1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118
5119
5120
5121
5122
5123
5124
5125
5126
5127
5128
5129
5130
5131
5132
5133
5134
5135
5136
5137
5138
5139
5140
5141
5142
5143
5144
5145
5146
5147
5148
5149
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160
5161
5162
5163
5164
5165
5166
5167
5168
5169
5170
5171
5172
5173
5174
5175
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220
5221
5222
5223
5224
5225
5226
5227
5228
5229
5230
5231
5232
5233
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
5271
5272
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289
5290
5291
5292
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308
5309
5310
5311
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334
5335
5336
5337
5338
5339
5340
5341
5342
5343
5344
5345
5346
5347
5348
5349
5350
5351
5352
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364
5365
5366
5367
5368
5369
5370
5371
5372
5373
5374
5375
5376
5377
5378
5379
5380
5381
5382
5383
5384
5385
5386
5387
5388
5389
5390
5391
5392
5393
5394
5395
5396
5397
5398
5399
5400
5401
5402
5403
5404
5405
5406
5407
5408
5409
5410
5411
5412
5413
5414
5415
5416
5417
5418
5419
5420
5421
5422
5423
5424
5425
5426
5427
5428
5429
5430
5431
5432
5433
5434
5435
5436
5437
5438
5439
5440
5441
5442
5443
5444
5445
5446
5447
5448
5449
5450
5451
5452
5453
5454
5455
5456
5457
5458
5459
5460
5461
5462
5463
5464
5465
5466
5467
5468
5469
5470
5471
5472
5473
5474
5475
5476
5477
5478
5479
5480
5481
5482
5483
5484
5485
5486
5487
5488
5489
5490
5491
5492
5493
5494
5495
5496
5497
5498
5499
5500
5501
5502
5503
5504
5505
5506
5507
5508
5509
5510
5511
5512
5513
5514
5515
5516
5517
5518
5519
5520
5521
5522
5523
5524
5525
5526
5527
5528
5529
5530
5531
5532
5533
5534
5535
5536
5537
5538
5539
5540
5541
5542
5543
5544
5545
5546
5547
5548
5549
5550
5551
5552
5553
5554
5555
5556
5557
5558
5559
5560
5561
5562
5563
5564
5565
5566
5567
5568
5569
5570
5571
5572
5573
5574
5575
5576
5577
5578
5579
5580
5581
5582
5583
5584
5585
5586
5587
5588
5589
5590
5591
5592
5593
5594
5595
5596
5597
5598
5599
5600
5601
5602
5603
5604
5605
5606
5607
5608
5609
5610
5611
5612
5613
5614
5615
5616
5617
5618
5619
5620
5621
5622
5623
5624
5625
5626
5627
5628
5629
5630
5631
5632
5633
5634
5635
5636
5637
5638
5639
5640
5641
5642
5643
5644
5645
5646
5647
5648
5649
5650
5651
5652
5653
5654
5655
5656
5657
5658
5659
5660
5661
5662
5663
5664
5665
5666
5667
5668
5669
5670
5671
5672
5673
5674
5675
5676
5677
5678
5679
5680
5681
5682
5683
5684
5685
5686
5687
5688
5689
5690
5691
5692
5693
5694
5695
5696
5697
5698
5699
5700
5701
5702
5703
5704
5705
5706
5707
5708
5709
5710
5711
5712
5713
5714
5715
5716
5717
5718
5719
5720
5721
5722
5723
5724
5725
5726
5727
5728
5729
5730
5731
5732
5733
5734
5735
5736
5737
5738
5739
5740
5741
5742
5743
5744
5745
5746
5747
5748
5749
5750
5751
5752
5753
5754
5755
5756
5757
5758
5759
5760
5761
5762
5763
5764
5765
5766
5767
5768
5769
5770
5771
5772
5773
5774
5775
5776
5777
5778
5779
5780
5781
5782
5783
5784
5785
5786
5787
5788
5789
5790
5791
5792
5793
5794
5795
5796
5797
5798
5799
5800
5801
5802
5803
5804
5805
