, 2> fn_ver_vec;
for (struct cgraph_function_version_info *vinfo = ninfo->next;
vinfo;
vinfo = vinfo->next)
{
struct cgraph_node *version = vinfo->this_node;
/* Check for virtual functions here again, as by this time it should
have been determined if this function needs a vtable index or
not. This happens for methods in derived classes that override
virtual methods in base classes but are not explicitly marked as
virtual. */
if (DECL_VINDEX (version->decl))
sorry ("Virtual function multiversioning not supported");
fn_ver_vec.safe_push (version->decl);
}
dispatch_function_versions (resolver, &fn_ver_vec, &empty_bb);
cgraph_edge::rebuild_edges ();
pop_cfun ();
return resolver;
}
/* Hook to determine if one function can safely inline another. */
static bool
rs6000_can_inline_p (tree caller, tree callee)
{
bool ret = false;
tree caller_tree = DECL_FUNCTION_SPECIFIC_TARGET (caller);
tree callee_tree = DECL_FUNCTION_SPECIFIC_TARGET (callee);
/* If callee has no option attributes, then it is ok to inline. */
if (!callee_tree)
ret = true;
/* If caller has no option attributes, but callee does then it is not ok to
inline. */
else if (!caller_tree)
ret = false;
else
{
struct cl_target_option *caller_opts = TREE_TARGET_OPTION (caller_tree);
struct cl_target_option *callee_opts = TREE_TARGET_OPTION (callee_tree);
/* Callee's options should a subset of the caller's, i.e. a vsx function
can inline an altivec function but a non-vsx function can't inline a
vsx function. */
if ((caller_opts->x_rs6000_isa_flags & callee_opts->x_rs6000_isa_flags)
== callee_opts->x_rs6000_isa_flags)
ret = true;
}
if (TARGET_DEBUG_TARGET)
fprintf (stderr, "rs6000_can_inline_p:, caller %s, callee %s, %s inline\n",
get_decl_name (caller), get_decl_name (callee),
(ret ? "can" : "cannot"));
return ret;
}
/* Allocate a stack temp and fixup the address so it meets the particular
memory requirements (either offetable or REG+REG addressing). */
rtx
rs6000_allocate_stack_temp (machine_mode mode,
bool offsettable_p,
bool reg_reg_p)
{
rtx stack = assign_stack_temp (mode, GET_MODE_SIZE (mode));
rtx addr = XEXP (stack, 0);
int strict_p = reload_completed;
if (!legitimate_indirect_address_p (addr, strict_p))
{
if (offsettable_p
&& !rs6000_legitimate_offset_address_p (mode, addr, strict_p, true))
stack = replace_equiv_address (stack, copy_addr_to_reg (addr));
else if (reg_reg_p && !legitimate_indexed_address_p (addr, strict_p))
stack = replace_equiv_address (stack, copy_addr_to_reg (addr));
}
return stack;
}
/* Given a memory reference, if it is not a reg or reg+reg addressing, convert
to such a form to deal with memory reference instructions like STFIWX that
only take reg+reg addressing. */
rtx
rs6000_address_for_fpconvert (rtx x)
{
rtx addr;
gcc_assert (MEM_P (x));
addr = XEXP (x, 0);
if (can_create_pseudo_p ()
&& ! legitimate_indirect_address_p (addr, reload_completed)
&& ! legitimate_indexed_address_p (addr, reload_completed))
{
if (GET_CODE (addr) == PRE_INC || GET_CODE (addr) == PRE_DEC)
{
rtx reg = XEXP (addr, 0);
HOST_WIDE_INT size = GET_MODE_SIZE (GET_MODE (x));
rtx size_rtx = GEN_INT ((GET_CODE (addr) == PRE_DEC) ? -size : size);
gcc_assert (REG_P (reg));
emit_insn (gen_add3_insn (reg, reg, size_rtx));
addr = reg;
}
else if (GET_CODE (addr) == PRE_MODIFY)
{
rtx reg = XEXP (addr, 0);
rtx expr = XEXP (addr, 1);
gcc_assert (REG_P (reg));
gcc_assert (GET_CODE (expr) == PLUS);
emit_insn (gen_add3_insn (reg, XEXP (expr, 0), XEXP (expr, 1)));
addr = reg;
}
x = replace_equiv_address (x, copy_addr_to_reg (addr));
}
return x;
}
/* Given a memory reference, if it is not in the form for altivec memory
reference instructions (i.e. reg or reg+reg addressing with AND of -16),
convert to the altivec format. */
rtx
rs6000_address_for_altivec (rtx x)
{
gcc_assert (MEM_P (x));
if (!altivec_indexed_or_indirect_operand (x, GET_MODE (x)))
{
rtx addr = XEXP (x, 0);
if (!legitimate_indexed_address_p (addr, reload_completed)
&& !legitimate_indirect_address_p (addr, reload_completed))
addr = copy_to_mode_reg (Pmode, addr);
addr = gen_rtx_AND (Pmode, addr, GEN_INT (-16));
x = change_address (x, GET_MODE (x), addr);
}
return x;
}
/* Implement TARGET_LEGITIMATE_CONSTANT_P.
On the RS/6000, all integer constants are acceptable, most won't be valid
for particular insns, though. Only easy FP constants are acceptable. */
static bool
rs6000_legitimate_constant_p (machine_mode mode, rtx x)
{
if (TARGET_ELF && tls_referenced_p (x))
return false;
return ((GET_CODE (x) != CONST_DOUBLE && GET_CODE (x) != CONST_VECTOR)
|| GET_MODE (x) == VOIDmode
|| (TARGET_POWERPC64 && mode == DImode)
|| easy_fp_constant (x, mode)
|| easy_vector_constant (x, mode));
}
/* Return TRUE iff the sequence ending in LAST sets the static chain. */
static bool
chain_already_loaded (rtx_insn *last)
{
for (; last != NULL; last = PREV_INSN (last))
{
if (NONJUMP_INSN_P (last))
{
rtx patt = PATTERN (last);
if (GET_CODE (patt) == SET)
{
rtx lhs = XEXP (patt, 0);
if (REG_P (lhs) && REGNO (lhs) == STATIC_CHAIN_REGNUM)
return true;
}
}
}
return false;
}
/* Expand code to perform a call under the AIX or ELFv2 ABI. */
void
rs6000_call_aix (rtx value, rtx func_desc, rtx flag, rtx cookie)
{
const bool direct_call_p
= GET_CODE (func_desc) == SYMBOL_REF && SYMBOL_REF_FUNCTION_P (func_desc);
rtx toc_reg = gen_rtx_REG (Pmode, TOC_REGNUM);
rtx toc_load = NULL_RTX;
rtx toc_restore = NULL_RTX;
rtx func_addr;
rtx abi_reg = NULL_RTX;
rtx call[4];
int n_call;
rtx insn;
/* Handle longcall attributes. */
if (INTVAL (cookie) & CALL_LONG)
func_desc = rs6000_longcall_ref (func_desc);
/* Handle indirect calls. */
if (GET_CODE (func_desc) != SYMBOL_REF
|| (DEFAULT_ABI == ABI_AIX && !SYMBOL_REF_FUNCTION_P (func_desc)))
{
/* Save the TOC into its reserved slot before the call,
and prepare to restore it after the call. */
rtx stack_ptr = gen_rtx_REG (Pmode, STACK_POINTER_REGNUM);
rtx stack_toc_offset = GEN_INT (RS6000_TOC_SAVE_SLOT);
rtx stack_toc_mem = gen_frame_mem (Pmode,
gen_rtx_PLUS (Pmode, stack_ptr,
stack_toc_offset));
rtx stack_toc_unspec = gen_rtx_UNSPEC (Pmode,
gen_rtvec (1, stack_toc_offset),
UNSPEC_TOCSLOT);
toc_restore = gen_rtx_SET (toc_reg, stack_toc_unspec);
/* Can we optimize saving the TOC in the prologue or
do we need to do it at every call? */
if (TARGET_SAVE_TOC_INDIRECT && !cfun->calls_alloca)
cfun->machine->save_toc_in_prologue = true;
else
{
MEM_VOLATILE_P (stack_toc_mem) = 1;
emit_move_insn (stack_toc_mem, toc_reg);
}
if (DEFAULT_ABI == ABI_ELFv2)
{
/* A function pointer in the ELFv2 ABI is just a plain address, but
the ABI requires it to be loaded into r12 before the call. */
func_addr = gen_rtx_REG (Pmode, 12);
emit_move_insn (func_addr, func_desc);
abi_reg = func_addr;
}
else
{
/* A function pointer under AIX is a pointer to a data area whose
first word contains the actual address of the function, whose
second word contains a pointer to its TOC, and whose third word
contains a value to place in the static chain register (r11).
