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
|
/* Medium-level subroutines: convert bit-field store and extract
and shifts, multiplies and divides to rtl instructions.
Copyright (C) 1987, 1988, 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
1999, 2000, 2001, 2002 Free Software Foundation, Inc.
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 2, 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 COPYING. If not, write to the Free
Software Foundation, 59 Temple Place - Suite 330, Boston, MA
02111-1307, USA. */
#include "config.h"
#include "system.h"
#include "toplev.h"
#include "rtl.h"
#include "tree.h"
#include "tm_p.h"
#include "flags.h"
#include "insn-config.h"
#include "expr.h"
#include "optabs.h"
#include "real.h"
#include "recog.h"
#include "langhooks.h"
static void store_fixed_bit_field PARAMS ((rtx, unsigned HOST_WIDE_INT,
unsigned HOST_WIDE_INT,
unsigned HOST_WIDE_INT, rtx));
static void store_split_bit_field PARAMS ((rtx, unsigned HOST_WIDE_INT,
unsigned HOST_WIDE_INT, rtx));
static rtx extract_fixed_bit_field PARAMS ((enum machine_mode, rtx,
unsigned HOST_WIDE_INT,
unsigned HOST_WIDE_INT,
unsigned HOST_WIDE_INT,
rtx, int));
static rtx mask_rtx PARAMS ((enum machine_mode, int,
int, int));
static rtx lshift_value PARAMS ((enum machine_mode, rtx,
int, int));
static rtx extract_split_bit_field PARAMS ((rtx, unsigned HOST_WIDE_INT,
unsigned HOST_WIDE_INT, int));
static void do_cmp_and_jump PARAMS ((rtx, rtx, enum rtx_code,
enum machine_mode, rtx));
/* Non-zero means divides or modulus operations are relatively cheap for
powers of two, so don't use branches; emit the operation instead.
Usually, this will mean that the MD file will emit non-branch
sequences. */
static int sdiv_pow2_cheap, smod_pow2_cheap;
#ifndef SLOW_UNALIGNED_ACCESS
#define SLOW_UNALIGNED_ACCESS(MODE, ALIGN) STRICT_ALIGNMENT
#endif
/* For compilers that support multiple targets with different word sizes,
MAX_BITS_PER_WORD contains the biggest value of BITS_PER_WORD. An example
is the H8/300(H) compiler. */
#ifndef MAX_BITS_PER_WORD
#define MAX_BITS_PER_WORD BITS_PER_WORD
#endif
/* Reduce conditional compilation elsewhere. */
#ifndef HAVE_insv
#define HAVE_insv 0
#define CODE_FOR_insv CODE_FOR_nothing
#define gen_insv(a,b,c,d) NULL_RTX
#endif
#ifndef HAVE_extv
#define HAVE_extv 0
#define CODE_FOR_extv CODE_FOR_nothing
#define gen_extv(a,b,c,d) NULL_RTX
#endif
#ifndef HAVE_extzv
#define HAVE_extzv 0
#define CODE_FOR_extzv CODE_FOR_nothing
#define gen_extzv(a,b,c,d) NULL_RTX
#endif
/* Cost of various pieces of RTL. Note that some of these are indexed by
shift count and some by mode. */
static int add_cost, negate_cost, zero_cost;
static int shift_cost[MAX_BITS_PER_WORD];
static int shiftadd_cost[MAX_BITS_PER_WORD];
static int shiftsub_cost[MAX_BITS_PER_WORD];
static int mul_cost[NUM_MACHINE_MODES];
static int div_cost[NUM_MACHINE_MODES];
static int mul_widen_cost[NUM_MACHINE_MODES];
static int mul_highpart_cost[NUM_MACHINE_MODES];
void
init_expmed ()
{
/* This is "some random pseudo register" for purposes of calling recog
to see what insns exist. */
rtx reg = gen_rtx_REG (word_mode, 10000);
rtx shift_insn, shiftadd_insn, shiftsub_insn;
int dummy;
int m;
enum machine_mode mode, wider_mode;
start_sequence ();
reg = gen_rtx_REG (word_mode, 10000);
zero_cost = rtx_cost (const0_rtx, 0);
add_cost = rtx_cost (gen_rtx_PLUS (word_mode, reg, reg), SET);
shift_insn = emit_insn (gen_rtx_SET (VOIDmode, reg,
gen_rtx_ASHIFT (word_mode, reg,
const0_rtx)));
shiftadd_insn
= emit_insn (gen_rtx_SET (VOIDmode, reg,
gen_rtx_PLUS (word_mode,
gen_rtx_MULT (word_mode,
reg, const0_rtx),
reg)));
shiftsub_insn
= emit_insn (gen_rtx_SET (VOIDmode, reg,
gen_rtx_MINUS (word_mode,
gen_rtx_MULT (word_mode,
reg, const0_rtx),
reg)));
init_recog ();
shift_cost[0] = 0;
shiftadd_cost[0] = shiftsub_cost[0] = add_cost;
for (m = 1; m < MAX_BITS_PER_WORD; m++)
{
shift_cost[m] = shiftadd_cost[m] = shiftsub_cost[m] = 32000;
XEXP (SET_SRC (PATTERN (shift_insn)), 1) = GEN_INT (m);
if (recog (PATTERN (shift_insn), shift_insn, &dummy) >= 0)
shift_cost[m] = rtx_cost (SET_SRC (PATTERN (shift_insn)), SET);
XEXP (XEXP (SET_SRC (PATTERN (shiftadd_insn)), 0), 1)
= GEN_INT ((HOST_WIDE_INT) 1 << m);
if (recog (PATTERN (shiftadd_insn), shiftadd_insn, &dummy) >= 0)
shiftadd_cost[m] = rtx_cost (SET_SRC (PATTERN (shiftadd_insn)), SET);
XEXP (XEXP (SET_SRC (PATTERN (shiftsub_insn)), 0), 1)
= GEN_INT ((HOST_WIDE_INT) 1 << m);
if (recog (PATTERN (shiftsub_insn), shiftsub_insn, &dummy) >= 0)
shiftsub_cost[m] = rtx_cost (SET_SRC (PATTERN (shiftsub_insn)), SET);
}
negate_cost = rtx_cost (gen_rtx_NEG (word_mode, reg), SET);
sdiv_pow2_cheap
= (rtx_cost (gen_rtx_DIV (word_mode, reg, GEN_INT (32)), SET)
<= 2 * add_cost);
smod_pow2_cheap
= (rtx_cost (gen_rtx_MOD (word_mode, reg, GEN_INT (32)), SET)
<= 2 * add_cost);
for (mode = GET_CLASS_NARROWEST_MODE (MODE_INT);
mode != VOIDmode;
mode = GET_MODE_WIDER_MODE (mode))
{
reg = gen_rtx_REG (mode, 10000);
div_cost[(int) mode] = rtx_cost (gen_rtx_UDIV (mode, reg, reg), SET);
mul_cost[(int) mode] = rtx_cost (gen_rtx_MULT (mode, reg, reg), SET);
wider_mode = GET_MODE_WIDER_MODE (mode);
if (wider_mode != VOIDmode)
{
mul_widen_cost[(int) wider_mode]
= rtx_cost (gen_rtx_MULT (wider_mode,
gen_rtx_ZERO_EXTEND (wider_mode, reg),
gen_rtx_ZERO_EXTEND (wider_mode, reg)),
SET);
mul_highpart_cost[(int) mode]
= rtx_cost (gen_rtx_TRUNCATE
(mode,
gen_rtx_LSHIFTRT (wider_mode,
gen_rtx_MULT (wider_mode,
gen_rtx_ZERO_EXTEND
(wider_mode, reg),
gen_rtx_ZERO_EXTEND
(wider_mode, reg)),
GEN_INT (GET_MODE_BITSIZE (mode)))),
SET);
}
}
end_sequence ();
}
/* Return an rtx representing minus the value of X.
MODE is the intended mode of the result,
useful if X is a CONST_INT. */
rtx
negate_rtx (mode, x)
enum machine_mode mode;
rtx x;
{
rtx result = simplify_unary_operation (NEG, mode, x, mode);
if (result == 0)
result = expand_unop (mode, neg_optab, x, NULL_RTX, 0);
return result;
}
/* Report on the availability of insv/extv/extzv and the desired mode
of each of their operands. Returns MAX_MACHINE_MODE if HAVE_foo
is false; else the mode of the specified operand. If OPNO is -1,
all the caller cares about is whether the insn is available. */
enum machine_mode
mode_for_extraction (pattern, opno)
enum extraction_pattern pattern;
int opno;
{
const struct insn_data *data;
switch (pattern)
{
case EP_insv:
if (HAVE_insv)
{
data = &insn_data[CODE_FOR_insv];
break;
}
return MAX_MACHINE_MODE;
case EP_extv:
if (HAVE_extv)
{
data = &insn_data[CODE_FOR_extv];
break;
}
return MAX_MACHINE_MODE;
case EP_extzv:
if (HAVE_extzv)
{
data = &insn_data[CODE_FOR_extzv];
break;
}
return MAX_MACHINE_MODE;
default:
abort ();
}
if (opno == -1)
return VOIDmode;
/* Everyone who uses this function used to follow it with
if (result == VOIDmode) result = word_mode; */
if (data->operand[opno].mode == VOIDmode)
return word_mode;
return data->operand[opno].mode;
}
/* Generate code to store value from rtx VALUE
into a bit-field within structure STR_RTX
containing BITSIZE bits starting at bit BITNUM.
FIELDMODE is the machine-mode of the FIELD_DECL node for this field.
ALIGN is the alignment that STR_RTX is known to have.
TOTAL_SIZE is the size of the structure in bytes, or -1 if varying. */
/* ??? Note that there are two different ideas here for how
to determine the size to count bits within, for a register.
One is BITS_PER_WORD, and the other is the size of operand 3
of the insv pattern.
If operand 3 of the insv pattern is VOIDmode, then we will use BITS_PER_WORD
else, we use the mode of operand 3. */
rtx
store_bit_field (str_rtx, bitsize, bitnum, fieldmode, value, total_size)
rtx str_rtx;
unsigned HOST_WIDE_INT bitsize;
unsigned HOST_WIDE_INT bitnum;
enum machine_mode fieldmode;
rtx value;
HOST_WIDE_INT total_size;
{
unsigned int unit
= (GET_CODE (str_rtx) == MEM) ? BITS_PER_UNIT : BITS_PER_WORD;
unsigned HOST_WIDE_INT offset = bitnum / unit;
unsigned HOST_WIDE_INT bitpos = bitnum % unit;
rtx op0 = str_rtx;
int byte_offset;
enum machine_mode op_mode = mode_for_extraction (EP_insv, 3);
/* Discount the part of the structure before the desired byte.
We need to know how many bytes are safe to reference after it. */
if (total_size >= 0)
total_size -= (bitpos / BIGGEST_ALIGNMENT
* (BIGGEST_ALIGNMENT / BITS_PER_UNIT));
while (GET_CODE (op0) == SUBREG)
{
/* The following line once was done only if WORDS_BIG_ENDIAN,
but I think that is a mistake. WORDS_BIG_ENDIAN is
meaningful at a much higher level; when structures are copied
between memory and regs, the higher-numbered regs
always get higher addresses. */
offset += (SUBREG_BYTE (op0) / UNITS_PER_WORD);
/* We used to adjust BITPOS here, but now we do the whole adjustment
right after the loop. */
op0 = SUBREG_REG (op0);
}
value = protect_from_queue (value, 0);
if (flag_force_mem)
{
int old_generating_concat_p = generating_concat_p;
generating_concat_p = 0;
value = force_not_mem (value);
generating_concat_p = old_generating_concat_p;
}
/* If the target is a register, overwriting the entire object, or storing
a full-word or multi-word field can be done with just a SUBREG.
If the target is memory, storing any naturally aligned field can be
done with a simple store. For targets that support fast unaligned
memory, any naturally sized, unit aligned field can be done directly. */
byte_offset = (bitnum % BITS_PER_WORD) / BITS_PER_UNIT
+ (offset * UNITS_PER_WORD);
if (bitpos == 0
&& bitsize == GET_MODE_BITSIZE (fieldmode)
&& (GET_CODE (op0) != MEM
? ((GET_MODE_SIZE (fieldmode) >= UNITS_PER_WORD
|| GET_MODE_SIZE (GET_MODE (op0)) == GET_MODE_SIZE (fieldmode))
&& byte_offset % GET_MODE_SIZE (fieldmode) == 0)
: (! SLOW_UNALIGNED_ACCESS (fieldmode, MEM_ALIGN (op0))
|| (offset * BITS_PER_UNIT % bitsize == 0
&& MEM_ALIGN (op0) % GET_MODE_BITSIZE (fieldmode) == 0))))
{
if (GET_MODE (op0) != fieldmode)
{
if (GET_CODE (op0) == SUBREG)
{
if (GET_MODE (SUBREG_REG (op0)) == fieldmode
|| GET_MODE_CLASS (fieldmode) == MODE_INT
|| GET_MODE_CLASS (fieldmode) == MODE_PARTIAL_INT)
op0 = SUBREG_REG (op0);
else
/* Else we've got some float mode source being extracted into
a different float mode destination -- this combination of
subregs results in Severe Tire Damage. */
abort ();
}
if (GET_CODE (op0) == REG)
op0 = gen_rtx_SUBREG (fieldmode, op0, byte_offset);
else
op0 = adjust_address (op0, fieldmode, offset);
}
emit_move_insn (op0, value);
return value;
}
/* Make sure we are playing with integral modes. Pun with subregs
if we aren't. This must come after the entire register case above,
since that case is valid for any mode. The following cases are only
valid for integral modes. */
{
enum machine_mode imode = int_mode_for_mode (GET_MODE (op0));
if (imode != GET_MODE (op0))
{
if (GET_CODE (op0) == MEM)
op0 = adjust_address (op0, imode, 0);
else if (imode != BLKmode)
op0 = gen_lowpart (imode, op0);
else
abort ();
}
}
/* We may be accessing data outside the field, which means
we can alias adjacent data. */
if (GET_CODE (op0) == MEM)
{
op0 = shallow_copy_rtx (op0);
set_mem_alias_set (op0, 0);
set_mem_expr (op0, 0);
}
/* If OP0 is a register, BITPOS must count within a word.
But as we have it, it counts within whatever size OP0 now has.
On a bigendian machine, these are not the same, so convert. */
if (BYTES_BIG_ENDIAN
&& GET_CODE (op0) != MEM
&& unit > GET_MODE_BITSIZE (GET_MODE (op0)))
bitpos += unit - GET_MODE_BITSIZE (GET_MODE (op0));
/* Storing an lsb-aligned field in a register
can be done with a movestrict instruction. */
if (GET_CODE (op0) != MEM
&& (BYTES_BIG_ENDIAN ? bitpos + bitsize == unit : bitpos == 0)
&& bitsize == GET_MODE_BITSIZE (fieldmode)
&& (movstrict_optab->handlers[(int) fieldmode].insn_code
!= CODE_FOR_nothing))
{
int icode = movstrict_optab->handlers[(int) fieldmode].insn_code;
/* Get appropriate low part of the value being stored. */
if (GET_CODE (value) == CONST_INT || GET_CODE (value) == REG)
value = gen_lowpart (fieldmode, value);
else if (!(GET_CODE (value) == SYMBOL_REF
|| GET_CODE (value) == LABEL_REF
|| GET_CODE (value) == CONST))
value = convert_to_mode (fieldmode, value, 0);
if (! (*insn_data[icode].operand[1].predicate) (value, fieldmode))
value = copy_to_mode_reg (fieldmode, value);
if (GET_CODE (op0) == SUBREG)
{
if (GET_MODE (SUBREG_REG (op0)) == fieldmode
|| GET_MODE_CLASS (fieldmode) == MODE_INT
|| GET_MODE_CLASS (fieldmode) == MODE_PARTIAL_INT)
op0 = SUBREG_REG (op0);
else
/* Else we've got some float mode source being extracted into
a different float mode destination -- this combination of
subregs results in Severe Tire Damage. */
abort ();
}
emit_insn (GEN_FCN (icode)
(gen_rtx_SUBREG (fieldmode, op0,
(bitnum % BITS_PER_WORD) / BITS_PER_UNIT
+ (offset * UNITS_PER_WORD)),
value));
return value;
}
/* Handle fields bigger than a word. */
if (bitsize > BITS_PER_WORD)
{
/* Here we transfer the words of the field
in the order least significant first.
This is because the most significant word is the one which may
be less than full.
However, only do that if the value is not BLKmode. */
unsigned int backwards = WORDS_BIG_ENDIAN && fieldmode != BLKmode;
unsigned int nwords = (bitsize + (BITS_PER_WORD - 1)) / BITS_PER_WORD;
unsigned int i;
/* This is the mode we must force value to, so that there will be enough
subwords to extract. Note that fieldmode will often (always?) be
VOIDmode, because that is what store_field uses to indicate that this
is a bit field, but passing VOIDmode to operand_subword_force will
result in an abort. */
fieldmode = smallest_mode_for_size (nwords * BITS_PER_WORD, MODE_INT);
for (i = 0; i < nwords; i++)
{
/* If I is 0, use the low-order word in both field and target;
if I is 1, use the next to lowest word; and so on. */
unsigned int wordnum = (backwards ? nwords - i - 1 : i);
unsigned int bit_offset = (backwards
? MAX ((int) bitsize - ((int) i + 1)
* BITS_PER_WORD,
0)
: (int) i * BITS_PER_WORD);
store_bit_field (op0, MIN (BITS_PER_WORD,
bitsize - i * BITS_PER_WORD),
bitnum + bit_offset, word_mode,
operand_subword_force (value, wordnum,
(GET_MODE (value) == VOIDmode
? fieldmode
: GET_MODE (value))),
total_size);
}
return value;
}
/* From here on we can assume that the field to be stored in is
a full-word (whatever type that is), since it is shorter than a word. */
/* OFFSET is the number of words or bytes (UNIT says which)
from STR_RTX to the first word or byte containing part of the field. */
if (GET_CODE (op0) != MEM)
{
if (offset != 0
|| GET_MODE_SIZE (GET_MODE (op0)) > UNITS_PER_WORD)
{
if (GET_CODE (op0) != REG)
{
/* Since this is a destination (lvalue), we can't copy it to a
pseudo. We can trivially remove a SUBREG that does not
change the size of the operand. Such a SUBREG may have been
added above. Otherwise, abort. */
if (GET_CODE (op0) == SUBREG
&& (GET_MODE_SIZE (GET_MODE (op0))
== GET_MODE_SIZE (GET_MODE (SUBREG_REG (op0)))))
op0 = SUBREG_REG (op0);
else
abort ();
}
op0 = gen_rtx_SUBREG (mode_for_size (BITS_PER_WORD, MODE_INT, 0),
op0, (offset * UNITS_PER_WORD));
}
offset = 0;
}
else
op0 = protect_from_queue (op0, 1);
/* If VALUE is a floating-point mode, access it as an integer of the
corresponding size. This can occur on a machine with 64 bit registers
that uses SFmode for float. This can also occur for unaligned float
structure fields. */
if (GET_MODE_CLASS (GET_MODE (value)) != MODE_INT
&& GET_MODE_CLASS (GET_MODE (value)) != MODE_PARTIAL_INT)
value = gen_lowpart (word_mode, value);
/* Now OFFSET is nonzero only if OP0 is memory
and is therefore always measured in bytes. */
if (HAVE_insv
&& GET_MODE (value) != BLKmode
&& !(bitsize == 1 && GET_CODE (value) == CONST_INT)
/* Ensure insv's size is wide enough for this field. */
&& (GET_MODE_BITSIZE (op_mode) >= bitsize)
&& ! ((GET_CODE (op0) == REG || GET_CODE (op0) == SUBREG)
&& (bitsize + bitpos > GET_MODE_BITSIZE (op_mode))))
{
int xbitpos = bitpos;
rtx value1;
rtx xop0 = op0;
rtx last = get_last_insn ();
rtx pat;
enum machine_mode maxmode = mode_for_extraction (EP_insv, 3);
int save_volatile_ok = volatile_ok;
volatile_ok = 1;
/* If this machine's insv can only insert into a register, copy OP0
into a register and save it back later. */
/* This used to check flag_force_mem, but that was a serious
de-optimization now that flag_force_mem is enabled by -O2. */
if (GET_CODE (op0) == MEM
&& ! ((*insn_data[(int) CODE_FOR_insv].operand[0].predicate)
(op0, VOIDmode)))
{
rtx tempreg;
enum machine_mode bestmode;
/* Get the mode to use for inserting into this field. If OP0 is
BLKmode, get the smallest mode consistent with the alignment. If
OP0 is a non-BLKmode object that is no wider than MAXMODE, use its
mode. Otherwise, use the smallest mode containing the field. */
if (GET_MODE (op0) == BLKmode
|| GET_MODE_SIZE (GET_MODE (op0)) > GET_MODE_SIZE (maxmode))
bestmode
= get_best_mode (bitsize, bitnum, MEM_ALIGN (op0), maxmode,
MEM_VOLATILE_P (op0));
else
bestmode = GET_MODE (op0);
if (bestmode == VOIDmode
|| (SLOW_UNALIGNED_ACCESS (bestmode, MEM_ALIGN (op0))
&& GET_MODE_BITSIZE (bestmode) > MEM_ALIGN (op0)))
goto insv_loses;
/* Adjust address to point to the containing unit of that mode.