5806
5807
5808
5809
5810
5811
5812
5813
5814
5815
5816
5817
5818
5819
5820
5821
5822
5823
5824
5825
5826
5827
5828
5829
5830
5831
5832
5833
5834
5835
5836
5837
5838
5839
5840
5841
5842
5843
5844
5845
5846
5847
5848
5849
5850
5851
5852
5853
5854
5855
5856
5857
5858
5859
5860
5861
5862
5863
5864
5865
5866
5867
5868
5869
5870
5871
5872
5873
5874
5875
5876
5877
5878
5879
5880
5881
5882
5883
5884
5885
5886
5887
5888
5889
5890
5891
5892
5893
5894
5895
5896
5897
5898
5899
5900
5901
5902
5903
5904
5905
5906
5907
5908
5909
5910
5911
5912
5913
5914
5915
5916
5917
5918
5919
5920
5921
5922
5923
5924
5925
5926
5927
5928
5929
5930
5931
5932
5933
5934
5935
5936
5937
5938
5939
5940
5941
5942
5943
5944
5945
5946
5947
5948
5949
5950
5951
5952
5953
5954
5955
5956
5957
5958
5959
5960
5961
5962
5963
5964
5965
5966
5967
5968
5969
5970
5971
5972
5973
5974
5975
5976
5977
5978
5979
5980
5981
5982
5983
5984
5985
5986
5987
5988
5989
5990
5991
5992
5993
5994
5995
5996
5997
5998
5999
6000
6001
6002
6003
6004
6005
6006
6007
6008
6009
6010
6011
6012
6013
6014
6015
6016
6017
6018
6019
6020
6021
6022
6023
6024
6025
6026
6027
6028
6029
6030
6031
6032
6033
6034
6035
6036
6037
6038
6039
6040
6041
6042
6043
6044
6045
6046
6047
6048
6049
6050
6051
6052
6053
6054
6055
6056
6057
6058
6059
6060
6061
6062
6063
6064
6065
6066
6067
6068
6069
6070
6071
6072
6073
6074
6075
6076
6077
6078
6079
6080
6081
6082
6083
6084
6085
6086
6087
6088
6089
6090
6091
6092
6093
6094
6095
6096
6097
6098
6099
6100
6101
6102
6103
6104
6105
6106
6107
6108
6109
6110
6111
6112
6113
6114
6115
6116
6117
6118
6119
6120
6121
6122
6123
6124
6125
6126
6127
6128
6129
6130
6131
6132
6133
6134
6135
6136
6137
6138
6139
6140
6141
6142
6143
6144
6145
6146
6147
6148
6149
6150
6151
6152
6153
6154
6155
6156
6157
6158
6159
6160
6161
6162
6163
6164
6165
6166
6167
6168
6169
6170
6171
6172
6173
6174
6175
6176
6177
6178
6179
6180
6181
6182
6183
6184
6185
6186
6187
6188
6189
6190
6191
6192
6193
6194
6195
6196
6197
6198
6199
6200
6201
6202
6203
6204
6205
6206
6207
6208
6209
6210
6211
6212
6213
6214
6215
6216
6217
6218
6219
6220
6221
6222
6223
6224
6225
6226
6227
6228
6229
6230
6231
6232
6233
6234
6235
6236
6237
6238
6239
6240
6241
6242
6243
6244
6245
6246
6247
6248
6249
6250
6251
6252
6253
6254
6255
6256
6257
6258
6259
6260
6261
6262
6263
6264
6265
6266
6267
6268
6269
6270
6271
6272
6273
6274
6275
6276
6277
6278
6279
6280
6281
6282
6283
6284
6285
6286
6287
6288
6289
6290
6291
6292
6293
6294
6295
6296
6297
6298
6299
6300
6301
6302
6303
6304
6305
6306
6307
6308
6309
6310
6311
6312
6313
6314
6315
6316
6317
6318
6319
6320
6321
6322
6323
6324
6325
6326
6327
6328
6329
6330
6331
6332
6333
6334
6335
6336
6337
6338
6339
6340
6341
6342
6343
6344
6345
6346
6347
6348
6349
6350
6351
6352
6353
6354
6355
6356
6357
6358
6359
6360
6361
6362
6363
6364
6365
6366
6367
6368
6369
6370
6371
6372
6373
6374
6375
6376
6377
6378
6379
6380
6381
6382
6383
6384
6385
6386
6387
6388
6389
6390
6391
6392
6393
6394
6395
6396
6397
6398
6399
6400
6401
6402
6403
6404
6405
6406
6407
6408
6409
6410
6411
6412
6413
6414
6415
6416
6417
6418
6419
6420
6421
