1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118
5119
5120
5121
5122
5123
5124
5125
5126
5127
5128
5129
5130
5131
5132
5133
5134
5135
5136
5137
5138
5139
5140
5141
5142
5143
5144
5145
5146
5147
5148
5149
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160
5161
5162
5163
5164
5165
5166
5167
5168
5169
5170
5171
5172
5173
5174
5175
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220
5221
5222
5223
5224
5225
5226
5227
5228
5229
5230
5231
5232
5233
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
5271
5272
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289
5290
5291
5292
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308
5309
5310
5311
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334
5335
5336
5337
5338
5339
5340
5341
5342
5343
5344
5345
5346
5347
5348
5349
5350
5351
5352
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364
5365
5366
5367
5368
5369
5370
5371
5372
5373
5374
5375
|
/* Analysis Utilities for Loop Vectorization.
Copyright (C) 2006-2019 Free Software Foundation, Inc.
Contributed by Dorit Nuzman <dorit@il.ibm.com>
This file is part of GCC.
GCC is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free
Software Foundation; either version 3, or (at your option) any later
version.
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
for more details.
You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING3. If not see
<http://www.gnu.org/licenses/>. */
#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "backend.h"
#include "rtl.h"
#include "tree.h"
#include "gimple.h"
#include "ssa.h"
#include "expmed.h"
#include "optabs-tree.h"
#include "insn-config.h"
#include "recog.h" /* FIXME: for insn_data */
#include "fold-const.h"
#include "stor-layout.h"
#include "tree-eh.h"
#include "gimplify.h"
#include "gimple-iterator.h"
#include "cfgloop.h"
#include "tree-vectorizer.h"
#include "dumpfile.h"
#include "builtins.h"
#include "internal-fn.h"
#include "case-cfn-macros.h"
#include "fold-const-call.h"
#include "attribs.h"
#include "cgraph.h"
#include "omp-simd-clone.h"
#include "predict.h"
#include "tree-vector-builder.h"
#include "vec-perm-indices.h"
/* Return true if we have a useful VR_RANGE range for VAR, storing it
in *MIN_VALUE and *MAX_VALUE if so. Note the range in the dump files. */
static bool
vect_get_range_info (tree var, wide_int *min_value, wide_int *max_value)
{
value_range_kind vr_type = get_range_info (var, min_value, max_value);
wide_int nonzero = get_nonzero_bits (var);
signop sgn = TYPE_SIGN (TREE_TYPE (var));
if (intersect_range_with_nonzero_bits (vr_type, min_value, max_value,
nonzero, sgn) == VR_RANGE)
{
if (dump_enabled_p ())
{
dump_generic_expr_loc (MSG_NOTE, vect_location, TDF_SLIM, var);
dump_printf (MSG_NOTE, " has range [");
dump_hex (MSG_NOTE, *min_value);
dump_printf (MSG_NOTE, ", ");
dump_hex (MSG_NOTE, *max_value);
dump_printf (MSG_NOTE, "]\n");
}
return true;
}
else
{
if (dump_enabled_p ())
{
dump_generic_expr_loc (MSG_NOTE, vect_location, TDF_SLIM, var);
dump_printf (MSG_NOTE, " has no range info\n");
}
return false;
}
}
/* Report that we've found an instance of pattern PATTERN in
statement STMT. */
static void
vect_pattern_detected (const char *name, gimple *stmt)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location, "%s: detected: %G", name, stmt);
}
/* Associate pattern statement PATTERN_STMT with ORIG_STMT_INFO and
return the pattern statement's stmt_vec_info. Set its vector type to
VECTYPE if it doesn't have one already. */
static stmt_vec_info
vect_init_pattern_stmt (gimple *pattern_stmt, stmt_vec_info orig_stmt_info,
tree vectype)
{
vec_info *vinfo = orig_stmt_info->vinfo;
stmt_vec_info pattern_stmt_info = vinfo->lookup_stmt (pattern_stmt);
if (pattern_stmt_info == NULL)
pattern_stmt_info = orig_stmt_info->vinfo->add_stmt (pattern_stmt);
gimple_set_bb (pattern_stmt, gimple_bb (orig_stmt_info->stmt));
pattern_stmt_info->pattern_stmt_p = true;
STMT_VINFO_RELATED_STMT (pattern_stmt_info) = orig_stmt_info;
STMT_VINFO_DEF_TYPE (pattern_stmt_info)
= STMT_VINFO_DEF_TYPE (orig_stmt_info);
if (!STMT_VINFO_VECTYPE (pattern_stmt_info))
STMT_VINFO_VECTYPE (pattern_stmt_info) = vectype;
return pattern_stmt_info;
}
/* Set the pattern statement of ORIG_STMT_INFO to PATTERN_STMT.
Also set the vector type of PATTERN_STMT to VECTYPE, if it doesn't
have one already. */
static void
vect_set_pattern_stmt (gimple *pattern_stmt, stmt_vec_info orig_stmt_info,
tree vectype)
{
STMT_VINFO_IN_PATTERN_P (orig_stmt_info) = true;
STMT_VINFO_RELATED_STMT (orig_stmt_info)
= vect_init_pattern_stmt (pattern_stmt, orig_stmt_info, vectype);
}
/* Add NEW_STMT to STMT_INFO's pattern definition statements. If VECTYPE
is nonnull, record that NEW_STMT's vector type is VECTYPE, which might
be different from the vector type of the final pattern statement. */
static inline void
append_pattern_def_seq (stmt_vec_info stmt_info, gimple *new_stmt,
tree vectype = NULL_TREE)
{
vec_info *vinfo = stmt_info->vinfo;
if (vectype)
{
stmt_vec_info new_stmt_info = vinfo->add_stmt (new_stmt);
STMT_VINFO_VECTYPE (new_stmt_info) = vectype;
}
gimple_seq_add_stmt_without_update (&STMT_VINFO_PATTERN_DEF_SEQ (stmt_info),
new_stmt);
}
/* The caller wants to perform new operations on vect_external variable
VAR, so that the result of the operations would also be vect_external.
Return the edge on which the operations can be performed, if one exists.
Return null if the operations should instead be treated as part of
the pattern that needs them. */
static edge
vect_get_external_def_edge (vec_info *vinfo, tree var)
{
edge e = NULL;
if (loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo))
{
e = loop_preheader_edge (loop_vinfo->loop);
if (!SSA_NAME_IS_DEFAULT_DEF (var))
{
basic_block bb = gimple_bb (SSA_NAME_DEF_STMT (var));
if (bb == NULL
|| !dominated_by_p (CDI_DOMINATORS, e->dest, bb))
e = NULL;
}
}
return e;
}
/* Return true if the target supports a vector version of CODE,
where CODE is known to map to a direct optab. ITYPE specifies
the type of (some of) the scalar inputs and OTYPE specifies the
type of the scalar result.
If CODE allows the inputs and outputs to have different type
(such as for WIDEN_SUM_EXPR), it is the input mode rather
than the output mode that determines the appropriate target pattern.
Operand 0 of the target pattern then specifies the mode that the output
must have.
When returning true, set *VECOTYPE_OUT to the vector version of OTYPE.
Also set *VECITYPE_OUT to the vector version of ITYPE if VECITYPE_OUT
is nonnull. */
static bool
vect_supportable_direct_optab_p (vec_info *vinfo, tree otype, tree_code code,
tree itype, tree *vecotype_out,
tree *vecitype_out = NULL)
{
tree vecitype = get_vectype_for_scalar_type (vinfo, itype);
if (!vecitype)
return false;
tree vecotype = get_vectype_for_scalar_type (vinfo, otype);
if (!vecotype)
return false;
optab optab = optab_for_tree_code (code, vecitype, optab_default);
if (!optab)
return false;
insn_code icode = optab_handler (optab, TYPE_MODE (vecitype));
if (icode == CODE_FOR_nothing
|| insn_data[icode].operand[0].mode != TYPE_MODE (vecotype))
return false;
*vecotype_out = vecotype;
if (vecitype_out)
*vecitype_out = vecitype;
return true;
}
/* Round bit precision PRECISION up to a full element. */
static unsigned int
vect_element_precision (unsigned int precision)
{
precision = 1 << ceil_log2 (precision);
return MAX (precision, BITS_PER_UNIT);
}
/* If OP is defined by a statement that's being considered for vectorization,
return information about that statement, otherwise return NULL. */
static stmt_vec_info
vect_get_internal_def (vec_info *vinfo, tree op)
{
stmt_vec_info def_stmt_info = vinfo->lookup_def (op);
if (def_stmt_info
&& STMT_VINFO_DEF_TYPE (def_stmt_info) == vect_internal_def)
return def_stmt_info;
return NULL;
}
/* Check whether NAME, an ssa-name used in STMT_VINFO,
is a result of a type promotion, such that:
DEF_STMT: NAME = NOP (name0)
If CHECK_SIGN is TRUE, check that either both types are signed or both are
unsigned. */
static bool
type_conversion_p (tree name, stmt_vec_info stmt_vinfo, bool check_sign,
tree *orig_type, gimple **def_stmt, bool *promotion)
{
tree type = TREE_TYPE (name);
tree oprnd0;
enum vect_def_type dt;
stmt_vec_info def_stmt_info;
if (!vect_is_simple_use (name, stmt_vinfo->vinfo, &dt, &def_stmt_info,
def_stmt))
return false;
if (dt != vect_internal_def
&& dt != vect_external_def && dt != vect_constant_def)
return false;
if (!*def_stmt)
return false;
if (!is_gimple_assign (*def_stmt))
return false;
if (!CONVERT_EXPR_CODE_P (gimple_assign_rhs_code (*def_stmt)))
return false;
oprnd0 = gimple_assign_rhs1 (*def_stmt);
*orig_type = TREE_TYPE (oprnd0);
if (!INTEGRAL_TYPE_P (type) || !INTEGRAL_TYPE_P (*orig_type)
|| ((TYPE_UNSIGNED (type) != TYPE_UNSIGNED (*orig_type)) && check_sign))
return false;
if (TYPE_PRECISION (type) >= (TYPE_PRECISION (*orig_type) * 2))
*promotion = true;
else
*promotion = false;
if (!vect_is_simple_use (oprnd0, stmt_vinfo->vinfo, &dt))
return false;
return true;
}
/* Holds information about an input operand after some sign changes
and type promotions have been peeled away. */
class vect_unpromoted_value {
public:
vect_unpromoted_value ();
void set_op (tree, vect_def_type, stmt_vec_info = NULL);
/* The value obtained after peeling away zero or more casts. */
tree op;
/* The type of OP. */
tree type;
/* The definition type of OP. */
vect_def_type dt;
/* If OP is the result of peeling at least one cast, and if the cast
of OP itself is a vectorizable statement, CASTER identifies that
statement, otherwise it is null. */
stmt_vec_info caster;
};
inline vect_unpromoted_value::vect_unpromoted_value ()
: op (NULL_TREE),
type (NULL_TREE),
dt (vect_uninitialized_def),
caster (NULL)
{
}
/* Set the operand to OP_IN, its definition type to DT_IN, and the
statement that casts it to CASTER_IN. */
inline void
vect_unpromoted_value::set_op (tree op_in, vect_def_type dt_in,
stmt_vec_info caster_in)
{
op = op_in;
type = TREE_TYPE (op);
dt = dt_in;
caster = caster_in;
}
/* If OP is a vectorizable SSA name, strip a sequence of integer conversions
to reach some vectorizable inner operand OP', continuing as long as it
is possible to convert OP' back to OP using a possible sign change
followed by a possible promotion P. Return this OP', or null if OP is
not a vectorizable SSA name. If there is a promotion P, describe its
input in UNPROM, otherwise describe OP' in UNPROM. If SINGLE_USE_P
is nonnull, set *SINGLE_USE_P to false if any of the SSA names involved
have more than one user.
A successful return means that it is possible to go from OP' to OP
via UNPROM. The cast from OP' to UNPROM is at most a sign change,
whereas the cast from UNPROM to OP might be a promotion, a sign
change, or a nop.
E.g. say we have:
signed short *ptr = ...;
signed short C = *ptr;
unsigned short B = (unsigned short) C; // sign change
signed int A = (signed int) B; // unsigned promotion
...possible other uses of A...
unsigned int OP = (unsigned int) A; // sign change
In this case it's possible to go directly from C to OP using:
OP = (unsigned int) (unsigned short) C;
+------------+ +--------------+
promotion sign change
so OP' would be C. The input to the promotion is B, so UNPROM
would describe B. */
static tree
vect_look_through_possible_promotion (vec_info *vinfo, tree op,
vect_unpromoted_value *unprom,
bool *single_use_p = NULL)
{
tree res = NULL_TREE;
tree op_type = TREE_TYPE (op);
unsigned int orig_precision = TYPE_PRECISION (op_type);
unsigned int min_precision = orig_precision;
stmt_vec_info caster = NULL;
while (TREE_CODE (op) == SSA_NAME && INTEGRAL_TYPE_P (op_type))
{
/* See whether OP is simple enough to vectorize. */
stmt_vec_info def_stmt_info;
gimple *def_stmt;
vect_def_type dt;
if (!vect_is_simple_use (op, vinfo, &dt, &def_stmt_info, &def_stmt))
break;
/* If OP is the input of a demotion, skip over it to see whether
OP is itself the result of a promotion. If so, the combined
effect of the promotion and the demotion might fit the required
pattern, otherwise neither operation fits.
This copes with cases such as the result of an arithmetic
operation being truncated before being stored, and where that
arithmetic operation has been recognized as an over-widened one. */
if (TYPE_PRECISION (op_type) <= min_precision)
{
/* Use OP as the UNPROM described above if we haven't yet
found a promotion, or if using the new input preserves the
sign of the previous promotion. */
if (!res
|| TYPE_PRECISION (unprom->type) == orig_precision
|| TYPE_SIGN (unprom->type) == TYPE_SIGN (op_type))
{
unprom->set_op (op, dt, caster);
min_precision = TYPE_PRECISION (op_type);
}
/* Stop if we've already seen a promotion and if this
conversion does more than change the sign. */
else if (TYPE_PRECISION (op_type)
!= TYPE_PRECISION (unprom->type))
break;
/* The sequence now extends to OP. */
res = op;
}
/* See whether OP is defined by a cast. Record it as CASTER if
the cast is potentially vectorizable. */
if (!def_stmt)
break;
caster = def_stmt_info;
/* Ignore pattern statements, since we don't link uses for them. */
if (caster
&& single_use_p
&& !STMT_VINFO_RELATED_STMT (caster)
&& !has_single_use (res))
*single_use_p = false;
gassign *assign = dyn_cast <gassign *> (def_stmt);
if (!assign || !CONVERT_EXPR_CODE_P (gimple_assign_rhs_code (def_stmt)))
break;
/* Continue with the input to the cast. */
op = gimple_assign_rhs1 (def_stmt);
op_type = TREE_TYPE (op);
}
return res;
}
/* OP is an integer operand to an operation that returns TYPE, and we
want to treat the operation as a widening one. So far we can treat
it as widening from *COMMON_TYPE.
Return true if OP is suitable for such a widening operation,
either widening from *COMMON_TYPE or from some supertype of it.
Update *COMMON_TYPE to the supertype in the latter case.
SHIFT_P is true if OP is a shift amount. */
static bool
vect_joust_widened_integer (tree type, bool shift_p, tree op,
tree *common_type)
{
/* Calculate the minimum precision required by OP, without changing
the sign of either operand. */
unsigned int precision;
if (shift_p)
{
if (!wi::leu_p (wi::to_widest (op), TYPE_PRECISION (type) / 2))
return false;
precision = TREE_INT_CST_LOW (op);
}
else
{
precision = wi::min_precision (wi::to_widest (op),
TYPE_SIGN (*common_type));
if (precision * 2 > TYPE_PRECISION (type))
return false;
}
/* If OP requires a wider type, switch to that type. The checks
above ensure that this is still narrower than the result. */
precision = vect_element_precision (precision);
if (TYPE_PRECISION (*common_type) < precision)
*common_type = build_nonstandard_integer_type
(precision, TYPE_UNSIGNED (*common_type));
return true;
}
/* Return true if the common supertype of NEW_TYPE and *COMMON_TYPE
is narrower than type, storing the supertype in *COMMON_TYPE if so. */
static bool
vect_joust_widened_type (tree type, tree new_type, tree *common_type)
{
if (types_compatible_p (*common_type, new_type))
return true;
/* See if *COMMON_TYPE can hold all values of NEW_TYPE. */
if ((TYPE_PRECISION (new_type) < TYPE_PRECISION (*common_type))
&& (TYPE_UNSIGNED (new_type) || !TYPE_UNSIGNED (*common_type)))
return true;
/* See if NEW_TYPE can hold all values of *COMMON_TYPE. */
if (TYPE_PRECISION (*common_type) < TYPE_PRECISION (new_type)
&& (TYPE_UNSIGNED (*common_type) || !TYPE_UNSIGNED (new_type)))
{
*common_type = new_type;
return true;
}
/* We have mismatched signs, with the signed type being
no wider than the unsigned type. In this case we need
a wider signed type. */
unsigned int precision = MAX (TYPE_PRECISION (*common_type),
TYPE_PRECISION (new_type));
precision *= 2;
if (precision * 2 > TYPE_PRECISION (type))
return false;
*common_type = build_nonstandard_integer_type (precision, false);
return true;
}
/* Check whether STMT_INFO can be viewed as a tree of integer operations
in which each node either performs CODE or WIDENED_CODE, and where
each leaf operand is narrower than the result of STMT_INFO. MAX_NOPS
specifies the maximum number of leaf operands. SHIFT_P says whether
CODE and WIDENED_CODE are some sort of shift.
If STMT_INFO is such a tree, return the number of leaf operands
and describe them in UNPROM[0] onwards. Also set *COMMON_TYPE
to a type that (a) is narrower than the result of STMT_INFO and
(b) can hold all leaf operand values.
Return 0 if STMT_INFO isn't such a tree, or if no such COMMON_TYPE
exists. */
static unsigned int
vect_widened_op_tree (stmt_vec_info stmt_info, tree_code code,
tree_code widened_code, bool shift_p,
unsigned int max_nops,
vect_unpromoted_value *unprom, tree *common_type)
{
/* Check for an integer operation with the right code. */
vec_info *vinfo = stmt_info->vinfo;
gassign *assign = dyn_cast <gassign *> (stmt_info->stmt);
if (!assign)
return 0;
tree_code rhs_code = gimple_assign_rhs_code (assign);
if (rhs_code != code && rhs_code != widened_code)
return 0;
tree type = gimple_expr_type (assign);
if (!INTEGRAL_TYPE_P (type))
return 0;
/* Assume that both operands will be leaf operands. */
max_nops -= 2;
/* Check the operands. */
unsigned int next_op = 0;
for (unsigned int i = 0; i < 2; ++i)
{
vect_unpromoted_value *this_unprom = &unprom[next_op];
unsigned int nops = 1;
tree op = gimple_op (assign, i + 1);
if (i == 1 && TREE_CODE (op) == INTEGER_CST)
{
/* We already have a common type from earlier operands.
Update it to account for OP. */
this_unprom->set_op (op, vect_constant_def);
if (!vect_joust_widened_integer (type, shift_p, op, common_type))
return 0;
}
else
{
/* Only allow shifts by constants. */
if (shift_p && i == 1)
return 0;
if (!vect_look_through_possible_promotion (stmt_info->vinfo, op,
this_unprom))
return 0;
if (TYPE_PRECISION (this_unprom->type) == TYPE_PRECISION (type))
{
/* The operand isn't widened. If STMT_INFO has the code
for an unwidened operation, recursively check whether
this operand is a node of the tree. */
if (rhs_code != code
|| max_nops == 0
|| this_unprom->dt != vect_internal_def)
return 0;
/* Give back the leaf slot allocated above now that we're
not treating this as a leaf operand. */
max_nops += 1;
/* Recursively process the definition of the operand. */
stmt_vec_info def_stmt_info
= vinfo->lookup_def (this_unprom->op);
nops = vect_widened_op_tree (def_stmt_info, code, widened_code,
shift_p, max_nops, this_unprom,
common_type);
if (nops == 0)
return 0;
max_nops -= nops;
}
else
{
/* Make sure that the operand is narrower than the result. */
if (TYPE_PRECISION (this_unprom->type) * 2
> TYPE_PRECISION (type))
return 0;
/* Update COMMON_TYPE for the new operand. */
if (i == 0)
*common_type = this_unprom->type;
else if (!vect_joust_widened_type (type, this_unprom->type,
common_type))
return 0;
}
}
next_op += nops;
}
return next_op;
}
/* Helper to return a new temporary for pattern of TYPE for STMT. If STMT
is NULL, the caller must set SSA_NAME_DEF_STMT for the returned SSA var. */
static tree
vect_recog_temp_ssa_var (tree type, gimple *stmt)
{
return make_temp_ssa_name (type, stmt, "patt");
}
/* STMT2_INFO describes a type conversion that could be split into STMT1
followed by a version of STMT2_INFO that takes NEW_RHS as its first
input. Try to do this using pattern statements, returning true on
success. */
static bool
vect_split_statement (stmt_vec_info stmt2_info, tree new_rhs,
gimple *stmt1, tree vectype)
{
vec_info *vinfo = stmt2_info->vinfo;
if (is_pattern_stmt_p (stmt2_info))
{
/* STMT2_INFO is part of a pattern. Get the statement to which
the pattern is attached. */
stmt_vec_info orig_stmt2_info = STMT_VINFO_RELATED_STMT (stmt2_info);
vect_init_pattern_stmt (stmt1, orig_stmt2_info, vectype);
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"Splitting pattern statement: %G", stmt2_info->stmt);
/* Since STMT2_INFO is a pattern statement, we can change it
in-situ without worrying about changing the code for the
containing block. */
gimple_assign_set_rhs1 (stmt2_info->stmt, new_rhs);
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location, "into: %G", stmt1);
dump_printf_loc (MSG_NOTE, vect_location, "and: %G",
stmt2_info->stmt);
}
gimple_seq *def_seq = &STMT_VINFO_PATTERN_DEF_SEQ (orig_stmt2_info);
if (STMT_VINFO_RELATED_STMT (orig_stmt2_info) == stmt2_info)
/* STMT2_INFO is the actual pattern statement. Add STMT1
to the end of the definition sequence. */
gimple_seq_add_stmt_without_update (def_seq, stmt1);
else
{
/* STMT2_INFO belongs to the definition sequence. Insert STMT1
before it. */
gimple_stmt_iterator gsi = gsi_for_stmt (stmt2_info->stmt, def_seq);
gsi_insert_before_without_update (&gsi, stmt1, GSI_SAME_STMT);
}
return true;
}
else
{
/* STMT2_INFO doesn't yet have a pattern. Try to create a
two-statement pattern now. */
gcc_assert (!STMT_VINFO_RELATED_STMT (stmt2_info));
tree lhs_type = TREE_TYPE (gimple_get_lhs (stmt2_info->stmt));
tree lhs_vectype = get_vectype_for_scalar_type (vinfo, lhs_type);
if (!lhs_vectype)
return false;
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"Splitting statement: %G", stmt2_info->stmt);
/* Add STMT1 as a singleton pattern definition sequence. */
gimple_seq *def_seq = &STMT_VINFO_PATTERN_DEF_SEQ (stmt2_info);
vect_init_pattern_stmt (stmt1, stmt2_info, vectype);
gimple_seq_add_stmt_without_update (def_seq, stmt1);
/* Build the second of the two pattern statements. */
tree new_lhs = vect_recog_temp_ssa_var (lhs_type, NULL);
gassign *new_stmt2 = gimple_build_assign (new_lhs, NOP_EXPR, new_rhs);
vect_set_pattern_stmt (new_stmt2, stmt2_info, lhs_vectype);
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"into pattern statements: %G", stmt1);
dump_printf_loc (MSG_NOTE, vect_location, "and: %G", new_stmt2);
}
return true;
}
}
/* Convert UNPROM to TYPE and return the result, adding new statements
to STMT_INFO's pattern definition statements if no better way is
available. VECTYPE is the vector form of TYPE. */
static tree
vect_convert_input (stmt_vec_info stmt_info, tree type,
vect_unpromoted_value *unprom, tree vectype)
{
vec_info *vinfo = stmt_info->vinfo;
/* Check for a no-op conversion. */
if (types_compatible_p (type, TREE_TYPE (unprom->op)))
return unprom->op;
/* Allow the caller to create constant vect_unpromoted_values. */
if (TREE_CODE (unprom->op) == INTEGER_CST)
return wide_int_to_tree (type, wi::to_widest (unprom->op));
tree input = unprom->op;
if (unprom->caster)
{
tree lhs = gimple_get_lhs (unprom->caster->stmt);
tree lhs_type = TREE_TYPE (lhs);
/* If the result of the existing cast is the right width, use it
instead of the source of the cast. */
if (TYPE_PRECISION (lhs_type) == TYPE_PRECISION (type))
input = lhs;
/* If the precision we want is between the source and result
precisions of the existing cast, try splitting the cast into
two and tapping into a mid-way point. */
else if (TYPE_PRECISION (lhs_type) > TYPE_PRECISION (type)
&& TYPE_PRECISION (type) > TYPE_PRECISION (unprom->type))
{
/* In order to preserve the semantics of the original cast,
give the mid-way point the same signedness as the input value.