Compute offset as multiple of this unit, counting in bytes. */
unit = GET_MODE_BITSIZE (bestmode);
offset = (bitnum / unit) * GET_MODE_SIZE (bestmode);
bitpos = bitnum % unit;
op0 = adjust_address (op0, bestmode, offset);
/* Fetch that unit, store the bitfield in it, then store
the unit. */
tempreg = copy_to_reg (op0);
store_bit_field (tempreg, bitsize, bitpos, fieldmode, value,
total_size);
emit_move_insn (op0, tempreg);
return value;
}
volatile_ok = save_volatile_ok;
/* Add OFFSET into OP0's address. */
if (GET_CODE (xop0) == MEM)
xop0 = adjust_address (xop0, byte_mode, offset);
/* If xop0 is a register, we need it in MAXMODE
to make it acceptable to the format of insv. */
if (GET_CODE (xop0) == SUBREG)
/* We can't just change the mode, because this might clobber op0,
and we will need the original value of op0 if insv fails. */
xop0 = gen_rtx_SUBREG (maxmode, SUBREG_REG (xop0), SUBREG_BYTE (xop0));
if (GET_CODE (xop0) == REG && GET_MODE (xop0) != maxmode)
xop0 = gen_rtx_SUBREG (maxmode, xop0, 0);
/* On big-endian machines, we count bits from the most significant.
If the bit field insn does not, we must invert. */
if (BITS_BIG_ENDIAN != BYTES_BIG_ENDIAN)
xbitpos = unit - bitsize - xbitpos;
/* We have been counting XBITPOS within UNIT.
Count instead within the size of the register. */
if (BITS_BIG_ENDIAN && GET_CODE (xop0) != MEM)
xbitpos += GET_MODE_BITSIZE (maxmode) - unit;
unit = GET_MODE_BITSIZE (maxmode);
/* Convert VALUE to maxmode (which insv insn wants) in VALUE1. */
value1 = value;
if (GET_MODE (value) != maxmode)
{
if (GET_MODE_BITSIZE (GET_MODE (value)) >= bitsize)
{
/* Optimization: Don't bother really extending VALUE
if it has all the bits we will actually use. However,
if we must narrow it, be sure we do it correctly. */
if (GET_MODE_SIZE (GET_MODE (value)) < GET_MODE_SIZE (maxmode))
{
rtx tmp;
tmp = simplify_subreg (maxmode, value1, GET_MODE (value), 0);
if (! tmp)
tmp = simplify_gen_subreg (maxmode,
force_reg (GET_MODE (value),
value1),
GET_MODE (value), 0);
value1 = tmp;
}
else
value1 = gen_lowpart (maxmode, value1);
}
else if (GET_CODE (value) == CONST_INT)
value1 = gen_int_mode (INTVAL (value), maxmode);
else if (!CONSTANT_P (value))
/* Parse phase is supposed to make VALUE's data type
match that of the component reference, which is a type
at least as wide as the field; so VALUE should have
a mode that corresponds to that type. */
abort ();
}
/* If this machine's insv insists on a register,
get VALUE1 into a register. */
if (! ((*insn_data[(int) CODE_FOR_insv].operand[3].predicate)
(value1, maxmode)))
value1 = force_reg (maxmode, value1);
pat = gen_insv (xop0, GEN_INT (bitsize), GEN_INT (xbitpos), value1);
if (pat)
emit_insn (pat);
else
{
delete_insns_since (last);
store_fixed_bit_field (op0, offset, bitsize, bitpos, value);
}
}
else
insv_loses:
/* Insv is not available; store using shifts and boolean ops. */
store_fixed_bit_field (op0, offset, bitsize, bitpos, value);
return value;
}
/* Use shifts and boolean operations to store VALUE
into a bit field of width BITSIZE
in a memory location specified by OP0 except offset by OFFSET bytes.
(OFFSET must be 0 if OP0 is a register.)
The field starts at position BITPOS within the byte.
(If OP0 is a register, it may be a full word or a narrower mode,
but BITPOS still counts within a full word,
which is significant on bigendian machines.)
Note that protect_from_queue has already been done on OP0 and VALUE. */
static void
store_fixed_bit_field (op0, offset, bitsize, bitpos, value)
rtx op0;
unsigned HOST_WIDE_INT offset, bitsize, bitpos;
rtx value;
{
enum machine_mode mode;
unsigned int total_bits = BITS_PER_WORD;
rtx subtarget, temp;
int all_zero = 0;
int all_one = 0;
/* There is a case not handled here:
a structure with a known alignment of just a halfword
and a field split across two aligned halfwords within the structure.
Or likewise a structure with a known alignment of just a byte
and a field split across two bytes.
Such cases are not supposed to be able to occur. */
if (GET_CODE (op0) == REG || GET_CODE (op0) == SUBREG)
{
if (offset != 0)
abort ();
/* Special treatment for a bit field split across two registers. */
if (bitsize + bitpos > BITS_PER_WORD)
{
store_split_bit_field (op0, bitsize, bitpos, value);
return;
}
}
else
{
/* Get the proper mode to use for this field. We want a mode that
includes the entire field. If such a mode would be larger than
a word, we won't be doing the extraction the normal way.
We don't want a mode bigger than the destination. */
mode = GET_MODE (op0);
if (GET_MODE_BITSIZE (mode) == 0
|| GET_MODE_BITSIZE (mode) > GET_MODE_BITSIZE (word_mode))
mode = word_mode;
mode = get_best_mode (bitsize, bitpos + offset * BITS_PER_UNIT,
MEM_ALIGN (op0), mode, MEM_VOLATILE_P (op0));
if (mode == VOIDmode)
{
/* The only way this should occur is if the field spans word
boundaries. */
store_split_bit_field (op0, bitsize, bitpos + offset * BITS_PER_UNIT,
value);
return;
}
total_bits = GET_MODE_BITSIZE (mode);
/* Make sure bitpos is valid for the chosen mode. Adjust BITPOS to
be in the range 0 to total_bits-1, and put any excess bytes in
OFFSET. */
if (bitpos >= total_bits)
{
offset += (bitpos / total_bits) * (total_bits / BITS_PER_UNIT);
bitpos -= ((bitpos / total_bits) * (total_bits / BITS_PER_UNIT)
* BITS_PER_UNIT);
}
/* Get ref to an aligned byte, halfword, or word containing the field.
Adjust BITPOS to be position within a word,
and OFFSET to be the offset of that word.
Then alter OP0 to refer to that word. */
bitpos += (offset % (total_bits / BITS_PER_UNIT)) * BITS_PER_UNIT;
offset -= (offset % (total_bits / BITS_PER_UNIT));
op0 = adjust_address (op0, mode, offset);
}
mode = GET_MODE (op0);
/* Now MODE is either some integral mode for a MEM as OP0,
or is a full-word for a REG as OP0. TOTAL_BITS corresponds.
The bit field is contained entirely within OP0.
BITPOS is the starting bit number within OP0.
(OP0's mode may actually be narrower than MODE.) */
if (BYTES_BIG_ENDIAN)
/* BITPOS is the distance between our msb
and that of the containing datum.
Convert it to the distance from the lsb. */
bitpos = total_bits - bitsize - bitpos;
/* Now BITPOS is always the distance between our lsb
and that of OP0. */
/* Shift VALUE left by BITPOS bits. If VALUE is not constant,
we must first convert its mode to MODE. */
if (GET_CODE (value) == CONST_INT)
{
HOST_WIDE_INT v = INTVAL (value);
if (bitsize < HOST_BITS_PER_WIDE_INT)
v &= ((HOST_WIDE_INT) 1 << bitsize) - 1;
if (v == 0)
all_zero = 1;
else if ((bitsize < HOST_BITS_PER_WIDE_INT
&& v == ((HOST_WIDE_INT) 1 << bitsize) - 1)
|| (bitsize == HOST_BITS_PER_WIDE_INT && v == -1))
all_one = 1;
value = lshift_value (mode, value, bitpos, bitsize);
}
else
{
int must_and = (GET_MODE_BITSIZE (GET_MODE (value)) != bitsize
&& bitpos + bitsize != GET_MODE_BITSIZE (mode));
if (GET_MODE (value) != mode)
{
if ((GET_CODE (value) == REG || GET_CODE (value) == SUBREG)
&& GET_MODE_SIZE (mode) < GET_MODE_SIZE (GET_MODE (value)))
value = gen_lowpart (mode, value);
else
value = convert_to_mode (mode, value, 1);
}
if (must_and)
value = expand_binop (mode, and_optab, value,
mask_rtx (mode, 0, bitsize, 0),
NULL_RTX, 1, OPTAB_LIB_WIDEN);
if (bitpos > 0)
value = expand_shift (LSHIFT_EXPR, mode, value,
build_int_2 (bitpos, 0), NULL_RTX, 1);
}
/* Now clear the chosen bits in OP0,
except that if VALUE is -1 we need not bother. */
subtarget = (GET_CODE (op0) == REG || ! flag_force_mem) ? op0 : 0;
if (! all_one)
{
temp = expand_binop (mode, and_optab, op0,
mask_rtx (mode, bitpos, bitsize, 1),
subtarget, 1, OPTAB_LIB_WIDEN);
subtarget = temp;
}
else
temp = op0;
/* Now logical-or VALUE into OP0, unless it is zero. */
if (! all_zero)
temp = expand_binop (mode, ior_optab, temp, value,
subtarget, 1, OPTAB_LIB_WIDEN);
if (op0 != temp)
emit_move_insn (op0, temp);
}
/* Store a bit field that is split across multiple accessible memory objects.
OP0 is the REG, SUBREG or MEM rtx for the first of the objects.
BITSIZE is the field width; BITPOS the position of its first bit
(within the word).
VALUE is the value to store.
This does not yet handle fields wider than BITS_PER_WORD. */
static void
store_split_bit_field (op0, bitsize, bitpos, value)
rtx op0;
unsigned HOST_WIDE_INT bitsize, bitpos;
rtx value;
{
unsigned int unit;
unsigned int bitsdone = 0;
/* Make sure UNIT isn't larger than BITS_PER_WORD, we can only handle that
much at a time. */
if (GET_CODE (op0) == REG || GET_CODE (op0) == SUBREG)
unit = BITS_PER_WORD;
else
unit = MIN (MEM_ALIGN (op0), BITS_PER_WORD);
/* If VALUE is a constant other than a CONST_INT, get it into a register in
WORD_MODE. If we can do this using gen_lowpart_common, do so. Note
that VALUE might be a floating-point constant. */
if (CONSTANT_P (value) && GET_CODE (value) != CONST_INT)
{
rtx word = gen_lowpart_common (word_mode, value);
if (word && (value != word))
value = word;
else
value = gen_lowpart_common (word_mode,
force_reg (GET_MODE (value) != VOIDmode
? GET_MODE (value)
: word_mode, value));
}
else if (GET_CODE (value) == ADDRESSOF)
value = copy_to_reg (value);
while (bitsdone < bitsize)
{
unsigned HOST_WIDE_INT thissize;
rtx part, word;
unsigned HOST_WIDE_INT thispos;
unsigned HOST_WIDE_INT offset;
offset = (bitpos + bitsdone) / unit;
thispos = (bitpos + bitsdone) % unit;
/* THISSIZE must not overrun a word boundary. Otherwise,
store_fixed_bit_field will call us again, and we will mutually
recurse forever. */
thissize = MIN (bitsize - bitsdone, BITS_PER_WORD);
thissize = MIN (thissize, unit - thispos);
if (BYTES_BIG_ENDIAN)
{
int total_bits;
/* We must do an endian conversion exactly the same way as it is
done in extract_bit_field, so that the two calls to
extract_fixed_bit_field will have comparable arguments. */
if (GET_CODE (value) != MEM || GET_MODE (value) == BLKmode)
total_bits = BITS_PER_WORD;
else
total_bits = GET_MODE_BITSIZE (GET_MODE (value));
/* Fetch successively less significant portions. */
if (GET_CODE (value) == CONST_INT)
part = GEN_INT (((unsigned HOST_WIDE_INT) (INTVAL (value))
>> (bitsize - bitsdone - thissize))
& (((HOST_WIDE_INT) 1 << thissize) - 1));
else
/* The args are chosen so that the last part includes the
lsb. Give extract_bit_field the value it needs (with
endianness compensation) to fetch the piece we want. */
part = extract_fixed_bit_field (word_mode, value, 0, thissize,
total_bits - bitsize + bitsdone,
NULL_RTX, 1);
}
else
{
/* Fetch successively more significant portions. */
if (GET_CODE (value) == CONST_INT)
part = GEN_INT (((unsigned HOST_WIDE_INT) (INTVAL (value))
>> bitsdone)
& (((HOST_WIDE_INT) 1 << thissize) - 1));
else
part = extract_fixed_bit_field (word_mode, value, 0, thissize,
bitsdone, NULL_RTX, 1);
}
/* If OP0 is a register, then handle OFFSET here.
When handling multiword bitfields, extract_bit_field may pass
down a word_mode SUBREG of a larger REG for a bitfield that actually
crosses a word boundary. Thus, for a SUBREG, we must find
the current word starting from the base register. */
if (GET_CODE (op0) == SUBREG)
{
int word_offset = (SUBREG_BYTE (op0) / UNITS_PER_WORD) + offset;
word = operand_subword_force (SUBREG_REG (op0), word_offset,
GET_MODE (SUBREG_REG (op0)));
offset = 0;
}
else if (GET_CODE (op0) == REG)
{
word = operand_subword_force (op0, offset, GET_MODE (op0));
offset = 0;
}
else
word = op0;
/* OFFSET is in UNITs, and UNIT is in bits.
store_fixed_bit_field wants offset in bytes. */
store_fixed_bit_field (word, offset * unit / BITS_PER_UNIT, thissize,
thispos, part);
bitsdone += thissize;
}
}
/* Generate code to extract a byte-field from STR_RTX
containing BITSIZE bits, starting at BITNUM,
and put it in TARGET if possible (if TARGET is nonzero).
Regardless of TARGET, we return the rtx for where the value is placed.
It may be a QUEUED.
STR_RTX is the structure containing the byte (a REG or MEM).
UNSIGNEDP is nonzero if this is an unsigned bit field.
MODE is the natural mode of the field value once extracted.
TMODE is the mode the caller would like the value to have;
but the value may be returned with type MODE instead.
TOTAL_SIZE is the size in bytes of the containing structure,
or -1 if varying.
If a TARGET is specified and we can store in it at no extra cost,
we do so, and return TARGET.
Otherwise, we return a REG of mode TMODE or MODE, with TMODE preferred
if they are equally easy. */
rtx
extract_bit_field (str_rtx, bitsize, bitnum, unsignedp,
target, mode, tmode, total_size)
rtx str_rtx;
unsigned HOST_WIDE_INT bitsize;
unsigned HOST_WIDE_INT bitnum;
int unsignedp;
rtx target;
enum machine_mode mode, tmode;
HOST_WIDE_INT total_size;
{
unsigned int unit
= (GET_CODE (str_rtx) == MEM) ? BITS_PER_UNIT : BITS_PER_WORD;
unsigned HOST_WIDE_INT offset = bitnum / unit;
unsigned HOST_WIDE_INT bitpos = bitnum % unit;
rtx op0 = str_rtx;
rtx spec_target = target;
rtx spec_target_subreg = 0;
enum machine_mode int_mode;
enum machine_mode extv_mode = mode_for_extraction (EP_extv, 0);
enum machine_mode extzv_mode = mode_for_extraction (EP_extzv, 0);
enum machine_mode mode1;
int byte_offset;
/* Discount the part of the structure before the desired byte.
We need to know how many bytes are safe to reference after it. */
if (total_size >= 0)
total_size -= (bitpos / BIGGEST_ALIGNMENT
* (BIGGEST_ALIGNMENT / BITS_PER_UNIT));
if (tmode == VOIDmode)
tmode = mode;
while (GET_CODE (op0) == SUBREG)
{
int outer_size = GET_MODE_BITSIZE (GET_MODE (op0));
int inner_size = GET_MODE_BITSIZE (GET_MODE (SUBREG_REG (op0)));
offset += SUBREG_BYTE (op0) / UNITS_PER_WORD;
inner_size = MIN (inner_size, BITS_PER_WORD);
if (BYTES_BIG_ENDIAN && (outer_size < inner_size))
{
bitpos += inner_size - outer_size;
if (bitpos > unit)
{
offset += (bitpos / unit);
bitpos %= unit;
}
}
op0 = SUBREG_REG (op0);
}
if (GET_CODE (op0) == REG
&& mode == GET_MODE (op0)
&& bitnum == 0
&& bitsize == GET_MODE_BITSIZE (GET_MODE (op0)))
{
/* We're trying to extract a full register from itself. */
return op0;
}
/* Make sure we are playing with integral modes. Pun with subregs
if we aren't. */
{
enum machine_mode imode = int_mode_for_mode (GET_MODE (op0));
if (imode != GET_MODE (op0))
{
if (GET_CODE (op0) == MEM)
op0 = adjust_address (op0, imode, 0);
else if (imode != BLKmode)
op0 = gen_lowpart (imode, op0);
else
abort ();
}
}
/* We may be accessing data outside the field, which means
we can alias adjacent data. */
if (GET_CODE (op0) == MEM)
{
op0 = shallow_copy_rtx (op0);
set_mem_alias_set (op0, 0);
set_mem_expr (op0, 0);
}
/* ??? We currently assume TARGET is at least as big as BITSIZE.
If that's wrong, the solution is to test for it and set TARGET to 0
if needed. */
/* If OP0 is a register, BITPOS must count within a word.
But as we have it, it counts within whatever size OP0 now has.
On a bigendian machine, these are not the same, so convert. */
if (BYTES_BIG_ENDIAN
&& GET_CODE (op0) != MEM
&& unit > GET_MODE_BITSIZE (GET_MODE (op0)))
bitpos += unit - GET_MODE_BITSIZE (GET_MODE (op0));
/* Extracting a full-word or multi-word value
from a structure in a register or aligned memory.
This can be done with just SUBREG.
So too extracting a subword value in
the least significant part of the register. */
byte_offset = (bitnum % BITS_PER_WORD) / BITS_PER_UNIT
+ (offset * UNITS_PER_WORD);
mode1 = (VECTOR_MODE_P (tmode)
? mode
: mode_for_size (bitsize, GET_MODE_CLASS (tmode), 0));
if (((GET_CODE (op0) != MEM
&& TRULY_NOOP_TRUNCATION (GET_MODE_BITSIZE (mode),
GET_MODE_BITSIZE (GET_MODE (op0)))
&& GET_MODE_SIZE (mode1) != 0
&& byte_offset % GET_MODE_SIZE (mode1) == 0)
|| (GET_CODE (op0) == MEM
&& (! SLOW_UNALIGNED_ACCESS (mode, MEM_ALIGN (op0))
|| (offset * BITS_PER_UNIT % bitsize == 0
&& MEM_ALIGN (op0) % bitsize == 0))))
&& ((bitsize >= BITS_PER_WORD && bitsize == GET_MODE_BITSIZE (mode)
&& bitpos % BITS_PER_WORD == 0)
|| (mode_for_size (bitsize, GET_MODE_CLASS (tmode), 0) != BLKmode
/* ??? The big endian test here is wrong. This is correct
if the value is in a register, and if mode_for_size is not
the same mode as op0. This causes us to get unnecessarily
inefficient code from the Thumb port when -mbig-endian. */
&& (BYTES_BIG_ENDIAN
? bitpos + bitsize == BITS_PER_WORD
: bitpos == 0))))
{
if (mode1 != GET_MODE (op0))
{
if (GET_CODE (op0) == SUBREG)
{
if (GET_MODE (SUBREG_REG (op0)) == mode1
|| GET_MODE_CLASS (mode1) == MODE_INT
|| GET_MODE_CLASS (mode1) == MODE_PARTIAL_INT)
op0 = SUBREG_REG (op0);
else
/* Else we've got some float mode source being extracted into
a different float mode destination -- this combination of
subregs results in Severe Tire Damage. */
abort ();
}
if (GET_CODE (op0) == REG)
op0 = gen_rtx_SUBREG (mode1, op0, byte_offset);
else
op0 = adjust_address (op0, mode1, offset);
}
if (mode1 != mode)
return convert_to_mode (tmode, op0, unsignedp);
return op0;
}
/* Handle fields bigger than a word. */
if (bitsize > BITS_PER_WORD)
{
/* Here we transfer the words of the field
in the order least significant first.