Note that if we load the static chain, our "trampoline" need
not have any executable code. */
/* Load up address of the actual function. */
func_desc = force_reg (Pmode, func_desc);
func_addr = gen_reg_rtx (Pmode);
emit_move_insn (func_addr, gen_rtx_MEM (Pmode, func_desc));
/* Prepare to load the TOC of the called function. Note that the
TOC load must happen immediately before the actual call so
that unwinding the TOC registers works correctly. See the
comment in frob_update_context. */
rtx func_toc_offset = GEN_INT (GET_MODE_SIZE (Pmode));
rtx func_toc_mem = gen_rtx_MEM (Pmode,
gen_rtx_PLUS (Pmode, func_desc,
func_toc_offset));
toc_load = gen_rtx_USE (VOIDmode, func_toc_mem);
/* If we have a static chain, load it up. But, if the call was
originally direct, the 3rd word has not been written since no
trampoline has been built, so we ought not to load it, lest we
override a static chain value. */
if (!direct_call_p
&& TARGET_POINTERS_TO_NESTED_FUNCTIONS
&& !chain_already_loaded (get_current_sequence ()->next->last))
{
rtx sc_reg = gen_rtx_REG (Pmode, STATIC_CHAIN_REGNUM);
rtx func_sc_offset = GEN_INT (2 * GET_MODE_SIZE (Pmode));
rtx func_sc_mem = gen_rtx_MEM (Pmode,
gen_rtx_PLUS (Pmode, func_desc,
func_sc_offset));
emit_move_insn (sc_reg, func_sc_mem);
abi_reg = sc_reg;
}
}
}
else
{
/* Direct calls use the TOC: for local calls, the callee will
assume the TOC register is set; for non-local calls, the
PLT stub needs the TOC register. */
abi_reg = toc_reg;
func_addr = func_desc;
}
/* Create the call. */
call[0] = gen_rtx_CALL (VOIDmode, gen_rtx_MEM (SImode, func_addr), flag);
if (value != NULL_RTX)
call[0] = gen_rtx_SET (value, call[0]);
n_call = 1;
if (toc_load)
call[n_call++] = toc_load;
if (toc_restore)
call[n_call++] = toc_restore;
call[n_call++] = gen_rtx_CLOBBER (VOIDmode, gen_rtx_REG (Pmode, LR_REGNO));
insn = gen_rtx_PARALLEL (VOIDmode, gen_rtvec_v (n_call, call));
insn = emit_call_insn (insn);
/* Mention all registers defined by the ABI to hold information
as uses in CALL_INSN_FUNCTION_USAGE. */
if (abi_reg)
use_reg (&CALL_INSN_FUNCTION_USAGE (insn), abi_reg);
}
/* Expand code to perform a sibling call under the AIX or ELFv2 ABI. */
void
rs6000_sibcall_aix (rtx value, rtx func_desc, rtx flag, rtx cookie)
{
rtx call[2];
rtx insn;
gcc_assert (INTVAL (cookie) == 0);
/* Create the call. */
call[0] = gen_rtx_CALL (VOIDmode, gen_rtx_MEM (SImode, func_desc), flag);
if (value != NULL_RTX)
call[0] = gen_rtx_SET (value, call[0]);
call[1] = simple_return_rtx;
insn = gen_rtx_PARALLEL (VOIDmode, gen_rtvec_v (2, call));
insn = emit_call_insn (insn);
/* Note use of the TOC register. */
use_reg (&CALL_INSN_FUNCTION_USAGE (insn), gen_rtx_REG (Pmode, TOC_REGNUM));
}
/* Return whether we need to always update the saved TOC pointer when we update
the stack pointer. */
static bool
rs6000_save_toc_in_prologue_p (void)
{
return (cfun && cfun->machine && cfun->machine->save_toc_in_prologue);
}
#ifdef HAVE_GAS_HIDDEN
# define USE_HIDDEN_LINKONCE 1
#else
# define USE_HIDDEN_LINKONCE 0
#endif
/* Fills in the label name that should be used for a 476 link stack thunk. */
void
get_ppc476_thunk_name (char name[32])
{
gcc_assert (TARGET_LINK_STACK);
if (USE_HIDDEN_LINKONCE)
sprintf (name, "__ppc476.get_thunk");
else
ASM_GENERATE_INTERNAL_LABEL (name, "LPPC476_", 0);
}
/* This function emits the simple thunk routine that is used to preserve
the link stack on the 476 cpu. */
static void rs6000_code_end (void) ATTRIBUTE_UNUSED;
static void
rs6000_code_end (void)
{
char name[32];
tree decl;
if (!TARGET_LINK_STACK)
return;
get_ppc476_thunk_name (name);
decl = build_decl (BUILTINS_LOCATION, FUNCTION_DECL, get_identifier (name),
build_function_type_list (void_type_node, NULL_TREE));
DECL_RESULT (decl) = build_decl (BUILTINS_LOCATION, RESULT_DECL,
NULL_TREE, void_type_node);
TREE_PUBLIC (decl) = 1;
TREE_STATIC (decl) = 1;
#if RS6000_WEAK
if (USE_HIDDEN_LINKONCE && !TARGET_XCOFF)
{
cgraph_node::create (decl)->set_comdat_group (DECL_ASSEMBLER_NAME (decl));
targetm.asm_out.unique_section (decl, 0);
switch_to_section (get_named_section (decl, NULL, 0));
DECL_WEAK (decl) = 1;
ASM_WEAKEN_DECL (asm_out_file, decl, name, 0);
targetm.asm_out.globalize_label (asm_out_file, name);
targetm.asm_out.assemble_visibility (decl, VISIBILITY_HIDDEN);
ASM_DECLARE_FUNCTION_NAME (asm_out_file, name, decl);
}
else
#endif
{
switch_to_section (text_section);
ASM_OUTPUT_LABEL (asm_out_file, name);
}
DECL_INITIAL (decl) = make_node (BLOCK);
current_function_decl = decl;
allocate_struct_function (decl, false);
init_function_start (decl);
first_function_block_is_cold = false;
/* Make sure unwind info is emitted for the thunk if needed. */
final_start_function (emit_barrier (), asm_out_file, 1);
fputs ("\tblr\n", asm_out_file);
final_end_function ();
init_insn_lengths ();
free_after_compilation (cfun);
set_cfun (NULL);
current_function_decl = NULL;
}
/* Add r30 to hard reg set if the prologue sets it up and it is not
pic_offset_table_rtx. */
static void
rs6000_set_up_by_prologue (struct hard_reg_set_container *set)
{
if (!TARGET_SINGLE_PIC_BASE
&& TARGET_TOC
&& TARGET_MINIMAL_TOC
&& !constant_pool_empty_p ())
add_to_hard_reg_set (&set->set, Pmode, RS6000_PIC_OFFSET_TABLE_REGNUM);
if (cfun->machine->split_stack_argp_used)
add_to_hard_reg_set (&set->set, Pmode, 12);
/* Make sure the hard reg set doesn't include r2, which was possibly added
via PIC_OFFSET_TABLE_REGNUM. */
if (TARGET_TOC)
remove_from_hard_reg_set (&set->set, Pmode, TOC_REGNUM);
}
/* Helper function for rs6000_split_logical to emit a logical instruction after
spliting the operation to single GPR registers.