It would be possible to use a signed type here instead if
TYPE is signed and UNPROM->TYPE is unsigned, but that would
make the sign of the midtype sensitive to the order in
which we process the statements, since the signedness of
TYPE is the signedness required by just one of possibly
many users. Also, unsigned promotions are usually as cheap
as or cheaper than signed ones, so it's better to keep an
unsigned promotion. */
tree midtype = build_nonstandard_integer_type
(TYPE_PRECISION (type), TYPE_UNSIGNED (unprom->type));
tree vec_midtype = get_vectype_for_scalar_type (vinfo, midtype);
if (vec_midtype)
{
input = vect_recog_temp_ssa_var (midtype, NULL);
gassign *new_stmt = gimple_build_assign (input, NOP_EXPR,
unprom->op);
if (!vect_split_statement (unprom->caster, input, new_stmt,
vec_midtype))
append_pattern_def_seq (stmt_info, new_stmt, vec_midtype);
}
}
/* See if we can reuse an existing result. */
if (types_compatible_p (type, TREE_TYPE (input)))
return input;
}
/* We need a new conversion statement. */
tree new_op = vect_recog_temp_ssa_var (type, NULL);
gassign *new_stmt = gimple_build_assign (new_op, NOP_EXPR, input);
/* If OP is an external value, see if we can insert the new statement
on an incoming edge. */
if (input == unprom->op && unprom->dt == vect_external_def)
if (edge e = vect_get_external_def_edge (stmt_info->vinfo, input))
{
basic_block new_bb = gsi_insert_on_edge_immediate (e, new_stmt);
gcc_assert (!new_bb);
return new_op;
}
/* As a (common) last resort, add the statement to the pattern itself. */
append_pattern_def_seq (stmt_info, new_stmt, vectype);
return new_op;
}
/* Invoke vect_convert_input for N elements of UNPROM and store the
result in the corresponding elements of RESULT. */
static void
vect_convert_inputs (stmt_vec_info stmt_info, unsigned int n,
tree *result, tree type, vect_unpromoted_value *unprom,
tree vectype)
{
for (unsigned int i = 0; i < n; ++i)
{
unsigned int j;
for (j = 0; j < i; ++j)
if (unprom[j].op == unprom[i].op)
break;
if (j < i)
result[i] = result[j];
else
result[i] = vect_convert_input (stmt_info, type, &unprom[i], vectype);
}
}
/* The caller has created a (possibly empty) sequence of pattern definition
statements followed by a single statement PATTERN_STMT. Cast the result
of this final statement to TYPE. If a new statement is needed, add
PATTERN_STMT to the end of STMT_INFO's pattern definition statements
and return the new statement, otherwise return PATTERN_STMT as-is.
VECITYPE is the vector form of PATTERN_STMT's result type. */
static gimple *
vect_convert_output (stmt_vec_info stmt_info, tree type, gimple *pattern_stmt,
tree vecitype)
{
tree lhs = gimple_get_lhs (pattern_stmt);
if (!types_compatible_p (type, TREE_TYPE (lhs)))
{
append_pattern_def_seq (stmt_info, pattern_stmt, vecitype);
tree cast_var = vect_recog_temp_ssa_var (type, NULL);
pattern_stmt = gimple_build_assign (cast_var, NOP_EXPR, lhs);
}
return pattern_stmt;
}
/* Return true if STMT_VINFO describes a reduction for which reassociation
is allowed. If STMT_INFO is part of a group, assume that it's part of
a reduction chain and optimistically assume that all statements
except the last allow reassociation.
Also require it to have code CODE and to be a reduction
in the outermost loop. When returning true, store the operands in
*OP0_OUT and *OP1_OUT. */
static bool
vect_reassociating_reduction_p (stmt_vec_info stmt_info, tree_code code,
tree *op0_out, tree *op1_out)
{
loop_vec_info loop_info = STMT_VINFO_LOOP_VINFO (stmt_info);
if (!loop_info)
return false;
gassign *assign = dyn_cast <gassign *> (stmt_info->stmt);
if (!assign || gimple_assign_rhs_code (assign) != code)
return false;
/* We don't allow changing the order of the computation in the inner-loop
when doing outer-loop vectorization. */
class loop *loop = LOOP_VINFO_LOOP (loop_info);
if (loop && nested_in_vect_loop_p (loop, stmt_info))
return false;
if (STMT_VINFO_DEF_TYPE (stmt_info) == vect_reduction_def)
{
if (needs_fold_left_reduction_p (TREE_TYPE (gimple_assign_lhs (assign)),
code))
return false;
}
else if (REDUC_GROUP_FIRST_ELEMENT (stmt_info) == NULL)
return false;
*op0_out = gimple_assign_rhs1 (assign);
*op1_out = gimple_assign_rhs2 (assign);
if (commutative_tree_code (code) && STMT_VINFO_REDUC_IDX (stmt_info) == 0)
std::swap (*op0_out, *op1_out);
return true;
}
/* Function vect_recog_dot_prod_pattern
Try to find the following pattern:
type x_t, y_t;
TYPE1 prod;
TYPE2 sum = init;
loop:
sum_0 = phi <init, sum_1>
S1 x_t = ...
S2 y_t = ...
S3 x_T = (TYPE1) x_t;
S4 y_T = (TYPE1) y_t;
S5 prod = x_T * y_T;
[S6 prod = (TYPE2) prod; #optional]
S7 sum_1 = prod + sum_0;
where 'TYPE1' is exactly double the size of type 'type', and 'TYPE2' is the
same size of 'TYPE1' or bigger. This is a special case of a reduction
computation.
Input:
* STMT_VINFO: The stmt from which the pattern search begins. In the
example, when this function is called with S7, the pattern {S3,S4,S5,S6,S7}
will be detected.
Output:
* TYPE_OUT: The type of the output of this pattern.
* Return value: A new stmt that will be used to replace the sequence of
stmts that constitute the pattern. In this case it will be:
WIDEN_DOT_PRODUCT <x_t, y_t, sum_0>
Note: The dot-prod idiom is a widening reduction pattern that is
vectorized without preserving all the intermediate results. It
produces only N/2 (widened) results (by summing up pairs of
intermediate results) rather than all N results. Therefore, we
cannot allow this pattern when we want to get all the results and in
the correct order (as is the case when this computation is in an
inner-loop nested in an outer-loop that us being vectorized). */
static gimple *
vect_recog_dot_prod_pattern (stmt_vec_info stmt_vinfo, tree *type_out)
{
tree oprnd0, oprnd1;
gimple *last_stmt = stmt_vinfo->stmt;
vec_info *vinfo = stmt_vinfo->vinfo;
tree type, half_type;
gimple *pattern_stmt;
tree var;
/* Look for the following pattern
DX = (TYPE1) X;
DY = (TYPE1) Y;
DPROD = DX * DY;
DDPROD = (TYPE2) DPROD;
sum_1 = DDPROD + sum_0;
In which
- DX is double the size of X
- DY is double the size of Y
- DX, DY, DPROD all have the same type
- sum is the same size of DPROD or bigger
- sum has been recognized as a reduction variable.
This is equivalent to:
DPROD = X w* Y; #widen mult
sum_1 = DPROD w+ sum_0; #widen summation
or
DPROD = X w* Y; #widen mult
sum_1 = DPROD + sum_0; #summation
*/
/* Starting from LAST_STMT, follow the defs of its uses in search
of the above pattern. */
if (!vect_reassociating_reduction_p (stmt_vinfo, PLUS_EXPR,
&oprnd0, &oprnd1))
return NULL;
type = gimple_expr_type (last_stmt);
vect_unpromoted_value unprom_mult;
oprnd0 = vect_look_through_possible_promotion (vinfo, oprnd0, &unprom_mult);
/* So far so good. Since last_stmt was detected as a (summation) reduction,
we know that oprnd1 is the reduction variable (defined by a loop-header
phi), and oprnd0 is an ssa-name defined by a stmt in the loop body.
Left to check that oprnd0 is defined by a (widen_)mult_expr */
if (!oprnd0)
return NULL;
stmt_vec_info mult_vinfo = vect_get_internal_def (vinfo, oprnd0);
if (!mult_vinfo)
return NULL;
/* FORNOW. Can continue analyzing the def-use chain when this stmt in a phi
inside the loop (in case we are analyzing an outer-loop). */
vect_unpromoted_value unprom0[2];
if (!vect_widened_op_tree (mult_vinfo, MULT_EXPR, WIDEN_MULT_EXPR,
false, 2, unprom0, &half_type))
return NULL;
/* If there are two widening operations, make sure they agree on
the sign of the extension. */
if (TYPE_PRECISION (unprom_mult.type) != TYPE_PRECISION (type)
&& TYPE_SIGN (unprom_mult.type) != TYPE_SIGN (half_type))
return NULL;
vect_pattern_detected ("vect_recog_dot_prod_pattern", last_stmt);
tree half_vectype;
if (!vect_supportable_direct_optab_p (vinfo, type, DOT_PROD_EXPR, half_type,
type_out, &half_vectype))
return NULL;
/* Get the inputs in the appropriate types. */
tree mult_oprnd[2];
vect_convert_inputs (stmt_vinfo, 2, mult_oprnd, half_type,
unprom0, half_vectype);
var = vect_recog_temp_ssa_var (type, NULL);
pattern_stmt = gimple_build_assign (var, DOT_PROD_EXPR,
mult_oprnd[0], mult_oprnd[1], oprnd1);
return pattern_stmt;
}
/* Function vect_recog_sad_pattern
Try to find the following Sum of Absolute Difference (SAD) pattern:
type x_t, y_t;
signed TYPE1 diff, abs_diff;
TYPE2 sum = init;
loop:
sum_0 = phi <init, sum_1>
S1 x_t = ...
S2 y_t = ...
S3 x_T = (TYPE1) x_t;
S4 y_T = (TYPE1) y_t;
S5 diff = x_T - y_T;
S6 abs_diff = ABS_EXPR <diff>;
[S7 abs_diff = (TYPE2) abs_diff; #optional]
S8 sum_1 = abs_diff + sum_0;
where 'TYPE1' is at least double the size of type 'type', and 'TYPE2' is the
same size of 'TYPE1' or bigger. This is a special case of a reduction
computation.
Input:
* STMT_VINFO: The stmt from which the pattern search begins. In the
example, when this function is called with S8, the pattern
{S3,S4,S5,S6,S7,S8} will be detected.
Output:
* TYPE_OUT: The type of the output of this pattern.
* Return value: A new stmt that will be used to replace the sequence of
stmts that constitute the pattern. In this case it will be:
SAD_EXPR <x_t, y_t, sum_0>
*/
static gimple *
vect_recog_sad_pattern (stmt_vec_info stmt_vinfo, tree *type_out)
{
gimple *last_stmt = stmt_vinfo->stmt;
vec_info *vinfo = stmt_vinfo->vinfo;
tree half_type;
/* Look for the following pattern
DX = (TYPE1) X;
DY = (TYPE1) Y;
DDIFF = DX - DY;
DAD = ABS_EXPR <DDIFF>;
DDPROD = (TYPE2) DPROD;
sum_1 = DAD + sum_0;
In which
- DX is at least double the size of X
- DY is at least double the size of Y
- DX, DY, DDIFF, DAD all have the same type
- sum is the same size of DAD or bigger
- sum has been recognized as a reduction variable.
This is equivalent to:
DDIFF = X w- Y; #widen sub
DAD = ABS_EXPR <DDIFF>;
sum_1 = DAD w+ sum_0; #widen summation
or
DDIFF = X w- Y; #widen sub
DAD = ABS_EXPR <DDIFF>;
sum_1 = DAD + sum_0; #summation
*/
/* Starting from LAST_STMT, follow the defs of its uses in search
of the above pattern. */
tree plus_oprnd0, plus_oprnd1;
if (!vect_reassociating_reduction_p (stmt_vinfo, PLUS_EXPR,
&plus_oprnd0, &plus_oprnd1))
return NULL;
tree sum_type = gimple_expr_type (last_stmt);
/* Any non-truncating sequence of conversions is OK here, since
with a successful match, the result of the ABS(U) is known to fit
within the nonnegative range of the result type. (It cannot be the
negative of the minimum signed value due to the range of the widening
MINUS_EXPR.) */
vect_unpromoted_value unprom_abs;
plus_oprnd0 = vect_look_through_possible_promotion (vinfo, plus_oprnd0,
&unprom_abs);
/* So far so good. Since last_stmt was detected as a (summation) reduction,
we know that plus_oprnd1 is the reduction variable (defined by a loop-header
phi), and plus_oprnd0 is an ssa-name defined by a stmt in the loop body.
Then check that plus_oprnd0 is defined by an abs_expr. */
if (!plus_oprnd0)
return NULL;
stmt_vec_info abs_stmt_vinfo = vect_get_internal_def (vinfo, plus_oprnd0);
if (!abs_stmt_vinfo)
return NULL;
/* FORNOW. Can continue analyzing the def-use chain when this stmt in a phi
inside the loop (in case we are analyzing an outer-loop). */
gassign *abs_stmt = dyn_cast <gassign *> (abs_stmt_vinfo->stmt);
if (!abs_stmt
|| (gimple_assign_rhs_code (abs_stmt) != ABS_EXPR
&& gimple_assign_rhs_code (abs_stmt) != ABSU_EXPR))
return NULL;
tree abs_oprnd = gimple_assign_rhs1 (abs_stmt);
tree abs_type = TREE_TYPE (abs_oprnd);
if (TYPE_UNSIGNED (abs_type))
return NULL;
/* Peel off conversions from the ABS input. This can involve sign
changes (e.g. from an unsigned subtraction to a signed ABS input)
or signed promotion, but it can't include unsigned promotion.
(Note that ABS of an unsigned promotion should have been folded
away before now anyway.) */
vect_unpromoted_value unprom_diff;
abs_oprnd = vect_look_through_possible_promotion (vinfo, abs_oprnd,
&unprom_diff);
if (!abs_oprnd)
return NULL;
if (TYPE_PRECISION (unprom_diff.type) != TYPE_PRECISION (abs_type)
&& TYPE_UNSIGNED (unprom_diff.type))
return NULL;
/* We then detect if the operand of abs_expr is defined by a minus_expr. */
stmt_vec_info diff_stmt_vinfo = vect_get_internal_def (vinfo, abs_oprnd);
if (!diff_stmt_vinfo)
return NULL;
/* FORNOW. Can continue analyzing the def-use chain when this stmt in a phi
inside the loop (in case we are analyzing an outer-loop). */
vect_unpromoted_value unprom[2];
if (!vect_widened_op_tree (diff_stmt_vinfo, MINUS_EXPR, MINUS_EXPR,
false, 2, unprom, &half_type))
return NULL;
vect_pattern_detected ("vect_recog_sad_pattern", last_stmt);
tree half_vectype;
if (!vect_supportable_direct_optab_p (vinfo, sum_type, SAD_EXPR, half_type,
type_out, &half_vectype))
return NULL;
/* Get the inputs to the SAD_EXPR in the appropriate types. */
tree sad_oprnd[2];
vect_convert_inputs (stmt_vinfo, 2, sad_oprnd, half_type,
unprom, half_vectype);
tree var = vect_recog_temp_ssa_var (sum_type, NULL);
gimple *pattern_stmt = gimple_build_assign (var, SAD_EXPR, sad_oprnd[0],
sad_oprnd[1], plus_oprnd1);
return pattern_stmt;
}
/* Recognize an operation that performs ORIG_CODE on widened inputs,
so that it can be treated as though it had the form:
A_TYPE a;
B_TYPE b;
HALF_TYPE a_cast = (HALF_TYPE) a; // possible no-op
HALF_TYPE b_cast = (HALF_TYPE) b; // possible no-op
| RES_TYPE a_extend = (RES_TYPE) a_cast; // promotion from HALF_TYPE
| RES_TYPE b_extend = (RES_TYPE) b_cast; // promotion from HALF_TYPE
| RES_TYPE res = a_extend ORIG_CODE b_extend;
Try to replace the pattern with:
A_TYPE a;
B_TYPE b;
HALF_TYPE a_cast = (HALF_TYPE) a; // possible no-op
HALF_TYPE b_cast = (HALF_TYPE) b; // possible no-op
| EXT_TYPE ext = a_cast WIDE_CODE b_cast;
| RES_TYPE res = (EXT_TYPE) ext; // possible no-op
where EXT_TYPE is wider than HALF_TYPE but has the same signedness.
SHIFT_P is true if ORIG_CODE and WIDE_CODE are shifts. NAME is the
name of the pattern being matched, for dump purposes. */
static gimple *
vect_recog_widen_op_pattern (stmt_vec_info last_stmt_info, tree *type_out,
tree_code orig_code, tree_code wide_code,
bool shift_p, const char *name)
{
vec_info *vinfo = last_stmt_info->vinfo;
gimple *last_stmt = last_stmt_info->stmt;
vect_unpromoted_value unprom[2];
tree half_type;
if (!vect_widened_op_tree (last_stmt_info, orig_code, orig_code,
shift_p, 2, unprom, &half_type))
return NULL;
/* Pattern detected. */
vect_pattern_detected (name, last_stmt);
tree type = gimple_expr_type (last_stmt);
tree itype = type;
if (TYPE_PRECISION (type) != TYPE_PRECISION (half_type) * 2
|| TYPE_UNSIGNED (type) != TYPE_UNSIGNED (half_type))
itype = build_nonstandard_integer_type (TYPE_PRECISION (half_type) * 2,
TYPE_UNSIGNED (half_type));
/* Check target support */
tree vectype = get_vectype_for_scalar_type (vinfo, half_type);
tree vecitype = get_vectype_for_scalar_type (vinfo, itype);
enum tree_code dummy_code;
int dummy_int;
auto_vec<tree> dummy_vec;
if (!vectype
|| !vecitype
|| !supportable_widening_operation (wide_code, last_stmt_info,
vecitype, vectype,
&dummy_code, &dummy_code,
&dummy_int, &dummy_vec))
return NULL;
*type_out = get_vectype_for_scalar_type (vinfo, type);
if (!*type_out)
return NULL;
tree oprnd[2];
vect_convert_inputs (last_stmt_info, 2, oprnd, half_type, unprom, vectype);
tree var = vect_recog_temp_ssa_var (itype, NULL);
gimple *pattern_stmt = gimple_build_assign (var, wide_code,
oprnd[0], oprnd[1]);
return vect_convert_output (last_stmt_info, type, pattern_stmt, vecitype);
}
/* Try to detect multiplication on widened inputs, converting MULT_EXPR
to WIDEN_MULT_EXPR. See vect_recog_widen_op_pattern for details. */
static gimple *
vect_recog_widen_mult_pattern (stmt_vec_info last_stmt_info, tree *type_out)
{
return vect_recog_widen_op_pattern (last_stmt_info, type_out, MULT_EXPR,
WIDEN_MULT_EXPR, false,
"vect_recog_widen_mult_pattern");
}
/* Function vect_recog_pow_pattern
Try to find the following pattern:
x = POW (y, N);
with POW being one of pow, powf, powi, powif and N being
either 2 or 0.5.
Input:
* STMT_VINFO: The stmt from which the pattern search begins.
Output:
* TYPE_OUT: The type of the output of this pattern.
* Return value: A new stmt that will be used to replace the sequence of
stmts that constitute the pattern. In this case it will be:
x = x * x
or
x = sqrt (x)
*/
static gimple *
vect_recog_pow_pattern (stmt_vec_info stmt_vinfo, tree *type_out)
{
vec_info *vinfo = stmt_vinfo->vinfo;
gimple *last_stmt = stmt_vinfo->stmt;
tree base, exp;
gimple *stmt;
tree var;
if (!is_gimple_call (last_stmt) || gimple_call_lhs (last_stmt) == NULL)
return NULL;
switch (gimple_call_combined_fn (last_stmt))
{
CASE_CFN_POW:
CASE_CFN_POWI:
break;
default:
return NULL;
}
base = gimple_call_arg (last_stmt, 0);
exp = gimple_call_arg (last_stmt, 1);
if (TREE_CODE (exp) != REAL_CST
&& TREE_CODE (exp) != INTEGER_CST)
{
if (flag_unsafe_math_optimizations
&& TREE_CODE (base) == REAL_CST
&& gimple_call_builtin_p (last_stmt, BUILT_IN_NORMAL))
{
combined_fn log_cfn;
built_in_function exp_bfn;
switch (DECL_FUNCTION_CODE (gimple_call_fndecl (last_stmt)))
{
case BUILT_IN_POW:
log_cfn = CFN_BUILT_IN_LOG;
exp_bfn = BUILT_IN_EXP;
break;
case BUILT_IN_POWF:
log_cfn = CFN_BUILT_IN_LOGF;
exp_bfn = BUILT_IN_EXPF;
break;
case BUILT_IN_POWL:
log_cfn = CFN_BUILT_IN_LOGL;
exp_bfn = BUILT_IN_EXPL;
break;
default:
return NULL;
}
tree logc = fold_const_call (log_cfn, TREE_TYPE (base), base);
tree exp_decl = builtin_decl_implicit (exp_bfn);
/* Optimize pow (C, x) as exp (log (C) * x). Normally match.pd
does that, but if C is a power of 2, we want to use
exp2 (log2 (C) * x) in the non-vectorized version, but for
vectorization we don't have vectorized exp2. */
if (logc
&& TREE_CODE (logc) == REAL_CST
&& exp_decl
&& lookup_attribute ("omp declare simd",
DECL_ATTRIBUTES (exp_decl)))
{
cgraph_node *node = cgraph_node::get_create (exp_decl);
if (node->simd_clones == NULL)
{
if (targetm.simd_clone.compute_vecsize_and_simdlen == NULL
|| node->definition)
return NULL;
expand_simd_clones (node);
if (node->simd_clones == NULL)
return NULL;
}
*type_out = get_vectype_for_scalar_type (vinfo, TREE_TYPE (base));
if (!*type_out)
return NULL;
tree def = vect_recog_temp_ssa_var (TREE_TYPE (base), NULL);
gimple *g = gimple_build_assign (def, MULT_EXPR, exp, logc);
append_pattern_def_seq (stmt_vinfo, g);
tree res = vect_recog_temp_ssa_var (TREE_TYPE (base), NULL);
g = gimple_build_call (exp_decl, 1, def);
gimple_call_set_lhs (g, res);
return g;
}
}
return NULL;
}
/* We now have a pow or powi builtin function call with a constant
exponent. */
/* Catch squaring. */
if ((tree_fits_shwi_p (exp)
&& tree_to_shwi (exp) == 2)
|| (TREE_CODE (exp) == REAL_CST
&& real_equal (&TREE_REAL_CST (exp), &dconst2)))
{
if (!vect_supportable_direct_optab_p (vinfo, TREE_TYPE (base), MULT_EXPR,
TREE_TYPE (base), type_out))
return NULL;
var = vect_recog_temp_ssa_var (TREE_TYPE (base), NULL);
stmt = gimple_build_assign (var, MULT_EXPR, base, base);
return stmt;
}
/* Catch square root. */
if (TREE_CODE (exp) == REAL_CST
&& real_equal (&TREE_REAL_CST (exp), &dconsthalf))
{
*type_out = get_vectype_for_scalar_type (vinfo, TREE_TYPE (base));
if (*type_out
&& direct_internal_fn_supported_p (IFN_SQRT, *type_out,
OPTIMIZE_FOR_SPEED))
{
gcall *stmt = gimple_build_call_internal (IFN_SQRT, 1, base);
var = vect_recog_temp_ssa_var (TREE_TYPE (base), stmt);
gimple_call_set_lhs (stmt, var);
gimple_call_set_nothrow (stmt, true);
return stmt;
}
}
return NULL;
}
/* Function vect_recog_widen_sum_pattern
Try to find the following pattern:
type x_t;
TYPE x_T, sum = init;
loop:
sum_0 = phi <init, sum_1>
S1 x_t = *p;
S2 x_T = (TYPE) x_t;
S3 sum_1 = x_T + sum_0;
where type 'TYPE' is at least double the size of type 'type', i.e - we're
summing elements of type 'type' into an accumulator of type 'TYPE'. This is
a special case of a reduction computation.