This is because the most significant word is the one which may
be less than full. */
unsigned int nwords = (bitsize + (BITS_PER_WORD - 1)) / BITS_PER_WORD;
unsigned int i;
if (target == 0 || GET_CODE (target) != REG)
target = gen_reg_rtx (mode);
/* Indicate for flow that the entire target reg is being set. */
emit_insn (gen_rtx_CLOBBER (VOIDmode, target));
for (i = 0; i < nwords; i++)
{
/* If I is 0, use the low-order word in both field and target;
if I is 1, use the next to lowest word; and so on. */
/* Word number in TARGET to use. */
unsigned int wordnum
= (WORDS_BIG_ENDIAN
? GET_MODE_SIZE (GET_MODE (target)) / UNITS_PER_WORD - i - 1
: i);
/* Offset from start of field in OP0. */
unsigned int bit_offset = (WORDS_BIG_ENDIAN
? MAX (0, ((int) bitsize - ((int) i + 1)
* (int) BITS_PER_WORD))
: (int) i * BITS_PER_WORD);
rtx target_part = operand_subword (target, wordnum, 1, VOIDmode);
rtx result_part
= extract_bit_field (op0, MIN (BITS_PER_WORD,
bitsize - i * BITS_PER_WORD),
bitnum + bit_offset, 1, target_part, mode,
word_mode, total_size);
if (target_part == 0)
abort ();
if (result_part != target_part)
emit_move_insn (target_part, result_part);
}
if (unsignedp)
{
/* Unless we've filled TARGET, the upper regs in a multi-reg value
need to be zero'd out. */
if (GET_MODE_SIZE (GET_MODE (target)) > nwords * UNITS_PER_WORD)
{
unsigned int i, total_words;
total_words = GET_MODE_SIZE (GET_MODE (target)) / UNITS_PER_WORD;
for (i = nwords; i < total_words; i++)
emit_move_insn
(operand_subword (target,
WORDS_BIG_ENDIAN ? total_words - i - 1 : i,
1, VOIDmode),
const0_rtx);
}
return target;
}
/* Signed bit field: sign-extend with two arithmetic shifts. */
target = expand_shift (LSHIFT_EXPR, mode, target,
build_int_2 (GET_MODE_BITSIZE (mode) - bitsize, 0),
NULL_RTX, 0);
return expand_shift (RSHIFT_EXPR, mode, target,
build_int_2 (GET_MODE_BITSIZE (mode) - bitsize, 0),
NULL_RTX, 0);
}
/* From here on we know the desired field is smaller than a word. */
/* Check if there is a correspondingly-sized integer field, so we can
safely extract it as one size of integer, if necessary; then
truncate or extend to the size that is wanted; then use SUBREGs or
convert_to_mode to get one of the modes we really wanted. */
int_mode = int_mode_for_mode (tmode);
if (int_mode == BLKmode)
int_mode = int_mode_for_mode (mode);
if (int_mode == BLKmode)
abort (); /* Should probably push op0 out to memory and then
do a load. */
/* OFFSET is the number of words or bytes (UNIT says which)
from STR_RTX to the first word or byte containing part of the field. */
if (GET_CODE (op0) != MEM)
{
if (offset != 0
|| GET_MODE_SIZE (GET_MODE (op0)) > UNITS_PER_WORD)
{
if (GET_CODE (op0) != REG)
op0 = copy_to_reg (op0);
op0 = gen_rtx_SUBREG (mode_for_size (BITS_PER_WORD, MODE_INT, 0),
op0, (offset * UNITS_PER_WORD));
}
offset = 0;
}
else
op0 = protect_from_queue (str_rtx, 1);
/* Now OFFSET is nonzero only for memory operands. */
if (unsignedp)
{
if (HAVE_extzv
&& (GET_MODE_BITSIZE (extzv_mode) >= bitsize)
&& ! ((GET_CODE (op0) == REG || GET_CODE (op0) == SUBREG)
&& (bitsize + bitpos > GET_MODE_BITSIZE (extzv_mode))))
{
unsigned HOST_WIDE_INT xbitpos = bitpos, xoffset = offset;
rtx bitsize_rtx, bitpos_rtx;
rtx last = get_last_insn ();
rtx xop0 = op0;
rtx xtarget = target;
rtx xspec_target = spec_target;
rtx xspec_target_subreg = spec_target_subreg;
rtx pat;
enum machine_mode maxmode = mode_for_extraction (EP_extzv, 0);
if (GET_CODE (xop0) == MEM)
{
int save_volatile_ok = volatile_ok;
volatile_ok = 1;
/* Is the memory operand acceptable? */
if (! ((*insn_data[(int) CODE_FOR_extzv].operand[1].predicate)
(xop0, GET_MODE (xop0))))
{
/* No, load into a reg and extract from there. */
enum machine_mode bestmode;
/* Get the mode to use for inserting into this field. If
OP0 is BLKmode, get the smallest mode consistent with the
alignment. If OP0 is a non-BLKmode object that is no
wider than MAXMODE, use its mode. Otherwise, use the
smallest mode containing the field. */
if (GET_MODE (xop0) == BLKmode
|| (GET_MODE_SIZE (GET_MODE (op0))
> GET_MODE_SIZE (maxmode)))
bestmode = get_best_mode (bitsize, bitnum,
MEM_ALIGN (xop0), maxmode,
MEM_VOLATILE_P (xop0));
else
bestmode = GET_MODE (xop0);
if (bestmode == VOIDmode
|| (SLOW_UNALIGNED_ACCESS (bestmode, MEM_ALIGN (xop0))
&& GET_MODE_BITSIZE (bestmode) > MEM_ALIGN (xop0)))
goto extzv_loses;
/* Compute offset as multiple of this unit,
counting in bytes. */
unit = GET_MODE_BITSIZE (bestmode);
xoffset = (bitnum / unit) * GET_MODE_SIZE (bestmode);
xbitpos = bitnum % unit;
xop0 = adjust_address (xop0, bestmode, xoffset);
/* Fetch it to a register in that size. */
xop0 = force_reg (bestmode, xop0);
/* XBITPOS counts within UNIT, which is what is expected. */
}
else
/* Get ref to first byte containing part of the field. */
xop0 = adjust_address (xop0, byte_mode, xoffset);
volatile_ok = save_volatile_ok;
}
/* If op0 is a register, we need it in MAXMODE (which is usually
SImode). to make it acceptable to the format of extzv. */
if (GET_CODE (xop0) == SUBREG && GET_MODE (xop0) != maxmode)
goto extzv_loses;
if (GET_CODE (xop0) == REG && GET_MODE (xop0) != maxmode)
xop0 = gen_rtx_SUBREG (maxmode, xop0, 0);
/* On big-endian machines, we count bits from the most significant.
If the bit field insn does not, we must invert. */
if (BITS_BIG_ENDIAN != BYTES_BIG_ENDIAN)
xbitpos = unit - bitsize - xbitpos;
/* Now convert from counting within UNIT to counting in MAXMODE. */
if (BITS_BIG_ENDIAN && GET_CODE (xop0) != MEM)
xbitpos += GET_MODE_BITSIZE (maxmode) - unit;
unit = GET_MODE_BITSIZE (maxmode);
if (xtarget == 0
|| (flag_force_mem && GET_CODE (xtarget) == MEM))
xtarget = xspec_target = gen_reg_rtx (tmode);
if (GET_MODE (xtarget) != maxmode)
{
if (GET_CODE (xtarget) == REG)
{
int wider = (GET_MODE_SIZE (maxmode)
> GET_MODE_SIZE (GET_MODE (xtarget)));
xtarget = gen_lowpart (maxmode, xtarget);
if (wider)
xspec_target_subreg = xtarget;
}
else
xtarget = gen_reg_rtx (maxmode);
}
/* If this machine's extzv insists on a register target,
make sure we have one. */
if (! ((*insn_data[(int) CODE_FOR_extzv].operand[0].predicate)
(xtarget, maxmode)))
xtarget = gen_reg_rtx (maxmode);
bitsize_rtx = GEN_INT (bitsize);
bitpos_rtx = GEN_INT (xbitpos);
pat = gen_extzv (protect_from_queue (xtarget, 1),
xop0, bitsize_rtx, bitpos_rtx);
if (pat)
{
emit_insn (pat);
target = xtarget;
spec_target = xspec_target;
spec_target_subreg = xspec_target_subreg;
}
else
{
delete_insns_since (last);
target = extract_fixed_bit_field (int_mode, op0, offset, bitsize,
bitpos, target, 1);
}
}
else
extzv_loses:
target = extract_fixed_bit_field (int_mode, op0, offset, bitsize,
bitpos, target, 1);
}
else
{
if (HAVE_extv
&& (GET_MODE_BITSIZE (extv_mode) >= bitsize)
&& ! ((GET_CODE (op0) == REG || GET_CODE (op0) == SUBREG)
&& (bitsize + bitpos > GET_MODE_BITSIZE (extv_mode))))
{
int xbitpos = bitpos, xoffset = offset;
rtx bitsize_rtx, bitpos_rtx;
rtx last = get_last_insn ();
rtx xop0 = op0, xtarget = target;
rtx xspec_target = spec_target;
rtx xspec_target_subreg = spec_target_subreg;
rtx pat;
enum machine_mode maxmode = mode_for_extraction (EP_extv, 0);
if (GET_CODE (xop0) == MEM)
{
/* Is the memory operand acceptable? */
if (! ((*insn_data[(int) CODE_FOR_extv].operand[1].predicate)
(xop0, GET_MODE (xop0))))
{
/* No, load into a reg and extract from there. */
enum machine_mode bestmode;
/* Get the mode to use for inserting into this field. If
OP0 is BLKmode, get the smallest mode consistent with the
alignment. If OP0 is a non-BLKmode object that is no
wider than MAXMODE, use its mode. Otherwise, use the
smallest mode containing the field. */
if (GET_MODE (xop0) == BLKmode
|| (GET_MODE_SIZE (GET_MODE (op0))
> GET_MODE_SIZE (maxmode)))
bestmode = get_best_mode (bitsize, bitnum,
MEM_ALIGN (xop0), maxmode,
MEM_VOLATILE_P (xop0));
else
bestmode = GET_MODE (xop0);
if (bestmode == VOIDmode
|| (SLOW_UNALIGNED_ACCESS (bestmode, MEM_ALIGN (xop0))
&& GET_MODE_BITSIZE (bestmode) > MEM_ALIGN (xop0)))
goto extv_loses;
/* Compute offset as multiple of this unit,
counting in bytes. */
unit = GET_MODE_BITSIZE (bestmode);
xoffset = (bitnum / unit) * GET_MODE_SIZE (bestmode);
xbitpos = bitnum % unit;
xop0 = adjust_address (xop0, bestmode, xoffset);
/* Fetch it to a register in that size. */
xop0 = force_reg (bestmode, xop0);
/* XBITPOS counts within UNIT, which is what is expected. */
}
else
/* Get ref to first byte containing part of the field. */
xop0 = adjust_address (xop0, byte_mode, xoffset);
}
/* If op0 is a register, we need it in MAXMODE (which is usually
SImode) to make it acceptable to the format of extv. */
if (GET_CODE (xop0) == SUBREG && GET_MODE (xop0) != maxmode)
goto extv_loses;
if (GET_CODE (xop0) == REG && GET_MODE (xop0) != maxmode)
xop0 = gen_rtx_SUBREG (maxmode, xop0, 0);
/* On big-endian machines, we count bits from the most significant.
If the bit field insn does not, we must invert. */
if (BITS_BIG_ENDIAN != BYTES_BIG_ENDIAN)
xbitpos = unit - bitsize - xbitpos;
/* XBITPOS counts within a size of UNIT.
Adjust to count within a size of MAXMODE. */
if (BITS_BIG_ENDIAN && GET_CODE (xop0) != MEM)
xbitpos += (GET_MODE_BITSIZE (maxmode) - unit);
unit = GET_MODE_BITSIZE (maxmode);
if (xtarget == 0
|| (flag_force_mem && GET_CODE (xtarget) == MEM))
xtarget = xspec_target = gen_reg_rtx (tmode);
if (GET_MODE (xtarget) != maxmode)
{
if (GET_CODE (xtarget) == REG)
{
int wider = (GET_MODE_SIZE (maxmode)
> GET_MODE_SIZE (GET_MODE (xtarget)));
xtarget = gen_lowpart (maxmode, xtarget);
if (wider)
xspec_target_subreg = xtarget;
}
else
xtarget = gen_reg_rtx (maxmode);
}
/* If this machine's extv insists on a register target,
make sure we have one. */
if (! ((*insn_data[(int) CODE_FOR_extv].operand[0].predicate)
(xtarget, maxmode)))
xtarget = gen_reg_rtx (maxmode);
bitsize_rtx = GEN_INT (bitsize);
bitpos_rtx = GEN_INT (xbitpos);
pat = gen_extv (protect_from_queue (xtarget, 1),
xop0, bitsize_rtx, bitpos_rtx);
if (pat)
{
emit_insn (pat);
target = xtarget;
spec_target = xspec_target;
spec_target_subreg = xspec_target_subreg;
}
else
{
delete_insns_since (last);
target = extract_fixed_bit_field (int_mode, op0, offset, bitsize,
bitpos, target, 0);
}
}
else
extv_loses:
target = extract_fixed_bit_field (int_mode, op0, offset, bitsize,
bitpos, target, 0);
}
if (target == spec_target)
return target;
if (target == spec_target_subreg)
return spec_target;
if (GET_MODE (target) != tmode && GET_MODE (target) != mode)
{
/* If the target mode is floating-point, first convert to the
integer mode of that size and then access it as a floating-point
value via a SUBREG. */
if (GET_MODE_CLASS (tmode) != MODE_INT
&& GET_MODE_CLASS (tmode) != MODE_PARTIAL_INT)
{
target = convert_to_mode (mode_for_size (GET_MODE_BITSIZE (tmode),
MODE_INT, 0),
target, unsignedp);
return gen_lowpart (tmode, target);
}
else
return convert_to_mode (tmode, target, unsignedp);
}
return target;
}
/* Extract a bit field using shifts and boolean operations
Returns an rtx to represent the value.
OP0 addresses a register (word) or memory (byte).
BITPOS says which bit within the word or byte the bit field starts in.
OFFSET says how many bytes farther the bit field starts;
it is 0 if OP0 is a register.
BITSIZE says how many bits long the bit field is.
(If OP0 is a register, it may be narrower than a full word,
but BITPOS still counts within a full word,
which is significant on bigendian machines.)
UNSIGNEDP is nonzero for an unsigned bit field (don't sign-extend value).
If TARGET is nonzero, attempts to store the value there
and return TARGET, but this is not guaranteed.
If TARGET is not used, create a pseudo-reg of mode TMODE for the value. */
static rtx
extract_fixed_bit_field (tmode, op0, offset, bitsize, bitpos,
target, unsignedp)
enum machine_mode tmode;
rtx op0, target;
unsigned HOST_WIDE_INT offset, bitsize, bitpos;
int unsignedp;
{
unsigned int total_bits = BITS_PER_WORD;
enum machine_mode mode;
if (GET_CODE (op0) == SUBREG || GET_CODE (op0) == REG)
{
/* Special treatment for a bit field split across two registers. */
if (bitsize + bitpos > BITS_PER_WORD)
return extract_split_bit_field (op0, bitsize, bitpos, unsignedp);
}
else
{
/* Get the proper mode to use for this field. We want a mode that
includes the entire field. If such a mode would be larger than
a word, we won't be doing the extraction the normal way. */
mode = get_best_mode (bitsize, bitpos + offset * BITS_PER_UNIT,
MEM_ALIGN (op0), word_mode, MEM_VOLATILE_P (op0));
if (mode == VOIDmode)
/* The only way this should occur is if the field spans word
boundaries. */
return extract_split_bit_field (op0, bitsize,
bitpos + offset * BITS_PER_UNIT,
unsignedp);
total_bits = GET_MODE_BITSIZE (mode);
/* Make sure bitpos is valid for the chosen mode. Adjust BITPOS to
be in the range 0 to total_bits-1, and put any excess bytes in
OFFSET. */
if (bitpos >= total_bits)
{
offset += (bitpos / total_bits) * (total_bits / BITS_PER_UNIT);
bitpos -= ((bitpos / total_bits) * (total_bits / BITS_PER_UNIT)
* BITS_PER_UNIT);
}
/* Get ref to an aligned byte, halfword, or word containing the field.
Adjust BITPOS to be position within a word,
and OFFSET to be the offset of that word.
Then alter OP0 to refer to that word. */
bitpos += (offset % (total_bits / BITS_PER_UNIT)) * BITS_PER_UNIT;
offset -= (offset % (total_bits / BITS_PER_UNIT));
op0 = adjust_address (op0, mode, offset);
}
mode = GET_MODE (op0);
if (BYTES_BIG_ENDIAN)
/* BITPOS is the distance between our msb and that of OP0.
Convert it to the distance from the lsb. */
bitpos = total_bits - bitsize - bitpos;
/* Now BITPOS is always the distance between the field's lsb and that of OP0.
We have reduced the big-endian case to the little-endian case. */
if (unsignedp)
{
if (bitpos)
{
/* If the field does not already start at the lsb,
shift it so it does. */
tree amount = build_int_2 (bitpos, 0);
/* Maybe propagate the target for the shift. */
/* But not if we will return it--could confuse integrate.c. */
rtx subtarget = (target != 0 && GET_CODE (target) == REG
&& !REG_FUNCTION_VALUE_P (target)
? target : 0);
if (tmode != mode) subtarget = 0;
op0 = expand_shift (RSHIFT_EXPR, mode, op0, amount, subtarget, 1);
}
/* Convert the value to the desired mode. */
if (mode != tmode)
op0 = convert_to_mode (tmode, op0, 1);
/* Unless the msb of the field used to be the msb when we shifted,
mask out the upper bits. */
if (GET_MODE_BITSIZE (mode) != bitpos + bitsize)
return expand_binop (GET_MODE (op0), and_optab, op0,
mask_rtx (GET_MODE (op0), 0, bitsize, 0),
target, 1, OPTAB_LIB_WIDEN);
return op0;
}
/* To extract a signed bit-field, first shift its msb to the msb of the word,
then arithmetic-shift its lsb to the lsb of the word. */
op0 = force_reg (mode, op0);
if (mode != tmode)
target = 0;
/* Find the narrowest integer mode that contains the field. */
for (mode = GET_CLASS_NARROWEST_MODE (MODE_INT); mode != VOIDmode;
mode = GET_MODE_WIDER_MODE (mode))
if (GET_MODE_BITSIZE (mode) >= bitsize + bitpos)
{
op0 = convert_to_mode (mode, op0, 0);
break;
}
if (GET_MODE_BITSIZE (mode) != (bitsize + bitpos))
{
tree amount
= build_int_2 (GET_MODE_BITSIZE (mode) - (bitsize + bitpos), 0);
/* Maybe propagate the target for the shift. */
/* But not if we will return the result--could confuse integrate.c. */
rtx subtarget = (target != 0 && GET_CODE (target) == REG
&& ! REG_FUNCTION_VALUE_P (target)
? target : 0);
op0 = expand_shift (LSHIFT_EXPR, mode, op0, amount, subtarget, 1);
}
return expand_shift (RSHIFT_EXPR, mode, op0,
build_int_2 (GET_MODE_BITSIZE (mode) - bitsize, 0),
target, 0);
}
/* Return a constant integer (CONST_INT or CONST_DOUBLE) mask value
of mode MODE with BITSIZE ones followed by BITPOS zeros, or the
complement of that if COMPLEMENT. The mask is truncated if
necessary to the width of mode MODE. The mask is zero-extended if
BITSIZE+BITPOS is too small for MODE. */
static rtx
mask_rtx (mode, bitpos, bitsize, complement)
enum machine_mode mode;
int bitpos, bitsize, complement;
{
HOST_WIDE_INT masklow, maskhigh;
if (bitpos < HOST_BITS_PER_WIDE_INT)
masklow = (HOST_WIDE_INT) -1 << bitpos;
else
masklow = 0;
if (bitpos + bitsize < HOST_BITS_PER_WIDE_INT)
masklow &= ((unsigned HOST_WIDE_INT) -1
>> (HOST_BITS_PER_WIDE_INT - bitpos - bitsize));
if (bitpos <= HOST_BITS_PER_WIDE_INT)
maskhigh = -1;
else
maskhigh = (HOST_WIDE_INT) -1 << (bitpos - HOST_BITS_PER_WIDE_INT);
if (bitpos + bitsize > HOST_BITS_PER_WIDE_INT)
maskhigh &= ((unsigned HOST_WIDE_INT) -1
>> (2 * HOST_BITS_PER_WIDE_INT - bitpos - bitsize));
else
maskhigh = 0;
if (complement)
{
maskhigh = ~maskhigh;
masklow = ~masklow;
}
return immed_double_const (masklow, maskhigh, mode);
}
/* Return a constant integer (CONST_INT or CONST_DOUBLE) rtx with the value
VALUE truncated to BITSIZE bits and then shifted left BITPOS bits. */
static rtx
lshift_value (mode, value, bitpos, bitsize)
enum machine_mode mode;
rtx value;
int bitpos, bitsize;
{
unsigned HOST_WIDE_INT v = INTVAL (value);
HOST_WIDE_INT low, high;
if (bitsize < HOST_BITS_PER_WIDE_INT)
v &= ~((HOST_WIDE_INT) -1 << bitsize);
if (bitpos < HOST_BITS_PER_WIDE_INT)
{
low = v << bitpos;
high = (bitpos > 0 ? (v >> (HOST_BITS_PER_WIDE_INT - bitpos)) : 0);
}
else
{
low = 0;
high = v << (bitpos - HOST_BITS_PER_WIDE_INT);
}
return immed_double_const (low, high, mode);
}
/* Extract a bit field that is split across two words
and return an RTX for the result.