6422
6423
6424
6425
6426
6427
6428
6429
6430
6431
6432
6433
6434
6435
6436
6437
6438
6439
6440
6441
6442
6443
6444
6445
6446
6447
6448
6449
6450
6451
6452
6453
6454
6455
6456
6457
6458
6459
6460
6461
6462
6463
6464
6465
6466
6467
6468
6469
6470
6471
6472
6473
6474
6475
6476
6477
6478
6479
6480
6481
6482
6483
6484
6485
6486
6487
6488
6489
6490
6491
6492
6493
6494
6495
6496
6497
6498
6499
6500
6501
6502
6503
6504
6505
6506
6507
6508
6509
6510
6511
6512
6513
6514
6515
6516
6517
6518
6519
6520
6521
6522
6523
6524
6525
6526
6527
6528
6529
6530
6531
6532
6533
6534
6535
6536
6537
6538
6539
6540
6541
6542
6543
6544
6545
6546
6547
6548
6549
6550
6551
6552
6553
6554
6555
6556
6557
6558
6559
6560
6561
6562
6563
6564
6565
6566
6567
6568
6569
6570
6571
6572
6573
6574
6575
6576
6577
6578
6579
6580
6581
6582
6583
6584
6585
6586
6587
6588
6589
6590
6591
6592
6593
6594
6595
6596
6597
6598
6599
6600
6601
6602
6603
6604
6605
6606
6607
6608
6609
6610
6611
6612
6613
6614
6615
|
/* Data References Analysis and Manipulation Utilities for Vectorization.
Copyright (C) 2003-2018 Free Software Foundation, Inc.
Contributed by Dorit Naishlos <dorit@il.ibm.com>
and Ira Rosen <irar@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 "target.h"
#include "rtl.h"
#include "tree.h"
#include "gimple.h"
#include "predict.h"
#include "memmodel.h"
#include "tm_p.h"
#include "ssa.h"
#include "optabs-tree.h"
#include "cgraph.h"
#include "dumpfile.h"
#include "alias.h"
#include "fold-const.h"
#include "stor-layout.h"
#include "tree-eh.h"
#include "gimplify.h"
#include "gimple-iterator.h"
#include "gimplify-me.h"
#include "tree-ssa-loop-ivopts.h"
#include "tree-ssa-loop-manip.h"
#include "tree-ssa-loop.h"
#include "cfgloop.h"
#include "tree-scalar-evolution.h"
#include "tree-vectorizer.h"
#include "expr.h"
#include "builtins.h"
#include "params.h"
#include "tree-cfg.h"
#include "tree-hash-traits.h"
#include "vec-perm-indices.h"
#include "internal-fn.h"
/* Return true if load- or store-lanes optab OPTAB is implemented for
COUNT vectors of type VECTYPE. NAME is the name of OPTAB. */
static bool
vect_lanes_optab_supported_p (const char *name, convert_optab optab,
tree vectype, unsigned HOST_WIDE_INT count)
{
machine_mode mode, array_mode;
bool limit_p;
mode = TYPE_MODE (vectype);
if (!targetm.array_mode (mode, count).exists (&array_mode))
{
poly_uint64 bits = count * GET_MODE_BITSIZE (mode);
limit_p = !targetm.array_mode_supported_p (mode, count);
if (!int_mode_for_size (bits, limit_p).exists (&array_mode))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"no array mode for %s["
HOST_WIDE_INT_PRINT_DEC "]\n",
GET_MODE_NAME (mode), count);
return false;
}
}
if (convert_optab_handler (optab, array_mode, mode) == CODE_FOR_nothing)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"cannot use %s<%s><%s>\n", name,
GET_MODE_NAME (array_mode), GET_MODE_NAME (mode));
return false;
}
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"can use %s<%s><%s>\n", name, GET_MODE_NAME (array_mode),
GET_MODE_NAME (mode));
return true;
}
/* Return the smallest scalar part of STMT.