Input:
* STMT_VINFO: The stmt from which the pattern search begins. In the example,
when this function is called with S3, the pattern {S2,S3} will be detected.
Output:
* TYPE_OUT: The type of the output of this pattern.
* Return value: A new stmt that will be used to replace the sequence of
stmts that constitute the pattern. In this case it will be:
WIDEN_SUM <x_t, sum_0>
Note: The widening-sum idiom is a widening reduction pattern that is
vectorized without preserving all the intermediate results. It
produces only N/2 (widened) results (by summing up pairs of
intermediate results) rather than all N results. Therefore, we
cannot allow this pattern when we want to get all the results and in
the correct order (as is the case when this computation is in an
inner-loop nested in an outer-loop that us being vectorized). */
static gimple *
vect_recog_widen_sum_pattern (stmt_vec_info stmt_vinfo, tree *type_out)
{
gimple *last_stmt = stmt_vinfo->stmt;
tree oprnd0, oprnd1;
vec_info *vinfo = stmt_vinfo->vinfo;
tree type;
gimple *pattern_stmt;
tree var;
/* Look for the following pattern
DX = (TYPE) X;
sum_1 = DX + sum_0;
In which DX is at least double the size of X, and sum_1 has been
recognized as a reduction variable.
*/
/* Starting from LAST_STMT, follow the defs of its uses in search
of the above pattern. */
if (!vect_reassociating_reduction_p (stmt_vinfo, PLUS_EXPR,
&oprnd0, &oprnd1))
return NULL;
type = gimple_expr_type (last_stmt);
/* So far so good. Since last_stmt was detected as a (summation) reduction,
we know that oprnd1 is the reduction variable (defined by a loop-header
phi), and oprnd0 is an ssa-name defined by a stmt in the loop body.
Left to check that oprnd0 is defined by a cast from type 'type' to type
'TYPE'. */
vect_unpromoted_value unprom0;
if (!vect_look_through_possible_promotion (vinfo, oprnd0, &unprom0)
|| TYPE_PRECISION (unprom0.type) * 2 > TYPE_PRECISION (type))
return NULL;
vect_pattern_detected ("vect_recog_widen_sum_pattern", last_stmt);
if (!vect_supportable_direct_optab_p (vinfo, type, WIDEN_SUM_EXPR,
unprom0.type, type_out))
return NULL;
var = vect_recog_temp_ssa_var (type, NULL);
pattern_stmt = gimple_build_assign (var, WIDEN_SUM_EXPR, unprom0.op, oprnd1);
return pattern_stmt;
}
/* Recognize cases in which an operation is performed in one type WTYPE
but could be done more efficiently in a narrower type NTYPE. For example,
if we have:
ATYPE a; // narrower than NTYPE
BTYPE b; // narrower than NTYPE
WTYPE aw = (WTYPE) a;
WTYPE bw = (WTYPE) b;
WTYPE res = aw + bw; // only uses of aw and bw
then it would be more efficient to do:
NTYPE an = (NTYPE) a;
NTYPE bn = (NTYPE) b;
NTYPE resn = an + bn;
WTYPE res = (WTYPE) resn;
Other situations include things like:
ATYPE a; // NTYPE or narrower
WTYPE aw = (WTYPE) a;
WTYPE res = aw + b;
when only "(NTYPE) res" is significant. In that case it's more efficient
to truncate "b" and do the operation on NTYPE instead:
NTYPE an = (NTYPE) a;
NTYPE bn = (NTYPE) b; // truncation
NTYPE resn = an + bn;
WTYPE res = (WTYPE) resn;
All users of "res" should then use "resn" instead, making the final
statement dead (not marked as relevant). The final statement is still
needed to maintain the type correctness of the IR.
vect_determine_precisions has already determined the minimum
precison of the operation and the minimum precision required
by users of the result. */
static gimple *
vect_recog_over_widening_pattern (stmt_vec_info last_stmt_info, tree *type_out)
{
gassign *last_stmt = dyn_cast <gassign *> (last_stmt_info->stmt);
if (!last_stmt)
return NULL;
/* See whether we have found that this operation can be done on a
narrower type without changing its semantics. */
unsigned int new_precision = last_stmt_info->operation_precision;
if (!new_precision)
return NULL;
vec_info *vinfo = last_stmt_info->vinfo;
tree lhs = gimple_assign_lhs (last_stmt);
tree type = TREE_TYPE (lhs);
tree_code code = gimple_assign_rhs_code (last_stmt);
/* Keep the first operand of a COND_EXPR as-is: only the other two
operands are interesting. */
unsigned int first_op = (code == COND_EXPR ? 2 : 1);
/* Check the operands. */
unsigned int nops = gimple_num_ops (last_stmt) - first_op;
auto_vec <vect_unpromoted_value, 3> unprom (nops);
unprom.quick_grow (nops);
unsigned int min_precision = 0;
bool single_use_p = false;
for (unsigned int i = 0; i < nops; ++i)
{
tree op = gimple_op (last_stmt, first_op + i);
if (TREE_CODE (op) == INTEGER_CST)
unprom[i].set_op (op, vect_constant_def);
else if (TREE_CODE (op) == SSA_NAME)
{
bool op_single_use_p = true;
if (!vect_look_through_possible_promotion (vinfo, op, &unprom[i],
&op_single_use_p))
return NULL;
/* If:
(1) N bits of the result are needed;
(2) all inputs are widened from M<N bits; and
(3) one operand OP is a single-use SSA name
we can shift the M->N widening from OP to the output
without changing the number or type of extensions involved.
This then reduces the number of copies of STMT_INFO.
If instead of (3) more than one operand is a single-use SSA name,
shifting the extension to the output is even more of a win.
If instead:
(1) N bits of the result are needed;
(2) one operand OP2 is widened from M2<N bits;
(3) another operand OP1 is widened from M1<M2 bits; and
(4) both OP1 and OP2 are single-use
the choice is between:
(a) truncating OP2 to M1, doing the operation on M1,
and then widening the result to N
(b) widening OP1 to M2, doing the operation on M2, and then
widening the result to N
Both shift the M2->N widening of the inputs to the output.
(a) additionally shifts the M1->M2 widening to the output;
it requires fewer copies of STMT_INFO but requires an extra
M2->M1 truncation.
Which is better will depend on the complexity and cost of
STMT_INFO, which is hard to predict at this stage. However,
a clear tie-breaker in favor of (b) is the fact that the
truncation in (a) increases the length of the operation chain.
If instead of (4) only one of OP1 or OP2 is single-use,
(b) is still a win over doing the operation in N bits:
it still shifts the M2->N widening on the single-use operand
to the output and reduces the number of STMT_INFO copies.
If neither operand is single-use then operating on fewer than
N bits might lead to more extensions overall. Whether it does
or not depends on global information about the vectorization
region, and whether that's a good trade-off would again
depend on the complexity and cost of the statements involved,
as well as things like register pressure that are not normally
modelled at this stage. We therefore ignore these cases
and just optimize the clear single-use wins above.
Thus we take the maximum precision of the unpromoted operands
and record whether any operand is single-use. */
if (unprom[i].dt == vect_internal_def)
{
min_precision = MAX (min_precision,
TYPE_PRECISION (unprom[i].type));
single_use_p |= op_single_use_p;
}
}
}
/* Although the operation could be done in operation_precision, we have
to balance that against introducing extra truncations or extensions.
Calculate the minimum precision that can be handled efficiently.
The loop above determined that the operation could be handled
efficiently in MIN_PRECISION if SINGLE_USE_P; this would shift an
extension from the inputs to the output without introducing more
instructions, and would reduce the number of instructions required
for STMT_INFO itself.
vect_determine_precisions has also determined that the result only
needs min_output_precision bits. Truncating by a factor of N times
requires a tree of N - 1 instructions, so if TYPE is N times wider
than min_output_precision, doing the operation in TYPE and truncating
the result requires N + (N - 1) = 2N - 1 instructions per output vector.
In contrast:
- truncating the input to a unary operation and doing the operation
in the new type requires at most N - 1 + 1 = N instructions per
output vector
- doing the same for a binary operation requires at most
(N - 1) * 2 + 1 = 2N - 1 instructions per output vector
Both unary and binary operations require fewer instructions than
this if the operands were extended from a suitable truncated form.
Thus there is usually nothing to lose by doing operations in
min_output_precision bits, but there can be something to gain. */
if (!single_use_p)
min_precision = last_stmt_info->min_output_precision;
else
min_precision = MIN (min_precision, last_stmt_info->min_output_precision);
/* Apply the minimum efficient precision we just calculated. */
if (new_precision < min_precision)
new_precision = min_precision;
if (new_precision >= TYPE_PRECISION (type))
return NULL;
vect_pattern_detected ("vect_recog_over_widening_pattern", last_stmt);
*type_out = get_vectype_for_scalar_type (vinfo, type);
if (!*type_out)
return NULL;
/* We've found a viable pattern. Get the new type of the operation. */
bool unsigned_p = (last_stmt_info->operation_sign == UNSIGNED);
tree new_type = build_nonstandard_integer_type (new_precision, unsigned_p);
/* If we're truncating an operation, we need to make sure that we
don't introduce new undefined overflow. The codes tested here are
a subset of those accepted by vect_truncatable_operation_p. */
tree op_type = new_type;
if (TYPE_OVERFLOW_UNDEFINED (new_type)
&& (code == PLUS_EXPR || code == MINUS_EXPR || code == MULT_EXPR))
op_type = build_nonstandard_integer_type (new_precision, true);
/* We specifically don't check here whether the target supports the
new operation, since it might be something that a later pattern
wants to rewrite anyway. If targets have a minimum element size
for some optabs, we should pattern-match smaller ops to larger ops
where beneficial. */
tree new_vectype = get_vectype_for_scalar_type (vinfo, new_type);
tree op_vectype = get_vectype_for_scalar_type (vinfo, op_type);
if (!new_vectype || !op_vectype)
return NULL;
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location, "demoting %T to %T\n",
type, new_type);
/* Calculate the rhs operands for an operation on OP_TYPE. */
tree ops[3] = {};
for (unsigned int i = 1; i < first_op; ++i)
ops[i - 1] = gimple_op (last_stmt, i);
vect_convert_inputs (last_stmt_info, nops, &ops[first_op - 1],
op_type, &unprom[0], op_vectype);
/* Use the operation to produce a result of type OP_TYPE. */
tree new_var = vect_recog_temp_ssa_var (op_type, NULL);
gimple *pattern_stmt = gimple_build_assign (new_var, code,
ops[0], ops[1], ops[2]);
gimple_set_location (pattern_stmt, gimple_location (last_stmt));
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"created pattern stmt: %G", pattern_stmt);
/* Convert back to the original signedness, if OP_TYPE is different
from NEW_TYPE. */
if (op_type != new_type)
pattern_stmt = vect_convert_output (last_stmt_info, new_type,
pattern_stmt, op_vectype);
/* Promote the result to the original type. */
pattern_stmt = vect_convert_output (last_stmt_info, type,
pattern_stmt, new_vectype);
return pattern_stmt;
}
/* Recognize the following patterns:
ATYPE a; // narrower than TYPE
BTYPE b; // narrower than TYPE
1) Multiply high with scaling
TYPE res = ((TYPE) a * (TYPE) b) >> c;
2) ... or also with rounding
TYPE res = (((TYPE) a * (TYPE) b) >> d + 1) >> 1;
where only the bottom half of res is used. */
static gimple *
vect_recog_mulhs_pattern (stmt_vec_info last_stmt_info, tree *type_out)
{
/* Check for a right shift. */
gassign *last_stmt = dyn_cast <gassign *> (last_stmt_info->stmt);
if (!last_stmt
|| gimple_assign_rhs_code (last_stmt) != RSHIFT_EXPR)
return NULL;
vec_info *vinfo = last_stmt_info->vinfo;
/* Check that the shift result is wider than the users of the
result need (i.e. that narrowing would be a natural choice). */
tree lhs_type = TREE_TYPE (gimple_assign_lhs (last_stmt));
unsigned int target_precision
= vect_element_precision (last_stmt_info->min_output_precision);
if (!INTEGRAL_TYPE_P (lhs_type)
|| target_precision >= TYPE_PRECISION (lhs_type))
return NULL;
/* Look through any change in sign on the outer shift input. */
vect_unpromoted_value unprom_rshift_input;
tree rshift_input = vect_look_through_possible_promotion
(vinfo, gimple_assign_rhs1 (last_stmt), &unprom_rshift_input);
if (!rshift_input
|| TYPE_PRECISION (TREE_TYPE (rshift_input))
!= TYPE_PRECISION (lhs_type))
return NULL;
/* Get the definition of the shift input. */
stmt_vec_info rshift_input_stmt_info
= vect_get_internal_def (vinfo, rshift_input);
if (!rshift_input_stmt_info)
return NULL;
gassign *rshift_input_stmt
= dyn_cast <gassign *> (rshift_input_stmt_info->stmt);
if (!rshift_input_stmt)
return NULL;
stmt_vec_info mulh_stmt_info;
tree scale_term;
internal_fn ifn;
unsigned int expect_offset;
/* Check for the presence of the rounding term. */
if (gimple_assign_rhs_code (rshift_input_stmt) == PLUS_EXPR)
{
/* Check that the outer shift was by 1. */
if (!integer_onep (gimple_assign_rhs2 (last_stmt)))
return NULL;
/* Check that the second operand of the PLUS_EXPR is 1. */
if (!integer_onep (gimple_assign_rhs2 (rshift_input_stmt)))
return NULL;
/* Look through any change in sign on the addition input. */
vect_unpromoted_value unprom_plus_input;
tree plus_input = vect_look_through_possible_promotion
(vinfo, gimple_assign_rhs1 (rshift_input_stmt), &unprom_plus_input);
if (!plus_input
|| TYPE_PRECISION (TREE_TYPE (plus_input))
!= TYPE_PRECISION (TREE_TYPE (rshift_input)))
return NULL;
/* Get the definition of the multiply-high-scale part. */
stmt_vec_info plus_input_stmt_info
= vect_get_internal_def (vinfo, plus_input);
if (!plus_input_stmt_info)
return NULL;
gassign *plus_input_stmt
= dyn_cast <gassign *> (plus_input_stmt_info->stmt);
if (!plus_input_stmt
|| gimple_assign_rhs_code (plus_input_stmt) != RSHIFT_EXPR)
return NULL;
/* Look through any change in sign on the scaling input. */
vect_unpromoted_value unprom_scale_input;
tree scale_input = vect_look_through_possible_promotion
(vinfo, gimple_assign_rhs1 (plus_input_stmt), &unprom_scale_input);
if (!scale_input
|| TYPE_PRECISION (TREE_TYPE (scale_input))
!= TYPE_PRECISION (TREE_TYPE (plus_input)))
return NULL;
/* Get the definition of the multiply-high part. */
mulh_stmt_info = vect_get_internal_def (vinfo, scale_input);
if (!mulh_stmt_info)
return NULL;
/* Get the scaling term. */
scale_term = gimple_assign_rhs2 (plus_input_stmt);
expect_offset = target_precision + 2;
ifn = IFN_MULHRS;
}
else
{
mulh_stmt_info = rshift_input_stmt_info;
scale_term = gimple_assign_rhs2 (last_stmt);
expect_offset = target_precision + 1;
ifn = IFN_MULHS;
}
/* Check that the scaling factor is correct. */
if (TREE_CODE (scale_term) != INTEGER_CST
|| wi::to_widest (scale_term) + expect_offset
!= TYPE_PRECISION (lhs_type))
return NULL;
/* Check whether the scaling input term can be seen as two widened
inputs multiplied together. */
vect_unpromoted_value unprom_mult[2];
tree new_type;
unsigned int nops
= vect_widened_op_tree (mulh_stmt_info, MULT_EXPR, WIDEN_MULT_EXPR,
false, 2, unprom_mult, &new_type);
if (nops != 2)
return NULL;
vect_pattern_detected ("vect_recog_mulhs_pattern", last_stmt);
/* Adjust output precision. */
if (TYPE_PRECISION (new_type) < target_precision)
new_type = build_nonstandard_integer_type
(target_precision, TYPE_UNSIGNED (new_type));
/* Check for target support. */
tree new_vectype = get_vectype_for_scalar_type (vinfo, new_type);
if (!new_vectype
|| !direct_internal_fn_supported_p
(ifn, new_vectype, OPTIMIZE_FOR_SPEED))
return NULL;
/* The IR requires a valid vector type for the cast result, even though
it's likely to be discarded. */
*type_out = get_vectype_for_scalar_type (vinfo, lhs_type);
if (!*type_out)
return NULL;
/* Generate the IFN_MULHRS call. */
tree new_var = vect_recog_temp_ssa_var (new_type, NULL);
tree new_ops[2];
vect_convert_inputs (last_stmt_info, 2, new_ops, new_type,
unprom_mult, new_vectype);
gcall *mulhrs_stmt
= gimple_build_call_internal (ifn, 2, new_ops[0], new_ops[1]);
gimple_call_set_lhs (mulhrs_stmt, new_var);
gimple_set_location (mulhrs_stmt, gimple_location (last_stmt));
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"created pattern stmt: %G", mulhrs_stmt);
return vect_convert_output (last_stmt_info, lhs_type,
mulhrs_stmt, new_vectype);
}
/* Recognize the patterns:
ATYPE a; // narrower than TYPE
BTYPE b; // narrower than TYPE
(1) TYPE avg = ((TYPE) a + (TYPE) b) >> 1;
or (2) TYPE avg = ((TYPE) a + (TYPE) b + 1) >> 1;
where only the bottom half of avg is used. Try to transform them into:
(1) NTYPE avg' = .AVG_FLOOR ((NTYPE) a, (NTYPE) b);
or (2) NTYPE avg' = .AVG_CEIL ((NTYPE) a, (NTYPE) b);
followed by:
TYPE avg = (TYPE) avg';
where NTYPE is no wider than half of TYPE. Since only the bottom half
of avg is used, all or part of the cast of avg' should become redundant. */
static gimple *
vect_recog_average_pattern (stmt_vec_info last_stmt_info, tree *type_out)
{
/* Check for a shift right by one bit. */
gassign *last_stmt = dyn_cast <gassign *> (last_stmt_info->stmt);
vec_info *vinfo = last_stmt_info->vinfo;
if (!last_stmt
|| gimple_assign_rhs_code (last_stmt) != RSHIFT_EXPR
|| !integer_onep (gimple_assign_rhs2 (last_stmt)))
return NULL;
/* Check that the shift result is wider than the users of the
result need (i.e. that narrowing would be a natural choice). */
tree lhs = gimple_assign_lhs (last_stmt);
tree type = TREE_TYPE (lhs);
unsigned int target_precision
= vect_element_precision (last_stmt_info->min_output_precision);
if (!INTEGRAL_TYPE_P (type) || target_precision >= TYPE_PRECISION (type))
return NULL;
/* Look through any change in sign on the shift input. */
tree rshift_rhs = gimple_assign_rhs1 (last_stmt);
vect_unpromoted_value unprom_plus;
rshift_rhs = vect_look_through_possible_promotion (vinfo, rshift_rhs,
&unprom_plus);
if (!rshift_rhs
|| TYPE_PRECISION (TREE_TYPE (rshift_rhs)) != TYPE_PRECISION (type))
return NULL;
/* Get the definition of the shift input. */
stmt_vec_info plus_stmt_info = vect_get_internal_def (vinfo, rshift_rhs);
if (!plus_stmt_info)
return NULL;
/* Check whether the shift input can be seen as a tree of additions on
2 or 3 widened inputs.
Note that the pattern should be a win even if the result of one or
more additions is reused elsewhere: if the pattern matches, we'd be
replacing 2N RSHIFT_EXPRs and N VEC_PACK_*s with N IFN_AVG_*s. */
internal_fn ifn = IFN_AVG_FLOOR;
vect_unpromoted_value unprom[3];
tree new_type;
unsigned int nops = vect_widened_op_tree (plus_stmt_info, PLUS_EXPR,
PLUS_EXPR, false, 3,
unprom, &new_type);
if (nops == 0)
return NULL;
if (nops == 3)
{
/* Check that one operand is 1. */
unsigned int i;
for (i = 0; i < 3; ++i)
if (integer_onep (unprom[i].op))
break;
if (i == 3)
return NULL;
/* Throw away the 1 operand and keep the other two. */
if (i < 2)
unprom[i] = unprom[2];
ifn = IFN_AVG_CEIL;
}
vect_pattern_detected ("vect_recog_average_pattern", last_stmt);
/* We know that:
(a) the operation can be viewed as:
TYPE widened0 = (TYPE) UNPROM[0];
TYPE widened1 = (TYPE) UNPROM[1];
TYPE tmp1 = widened0 + widened1 {+ 1};
TYPE tmp2 = tmp1 >> 1; // LAST_STMT_INFO
(b) the first two statements are equivalent to:
TYPE widened0 = (TYPE) (NEW_TYPE) UNPROM[0];
TYPE widened1 = (TYPE) (NEW_TYPE) UNPROM[1];
(c) vect_recog_over_widening_pattern has already tried to narrow TYPE
where sensible;
(d) all the operations can be performed correctly at twice the width of
NEW_TYPE, due to the nature of the average operation; and
(e) users of the result of the right shift need only TARGET_PRECISION
bits, where TARGET_PRECISION is no more than half of TYPE's
precision.