OP0 is the REG, SUBREG or MEM rtx for the first of the two words.
BITSIZE is the field width; BITPOS, position of its first bit, in the word.
UNSIGNEDP is 1 if should zero-extend the contents; else sign-extend. */
static rtx
extract_split_bit_field (op0, bitsize, bitpos, unsignedp)
rtx op0;
unsigned HOST_WIDE_INT bitsize, bitpos;
int unsignedp;
{
unsigned int unit;
unsigned int bitsdone = 0;
rtx result = NULL_RTX;
int first = 1;
/* Make sure UNIT isn't larger than BITS_PER_WORD, we can only handle that
much at a time. */
if (GET_CODE (op0) == REG || GET_CODE (op0) == SUBREG)
unit = BITS_PER_WORD;
else
unit = MIN (MEM_ALIGN (op0), BITS_PER_WORD);
while (bitsdone < bitsize)
{
unsigned HOST_WIDE_INT thissize;
rtx part, word;
unsigned HOST_WIDE_INT thispos;
unsigned HOST_WIDE_INT offset;
offset = (bitpos + bitsdone) / unit;
thispos = (bitpos + bitsdone) % unit;
/* THISSIZE must not overrun a word boundary. Otherwise,
extract_fixed_bit_field will call us again, and we will mutually
recurse forever. */
thissize = MIN (bitsize - bitsdone, BITS_PER_WORD);
thissize = MIN (thissize, unit - thispos);
/* If OP0 is a register, then handle OFFSET here.
When handling multiword bitfields, extract_bit_field may pass
down a word_mode SUBREG of a larger REG for a bitfield that actually
crosses a word boundary. Thus, for a SUBREG, we must find
the current word starting from the base register. */
if (GET_CODE (op0) == SUBREG)
{
int word_offset = (SUBREG_BYTE (op0) / UNITS_PER_WORD) + offset;
word = operand_subword_force (SUBREG_REG (op0), word_offset,
GET_MODE (SUBREG_REG (op0)));
offset = 0;
}
else if (GET_CODE (op0) == REG)
{
word = operand_subword_force (op0, offset, GET_MODE (op0));
offset = 0;
}
else
word = op0;
/* Extract the parts in bit-counting order,
whose meaning is determined by BYTES_PER_UNIT.
OFFSET is in UNITs, and UNIT is in bits.
extract_fixed_bit_field wants offset in bytes. */
part = extract_fixed_bit_field (word_mode, word,
offset * unit / BITS_PER_UNIT,
thissize, thispos, 0, 1);
bitsdone += thissize;
/* Shift this part into place for the result. */
if (BYTES_BIG_ENDIAN)
{
if (bitsize != bitsdone)
part = expand_shift (LSHIFT_EXPR, word_mode, part,
build_int_2 (bitsize - bitsdone, 0), 0, 1);
}
else
{
if (bitsdone != thissize)
part = expand_shift (LSHIFT_EXPR, word_mode, part,
build_int_2 (bitsdone - thissize, 0), 0, 1);
}
if (first)
result = part;
else
/* Combine the parts with bitwise or. This works
because we extracted each part as an unsigned bit field. */
result = expand_binop (word_mode, ior_optab, part, result, NULL_RTX, 1,
OPTAB_LIB_WIDEN);
first = 0;
}
/* Unsigned bit field: we are done. */
if (unsignedp)
return result;
/* Signed bit field: sign-extend with two arithmetic shifts. */
result = expand_shift (LSHIFT_EXPR, word_mode, result,
build_int_2 (BITS_PER_WORD - bitsize, 0),
NULL_RTX, 0);
return expand_shift (RSHIFT_EXPR, word_mode, result,
build_int_2 (BITS_PER_WORD - bitsize, 0), NULL_RTX, 0);
}
/* Add INC into TARGET. */
void
expand_inc (target, inc)
rtx target, inc;
{
rtx value = expand_binop (GET_MODE (target), add_optab,
target, inc,
target, 0, OPTAB_LIB_WIDEN);
if (value != target)
emit_move_insn (target, value);
}
/* Subtract DEC from TARGET. */
void
expand_dec (target, dec)
rtx target, dec;
{
rtx value = expand_binop (GET_MODE (target), sub_optab,
target, dec,
target, 0, OPTAB_LIB_WIDEN);
if (value != target)
emit_move_insn (target, value);
}
/* Output a shift instruction for expression code CODE,
with SHIFTED being the rtx for the value to shift,
and AMOUNT the tree for the amount to shift by.
Store the result in the rtx TARGET, if that is convenient.
If UNSIGNEDP is nonzero, do a logical shift; otherwise, arithmetic.
Return the rtx for where the value is. */
rtx
expand_shift (code, mode, shifted, amount, target, unsignedp)
enum tree_code code;
enum machine_mode mode;
rtx shifted;
tree amount;
rtx target;
int unsignedp;
{
rtx op1, temp = 0;
int left = (code == LSHIFT_EXPR || code == LROTATE_EXPR);
int rotate = (code == LROTATE_EXPR || code == RROTATE_EXPR);
int try;
/* Previously detected shift-counts computed by NEGATE_EXPR
and shifted in the other direction; but that does not work
on all machines. */
op1 = expand_expr (amount, NULL_RTX, VOIDmode, 0);
#ifdef SHIFT_COUNT_TRUNCATED
if (SHIFT_COUNT_TRUNCATED)
{
if (GET_CODE (op1) == CONST_INT
&& ((unsigned HOST_WIDE_INT) INTVAL (op1) >=
(unsigned HOST_WIDE_INT) GET_MODE_BITSIZE (mode)))
op1 = GEN_INT ((unsigned HOST_WIDE_INT) INTVAL (op1)
% GET_MODE_BITSIZE (mode));
else if (GET_CODE (op1) == SUBREG
&& subreg_lowpart_p (op1))
op1 = SUBREG_REG (op1);
}
#endif
if (op1 == const0_rtx)
return shifted;
for (try = 0; temp == 0 && try < 3; try++)
{
enum optab_methods methods;
if (try == 0)
methods = OPTAB_DIRECT;
else if (try == 1)
methods = OPTAB_WIDEN;
else
methods = OPTAB_LIB_WIDEN;
if (rotate)
{
/* Widening does not work for rotation. */
if (methods == OPTAB_WIDEN)
continue;
else if (methods == OPTAB_LIB_WIDEN)
{
/* If we have been unable to open-code this by a rotation,
do it as the IOR of two shifts. I.e., to rotate A
by N bits, compute (A << N) | ((unsigned) A >> (C - N))
where C is the bitsize of A.
It is theoretically possible that the target machine might
not be able to perform either shift and hence we would
be making two libcalls rather than just the one for the
shift (similarly if IOR could not be done). We will allow
this extremely unlikely lossage to avoid complicating the
code below. */
rtx subtarget = target == shifted ? 0 : target;
rtx temp1;
tree type = TREE_TYPE (amount);
tree new_amount = make_tree (type, op1);
tree other_amount
= fold (build (MINUS_EXPR, type,
convert (type,
build_int_2 (GET_MODE_BITSIZE (mode),
0)),
amount));
shifted = force_reg (mode, shifted);
temp = expand_shift (left ? LSHIFT_EXPR : RSHIFT_EXPR,
mode, shifted, new_amount, subtarget, 1);
temp1 = expand_shift (left ? RSHIFT_EXPR : LSHIFT_EXPR,
mode, shifted, other_amount, 0, 1);
return expand_binop (mode, ior_optab, temp, temp1, target,
unsignedp, methods);
}
temp = expand_binop (mode,
left ? rotl_optab : rotr_optab,
shifted, op1, target, unsignedp, methods);
/* If we don't have the rotate, but we are rotating by a constant
that is in range, try a rotate in the opposite direction. */
if (temp == 0 && GET_CODE (op1) == CONST_INT
&& INTVAL (op1) > 0
&& (unsigned int) INTVAL (op1) < GET_MODE_BITSIZE (mode))
temp = expand_binop (mode,
left ? rotr_optab : rotl_optab,
shifted,
GEN_INT (GET_MODE_BITSIZE (mode)
- INTVAL (op1)),
target, unsignedp, methods);
}
else if (unsignedp)
temp = expand_binop (mode,
left ? ashl_optab : lshr_optab,
shifted, op1, target, unsignedp, methods);
/* Do arithmetic shifts.
Also, if we are going to widen the operand, we can just as well
use an arithmetic right-shift instead of a logical one. */
if (temp == 0 && ! rotate
&& (! unsignedp || (! left && methods == OPTAB_WIDEN)))
{
enum optab_methods methods1 = methods;
/* If trying to widen a log shift to an arithmetic shift,
don't accept an arithmetic shift of the same size. */
if (unsignedp)
methods1 = OPTAB_MUST_WIDEN;
/* Arithmetic shift */
temp = expand_binop (mode,
left ? ashl_optab : ashr_optab,
shifted, op1, target, unsignedp, methods1);
}
/* We used to try extzv here for logical right shifts, but that was
only useful for one machine, the VAX, and caused poor code
generation there for lshrdi3, so the code was deleted and a
define_expand for lshrsi3 was added to vax.md. */
}
if (temp == 0)
abort ();
return temp;
}
enum alg_code { alg_zero, alg_m, alg_shift,
alg_add_t_m2, alg_sub_t_m2,
alg_add_factor, alg_sub_factor,
alg_add_t2_m, alg_sub_t2_m,
alg_add, alg_subtract, alg_factor, alg_shiftop };
/* This structure records a sequence of operations.
`ops' is the number of operations recorded.
`cost' is their total cost.
The operations are stored in `op' and the corresponding
logarithms of the integer coefficients in `log'.
These are the operations:
alg_zero total := 0;
alg_m total := multiplicand;
alg_shift total := total * coeff
alg_add_t_m2 total := total + multiplicand * coeff;
alg_sub_t_m2 total := total - multiplicand * coeff;
alg_add_factor total := total * coeff + total;
alg_sub_factor total := total * coeff - total;
alg_add_t2_m total := total * coeff + multiplicand;
alg_sub_t2_m total := total * coeff - multiplicand;
The first operand must be either alg_zero or alg_m. */
struct algorithm
{
short cost;
short ops;
/* The size of the OP and LOG fields are not directly related to the
word size, but the worst-case algorithms will be if we have few
consecutive ones or zeros, i.e., a multiplicand like 10101010101...
In that case we will generate shift-by-2, add, shift-by-2, add,...,
in total wordsize operations. */
enum alg_code op[MAX_BITS_PER_WORD];
char log[MAX_BITS_PER_WORD];
};
static void synth_mult PARAMS ((struct algorithm *,
unsigned HOST_WIDE_INT,
int));
static unsigned HOST_WIDE_INT choose_multiplier PARAMS ((unsigned HOST_WIDE_INT,
int, int,
unsigned HOST_WIDE_INT *,
int *, int *));
static unsigned HOST_WIDE_INT invert_mod2n PARAMS ((unsigned HOST_WIDE_INT,
int));
/* Compute and return the best algorithm for multiplying by T.
The algorithm must cost less than cost_limit
If retval.cost >= COST_LIMIT, no algorithm was found and all
other field of the returned struct are undefined. */
static void
synth_mult (alg_out, t, cost_limit)
struct algorithm *alg_out;
unsigned HOST_WIDE_INT t;
int cost_limit;
{
int m;
struct algorithm *alg_in, *best_alg;
int cost;
unsigned HOST_WIDE_INT q;
/* Indicate that no algorithm is yet found. If no algorithm
is found, this value will be returned and indicate failure. */
alg_out->cost = cost_limit;
if (cost_limit <= 0)
return;
/* t == 1 can be done in zero cost. */
if (t == 1)
{
alg_out->ops = 1;
alg_out->cost = 0;
alg_out->op[0] = alg_m;
return;
}
/* t == 0 sometimes has a cost. If it does and it exceeds our limit,
fail now. */
if (t == 0)
{
if (zero_cost >= cost_limit)
return;
else
{
alg_out->ops = 1;
alg_out->cost = zero_cost;
alg_out->op[0] = alg_zero;
return;
}
}
/* We'll be needing a couple extra algorithm structures now. */
alg_in = (struct algorithm *)alloca (sizeof (struct algorithm));
best_alg = (struct algorithm *)alloca (sizeof (struct algorithm));
/* If we have a group of zero bits at the low-order part of T, try
multiplying by the remaining bits and then doing a shift. */
if ((t & 1) == 0)
{
m = floor_log2 (t & -t); /* m = number of low zero bits */
if (m < BITS_PER_WORD)
{
q = t >> m;
cost = shift_cost[m];
synth_mult (alg_in, q, cost_limit - cost);
cost += alg_in->cost;
if (cost < cost_limit)
{
struct algorithm *x;
x = alg_in, alg_in = best_alg, best_alg = x;
best_alg->log[best_alg->ops] = m;
best_alg->op[best_alg->ops] = alg_shift;
cost_limit = cost;
}
}
}
/* If we have an odd number, add or subtract one. */
if ((t & 1) != 0)
{
unsigned HOST_WIDE_INT w;
for (w = 1; (w & t) != 0; w <<= 1)
;
/* If T was -1, then W will be zero after the loop. This is another
case where T ends with ...111. Handling this with (T + 1) and
subtract 1 produces slightly better code and results in algorithm
selection much faster than treating it like the ...0111 case
below. */
if (w == 0
|| (w > 2
/* Reject the case where t is 3.
Thus we prefer addition in that case. */
&& t != 3))
{
/* T ends with ...111. Multiply by (T + 1) and subtract 1. */
cost = add_cost;
synth_mult (alg_in, t + 1, cost_limit - cost);
cost += alg_in->cost;
if (cost < cost_limit)
{
struct algorithm *x;
x = alg_in, alg_in = best_alg, best_alg = x;
best_alg->log[best_alg->ops] = 0;
best_alg->op[best_alg->ops] = alg_sub_t_m2;
cost_limit = cost;
}
}
else
{
/* T ends with ...01 or ...011. Multiply by (T - 1) and add 1. */
cost = add_cost;
synth_mult (alg_in, t - 1, cost_limit - cost);
cost += alg_in->cost;
if (cost < cost_limit)
{
struct algorithm *x;
x = alg_in, alg_in = best_alg, best_alg = x;
best_alg->log[best_alg->ops] = 0;
best_alg->op[best_alg->ops] = alg_add_t_m2;
cost_limit = cost;
}
}
}
/* Look for factors of t of the form
t = q(2**m +- 1), 2 <= m <= floor(log2(t - 1)).
If we find such a factor, we can multiply by t using an algorithm that
multiplies by q, shift the result by m and add/subtract it to itself.
We search for large factors first and loop down, even if large factors
are less probable than small; if we find a large factor we will find a
good sequence quickly, and therefore be able to prune (by decreasing
COST_LIMIT) the search. */
for (m = floor_log2 (t - 1); m >= 2; m--)
{
unsigned HOST_WIDE_INT d;
d = ((unsigned HOST_WIDE_INT) 1 << m) + 1;
if (t % d == 0 && t > d && m < BITS_PER_WORD)
{
cost = MIN (shiftadd_cost[m], add_cost + shift_cost[m]);
synth_mult (alg_in, t / d, cost_limit - cost);
cost += alg_in->cost;
if (cost < cost_limit)
{
struct algorithm *x;
x = alg_in, alg_in = best_alg, best_alg = x;
best_alg->log[best_alg->ops] = m;
best_alg->op[best_alg->ops] = alg_add_factor;
cost_limit = cost;
}
/* Other factors will have been taken care of in the recursion. */
break;
}
d = ((unsigned HOST_WIDE_INT) 1 << m) - 1;
if (t % d == 0 && t > d && m < BITS_PER_WORD)
{
cost = MIN (shiftsub_cost[m], add_cost + shift_cost[m]);
synth_mult (alg_in, t / d, cost_limit - cost);
cost += alg_in->cost;
if (cost < cost_limit)
{
struct algorithm *x;
x = alg_in, alg_in = best_alg, best_alg = x;
best_alg->log[best_alg->ops] = m;
best_alg->op[best_alg->ops] = alg_sub_factor;
cost_limit = cost;
}
break;
}
}
/* Try shift-and-add (load effective address) instructions,
i.e. do a*3, a*5, a*9. */
if ((t & 1) != 0)
{
q = t - 1;
q = q & -q;
m = exact_log2 (q);
if (m >= 0 && m < BITS_PER_WORD)
{
cost = shiftadd_cost[m];
synth_mult (alg_in, (t - 1) >> m, cost_limit - cost);
cost += alg_in->cost;
if (cost < cost_limit)
{
struct algorithm *x;
x = alg_in, alg_in = best_alg, best_alg = x;
best_alg->log[best_alg->ops] = m;
best_alg->op[best_alg->ops] = alg_add_t2_m;
cost_limit = cost;
}
}
q = t + 1;
q = q & -q;
m = exact_log2 (q);
if (m >= 0 && m < BITS_PER_WORD)
{
cost = shiftsub_cost[m];
synth_mult (alg_in, (t + 1) >> m, cost_limit - cost);
cost += alg_in->cost;
if (cost < cost_limit)
{
struct algorithm *x;
x = alg_in, alg_in = best_alg, best_alg = x;
best_alg->log[best_alg->ops] = m;
best_alg->op[best_alg->ops] = alg_sub_t2_m;
cost_limit = cost;
}
}
}
/* If cost_limit has not decreased since we stored it in alg_out->cost,
we have not found any algorithm. */
if (cost_limit == alg_out->cost)
return;
/* If we are getting a too long sequence for `struct algorithm'
to record, make this search fail. */
if (best_alg->ops == MAX_BITS_PER_WORD)
return;
/* Copy the algorithm from temporary space to the space at alg_out.
We avoid using structure assignment because the majority of
best_alg is normally undefined, and this is a critical function. */
alg_out->ops = best_alg->ops + 1;
alg_out->cost = cost_limit;
memcpy (alg_out->op, best_alg->op,
alg_out->ops * sizeof *alg_out->op);
memcpy (alg_out->log, best_alg->log,
alg_out->ops * sizeof *alg_out->log);
}
/* Perform a multiplication and return an rtx for the result.
MODE is mode of value; OP0 and OP1 are what to multiply (rtx's);
TARGET is a suggestion for where to store the result (an rtx).
We check specially for a constant integer as OP1.
If you want this check for OP0 as well, then before calling
you should swap the two operands if OP0 would be constant. */
rtx
expand_mult (mode, op0, op1, target, unsignedp)
enum machine_mode mode;
rtx op0, op1, target;
int unsignedp;
{
rtx const_op1 = op1;
/* synth_mult does an `unsigned int' multiply. As long as the mode is
less than or equal in size to `unsigned int' this doesn't matter.