DEST is the destination register.
OP1 and OP2 are the input source registers.
CODE is the base operation (AND, IOR, XOR, NOT).
MODE is the machine mode.
If COMPLEMENT_FINAL_P is true, wrap the whole operation with NOT.
If COMPLEMENT_OP1_P is true, wrap operand1 with NOT.
If COMPLEMENT_OP2_P is true, wrap operand2 with NOT. */
static void
rs6000_split_logical_inner (rtx dest,
rtx op1,
rtx op2,
enum rtx_code code,
machine_mode mode,
bool complement_final_p,
bool complement_op1_p,
bool complement_op2_p)
{
rtx bool_rtx;
/* Optimize AND of 0/0xffffffff and IOR/XOR of 0. */
if (op2 && GET_CODE (op2) == CONST_INT
&& (mode == SImode || (mode == DImode && TARGET_POWERPC64))
&& !complement_final_p && !complement_op1_p && !complement_op2_p)
{
HOST_WIDE_INT mask = GET_MODE_MASK (mode);
HOST_WIDE_INT value = INTVAL (op2) & mask;
/* Optimize AND of 0 to just set 0. Optimize AND of -1 to be a move. */
if (code == AND)
{
if (value == 0)
{
emit_insn (gen_rtx_SET (dest, const0_rtx));
return;
}
else if (value == mask)
{
if (!rtx_equal_p (dest, op1))
emit_insn (gen_rtx_SET (dest, op1));
return;
}
}
/* Optimize IOR/XOR of 0 to be a simple move. Split large operations
into separate ORI/ORIS or XORI/XORIS instrucitons. */
else if (code == IOR || code == XOR)
{
if (value == 0)
{
if (!rtx_equal_p (dest, op1))
emit_insn (gen_rtx_SET (dest, op1));
return;
}
}
}
if (code == AND && mode == SImode
&& !complement_final_p && !complement_op1_p && !complement_op2_p)
{
emit_insn (gen_andsi3 (dest, op1, op2));
return;
}
if (complement_op1_p)
op1 = gen_rtx_NOT (mode, op1);
if (complement_op2_p)
op2 = gen_rtx_NOT (mode, op2);
/* For canonical RTL, if only one arm is inverted it is the first. */
if (!complement_op1_p && complement_op2_p)
std::swap (op1, op2);
bool_rtx = ((code == NOT)
? gen_rtx_NOT (mode, op1)
: gen_rtx_fmt_ee (code, mode, op1, op2));
if (complement_final_p)
bool_rtx = gen_rtx_NOT (mode, bool_rtx);
emit_insn (gen_rtx_SET (dest, bool_rtx));
}
/* Split a DImode AND/IOR/XOR with a constant on a 32-bit system. These
operations are split immediately during RTL generation to allow for more
optimizations of the AND/IOR/XOR.
OPERANDS is an array containing the destination and two input operands.
CODE is the base operation (AND, IOR, XOR, NOT).
MODE is the machine mode.
If COMPLEMENT_FINAL_P is true, wrap the whole operation with NOT.
If COMPLEMENT_OP1_P is true, wrap operand1 with NOT.
If COMPLEMENT_OP2_P is true, wrap operand2 with NOT.
CLOBBER_REG is either NULL or a scratch register of type CC to allow
formation of the AND instructions. */
static void
rs6000_split_logical_di (rtx operands[3],
enum rtx_code code,
bool complement_final_p,
bool complement_op1_p,
bool complement_op2_p)
{
const HOST_WIDE_INT lower_32bits = HOST_WIDE_INT_C(0xffffffff);
const HOST_WIDE_INT upper_32bits = ~ lower_32bits;
const HOST_WIDE_INT sign_bit = HOST_WIDE_INT_C(0x80000000);
enum hi_lo { hi = 0, lo = 1 };
rtx op0_hi_lo[2], op1_hi_lo[2], op2_hi_lo[2];
size_t i;
op0_hi_lo[hi] = gen_highpart (SImode, operands[0]);
op1_hi_lo[hi] = gen_highpart (SImode, operands[1]);
op0_hi_lo[lo] = gen_lowpart (SImode, operands[0]);
op1_hi_lo[lo] = gen_lowpart (SImode, operands[1]);
if (code == NOT)
op2_hi_lo[hi] = op2_hi_lo[lo] = NULL_RTX;
else
{
if (GET_CODE (operands[2]) != CONST_INT)
{
op2_hi_lo[hi] = gen_highpart_mode (SImode, DImode, operands[2]);
op2_hi_lo[lo] = gen_lowpart (SImode, operands[2]);
}
else
{
HOST_WIDE_INT value = INTVAL (operands[2]);
HOST_WIDE_INT value_hi_lo[2];
gcc_assert (!complement_final_p);
gcc_assert (!complement_op1_p);
gcc_assert (!complement_op2_p);
value_hi_lo[hi] = value >> 32;
value_hi_lo[lo] = value & lower_32bits;
for (i = 0; i < 2; i++)
{
HOST_WIDE_INT sub_value = value_hi_lo[i];
if (sub_value & sign_bit)
sub_value |= upper_32bits;
op2_hi_lo[i] = GEN_INT (sub_value);
/* If this is an AND instruction, check to see if we need to load
the value in a register. */
if (code == AND && sub_value != -1 && sub_value != 0
&& !and_operand (op2_hi_lo[i], SImode))
op2_hi_lo[i] = force_reg (SImode, op2_hi_lo[i]);
}
}
}
for (i = 0; i < 2; i++)
{
/* Split large IOR/XOR operations. */
if ((code == IOR || code == XOR)
&& GET_CODE (op2_hi_lo[i]) == CONST_INT
&& !complement_final_p
&& !complement_op1_p
&& !complement_op2_p
&& !logical_const_operand (op2_hi_lo[i], SImode))
{
HOST_WIDE_INT value = INTVAL (op2_hi_lo[i]);
HOST_WIDE_INT hi_16bits = value & HOST_WIDE_INT_C(0xffff0000);
HOST_WIDE_INT lo_16bits = value & HOST_WIDE_INT_C(0x0000ffff);
rtx tmp = gen_reg_rtx (SImode);
/* Make sure the constant is sign extended. */
if ((hi_16bits & sign_bit) != 0)
hi_16bits |= upper_32bits;
rs6000_split_logical_inner (tmp, op1_hi_lo[i], GEN_INT (hi_16bits),
code, SImode, false, false, false);
rs6000_split_logical_inner (op0_hi_lo[i], tmp, GEN_INT (lo_16bits),
code, SImode, false, false, false);
}
else
rs6000_split_logical_inner (op0_hi_lo[i], op1_hi_lo[i], op2_hi_lo[i],
code, SImode, complement_final_p,
complement_op1_p, complement_op2_p);
}
return;
}
/* Split the insns that make up boolean operations operating on multiple GPR
registers. The boolean MD patterns ensure that the inputs either are
exactly the same as the output registers, or there is no overlap.