Under these circumstances, the only situation in which NEW_TYPE
could be narrower than TARGET_PRECISION is if widened0, widened1
and an addition result are all used more than once. Thus we can
treat any widening of UNPROM[0] and UNPROM[1] to TARGET_PRECISION
as "free", whereas widening the result of the average instruction
from NEW_TYPE to TARGET_PRECISION would be a new operation. It's
therefore better not to go narrower than TARGET_PRECISION. */
if (TYPE_PRECISION (new_type) < target_precision)
new_type = build_nonstandard_integer_type (target_precision,
TYPE_UNSIGNED (new_type));
/* Check for target support. */
tree new_vectype = get_vectype_for_scalar_type (vinfo, new_type);
if (!new_vectype
|| !direct_internal_fn_supported_p (ifn, new_vectype,
OPTIMIZE_FOR_SPEED))
return NULL;
/* The IR requires a valid vector type for the cast result, even though
it's likely to be discarded. */
*type_out = get_vectype_for_scalar_type (vinfo, type);
if (!*type_out)
return NULL;
/* Generate the IFN_AVG* call. */
tree new_var = vect_recog_temp_ssa_var (new_type, NULL);
tree new_ops[2];
vect_convert_inputs (last_stmt_info, 2, new_ops, new_type,
unprom, new_vectype);
gcall *average_stmt = gimple_build_call_internal (ifn, 2, new_ops[0],
new_ops[1]);
gimple_call_set_lhs (average_stmt, new_var);
gimple_set_location (average_stmt, gimple_location (last_stmt));
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"created pattern stmt: %G", average_stmt);
return vect_convert_output (last_stmt_info, type, average_stmt, new_vectype);
}
/* Recognize cases in which the input to a cast is wider than its
output, and the input is fed by a widening operation. Fold this
by removing the unnecessary intermediate widening. E.g.:
unsigned char a;
unsigned int b = (unsigned int) a;
unsigned short c = (unsigned short) b;
-->
unsigned short c = (unsigned short) a;
Although this is rare in input IR, it is an expected side-effect
of the over-widening pattern above.
This is beneficial also for integer-to-float conversions, if the
widened integer has more bits than the float, and if the unwidened
input doesn't. */
static gimple *
vect_recog_cast_forwprop_pattern (stmt_vec_info last_stmt_info, tree *type_out)
{
/* Check for a cast, including an integer-to-float conversion. */
gassign *last_stmt = dyn_cast <gassign *> (last_stmt_info->stmt);
if (!last_stmt)
return NULL;
tree_code code = gimple_assign_rhs_code (last_stmt);
if (!CONVERT_EXPR_CODE_P (code) && code != FLOAT_EXPR)
return NULL;
/* Make sure that the rhs is a scalar with a natural bitsize. */
tree lhs = gimple_assign_lhs (last_stmt);
if (!lhs)
return NULL;
tree lhs_type = TREE_TYPE (lhs);
scalar_mode lhs_mode;
if (VECT_SCALAR_BOOLEAN_TYPE_P (lhs_type)
|| !is_a <scalar_mode> (TYPE_MODE (lhs_type), &lhs_mode))
return NULL;
/* Check for a narrowing operation (from a vector point of view). */
tree rhs = gimple_assign_rhs1 (last_stmt);
tree rhs_type = TREE_TYPE (rhs);
if (!INTEGRAL_TYPE_P (rhs_type)
|| VECT_SCALAR_BOOLEAN_TYPE_P (rhs_type)
|| TYPE_PRECISION (rhs_type) <= GET_MODE_BITSIZE (lhs_mode))
return NULL;
/* Try to find an unpromoted input. */
vec_info *vinfo = last_stmt_info->vinfo;
vect_unpromoted_value unprom;
if (!vect_look_through_possible_promotion (vinfo, rhs, &unprom)
|| TYPE_PRECISION (unprom.type) >= TYPE_PRECISION (rhs_type))
return NULL;
/* If the bits above RHS_TYPE matter, make sure that they're the
same when extending from UNPROM as they are when extending from RHS. */
if (!INTEGRAL_TYPE_P (lhs_type)
&& TYPE_SIGN (rhs_type) != TYPE_SIGN (unprom.type))
return NULL;
/* We can get the same result by casting UNPROM directly, to avoid
the unnecessary widening and narrowing. */
vect_pattern_detected ("vect_recog_cast_forwprop_pattern", last_stmt);
*type_out = get_vectype_for_scalar_type (vinfo, lhs_type);
if (!*type_out)
return NULL;
tree new_var = vect_recog_temp_ssa_var (lhs_type, NULL);
gimple *pattern_stmt = gimple_build_assign (new_var, code, unprom.op);
gimple_set_location (pattern_stmt, gimple_location (last_stmt));
return pattern_stmt;
}
/* Try to detect a shift left of a widened input, converting LSHIFT_EXPR
to WIDEN_LSHIFT_EXPR. See vect_recog_widen_op_pattern for details. */
static gimple *
vect_recog_widen_shift_pattern (stmt_vec_info last_stmt_info, tree *type_out)
{
return vect_recog_widen_op_pattern (last_stmt_info, type_out, LSHIFT_EXPR,
WIDEN_LSHIFT_EXPR, true,
"vect_recog_widen_shift_pattern");
}
/* Detect a rotate pattern wouldn't be otherwise vectorized:
type a_t, b_t, c_t;
S0 a_t = b_t r<< c_t;
Input/Output:
* STMT_VINFO: The stmt from which the pattern search begins,
i.e. the shift/rotate stmt. The original stmt (S0) is replaced
with a sequence:
S1 d_t = -c_t;
S2 e_t = d_t & (B - 1);
S3 f_t = b_t << c_t;
S4 g_t = b_t >> e_t;
S0 a_t = f_t | g_t;
where B is element bitsize of type.
Output:
* TYPE_OUT: The type of the output of this pattern.
* Return value: A new stmt that will be used to replace the rotate
S0 stmt. */
static gimple *
vect_recog_rotate_pattern (stmt_vec_info stmt_vinfo, tree *type_out)
{
gimple *last_stmt = stmt_vinfo->stmt;
tree oprnd0, oprnd1, lhs, var, var1, var2, vectype, type, stype, def, def2;
gimple *pattern_stmt, *def_stmt;
enum tree_code rhs_code;
vec_info *vinfo = stmt_vinfo->vinfo;
enum vect_def_type dt;
optab optab1, optab2;
edge ext_def = NULL;
bool bswap16_p = false;
if (is_gimple_assign (last_stmt))
{
rhs_code = gimple_assign_rhs_code (last_stmt);
switch (rhs_code)
{
case LROTATE_EXPR:
case RROTATE_EXPR:
break;
default:
return NULL;
}
lhs = gimple_assign_lhs (last_stmt);
oprnd0 = gimple_assign_rhs1 (last_stmt);
type = TREE_TYPE (oprnd0);
oprnd1 = gimple_assign_rhs2 (last_stmt);
}
else if (gimple_call_builtin_p (last_stmt, BUILT_IN_BSWAP16))
{
/* __builtin_bswap16 (x) is another form of x r>> 8.
The vectorizer has bswap support, but only if the argument isn't
promoted. */
lhs = gimple_call_lhs (last_stmt);
oprnd0 = gimple_call_arg (last_stmt, 0);
type = TREE_TYPE (oprnd0);
if (!lhs
|| TYPE_PRECISION (TREE_TYPE (lhs)) != 16
|| TYPE_PRECISION (type) <= 16
|| TREE_CODE (oprnd0) != SSA_NAME
|| BITS_PER_UNIT != 8
|| !TYPE_UNSIGNED (TREE_TYPE (lhs)))
return NULL;
stmt_vec_info def_stmt_info;
if (!vect_is_simple_use (oprnd0, vinfo, &dt, &def_stmt_info, &def_stmt))
return NULL;
if (dt != vect_internal_def)
return NULL;
if (gimple_assign_cast_p (def_stmt))
{
def = gimple_assign_rhs1 (def_stmt);
if (INTEGRAL_TYPE_P (TREE_TYPE (def))
&& TYPE_PRECISION (TREE_TYPE (def)) == 16)
oprnd0 = def;
}
type = TREE_TYPE (lhs);
vectype = get_vectype_for_scalar_type (vinfo, type);
if (vectype == NULL_TREE)
return NULL;
if (tree char_vectype = get_same_sized_vectype (char_type_node, vectype))
{
/* The encoding uses one stepped pattern for each byte in the
16-bit word. */
vec_perm_builder elts (TYPE_VECTOR_SUBPARTS (char_vectype), 2, 3);
for (unsigned i = 0; i < 3; ++i)
for (unsigned j = 0; j < 2; ++j)
elts.quick_push ((i + 1) * 2 - j - 1);
vec_perm_indices indices (elts, 1,
TYPE_VECTOR_SUBPARTS (char_vectype));
if (can_vec_perm_const_p (TYPE_MODE (char_vectype), indices))
{
/* vectorizable_bswap can handle the __builtin_bswap16 if we
undo the argument promotion. */
if (!useless_type_conversion_p (type, TREE_TYPE (oprnd0)))
{
def = vect_recog_temp_ssa_var (type, NULL);
def_stmt = gimple_build_assign (def, NOP_EXPR, oprnd0);
append_pattern_def_seq (stmt_vinfo, def_stmt);
oprnd0 = def;
}
/* Pattern detected. */
vect_pattern_detected ("vect_recog_rotate_pattern", last_stmt);
*type_out = vectype;
/* Pattern supported. Create a stmt to be used to replace the
pattern, with the unpromoted argument. */
var = vect_recog_temp_ssa_var (type, NULL);
pattern_stmt = gimple_build_call (gimple_call_fndecl (last_stmt),
1, oprnd0);
gimple_call_set_lhs (pattern_stmt, var);
gimple_call_set_fntype (as_a <gcall *> (pattern_stmt),
gimple_call_fntype (last_stmt));
return pattern_stmt;
}
}
oprnd1 = build_int_cst (integer_type_node, 8);
rhs_code = LROTATE_EXPR;
bswap16_p = true;
}
else
return NULL;
if (TREE_CODE (oprnd0) != SSA_NAME
|| TYPE_PRECISION (TREE_TYPE (lhs)) != TYPE_PRECISION (type)
|| !INTEGRAL_TYPE_P (type)
|| !TYPE_UNSIGNED (type))
return NULL;
stmt_vec_info def_stmt_info;
if (!vect_is_simple_use (oprnd1, vinfo, &dt, &def_stmt_info, &def_stmt))
return NULL;
if (dt != vect_internal_def
&& dt != vect_constant_def
&& dt != vect_external_def)
return NULL;
vectype = get_vectype_for_scalar_type (vinfo, type);
if (vectype == NULL_TREE)
return NULL;
/* If vector/vector or vector/scalar rotate is supported by the target,
don't do anything here. */
optab1 = optab_for_tree_code (rhs_code, vectype, optab_vector);
if (optab1
&& optab_handler (optab1, TYPE_MODE (vectype)) != CODE_FOR_nothing)
{
use_rotate:
if (bswap16_p)
{
if (!useless_type_conversion_p (type, TREE_TYPE (oprnd0)))
{
def = vect_recog_temp_ssa_var (type, NULL);
def_stmt = gimple_build_assign (def, NOP_EXPR, oprnd0);
append_pattern_def_seq (stmt_vinfo, def_stmt);
oprnd0 = def;
}
/* Pattern detected. */
vect_pattern_detected ("vect_recog_rotate_pattern", last_stmt);
*type_out = vectype;
/* Pattern supported. Create a stmt to be used to replace the
pattern. */
var = vect_recog_temp_ssa_var (type, NULL);
pattern_stmt = gimple_build_assign (var, LROTATE_EXPR, oprnd0,
oprnd1);
return pattern_stmt;
}
return NULL;
}
if (is_a <bb_vec_info> (vinfo) || dt != vect_internal_def)
{
optab2 = optab_for_tree_code (rhs_code, vectype, optab_scalar);
if (optab2
&& optab_handler (optab2, TYPE_MODE (vectype)) != CODE_FOR_nothing)
goto use_rotate;
}
/* If vector/vector or vector/scalar shifts aren't supported by the target,
don't do anything here either. */
optab1 = optab_for_tree_code (LSHIFT_EXPR, vectype, optab_vector);
optab2 = optab_for_tree_code (RSHIFT_EXPR, vectype, optab_vector);
if (!optab1
|| optab_handler (optab1, TYPE_MODE (vectype)) == CODE_FOR_nothing
|| !optab2
|| optab_handler (optab2, TYPE_MODE (vectype)) == CODE_FOR_nothing)
{
if (! is_a <bb_vec_info> (vinfo) && dt == vect_internal_def)
return NULL;
optab1 = optab_for_tree_code (LSHIFT_EXPR, vectype, optab_scalar);
optab2 = optab_for_tree_code (RSHIFT_EXPR, vectype, optab_scalar);
if (!optab1
|| optab_handler (optab1, TYPE_MODE (vectype)) == CODE_FOR_nothing
|| !optab2
|| optab_handler (optab2, TYPE_MODE (vectype)) == CODE_FOR_nothing)
return NULL;
}
*type_out = vectype;
if (bswap16_p && !useless_type_conversion_p (type, TREE_TYPE (oprnd0)))
{
def = vect_recog_temp_ssa_var (type, NULL);
def_stmt = gimple_build_assign (def, NOP_EXPR, oprnd0);
append_pattern_def_seq (stmt_vinfo, def_stmt);
oprnd0 = def;
}
if (dt == vect_external_def
&& TREE_CODE (oprnd1) == SSA_NAME)
ext_def = vect_get_external_def_edge (vinfo, oprnd1);
def = NULL_TREE;
scalar_int_mode mode = SCALAR_INT_TYPE_MODE (type);
if (TREE_CODE (oprnd1) == INTEGER_CST
|| TYPE_MODE (TREE_TYPE (oprnd1)) == mode)
def = oprnd1;
else if (def_stmt && gimple_assign_cast_p (def_stmt))
{
tree rhs1 = gimple_assign_rhs1 (def_stmt);
if (TYPE_MODE (TREE_TYPE (rhs1)) == mode
&& TYPE_PRECISION (TREE_TYPE (rhs1))
== TYPE_PRECISION (type))
def = rhs1;
}
if (def == NULL_TREE)
{
def = vect_recog_temp_ssa_var (type, NULL);
def_stmt = gimple_build_assign (def, NOP_EXPR, oprnd1);
if (ext_def)
{
basic_block new_bb
= gsi_insert_on_edge_immediate (ext_def, def_stmt);
gcc_assert (!new_bb);
}
else
append_pattern_def_seq (stmt_vinfo, def_stmt);
}
stype = TREE_TYPE (def);
scalar_int_mode smode = SCALAR_INT_TYPE_MODE (stype);
if (TREE_CODE (def) == INTEGER_CST)
{
if (!tree_fits_uhwi_p (def)
|| tree_to_uhwi (def) >= GET_MODE_PRECISION (mode)
|| integer_zerop (def))
return NULL;
def2 = build_int_cst (stype,
GET_MODE_PRECISION (mode) - tree_to_uhwi (def));
}
else
{
tree vecstype = get_vectype_for_scalar_type (vinfo, stype);
if (vecstype == NULL_TREE)
return NULL;
def2 = vect_recog_temp_ssa_var (stype, NULL);
def_stmt = gimple_build_assign (def2, NEGATE_EXPR, def);
if (ext_def)
{
basic_block new_bb
= gsi_insert_on_edge_immediate (ext_def, def_stmt);
gcc_assert (!new_bb);
}
else
append_pattern_def_seq (stmt_vinfo, def_stmt, vecstype);
def2 = vect_recog_temp_ssa_var (stype, NULL);
tree mask = build_int_cst (stype, GET_MODE_PRECISION (smode) - 1);
def_stmt = gimple_build_assign (def2, BIT_AND_EXPR,
gimple_assign_lhs (def_stmt), mask);
if (ext_def)
{
basic_block new_bb
= gsi_insert_on_edge_immediate (ext_def, def_stmt);
gcc_assert (!new_bb);
}
else
append_pattern_def_seq (stmt_vinfo, def_stmt, vecstype);
}
var1 = vect_recog_temp_ssa_var (type, NULL);
def_stmt = gimple_build_assign (var1, rhs_code == LROTATE_EXPR
? LSHIFT_EXPR : RSHIFT_EXPR,
oprnd0, def);
append_pattern_def_seq (stmt_vinfo, def_stmt);
var2 = vect_recog_temp_ssa_var (type, NULL);
def_stmt = gimple_build_assign (var2, rhs_code == LROTATE_EXPR
? RSHIFT_EXPR : LSHIFT_EXPR,
oprnd0, def2);
append_pattern_def_seq (stmt_vinfo, def_stmt);
/* Pattern detected. */
vect_pattern_detected ("vect_recog_rotate_pattern", last_stmt);
/* Pattern supported. Create a stmt to be used to replace the pattern. */
var = vect_recog_temp_ssa_var (type, NULL);
pattern_stmt = gimple_build_assign (var, BIT_IOR_EXPR, var1, var2);
return pattern_stmt;
}
/* Detect a vector by vector shift pattern that wouldn't be otherwise
vectorized:
type a_t;
TYPE b_T, res_T;
S1 a_t = ;
S2 b_T = ;
S3 res_T = b_T op a_t;
where type 'TYPE' is a type with different size than 'type',
and op is <<, >> or rotate.
Also detect cases:
type a_t;
TYPE b_T, c_T, res_T;
S0 c_T = ;
S1 a_t = (type) c_T;
S2 b_T = ;
S3 res_T = b_T op a_t;
Input/Output:
* STMT_VINFO: The stmt from which the pattern search begins,
i.e. the shift/rotate stmt. The original stmt (S3) is replaced
with a shift/rotate which has same type on both operands, in the
second case just b_T op c_T, in the first case with added cast
from a_t to c_T in STMT_VINFO_PATTERN_DEF_SEQ.
Output:
* TYPE_OUT: The type of the output of this pattern.
* Return value: A new stmt that will be used to replace the shift/rotate
S3 stmt. */
static gimple *
vect_recog_vector_vector_shift_pattern (stmt_vec_info stmt_vinfo,
tree *type_out)
{
gimple *last_stmt = stmt_vinfo->stmt;
tree oprnd0, oprnd1, lhs, var;
gimple *pattern_stmt;
enum tree_code rhs_code;
vec_info *vinfo = stmt_vinfo->vinfo;
if (!is_gimple_assign (last_stmt))
return NULL;
rhs_code = gimple_assign_rhs_code (last_stmt);
switch (rhs_code)
{
case LSHIFT_EXPR:
case RSHIFT_EXPR:
case LROTATE_EXPR:
case RROTATE_EXPR:
break;
default:
return NULL;
}
lhs = gimple_assign_lhs (last_stmt);
oprnd0 = gimple_assign_rhs1 (last_stmt);
oprnd1 = gimple_assign_rhs2 (last_stmt);
if (TREE_CODE (oprnd0) != SSA_NAME
|| TREE_CODE (oprnd1) != SSA_NAME
|| TYPE_MODE (TREE_TYPE (oprnd0)) == TYPE_MODE (TREE_TYPE (oprnd1))
|| !type_has_mode_precision_p (TREE_TYPE (oprnd1))
|| TYPE_PRECISION (TREE_TYPE (lhs))
!= TYPE_PRECISION (TREE_TYPE (oprnd0)))
return NULL;
stmt_vec_info def_vinfo = vect_get_internal_def (vinfo, oprnd1);
if (!def_vinfo)
return NULL;
*type_out = get_vectype_for_scalar_type (vinfo, TREE_TYPE (oprnd0));
if (*type_out == NULL_TREE)
return NULL;
tree def = NULL_TREE;
gassign *def_stmt = dyn_cast <gassign *> (def_vinfo->stmt);
if (def_stmt && gimple_assign_cast_p (def_stmt))
{
tree rhs1 = gimple_assign_rhs1 (def_stmt);
if (TYPE_MODE (TREE_TYPE (rhs1)) == TYPE_MODE (TREE_TYPE (oprnd0))
&& TYPE_PRECISION (TREE_TYPE (rhs1))
== TYPE_PRECISION (TREE_TYPE (oprnd0)))
{
if (TYPE_PRECISION (TREE_TYPE (oprnd1))
>= TYPE_PRECISION (TREE_TYPE (rhs1)))
def = rhs1;
else
{
tree mask
= build_low_bits_mask (TREE_TYPE (rhs1),
TYPE_PRECISION (TREE_TYPE (oprnd1)));
def = vect_recog_temp_ssa_var (TREE_TYPE (rhs1), NULL);
def_stmt = gimple_build_assign (def, BIT_AND_EXPR, rhs1, mask);
tree vecstype = get_vectype_for_scalar_type (vinfo,
TREE_TYPE (rhs1));
append_pattern_def_seq (stmt_vinfo, def_stmt, vecstype);
}
}
}
if (def == NULL_TREE)
{
def = vect_recog_temp_ssa_var (TREE_TYPE (oprnd0), NULL);
def_stmt = gimple_build_assign (def, NOP_EXPR, oprnd1);
append_pattern_def_seq (stmt_vinfo, def_stmt);
}
/* Pattern detected. */
vect_pattern_detected ("vect_recog_vector_vector_shift_pattern", last_stmt);
/* Pattern supported. Create a stmt to be used to replace the pattern. */
var = vect_recog_temp_ssa_var (TREE_TYPE (oprnd0), NULL);
pattern_stmt = gimple_build_assign (var, rhs_code, oprnd0, def);
return pattern_stmt;
}
/* Return true iff the target has a vector optab implementing the operation
CODE on type VECTYPE. */
static bool
target_has_vecop_for_code (tree_code code, tree vectype)
{
optab voptab = optab_for_tree_code (code, vectype, optab_vector);
return voptab
&& optab_handler (voptab, TYPE_MODE (vectype)) != CODE_FOR_nothing;
}
/* Verify that the target has optabs of VECTYPE to perform all the steps
needed by the multiplication-by-immediate synthesis algorithm described by
ALG and VAR. If SYNTH_SHIFT_P is true ensure that vector addition is
present. Return true iff the target supports all the steps. */
static bool
target_supports_mult_synth_alg (struct algorithm *alg, mult_variant var,
tree vectype, bool synth_shift_p)
{
if (alg->op[0] != alg_zero && alg->op[0] != alg_m)
return false;
bool supports_vminus = target_has_vecop_for_code (MINUS_EXPR, vectype);
bool supports_vplus = target_has_vecop_for_code (PLUS_EXPR, vectype);
if (var == negate_variant
&& !target_has_vecop_for_code (NEGATE_EXPR, vectype))
return false;
/* If we must synthesize shifts with additions make sure that vector
addition is available. */
if ((var == add_variant || synth_shift_p) && !supports_vplus)
return false;
for (int i = 1; i < alg->ops; i++)
{
switch (alg->op[i])
{
case alg_shift:
break;
case alg_add_t_m2:
case alg_add_t2_m:
case alg_add_factor:
if (!supports_vplus)
return false;
break;
case alg_sub_t_m2:
case alg_sub_t2_m:
case alg_sub_factor:
if (!supports_vminus)
return false;
break;
case alg_unknown:
case alg_m:
case alg_zero:
case alg_impossible:
return false;
default:
gcc_unreachable ();
}
}
return true;
}
/* Synthesize a left shift of OP by AMNT bits using a series of additions and
putting the final result in DEST. Append all statements but the last into
VINFO. Return the last statement. */
static gimple *
synth_lshift_by_additions (tree dest, tree op, HOST_WIDE_INT amnt,
stmt_vec_info vinfo)
{
HOST_WIDE_INT i;
tree itype = TREE_TYPE (op);
tree prev_res = op;
gcc_assert (amnt >= 0);
for (i = 0; i < amnt; i++)
{
tree tmp_var = (i < amnt - 1) ? vect_recog_temp_ssa_var (itype, NULL)
: dest;
gimple *stmt
= gimple_build_assign (tmp_var, PLUS_EXPR, prev_res, prev_res);
prev_res = tmp_var;
if (i < amnt - 1)
append_pattern_def_seq (vinfo, stmt);
else
return stmt;
}
gcc_unreachable ();
return NULL;
}
/* Helper for vect_synth_mult_by_constant. Apply a binary operation
CODE to operands OP1 and OP2, creating a new temporary SSA var in
the process if necessary. Append the resulting assignment statements
to the sequence in STMT_VINFO. Return the SSA variable that holds the
result of the binary operation. If SYNTH_SHIFT_P is true synthesize
left shifts using additions. */
static tree
apply_binop_and_append_stmt (tree_code code, tree op1, tree op2,
stmt_vec_info stmt_vinfo, bool synth_shift_p)
{
if (integer_zerop (op2)
&& (code == LSHIFT_EXPR
|| code == PLUS_EXPR))
{
gcc_assert (TREE_CODE (op1) == SSA_NAME);
return op1;
}
gimple *stmt;
tree itype = TREE_TYPE (op1);
tree tmp_var = vect_recog_temp_ssa_var (itype, NULL);
if (code == LSHIFT_EXPR
&& synth_shift_p)
{
stmt = synth_lshift_by_additions (tmp_var, op1, TREE_INT_CST_LOW (op2),
stmt_vinfo);
append_pattern_def_seq (stmt_vinfo, stmt);
return tmp_var;
}
stmt = gimple_build_assign (tmp_var, code, op1, op2);
append_pattern_def_seq (stmt_vinfo, stmt);
return tmp_var;
}
/* Synthesize a multiplication of OP by an INTEGER_CST VAL using shifts
and simple arithmetic operations to be vectorized. Record the statements
produced in STMT_VINFO and return the last statement in the sequence or
NULL if it's not possible to synthesize such a multiplication.