This is used to determine the vectype of the stmt. We generally set the
vectype according to the type of the result (lhs). For stmts whose
result-type is different than the type of the arguments (e.g., demotion,
promotion), vectype will be reset appropriately (later). Note that we have
to visit the smallest datatype in this function, because that determines the
VF. If the smallest datatype in the loop is present only as the rhs of a
promotion operation - we'd miss it.
Such a case, where a variable of this datatype does not appear in the lhs
anywhere in the loop, can only occur if it's an invariant: e.g.:
'int_x = (int) short_inv', which we'd expect to have been optimized away by
invariant motion. However, we cannot rely on invariant motion to always
take invariants out of the loop, and so in the case of promotion we also
have to check the rhs.
LHS_SIZE_UNIT and RHS_SIZE_UNIT contain the sizes of the corresponding
types. */
tree
vect_get_smallest_scalar_type (gimple *stmt, HOST_WIDE_INT *lhs_size_unit,
HOST_WIDE_INT *rhs_size_unit)
{
tree scalar_type = gimple_expr_type (stmt);
HOST_WIDE_INT lhs, rhs;
/* During the analysis phase, this function is called on arbitrary
statements that might not have scalar results. */
if (!tree_fits_uhwi_p (TYPE_SIZE_UNIT (scalar_type)))
return scalar_type;
lhs = rhs = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (scalar_type));
if (is_gimple_assign (stmt)
&& (gimple_assign_cast_p (stmt)
|| gimple_assign_rhs_code (stmt) == DOT_PROD_EXPR
|| gimple_assign_rhs_code (stmt) == WIDEN_SUM_EXPR
|| gimple_assign_rhs_code (stmt) == WIDEN_MULT_EXPR
|| gimple_assign_rhs_code (stmt) == WIDEN_LSHIFT_EXPR
|| gimple_assign_rhs_code (stmt) == FLOAT_EXPR))
{
tree rhs_type = TREE_TYPE (gimple_assign_rhs1 (stmt));
rhs = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (rhs_type));
if (rhs < lhs)
scalar_type = rhs_type;
}
*lhs_size_unit = lhs;
*rhs_size_unit = rhs;
return scalar_type;
}
/* Insert DDR into LOOP_VINFO list of ddrs that may alias and need to be
tested at run-time. Return TRUE if DDR was successfully inserted.
Return false if versioning is not supported. */
static bool
vect_mark_for_runtime_alias_test (ddr_p ddr, loop_vec_info loop_vinfo)
{
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
if ((unsigned) PARAM_VALUE (PARAM_VECT_MAX_VERSION_FOR_ALIAS_CHECKS) == 0)
return false;
if (!runtime_alias_check_p (ddr, loop,
optimize_loop_nest_for_speed_p (loop)))
return false;
LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo).safe_push (ddr);
return true;
}
/* Record that loop LOOP_VINFO needs to check that VALUE is nonzero. */
static void
vect_check_nonzero_value (loop_vec_info loop_vinfo, tree value)
{
vec<tree> checks = LOOP_VINFO_CHECK_NONZERO (loop_vinfo);
for (unsigned int i = 0; i < checks.length(); ++i)
if (checks[i] == value)
return;
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location, "need run-time check that ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, value);
dump_printf (MSG_NOTE, " is nonzero\n");
}
LOOP_VINFO_CHECK_NONZERO (loop_vinfo).safe_push (value);
}
/* Return true if we know that the order of vectorized STMT_A and
vectorized STMT_B will be the same as the order of STMT_A and STMT_B.
At least one of the statements is a write. */
static bool
vect_preserves_scalar_order_p (gimple *stmt_a, gimple *stmt_b)
{
stmt_vec_info stmtinfo_a = vinfo_for_stmt (stmt_a);
stmt_vec_info stmtinfo_b = vinfo_for_stmt (stmt_b);
/* Single statements are always kept in their original order. */
if (!STMT_VINFO_GROUPED_ACCESS (stmtinfo_a)
&& !STMT_VINFO_GROUPED_ACCESS (stmtinfo_b))
return true;
/* STMT_A and STMT_B belong to overlapping groups. All loads in a
group are emitted at the position of the first scalar load and all
stores in a group are emitted at the position of the last scalar store.