If the mode is larger than `unsigned int', then synth_mult works only
if the constant value exactly fits in an `unsigned int' without any
truncation. This means that multiplying by negative values does
not work; results are off by 2^32 on a 32 bit machine. */
/* If we are multiplying in DImode, it may still be a win
to try to work with shifts and adds. */
if (GET_CODE (op1) == CONST_DOUBLE
&& GET_MODE_CLASS (GET_MODE (op1)) == MODE_INT
&& HOST_BITS_PER_INT >= BITS_PER_WORD
&& CONST_DOUBLE_HIGH (op1) == 0)
const_op1 = GEN_INT (CONST_DOUBLE_LOW (op1));
else if (HOST_BITS_PER_INT < GET_MODE_BITSIZE (mode)
&& GET_CODE (op1) == CONST_INT
&& INTVAL (op1) < 0)
const_op1 = 0;
/* We used to test optimize here, on the grounds that it's better to
produce a smaller program when -O is not used.
But this causes such a terrible slowdown sometimes
that it seems better to use synth_mult always. */
if (const_op1 && GET_CODE (const_op1) == CONST_INT
&& (unsignedp || ! flag_trapv))
{
struct algorithm alg;
struct algorithm alg2;
HOST_WIDE_INT val = INTVAL (op1);
HOST_WIDE_INT val_so_far;
rtx insn;
int mult_cost;
enum {basic_variant, negate_variant, add_variant} variant = basic_variant;
/* op0 must be register to make mult_cost match the precomputed
shiftadd_cost array. */
op0 = force_reg (mode, op0);
/* Try to do the computation three ways: multiply by the negative of OP1
and then negate, do the multiplication directly, or do multiplication
by OP1 - 1. */
mult_cost = rtx_cost (gen_rtx_MULT (mode, op0, op1), SET);
mult_cost = MIN (12 * add_cost, mult_cost);
synth_mult (&alg, val, mult_cost);
/* This works only if the inverted value actually fits in an
`unsigned int' */
if (HOST_BITS_PER_INT >= GET_MODE_BITSIZE (mode))
{
synth_mult (&alg2, - val,
(alg.cost < mult_cost ? alg.cost : mult_cost) - negate_cost);
if (alg2.cost + negate_cost < alg.cost)
alg = alg2, variant = negate_variant;
}
/* This proves very useful for division-by-constant. */
synth_mult (&alg2, val - 1,
(alg.cost < mult_cost ? alg.cost : mult_cost) - add_cost);
if (alg2.cost + add_cost < alg.cost)
alg = alg2, variant = add_variant;
if (alg.cost < mult_cost)
{
/* We found something cheaper than a multiply insn. */
int opno;
rtx accum, tem;
enum machine_mode nmode;
op0 = protect_from_queue (op0, 0);
/* Avoid referencing memory over and over.
For speed, but also for correctness when mem is volatile. */
if (GET_CODE (op0) == MEM)
op0 = force_reg (mode, op0);
/* ACCUM starts out either as OP0 or as a zero, depending on
the first operation. */
if (alg.op[0] == alg_zero)
{
accum = copy_to_mode_reg (mode, const0_rtx);
val_so_far = 0;
}
else if (alg.op[0] == alg_m)
{
accum = copy_to_mode_reg (mode, op0);
val_so_far = 1;
}
else
abort ();
for (opno = 1; opno < alg.ops; opno++)
{
int log = alg.log[opno];
int preserve = preserve_subexpressions_p ();
rtx shift_subtarget = preserve ? 0 : accum;
rtx add_target
= (opno == alg.ops - 1 && target != 0 && variant != add_variant
&& ! preserve)
? target : 0;
rtx accum_target = preserve ? 0 : accum;
switch (alg.op[opno])
{
case alg_shift:
accum = expand_shift (LSHIFT_EXPR, mode, accum,
build_int_2 (log, 0), NULL_RTX, 0);
val_so_far <<= log;
break;
case alg_add_t_m2:
tem = expand_shift (LSHIFT_EXPR, mode, op0,
build_int_2 (log, 0), NULL_RTX, 0);
accum = force_operand (gen_rtx_PLUS (mode, accum, tem),
add_target
? add_target : accum_target);
val_so_far += (HOST_WIDE_INT) 1 << log;
break;
case alg_sub_t_m2:
tem = expand_shift (LSHIFT_EXPR, mode, op0,
build_int_2 (log, 0), NULL_RTX, 0);
accum = force_operand (gen_rtx_MINUS (mode, accum, tem),
add_target
? add_target : accum_target);
val_so_far -= (HOST_WIDE_INT) 1 << log;
break;
case alg_add_t2_m:
accum = expand_shift (LSHIFT_EXPR, mode, accum,
build_int_2 (log, 0), shift_subtarget,
0);
accum = force_operand (gen_rtx_PLUS (mode, accum, op0),
add_target
? add_target : accum_target);
val_so_far = (val_so_far << log) + 1;
break;
case alg_sub_t2_m:
accum = expand_shift (LSHIFT_EXPR, mode, accum,
build_int_2 (log, 0), shift_subtarget,
0);
accum = force_operand (gen_rtx_MINUS (mode, accum, op0),
add_target
? add_target : accum_target);
val_so_far = (val_so_far << log) - 1;
break;
case alg_add_factor:
tem = expand_shift (LSHIFT_EXPR, mode, accum,
build_int_2 (log, 0), NULL_RTX, 0);
accum = force_operand (gen_rtx_PLUS (mode, accum, tem),
add_target
? add_target : accum_target);
val_so_far += val_so_far << log;
break;
case alg_sub_factor:
tem = expand_shift (LSHIFT_EXPR, mode, accum,
build_int_2 (log, 0), NULL_RTX, 0);
accum = force_operand (gen_rtx_MINUS (mode, tem, accum),
(add_target ? add_target
: preserve ? 0 : tem));
val_so_far = (val_so_far << log) - val_so_far;
break;
default:
abort ();
}
/* Write a REG_EQUAL note on the last insn so that we can cse
multiplication sequences. Note that if ACCUM is a SUBREG,
we've set the inner register and must properly indicate
that. */
tem = op0, nmode = mode;
if (GET_CODE (accum) == SUBREG)
{
nmode = GET_MODE (SUBREG_REG (accum));
tem = gen_lowpart (nmode, op0);
}
insn = get_last_insn ();
set_unique_reg_note (insn,
REG_EQUAL,
gen_rtx_MULT (nmode, tem,
GEN_INT (val_so_far)));
}
if (variant == negate_variant)
{
val_so_far = - val_so_far;
accum = expand_unop (mode, neg_optab, accum, target, 0);
}
else if (variant == add_variant)
{
val_so_far = val_so_far + 1;
accum = force_operand (gen_rtx_PLUS (mode, accum, op0), target);
}
if (val != val_so_far)
abort ();
return accum;
}
}
/* This used to use umul_optab if unsigned, but for non-widening multiply
there is no difference between signed and unsigned. */
op0 = expand_binop (mode,
! unsignedp
&& flag_trapv && (GET_MODE_CLASS(mode) == MODE_INT)
? smulv_optab : smul_optab,
op0, op1, target, unsignedp, OPTAB_LIB_WIDEN);
if (op0 == 0)
abort ();
return op0;
}
/* Return the smallest n such that 2**n >= X. */
int
ceil_log2 (x)
unsigned HOST_WIDE_INT x;
{
return floor_log2 (x - 1) + 1;
}
/* Choose a minimal N + 1 bit approximation to 1/D that can be used to
replace division by D, and put the least significant N bits of the result
in *MULTIPLIER_PTR and return the most significant bit.
The width of operations is N (should be <= HOST_BITS_PER_WIDE_INT), the
needed precision is in PRECISION (should be <= N).
PRECISION should be as small as possible so this function can choose
multiplier more freely.
The rounded-up logarithm of D is placed in *lgup_ptr. A shift count that
is to be used for a final right shift is placed in *POST_SHIFT_PTR.
Using this function, x/D will be equal to (x * m) >> (*POST_SHIFT_PTR),
where m is the full HOST_BITS_PER_WIDE_INT + 1 bit multiplier. */
static
unsigned HOST_WIDE_INT
choose_multiplier (d, n, precision, multiplier_ptr, post_shift_ptr, lgup_ptr)
unsigned HOST_WIDE_INT d;
int n;
int precision;
unsigned HOST_WIDE_INT *multiplier_ptr;
int *post_shift_ptr;
int *lgup_ptr;
{
HOST_WIDE_INT mhigh_hi, mlow_hi;
unsigned HOST_WIDE_INT mhigh_lo, mlow_lo;
int lgup, post_shift;
int pow, pow2;
unsigned HOST_WIDE_INT nl, dummy1;
HOST_WIDE_INT nh, dummy2;
/* lgup = ceil(log2(divisor)); */
lgup = ceil_log2 (d);
if (lgup > n)
abort ();
pow = n + lgup;
pow2 = n + lgup - precision;
if (pow == 2 * HOST_BITS_PER_WIDE_INT)
{
/* We could handle this with some effort, but this case is much better
handled directly with a scc insn, so rely on caller using that. */
abort ();
}
/* mlow = 2^(N + lgup)/d */
if (pow >= HOST_BITS_PER_WIDE_INT)
{
nh = (HOST_WIDE_INT) 1 << (pow - HOST_BITS_PER_WIDE_INT);
nl = 0;
}
else
{
nh = 0;
nl = (unsigned HOST_WIDE_INT) 1 << pow;
}
div_and_round_double (TRUNC_DIV_EXPR, 1, nl, nh, d, (HOST_WIDE_INT) 0,
&mlow_lo, &mlow_hi, &dummy1, &dummy2);
/* mhigh = (2^(N + lgup) + 2^N + lgup - precision)/d */
if (pow2 >= HOST_BITS_PER_WIDE_INT)
nh |= (HOST_WIDE_INT) 1 << (pow2 - HOST_BITS_PER_WIDE_INT);
else
nl |= (unsigned HOST_WIDE_INT) 1 << pow2;
div_and_round_double (TRUNC_DIV_EXPR, 1, nl, nh, d, (HOST_WIDE_INT) 0,
&mhigh_lo, &mhigh_hi, &dummy1, &dummy2);
if (mhigh_hi && nh - d >= d)
abort ();
if (mhigh_hi > 1 || mlow_hi > 1)
abort ();
/* assert that mlow < mhigh. */
if (! (mlow_hi < mhigh_hi || (mlow_hi == mhigh_hi && mlow_lo < mhigh_lo)))
abort ();
/* If precision == N, then mlow, mhigh exceed 2^N
(but they do not exceed 2^(N+1)). */
/* Reduce to lowest terms */
for (post_shift = lgup; post_shift > 0; post_shift--)
{
unsigned HOST_WIDE_INT ml_lo = (mlow_hi << (HOST_BITS_PER_WIDE_INT - 1)) | (mlow_lo >> 1);
unsigned HOST_WIDE_INT mh_lo = (mhigh_hi << (HOST_BITS_PER_WIDE_INT - 1)) | (mhigh_lo >> 1);
if (ml_lo >= mh_lo)
break;
mlow_hi = 0;
mlow_lo = ml_lo;
mhigh_hi = 0;
mhigh_lo = mh_lo;
}
*post_shift_ptr = post_shift;
*lgup_ptr = lgup;
if (n < HOST_BITS_PER_WIDE_INT)
{
unsigned HOST_WIDE_INT mask = ((unsigned HOST_WIDE_INT) 1 << n) - 1;
*multiplier_ptr = mhigh_lo & mask;
return mhigh_lo >= mask;
}
else
{
*multiplier_ptr = mhigh_lo;
return mhigh_hi;
}
}
/* Compute the inverse of X mod 2**n, i.e., find Y such that X * Y is
congruent to 1 (mod 2**N). */
static unsigned HOST_WIDE_INT
invert_mod2n (x, n)
unsigned HOST_WIDE_INT x;
int n;
{
/* Solve x*y == 1 (mod 2^n), where x is odd. Return y. */
/* The algorithm notes that the choice y = x satisfies
x*y == 1 mod 2^3, since x is assumed odd.
Each iteration doubles the number of bits of significance in y. */
unsigned HOST_WIDE_INT mask;
unsigned HOST_WIDE_INT y = x;
int nbit = 3;
mask = (n == HOST_BITS_PER_WIDE_INT
? ~(unsigned HOST_WIDE_INT) 0
: ((unsigned HOST_WIDE_INT) 1 << n) - 1);
while (nbit < n)
{
y = y * (2 - x*y) & mask; /* Modulo 2^N */
nbit *= 2;
}
return y;
}
/* Emit code to adjust ADJ_OPERAND after multiplication of wrong signedness
flavor of OP0 and OP1. ADJ_OPERAND is already the high half of the
product OP0 x OP1. If UNSIGNEDP is nonzero, adjust the signed product
to become unsigned, if UNSIGNEDP is zero, adjust the unsigned product to
become signed.
The result is put in TARGET if that is convenient.
MODE is the mode of operation. */
rtx
expand_mult_highpart_adjust (mode, adj_operand, op0, op1, target, unsignedp)
enum machine_mode mode;
rtx adj_operand, op0, op1, target;
int unsignedp;
{
rtx tem;
enum rtx_code adj_code = unsignedp ? PLUS : MINUS;
tem = expand_shift (RSHIFT_EXPR, mode, op0,
build_int_2 (GET_MODE_BITSIZE (mode) - 1, 0),
NULL_RTX, 0);
tem = expand_and (mode, tem, op1, NULL_RTX);
adj_operand
= force_operand (gen_rtx_fmt_ee (adj_code, mode, adj_operand, tem),
adj_operand);
tem = expand_shift (RSHIFT_EXPR, mode, op1,
build_int_2 (GET_MODE_BITSIZE (mode) - 1, 0),
NULL_RTX, 0);
tem = expand_and (mode, tem, op0, NULL_RTX);
target = force_operand (gen_rtx_fmt_ee (adj_code, mode, adj_operand, tem),
target);
return target;
}
/* Emit code to multiply OP0 and CNST1, putting the high half of the result
in TARGET if that is convenient, and return where the result is. If the
operation can not be performed, 0 is returned.
MODE is the mode of operation and result.
UNSIGNEDP nonzero means unsigned multiply.
MAX_COST is the total allowed cost for the expanded RTL. */
rtx
expand_mult_highpart (mode, op0, cnst1, target, unsignedp, max_cost)
enum machine_mode mode;
rtx op0, target;
unsigned HOST_WIDE_INT cnst1;
int unsignedp;
int max_cost;
{
enum machine_mode wider_mode = GET_MODE_WIDER_MODE (mode);
optab mul_highpart_optab;
optab moptab;
rtx tem;
int size = GET_MODE_BITSIZE (mode);
rtx op1, wide_op1;
/* We can't support modes wider than HOST_BITS_PER_INT. */
if (size > HOST_BITS_PER_WIDE_INT)
abort ();
op1 = gen_int_mode (cnst1, mode);
wide_op1
= immed_double_const (cnst1,
(unsignedp
? (HOST_WIDE_INT) 0
: -(cnst1 >> (HOST_BITS_PER_WIDE_INT - 1))),
wider_mode);
/* expand_mult handles constant multiplication of word_mode
or narrower. It does a poor job for large modes. */
if (size < BITS_PER_WORD
&& mul_cost[(int) wider_mode] + shift_cost[size-1] < max_cost)
{
/* We have to do this, since expand_binop doesn't do conversion for
multiply. Maybe change expand_binop to handle widening multiply? */
op0 = convert_to_mode (wider_mode, op0, unsignedp);
/* We know that this can't have signed overflow, so pretend this is
an unsigned multiply. */
tem = expand_mult (wider_mode, op0, wide_op1, NULL_RTX, 0);
tem = expand_shift (RSHIFT_EXPR, wider_mode, tem,
build_int_2 (size, 0), NULL_RTX, 1);
return convert_modes (mode, wider_mode, tem, unsignedp);
}
if (target == 0)
target = gen_reg_rtx (mode);
/* Firstly, try using a multiplication insn that only generates the needed
high part of the product, and in the sign flavor of unsignedp. */
if (mul_highpart_cost[(int) mode] < max_cost)
{
mul_highpart_optab = unsignedp ? umul_highpart_optab : smul_highpart_optab;
target = expand_binop (mode, mul_highpart_optab,
op0, op1, target, unsignedp, OPTAB_DIRECT);
if (target)
return target;
}
/* Secondly, same as above, but use sign flavor opposite of unsignedp.
Need to adjust the result after the multiplication. */
if (size - 1 < BITS_PER_WORD
&& (mul_highpart_cost[(int) mode] + 2 * shift_cost[size-1] + 4 * add_cost
< max_cost))
{
mul_highpart_optab = unsignedp ? smul_highpart_optab : umul_highpart_optab;
target = expand_binop (mode, mul_highpart_optab,
op0, op1, target, unsignedp, OPTAB_DIRECT);
if (target)
/* We used the wrong signedness. Adjust the result. */
return expand_mult_highpart_adjust (mode, target, op0,
op1, target, unsignedp);
}
/* Try widening multiplication. */
moptab = unsignedp ? umul_widen_optab : smul_widen_optab;
if (moptab->handlers[(int) wider_mode].insn_code != CODE_FOR_nothing
&& mul_widen_cost[(int) wider_mode] < max_cost)
{
op1 = force_reg (mode, op1);
goto try;
}
/* Try widening the mode and perform a non-widening multiplication. */
moptab = smul_optab;
if (smul_optab->handlers[(int) wider_mode].insn_code != CODE_FOR_nothing
&& size - 1 < BITS_PER_WORD
&& mul_cost[(int) wider_mode] + shift_cost[size-1] < max_cost)
{
op1 = wide_op1;
goto try;
}
/* Try widening multiplication of opposite signedness, and adjust. */
moptab = unsignedp ? smul_widen_optab : umul_widen_optab;
if (moptab->handlers[(int) wider_mode].insn_code != CODE_FOR_nothing
&& size - 1 < BITS_PER_WORD
&& (mul_widen_cost[(int) wider_mode]
+ 2 * shift_cost[size-1] + 4 * add_cost < max_cost))
{
rtx regop1 = force_reg (mode, op1);
tem = expand_binop (wider_mode, moptab, op0, regop1,
NULL_RTX, ! unsignedp, OPTAB_WIDEN);
if (tem != 0)
{
/* Extract the high half of the just generated product. */
tem = expand_shift (RSHIFT_EXPR, wider_mode, tem,
build_int_2 (size, 0), NULL_RTX, 1);
tem = convert_modes (mode, wider_mode, tem, unsignedp);
/* We used the wrong signedness. Adjust the result. */
return expand_mult_highpart_adjust (mode, tem, op0, op1,
target, unsignedp);
}
}
return 0;
try:
/* Pass NULL_RTX as target since TARGET has wrong mode. */
tem = expand_binop (wider_mode, moptab, op0, op1,
NULL_RTX, unsignedp, OPTAB_WIDEN);
if (tem == 0)
return 0;
/* Extract the high half of the just generated product. */
if (mode == word_mode)
{
return gen_highpart (mode, tem);
}
else
{
tem = expand_shift (RSHIFT_EXPR, wider_mode, tem,
build_int_2 (size, 0), NULL_RTX, 1);
return convert_modes (mode, wider_mode, tem, unsignedp);
}
}
/* Emit the code to divide OP0 by OP1, putting the result in TARGET
if that is convenient, and returning where the result is.
You may request either the quotient or the remainder as the result;
specify REM_FLAG nonzero to get the remainder.
CODE is the expression code for which kind of division this is;
it controls how rounding is done. MODE is the machine mode to use.
UNSIGNEDP nonzero means do unsigned division. */
/* ??? For CEIL_MOD_EXPR, can compute incorrect remainder with ANDI
and then correct it by or'ing in missing high bits
if result of ANDI is nonzero.
For ROUND_MOD_EXPR, can use ANDI and then sign-extend the result.
This could optimize to a bfexts instruction.
But C doesn't use these operations, so their optimizations are
left for later. */
/* ??? For modulo, we don't actually need the highpart of the first product,
the low part will do nicely. And for small divisors, the second multiply
can also be a low-part only multiply or even be completely left out.
E.g. to calculate the remainder of a division by 3 with a 32 bit
multiply, multiply with 0x55555556 and extract the upper two bits;
the result is exact for inputs up to 0x1fffffff.
The input range can be reduced by using cross-sum rules.
For odd divisors >= 3, the following table gives right shift counts
so that if an number is shifted by an integer multiple of the given
amount, the remainder stays the same:
2, 4, 3, 6, 10, 12, 4, 8, 18, 6, 11, 20, 18, 0, 5, 10, 12, 0, 12, 20,
14, 12, 23, 21, 8, 0, 20, 18, 0, 0, 6, 12, 0, 22, 0, 18, 20, 30, 0, 0,
0, 8, 0, 11, 12, 10, 36, 0, 30, 0, 0, 12, 0, 0, 0, 0, 44, 12, 24, 0,
20, 0, 7, 14, 0, 18, 36, 0, 0, 46, 60, 0, 42, 0, 15, 24, 20, 0, 0, 33,
0, 20, 0, 0, 18, 0, 60, 0, 0, 0, 0, 0, 40, 18, 0, 0, 12
Cross-sum rules for even numbers can be derived by leaving as many bits
to the right alone as the divisor has zeros to the right.