This function mirrors the behavior of expand_mult_const in expmed.c but
works on tree-ssa form. */
static gimple *
vect_synth_mult_by_constant (tree op, tree val,
stmt_vec_info stmt_vinfo)
{
vec_info *vinfo = stmt_vinfo->vinfo;
tree itype = TREE_TYPE (op);
machine_mode mode = TYPE_MODE (itype);
struct algorithm alg;
mult_variant variant;
if (!tree_fits_shwi_p (val))
return NULL;
/* Multiplication synthesis by shifts, adds and subs can introduce
signed overflow where the original operation didn't. Perform the
operations on an unsigned type and cast back to avoid this.
In the future we may want to relax this for synthesis algorithms
that we can prove do not cause unexpected overflow. */
bool cast_to_unsigned_p = !TYPE_OVERFLOW_WRAPS (itype);
tree multtype = cast_to_unsigned_p ? unsigned_type_for (itype) : itype;
/* Targets that don't support vector shifts but support vector additions
can synthesize shifts that way. */
bool synth_shift_p = !vect_supportable_shift (vinfo, LSHIFT_EXPR, multtype);
HOST_WIDE_INT hwval = tree_to_shwi (val);
/* Use MAX_COST here as we don't want to limit the sequence on rtx costs.
The vectorizer's benefit analysis will decide whether it's beneficial
to do this. */
bool possible = choose_mult_variant (mode, hwval, &alg,
&variant, MAX_COST);
if (!possible)
return NULL;
tree vectype = get_vectype_for_scalar_type (vinfo, multtype);
if (!vectype
|| !target_supports_mult_synth_alg (&alg, variant,
vectype, synth_shift_p))
return NULL;
tree accumulator;
/* Clear out the sequence of statements so we can populate it below. */
gimple *stmt = NULL;
if (cast_to_unsigned_p)
{
tree tmp_op = vect_recog_temp_ssa_var (multtype, NULL);
stmt = gimple_build_assign (tmp_op, CONVERT_EXPR, op);
append_pattern_def_seq (stmt_vinfo, stmt);
op = tmp_op;
}
if (alg.op[0] == alg_zero)
accumulator = build_int_cst (multtype, 0);
else
accumulator = op;
bool needs_fixup = (variant == negate_variant)
|| (variant == add_variant);
for (int i = 1; i < alg.ops; i++)
{
tree shft_log = build_int_cst (multtype, alg.log[i]);
tree accum_tmp = vect_recog_temp_ssa_var (multtype, NULL);
tree tmp_var = NULL_TREE;
switch (alg.op[i])
{
case alg_shift:
if (synth_shift_p)
stmt
= synth_lshift_by_additions (accum_tmp, accumulator, alg.log[i],
stmt_vinfo);
else
stmt = gimple_build_assign (accum_tmp, LSHIFT_EXPR, accumulator,
shft_log);
break;
case alg_add_t_m2:
tmp_var
= apply_binop_and_append_stmt (LSHIFT_EXPR, op, shft_log,
stmt_vinfo, synth_shift_p);
stmt = gimple_build_assign (accum_tmp, PLUS_EXPR, accumulator,
tmp_var);
break;
case alg_sub_t_m2:
tmp_var = apply_binop_and_append_stmt (LSHIFT_EXPR, op,
shft_log, stmt_vinfo,
synth_shift_p);
/* In some algorithms the first step involves zeroing the
accumulator. If subtracting from such an accumulator
just emit the negation directly. */
if (integer_zerop (accumulator))
stmt = gimple_build_assign (accum_tmp, NEGATE_EXPR, tmp_var);
else
stmt = gimple_build_assign (accum_tmp, MINUS_EXPR, accumulator,
tmp_var);
break;
case alg_add_t2_m:
tmp_var
= apply_binop_and_append_stmt (LSHIFT_EXPR, accumulator, shft_log,
stmt_vinfo, synth_shift_p);
stmt = gimple_build_assign (accum_tmp, PLUS_EXPR, tmp_var, op);
break;
case alg_sub_t2_m:
tmp_var
= apply_binop_and_append_stmt (LSHIFT_EXPR, accumulator, shft_log,
stmt_vinfo, synth_shift_p);
stmt = gimple_build_assign (accum_tmp, MINUS_EXPR, tmp_var, op);
break;
case alg_add_factor:
tmp_var
= apply_binop_and_append_stmt (LSHIFT_EXPR, accumulator, shft_log,
stmt_vinfo, synth_shift_p);
stmt = gimple_build_assign (accum_tmp, PLUS_EXPR, accumulator,
tmp_var);
break;
case alg_sub_factor:
tmp_var
= apply_binop_and_append_stmt (LSHIFT_EXPR, accumulator, shft_log,
stmt_vinfo, synth_shift_p);
stmt = gimple_build_assign (accum_tmp, MINUS_EXPR, tmp_var,
accumulator);
break;
default:
gcc_unreachable ();
}
/* We don't want to append the last stmt in the sequence to stmt_vinfo
but rather return it directly. */
if ((i < alg.ops - 1) || needs_fixup || cast_to_unsigned_p)
append_pattern_def_seq (stmt_vinfo, stmt);
accumulator = accum_tmp;
}
if (variant == negate_variant)
{
tree accum_tmp = vect_recog_temp_ssa_var (multtype, NULL);
stmt = gimple_build_assign (accum_tmp, NEGATE_EXPR, accumulator);
accumulator = accum_tmp;
if (cast_to_unsigned_p)
append_pattern_def_seq (stmt_vinfo, stmt);
}
else if (variant == add_variant)
{
tree accum_tmp = vect_recog_temp_ssa_var (multtype, NULL);
stmt = gimple_build_assign (accum_tmp, PLUS_EXPR, accumulator, op);
accumulator = accum_tmp;
if (cast_to_unsigned_p)
append_pattern_def_seq (stmt_vinfo, stmt);
}
/* Move back to a signed if needed. */
if (cast_to_unsigned_p)
{
tree accum_tmp = vect_recog_temp_ssa_var (itype, NULL);
stmt = gimple_build_assign (accum_tmp, CONVERT_EXPR, accumulator);
}
return stmt;
}
/* Detect multiplication by constant and convert it into a sequence of
shifts and additions, subtractions, negations. We reuse the
choose_mult_variant algorithms from expmed.c
Input/Output:
STMT_VINFO: The stmt from which the pattern search begins,
i.e. the mult stmt.
Output:
* TYPE_OUT: The type of the output of this pattern.
* Return value: A new stmt that will be used to replace
the multiplication. */
static gimple *
vect_recog_mult_pattern (stmt_vec_info stmt_vinfo, tree *type_out)
{
vec_info *vinfo = stmt_vinfo->vinfo;
gimple *last_stmt = stmt_vinfo->stmt;
tree oprnd0, oprnd1, vectype, itype;
gimple *pattern_stmt;
if (!is_gimple_assign (last_stmt))
return NULL;
if (gimple_assign_rhs_code (last_stmt) != MULT_EXPR)
return NULL;
oprnd0 = gimple_assign_rhs1 (last_stmt);
oprnd1 = gimple_assign_rhs2 (last_stmt);
itype = TREE_TYPE (oprnd0);
if (TREE_CODE (oprnd0) != SSA_NAME
|| TREE_CODE (oprnd1) != INTEGER_CST
|| !INTEGRAL_TYPE_P (itype)
|| !type_has_mode_precision_p (itype))
return NULL;
vectype = get_vectype_for_scalar_type (vinfo, itype);
if (vectype == NULL_TREE)
return NULL;
/* If the target can handle vectorized multiplication natively,
don't attempt to optimize this. */
optab mul_optab = optab_for_tree_code (MULT_EXPR, vectype, optab_default);
if (mul_optab != unknown_optab)
{
machine_mode vec_mode = TYPE_MODE (vectype);
int icode = (int) optab_handler (mul_optab, vec_mode);
if (icode != CODE_FOR_nothing)
return NULL;
}
pattern_stmt = vect_synth_mult_by_constant (oprnd0, oprnd1, stmt_vinfo);
if (!pattern_stmt)
return NULL;
/* Pattern detected. */
vect_pattern_detected ("vect_recog_mult_pattern", last_stmt);
*type_out = vectype;
return pattern_stmt;
}
/* Detect a signed division by a constant that wouldn't be
otherwise vectorized:
type a_t, b_t;
S1 a_t = b_t / N;
where type 'type' is an integral type and N is a constant.
Similarly handle modulo by a constant:
S4 a_t = b_t % N;
Input/Output:
* STMT_VINFO: The stmt from which the pattern search begins,
i.e. the division stmt. S1 is replaced by if N is a power
of two constant and type is signed:
S3 y_t = b_t < 0 ? N - 1 : 0;
S2 x_t = b_t + y_t;
S1' a_t = x_t >> log2 (N);
S4 is replaced if N is a power of two constant and
type is signed by (where *_T temporaries have unsigned type):
S9 y_T = b_t < 0 ? -1U : 0U;
S8 z_T = y_T >> (sizeof (type_t) * CHAR_BIT - log2 (N));
S7 z_t = (type) z_T;
S6 w_t = b_t + z_t;
S5 x_t = w_t & (N - 1);
S4' a_t = x_t - z_t;
Output:
* TYPE_OUT: The type of the output of this pattern.
* Return value: A new stmt that will be used to replace the division
S1 or modulo S4 stmt. */
static gimple *
vect_recog_divmod_pattern (stmt_vec_info stmt_vinfo, tree *type_out)
{
vec_info *vinfo = stmt_vinfo->vinfo;
gimple *last_stmt = stmt_vinfo->stmt;
tree oprnd0, oprnd1, vectype, itype, cond;
gimple *pattern_stmt, *def_stmt;
enum tree_code rhs_code;
optab optab;
tree q;
int dummy_int, prec;
if (!is_gimple_assign (last_stmt))
return NULL;
rhs_code = gimple_assign_rhs_code (last_stmt);
switch (rhs_code)
{
case TRUNC_DIV_EXPR:
case EXACT_DIV_EXPR:
case TRUNC_MOD_EXPR:
break;
default:
return NULL;
}
oprnd0 = gimple_assign_rhs1 (last_stmt);
oprnd1 = gimple_assign_rhs2 (last_stmt);
itype = TREE_TYPE (oprnd0);
if (TREE_CODE (oprnd0) != SSA_NAME
|| TREE_CODE (oprnd1) != INTEGER_CST
|| TREE_CODE (itype) != INTEGER_TYPE
|| !type_has_mode_precision_p (itype))
return NULL;
scalar_int_mode itype_mode = SCALAR_INT_TYPE_MODE (itype);
vectype = get_vectype_for_scalar_type (vinfo, itype);
if (vectype == NULL_TREE)
return NULL;
if (optimize_bb_for_size_p (gimple_bb (last_stmt)))
{
/* If the target can handle vectorized division or modulo natively,
don't attempt to optimize this, since native division is likely
to give smaller code. */
optab = optab_for_tree_code (rhs_code, vectype, optab_default);
if (optab != unknown_optab)
{
machine_mode vec_mode = TYPE_MODE (vectype);
int icode = (int) optab_handler (optab, vec_mode);
if (icode != CODE_FOR_nothing)
return NULL;
}
}
prec = TYPE_PRECISION (itype);
if (integer_pow2p (oprnd1))
{
if (TYPE_UNSIGNED (itype) || tree_int_cst_sgn (oprnd1) != 1)
return NULL;
/* Pattern detected. */
vect_pattern_detected ("vect_recog_divmod_pattern", last_stmt);
*type_out = vectype;
/* Check if the target supports this internal function. */
internal_fn ifn = IFN_DIV_POW2;
if (direct_internal_fn_supported_p (ifn, vectype, OPTIMIZE_FOR_SPEED))
{
tree shift = build_int_cst (itype, tree_log2 (oprnd1));
tree var_div = vect_recog_temp_ssa_var (itype, NULL);
gimple *div_stmt = gimple_build_call_internal (ifn, 2, oprnd0, shift);
gimple_call_set_lhs (div_stmt, var_div);
if (rhs_code == TRUNC_MOD_EXPR)
{
append_pattern_def_seq (stmt_vinfo, div_stmt);
def_stmt
= gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
LSHIFT_EXPR, var_div, shift);
append_pattern_def_seq (stmt_vinfo, def_stmt);
pattern_stmt
= gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
MINUS_EXPR, oprnd0,
gimple_assign_lhs (def_stmt));
}
else
pattern_stmt = div_stmt;
gimple_set_location (pattern_stmt, gimple_location (last_stmt));
return pattern_stmt;
}
cond = build2 (LT_EXPR, boolean_type_node, oprnd0,
build_int_cst (itype, 0));
if (rhs_code == TRUNC_DIV_EXPR
|| rhs_code == EXACT_DIV_EXPR)
{
tree var = vect_recog_temp_ssa_var (itype, NULL);
tree shift;
def_stmt
= gimple_build_assign (var, COND_EXPR, cond,
fold_build2 (MINUS_EXPR, itype, oprnd1,
build_int_cst (itype, 1)),
build_int_cst (itype, 0));
append_pattern_def_seq (stmt_vinfo, def_stmt);
var = vect_recog_temp_ssa_var (itype, NULL);
def_stmt
= gimple_build_assign (var, PLUS_EXPR, oprnd0,
gimple_assign_lhs (def_stmt));
append_pattern_def_seq (stmt_vinfo, def_stmt);
shift = build_int_cst (itype, tree_log2 (oprnd1));
pattern_stmt
= gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
RSHIFT_EXPR, var, shift);
}
else
{
tree signmask;
if (compare_tree_int (oprnd1, 2) == 0)
{
signmask = vect_recog_temp_ssa_var (itype, NULL);
def_stmt = gimple_build_assign (signmask, COND_EXPR, cond,
build_int_cst (itype, 1),
build_int_cst (itype, 0));
append_pattern_def_seq (stmt_vinfo, def_stmt);
}
else
{
tree utype
= build_nonstandard_integer_type (prec, 1);
tree vecutype = get_vectype_for_scalar_type (vinfo, utype);
tree shift
= build_int_cst (utype, GET_MODE_BITSIZE (itype_mode)
- tree_log2 (oprnd1));
tree var = vect_recog_temp_ssa_var (utype, NULL);
def_stmt = gimple_build_assign (var, COND_EXPR, cond,
build_int_cst (utype, -1),
build_int_cst (utype, 0));
append_pattern_def_seq (stmt_vinfo, def_stmt, vecutype);
var = vect_recog_temp_ssa_var (utype, NULL);
def_stmt = gimple_build_assign (var, RSHIFT_EXPR,
gimple_assign_lhs (def_stmt),
shift);
append_pattern_def_seq (stmt_vinfo, def_stmt, vecutype);
signmask = vect_recog_temp_ssa_var (itype, NULL);
def_stmt
= gimple_build_assign (signmask, NOP_EXPR, var);
append_pattern_def_seq (stmt_vinfo, def_stmt);
}
def_stmt
= gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
PLUS_EXPR, oprnd0, signmask);
append_pattern_def_seq (stmt_vinfo, def_stmt);
def_stmt
= gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
BIT_AND_EXPR, gimple_assign_lhs (def_stmt),
fold_build2 (MINUS_EXPR, itype, oprnd1,
build_int_cst (itype, 1)));
append_pattern_def_seq (stmt_vinfo, def_stmt);
pattern_stmt
= gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
MINUS_EXPR, gimple_assign_lhs (def_stmt),
signmask);
}
return pattern_stmt;
}
if (prec > HOST_BITS_PER_WIDE_INT
|| integer_zerop (oprnd1))
return NULL;
if (!can_mult_highpart_p (TYPE_MODE (vectype), TYPE_UNSIGNED (itype)))
return NULL;
if (TYPE_UNSIGNED (itype))
{
unsigned HOST_WIDE_INT mh, ml;
int pre_shift, post_shift;
unsigned HOST_WIDE_INT d = (TREE_INT_CST_LOW (oprnd1)
& GET_MODE_MASK (itype_mode));
tree t1, t2, t3, t4;
if (d >= (HOST_WIDE_INT_1U << (prec - 1)))
/* FIXME: Can transform this into oprnd0 >= oprnd1 ? 1 : 0. */
return NULL;
/* Find a suitable multiplier and right shift count
instead of multiplying with D. */
mh = choose_multiplier (d, prec, prec, &ml, &post_shift, &dummy_int);
/* 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 = ctz_or_zero (d);
mh = choose_multiplier (d >> pre_shift, prec, prec - pre_shift,
&ml, &post_shift, &dummy_int);
gcc_assert (!mh);
}
else
pre_shift = 0;
if (mh != 0)
{
if (post_shift - 1 >= prec)
return NULL;
/* t1 = oprnd0 h* ml;
t2 = oprnd0 - t1;
t3 = t2 >> 1;
t4 = t1 + t3;
q = t4 >> (post_shift - 1); */
t1 = vect_recog_temp_ssa_var (itype, NULL);
def_stmt = gimple_build_assign (t1, MULT_HIGHPART_EXPR, oprnd0,
build_int_cst (itype, ml));
append_pattern_def_seq (stmt_vinfo, def_stmt);
t2 = vect_recog_temp_ssa_var (itype, NULL);
def_stmt
= gimple_build_assign (t2, MINUS_EXPR, oprnd0, t1);
append_pattern_def_seq (stmt_vinfo, def_stmt);
t3 = vect_recog_temp_ssa_var (itype, NULL);
def_stmt
= gimple_build_assign (t3, RSHIFT_EXPR, t2, integer_one_node);
append_pattern_def_seq (stmt_vinfo, def_stmt);
t4 = vect_recog_temp_ssa_var (itype, NULL);
def_stmt
= gimple_build_assign (t4, PLUS_EXPR, t1, t3);
if (post_shift != 1)
{
append_pattern_def_seq (stmt_vinfo, def_stmt);
q = vect_recog_temp_ssa_var (itype, NULL);
pattern_stmt
= gimple_build_assign (q, RSHIFT_EXPR, t4,
build_int_cst (itype, post_shift - 1));
}
else
{
q = t4;
pattern_stmt = def_stmt;
}
}
else
{
if (pre_shift >= prec || post_shift >= prec)
return NULL;
/* t1 = oprnd0 >> pre_shift;
t2 = t1 h* ml;
q = t2 >> post_shift; */
if (pre_shift)
{
t1 = vect_recog_temp_ssa_var (itype, NULL);
def_stmt
= gimple_build_assign (t1, RSHIFT_EXPR, oprnd0,
build_int_cst (NULL, pre_shift));
append_pattern_def_seq (stmt_vinfo, def_stmt);
}
else
t1 = oprnd0;
t2 = vect_recog_temp_ssa_var (itype, NULL);
def_stmt = gimple_build_assign (t2, MULT_HIGHPART_EXPR, t1,
build_int_cst (itype, ml));
if (post_shift)
{
append_pattern_def_seq (stmt_vinfo, def_stmt);
q = vect_recog_temp_ssa_var (itype, NULL);
def_stmt
= gimple_build_assign (q, RSHIFT_EXPR, t2,
build_int_cst (itype, post_shift));
}
else
q = t2;
pattern_stmt = def_stmt;
}
}
else
{
unsigned HOST_WIDE_INT ml;
int post_shift;
HOST_WIDE_INT d = TREE_INT_CST_LOW (oprnd1);
unsigned HOST_WIDE_INT abs_d;
bool add = false;
tree t1, t2, t3, t4;
/* Give up for -1. */
if (d == -1)
return NULL;
/* Since d might be INT_MIN, we have to cast to
unsigned HOST_WIDE_INT before negating to avoid
undefined signed overflow. */
abs_d = (d >= 0
? (unsigned HOST_WIDE_INT) d
: - (unsigned HOST_WIDE_INT) d);
/* n rem d = n rem -d */
if (rhs_code == TRUNC_MOD_EXPR && d < 0)
{
d = abs_d;
oprnd1 = build_int_cst (itype, abs_d);
}
else if (HOST_BITS_PER_WIDE_INT >= prec
&& abs_d == HOST_WIDE_INT_1U << (prec - 1))
/* This case is not handled correctly below. */
return NULL;
choose_multiplier (abs_d, prec, prec - 1, &ml, &post_shift, &dummy_int);
if (ml >= HOST_WIDE_INT_1U << (prec - 1))
{
add = true;
ml |= HOST_WIDE_INT_M1U << (prec - 1);
}
if (post_shift >= prec)
return NULL;
/* t1 = oprnd0 h* ml; */
t1 = vect_recog_temp_ssa_var (itype, NULL);
def_stmt = gimple_build_assign (t1, MULT_HIGHPART_EXPR, oprnd0,
build_int_cst (itype, ml));
if (add)
{
/* t2 = t1 + oprnd0; */
append_pattern_def_seq (stmt_vinfo, def_stmt);
t2 = vect_recog_temp_ssa_var (itype, NULL);
def_stmt = gimple_build_assign (t2, PLUS_EXPR, t1, oprnd0);
}
else
t2 = t1;
if (post_shift)
{
/* t3 = t2 >> post_shift; */
append_pattern_def_seq (stmt_vinfo, def_stmt);
t3 = vect_recog_temp_ssa_var (itype, NULL);
def_stmt = gimple_build_assign (t3, RSHIFT_EXPR, t2,
build_int_cst (itype, post_shift));
}
else
t3 = t2;
wide_int oprnd0_min, oprnd0_max;
int msb = 1;
if (get_range_info (oprnd0, &oprnd0_min, &oprnd0_max) == VR_RANGE)
{
if (!wi::neg_p (oprnd0_min, TYPE_SIGN (itype)))
msb = 0;
else if (wi::neg_p (oprnd0_max, TYPE_SIGN (itype)))
msb = -1;
}
if (msb == 0 && d >= 0)
{
/* q = t3; */
q = t3;
pattern_stmt = def_stmt;
}
else
{
/* t4 = oprnd0 >> (prec - 1);
or if we know from VRP that oprnd0 >= 0
t4 = 0;
or if we know from VRP that oprnd0 < 0
t4 = -1; */
append_pattern_def_seq (stmt_vinfo, def_stmt);
t4 = vect_recog_temp_ssa_var (itype, NULL);
if (msb != 1)
def_stmt = gimple_build_assign (t4, INTEGER_CST,
build_int_cst (itype, msb));
else
def_stmt = gimple_build_assign (t4, RSHIFT_EXPR, oprnd0,
build_int_cst (itype, prec - 1));
append_pattern_def_seq (stmt_vinfo, def_stmt);
/* q = t3 - t4; or q = t4 - t3; */
q = vect_recog_temp_ssa_var (itype, NULL);
pattern_stmt = gimple_build_assign (q, MINUS_EXPR, d < 0 ? t4 : t3,
d < 0 ? t3 : t4);
}
}
if (rhs_code == TRUNC_MOD_EXPR)
{
tree r, t1;
/* We divided. Now finish by:
t1 = q * oprnd1;
r = oprnd0 - t1; */
append_pattern_def_seq (stmt_vinfo, pattern_stmt);
t1 = vect_recog_temp_ssa_var (itype, NULL);
def_stmt = gimple_build_assign (t1, MULT_EXPR, q, oprnd1);
append_pattern_def_seq (stmt_vinfo, def_stmt);
r = vect_recog_temp_ssa_var (itype, NULL);
pattern_stmt = gimple_build_assign (r, MINUS_EXPR, oprnd0, t1);
}
/* Pattern detected. */
vect_pattern_detected ("vect_recog_divmod_pattern", last_stmt);
*type_out = vectype;
return pattern_stmt;
}
/* Function vect_recog_mixed_size_cond_pattern
Try to find the following pattern:
type x_t, y_t;
TYPE a_T, b_T, c_T;
loop:
S1 a_T = x_t CMP y_t ? b_T : c_T;
where type 'TYPE' is an integral type which has different size
from 'type'. b_T and c_T are either constants (and if 'TYPE' is wider
than 'type', the constants need to fit into an integer type
with the same width as 'type') or results of conversion from 'type'.