Thus writes will happen no earlier than their current position
(but could happen later) while reads will happen no later than their
current position (but could happen earlier). Reordering is therefore
only possible if the first access is a write. */
if (is_pattern_stmt_p (stmtinfo_a))
stmt_a = STMT_VINFO_RELATED_STMT (stmtinfo_a);
if (is_pattern_stmt_p (stmtinfo_b))
stmt_b = STMT_VINFO_RELATED_STMT (stmtinfo_b);
gimple *earlier_stmt = get_earlier_stmt (stmt_a, stmt_b);
return !DR_IS_WRITE (STMT_VINFO_DATA_REF (vinfo_for_stmt (earlier_stmt)));
}
/* A subroutine of vect_analyze_data_ref_dependence. Handle
DDR_COULD_BE_INDEPENDENT_P ddr DDR that has a known set of dependence
distances. These distances are conservatively correct but they don't
reflect a guaranteed dependence.
Return true if this function does all the work necessary to avoid
an alias or false if the caller should use the dependence distances
to limit the vectorization factor in the usual way. LOOP_DEPTH is
the depth of the loop described by LOOP_VINFO and the other arguments
are as for vect_analyze_data_ref_dependence. */
static bool
vect_analyze_possibly_independent_ddr (data_dependence_relation *ddr,
loop_vec_info loop_vinfo,
int loop_depth, unsigned int *max_vf)
{
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
lambda_vector dist_v;
unsigned int i;
FOR_EACH_VEC_ELT (DDR_DIST_VECTS (ddr), i, dist_v)
{
int dist = dist_v[loop_depth];
if (dist != 0 && !(dist > 0 && DDR_REVERSED_P (ddr)))
{
/* If the user asserted safelen >= DIST consecutive iterations
can be executed concurrently, assume independence.
??? An alternative would be to add the alias check even
in this case, and vectorize the fallback loop with the
maximum VF set to safelen. However, if the user has
explicitly given a length, it's less likely that that
would be a win. */
if (loop->safelen >= 2 && abs_hwi (dist) <= loop->safelen)
{
if ((unsigned int) loop->safelen < *max_vf)
*max_vf = loop->safelen;
LOOP_VINFO_NO_DATA_DEPENDENCIES (loop_vinfo) = false;
continue;
}
/* For dependence distances of 2 or more, we have the option
of limiting VF or checking for an alias at runtime.
Prefer to check at runtime if we can, to avoid limiting
the VF unnecessarily when the bases are in fact independent.
Note that the alias checks will be removed if the VF ends up
being small enough. */
return vect_mark_for_runtime_alias_test (ddr, loop_vinfo);
}
}
return true;
}
/* Function vect_analyze_data_ref_dependence.
Return TRUE if there (might) exist a dependence between a memory-reference
DRA and a memory-reference DRB. When versioning for alias may check a
dependence at run-time, return FALSE. Adjust *MAX_VF according to
the data dependence. */
static bool
vect_analyze_data_ref_dependence (struct data_dependence_relation *ddr,
loop_vec_info loop_vinfo,
unsigned int *max_vf)
{
unsigned int i;
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
struct data_reference *dra = DDR_A (ddr);
struct data_reference *drb = DDR_B (ddr);
stmt_vec_info stmtinfo_a = vinfo_for_stmt (vect_dr_stmt (dra));
stmt_vec_info stmtinfo_b = vinfo_for_stmt (vect_dr_stmt (drb));
lambda_vector dist_v;
unsigned int loop_depth;
/* In loop analysis all data references should be vectorizable. */
if (!STMT_VINFO_VECTORIZABLE (stmtinfo_a)
|| !STMT_VINFO_VECTORIZABLE (stmtinfo_b))
gcc_unreachable ();
/* Independent data accesses. */
if (DDR_ARE_DEPENDENT (ddr) == chrec_known)
return false;
if (dra == drb
|| (DR_IS_READ (dra) && DR_IS_READ (drb)))
return false;
/* We do not have to consider dependences between accesses that belong
to the same group, unless the stride could be smaller than the
group size. */
if (DR_GROUP_FIRST_ELEMENT (stmtinfo_a)
&& (DR_GROUP_FIRST_ELEMENT (stmtinfo_a)
== DR_GROUP_FIRST_ELEMENT (stmtinfo_b))
&& !STMT_VINFO_STRIDED_P (stmtinfo_a))
return false;
/* Even if we have an anti-dependence then, as the vectorized loop covers at
least two scalar iterations, there is always also a true dependence.