OPERANDS is an array containing the destination and two input operands.
CODE is the base operation (AND, IOR, XOR, NOT).
If COMPLEMENT_FINAL_P is true, wrap the whole operation with NOT.
If COMPLEMENT_OP1_P is true, wrap operand1 with NOT.
If COMPLEMENT_OP2_P is true, wrap operand2 with NOT. */
void
rs6000_split_logical (rtx operands[3],
enum rtx_code code,
bool complement_final_p,
bool complement_op1_p,
bool complement_op2_p)
{
machine_mode mode = GET_MODE (operands[0]);
machine_mode sub_mode;
rtx op0, op1, op2;
int sub_size, regno0, regno1, nregs, i;
/* If this is DImode, use the specialized version that can run before
register allocation. */
if (mode == DImode && !TARGET_POWERPC64)
{
rs6000_split_logical_di (operands, code, complement_final_p,
complement_op1_p, complement_op2_p);
return;
}
op0 = operands[0];
op1 = operands[1];
op2 = (code == NOT) ? NULL_RTX : operands[2];
sub_mode = (TARGET_POWERPC64) ? DImode : SImode;
sub_size = GET_MODE_SIZE (sub_mode);
regno0 = REGNO (op0);
regno1 = REGNO (op1);
gcc_assert (reload_completed);
gcc_assert (IN_RANGE (regno0, FIRST_GPR_REGNO, LAST_GPR_REGNO));
gcc_assert (IN_RANGE (regno1, FIRST_GPR_REGNO, LAST_GPR_REGNO));
nregs = rs6000_hard_regno_nregs[(int)mode][regno0];
gcc_assert (nregs > 1);
if (op2 && REG_P (op2))
gcc_assert (IN_RANGE (REGNO (op2), FIRST_GPR_REGNO, LAST_GPR_REGNO));
for (i = 0; i < nregs; i++)
{
int offset = i * sub_size;
rtx sub_op0 = simplify_subreg (sub_mode, op0, mode, offset);
rtx sub_op1 = simplify_subreg (sub_mode, op1, mode, offset);
rtx sub_op2 = ((code == NOT)
? NULL_RTX
: simplify_subreg (sub_mode, op2, mode, offset));
rs6000_split_logical_inner (sub_op0, sub_op1, sub_op2, code, sub_mode,
complement_final_p, complement_op1_p,
complement_op2_p);
}
return;
}
/* Return true if the peephole2 can combine a load involving a combination of
an addis instruction and a load with an offset that can be fused together on
a power8. */
bool
fusion_gpr_load_p (rtx addis_reg, /* register set via addis. */
rtx addis_value, /* addis value. */
rtx target, /* target register that is loaded. */
rtx mem) /* bottom part of the memory addr. */
{
rtx addr;
rtx base_reg;
/* Validate arguments. */
if (!base_reg_operand (addis_reg, GET_MODE (addis_reg)))
return false;
if (!base_reg_operand (target, GET_MODE (target)))
return false;
if (!fusion_gpr_addis (addis_value, GET_MODE (addis_value)))
return false;
/* Allow sign/zero extension. */
if (GET_CODE (mem) == ZERO_EXTEND
|| (GET_CODE (mem) == SIGN_EXTEND && TARGET_P8_FUSION_SIGN))
mem = XEXP (mem, 0);
if (!MEM_P (mem))
return false;
if (!fusion_gpr_mem_load (mem, GET_MODE (mem)))
return false;
addr = XEXP (mem, 0); /* either PLUS or LO_SUM. */
if (GET_CODE (addr) != PLUS && GET_CODE (addr) != LO_SUM)
return false;
/* Validate that the register used to load the high value is either the
register being loaded, or we can safely replace its use.
This function is only called from the peephole2 pass and we assume that
there are 2 instructions in the peephole (addis and load), so we want to
check if the target register was not used in the memory address and the
register to hold the addis result is dead after the peephole. */
if (REGNO (addis_reg) != REGNO (target))
{
if (reg_mentioned_p (target, mem))
return false;
if (!peep2_reg_dead_p (2, addis_reg))
return false;
/* If the target register being loaded is the stack pointer, we must
avoid loading any other value into it, even temporarily. */
if (REG_P (target) && REGNO (target) == STACK_POINTER_REGNUM)
return false;
}
base_reg = XEXP (addr, 0);
return REGNO (addis_reg) == REGNO (base_reg);
}
/* During the peephole2 pass, adjust and expand the insns for a load fusion
sequence. We adjust the addis register to use the target register. If the
load sign extends, we adjust the code to do the zero extending load, and an
explicit sign extension later since the fusion only covers zero extending
loads.
The operands are:
operands[0] register set with addis (to be replaced with target)
operands[1] value set via addis
operands[2] target register being loaded
operands[3] D-form memory reference using operands[0]. */
void
expand_fusion_gpr_load (rtx *operands)
{
rtx addis_value = operands[1];
rtx target = operands[2];
rtx orig_mem = operands[3];
rtx new_addr, new_mem, orig_addr, offset;
enum rtx_code plus_or_lo_sum;
machine_mode target_mode = GET_MODE (target);
machine_mode extend_mode = target_mode;
machine_mode ptr_mode = Pmode;
enum rtx_code extend = UNKNOWN;
if (GET_CODE (orig_mem) == ZERO_EXTEND
|| (TARGET_P8_FUSION_SIGN && GET_CODE (orig_mem) == SIGN_EXTEND))
{
extend = GET_CODE (orig_mem);
orig_mem = XEXP (orig_mem, 0);
target_mode = GET_MODE (orig_mem);
}
gcc_assert (MEM_P (orig_mem));
orig_addr = XEXP (orig_mem, 0);
plus_or_lo_sum = GET_CODE (orig_addr);
gcc_assert (plus_or_lo_sum == PLUS || plus_or_lo_sum == LO_SUM);
offset = XEXP (orig_addr, 1);
new_addr = gen_rtx_fmt_ee (plus_or_lo_sum, ptr_mode, addis_value, offset);
new_mem = replace_equiv_address_nv (orig_mem, new_addr, false);
if (extend != UNKNOWN)
new_mem = gen_rtx_fmt_e (ZERO_EXTEND, extend_mode, new_mem);
new_mem = gen_rtx_UNSPEC (extend_mode, gen_rtvec (1, new_mem),
UNSPEC_FUSION_GPR);
emit_insn (gen_rtx_SET (target, new_mem));
if (extend == SIGN_EXTEND)
{
int sub_off = ((BYTES_BIG_ENDIAN)
? GET_MODE_SIZE (extend_mode) - GET_MODE_SIZE (target_mode)
: 0);
rtx sign_reg
= simplify_subreg (target_mode, target, extend_mode, sub_off);
emit_insn (gen_rtx_SET (target,