Input:
* STMT_VINFO: The stmt from which the pattern search begins.
Output:
* TYPE_OUT: The type of the output of this pattern.
* Return value: A new stmt that will be used to replace the pattern.
Additionally a def_stmt is added.
a_it = x_t CMP y_t ? b_it : c_it;
a_T = (TYPE) a_it; */
static gimple *
vect_recog_mixed_size_cond_pattern (stmt_vec_info stmt_vinfo, tree *type_out)
{
vec_info *vinfo = stmt_vinfo->vinfo;
gimple *last_stmt = stmt_vinfo->stmt;
tree cond_expr, then_clause, else_clause;
tree type, vectype, comp_vectype, itype = NULL_TREE, vecitype;
gimple *pattern_stmt, *def_stmt;
tree orig_type0 = NULL_TREE, orig_type1 = NULL_TREE;
gimple *def_stmt0 = NULL, *def_stmt1 = NULL;
bool promotion;
tree comp_scalar_type;
if (!is_gimple_assign (last_stmt)
|| gimple_assign_rhs_code (last_stmt) != COND_EXPR
|| STMT_VINFO_DEF_TYPE (stmt_vinfo) != vect_internal_def)
return NULL;
cond_expr = gimple_assign_rhs1 (last_stmt);
then_clause = gimple_assign_rhs2 (last_stmt);
else_clause = gimple_assign_rhs3 (last_stmt);
if (!COMPARISON_CLASS_P (cond_expr))
return NULL;
comp_scalar_type = TREE_TYPE (TREE_OPERAND (cond_expr, 0));
comp_vectype = get_vectype_for_scalar_type (vinfo, comp_scalar_type);
if (comp_vectype == NULL_TREE)
return NULL;
type = gimple_expr_type (last_stmt);
if (types_compatible_p (type, comp_scalar_type)
|| ((TREE_CODE (then_clause) != INTEGER_CST
|| TREE_CODE (else_clause) != INTEGER_CST)
&& !INTEGRAL_TYPE_P (comp_scalar_type))
|| !INTEGRAL_TYPE_P (type))
return NULL;
if ((TREE_CODE (then_clause) != INTEGER_CST
&& !type_conversion_p (then_clause, stmt_vinfo, false, &orig_type0,
&def_stmt0, &promotion))
|| (TREE_CODE (else_clause) != INTEGER_CST
&& !type_conversion_p (else_clause, stmt_vinfo, false, &orig_type1,
&def_stmt1, &promotion)))
return NULL;
if (orig_type0 && orig_type1
&& !types_compatible_p (orig_type0, orig_type1))
return NULL;
if (orig_type0)
{
if (!types_compatible_p (orig_type0, comp_scalar_type))
return NULL;
then_clause = gimple_assign_rhs1 (def_stmt0);
itype = orig_type0;
}
if (orig_type1)
{
if (!types_compatible_p (orig_type1, comp_scalar_type))
return NULL;
else_clause = gimple_assign_rhs1 (def_stmt1);
itype = orig_type1;
}
HOST_WIDE_INT cmp_mode_size
= GET_MODE_UNIT_BITSIZE (TYPE_MODE (comp_vectype));
scalar_int_mode type_mode = SCALAR_INT_TYPE_MODE (type);
if (GET_MODE_BITSIZE (type_mode) == cmp_mode_size)
return NULL;
vectype = get_vectype_for_scalar_type (vinfo, type);
if (vectype == NULL_TREE)
return NULL;
if (expand_vec_cond_expr_p (vectype, comp_vectype, TREE_CODE (cond_expr)))
return NULL;
if (itype == NULL_TREE)
itype = build_nonstandard_integer_type (cmp_mode_size,
TYPE_UNSIGNED (type));
if (itype == NULL_TREE
|| GET_MODE_BITSIZE (SCALAR_TYPE_MODE (itype)) != cmp_mode_size)
return NULL;
vecitype = get_vectype_for_scalar_type (vinfo, itype);
if (vecitype == NULL_TREE)
return NULL;
if (!expand_vec_cond_expr_p (vecitype, comp_vectype, TREE_CODE (cond_expr)))
return NULL;
if (GET_MODE_BITSIZE (type_mode) > cmp_mode_size)
{
if ((TREE_CODE (then_clause) == INTEGER_CST
&& !int_fits_type_p (then_clause, itype))
|| (TREE_CODE (else_clause) == INTEGER_CST
&& !int_fits_type_p (else_clause, itype)))
return NULL;
}
def_stmt = gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
COND_EXPR, unshare_expr (cond_expr),
fold_convert (itype, then_clause),
fold_convert (itype, else_clause));
pattern_stmt = gimple_build_assign (vect_recog_temp_ssa_var (type, NULL),
NOP_EXPR, gimple_assign_lhs (def_stmt));
append_pattern_def_seq (stmt_vinfo, def_stmt, vecitype);
*type_out = vectype;
vect_pattern_detected ("vect_recog_mixed_size_cond_pattern", last_stmt);
return pattern_stmt;
}
/* Helper function of vect_recog_bool_pattern. Called recursively, return
true if bool VAR can and should be optimized that way. Assume it shouldn't
in case it's a result of a comparison which can be directly vectorized into
a vector comparison. Fills in STMTS with all stmts visited during the
walk. */
static bool
check_bool_pattern (tree var, vec_info *vinfo, hash_set<gimple *> &stmts)
{
tree rhs1;
enum tree_code rhs_code;
stmt_vec_info def_stmt_info = vect_get_internal_def (vinfo, var);
if (!def_stmt_info)
return false;
gassign *def_stmt = dyn_cast <gassign *> (def_stmt_info->stmt);
if (!def_stmt)
return false;
if (stmts.contains (def_stmt))
return true;
rhs1 = gimple_assign_rhs1 (def_stmt);
rhs_code = gimple_assign_rhs_code (def_stmt);
switch (rhs_code)
{
case SSA_NAME:
if (! check_bool_pattern (rhs1, vinfo, stmts))
return false;
break;
CASE_CONVERT:
if (!VECT_SCALAR_BOOLEAN_TYPE_P (TREE_TYPE (rhs1)))
return false;
if (! check_bool_pattern (rhs1, vinfo, stmts))
return false;
break;
case BIT_NOT_EXPR:
if (! check_bool_pattern (rhs1, vinfo, stmts))
return false;
break;
case BIT_AND_EXPR:
case BIT_IOR_EXPR:
case BIT_XOR_EXPR:
if (! check_bool_pattern (rhs1, vinfo, stmts)
|| ! check_bool_pattern (gimple_assign_rhs2 (def_stmt), vinfo, stmts))
return false;
break;
default:
if (TREE_CODE_CLASS (rhs_code) == tcc_comparison)
{
tree vecitype, comp_vectype;
/* If the comparison can throw, then is_gimple_condexpr will be
false and we can't make a COND_EXPR/VEC_COND_EXPR out of it. */
if (stmt_could_throw_p (cfun, def_stmt))
return false;
comp_vectype = get_vectype_for_scalar_type (vinfo, TREE_TYPE (rhs1));
if (comp_vectype == NULL_TREE)
return false;
tree mask_type = get_mask_type_for_scalar_type (vinfo,
TREE_TYPE (rhs1));
if (mask_type
&& expand_vec_cmp_expr_p (comp_vectype, mask_type, rhs_code))
return false;
if (TREE_CODE (TREE_TYPE (rhs1)) != INTEGER_TYPE)
{
scalar_mode mode = SCALAR_TYPE_MODE (TREE_TYPE (rhs1));
tree itype
= build_nonstandard_integer_type (GET_MODE_BITSIZE (mode), 1);
vecitype = get_vectype_for_scalar_type (vinfo, itype);
if (vecitype == NULL_TREE)
return false;
}
else
vecitype = comp_vectype;
if (! expand_vec_cond_expr_p (vecitype, comp_vectype, rhs_code))
return false;
}
else
return false;
break;
}
bool res = stmts.add (def_stmt);
/* We can't end up recursing when just visiting SSA defs but not PHIs. */
gcc_assert (!res);
return true;
}
/* Helper function of adjust_bool_pattern. Add a cast to TYPE to a previous
stmt (SSA_NAME_DEF_STMT of VAR) adding a cast to STMT_INFOs
pattern sequence. */
static tree
adjust_bool_pattern_cast (tree type, tree var, stmt_vec_info stmt_info)
{
vec_info *vinfo = stmt_info->vinfo;
gimple *cast_stmt = gimple_build_assign (vect_recog_temp_ssa_var (type, NULL),
NOP_EXPR, var);
append_pattern_def_seq (stmt_info, cast_stmt,
get_vectype_for_scalar_type (vinfo, type));
return gimple_assign_lhs (cast_stmt);
}
/* Helper function of vect_recog_bool_pattern. Do the actual transformations.
VAR is an SSA_NAME that should be transformed from bool to a wider integer
type, OUT_TYPE is the desired final integer type of the whole pattern.
STMT_INFO is the info of the pattern root and is where pattern stmts should
be associated with. DEFS is a map of pattern defs. */
static void
adjust_bool_pattern (tree var, tree out_type,
stmt_vec_info stmt_info, hash_map <tree, tree> &defs)
{
vec_info *vinfo = stmt_info->vinfo;
gimple *stmt = SSA_NAME_DEF_STMT (var);
enum tree_code rhs_code, def_rhs_code;
tree itype, cond_expr, rhs1, rhs2, irhs1, irhs2;
location_t loc;
gimple *pattern_stmt, *def_stmt;
tree trueval = NULL_TREE;
rhs1 = gimple_assign_rhs1 (stmt);
rhs2 = gimple_assign_rhs2 (stmt);
rhs_code = gimple_assign_rhs_code (stmt);
loc = gimple_location (stmt);
switch (rhs_code)
{
case SSA_NAME:
CASE_CONVERT:
irhs1 = *defs.get (rhs1);
itype = TREE_TYPE (irhs1);
pattern_stmt
= gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
SSA_NAME, irhs1);
break;
case BIT_NOT_EXPR:
irhs1 = *defs.get (rhs1);
itype = TREE_TYPE (irhs1);
pattern_stmt
= gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
BIT_XOR_EXPR, irhs1, build_int_cst (itype, 1));
break;
case BIT_AND_EXPR:
/* Try to optimize x = y & (a < b ? 1 : 0); into
x = (a < b ? y : 0);
E.g. for:
bool a_b, b_b, c_b;
TYPE d_T;
S1 a_b = x1 CMP1 y1;
S2 b_b = x2 CMP2 y2;
S3 c_b = a_b & b_b;
S4 d_T = (TYPE) c_b;
we would normally emit:
S1' a_T = x1 CMP1 y1 ? 1 : 0;
S2' b_T = x2 CMP2 y2 ? 1 : 0;
S3' c_T = a_T & b_T;
S4' d_T = c_T;
but we can save one stmt by using the
result of one of the COND_EXPRs in the other COND_EXPR and leave
BIT_AND_EXPR stmt out:
S1' a_T = x1 CMP1 y1 ? 1 : 0;
S3' c_T = x2 CMP2 y2 ? a_T : 0;
S4' f_T = c_T;
At least when VEC_COND_EXPR is implemented using masks
cond ? 1 : 0 is as expensive as cond ? var : 0, in both cases it
computes the comparison masks and ands it, in one case with
all ones vector, in the other case with a vector register.
Don't do this for BIT_IOR_EXPR, because cond ? 1 : var; is
often more expensive. */
def_stmt = SSA_NAME_DEF_STMT (rhs2);
def_rhs_code = gimple_assign_rhs_code (def_stmt);
if (TREE_CODE_CLASS (def_rhs_code) == tcc_comparison)
{
irhs1 = *defs.get (rhs1);
tree def_rhs1 = gimple_assign_rhs1 (def_stmt);
if (TYPE_PRECISION (TREE_TYPE (irhs1))
== GET_MODE_BITSIZE (SCALAR_TYPE_MODE (TREE_TYPE (def_rhs1))))
{
rhs_code = def_rhs_code;
rhs1 = def_rhs1;
rhs2 = gimple_assign_rhs2 (def_stmt);
trueval = irhs1;
goto do_compare;
}
else
irhs2 = *defs.get (rhs2);
goto and_ior_xor;
}
def_stmt = SSA_NAME_DEF_STMT (rhs1);
def_rhs_code = gimple_assign_rhs_code (def_stmt);
if (TREE_CODE_CLASS (def_rhs_code) == tcc_comparison)
{
irhs2 = *defs.get (rhs2);
tree def_rhs1 = gimple_assign_rhs1 (def_stmt);
if (TYPE_PRECISION (TREE_TYPE (irhs2))
== GET_MODE_BITSIZE (SCALAR_TYPE_MODE (TREE_TYPE (def_rhs1))))
{
rhs_code = def_rhs_code;
rhs1 = def_rhs1;
rhs2 = gimple_assign_rhs2 (def_stmt);
trueval = irhs2;
goto do_compare;
}
else
irhs1 = *defs.get (rhs1);
goto and_ior_xor;
}
/* FALLTHRU */
case BIT_IOR_EXPR:
case BIT_XOR_EXPR:
irhs1 = *defs.get (rhs1);
irhs2 = *defs.get (rhs2);
and_ior_xor:
if (TYPE_PRECISION (TREE_TYPE (irhs1))
!= TYPE_PRECISION (TREE_TYPE (irhs2)))
{
int prec1 = TYPE_PRECISION (TREE_TYPE (irhs1));
int prec2 = TYPE_PRECISION (TREE_TYPE (irhs2));
int out_prec = TYPE_PRECISION (out_type);
if (absu_hwi (out_prec - prec1) < absu_hwi (out_prec - prec2))
irhs2 = adjust_bool_pattern_cast (TREE_TYPE (irhs1), irhs2,
stmt_info);
else if (absu_hwi (out_prec - prec1) > absu_hwi (out_prec - prec2))
irhs1 = adjust_bool_pattern_cast (TREE_TYPE (irhs2), irhs1,
stmt_info);
else
{
irhs1 = adjust_bool_pattern_cast (out_type, irhs1, stmt_info);
irhs2 = adjust_bool_pattern_cast (out_type, irhs2, stmt_info);
}
}
itype = TREE_TYPE (irhs1);
pattern_stmt
= gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
rhs_code, irhs1, irhs2);
break;
default:
do_compare:
gcc_assert (TREE_CODE_CLASS (rhs_code) == tcc_comparison);
if (TREE_CODE (TREE_TYPE (rhs1)) != INTEGER_TYPE
|| !TYPE_UNSIGNED (TREE_TYPE (rhs1))
|| maybe_ne (TYPE_PRECISION (TREE_TYPE (rhs1)),
GET_MODE_BITSIZE (TYPE_MODE (TREE_TYPE (rhs1)))))
{
scalar_mode mode = SCALAR_TYPE_MODE (TREE_TYPE (rhs1));
itype
= build_nonstandard_integer_type (GET_MODE_BITSIZE (mode), 1);
}
else
itype = TREE_TYPE (rhs1);
cond_expr = build2_loc (loc, rhs_code, itype, rhs1, rhs2);
if (trueval == NULL_TREE)
trueval = build_int_cst (itype, 1);
else
gcc_checking_assert (useless_type_conversion_p (itype,
TREE_TYPE (trueval)));
pattern_stmt
= gimple_build_assign (vect_recog_temp_ssa_var (itype, NULL),
COND_EXPR, cond_expr, trueval,
build_int_cst (itype, 0));
break;
}
gimple_set_location (pattern_stmt, loc);
append_pattern_def_seq (stmt_info, pattern_stmt,
get_vectype_for_scalar_type (vinfo, itype));
defs.put (var, gimple_assign_lhs (pattern_stmt));
}
/* Comparison function to qsort a vector of gimple stmts after UID. */
static int
sort_after_uid (const void *p1, const void *p2)
{
const gimple *stmt1 = *(const gimple * const *)p1;
const gimple *stmt2 = *(const gimple * const *)p2;
return gimple_uid (stmt1) - gimple_uid (stmt2);
}
/* Create pattern stmts for all stmts participating in the bool pattern
specified by BOOL_STMT_SET and its root STMT_INFO with the desired type
OUT_TYPE. Return the def of the pattern root. */
static tree
adjust_bool_stmts (hash_set <gimple *> &bool_stmt_set,
tree out_type, stmt_vec_info stmt_info)
{
/* Gather original stmts in the bool pattern in their order of appearance
in the IL. */
auto_vec<gimple *> bool_stmts (bool_stmt_set.elements ());
for (hash_set <gimple *>::iterator i = bool_stmt_set.begin ();
i != bool_stmt_set.end (); ++i)
bool_stmts.quick_push (*i);
bool_stmts.qsort (sort_after_uid);
/* Now process them in that order, producing pattern stmts. */
hash_map <tree, tree> defs;
for (unsigned i = 0; i < bool_stmts.length (); ++i)
adjust_bool_pattern (gimple_assign_lhs (bool_stmts[i]),
out_type, stmt_info, defs);
/* Pop the last pattern seq stmt and install it as pattern root for STMT. */
gimple *pattern_stmt
= gimple_seq_last_stmt (STMT_VINFO_PATTERN_DEF_SEQ (stmt_info));
return gimple_assign_lhs (pattern_stmt);
}
/* Helper for search_type_for_mask. */
static tree
search_type_for_mask_1 (tree var, vec_info *vinfo,
hash_map<gimple *, tree> &cache)
{
tree rhs1;
enum tree_code rhs_code;
tree res = NULL_TREE, res2;
if (!VECT_SCALAR_BOOLEAN_TYPE_P (TREE_TYPE (var)))
return NULL_TREE;
stmt_vec_info def_stmt_info = vect_get_internal_def (vinfo, var);
if (!def_stmt_info)
return NULL_TREE;
gassign *def_stmt = dyn_cast <gassign *> (def_stmt_info->stmt);
if (!def_stmt)
return NULL_TREE;
tree *c = cache.get (def_stmt);
if (c)
return *c;
rhs_code = gimple_assign_rhs_code (def_stmt);
rhs1 = gimple_assign_rhs1 (def_stmt);
switch (rhs_code)
{
case SSA_NAME:
case BIT_NOT_EXPR:
CASE_CONVERT:
res = search_type_for_mask_1 (rhs1, vinfo, cache);
break;
case BIT_AND_EXPR:
case BIT_IOR_EXPR:
case BIT_XOR_EXPR:
res = search_type_for_mask_1 (rhs1, vinfo, cache);
res2 = search_type_for_mask_1 (gimple_assign_rhs2 (def_stmt), vinfo,
cache);
if (!res || (res2 && TYPE_PRECISION (res) > TYPE_PRECISION (res2)))
res = res2;
break;
default:
if (TREE_CODE_CLASS (rhs_code) == tcc_comparison)
{
tree comp_vectype, mask_type;
if (VECT_SCALAR_BOOLEAN_TYPE_P (TREE_TYPE (rhs1)))
{
res = search_type_for_mask_1 (rhs1, vinfo, cache);
res2 = search_type_for_mask_1 (gimple_assign_rhs2 (def_stmt),
vinfo, cache);
if (!res || (res2 && TYPE_PRECISION (res) > TYPE_PRECISION (res2)))
res = res2;
break;
}
comp_vectype = get_vectype_for_scalar_type (vinfo, TREE_TYPE (rhs1));
if (comp_vectype == NULL_TREE)
{
res = NULL_TREE;
break;
}
mask_type = get_mask_type_for_scalar_type (vinfo, TREE_TYPE (rhs1));
if (!mask_type
|| !expand_vec_cmp_expr_p (comp_vectype, mask_type, rhs_code))
{
res = NULL_TREE;
break;
}
if (TREE_CODE (TREE_TYPE (rhs1)) != INTEGER_TYPE
|| !TYPE_UNSIGNED (TREE_TYPE (rhs1)))
{
scalar_mode mode = SCALAR_TYPE_MODE (TREE_TYPE (rhs1));
res = build_nonstandard_integer_type (GET_MODE_BITSIZE (mode), 1);
}
else
res = TREE_TYPE (rhs1);
}
}
cache.put (def_stmt, res);
return res;
}
/* Return the proper type for converting bool VAR into
an integer value or NULL_TREE if no such type exists.
The type is chosen so that converted value has the
same number of elements as VAR's vector type. */
static tree
search_type_for_mask (tree var, vec_info *vinfo)
{
hash_map<gimple *, tree> cache;
return search_type_for_mask_1 (var, vinfo, cache);
}
/* Function vect_recog_bool_pattern
Try to find pattern like following:
bool a_b, b_b, c_b, d_b, e_b;
TYPE f_T;
loop:
S1 a_b = x1 CMP1 y1;
S2 b_b = x2 CMP2 y2;
S3 c_b = a_b & b_b;
S4 d_b = x3 CMP3 y3;
S5 e_b = c_b | d_b;
S6 f_T = (TYPE) e_b;
where type 'TYPE' is an integral type. Or a similar pattern
ending in
S6 f_Y = e_b ? r_Y : s_Y;
as results from if-conversion of a complex condition.
Input:
* STMT_VINFO: The stmt at the end from which the pattern
search begins, i.e. cast of a bool to
an integer type.
Output:
* TYPE_OUT: The type of the output of this pattern.
* Return value: A new stmt that will be used to replace the pattern.