As the vectorizer does not re-order loads and stores we can ignore
the anti-dependence if TBAA can disambiguate both DRs similar to the
case with known negative distance anti-dependences (positive
distance anti-dependences would violate TBAA constraints). */
if (((DR_IS_READ (dra) && DR_IS_WRITE (drb))
|| (DR_IS_WRITE (dra) && DR_IS_READ (drb)))
&& !alias_sets_conflict_p (get_alias_set (DR_REF (dra)),
get_alias_set (DR_REF (drb))))
return false;
/* Unknown data dependence. */
if (DDR_ARE_DEPENDENT (ddr) == chrec_dont_know)
{
/* If user asserted safelen consecutive iterations can be
executed concurrently, assume independence. */
if (loop->safelen >= 2)
{
if ((unsigned int) loop->safelen < *max_vf)
*max_vf = loop->safelen;
LOOP_VINFO_NO_DATA_DEPENDENCIES (loop_vinfo) = false;
return false;
}
if (STMT_VINFO_GATHER_SCATTER_P (stmtinfo_a)
|| STMT_VINFO_GATHER_SCATTER_P (stmtinfo_b))
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"versioning for alias not supported for: "
"can't determine dependence between ");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM,
DR_REF (dra));
dump_printf (MSG_MISSED_OPTIMIZATION, " and ");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM,
DR_REF (drb));
dump_printf (MSG_MISSED_OPTIMIZATION, "\n");
}
return true;
}
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"versioning for alias required: "
"can't determine dependence between ");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM,
DR_REF (dra));
dump_printf (MSG_MISSED_OPTIMIZATION, " and ");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM,
DR_REF (drb));
dump_printf (MSG_MISSED_OPTIMIZATION, "\n");
}
/* Add to list of ddrs that need to be tested at run-time. */
return !vect_mark_for_runtime_alias_test (ddr, loop_vinfo);
}
/* Known data dependence. */
if (DDR_NUM_DIST_VECTS (ddr) == 0)
{
/* If user asserted safelen consecutive iterations can be
executed concurrently, assume independence. */
if (loop->safelen >= 2)
{
if ((unsigned int) loop->safelen < *max_vf)
*max_vf = loop->safelen;
LOOP_VINFO_NO_DATA_DEPENDENCIES (loop_vinfo) = false;
return false;
}
if (STMT_VINFO_GATHER_SCATTER_P (stmtinfo_a)
|| STMT_VINFO_GATHER_SCATTER_P (stmtinfo_b))
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"versioning for alias not supported for: "
"bad dist vector for ");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM,
DR_REF (dra));
dump_printf (MSG_MISSED_OPTIMIZATION, " and ");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM,
DR_REF (drb));
dump_printf (MSG_MISSED_OPTIMIZATION, "\n");
}
return true;
}
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"versioning for alias required: "
"bad dist vector for ");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM, DR_REF (dra));
dump_printf (MSG_MISSED_OPTIMIZATION, " and ");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM, DR_REF (drb));
dump_printf (MSG_MISSED_OPTIMIZATION, "\n");
}
/* Add to list of ddrs that need to be tested at run-time. */
return !vect_mark_for_runtime_alias_test (ddr, loop_vinfo);
}
loop_depth = index_in_loop_nest (loop->num, DDR_LOOP_NEST (ddr));
if (DDR_COULD_BE_INDEPENDENT_P (ddr)
&& vect_analyze_possibly_independent_ddr (ddr, loop_vinfo,
loop_depth, max_vf))
return false;
FOR_EACH_VEC_ELT (DDR_DIST_VECTS (ddr), i, dist_v)
{
int dist = dist_v[loop_depth];
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"dependence distance = %d.\n", dist);
if (dist == 0)
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"dependence distance == 0 between ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (dra));
dump_printf (MSG_NOTE, " and ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (drb));
dump_printf (MSG_MISSED_OPTIMIZATION, "\n");
}
/* When we perform grouped accesses and perform implicit CSE
by detecting equal accesses and doing disambiguation with
runtime alias tests like for
.. = a[i];
.. = a[i+1];
a[i] = ..;
a[i+1] = ..;
*p = ..;
.. = a[i];
.. = a[i+1];
where we will end up loading { a[i], a[i+1] } once, make
sure that inserting group loads before the first load and
stores after the last store will do the right thing.