gen_rtx_SIGN_EXTEND (extend_mode, sign_reg)));
}
return;
}
/* Emit the addis instruction that will be part of a fused instruction
sequence. */
void
emit_fusion_addis (rtx target, rtx addis_value)
{
rtx fuse_ops[10];
const char *addis_str = NULL;
/* Emit the addis instruction. */
fuse_ops[0] = target;
if (satisfies_constraint_L (addis_value))
{
fuse_ops[1] = addis_value;
addis_str = "lis %0,%v1";
}
else if (GET_CODE (addis_value) == PLUS)
{
rtx op0 = XEXP (addis_value, 0);
rtx op1 = XEXP (addis_value, 1);
if (REG_P (op0) && CONST_INT_P (op1)
&& satisfies_constraint_L (op1))
{
fuse_ops[1] = op0;
fuse_ops[2] = op1;
addis_str = "addis %0,%1,%v2";
}
}
else if (GET_CODE (addis_value) == HIGH)
{
rtx value = XEXP (addis_value, 0);
if (GET_CODE (value) == UNSPEC && XINT (value, 1) == UNSPEC_TOCREL)
{
fuse_ops[1] = XVECEXP (value, 0, 0); /* symbol ref. */
fuse_ops[2] = XVECEXP (value, 0, 1); /* TOC register. */
if (TARGET_ELF)
addis_str = "addis %0,%2,%1@toc@ha";
else if (TARGET_XCOFF)
addis_str = "addis %0,%1@u(%2)";
else
gcc_unreachable ();
}
else if (GET_CODE (value) == PLUS)
{
rtx op0 = XEXP (value, 0);
rtx op1 = XEXP (value, 1);
if (GET_CODE (op0) == UNSPEC
&& XINT (op0, 1) == UNSPEC_TOCREL
&& CONST_INT_P (op1))
{
fuse_ops[1] = XVECEXP (op0, 0, 0); /* symbol ref. */
fuse_ops[2] = XVECEXP (op0, 0, 1); /* TOC register. */
fuse_ops[3] = op1;
if (TARGET_ELF)
addis_str = "addis %0,%2,%1+%3@toc@ha";
else if (TARGET_XCOFF)
addis_str = "addis %0,%1+%3@u(%2)";
else
gcc_unreachable ();
}
}
else if (satisfies_constraint_L (value))
{
fuse_ops[1] = value;
addis_str = "lis %0,%v1";
}
else if (TARGET_ELF && !TARGET_POWERPC64 && CONSTANT_P (value))
{
fuse_ops[1] = value;
addis_str = "lis %0,%1@ha";
}
}
if (!addis_str)
fatal_insn ("Could not generate addis value for fusion", addis_value);
output_asm_insn (addis_str, fuse_ops);
}
/* Emit a D-form load or store instruction that is the second instruction
of a fusion sequence. */
void
emit_fusion_load_store (rtx load_store_reg, rtx addis_reg, rtx offset,
const char *insn_str)
{
rtx fuse_ops[10];
char insn_template[80];
fuse_ops[0] = load_store_reg;
fuse_ops[1] = addis_reg;
if (CONST_INT_P (offset) && satisfies_constraint_I (offset))
{
sprintf (insn_template, "%s %%0,%%2(%%1)", insn_str);
fuse_ops[2] = offset;
output_asm_insn (insn_template, fuse_ops);
}
else if (GET_CODE (offset) == UNSPEC
&& XINT (offset, 1) == UNSPEC_TOCREL)
{
if (TARGET_ELF)
sprintf (insn_template, "%s %%0,%%2@toc@l(%%1)", insn_str);
else if (TARGET_XCOFF)
sprintf (insn_template, "%s %%0,%%2@l(%%1)", insn_str);
else
gcc_unreachable ();
fuse_ops[2] = XVECEXP (offset, 0, 0);
output_asm_insn (insn_template, fuse_ops);
}
else if (GET_CODE (offset) == PLUS
&& GET_CODE (XEXP (offset, 0)) == UNSPEC
&& XINT (XEXP (offset, 0), 1) == UNSPEC_TOCREL
&& CONST_INT_P (XEXP (offset, 1)))
{
rtx tocrel_unspec = XEXP (offset, 0);
if (TARGET_ELF)
sprintf (insn_template, "%s %%0,%%2+%%3@toc@l(%%1)", insn_str);
else if (TARGET_XCOFF)
sprintf (insn_template, "%s %%0,%%2+%%3@l(%%1)", insn_str);
else
gcc_unreachable ();
fuse_ops[2] = XVECEXP (tocrel_unspec, 0, 0);
fuse_ops[3] = XEXP (offset, 1);
output_asm_insn (insn_template, fuse_ops);
}
else if (TARGET_ELF && !TARGET_POWERPC64 && CONSTANT_P (offset))
{
sprintf (insn_template, "%s %%0,%%2@l(%%1)", insn_str);
fuse_ops[2] = offset;
output_asm_insn (insn_template, fuse_ops);
}
else
fatal_insn ("Unable to generate load/store offset for fusion", offset);
return;
}
/* Wrap a TOC address that can be fused to indicate that special fusion
processing is needed. */
rtx
fusion_wrap_memory_address (rtx old_mem)
{
rtx old_addr = XEXP (old_mem, 0);
rtvec v = gen_rtvec (1, old_addr);
rtx new_addr = gen_rtx_UNSPEC (Pmode, v, UNSPEC_FUSION_ADDIS);
return replace_equiv_address_nv (old_mem, new_addr, false);
}
/* Given an address, convert it into the addis and load offset parts. Addresses
created during the peephole2 process look like:
(lo_sum (high (unspec [(sym)] UNSPEC_TOCREL))
(unspec [(...)] UNSPEC_TOCREL))
Addresses created via toc fusion look like:
(unspec [(unspec [(...)] UNSPEC_TOCREL)] UNSPEC_FUSION_ADDIS)) */
static void
fusion_split_address (rtx addr, rtx *p_hi, rtx *p_lo)
{
rtx hi, lo;
if (GET_CODE (addr) == UNSPEC && XINT (addr, 1) == UNSPEC_FUSION_ADDIS)
{
lo = XVECEXP (addr, 0, 0);
hi = gen_rtx_HIGH (Pmode, lo);
}
else if (GET_CODE (addr) == PLUS || GET_CODE (addr) == LO_SUM)
{
hi = XEXP (addr, 0);
lo = XEXP (addr, 1);
}
else
gcc_unreachable ();
*p_hi = hi;
*p_lo = lo;
}
/* Return a string to fuse an addis instruction with a gpr load to the same
register that we loaded up the addis instruction. The address that is used
is the logical address that was formed during peephole2:
(lo_sum (high) (low-part))
Or the address is the TOC address that is wrapped before register allocation:
(unspec [(addr) (toc-reg)] UNSPEC_FUSION_ADDIS)
The code is complicated, so we call output_asm_insn directly, and just
return "". */
const char *
emit_fusion_gpr_load (rtx target, rtx mem)
{
rtx addis_value;
rtx addr;
rtx load_offset;
const char *load_str = NULL;
machine_mode mode;
if (GET_CODE (mem) == ZERO_EXTEND)
mem = XEXP (mem, 0);
gcc_assert (REG_P (target) && MEM_P (mem));
addr = XEXP (mem, 0);
fusion_split_address (addr, &addis_value, &load_offset);
/* Now emit the load instruction to the same register. */
mode = GET_MODE (mem);
switch (mode)
{
case E_QImode:
load_str = "lbz";
break;
case E_HImode:
load_str = "lhz";
break;
case E_SImode:
case E_SFmode:
load_str = "lwz";
break;