E.g. if x is an unsigned 32 bit number:
(x mod 12) == (((x & 1023) + ((x >> 8) & ~3)) * 0x15555558 >> 2 * 3) >> 28
*/
#define EXACT_POWER_OF_2_OR_ZERO_P(x) (((x) & ((x) - 1)) == 0)
rtx
expand_divmod (rem_flag, code, mode, op0, op1, target, unsignedp)
int rem_flag;
enum tree_code code;
enum machine_mode mode;
rtx op0, op1, target;
int unsignedp;
{
enum machine_mode compute_mode;
rtx tquotient;
rtx quotient = 0, remainder = 0;
rtx last;
int size;
rtx insn, set;
optab optab1, optab2;
int op1_is_constant, op1_is_pow2;
int max_cost, extra_cost;
static HOST_WIDE_INT last_div_const = 0;
op1_is_constant = GET_CODE (op1) == CONST_INT;
op1_is_pow2 = (op1_is_constant
&& ((EXACT_POWER_OF_2_OR_ZERO_P (INTVAL (op1))
|| (! unsignedp && EXACT_POWER_OF_2_OR_ZERO_P (-INTVAL (op1))))));
/*
This is the structure of expand_divmod:
First comes code to fix up the operands so we can perform the operations
correctly and efficiently.
Second comes a switch statement with code specific for each rounding mode.
For some special operands this code emits all RTL for the desired
operation, for other cases, it generates only a quotient and stores it in
QUOTIENT. The case for trunc division/remainder might leave quotient = 0,
to indicate that it has not done anything.
Last comes code that finishes the operation. If QUOTIENT is set and
REM_FLAG is set, the remainder is computed as OP0 - QUOTIENT * OP1. If
QUOTIENT is not set, it is computed using trunc rounding.
We try to generate special code for division and remainder when OP1 is a
constant. If |OP1| = 2**n we can use shifts and some other fast
operations. For other values of OP1, we compute a carefully selected
fixed-point approximation m = 1/OP1, and generate code that multiplies OP0
by m.
In all cases but EXACT_DIV_EXPR, this multiplication requires the upper
half of the product. Different strategies for generating the product are
implemented in expand_mult_highpart.
If what we actually want is the remainder, we generate that by another
by-constant multiplication and a subtraction. */
/* We shouldn't be called with OP1 == const1_rtx, but some of the
code below will malfunction if we are, so check here and handle
the special case if so. */
if (op1 == const1_rtx)
return rem_flag ? const0_rtx : op0;
/* When dividing by -1, we could get an overflow.
negv_optab can handle overflows. */
if (! unsignedp && op1 == constm1_rtx)
{
if (rem_flag)
return const0_rtx;
return expand_unop (mode, flag_trapv && GET_MODE_CLASS(mode) == MODE_INT
? negv_optab : neg_optab, op0, target, 0);
}
if (target
/* Don't use the function value register as a target
since we have to read it as well as write it,
and function-inlining gets confused by this. */
&& ((REG_P (target) && REG_FUNCTION_VALUE_P (target))
/* Don't clobber an operand while doing a multi-step calculation. */
|| ((rem_flag || op1_is_constant)
&& (reg_mentioned_p (target, op0)
|| (GET_CODE (op0) == MEM && GET_CODE (target) == MEM)))
|| reg_mentioned_p (target, op1)
|| (GET_CODE (op1) == MEM && GET_CODE (target) == MEM)))
target = 0;
/* Get the mode in which to perform this computation. Normally it will
be MODE, but sometimes we can't do the desired operation in MODE.
If so, pick a wider mode in which we can do the operation. Convert
to that mode at the start to avoid repeated conversions.
First see what operations we need. These depend on the expression
we are evaluating. (We assume that divxx3 insns exist under the
same conditions that modxx3 insns and that these insns don't normally
fail. If these assumptions are not correct, we may generate less
efficient code in some cases.)
Then see if we find a mode in which we can open-code that operation
(either a division, modulus, or shift). Finally, check for the smallest
mode for which we can do the operation with a library call. */
/* We might want to refine this now that we have division-by-constant
optimization. Since expand_mult_highpart tries so many variants, it is
not straightforward to generalize this. Maybe we should make an array
of possible modes in init_expmed? Save this for GCC 2.7. */
optab1 = (op1_is_pow2 ? (unsignedp ? lshr_optab : ashr_optab)
: (unsignedp ? udiv_optab : sdiv_optab));
optab2 = (op1_is_pow2 ? optab1 : (unsignedp ? udivmod_optab : sdivmod_optab));
for (compute_mode = mode; compute_mode != VOIDmode;
compute_mode = GET_MODE_WIDER_MODE (compute_mode))
if (optab1->handlers[(int) compute_mode].insn_code != CODE_FOR_nothing
|| optab2->handlers[(int) compute_mode].insn_code != CODE_FOR_nothing)
break;
if (compute_mode == VOIDmode)
for (compute_mode = mode; compute_mode != VOIDmode;
compute_mode = GET_MODE_WIDER_MODE (compute_mode))
if (optab1->handlers[(int) compute_mode].libfunc
|| optab2->handlers[(int) compute_mode].libfunc)
break;
/* If we still couldn't find a mode, use MODE, but we'll probably abort
in expand_binop. */
if (compute_mode == VOIDmode)
compute_mode = mode;
if (target && GET_MODE (target) == compute_mode)
tquotient = target;
else
tquotient = gen_reg_rtx (compute_mode);
size = GET_MODE_BITSIZE (compute_mode);
#if 0
/* It should be possible to restrict the precision to GET_MODE_BITSIZE
(mode), and thereby get better code when OP1 is a constant. Do that
later. It will require going over all usages of SIZE below. */
size = GET_MODE_BITSIZE (mode);
#endif
/* Only deduct something for a REM if the last divide done was
for a different constant. Then set the constant of the last
divide. */
max_cost = div_cost[(int) compute_mode]
- (rem_flag && ! (last_div_const != 0 && op1_is_constant
&& INTVAL (op1) == last_div_const)
? mul_cost[(int) compute_mode] + add_cost : 0);
last_div_const = ! rem_flag && op1_is_constant ? INTVAL (op1) : 0;
/* Now convert to the best mode to use. */
if (compute_mode != mode)
{
op0 = convert_modes (compute_mode, mode, op0, unsignedp);
op1 = convert_modes (compute_mode, mode, op1, unsignedp);
/* convert_modes may have placed op1 into a register, so we
must recompute the following. */
op1_is_constant = GET_CODE (op1) == CONST_INT;
op1_is_pow2 = (op1_is_constant
&& ((EXACT_POWER_OF_2_OR_ZERO_P (INTVAL (op1))
|| (! unsignedp
&& EXACT_POWER_OF_2_OR_ZERO_P (-INTVAL (op1)))))) ;
}
/* If one of the operands is a volatile MEM, copy it into a register. */
if (GET_CODE (op0) == MEM && MEM_VOLATILE_P (op0))
op0 = force_reg (compute_mode, op0);
if (GET_CODE (op1) == MEM && MEM_VOLATILE_P (op1))
op1 = force_reg (compute_mode, op1);
/* If we need the remainder or if OP1 is constant, we need to
put OP0 in a register in case it has any queued subexpressions. */
if (rem_flag || op1_is_constant)
op0 = force_reg (compute_mode, op0);
last = get_last_insn ();
/* Promote floor rounding to trunc rounding for unsigned operations. */
if (unsignedp)
{
if (code == FLOOR_DIV_EXPR)
code = TRUNC_DIV_EXPR;
if (code == FLOOR_MOD_EXPR)
code = TRUNC_MOD_EXPR;
if (code == EXACT_DIV_EXPR && op1_is_pow2)
code = TRUNC_DIV_EXPR;
}
if (op1 != const0_rtx)
switch (code)
{
case TRUNC_MOD_EXPR:
case TRUNC_DIV_EXPR:
if (op1_is_constant)
{
if (unsignedp)
{
unsigned HOST_WIDE_INT mh, ml;
int pre_shift, post_shift;
int dummy;
unsigned HOST_WIDE_INT d = INTVAL (op1);
if (EXACT_POWER_OF_2_OR_ZERO_P (d))
{
pre_shift = floor_log2 (d);
if (rem_flag)
{
remainder
= expand_binop (compute_mode, and_optab, op0,
GEN_INT (((HOST_WIDE_INT) 1 << pre_shift) - 1),
remainder, 1,
OPTAB_LIB_WIDEN);
if (remainder)
return gen_lowpart (mode, remainder);
}
quotient = expand_shift (RSHIFT_EXPR, compute_mode, op0,
build_int_2 (pre_shift, 0),
tquotient, 1);
}
else if (size <= HOST_BITS_PER_WIDE_INT)
{
if (d >= ((unsigned HOST_WIDE_INT) 1 << (size - 1)))
{
/* Most significant bit of divisor is set; emit an scc
insn. */
quotient = emit_store_flag (tquotient, GEU, op0, op1,
compute_mode, 1, 1);
if (quotient == 0)
goto fail1;
}
else
{
/* Find a suitable multiplier and right shift count
instead of multiplying with D. */
mh = choose_multiplier (d, size, size,
&ml, &post_shift, &dummy);
/* If the suggested multiplier is more than SIZE bits,
we can do better for even divisors, using an
initial right shift. */
if (mh != 0 && (d & 1) == 0)
{
pre_shift = floor_log2 (d & -d);
mh = choose_multiplier (d >> pre_shift, size,
size - pre_shift,
&ml, &post_shift, &dummy);
if (mh)
abort ();
}
else
pre_shift = 0;
if (mh != 0)
{
rtx t1, t2, t3, t4;
if (post_shift - 1 >= BITS_PER_WORD)
goto fail1;
extra_cost = (shift_cost[post_shift - 1]
+ shift_cost[1] + 2 * add_cost);
t1 = expand_mult_highpart (compute_mode, op0, ml,
NULL_RTX, 1,
max_cost - extra_cost);
if (t1 == 0)
goto fail1;
t2 = force_operand (gen_rtx_MINUS (compute_mode,
op0, t1),
NULL_RTX);
t3 = expand_shift (RSHIFT_EXPR, compute_mode, t2,
build_int_2 (1, 0), NULL_RTX,1);
t4 = force_operand (gen_rtx_PLUS (compute_mode,
t1, t3),
NULL_RTX);
quotient
= expand_shift (RSHIFT_EXPR, compute_mode, t4,
build_int_2 (post_shift - 1, 0),
tquotient, 1);
}
else
{
rtx t1, t2;
if (pre_shift >= BITS_PER_WORD
|| post_shift >= BITS_PER_WORD)
goto fail1;
t1 = expand_shift (RSHIFT_EXPR, compute_mode, op0,
build_int_2 (pre_shift, 0),
NULL_RTX, 1);
extra_cost = (shift_cost[pre_shift]
+ shift_cost[post_shift]);
t2 = expand_mult_highpart (compute_mode, t1, ml,
NULL_RTX, 1,
max_cost - extra_cost);
if (t2 == 0)
goto fail1;
quotient
= expand_shift (RSHIFT_EXPR, compute_mode, t2,
build_int_2 (post_shift, 0),
tquotient, 1);
}
}
}
else /* Too wide mode to use tricky code */
break;
insn = get_last_insn ();
if (insn != last
&& (set = single_set (insn)) != 0
&& SET_DEST (set) == quotient)
set_unique_reg_note (insn,
REG_EQUAL,
gen_rtx_UDIV (compute_mode, op0, op1));
}
else /* TRUNC_DIV, signed */
{
unsigned HOST_WIDE_INT ml;
int lgup, post_shift;
HOST_WIDE_INT d = INTVAL (op1);
unsigned HOST_WIDE_INT abs_d = d >= 0 ? d : -d;
/* n rem d = n rem -d */
if (rem_flag && d < 0)
{
d = abs_d;
op1 = gen_int_mode (abs_d, compute_mode);
}
if (d == 1)
quotient = op0;
else if (d == -1)
quotient = expand_unop (compute_mode, neg_optab, op0,
tquotient, 0);
else if (abs_d == (unsigned HOST_WIDE_INT) 1 << (size - 1))
{
/* This case is not handled correctly below. */
quotient = emit_store_flag (tquotient, EQ, op0, op1,
compute_mode, 1, 1);
if (quotient == 0)
goto fail1;
}
else if (EXACT_POWER_OF_2_OR_ZERO_P (d)
&& (rem_flag ? smod_pow2_cheap : sdiv_pow2_cheap)
/* ??? The cheap metric is computed only for
word_mode. If this operation is wider, this may
not be so. Assume true if the optab has an
expander for this mode. */
&& (((rem_flag ? smod_optab : sdiv_optab)
->handlers[(int) compute_mode].insn_code
!= CODE_FOR_nothing)
|| (sdivmod_optab->handlers[(int) compute_mode]
.insn_code != CODE_FOR_nothing)))
;
else if (EXACT_POWER_OF_2_OR_ZERO_P (abs_d))
{
lgup = floor_log2 (abs_d);
if (BRANCH_COST < 1 || (abs_d != 2 && BRANCH_COST < 3))
{
rtx label = gen_label_rtx ();
rtx t1;
t1 = copy_to_mode_reg (compute_mode, op0);
do_cmp_and_jump (t1, const0_rtx, GE,
compute_mode, label);
expand_inc (t1, gen_int_mode (abs_d - 1,
compute_mode));
emit_label (label);
quotient = expand_shift (RSHIFT_EXPR, compute_mode, t1,
build_int_2 (lgup, 0),
tquotient, 0);
}
else
{
rtx t1, t2, t3;
t1 = expand_shift (RSHIFT_EXPR, compute_mode, op0,
build_int_2 (size - 1, 0),
NULL_RTX, 0);
t2 = expand_shift (RSHIFT_EXPR, compute_mode, t1,
build_int_2 (size - lgup, 0),
NULL_RTX, 1);
t3 = force_operand (gen_rtx_PLUS (compute_mode,
op0, t2),
NULL_RTX);
quotient = expand_shift (RSHIFT_EXPR, compute_mode, t3,
build_int_2 (lgup, 0),
tquotient, 0);
}
/* We have computed OP0 / abs(OP1). If OP1 is negative, negate
the quotient. */
if (d < 0)
{
insn = get_last_insn ();
if (insn != last
&& (set = single_set (insn)) != 0
&& SET_DEST (set) == quotient
&& abs_d < ((unsigned HOST_WIDE_INT) 1
<< (HOST_BITS_PER_WIDE_INT - 1)))
set_unique_reg_note (insn,
REG_EQUAL,
gen_rtx_DIV (compute_mode,
op0,
GEN_INT
(trunc_int_for_mode
(abs_d,
compute_mode))));
quotient = expand_unop (compute_mode, neg_optab,
quotient, quotient, 0);
}
}
else if (size <= HOST_BITS_PER_WIDE_INT)
{
choose_multiplier (abs_d, size, size - 1,
&ml, &post_shift, &lgup);
if (ml < (unsigned HOST_WIDE_INT) 1 << (size - 1))
{
rtx t1, t2, t3;
if (post_shift >= BITS_PER_WORD
|| size - 1 >= BITS_PER_WORD)
goto fail1;
extra_cost = (shift_cost[post_shift]
+ shift_cost[size - 1] + add_cost);
t1 = expand_mult_highpart (compute_mode, op0, ml,
NULL_RTX, 0,
max_cost - extra_cost);
if (t1 == 0)
goto fail1;
t2 = expand_shift (RSHIFT_EXPR, compute_mode, t1,
build_int_2 (post_shift, 0), NULL_RTX, 0);
t3 = expand_shift (RSHIFT_EXPR, compute_mode, op0,
build_int_2 (size - 1, 0), NULL_RTX, 0);
if (d < 0)
quotient
= force_operand (gen_rtx_MINUS (compute_mode,
t3, t2),
tquotient);
else
quotient
= force_operand (gen_rtx_MINUS (compute_mode,
t2, t3),
tquotient);
}
else
{
rtx t1, t2, t3, t4;
if (post_shift >= BITS_PER_WORD
|| size - 1 >= BITS_PER_WORD)
goto fail1;
ml |= (~(unsigned HOST_WIDE_INT) 0) << (size - 1);
extra_cost = (shift_cost[post_shift]
+ shift_cost[size - 1] + 2 * add_cost);
t1 = expand_mult_highpart (compute_mode, op0, ml,
NULL_RTX, 0,
max_cost - extra_cost);
if (t1 == 0)
goto fail1;
t2 = force_operand (gen_rtx_PLUS (compute_mode,
t1, op0),
NULL_RTX);
t3 = expand_shift (RSHIFT_EXPR, compute_mode, t2,
build_int_2 (post_shift, 0),
NULL_RTX, 0);
t4 = expand_shift (RSHIFT_EXPR, compute_mode, op0,
build_int_2 (size - 1, 0),
NULL_RTX, 0);
if (d < 0)
quotient
= force_operand (gen_rtx_MINUS (compute_mode,
t4, t3),
tquotient);
else
quotient
= force_operand (gen_rtx_MINUS (compute_mode,
t3, t4),
tquotient);
}
}
else /* Too wide mode to use tricky code */
break;
insn = get_last_insn ();
if (insn != last
&& (set = single_set (insn)) != 0
&& SET_DEST (set) == quotient)
set_unique_reg_note (insn,
REG_EQUAL,
gen_rtx_DIV (compute_mode, op0, op1));
}
break;
}
fail1:
delete_insns_since (last);
break;
case FLOOR_DIV_EXPR:
case FLOOR_MOD_EXPR:
/* We will come here only for signed operations. */
if (op1_is_constant && HOST_BITS_PER_WIDE_INT >= size)
{
unsigned HOST_WIDE_INT mh, ml;
int pre_shift, lgup, post_shift;
HOST_WIDE_INT d = INTVAL (op1);
if (d > 0)
{
/* We could just as easily deal with negative constants here,
but it does not seem worth the trouble for GCC 2.6. */
if (EXACT_POWER_OF_2_OR_ZERO_P (d))
{
pre_shift = floor_log2 (d);
if (rem_flag)
{
remainder = expand_binop (compute_mode, and_optab, op0,
GEN_INT (((HOST_WIDE_INT) 1 << pre_shift) - 1),
remainder, 0, OPTAB_LIB_WIDEN);
if (remainder)
return gen_lowpart (mode, remainder);
}
quotient = expand_shift (RSHIFT_EXPR, compute_mode, op0,
build_int_2 (pre_shift, 0),
tquotient, 0);
}
else
{
rtx t1, t2, t3, t4;
mh = choose_multiplier (d, size, size - 1,
&ml, &post_shift, &lgup);
if (mh)
abort ();
if (post_shift < BITS_PER_WORD
&& size - 1 < BITS_PER_WORD)
{
t1 = expand_shift (RSHIFT_EXPR, compute_mode, op0,
build_int_2 (size - 1, 0),
NULL_RTX, 0);
t2 = expand_binop (compute_mode, xor_optab, op0, t1,
NULL_RTX, 0, OPTAB_WIDEN);
extra_cost = (shift_cost[post_shift]
+ shift_cost[size - 1] + 2 * add_cost);
t3 = expand_mult_highpart (compute_mode, t2, ml,
NULL_RTX, 1,
max_cost - extra_cost);
if (t3 != 0)
{
t4 = expand_shift (RSHIFT_EXPR, compute_mode, t3,
build_int_2 (post_shift, 0),
NULL_RTX, 1);
quotient = expand_binop (compute_mode, xor_optab,
t4, t1, tquotient, 0,
OPTAB_WIDEN);
}
}
}
}
else
{
rtx nsign, t1, t2, t3, t4;
t1 = force_operand (gen_rtx_PLUS (compute_mode,
op0, constm1_rtx), NULL_RTX);
t2 = expand_binop (compute_mode, ior_optab, op0, t1, NULL_RTX,
0, OPTAB_WIDEN);
nsign = expand_shift (RSHIFT_EXPR, compute_mode, t2,
build_int_2 (size - 1, 0), NULL_RTX, 0);
t3 = force_operand (gen_rtx_MINUS (compute_mode, t1, nsign),
NULL_RTX);
t4 = expand_divmod (0, TRUNC_DIV_EXPR, compute_mode, t3, op1,
NULL_RTX, 0);
if (t4)
{
rtx t5;
t5 = expand_unop (compute_mode, one_cmpl_optab, nsign,
NULL_RTX, 0);
quotient = force_operand (gen_rtx_PLUS (compute_mode,
t4, t5),
tquotient);
}
}
}
if (quotient != 0)
break;
delete_insns_since (last);
/* Try using an instruction that produces both the quotient and
remainder, using truncation. We can easily compensate the quotient
or remainder to get floor rounding, once we have the remainder.