Assuming size of TYPE is the same as size of all comparisons
(otherwise some casts would be added where needed), the above
sequence we create related pattern stmts:
S1' a_T = x1 CMP1 y1 ? 1 : 0;
S3' c_T = x2 CMP2 y2 ? a_T : 0;
S4' d_T = x3 CMP3 y3 ? 1 : 0;
S5' e_T = c_T | d_T;
S6' f_T = e_T;
Instead of the above S3' we could emit:
S2' b_T = x2 CMP2 y2 ? 1 : 0;
S3' c_T = a_T | b_T;
but the above is more efficient. */
static gimple *
vect_recog_bool_pattern (stmt_vec_info stmt_vinfo, tree *type_out)
{
gimple *last_stmt = stmt_vinfo->stmt;
enum tree_code rhs_code;
tree var, lhs, rhs, vectype;
vec_info *vinfo = stmt_vinfo->vinfo;
gimple *pattern_stmt;
if (!is_gimple_assign (last_stmt))
return NULL;
var = gimple_assign_rhs1 (last_stmt);
lhs = gimple_assign_lhs (last_stmt);
if (!VECT_SCALAR_BOOLEAN_TYPE_P (TREE_TYPE (var)))
return NULL;
hash_set<gimple *> bool_stmts;
rhs_code = gimple_assign_rhs_code (last_stmt);
if (CONVERT_EXPR_CODE_P (rhs_code))
{
if (! INTEGRAL_TYPE_P (TREE_TYPE (lhs))
|| TYPE_PRECISION (TREE_TYPE (lhs)) == 1)
return NULL;
vectype = get_vectype_for_scalar_type (vinfo, TREE_TYPE (lhs));
if (vectype == NULL_TREE)
return NULL;
if (check_bool_pattern (var, vinfo, bool_stmts))
{
rhs = adjust_bool_stmts (bool_stmts, TREE_TYPE (lhs), stmt_vinfo);
lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
if (useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (rhs)))
pattern_stmt = gimple_build_assign (lhs, SSA_NAME, rhs);
else
pattern_stmt
= gimple_build_assign (lhs, NOP_EXPR, rhs);
}
else
{
tree type = search_type_for_mask (var, vinfo);
tree cst0, cst1, tmp;
if (!type)
return NULL;
/* We may directly use cond with narrowed type to avoid
multiple cond exprs with following result packing and
perform single cond with packed mask instead. In case
of widening we better make cond first and then extract
results. */
if (TYPE_MODE (type) == TYPE_MODE (TREE_TYPE (lhs)))
type = TREE_TYPE (lhs);
cst0 = build_int_cst (type, 0);
cst1 = build_int_cst (type, 1);
tmp = vect_recog_temp_ssa_var (type, NULL);
pattern_stmt = gimple_build_assign (tmp, COND_EXPR, var, cst1, cst0);
if (!useless_type_conversion_p (type, TREE_TYPE (lhs)))
{
tree new_vectype = get_vectype_for_scalar_type (vinfo, type);
append_pattern_def_seq (stmt_vinfo, pattern_stmt, new_vectype);
lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
pattern_stmt = gimple_build_assign (lhs, CONVERT_EXPR, tmp);
}
}
*type_out = vectype;
vect_pattern_detected ("vect_recog_bool_pattern", last_stmt);
return pattern_stmt;
}
else if (rhs_code == COND_EXPR
&& TREE_CODE (var) == SSA_NAME)
{
vectype = get_vectype_for_scalar_type (vinfo, TREE_TYPE (lhs));
if (vectype == NULL_TREE)
return NULL;
/* Build a scalar type for the boolean result that when
vectorized matches the vector type of the result in
size and number of elements. */
unsigned prec
= vector_element_size (tree_to_poly_uint64 (TYPE_SIZE (vectype)),
TYPE_VECTOR_SUBPARTS (vectype));
tree type
= build_nonstandard_integer_type (prec,
TYPE_UNSIGNED (TREE_TYPE (var)));
if (get_vectype_for_scalar_type (vinfo, type) == NULL_TREE)
return NULL;
if (!check_bool_pattern (var, vinfo, bool_stmts))
return NULL;
rhs = adjust_bool_stmts (bool_stmts, type, stmt_vinfo);
lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
pattern_stmt
= gimple_build_assign (lhs, COND_EXPR,
build2 (NE_EXPR, boolean_type_node,
rhs, build_int_cst (type, 0)),
gimple_assign_rhs2 (last_stmt),
gimple_assign_rhs3 (last_stmt));
*type_out = vectype;
vect_pattern_detected ("vect_recog_bool_pattern", last_stmt);
return pattern_stmt;
}
else if (rhs_code == SSA_NAME
&& STMT_VINFO_DATA_REF (stmt_vinfo))
{
stmt_vec_info pattern_stmt_info;
vectype = STMT_VINFO_VECTYPE (stmt_vinfo);
gcc_assert (vectype != NULL_TREE);
if (!VECTOR_MODE_P (TYPE_MODE (vectype)))
return NULL;
if (check_bool_pattern (var, vinfo, bool_stmts))
rhs = adjust_bool_stmts (bool_stmts, TREE_TYPE (vectype), stmt_vinfo);
else
{
tree type = search_type_for_mask (var, vinfo);
tree cst0, cst1, new_vectype;
if (!type)
return NULL;
if (TYPE_MODE (type) == TYPE_MODE (TREE_TYPE (vectype)))
type = TREE_TYPE (vectype);
cst0 = build_int_cst (type, 0);
cst1 = build_int_cst (type, 1);
new_vectype = get_vectype_for_scalar_type (vinfo, type);
rhs = vect_recog_temp_ssa_var (type, NULL);
pattern_stmt = gimple_build_assign (rhs, COND_EXPR, var, cst1, cst0);
append_pattern_def_seq (stmt_vinfo, pattern_stmt, new_vectype);
}
lhs = build1 (VIEW_CONVERT_EXPR, TREE_TYPE (vectype), lhs);
if (!useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (rhs)))
{
tree rhs2 = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
gimple *cast_stmt = gimple_build_assign (rhs2, NOP_EXPR, rhs);
append_pattern_def_seq (stmt_vinfo, cast_stmt);
rhs = rhs2;
}
pattern_stmt = gimple_build_assign (lhs, SSA_NAME, rhs);
pattern_stmt_info = vinfo->add_stmt (pattern_stmt);
vinfo->move_dr (pattern_stmt_info, stmt_vinfo);
*type_out = vectype;
vect_pattern_detected ("vect_recog_bool_pattern", last_stmt);
return pattern_stmt;
}
else
return NULL;
}
/* A helper for vect_recog_mask_conversion_pattern. Build
conversion of MASK to a type suitable for masking VECTYPE.
Built statement gets required vectype and is appended to
a pattern sequence of STMT_VINFO.
Return converted mask. */
static tree
build_mask_conversion (tree mask, tree vectype, stmt_vec_info stmt_vinfo)
{
gimple *stmt;
tree masktype, tmp;
masktype = truth_type_for (vectype);
tmp = vect_recog_temp_ssa_var (TREE_TYPE (masktype), NULL);
stmt = gimple_build_assign (tmp, CONVERT_EXPR, mask);
append_pattern_def_seq (stmt_vinfo, stmt, masktype);
return tmp;
}
/* Function vect_recog_mask_conversion_pattern
Try to find statements which require boolean type
converison. Additional conversion statements are
added to handle such cases. For example:
bool m_1, m_2, m_3;
int i_4, i_5;
double d_6, d_7;
char c_1, c_2, c_3;
S1 m_1 = i_4 > i_5;
S2 m_2 = d_6 < d_7;
S3 m_3 = m_1 & m_2;
S4 c_1 = m_3 ? c_2 : c_3;
Will be transformed into:
S1 m_1 = i_4 > i_5;
S2 m_2 = d_6 < d_7;
S3'' m_2' = (_Bool[bitsize=32])m_2
S3' m_3' = m_1 & m_2';
S4'' m_3'' = (_Bool[bitsize=8])m_3'
S4' c_1' = m_3'' ? c_2 : c_3; */
static gimple *
vect_recog_mask_conversion_pattern (stmt_vec_info stmt_vinfo, tree *type_out)
{
gimple *last_stmt = stmt_vinfo->stmt;
enum tree_code rhs_code;
tree lhs = NULL_TREE, rhs1, rhs2, tmp, rhs1_type, rhs2_type;
tree vectype1, vectype2;
stmt_vec_info pattern_stmt_info;
vec_info *vinfo = stmt_vinfo->vinfo;
/* Check for MASK_LOAD ans MASK_STORE calls requiring mask conversion. */
if (is_gimple_call (last_stmt)
&& gimple_call_internal_p (last_stmt))
{
gcall *pattern_stmt;
internal_fn ifn = gimple_call_internal_fn (last_stmt);
int mask_argno = internal_fn_mask_index (ifn);
if (mask_argno < 0)
return NULL;
bool store_p = internal_store_fn_p (ifn);
if (store_p)
{
int rhs_index = internal_fn_stored_value_index (ifn);
tree rhs = gimple_call_arg (last_stmt, rhs_index);
vectype1 = get_vectype_for_scalar_type (vinfo, TREE_TYPE (rhs));
}
else
{
lhs = gimple_call_lhs (last_stmt);
vectype1 = get_vectype_for_scalar_type (vinfo, TREE_TYPE (lhs));
}
tree mask_arg = gimple_call_arg (last_stmt, mask_argno);
tree mask_arg_type = search_type_for_mask (mask_arg, vinfo);
if (!mask_arg_type)
return NULL;
vectype2 = get_mask_type_for_scalar_type (vinfo, mask_arg_type);
if (!vectype1 || !vectype2
|| known_eq (TYPE_VECTOR_SUBPARTS (vectype1),
TYPE_VECTOR_SUBPARTS (vectype2)))
return NULL;
tmp = build_mask_conversion (mask_arg, vectype1, stmt_vinfo);
auto_vec<tree, 8> args;
unsigned int nargs = gimple_call_num_args (last_stmt);
args.safe_grow (nargs);
for (unsigned int i = 0; i < nargs; ++i)
args[i] = ((int) i == mask_argno
? tmp
: gimple_call_arg (last_stmt, i));
pattern_stmt = gimple_build_call_internal_vec (ifn, args);
if (!store_p)
{
lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
gimple_call_set_lhs (pattern_stmt, lhs);
}
gimple_call_set_nothrow (pattern_stmt, true);
pattern_stmt_info = vinfo->add_stmt (pattern_stmt);
if (STMT_VINFO_DATA_REF (stmt_vinfo))
vinfo->move_dr (pattern_stmt_info, stmt_vinfo);
*type_out = vectype1;
vect_pattern_detected ("vect_recog_mask_conversion_pattern", last_stmt);
return pattern_stmt;
}
if (!is_gimple_assign (last_stmt))
return NULL;
gimple *pattern_stmt;
lhs = gimple_assign_lhs (last_stmt);
rhs1 = gimple_assign_rhs1 (last_stmt);
rhs_code = gimple_assign_rhs_code (last_stmt);
/* Check for cond expression requiring mask conversion. */
if (rhs_code == COND_EXPR)
{
vectype1 = get_vectype_for_scalar_type (vinfo, TREE_TYPE (lhs));
if (TREE_CODE (rhs1) == SSA_NAME)
{
rhs1_type = search_type_for_mask (rhs1, vinfo);
if (!rhs1_type)
return NULL;
}
else if (COMPARISON_CLASS_P (rhs1))
{
/* Check whether we're comparing scalar booleans and (if so)
whether a better mask type exists than the mask associated
with boolean-sized elements. This avoids unnecessary packs
and unpacks if the booleans are set from comparisons of
wider types. E.g. in:
int x1, x2, x3, x4, y1, y1;
...
bool b1 = (x1 == x2);
bool b2 = (x3 == x4);
... = b1 == b2 ? y1 : y2;
it is better for b1 and b2 to use the mask type associated
with int elements rather bool (byte) elements. */
rhs1_type = search_type_for_mask (TREE_OPERAND (rhs1, 0), vinfo);
if (!rhs1_type)
rhs1_type = TREE_TYPE (TREE_OPERAND (rhs1, 0));
}
else
return NULL;
vectype2 = get_mask_type_for_scalar_type (vinfo, rhs1_type);
if (!vectype1 || !vectype2)
return NULL;
/* Continue if a conversion is needed. Also continue if we have
a comparison whose vector type would normally be different from
VECTYPE2 when considered in isolation. In that case we'll
replace the comparison with an SSA name (so that we can record
its vector type) and behave as though the comparison was an SSA
name from the outset. */
if (known_eq (TYPE_VECTOR_SUBPARTS (vectype1),
TYPE_VECTOR_SUBPARTS (vectype2))
&& (TREE_CODE (rhs1) == SSA_NAME
|| rhs1_type == TREE_TYPE (TREE_OPERAND (rhs1, 0))))
return NULL;
/* If rhs1 is invariant and we can promote it leave the COND_EXPR
in place, we can handle it in vectorizable_condition. This avoids
unnecessary promotion stmts and increased vectorization factor. */
if (COMPARISON_CLASS_P (rhs1)
&& INTEGRAL_TYPE_P (rhs1_type)
&& known_le (TYPE_VECTOR_SUBPARTS (vectype1),
TYPE_VECTOR_SUBPARTS (vectype2)))
{
enum vect_def_type dt;
if (vect_is_simple_use (TREE_OPERAND (rhs1, 0), vinfo, &dt)
&& dt == vect_external_def
&& vect_is_simple_use (TREE_OPERAND (rhs1, 1), vinfo, &dt)
&& (dt == vect_external_def
|| dt == vect_constant_def))
{
tree wide_scalar_type = build_nonstandard_integer_type
(tree_to_uhwi (TYPE_SIZE (TREE_TYPE (vectype1))),
TYPE_UNSIGNED (rhs1_type));
tree vectype3 = get_vectype_for_scalar_type (vinfo,
wide_scalar_type);
if (expand_vec_cond_expr_p (vectype1, vectype3, TREE_CODE (rhs1)))
return NULL;
}
}
/* If rhs1 is a comparison we need to move it into a
separate statement. */
if (TREE_CODE (rhs1) != SSA_NAME)
{
tmp = vect_recog_temp_ssa_var (TREE_TYPE (rhs1), NULL);
pattern_stmt = gimple_build_assign (tmp, rhs1);
rhs1 = tmp;
append_pattern_def_seq (stmt_vinfo, pattern_stmt, vectype2);
}
if (maybe_ne (TYPE_VECTOR_SUBPARTS (vectype1),
TYPE_VECTOR_SUBPARTS (vectype2)))
tmp = build_mask_conversion (rhs1, vectype1, stmt_vinfo);
else
tmp = rhs1;
lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
pattern_stmt = gimple_build_assign (lhs, COND_EXPR, tmp,
gimple_assign_rhs2 (last_stmt),
gimple_assign_rhs3 (last_stmt));
*type_out = vectype1;
vect_pattern_detected ("vect_recog_mask_conversion_pattern", last_stmt);
return pattern_stmt;
}
/* Now check for binary boolean operations requiring conversion for
one of operands. */
if (!VECT_SCALAR_BOOLEAN_TYPE_P (TREE_TYPE (lhs)))
return NULL;
if (rhs_code != BIT_IOR_EXPR
&& rhs_code != BIT_XOR_EXPR
&& rhs_code != BIT_AND_EXPR
&& TREE_CODE_CLASS (rhs_code) != tcc_comparison)
return NULL;
rhs2 = gimple_assign_rhs2 (last_stmt);
rhs1_type = search_type_for_mask (rhs1, vinfo);
rhs2_type = search_type_for_mask (rhs2, vinfo);
if (!rhs1_type || !rhs2_type
|| TYPE_PRECISION (rhs1_type) == TYPE_PRECISION (rhs2_type))
return NULL;
if (TYPE_PRECISION (rhs1_type) < TYPE_PRECISION (rhs2_type))
{
vectype1 = get_mask_type_for_scalar_type (vinfo, rhs1_type);
if (!vectype1)
return NULL;
rhs2 = build_mask_conversion (rhs2, vectype1, stmt_vinfo);
}
else
{
vectype1 = get_mask_type_for_scalar_type (vinfo, rhs2_type);
if (!vectype1)
return NULL;
rhs1 = build_mask_conversion (rhs1, vectype1, stmt_vinfo);
}
lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
pattern_stmt = gimple_build_assign (lhs, rhs_code, rhs1, rhs2);
*type_out = vectype1;
vect_pattern_detected ("vect_recog_mask_conversion_pattern", last_stmt);
return pattern_stmt;
}
/* STMT_INFO is a load or store. If the load or store is conditional, return
the boolean condition under which it occurs, otherwise return null. */
static tree
vect_get_load_store_mask (stmt_vec_info stmt_info)
{
if (gassign *def_assign = dyn_cast <gassign *> (stmt_info->stmt))
{
gcc_assert (gimple_assign_single_p (def_assign));
return NULL_TREE;
}
if (gcall *def_call = dyn_cast <gcall *> (stmt_info->stmt))
{
internal_fn ifn = gimple_call_internal_fn (def_call);
int mask_index = internal_fn_mask_index (ifn);
return gimple_call_arg (def_call, mask_index);
}
gcc_unreachable ();
}
/* Return MASK if MASK is suitable for masking an operation on vectors
of type VECTYPE, otherwise convert it into such a form and return
the result. Associate any conversion statements with STMT_INFO's
pattern. */
static tree
vect_convert_mask_for_vectype (tree mask, tree vectype,
stmt_vec_info stmt_info, vec_info *vinfo)
{
tree mask_type = search_type_for_mask (mask, vinfo);
if (mask_type)
{
tree mask_vectype = get_mask_type_for_scalar_type (vinfo, mask_type);
if (mask_vectype
&& maybe_ne (TYPE_VECTOR_SUBPARTS (vectype),
TYPE_VECTOR_SUBPARTS (mask_vectype)))
mask = build_mask_conversion (mask, vectype, stmt_info);
}
return mask;
}
/* Return the equivalent of:
fold_convert (TYPE, VALUE)
with the expectation that the operation will be vectorized.
If new statements are needed, add them as pattern statements
to STMT_INFO. */
static tree
vect_add_conversion_to_pattern (tree type, tree value, stmt_vec_info stmt_info)
{
if (useless_type_conversion_p (type, TREE_TYPE (value)))
return value;
vec_info *vinfo = stmt_info->vinfo;
tree new_value = vect_recog_temp_ssa_var (type, NULL);
gassign *conversion = gimple_build_assign (new_value, CONVERT_EXPR, value);
append_pattern_def_seq (stmt_info, conversion,
get_vectype_for_scalar_type (vinfo, type));
return new_value;
}
/* Try to convert STMT_INFO into a call to a gather load or scatter store
internal function. Return the final statement on success and set
*TYPE_OUT to the vector type being loaded or stored.
This function only handles gathers and scatters that were recognized
as such from the outset (indicated by STMT_VINFO_GATHER_SCATTER_P). */
static gimple *
vect_recog_gather_scatter_pattern (stmt_vec_info stmt_info, tree *type_out)
{
/* Currently we only support this for loop vectorization. */
loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (stmt_info->vinfo);
if (!loop_vinfo)
return NULL;
/* Make sure that we're looking at a gather load or scatter store. */
data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
if (!dr || !STMT_VINFO_GATHER_SCATTER_P (stmt_info))
return NULL;
/* Get the boolean that controls whether the load or store happens.
This is null if the operation is unconditional. */
tree mask = vect_get_load_store_mask (stmt_info);
/* Make sure that the target supports an appropriate internal
function for the gather/scatter operation. */
gather_scatter_info gs_info;
if (!vect_check_gather_scatter (stmt_info, loop_vinfo, &gs_info)
|| gs_info.decl)
return NULL;
/* Convert the mask to the right form. */
tree gs_vectype = get_vectype_for_scalar_type (loop_vinfo,
gs_info.element_type);
if (mask)
mask = vect_convert_mask_for_vectype (mask, gs_vectype, stmt_info,
loop_vinfo);
/* Get the invariant base and non-invariant offset, converting the
latter to the same width as the vector elements. */
tree base = gs_info.base;
tree offset_type = TREE_TYPE (gs_info.offset_vectype);
tree offset = vect_add_conversion_to_pattern (offset_type, gs_info.offset,
stmt_info);
/* Build the new pattern statement. */
tree scale = size_int (gs_info.scale);
gcall *pattern_stmt;
if (DR_IS_READ (dr))
{
tree zero = build_zero_cst (gs_info.element_type);
if (mask != NULL)
pattern_stmt = gimple_build_call_internal (gs_info.ifn, 5, base,
offset, scale, zero, mask);
else
pattern_stmt = gimple_build_call_internal (gs_info.ifn, 4, base,
offset, scale, zero);
tree load_lhs = vect_recog_temp_ssa_var (gs_info.element_type, NULL);
gimple_call_set_lhs (pattern_stmt, load_lhs);
}
else
{
tree rhs = vect_get_store_rhs (stmt_info);
if (mask != NULL)
pattern_stmt = gimple_build_call_internal (IFN_MASK_SCATTER_STORE, 5,
base, offset, scale, rhs,
mask);
else
pattern_stmt = gimple_build_call_internal (IFN_SCATTER_STORE, 4,
base, offset, scale, rhs);
}
gimple_call_set_nothrow (pattern_stmt, true);
/* Copy across relevant vectorization info and associate DR with the
new pattern statement instead of the original statement. */
stmt_vec_info pattern_stmt_info = loop_vinfo->add_stmt (pattern_stmt);
loop_vinfo->move_dr (pattern_stmt_info, stmt_info);
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
*type_out = vectype;
vect_pattern_detected ("gather/scatter pattern", stmt_info->stmt);
return pattern_stmt;
}
/* Return true if TYPE is a non-boolean integer type. These are the types
that we want to consider for narrowing. */
static bool
vect_narrowable_type_p (tree type)
{
return INTEGRAL_TYPE_P (type) && !VECT_SCALAR_BOOLEAN_TYPE_P (type);
}
/* Return true if the operation given by CODE can be truncated to N bits
when only N bits of the output are needed. This is only true if bit N+1
of the inputs has no effect on the low N bits of the result. */
static bool
vect_truncatable_operation_p (tree_code code)
{
switch (code)
{
case PLUS_EXPR:
case MINUS_EXPR:
case MULT_EXPR:
case BIT_AND_EXPR:
case BIT_IOR_EXPR:
case BIT_XOR_EXPR:
case COND_EXPR:
return true;
default:
return false;
}
}
/* Record that STMT_INFO could be changed from operating on TYPE to
operating on a type with the precision and sign given by PRECISION
and SIGN respectively. PRECISION is an arbitrary bit precision;
it might not be a whole number of bytes. */
static void
vect_set_operation_type (stmt_vec_info stmt_info, tree type,
unsigned int precision, signop sign)
{
/* Round the precision up to a whole number of bytes. */
precision = vect_element_precision (precision);
if (precision < TYPE_PRECISION (type)
&& (!stmt_info->operation_precision
|| stmt_info->operation_precision > precision))
{
stmt_info->operation_precision = precision;
stmt_info->operation_sign = sign;
}
}
/* Record that STMT_INFO only requires MIN_INPUT_PRECISION from its
non-boolean inputs, all of which have type TYPE. MIN_INPUT_PRECISION
is an arbitrary bit precision; it might not be a whole number of bytes. */
static void
vect_set_min_input_precision (stmt_vec_info stmt_info, tree type,
unsigned int min_input_precision)
{
/* This operation in isolation only requires the inputs to have
MIN_INPUT_PRECISION of precision, However, that doesn't mean
that MIN_INPUT_PRECISION is a natural precision for the chain
as a whole. E.g. consider something like:
unsigned short *x, *y;
*y = ((*x & 0xf0) >> 4) | (*y << 4);
The right shift can be done on unsigned chars, and only requires the
result of "*x & 0xf0" to be done on unsigned chars. But taking that
approach would mean turning a natural chain of single-vector unsigned
short operations into one that truncates "*x" and then extends
"(*x & 0xf0) >> 4", with two vectors for each unsigned short
operation and one vector for each unsigned char operation.