case E_DImode:
case E_DFmode:
gcc_assert (TARGET_POWERPC64);
load_str = "ld";
break;
default:
fatal_insn ("Bad GPR fusion", gen_rtx_SET (target, mem));
}
/* Emit the addis instruction. */
emit_fusion_addis (target, addis_value);
/* Emit the D-form load instruction. */
emit_fusion_load_store (target, target, load_offset, load_str);
return "";
}
/* Return true if the peephole2 can combine a load/store involving a
combination of an addis instruction and the memory operation. This was
added to the ISA 3.0 (power9) hardware. */
bool
fusion_p9_p (rtx addis_reg, /* register set via addis. */
rtx addis_value, /* addis value. */
rtx dest, /* destination (memory or register). */
rtx src) /* source (register or memory). */
{
rtx addr, mem, offset;
machine_mode mode = GET_MODE (src);
/* Validate arguments. */
if (!base_reg_operand (addis_reg, GET_MODE (addis_reg)))
return false;
if (!fusion_gpr_addis (addis_value, GET_MODE (addis_value)))
return false;
/* Ignore extend operations that are part of the load. */
if (GET_CODE (src) == FLOAT_EXTEND || GET_CODE (src) == ZERO_EXTEND)
src = XEXP (src, 0);
/* Test for memory<-register or register<-memory. */
if (fpr_reg_operand (src, mode) || int_reg_operand (src, mode))
{
if (!MEM_P (dest))
return false;
mem = dest;
}
else if (MEM_P (src))
{
if (!fpr_reg_operand (dest, mode) && !int_reg_operand (dest, mode))
return false;
mem = src;
}
else
return false;
addr = XEXP (mem, 0); /* either PLUS or LO_SUM. */
if (GET_CODE (addr) == PLUS)
{
if (!rtx_equal_p (addis_reg, XEXP (addr, 0)))
return false;
return satisfies_constraint_I (XEXP (addr, 1));
}
else if (GET_CODE (addr) == LO_SUM)
{
if (!rtx_equal_p (addis_reg, XEXP (addr, 0)))
return false;
offset = XEXP (addr, 1);
if (TARGET_XCOFF || (TARGET_ELF && TARGET_POWERPC64))
return small_toc_ref (offset, GET_MODE (offset));
else if (TARGET_ELF && !TARGET_POWERPC64)
return CONSTANT_P (offset);
}
return false;
}
/* During the peephole2 pass, adjust and expand the insns for an extended fusion
load sequence.
The operands are:
operands[0] register set with addis
operands[1] value set via addis
operands[2] target register being loaded
operands[3] D-form memory reference using operands[0].
This is similar to the fusion introduced with power8, except it scales to
both loads/stores and does not require the result register to be the same as
the base register. At the moment, we only do this if register set with addis
is dead. */
void
expand_fusion_p9_load (rtx *operands)
{
rtx tmp_reg = operands[0];
rtx addis_value = operands[1];
rtx target = operands[2];
rtx orig_mem = operands[3];
rtx new_addr, new_mem, orig_addr, offset, set, clobber, insn;
enum rtx_code plus_or_lo_sum;
machine_mode target_mode = GET_MODE (target);
machine_mode extend_mode = target_mode;
machine_mode ptr_mode = Pmode;
enum rtx_code extend = UNKNOWN;
if (GET_CODE (orig_mem) == FLOAT_EXTEND || GET_CODE (orig_mem) == ZERO_EXTEND)
{
extend = GET_CODE (orig_mem);
orig_mem = XEXP (orig_mem, 0);
target_mode = GET_MODE (orig_mem);
}
gcc_assert (MEM_P (orig_mem));
orig_addr = XEXP (orig_mem, 0);
plus_or_lo_sum = GET_CODE (orig_addr);
gcc_assert (plus_or_lo_sum == PLUS || plus_or_lo_sum == LO_SUM);
offset = XEXP (orig_addr, 1);
new_addr = gen_rtx_fmt_ee (plus_or_lo_sum, ptr_mode, addis_value, offset);
new_mem = replace_equiv_address_nv (orig_mem, new_addr, false);
if (extend != UNKNOWN)
new_mem = gen_rtx_fmt_e (extend, extend_mode, new_mem);
new_mem = gen_rtx_UNSPEC (extend_mode, gen_rtvec (1, new_mem),
UNSPEC_FUSION_P9);
set = gen_rtx_SET (target, new_mem);
clobber = gen_rtx_CLOBBER (VOIDmode, tmp_reg);
insn = gen_rtx_PARALLEL (VOIDmode, gen_rtvec (2, set, clobber));
emit_insn (insn);
return;
}
/* During the peephole2 pass, adjust and expand the insns for an extended fusion
store sequence.
The operands are:
operands[0] register set with addis
operands[1] value set via addis
operands[2] target D-form memory being stored to
operands[3] register being stored
This is similar to the fusion introduced with power8, except it scales to
both loads/stores and does not require the result register to be the same as
the base register. At the moment, we only do this if register set with addis
is dead. */
void
expand_fusion_p9_store (rtx *operands)
{
rtx tmp_reg = operands[0];
rtx addis_value = operands[1];
rtx orig_mem = operands[2];
rtx src = operands[3];
rtx new_addr, new_mem, orig_addr, offset, set, clobber, insn, new_src;
enum rtx_code plus_or_lo_sum;
machine_mode target_mode = GET_MODE (orig_mem);
machine_mode ptr_mode = Pmode;
gcc_assert (MEM_P (orig_mem));
orig_addr = XEXP (orig_mem, 0);
plus_or_lo_sum = GET_CODE (orig_addr);
gcc_assert (plus_or_lo_sum == PLUS || plus_or_lo_sum == LO_SUM);
offset = XEXP (orig_addr, 1);
new_addr = gen_rtx_fmt_ee (plus_or_lo_sum, ptr_mode, addis_value, offset);
new_mem = replace_equiv_address_nv (orig_mem, new_addr, false);
new_src = gen_rtx_UNSPEC (target_mode, gen_rtvec (1, src),
UNSPEC_FUSION_P9);
set = gen_rtx_SET (new_mem, new_src);
clobber = gen_rtx_CLOBBER (VOIDmode, tmp_reg);
insn = gen_rtx_PARALLEL (VOIDmode, gen_rtvec (2, set, clobber));
emit_insn (insn);
return;
}
/* Return a string to fuse an addis instruction with a load using extended
fusion. The address that is used is the logical address that was formed
during peephole2: (lo_sum (high) (low-part))
The code is complicated, so we call output_asm_insn directly, and just
return "". */
const char *
emit_fusion_p9_load (rtx reg, rtx mem, rtx tmp_reg)
{