Notice that we compute also the final remainder value here,
and return the result right away. */
if (target == 0 || GET_MODE (target) != compute_mode)
target = gen_reg_rtx (compute_mode);
if (rem_flag)
{
remainder
= GET_CODE (target) == REG ? target : gen_reg_rtx (compute_mode);
quotient = gen_reg_rtx (compute_mode);
}
else
{
quotient
= GET_CODE (target) == REG ? target : gen_reg_rtx (compute_mode);
remainder = gen_reg_rtx (compute_mode);
}
if (expand_twoval_binop (sdivmod_optab, op0, op1,
quotient, remainder, 0))
{
/* This could be computed with a branch-less sequence.
Save that for later. */
rtx tem;
rtx label = gen_label_rtx ();
do_cmp_and_jump (remainder, const0_rtx, EQ, compute_mode, label);
tem = expand_binop (compute_mode, xor_optab, op0, op1,
NULL_RTX, 0, OPTAB_WIDEN);
do_cmp_and_jump (tem, const0_rtx, GE, compute_mode, label);
expand_dec (quotient, const1_rtx);
expand_inc (remainder, op1);
emit_label (label);
return gen_lowpart (mode, rem_flag ? remainder : quotient);
}
/* No luck with division elimination or divmod. Have to do it
by conditionally adjusting op0 *and* the result. */
{
rtx label1, label2, label3, label4, label5;
rtx adjusted_op0;
rtx tem;
quotient = gen_reg_rtx (compute_mode);
adjusted_op0 = copy_to_mode_reg (compute_mode, op0);
label1 = gen_label_rtx ();
label2 = gen_label_rtx ();
label3 = gen_label_rtx ();
label4 = gen_label_rtx ();
label5 = gen_label_rtx ();
do_cmp_and_jump (op1, const0_rtx, LT, compute_mode, label2);
do_cmp_and_jump (adjusted_op0, const0_rtx, LT, compute_mode, label1);
tem = expand_binop (compute_mode, sdiv_optab, adjusted_op0, op1,
quotient, 0, OPTAB_LIB_WIDEN);
if (tem != quotient)
emit_move_insn (quotient, tem);
emit_jump_insn (gen_jump (label5));
emit_barrier ();
emit_label (label1);
expand_inc (adjusted_op0, const1_rtx);
emit_jump_insn (gen_jump (label4));
emit_barrier ();
emit_label (label2);
do_cmp_and_jump (adjusted_op0, const0_rtx, GT, compute_mode, label3);
tem = expand_binop (compute_mode, sdiv_optab, adjusted_op0, op1,
quotient, 0, OPTAB_LIB_WIDEN);
if (tem != quotient)
emit_move_insn (quotient, tem);
emit_jump_insn (gen_jump (label5));
emit_barrier ();
emit_label (label3);
expand_dec (adjusted_op0, const1_rtx);
emit_label (label4);
tem = expand_binop (compute_mode, sdiv_optab, adjusted_op0, op1,
quotient, 0, OPTAB_LIB_WIDEN);
if (tem != quotient)
emit_move_insn (quotient, tem);
expand_dec (quotient, const1_rtx);
emit_label (label5);
}
break;
case CEIL_DIV_EXPR:
case CEIL_MOD_EXPR:
if (unsignedp)
{
if (op1_is_constant && EXACT_POWER_OF_2_OR_ZERO_P (INTVAL (op1)))
{
rtx t1, t2, t3;
unsigned HOST_WIDE_INT d = INTVAL (op1);
t1 = expand_shift (RSHIFT_EXPR, compute_mode, op0,
build_int_2 (floor_log2 (d), 0),
tquotient, 1);
t2 = expand_binop (compute_mode, and_optab, op0,
GEN_INT (d - 1),
NULL_RTX, 1, OPTAB_LIB_WIDEN);
t3 = gen_reg_rtx (compute_mode);
t3 = emit_store_flag (t3, NE, t2, const0_rtx,
compute_mode, 1, 1);
if (t3 == 0)
{
rtx lab;
lab = gen_label_rtx ();
do_cmp_and_jump (t2, const0_rtx, EQ, compute_mode, lab);
expand_inc (t1, const1_rtx);
emit_label (lab);
quotient = t1;
}
else
quotient = force_operand (gen_rtx_PLUS (compute_mode,
t1, t3),
tquotient);
break;
}
/* Try using an instruction that produces both the quotient and
remainder, using truncation. We can easily compensate the
quotient or remainder to get ceiling rounding, once we have the
remainder. Notice that we compute also the final remainder
value here, and return the result right away. */
if (target == 0 || GET_MODE (target) != compute_mode)
target = gen_reg_rtx (compute_mode);
if (rem_flag)
{
remainder = (GET_CODE (target) == REG
? target : gen_reg_rtx (compute_mode));
quotient = gen_reg_rtx (compute_mode);
}
else
{
quotient = (GET_CODE (target) == REG
? target : gen_reg_rtx (compute_mode));
remainder = gen_reg_rtx (compute_mode);
}
if (expand_twoval_binop (udivmod_optab, op0, op1, quotient,
remainder, 1))
{
/* This could be computed with a branch-less sequence.
Save that for later. */
rtx label = gen_label_rtx ();
do_cmp_and_jump (remainder, const0_rtx, EQ,
compute_mode, label);
expand_inc (quotient, const1_rtx);
expand_dec (remainder, op1);
emit_label (label);
return gen_lowpart (mode, rem_flag ? remainder : quotient);
}
/* No luck with division elimination or divmod. Have to do it
by conditionally adjusting op0 *and* the result. */
{
rtx label1, label2;
rtx adjusted_op0, tem;
quotient = gen_reg_rtx (compute_mode);
adjusted_op0 = copy_to_mode_reg (compute_mode, op0);
label1 = gen_label_rtx ();
label2 = gen_label_rtx ();
do_cmp_and_jump (adjusted_op0, const0_rtx, NE,
compute_mode, label1);
emit_move_insn (quotient, const0_rtx);
emit_jump_insn (gen_jump (label2));
emit_barrier ();
emit_label (label1);
expand_dec (adjusted_op0, const1_rtx);
tem = expand_binop (compute_mode, udiv_optab, adjusted_op0, op1,
quotient, 1, OPTAB_LIB_WIDEN);
if (tem != quotient)
emit_move_insn (quotient, tem);
expand_inc (quotient, const1_rtx);
emit_label (label2);
}
}
else /* signed */
{
if (op1_is_constant && EXACT_POWER_OF_2_OR_ZERO_P (INTVAL (op1))
&& INTVAL (op1) >= 0)
{
/* This is extremely similar to the code for the unsigned case
above. For 2.7 we should merge these variants, but for
2.6.1 I don't want to touch the code for unsigned since that
get used in C. The signed case will only be used by other
languages (Ada). */
rtx t1, t2, t3;
unsigned HOST_WIDE_INT d = INTVAL (op1);
t1 = expand_shift (RSHIFT_EXPR, compute_mode, op0,
build_int_2 (floor_log2 (d), 0),
tquotient, 0);
t2 = expand_binop (compute_mode, and_optab, op0,
GEN_INT (d - 1),
NULL_RTX, 1, OPTAB_LIB_WIDEN);
t3 = gen_reg_rtx (compute_mode);
t3 = emit_store_flag (t3, NE, t2, const0_rtx,
compute_mode, 1, 1);
if (t3 == 0)
{
rtx lab;
lab = gen_label_rtx ();
do_cmp_and_jump (t2, const0_rtx, EQ, compute_mode, lab);
expand_inc (t1, const1_rtx);
emit_label (lab);
quotient = t1;
}
else
quotient = force_operand (gen_rtx_PLUS (compute_mode,
t1, t3),
tquotient);
break;
}
/* Try using an instruction that produces both the quotient and
remainder, using truncation. We can easily compensate the
quotient or remainder to get ceiling rounding, once we have the
remainder. Notice that we compute also the final remainder
value here, and return the result right away. */
if (target == 0 || GET_MODE (target) != compute_mode)
target = gen_reg_rtx (compute_mode);
if (rem_flag)
{
remainder= (GET_CODE (target) == REG
? target : gen_reg_rtx (compute_mode));
quotient = gen_reg_rtx (compute_mode);
}
else
{
quotient = (GET_CODE (target) == REG
? target : gen_reg_rtx (compute_mode));
remainder = gen_reg_rtx (compute_mode);
}
if (expand_twoval_binop (sdivmod_optab, op0, op1, quotient,
remainder, 0))
{
/* This could be computed with a branch-less sequence.
Save that for later. */
rtx tem;
rtx label = gen_label_rtx ();
do_cmp_and_jump (remainder, const0_rtx, EQ,
compute_mode, label);
tem = expand_binop (compute_mode, xor_optab, op0, op1,
NULL_RTX, 0, OPTAB_WIDEN);
do_cmp_and_jump (tem, const0_rtx, LT, compute_mode, label);
expand_inc (quotient, const1_rtx);
expand_dec (remainder, op1);
emit_label (label);
return gen_lowpart (mode, rem_flag ? remainder : quotient);
}
/* No luck with division elimination or divmod. Have to do it
by conditionally adjusting op0 *and* the result. */
{
rtx label1, label2, label3, label4, label5;
rtx adjusted_op0;
rtx tem;
quotient = gen_reg_rtx (compute_mode);
adjusted_op0 = copy_to_mode_reg (compute_mode, op0);
label1 = gen_label_rtx ();
label2 = gen_label_rtx ();
label3 = gen_label_rtx ();
label4 = gen_label_rtx ();
label5 = gen_label_rtx ();
do_cmp_and_jump (op1, const0_rtx, LT, compute_mode, label2);
do_cmp_and_jump (adjusted_op0, const0_rtx, GT,
compute_mode, label1);
tem = expand_binop (compute_mode, sdiv_optab, adjusted_op0, op1,
quotient, 0, OPTAB_LIB_WIDEN);
if (tem != quotient)
emit_move_insn (quotient, tem);
emit_jump_insn (gen_jump (label5));
emit_barrier ();
emit_label (label1);
expand_dec (adjusted_op0, const1_rtx);
emit_jump_insn (gen_jump (label4));
emit_barrier ();
emit_label (label2);
do_cmp_and_jump (adjusted_op0, const0_rtx, LT,
compute_mode, label3);
tem = expand_binop (compute_mode, sdiv_optab, adjusted_op0, op1,
quotient, 0, OPTAB_LIB_WIDEN);
if (tem != quotient)
emit_move_insn (quotient, tem);
emit_jump_insn (gen_jump (label5));
emit_barrier ();
emit_label (label3);
expand_inc (adjusted_op0, const1_rtx);
emit_label (label4);
tem = expand_binop (compute_mode, sdiv_optab, adjusted_op0, op1,
quotient, 0, OPTAB_LIB_WIDEN);
if (tem != quotient)
emit_move_insn (quotient, tem);
expand_inc (quotient, const1_rtx);
emit_label (label5);
}
}
break;
case EXACT_DIV_EXPR:
if (op1_is_constant && HOST_BITS_PER_WIDE_INT >= size)
{
HOST_WIDE_INT d = INTVAL (op1);
unsigned HOST_WIDE_INT ml;
int pre_shift;
rtx t1;
pre_shift = floor_log2 (d & -d);
ml = invert_mod2n (d >> pre_shift, size);
t1 = expand_shift (RSHIFT_EXPR, compute_mode, op0,
build_int_2 (pre_shift, 0), NULL_RTX, unsignedp);
quotient = expand_mult (compute_mode, t1,
gen_int_mode (ml, compute_mode),
NULL_RTX, 0);
insn = get_last_insn ();
set_unique_reg_note (insn,
REG_EQUAL,
gen_rtx_fmt_ee (unsignedp ? UDIV : DIV,
compute_mode,
op0, op1));
}
break;
case ROUND_DIV_EXPR:
case ROUND_MOD_EXPR:
if (unsignedp)
{
rtx tem;
rtx label;
label = gen_label_rtx ();
quotient = gen_reg_rtx (compute_mode);
remainder = gen_reg_rtx (compute_mode);
if (expand_twoval_binop (udivmod_optab, op0, op1, quotient, remainder, 1) == 0)
{
rtx tem;
quotient = expand_binop (compute_mode, udiv_optab, op0, op1,
quotient, 1, OPTAB_LIB_WIDEN);
tem = expand_mult (compute_mode, quotient, op1, NULL_RTX, 1);
remainder = expand_binop (compute_mode, sub_optab, op0, tem,
remainder, 1, OPTAB_LIB_WIDEN);
}
tem = plus_constant (op1, -1);
tem = expand_shift (RSHIFT_EXPR, compute_mode, tem,
build_int_2 (1, 0), NULL_RTX, 1);
do_cmp_and_jump (remainder, tem, LEU, compute_mode, label);
expand_inc (quotient, const1_rtx);
expand_dec (remainder, op1);
emit_label (label);
}
else
{
rtx abs_rem, abs_op1, tem, mask;
rtx label;
label = gen_label_rtx ();
quotient = gen_reg_rtx (compute_mode);
remainder = gen_reg_rtx (compute_mode);
if (expand_twoval_binop (sdivmod_optab, op0, op1, quotient, remainder, 0) == 0)
{
rtx tem;
quotient = expand_binop (compute_mode, sdiv_optab, op0, op1,
quotient, 0, OPTAB_LIB_WIDEN);
tem = expand_mult (compute_mode, quotient, op1, NULL_RTX, 0);
remainder = expand_binop (compute_mode, sub_optab, op0, tem,
remainder, 0, OPTAB_LIB_WIDEN);
}
abs_rem = expand_abs (compute_mode, remainder, NULL_RTX, 1, 0);
abs_op1 = expand_abs (compute_mode, op1, NULL_RTX, 1, 0);
tem = expand_shift (LSHIFT_EXPR, compute_mode, abs_rem,
build_int_2 (1, 0), NULL_RTX, 1);
do_cmp_and_jump (tem, abs_op1, LTU, compute_mode, label);
tem = expand_binop (compute_mode, xor_optab, op0, op1,
NULL_RTX, 0, OPTAB_WIDEN);
mask = expand_shift (RSHIFT_EXPR, compute_mode, tem,
build_int_2 (size - 1, 0), NULL_RTX, 0);
tem = expand_binop (compute_mode, xor_optab, mask, const1_rtx,
NULL_RTX, 0, OPTAB_WIDEN);
tem = expand_binop (compute_mode, sub_optab, tem, mask,
NULL_RTX, 0, OPTAB_WIDEN);
expand_inc (quotient, tem);
tem = expand_binop (compute_mode, xor_optab, mask, op1,
NULL_RTX, 0, OPTAB_WIDEN);
tem = expand_binop (compute_mode, sub_optab, tem, mask,
NULL_RTX, 0, OPTAB_WIDEN);
expand_dec (remainder, tem);
emit_label (label);
}
return gen_lowpart (mode, rem_flag ? remainder : quotient);
default:
abort ();
}
if (quotient == 0)
{
if (target && GET_MODE (target) != compute_mode)
target = 0;
if (rem_flag)
{
/* Try to produce the remainder without producing the quotient.
If we seem to have a divmod pattern that does not require widening,
don't try widening here. We should really have an WIDEN argument
to expand_twoval_binop, since what we'd really like to do here is
1) try a mod insn in compute_mode
2) try a divmod insn in compute_mode
3) try a div insn in compute_mode and multiply-subtract to get
remainder
4) try the same things with widening allowed. */
remainder
= sign_expand_binop (compute_mode, umod_optab, smod_optab,
op0, op1, target,
unsignedp,
((optab2->handlers[(int) compute_mode].insn_code
!= CODE_FOR_nothing)
? OPTAB_DIRECT : OPTAB_WIDEN));
if (remainder == 0)
{
/* No luck there. Can we do remainder and divide at once
without a library call? */
remainder = gen_reg_rtx (compute_mode);
if (! expand_twoval_binop ((unsignedp
? udivmod_optab
: sdivmod_optab),
op0, op1,
NULL_RTX, remainder, unsignedp))
remainder = 0;
}
if (remainder)
return gen_lowpart (mode, remainder);
}
/* Produce the quotient. Try a quotient insn, but not a library call.
If we have a divmod in this mode, use it in preference to widening
the div (for this test we assume it will not fail). Note that optab2
is set to the one of the two optabs that the call below will use. */
quotient
= sign_expand_binop (compute_mode, udiv_optab, sdiv_optab,
op0, op1, rem_flag ? NULL_RTX : target,
unsignedp,
((optab2->handlers[(int) compute_mode].insn_code
!= CODE_FOR_nothing)
? OPTAB_DIRECT : OPTAB_WIDEN));
if (quotient == 0)
{
/* No luck there. Try a quotient-and-remainder insn,
keeping the quotient alone. */
quotient = gen_reg_rtx (compute_mode);
if (! expand_twoval_binop (unsignedp ? udivmod_optab : sdivmod_optab,
op0, op1,
quotient, NULL_RTX, unsignedp))
{
quotient = 0;
if (! rem_flag)
/* Still no luck. If we are not computing the remainder,
use a library call for the quotient. */
quotient = sign_expand_binop (compute_mode,
udiv_optab, sdiv_optab,
op0, op1, target,
unsignedp, OPTAB_LIB_WIDEN);
}
}
}
if (rem_flag)
{
if (target && GET_MODE (target) != compute_mode)
target = 0;
if (quotient == 0)
/* No divide instruction either. Use library for remainder. */
remainder = sign_expand_binop (compute_mode, umod_optab, smod_optab,
op0, op1, target,
unsignedp, OPTAB_LIB_WIDEN);
else
{
/* We divided. Now finish doing X - Y * (X / Y). */
remainder = expand_mult (compute_mode, quotient, op1,
NULL_RTX, unsignedp);
remainder = expand_binop (compute_mode, sub_optab, op0,
remainder, target, unsignedp,
OPTAB_LIB_WIDEN);
}
}
return gen_lowpart (mode, rem_flag ? remainder : quotient);
}
/* Return a tree node with data type TYPE, describing the value of X.
Usually this is an RTL_EXPR, if there is no obvious better choice.
X may be an expression, however we only support those expressions
generated by loop.c. */
tree
make_tree (type, x)
tree type;
rtx x;
{
tree t;
switch (GET_CODE (x))
{
case CONST_INT:
t = build_int_2 (INTVAL (x),
(TREE_UNSIGNED (type)
&& (GET_MODE_BITSIZE (TYPE_MODE (type)) < HOST_BITS_PER_WIDE_INT))
|| INTVAL (x) >= 0 ? 0 : -1);
TREE_TYPE (t) = type;
return t;
case CONST_DOUBLE:
if (GET_MODE (x) == VOIDmode)
{
t = build_int_2 (CONST_DOUBLE_LOW (x), CONST_DOUBLE_HIGH (x));
TREE_TYPE (t) = type;
}
else
{
REAL_VALUE_TYPE d;
REAL_VALUE_FROM_CONST_DOUBLE (d, x);
t = build_real (type, d);
}
return t;
case CONST_VECTOR:
{
int i, units;
rtx elt;
tree t = NULL_TREE;
units = CONST_VECTOR_NUNITS (x);
/* Build a tree with vector elements. */
for (i = units - 1; i >= 0; --i)
{
elt = CONST_VECTOR_ELT (x, i);
t = tree_cons (NULL_TREE, make_tree (type, elt), t);
}
return build_vector (type, t);
}
case PLUS:
return fold (build (PLUS_EXPR, type, make_tree (type, XEXP (x, 0)),
make_tree (type, XEXP (x, 1))));
case MINUS:
return fold (build (MINUS_EXPR, type, make_tree (type, XEXP (x, 0)),
make_tree (type, XEXP (x, 1))));
case NEG:
return fold (build1 (NEGATE_EXPR, type, make_tree (type, XEXP (x, 0))));
case MULT:
return fold (build (MULT_EXPR, type, make_tree (type, XEXP (x, 0)),
make_tree (type, XEXP (x, 1))));
case ASHIFT:
return fold (build (LSHIFT_EXPR, type, make_tree (type, XEXP (x, 0)),
make_tree (type, XEXP (x, 1))));
case LSHIFTRT:
t = (*lang_hooks.types.unsigned_type) (type);
return fold (convert (type,
build (RSHIFT_EXPR, t,
make_tree (t, XEXP (x, 0)),
make_tree (type, XEXP (x, 1)))));
case ASHIFTRT:
t = (*lang_hooks.types.signed_type) (type);
return fold (convert (type,
build (RSHIFT_EXPR, t,
make_tree (t, XEXP (x, 0)),
make_tree (type, XEXP (x, 1)))));
case DIV:
if (TREE_CODE (type) != REAL_TYPE)
t = (*lang_hooks.types.signed_type) (type);
else
t = type;
return fold (convert (type,
build (TRUNC_DIV_EXPR, t,
make_tree (t, XEXP (x, 0)),
make_tree (t, XEXP (x, 1)))));
case UDIV:
t = (*lang_hooks.types.unsigned_type) (type);
return fold (convert (type,
build (TRUNC_DIV_EXPR, t,
make_tree (t, XEXP (x, 0)),
make_tree (t, XEXP (x, 1)))));
default:
t = make_node (RTL_EXPR);
TREE_TYPE (t) = type;
#ifdef POINTERS_EXTEND_UNSIGNED
/* If TYPE is a POINTER_TYPE, X might be Pmode with TYPE_MODE being
ptr_mode. So convert. */
if (POINTER_TYPE_P (type) && GET_MODE (x) != TYPE_MODE (type))
x = convert_memory_address (TYPE_MODE (type), x);
#endif
RTL_EXPR_RTL (t) = x;
/* There are no insns to be output
when this rtl_expr is used. */
RTL_EXPR_SEQUENCE (t) = 0;
return t;
}
}
/* Return an rtx representing the value of X * MULT + ADD.