This would be a significant pessimization.
Instead only propagate the maximum of this precision and the precision
required by the users of the result. This means that we don't pessimize
the case above but continue to optimize things like:
unsigned char *y;
unsigned short *x;
*y = ((*x & 0xf0) >> 4) | (*y << 4);
Here we would truncate two vectors of *x to a single vector of
unsigned chars and use single-vector unsigned char operations for
everything else, rather than doing two unsigned short copies of
"(*x & 0xf0) >> 4" and then truncating the result. */
min_input_precision = MAX (min_input_precision,
stmt_info->min_output_precision);
if (min_input_precision < TYPE_PRECISION (type)
&& (!stmt_info->min_input_precision
|| stmt_info->min_input_precision > min_input_precision))
stmt_info->min_input_precision = min_input_precision;
}
/* Subroutine of vect_determine_min_output_precision. Return true if
we can calculate a reduced number of output bits for STMT_INFO,
whose result is LHS. */
static bool
vect_determine_min_output_precision_1 (stmt_vec_info stmt_info, tree lhs)
{
/* Take the maximum precision required by users of the result. */
vec_info *vinfo = stmt_info->vinfo;
unsigned int precision = 0;
imm_use_iterator iter;
use_operand_p use;
FOR_EACH_IMM_USE_FAST (use, iter, lhs)
{
gimple *use_stmt = USE_STMT (use);
if (is_gimple_debug (use_stmt))
continue;
stmt_vec_info use_stmt_info = vinfo->lookup_stmt (use_stmt);
if (!use_stmt_info || !use_stmt_info->min_input_precision)
return false;
/* The input precision recorded for COND_EXPRs applies only to the
"then" and "else" values. */
gassign *assign = dyn_cast <gassign *> (stmt_info->stmt);
if (assign
&& gimple_assign_rhs_code (assign) == COND_EXPR
&& use->use != gimple_assign_rhs2_ptr (assign)
&& use->use != gimple_assign_rhs3_ptr (assign))
return false;
precision = MAX (precision, use_stmt_info->min_input_precision);
}
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"only the low %d bits of %T are significant\n",
precision, lhs);
stmt_info->min_output_precision = precision;
return true;
}
/* Calculate min_output_precision for STMT_INFO. */
static void
vect_determine_min_output_precision (stmt_vec_info stmt_info)
{
/* We're only interested in statements with a narrowable result. */
tree lhs = gimple_get_lhs (stmt_info->stmt);
if (!lhs
|| TREE_CODE (lhs) != SSA_NAME
|| !vect_narrowable_type_p (TREE_TYPE (lhs)))
return;
if (!vect_determine_min_output_precision_1 (stmt_info, lhs))
stmt_info->min_output_precision = TYPE_PRECISION (TREE_TYPE (lhs));
}
/* Use range information to decide whether STMT (described by STMT_INFO)
could be done in a narrower type. This is effectively a forward
propagation, since it uses context-independent information that applies
to all users of an SSA name. */
static void
vect_determine_precisions_from_range (stmt_vec_info stmt_info, gassign *stmt)
{
tree lhs = gimple_assign_lhs (stmt);
if (!lhs || TREE_CODE (lhs) != SSA_NAME)
return;
tree type = TREE_TYPE (lhs);
if (!vect_narrowable_type_p (type))
return;
/* First see whether we have any useful range information for the result. */
unsigned int precision = TYPE_PRECISION (type);
signop sign = TYPE_SIGN (type);
wide_int min_value, max_value;
if (!vect_get_range_info (lhs, &min_value, &max_value))
return;
tree_code code = gimple_assign_rhs_code (stmt);
unsigned int nops = gimple_num_ops (stmt);
if (!vect_truncatable_operation_p (code))
/* Check that all relevant input operands are compatible, and update
[MIN_VALUE, MAX_VALUE] to include their ranges. */
for (unsigned int i = 1; i < nops; ++i)
{
tree op = gimple_op (stmt, i);
if (TREE_CODE (op) == INTEGER_CST)
{
/* Don't require the integer to have RHS_TYPE (which it might
not for things like shift amounts, etc.), but do require it
to fit the type. */
if (!int_fits_type_p (op, type))
return;
min_value = wi::min (min_value, wi::to_wide (op, precision), sign);
max_value = wi::max (max_value, wi::to_wide (op, precision), sign);
}
else if (TREE_CODE (op) == SSA_NAME)
{
/* Ignore codes that don't take uniform arguments. */
if (!types_compatible_p (TREE_TYPE (op), type))
return;
wide_int op_min_value, op_max_value;
if (!vect_get_range_info (op, &op_min_value, &op_max_value))
return;
min_value = wi::min (min_value, op_min_value, sign);
max_value = wi::max (max_value, op_max_value, sign);
}
else
return;
}
/* Try to switch signed types for unsigned types if we can.
This is better for two reasons. First, unsigned ops tend
to be cheaper than signed ops. Second, it means that we can
handle things like:
signed char c;
int res = (int) c & 0xff00; // range [0x0000, 0xff00]
as:
signed char c;
unsigned short res_1 = (unsigned short) c & 0xff00;
int res = (int) res_1;
where the intermediate result res_1 has unsigned rather than
signed type. */
if (sign == SIGNED && !wi::neg_p (min_value))
sign = UNSIGNED;
/* See what precision is required for MIN_VALUE and MAX_VALUE. */
unsigned int precision1 = wi::min_precision (min_value, sign);
unsigned int precision2 = wi::min_precision (max_value, sign);
unsigned int value_precision = MAX (precision1, precision2);
if (value_precision >= precision)
return;
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location, "can narrow to %s:%d"
" without loss of precision: %G",
sign == SIGNED ? "signed" : "unsigned",
value_precision, stmt);
vect_set_operation_type (stmt_info, type, value_precision, sign);
vect_set_min_input_precision (stmt_info, type, value_precision);
}
/* Use information about the users of STMT's result to decide whether
STMT (described by STMT_INFO) could be done in a narrower type.
This is effectively a backward propagation. */
static void
vect_determine_precisions_from_users (stmt_vec_info stmt_info, gassign *stmt)
{
tree_code code = gimple_assign_rhs_code (stmt);
unsigned int opno = (code == COND_EXPR ? 2 : 1);
tree type = TREE_TYPE (gimple_op (stmt, opno));
if (!vect_narrowable_type_p (type))
return;
unsigned int precision = TYPE_PRECISION (type);
unsigned int operation_precision, min_input_precision;
switch (code)
{
CASE_CONVERT:
/* Only the bits that contribute to the output matter. Don't change
the precision of the operation itself. */
operation_precision = precision;
min_input_precision = stmt_info->min_output_precision;
break;
case LSHIFT_EXPR:
case RSHIFT_EXPR:
{
tree shift = gimple_assign_rhs2 (stmt);
if (TREE_CODE (shift) != INTEGER_CST
|| !wi::ltu_p (wi::to_widest (shift), precision))
return;
unsigned int const_shift = TREE_INT_CST_LOW (shift);
if (code == LSHIFT_EXPR)
{
/* We need CONST_SHIFT fewer bits of the input. */
operation_precision = stmt_info->min_output_precision;
min_input_precision = (MAX (operation_precision, const_shift)
- const_shift);
}
else
{
/* We need CONST_SHIFT extra bits to do the operation. */
operation_precision = (stmt_info->min_output_precision
+ const_shift);
min_input_precision = operation_precision;
}
break;
}
default:
if (vect_truncatable_operation_p (code))
{
/* Input bit N has no effect on output bits N-1 and lower. */
operation_precision = stmt_info->min_output_precision;
min_input_precision = operation_precision;
break;
}
return;
}
if (operation_precision < precision)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location, "can narrow to %s:%d"
" without affecting users: %G",
TYPE_UNSIGNED (type) ? "unsigned" : "signed",
operation_precision, stmt);
vect_set_operation_type (stmt_info, type, operation_precision,
TYPE_SIGN (type));
}
vect_set_min_input_precision (stmt_info, type, min_input_precision);
}
/* Handle vect_determine_precisions for STMT_INFO, given that we
have already done so for the users of its result. */
void
vect_determine_stmt_precisions (stmt_vec_info stmt_info)
{
vect_determine_min_output_precision (stmt_info);
if (gassign *stmt = dyn_cast <gassign *> (stmt_info->stmt))
{
vect_determine_precisions_from_range (stmt_info, stmt);
vect_determine_precisions_from_users (stmt_info, stmt);
}
}
/* Walk backwards through the vectorizable region to determine the
values of these fields:
- min_output_precision
- min_input_precision
- operation_precision
- operation_sign. */
void
vect_determine_precisions (vec_info *vinfo)
{
DUMP_VECT_SCOPE ("vect_determine_precisions");
if (loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo))
{
class loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
basic_block *bbs = LOOP_VINFO_BBS (loop_vinfo);
unsigned int nbbs = loop->num_nodes;
for (unsigned int i = 0; i < nbbs; i++)
{
basic_block bb = bbs[nbbs - i - 1];
for (gimple_stmt_iterator si = gsi_last_bb (bb);
!gsi_end_p (si); gsi_prev (&si))
vect_determine_stmt_precisions
(vinfo->lookup_stmt (gsi_stmt (si)));
}
}
else
{
bb_vec_info bb_vinfo = as_a <bb_vec_info> (vinfo);
gimple_stmt_iterator si = bb_vinfo->region_end;
gimple *stmt;
do
{
if (!gsi_stmt (si))
si = gsi_last_bb (bb_vinfo->bb);
else
gsi_prev (&si);
stmt = gsi_stmt (si);
stmt_vec_info stmt_info = vinfo->lookup_stmt (stmt);
if (stmt_info && STMT_VINFO_VECTORIZABLE (stmt_info))
vect_determine_stmt_precisions (stmt_info);
}
while (stmt != gsi_stmt (bb_vinfo->region_begin));
}
}
typedef gimple *(*vect_recog_func_ptr) (stmt_vec_info, tree *);
struct vect_recog_func
{
vect_recog_func_ptr fn;
const char *name;
};
/* Note that ordering matters - the first pattern matching on a stmt is
taken which means usually the more complex one needs to preceed the
less comples onex (widen_sum only after dot_prod or sad for example). */
static vect_recog_func vect_vect_recog_func_ptrs[] = {
{ vect_recog_over_widening_pattern, "over_widening" },
/* Must come after over_widening, which narrows the shift as much as
possible beforehand. */
{ vect_recog_average_pattern, "average" },
{ vect_recog_mulhs_pattern, "mult_high" },
{ vect_recog_cast_forwprop_pattern, "cast_forwprop" },
{ vect_recog_widen_mult_pattern, "widen_mult" },
{ vect_recog_dot_prod_pattern, "dot_prod" },
{ vect_recog_sad_pattern, "sad" },
{ vect_recog_widen_sum_pattern, "widen_sum" },
{ vect_recog_pow_pattern, "pow" },
{ vect_recog_widen_shift_pattern, "widen_shift" },
{ vect_recog_rotate_pattern, "rotate" },
{ vect_recog_vector_vector_shift_pattern, "vector_vector_shift" },
{ vect_recog_divmod_pattern, "divmod" },
{ vect_recog_mult_pattern, "mult" },
{ vect_recog_mixed_size_cond_pattern, "mixed_size_cond" },
{ vect_recog_bool_pattern, "bool" },
/* This must come before mask conversion, and includes the parts
of mask conversion that are needed for gather and scatter
internal functions. */
{ vect_recog_gather_scatter_pattern, "gather_scatter" },
{ vect_recog_mask_conversion_pattern, "mask_conversion" }
};
const unsigned int NUM_PATTERNS = ARRAY_SIZE (vect_vect_recog_func_ptrs);
/* Mark statements that are involved in a pattern. */
static inline void
vect_mark_pattern_stmts (stmt_vec_info orig_stmt_info, gimple *pattern_stmt,
tree pattern_vectype)
{
stmt_vec_info orig_stmt_info_saved = orig_stmt_info;
gimple *def_seq = STMT_VINFO_PATTERN_DEF_SEQ (orig_stmt_info);
gimple *orig_pattern_stmt = NULL;
if (is_pattern_stmt_p (orig_stmt_info))
{
/* We're replacing a statement in an existing pattern definition
sequence. */
orig_pattern_stmt = orig_stmt_info->stmt;
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"replacing earlier pattern %G", orig_pattern_stmt);
/* To keep the book-keeping simple, just swap the lhs of the
old and new statements, so that the old one has a valid but
unused lhs. */
tree old_lhs = gimple_get_lhs (orig_pattern_stmt);
gimple_set_lhs (orig_pattern_stmt, gimple_get_lhs (pattern_stmt));
gimple_set_lhs (pattern_stmt, old_lhs);
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location, "with %G", pattern_stmt);
/* Switch to the statement that ORIG replaces. */
orig_stmt_info = STMT_VINFO_RELATED_STMT (orig_stmt_info);
/* We shouldn't be replacing the main pattern statement. */
gcc_assert (STMT_VINFO_RELATED_STMT (orig_stmt_info)->stmt
!= orig_pattern_stmt);
}
if (def_seq)
for (gimple_stmt_iterator si = gsi_start (def_seq);
!gsi_end_p (si); gsi_next (&si))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"extra pattern stmt: %G", gsi_stmt (si));
stmt_vec_info pattern_stmt_info
= vect_init_pattern_stmt (gsi_stmt (si),
orig_stmt_info, pattern_vectype);
/* Stmts in the def sequence are not vectorizable cycle or
induction defs, instead they should all be vect_internal_def
feeding the main pattern stmt which retains this def type. */
STMT_VINFO_DEF_TYPE (pattern_stmt_info) = vect_internal_def;
}
if (orig_pattern_stmt)
{
vect_init_pattern_stmt (pattern_stmt, orig_stmt_info, pattern_vectype);
/* Insert all the new pattern statements before the original one. */
gimple_seq *orig_def_seq = &STMT_VINFO_PATTERN_DEF_SEQ (orig_stmt_info);
gimple_stmt_iterator gsi = gsi_for_stmt (orig_pattern_stmt,
orig_def_seq);
gsi_insert_seq_before_without_update (&gsi, def_seq, GSI_SAME_STMT);
gsi_insert_before_without_update (&gsi, pattern_stmt, GSI_SAME_STMT);
/* Remove the pattern statement that this new pattern replaces. */
gsi_remove (&gsi, false);
}
else
vect_set_pattern_stmt (pattern_stmt, orig_stmt_info, pattern_vectype);
/* Transfer reduction path info to the pattern. */
if (STMT_VINFO_REDUC_IDX (orig_stmt_info_saved) != -1)
{
vec_info *vinfo = orig_stmt_info_saved->vinfo;
tree lookfor = gimple_op (orig_stmt_info_saved->stmt,
1 + STMT_VINFO_REDUC_IDX (orig_stmt_info));
/* Search the pattern def sequence and the main pattern stmt. Note
we may have inserted all into a containing pattern def sequence
so the following is a bit awkward. */
gimple_stmt_iterator si;
gimple *s;
if (def_seq)
{
si = gsi_start (def_seq);
s = gsi_stmt (si);
gsi_next (&si);
}
else
{
si = gsi_none ();
s = pattern_stmt;
}
do
{
bool found = false;
for (unsigned i = 1; i < gimple_num_ops (s); ++i)
if (gimple_op (s, i) == lookfor)
{
STMT_VINFO_REDUC_IDX (vinfo->lookup_stmt (s)) = i - 1;
lookfor = gimple_get_lhs (s);
found = true;
break;
}
if (s == pattern_stmt)
{
if (!found && dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"failed to update reduction index.\n");
break;
}
if (gsi_end_p (si))
s = pattern_stmt;
else
{
s = gsi_stmt (si);
if (s == pattern_stmt)
/* Found the end inside a bigger pattern def seq. */
si = gsi_none ();
else
gsi_next (&si);
}
} while (1);
}
}
/* Function vect_pattern_recog_1
Input:
PATTERN_RECOG_FUNC: A pointer to a function that detects a certain
computation pattern.
STMT_INFO: A stmt from which the pattern search should start.
If PATTERN_RECOG_FUNC successfully detected the pattern, it creates
a sequence of statements that has the same functionality and can be
used to replace STMT_INFO. It returns the last statement in the sequence
and adds any earlier statements to STMT_INFO's STMT_VINFO_PATTERN_DEF_SEQ.
PATTERN_RECOG_FUNC also sets *TYPE_OUT to the vector type of the final
statement, having first checked that the target supports the new operation
in that type.
This function also does some bookkeeping, as explained in the documentation
for vect_recog_pattern. */
static void
vect_pattern_recog_1 (vect_recog_func *recog_func, stmt_vec_info stmt_info)
{
vec_info *vinfo = stmt_info->vinfo;
gimple *pattern_stmt;
loop_vec_info loop_vinfo;
tree pattern_vectype;
/* If this statement has already been replaced with pattern statements,
leave the original statement alone, since the first match wins.
Instead try to match against the definition statements that feed
the main pattern statement. */
if (STMT_VINFO_IN_PATTERN_P (stmt_info))
{
gimple_stmt_iterator gsi;
for (gsi = gsi_start (STMT_VINFO_PATTERN_DEF_SEQ (stmt_info));
!gsi_end_p (gsi); gsi_next (&gsi))
vect_pattern_recog_1 (recog_func, vinfo->lookup_stmt (gsi_stmt (gsi)));
return;
}
gcc_assert (!STMT_VINFO_PATTERN_DEF_SEQ (stmt_info));
pattern_stmt = recog_func->fn (stmt_info, &pattern_vectype);
if (!pattern_stmt)
{
/* Clear any half-formed pattern definition sequence. */
STMT_VINFO_PATTERN_DEF_SEQ (stmt_info) = NULL;
return;
}
loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
gcc_assert (pattern_vectype);
/* Found a vectorizable pattern. */
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"%s pattern recognized: %G",
recog_func->name, pattern_stmt);
/* Mark the stmts that are involved in the pattern. */
vect_mark_pattern_stmts (stmt_info, pattern_stmt, pattern_vectype);
/* Patterns cannot be vectorized using SLP, because they change the order of
computation. */
if (loop_vinfo)
{
unsigned ix, ix2;
stmt_vec_info *elem_ptr;
VEC_ORDERED_REMOVE_IF (LOOP_VINFO_REDUCTIONS (loop_vinfo), ix, ix2,
elem_ptr, *elem_ptr == stmt_info);
}
}
/* Function vect_pattern_recog
Input:
LOOP_VINFO - a struct_loop_info of a loop in which we want to look for
computation idioms.
Output - for each computation idiom that is detected we create a new stmt
that provides the same functionality and that can be vectorized. We
also record some information in the struct_stmt_info of the relevant
stmts, as explained below:
At the entry to this function we have the following stmts, with the
following initial value in the STMT_VINFO fields:
stmt in_pattern_p related_stmt vec_stmt
S1: a_i = .... - - -
S2: a_2 = ..use(a_i).. - - -
S3: a_1 = ..use(a_2).. - - -
S4: a_0 = ..use(a_1).. - - -
S5: ... = ..use(a_0).. - - -
Say the sequence {S1,S2,S3,S4} was detected as a pattern that can be
represented by a single stmt. We then:
- create a new stmt S6 equivalent to the pattern (the stmt is not
inserted into the code)
- fill in the STMT_VINFO fields as follows:
in_pattern_p related_stmt vec_stmt
S1: a_i = .... - - -
S2: a_2 = ..use(a_i).. - - -
S3: a_1 = ..use(a_2).. - - -
S4: a_0 = ..use(a_1).. true S6 -
'---> S6: a_new = .... - S4 -
S5: ... = ..use(a_0).. - - -
(the last stmt in the pattern (S4) and the new pattern stmt (S6) point
to each other through the RELATED_STMT field).
S6 will be marked as relevant in vect_mark_stmts_to_be_vectorized instead
of S4 because it will replace all its uses. Stmts {S1,S2,S3} will
remain irrelevant unless used by stmts other than S4.
If vectorization succeeds, vect_transform_stmt will skip over {S1,S2,S3}
(because they are marked as irrelevant). It will vectorize S6, and record
a pointer to the new vector stmt VS6 from S6 (as usual).
S4 will be skipped, and S5 will be vectorized as usual:
in_pattern_p related_stmt vec_stmt
S1: a_i = .... - - -
S2: a_2 = ..use(a_i).. - - -
S3: a_1 = ..use(a_2).. - - -
> VS6: va_new = .... - - -
S4: a_0 = ..use(a_1).. true S6 VS6
'---> S6: a_new = .... - S4 VS6
> VS5: ... = ..vuse(va_new).. - - -
S5: ... = ..use(a_0).. - - -
DCE could then get rid of {S1,S2,S3,S4,S5} (if their defs are not used
elsewhere), and we'll end up with:
VS6: va_new = ....
VS5: ... = ..vuse(va_new)..
In case of more than one pattern statements, e.g., widen-mult with
intermediate type:
S1 a_t = ;
S2 a_T = (TYPE) a_t;
'--> S3: a_it = (interm_type) a_t;
S4 prod_T = a_T * CONST;
'--> S5: prod_T' = a_it w* CONST;
there may be other users of a_T outside the pattern. In that case S2 will
be marked as relevant (as well as S3), and both S2 and S3 will be analyzed
and vectorized. The vector stmt VS2 will be recorded in S2, and VS3 will
be recorded in S3. */
void
vect_pattern_recog (vec_info *vinfo)
{
class loop *loop;
basic_block *bbs;
unsigned int nbbs;
gimple_stmt_iterator si;
unsigned int i, j;
vect_determine_precisions (vinfo);
DUMP_VECT_SCOPE ("vect_pattern_recog");
if (loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo))
{
loop = LOOP_VINFO_LOOP (loop_vinfo);
bbs = LOOP_VINFO_BBS (loop_vinfo);
nbbs = loop->num_nodes;
/* Scan through the loop stmts, applying the pattern recognition
functions starting at each stmt visited: */
for (i = 0; i < nbbs; i++)
{
basic_block bb = bbs[i];
for (si = gsi_start_bb (bb); !gsi_end_p (si); gsi_next (&si))
{
stmt_vec_info stmt_info = vinfo->lookup_stmt (gsi_stmt (si));
/* Scan over all generic vect_recog_xxx_pattern functions. */
for (j = 0; j < NUM_PATTERNS; j++)
vect_pattern_recog_1 (&vect_vect_recog_func_ptrs[j],
stmt_info);
}
}
}
else
{
bb_vec_info bb_vinfo = as_a <bb_vec_info> (vinfo);
for (si = bb_vinfo->region_begin;
gsi_stmt (si) != gsi_stmt (bb_vinfo->region_end); gsi_next (&si))
{
gimple *stmt = gsi_stmt (si);
stmt_vec_info stmt_info = bb_vinfo->lookup_stmt (stmt);
if (stmt_info && !STMT_VINFO_VECTORIZABLE (stmt_info))
continue;
/* Scan over all generic vect_recog_xxx_pattern functions. */
for (j = 0; j < NUM_PATTERNS; j++)
vect_pattern_recog_1 (&vect_vect_recog_func_ptrs[j], stmt_info);
}
}
}
|