machine_mode mode = GET_MODE (reg);
rtx hi;
rtx lo;
rtx addr;
const char *load_string;
int r;
if (GET_CODE (mem) == FLOAT_EXTEND || GET_CODE (mem) == ZERO_EXTEND)
{
mem = XEXP (mem, 0);
mode = GET_MODE (mem);
}
if (GET_CODE (reg) == SUBREG)
{
gcc_assert (SUBREG_BYTE (reg) == 0);
reg = SUBREG_REG (reg);
}
if (!REG_P (reg))
fatal_insn ("emit_fusion_p9_load, bad reg #1", reg);
r = REGNO (reg);
if (FP_REGNO_P (r))
{
if (mode == SFmode)
load_string = "lfs";
else if (mode == DFmode || mode == DImode)
load_string = "lfd";
else
gcc_unreachable ();
}
else if (ALTIVEC_REGNO_P (r) && TARGET_P9_VECTOR)
{
if (mode == SFmode)
load_string = "lxssp";
else if (mode == DFmode || mode == DImode)
load_string = "lxsd";
else
gcc_unreachable ();
}
else if (INT_REGNO_P (r))
{
switch (mode)
{
case E_QImode:
load_string = "lbz";
break;
case E_HImode:
load_string = "lhz";
break;
case E_SImode:
case E_SFmode:
load_string = "lwz";
break;
case E_DImode:
case E_DFmode:
if (!TARGET_POWERPC64)
gcc_unreachable ();
load_string = "ld";
break;
default:
gcc_unreachable ();
}
}
else
fatal_insn ("emit_fusion_p9_load, bad reg #2", reg);
if (!MEM_P (mem))
fatal_insn ("emit_fusion_p9_load not MEM", mem);
addr = XEXP (mem, 0);
fusion_split_address (addr, &hi, &lo);
/* Emit the addis instruction. */
emit_fusion_addis (tmp_reg, hi);
/* Emit the D-form load instruction. */
emit_fusion_load_store (reg, tmp_reg, lo, load_string);
return "";
}
/* Return a string to fuse an addis instruction with a store using extended
fusion. The address that is used is the logical address that was formed
during peephole2: (lo_sum (high) (low-part))
The code is complicated, so we call output_asm_insn directly, and just
return "". */
const char *
emit_fusion_p9_store (rtx mem, rtx reg, rtx tmp_reg)
{
machine_mode mode = GET_MODE (reg);
rtx hi;
rtx lo;
rtx addr;
const char *store_string;
int r;
if (GET_CODE (reg) == SUBREG)
{
gcc_assert (SUBREG_BYTE (reg) == 0);
reg = SUBREG_REG (reg);
}
if (!REG_P (reg))
fatal_insn ("emit_fusion_p9_store, bad reg #1", reg);
r = REGNO (reg);
if (FP_REGNO_P (r))
{
if (mode == SFmode)
store_string = "stfs";
else if (mode == DFmode)
store_string = "stfd";
else
gcc_unreachable ();
}
else if (ALTIVEC_REGNO_P (r) && TARGET_P9_VECTOR)
{
if (mode == SFmode)
store_string = "stxssp";
else if (mode == DFmode || mode == DImode)
store_string = "stxsd";
else
gcc_unreachable ();
}
else if (INT_REGNO_P (r))
{
switch (mode)
{
case E_QImode:
store_string = "stb";
break;
case E_HImode:
store_string = "sth";
break;
case E_SImode:
case E_SFmode:
store_string = "stw";
break;
case E_DImode:
case E_DFmode:
if (!TARGET_POWERPC64)
gcc_unreachable ();
store_string = "std";
break;
default:
gcc_unreachable ();
}
}
else
fatal_insn ("emit_fusion_p9_store, bad reg #2", reg);
if (!MEM_P (mem))
fatal_insn ("emit_fusion_p9_store not MEM", mem);
addr = XEXP (mem, 0);
fusion_split_address (addr, &hi, &lo);
/* Emit the addis instruction. */
emit_fusion_addis (tmp_reg, hi);
/* Emit the D-form load instruction. */
emit_fusion_load_store (reg, tmp_reg, lo, store_string);
return "";
}
#ifdef RS6000_GLIBC_ATOMIC_FENV
/* Function declarations for rs6000_atomic_assign_expand_fenv. */
static tree atomic_hold_decl, atomic_clear_decl, atomic_update_decl;
#endif
/* Implement TARGET_ATOMIC_ASSIGN_EXPAND_FENV hook. */
static void
rs6000_atomic_assign_expand_fenv (tree *hold, tree *clear, tree *update)
{
if (!TARGET_HARD_FLOAT)
{
#ifdef RS6000_GLIBC_ATOMIC_FENV
if (atomic_hold_decl == NULL_TREE)
{
atomic_hold_decl
= build_decl (BUILTINS_LOCATION, FUNCTION_DECL,
get_identifier ("__atomic_feholdexcept"),
build_function_type_list (void_type_node,
double_ptr_type_node,
NULL_TREE));
TREE_PUBLIC (atomic_hold_decl) = 1;
DECL_EXTERNAL (atomic_hold_decl) = 1;
}
if (atomic_clear_decl == NULL_TREE)
{
atomic_clear_decl
= build_decl (BUILTINS_LOCATION, FUNCTION_DECL,
get_identifier ("__atomic_feclearexcept"),
build_function_type_list (void_type_node,
NULL_TREE));
TREE_PUBLIC (atomic_clear_decl) = 1;
DECL_EXTERNAL (atomic_clear_decl) = 1;
}
tree const_double = build_qualified_type (double_type_node,
TYPE_QUAL_CONST);
tree const_double_ptr = build_pointer_type (const_double);
if (atomic_update_decl == NULL_TREE)
{
atomic_update_decl
= build_decl (BUILTINS_LOCATION, FUNCTION_DECL,
get_identifier ("__atomic_feupdateenv"),
build_function_type_list (void_type_node,
const_double_ptr,
NULL_TREE));
TREE_PUBLIC (atomic_update_decl) = 1;
DECL_EXTERNAL (atomic_update_decl) = 1;
}
tree fenv_var = create_tmp_var_raw (double_type_node);
TREE_ADDRESSABLE (fenv_var) = 1;
tree fenv_addr = build1 (ADDR_EXPR, double_ptr_type_node, fenv_var);
*hold = build_call_expr (atomic_hold_decl, 1, fenv_addr);
*clear = build_call_expr (atomic_clear_decl, 0);
*update = build_call_expr (atomic_update_decl, 1,
fold_convert (const_double_ptr, fenv_addr));
#endif
return;
}
tree mffs = rs6000_builtin_decls[RS6000_BUILTIN_MFFS];
tree mtfsf = rs6000_builtin_decls[RS6000_BUILTIN_MTFSF];
tree call_mffs = build_call_expr (mffs, 0);
/* Generates the equivalent of feholdexcept (&fenv_var)
*fenv_var = __builtin_mffs ();
double fenv_hold;
*(uint64_t*)&fenv_hold = *(uint64_t*)fenv_var & 0xffffffff00000007LL;
__builtin_mtfsf (0xff, fenv_hold); */
/* Mask to clear everything except for the rounding modes and non-IEEE
arithmetic flag. */
const unsigned HOST_WIDE_INT hold_exception_mask =
HOST_WIDE_INT_C (0xffffffff00000007);
tree fenv_var = create_tmp_var_raw (double_type_node);