TARGET is a suggestion for where to store the result (an rtx).
MODE is the machine mode for the computation.
X and MULT must have mode MODE. ADD may have a different mode.
So can X (defaults to same as MODE).
UNSIGNEDP is non-zero to do unsigned multiplication.
This may emit insns. */
rtx
expand_mult_add (x, target, mult, add, mode, unsignedp)
rtx x, target, mult, add;
enum machine_mode mode;
int unsignedp;
{
tree type = (*lang_hooks.types.type_for_mode) (mode, unsignedp);
tree add_type = (GET_MODE (add) == VOIDmode
? type: (*lang_hooks.types.type_for_mode) (GET_MODE (add),
unsignedp));
tree result = fold (build (PLUS_EXPR, type,
fold (build (MULT_EXPR, type,
make_tree (type, x),
make_tree (type, mult))),
make_tree (add_type, add)));
return expand_expr (result, target, VOIDmode, 0);
}
/* Compute the logical-and of OP0 and OP1, storing it in TARGET
and returning TARGET.
If TARGET is 0, a pseudo-register or constant is returned. */
rtx
expand_and (mode, op0, op1, target)
enum machine_mode mode;
rtx op0, op1, target;
{
rtx tem = 0;
if (GET_MODE (op0) == VOIDmode && GET_MODE (op1) == VOIDmode)
tem = simplify_binary_operation (AND, mode, op0, op1);
if (tem == 0)
tem = expand_binop (mode, and_optab, op0, op1, target, 0, OPTAB_LIB_WIDEN);
if (target == 0)
target = tem;
else if (tem != target)
emit_move_insn (target, tem);
return target;
}
/* Emit a store-flags instruction for comparison CODE on OP0 and OP1
and storing in TARGET. Normally return TARGET.
Return 0 if that cannot be done.
MODE is the mode to use for OP0 and OP1 should they be CONST_INTs. If
it is VOIDmode, they cannot both be CONST_INT.
UNSIGNEDP is for the case where we have to widen the operands
to perform the operation. It says to use zero-extension.
NORMALIZEP is 1 if we should convert the result to be either zero
or one. Normalize is -1 if we should convert the result to be
either zero or -1. If NORMALIZEP is zero, the result will be left
"raw" out of the scc insn. */
rtx
emit_store_flag (target, code, op0, op1, mode, unsignedp, normalizep)
rtx target;
enum rtx_code code;
rtx op0, op1;
enum machine_mode mode;
int unsignedp;
int normalizep;
{
rtx subtarget;
enum insn_code icode;
enum machine_mode compare_mode;
enum machine_mode target_mode = GET_MODE (target);
rtx tem;
rtx last = get_last_insn ();
rtx pattern, comparison;
/* ??? Ok to do this and then fail? */
op0 = protect_from_queue (op0, 0);
op1 = protect_from_queue (op1, 0);
if (unsignedp)
code = unsigned_condition (code);
/* If one operand is constant, make it the second one. Only do this
if the other operand is not constant as well. */
if (swap_commutative_operands_p (op0, op1))
{
tem = op0;
op0 = op1;
op1 = tem;
code = swap_condition (code);
}
if (mode == VOIDmode)
mode = GET_MODE (op0);
/* For some comparisons with 1 and -1, we can convert this to
comparisons with zero. This will often produce more opportunities for
store-flag insns. */
switch (code)
{
case LT:
if (op1 == const1_rtx)
op1 = const0_rtx, code = LE;
break;
case LE:
if (op1 == constm1_rtx)
op1 = const0_rtx, code = LT;
break;
case GE:
if (op1 == const1_rtx)
op1 = const0_rtx, code = GT;
break;
case GT:
if (op1 == constm1_rtx)
op1 = const0_rtx, code = GE;
break;
case GEU:
if (op1 == const1_rtx)
op1 = const0_rtx, code = NE;
break;
case LTU:
if (op1 == const1_rtx)
op1 = const0_rtx, code = EQ;
break;
default:
break;
}
/* If we are comparing a double-word integer with zero, we can convert
the comparison into one involving a single word. */
if (GET_MODE_BITSIZE (mode) == BITS_PER_WORD * 2
&& GET_MODE_CLASS (mode) == MODE_INT
&& op1 == const0_rtx
&& (GET_CODE (op0) != MEM || ! MEM_VOLATILE_P (op0)))
{
if (code == EQ || code == NE)
{
/* Do a logical OR of the two words and compare the result. */
rtx op0h = gen_highpart (word_mode, op0);
rtx op0l = gen_lowpart (word_mode, op0);
rtx op0both = expand_binop (word_mode, ior_optab, op0h, op0l,
NULL_RTX, unsignedp, OPTAB_DIRECT);
if (op0both != 0)
return emit_store_flag (target, code, op0both, op1, word_mode,
unsignedp, normalizep);
}
else if (code == LT || code == GE)
/* If testing the sign bit, can just test on high word. */
return emit_store_flag (target, code, gen_highpart (word_mode, op0),
op1, word_mode, unsignedp, normalizep);
}
/* From now on, we won't change CODE, so set ICODE now. */
icode = setcc_gen_code[(int) code];
/* If this is A < 0 or A >= 0, we can do this by taking the ones
complement of A (for GE) and shifting the sign bit to the low bit. */
if (op1 == const0_rtx && (code == LT || code == GE)
&& GET_MODE_CLASS (mode) == MODE_INT
&& (normalizep || STORE_FLAG_VALUE == 1
|| (GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
&& ((STORE_FLAG_VALUE & GET_MODE_MASK (mode))
== (HOST_WIDE_INT) 1 << (GET_MODE_BITSIZE (mode) - 1)))))
{
subtarget = target;
/* If the result is to be wider than OP0, it is best to convert it
first. If it is to be narrower, it is *incorrect* to convert it
first. */
if (GET_MODE_SIZE (target_mode) > GET_MODE_SIZE (mode))
{
op0 = protect_from_queue (op0, 0);
op0 = convert_modes (target_mode, mode, op0, 0);
mode = target_mode;
}
if (target_mode != mode)
subtarget = 0;
if (code == GE)
op0 = expand_unop (mode, one_cmpl_optab, op0,
((STORE_FLAG_VALUE == 1 || normalizep)
? 0 : subtarget), 0);
if (STORE_FLAG_VALUE == 1 || normalizep)
/* If we are supposed to produce a 0/1 value, we want to do
a logical shift from the sign bit to the low-order bit; for
a -1/0 value, we do an arithmetic shift. */
op0 = expand_shift (RSHIFT_EXPR, mode, op0,
size_int (GET_MODE_BITSIZE (mode) - 1),
subtarget, normalizep != -1);
if (mode != target_mode)
op0 = convert_modes (target_mode, mode, op0, 0);
return op0;
}
if (icode != CODE_FOR_nothing)
{
insn_operand_predicate_fn pred;
/* We think we may be able to do this with a scc insn. Emit the
comparison and then the scc insn.
compare_from_rtx may call emit_queue, which would be deleted below
if the scc insn fails. So call it ourselves before setting LAST.
Likewise for do_pending_stack_adjust. */
emit_queue ();
do_pending_stack_adjust ();
last = get_last_insn ();
comparison
= compare_from_rtx (op0, op1, code, unsignedp, mode, NULL_RTX);
if (GET_CODE (comparison) == CONST_INT)
return (comparison == const0_rtx ? const0_rtx
: normalizep == 1 ? const1_rtx
: normalizep == -1 ? constm1_rtx
: const_true_rtx);
/* The code of COMPARISON may not match CODE if compare_from_rtx
decided to swap its operands and reverse the original code.
We know that compare_from_rtx returns either a CONST_INT or
a new comparison code, so it is safe to just extract the
code from COMPARISON. */
code = GET_CODE (comparison);
/* Get a reference to the target in the proper mode for this insn. */
compare_mode = insn_data[(int) icode].operand[0].mode;
subtarget = target;
pred = insn_data[(int) icode].operand[0].predicate;
if (preserve_subexpressions_p ()
|| ! (*pred) (subtarget, compare_mode))
subtarget = gen_reg_rtx (compare_mode);
pattern = GEN_FCN (icode) (subtarget);
if (pattern)
{
emit_insn (pattern);
/* If we are converting to a wider mode, first convert to
TARGET_MODE, then normalize. This produces better combining
opportunities on machines that have a SIGN_EXTRACT when we are
testing a single bit. This mostly benefits the 68k.
If STORE_FLAG_VALUE does not have the sign bit set when
interpreted in COMPARE_MODE, we can do this conversion as
unsigned, which is usually more efficient. */
if (GET_MODE_SIZE (target_mode) > GET_MODE_SIZE (compare_mode))
{
convert_move (target, subtarget,
(GET_MODE_BITSIZE (compare_mode)
<= HOST_BITS_PER_WIDE_INT)
&& 0 == (STORE_FLAG_VALUE
& ((HOST_WIDE_INT) 1
<< (GET_MODE_BITSIZE (compare_mode) -1))));
op0 = target;
compare_mode = target_mode;
}
else
op0 = subtarget;
/* If we want to keep subexpressions around, don't reuse our
last target. */
if (preserve_subexpressions_p ())
subtarget = 0;
/* Now normalize to the proper value in COMPARE_MODE. Sometimes
we don't have to do anything. */
if (normalizep == 0 || normalizep == STORE_FLAG_VALUE)
;
/* STORE_FLAG_VALUE might be the most negative number, so write
the comparison this way to avoid a compiler-time warning. */
else if (- normalizep == STORE_FLAG_VALUE)
op0 = expand_unop (compare_mode, neg_optab, op0, subtarget, 0);
/* We don't want to use STORE_FLAG_VALUE < 0 below since this
makes it hard to use a value of just the sign bit due to
ANSI integer constant typing rules. */
else if (GET_MODE_BITSIZE (compare_mode) <= HOST_BITS_PER_WIDE_INT
&& (STORE_FLAG_VALUE
& ((HOST_WIDE_INT) 1
<< (GET_MODE_BITSIZE (compare_mode) - 1))))
op0 = expand_shift (RSHIFT_EXPR, compare_mode, op0,
size_int (GET_MODE_BITSIZE (compare_mode) - 1),
subtarget, normalizep == 1);
else if (STORE_FLAG_VALUE & 1)
{
op0 = expand_and (compare_mode, op0, const1_rtx, subtarget);
if (normalizep == -1)
op0 = expand_unop (compare_mode, neg_optab, op0, op0, 0);
}
else
abort ();
/* If we were converting to a smaller mode, do the
conversion now. */
if (target_mode != compare_mode)
{
convert_move (target, op0, 0);
return target;
}
else
return op0;
}
}
delete_insns_since (last);
/* If expensive optimizations, use different pseudo registers for each
insn, instead of reusing the same pseudo. This leads to better CSE,
but slows down the compiler, since there are more pseudos */
subtarget = (!flag_expensive_optimizations
&& (target_mode == mode)) ? target : NULL_RTX;
/* If we reached here, we can't do this with a scc insn. However, there
are some comparisons that can be done directly. For example, if
this is an equality comparison of integers, we can try to exclusive-or
(or subtract) the two operands and use a recursive call to try the
comparison with zero. Don't do any of these cases if branches are
very cheap. */
if (BRANCH_COST > 0
&& GET_MODE_CLASS (mode) == MODE_INT && (code == EQ || code == NE)
&& op1 != const0_rtx)
{
tem = expand_binop (mode, xor_optab, op0, op1, subtarget, 1,
OPTAB_WIDEN);
if (tem == 0)
tem = expand_binop (mode, sub_optab, op0, op1, subtarget, 1,
OPTAB_WIDEN);
if (tem != 0)
tem = emit_store_flag (target, code, tem, const0_rtx,
mode, unsignedp, normalizep);
if (tem == 0)
delete_insns_since (last);
return tem;
}
/* Some other cases we can do are EQ, NE, LE, and GT comparisons with
the constant zero. Reject all other comparisons at this point. Only
do LE and GT if branches are expensive since they are expensive on
2-operand machines. */
if (BRANCH_COST == 0
|| GET_MODE_CLASS (mode) != MODE_INT || op1 != const0_rtx
|| (code != EQ && code != NE
&& (BRANCH_COST <= 1 || (code != LE && code != GT))))
return 0;
/* See what we need to return. We can only return a 1, -1, or the
sign bit. */
if (normalizep == 0)
{
if (STORE_FLAG_VALUE == 1 || STORE_FLAG_VALUE == -1)
normalizep = STORE_FLAG_VALUE;
else if (GET_MODE_BITSIZE (mode) <= HOST_BITS_PER_WIDE_INT
&& ((STORE_FLAG_VALUE & GET_MODE_MASK (mode))
== (unsigned HOST_WIDE_INT) 1 << (GET_MODE_BITSIZE (mode) - 1)))
;
else
return 0;
}
/* Try to put the result of the comparison in the sign bit. Assume we can't
do the necessary operation below. */
tem = 0;
/* To see if A <= 0, compute (A | (A - 1)). A <= 0 iff that result has
the sign bit set. */
if (code == LE)
{
/* This is destructive, so SUBTARGET can't be OP0. */
if (rtx_equal_p (subtarget, op0))
subtarget = 0;
tem = expand_binop (mode, sub_optab, op0, const1_rtx, subtarget, 0,
OPTAB_WIDEN);
if (tem)
tem = expand_binop (mode, ior_optab, op0, tem, subtarget, 0,
OPTAB_WIDEN);
}
/* To see if A > 0, compute (((signed) A) << BITS) - A, where BITS is the
number of bits in the mode of OP0, minus one. */
if (code == GT)
{
if (rtx_equal_p (subtarget, op0))
subtarget = 0;
tem = expand_shift (RSHIFT_EXPR, mode, op0,
size_int (GET_MODE_BITSIZE (mode) - 1),
subtarget, 0);
tem = expand_binop (mode, sub_optab, tem, op0, subtarget, 0,
OPTAB_WIDEN);
}
if (code == EQ || code == NE)
{
/* For EQ or NE, one way to do the comparison is to apply an operation
that converts the operand into a positive number if it is non-zero
or zero if it was originally zero. Then, for EQ, we subtract 1 and
for NE we negate. This puts the result in the sign bit. Then we
normalize with a shift, if needed.
Two operations that can do the above actions are ABS and FFS, so try
them. If that doesn't work, and MODE is smaller than a full word,
we can use zero-extension to the wider mode (an unsigned conversion)
as the operation. */
/* Note that ABS doesn't yield a positive number for INT_MIN, but
that is compensated by the subsequent overflow when subtracting
one / negating. */
if (abs_optab->handlers[(int) mode].insn_code != CODE_FOR_nothing)
tem = expand_unop (mode, abs_optab, op0, subtarget, 1);
else if (ffs_optab->handlers[(int) mode].insn_code != CODE_FOR_nothing)
tem = expand_unop (mode, ffs_optab, op0, subtarget, 1);
else if (GET_MODE_SIZE (mode) < UNITS_PER_WORD)
{
op0 = protect_from_queue (op0, 0);
tem = convert_modes (word_mode, mode, op0, 1);
mode = word_mode;
}
if (tem != 0)
{
if (code == EQ)
tem = expand_binop (mode, sub_optab, tem, const1_rtx, subtarget,
0, OPTAB_WIDEN);
else
tem = expand_unop (mode, neg_optab, tem, subtarget, 0);
}
/* If we couldn't do it that way, for NE we can "or" the two's complement
of the value with itself. For EQ, we take the one's complement of
that "or", which is an extra insn, so we only handle EQ if branches
are expensive. */
if (tem == 0 && (code == NE || BRANCH_COST > 1))
{
if (rtx_equal_p (subtarget, op0))
subtarget = 0;
tem = expand_unop (mode, neg_optab, op0, subtarget, 0);
tem = expand_binop (mode, ior_optab, tem, op0, subtarget, 0,
OPTAB_WIDEN);
if (tem && code == EQ)
tem = expand_unop (mode, one_cmpl_optab, tem, subtarget, 0);
}
}
if (tem && normalizep)
tem = expand_shift (RSHIFT_EXPR, mode, tem,
size_int (GET_MODE_BITSIZE (mode) - 1),
subtarget, normalizep == 1);
if (tem)
{
if (GET_MODE (tem) != target_mode)
{
convert_move (target, tem, 0);
tem = target;
}
else if (!subtarget)
{
emit_move_insn (target, tem);
tem = target;
}
}
else
delete_insns_since (last);
return tem;
}
/* Like emit_store_flag, but always succeeds. */
rtx
emit_store_flag_force (target, code, op0, op1, mode, unsignedp, normalizep)
rtx target;
enum rtx_code code;
rtx op0, op1;
enum machine_mode mode;
int unsignedp;
int normalizep;
{
rtx tem, label;
/* First see if emit_store_flag can do the job. */
tem = emit_store_flag (target, code, op0, op1, mode, unsignedp, normalizep);
if (tem != 0)
return tem;
if (normalizep == 0)
normalizep = 1;
/* If this failed, we have to do this with set/compare/jump/set code. */
if (GET_CODE (target) != REG
|| reg_mentioned_p (target, op0) || reg_mentioned_p (target, op1))
target = gen_reg_rtx (GET_MODE (target));
emit_move_insn (target, const1_rtx);
label = gen_label_rtx ();
do_compare_rtx_and_jump (op0, op1, code, unsignedp, mode, NULL_RTX,
NULL_RTX, label);
emit_move_insn (target, const0_rtx);
emit_label (label);
return target;
}
/* Perform possibly multi-word comparison and conditional jump to LABEL
if ARG1 OP ARG2 true where ARG1 and ARG2 are of mode MODE
The algorithm is based on the code in expr.c:do_jump.
Note that this does not perform a general comparison. Only variants
generated within expmed.c are correctly handled, others abort (but could
be handled if needed). */
static void
do_cmp_and_jump (arg1, arg2, op, mode, label)
rtx arg1, arg2, label;
enum rtx_code op;
enum machine_mode mode;
{
/* If this mode is an integer too wide to compare properly,
compare word by word. Rely on cse to optimize constant cases. */
if (GET_MODE_CLASS (mode) == MODE_INT
&& ! can_compare_p (op, mode, ccp_jump))
{
rtx label2 = gen_label_rtx ();
switch (op)
{
case LTU:
do_jump_by_parts_greater_rtx (mode, 1, arg2, arg1, label2, label);
break;
case LEU:
do_jump_by_parts_greater_rtx (mode, 1, arg1, arg2, label, label2);
break;
case LT:
do_jump_by_parts_greater_rtx (mode, 0, arg2, arg1, label2, label);
break;
case GT:
do_jump_by_parts_greater_rtx (mode, 0, arg1, arg2, label2, label);
break;
case GE:
do_jump_by_parts_greater_rtx (mode, 0, arg2, arg1, label, label2);
break;
/* do_jump_by_parts_equality_rtx compares with zero. Luckily
that's the only equality operations we do */
case EQ:
if (arg2 != const0_rtx || mode != GET_MODE(arg1))
abort ();
do_jump_by_parts_equality_rtx (arg1, label2, label);
break;
case NE:
if (arg2 != const0_rtx || mode != GET_MODE(arg1))
abort ();
do_jump_by_parts_equality_rtx (arg1, label, label2);
break;
default:
abort ();
}
emit_label (label2);
}
else
emit_cmp_and_jump_insns (arg1, arg2, op, NULL_RTX, mode, 0, label);
}
|