tree hold_mffs = build2 (MODIFY_EXPR, void_type_node, fenv_var, call_mffs);
tree fenv_llu = build1 (VIEW_CONVERT_EXPR, uint64_type_node, fenv_var);
tree fenv_llu_and = build2 (BIT_AND_EXPR, uint64_type_node, fenv_llu,
build_int_cst (uint64_type_node,
hold_exception_mask));
tree fenv_hold_mtfsf = build1 (VIEW_CONVERT_EXPR, double_type_node,
fenv_llu_and);
tree hold_mtfsf = build_call_expr (mtfsf, 2,
build_int_cst (unsigned_type_node, 0xff),
fenv_hold_mtfsf);
*hold = build2 (COMPOUND_EXPR, void_type_node, hold_mffs, hold_mtfsf);
/* Generates the equivalent of feclearexcept (FE_ALL_EXCEPT):
double fenv_clear = __builtin_mffs ();
*(uint64_t)&fenv_clear &= 0xffffffff00000000LL;
__builtin_mtfsf (0xff, fenv_clear); */
/* Mask to clear everything except for the rounding modes and non-IEEE
arithmetic flag. */
const unsigned HOST_WIDE_INT clear_exception_mask =
HOST_WIDE_INT_C (0xffffffff00000000);
tree fenv_clear = create_tmp_var_raw (double_type_node);
tree clear_mffs = build2 (MODIFY_EXPR, void_type_node, fenv_clear, call_mffs);
tree fenv_clean_llu = build1 (VIEW_CONVERT_EXPR, uint64_type_node, fenv_clear);
tree fenv_clear_llu_and = build2 (BIT_AND_EXPR, uint64_type_node,
fenv_clean_llu,
build_int_cst (uint64_type_node,
clear_exception_mask));
tree fenv_clear_mtfsf = build1 (VIEW_CONVERT_EXPR, double_type_node,
fenv_clear_llu_and);
tree clear_mtfsf = build_call_expr (mtfsf, 2,
build_int_cst (unsigned_type_node, 0xff),
fenv_clear_mtfsf);
*clear = build2 (COMPOUND_EXPR, void_type_node, clear_mffs, clear_mtfsf);
/* Generates the equivalent of feupdateenv (&fenv_var)
double old_fenv = __builtin_mffs ();
double fenv_update;
*(uint64_t*)&fenv_update = (*(uint64_t*)&old & 0xffffffff1fffff00LL) |
(*(uint64_t*)fenv_var 0x1ff80fff);
__builtin_mtfsf (0xff, fenv_update); */
const unsigned HOST_WIDE_INT update_exception_mask =
HOST_WIDE_INT_C (0xffffffff1fffff00);
const unsigned HOST_WIDE_INT new_exception_mask =
HOST_WIDE_INT_C (0x1ff80fff);
tree old_fenv = create_tmp_var_raw (double_type_node);
tree update_mffs = build2 (MODIFY_EXPR, void_type_node, old_fenv, call_mffs);
tree old_llu = build1 (VIEW_CONVERT_EXPR, uint64_type_node, old_fenv);
tree old_llu_and = build2 (BIT_AND_EXPR, uint64_type_node, old_llu,
build_int_cst (uint64_type_node,
update_exception_mask));
tree new_llu_and = build2 (BIT_AND_EXPR, uint64_type_node, fenv_llu,
build_int_cst (uint64_type_node,
new_exception_mask));
tree new_llu_mask = build2 (BIT_IOR_EXPR, uint64_type_node,
old_llu_and, new_llu_and);
tree fenv_update_mtfsf = build1 (VIEW_CONVERT_EXPR, double_type_node,
new_llu_mask);
tree update_mtfsf = build_call_expr (mtfsf, 2,
build_int_cst (unsigned_type_node, 0xff),
fenv_update_mtfsf);
*update = build2 (COMPOUND_EXPR, void_type_node, update_mffs, update_mtfsf);
}
void
rs6000_generate_float2_code (bool signed_convert, rtx dst, rtx src1, rtx src2)
{
rtx rtx_tmp0, rtx_tmp1, rtx_tmp2, rtx_tmp3;
rtx_tmp0 = gen_reg_rtx (V2DImode);
rtx_tmp1 = gen_reg_rtx (V2DImode);
/* The destination of the vmrgew instruction layout is:
rtx_tmp2[0] rtx_tmp3[0] rtx_tmp2[1] rtx_tmp3[0].
Setup rtx_tmp0 and rtx_tmp1 to ensure the order of the elements after the
vmrgew instruction will be correct. */
if (VECTOR_ELT_ORDER_BIG)
{
emit_insn (gen_vsx_xxpermdi_v2di_be (rtx_tmp0, src1, src2, GEN_INT (0)));
emit_insn (gen_vsx_xxpermdi_v2di_be (rtx_tmp1, src1, src2, GEN_INT (3)));
}
else
{
emit_insn (gen_vsx_xxpermdi_v2di (rtx_tmp0, src1, src2, GEN_INT (3)));
emit_insn (gen_vsx_xxpermdi_v2di (rtx_tmp1, src1, src2, GEN_INT (0)));
}
rtx_tmp2 = gen_reg_rtx (V4SFmode);
rtx_tmp3 = gen_reg_rtx (V4SFmode);
if (signed_convert)
{
emit_insn (gen_vsx_xvcvsxdsp (rtx_tmp2, rtx_tmp0));
emit_insn (gen_vsx_xvcvsxdsp (rtx_tmp3, rtx_tmp1));
}
else
{
emit_insn (gen_vsx_xvcvuxdsp (rtx_tmp2, rtx_tmp0));
emit_insn (gen_vsx_xvcvuxdsp (rtx_tmp3, rtx_tmp1));
}
if (VECTOR_ELT_ORDER_BIG)
emit_insn (gen_p8_vmrgew_v4sf (dst, rtx_tmp2, rtx_tmp3));
else
emit_insn (gen_p8_vmrgew_v4sf (dst, rtx_tmp3, rtx_tmp2));
}
void
rs6000_generate_vsigned2_code (bool signed_convert, rtx dst, rtx src1,
rtx src2)
{
rtx rtx_tmp0, rtx_tmp1, rtx_tmp2, rtx_tmp3;
rtx_tmp0 = gen_reg_rtx (V2DFmode);
rtx_tmp1 = gen_reg_rtx (V2DFmode);
emit_insn (gen_vsx_xxpermdi_v2df (rtx_tmp0, src1, src2, GEN_INT (0)));
emit_insn (gen_vsx_xxpermdi_v2df (rtx_tmp1, src1, src2, GEN_INT (3)));
rtx_tmp2 = gen_reg_rtx (V4SImode);
rtx_tmp3 = gen_reg_rtx (V4SImode);
if (signed_convert)
{
emit_insn (gen_vsx_xvcvdpsxws (rtx_tmp2, rtx_tmp0));
emit_insn (gen_vsx_xvcvdpsxws (rtx_tmp3, rtx_tmp1));
}
else
{
emit_insn (gen_vsx_xvcvdpuxws (rtx_tmp2, rtx_tmp0));
emit_insn (gen_vsx_xvcvdpuxws (rtx_tmp3, rtx_tmp1));
}
emit_insn (gen_p8_vmrgew_v4si (dst, rtx_tmp2, rtx_tmp3));
}
/* Implement the TARGET_OPTAB_SUPPORTED_P hook. */
static bool
rs6000_optab_supported_p (int op, machine_mode mode1, machine_mode,
optimization_type opt_type)
{
switch (op)
{
case rsqrt_optab:
return (opt_type == OPTIMIZE_FOR_SPEED
&& RS6000_RECIP_AUTO_RSQRTE_P (mode1));
default:
return true;
}
}
/* Implement TARGET_CONSTANT_ALIGNMENT. */
static HOST_WIDE_INT
rs6000_constant_alignment (const_tree exp, HOST_WIDE_INT align)
{
if (TREE_CODE (exp) == STRING_CST
&& (STRICT_ALIGNMENT || !optimize_size))
return MAX (align, BITS_PER_WORD);
return align;
}
/* Implement TARGET_STARTING_FRAME_OFFSET. */
static HOST_WIDE_INT
rs6000_starting_frame_offset (void)
{
if (FRAME_GROWS_DOWNWARD)
return 0;
return RS6000_STARTING_FRAME_OFFSET;
}
struct gcc_target targetm = TARGET_INITIALIZER;
#include "gt-rs6000.h"
|