Similar for groups like
a[i] = ...;
... = a[i];
a[i+1] = ...;
where loads from the group interleave with the store. */
if (!vect_preserves_scalar_order_p (vect_dr_stmt(dra),
vect_dr_stmt (drb)))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"READ_WRITE dependence in interleaving.\n");
return true;
}
if (loop->safelen < 2)
{
tree indicator = dr_zero_step_indicator (dra);
if (TREE_CODE (indicator) != INTEGER_CST)
vect_check_nonzero_value (loop_vinfo, indicator);
else if (integer_zerop (indicator))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"access also has a zero step\n");
return true;
}
}
continue;
}
if (dist > 0 && DDR_REVERSED_P (ddr))
{
/* If DDR_REVERSED_P the order of the data-refs in DDR was
reversed (to make distance vector positive), and the actual
distance is negative. */
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"dependence distance negative.\n");
/* Record a negative dependence distance to later limit the
amount of stmt copying / unrolling we can perform.
Only need to handle read-after-write dependence. */
if (DR_IS_READ (drb)
&& (STMT_VINFO_MIN_NEG_DIST (stmtinfo_b) == 0
|| STMT_VINFO_MIN_NEG_DIST (stmtinfo_b) > (unsigned)dist))
STMT_VINFO_MIN_NEG_DIST (stmtinfo_b) = dist;
continue;
}
unsigned int abs_dist = abs (dist);
if (abs_dist >= 2 && abs_dist < *max_vf)
{
/* The dependence distance requires reduction of the maximal
vectorization factor. */
*max_vf = abs (dist);
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"adjusting maximal vectorization factor to %i\n",
*max_vf);
}
if (abs_dist >= *max_vf)
{
/* Dependence distance does not create dependence, as far as
vectorization is concerned, in this case. */
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"dependence distance >= VF.\n");
continue;
}
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized, possible dependence "
"between data-refs ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (dra));
dump_printf (MSG_NOTE, " and ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (drb));
dump_printf (MSG_NOTE, "\n");
}
return true;
}
return false;
}
/* Function vect_analyze_data_ref_dependences.
Examine all the data references in the loop, and make sure there do not
exist any data dependences between them. Set *MAX_VF according to
the maximum vectorization factor the data dependences allow. */
bool
vect_analyze_data_ref_dependences (loop_vec_info loop_vinfo,
unsigned int *max_vf)
{
unsigned int i;
struct data_dependence_relation *ddr;
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"=== vect_analyze_data_ref_dependences ===\n");
LOOP_VINFO_DDRS (loop_vinfo)
.create (LOOP_VINFO_DATAREFS (loop_vinfo).length ()
* LOOP_VINFO_DATAREFS (loop_vinfo).length ());
LOOP_VINFO_NO_DATA_DEPENDENCIES (loop_vinfo) = true;
/* We need read-read dependences to compute STMT_VINFO_SAME_ALIGN_REFS. */
if (!compute_all_dependences (LOOP_VINFO_DATAREFS (loop_vinfo),
&LOOP_VINFO_DDRS (loop_vinfo),
LOOP_VINFO_LOOP_NEST (loop_vinfo), true))
return false;
/* For epilogues we either have no aliases or alias versioning
was applied to original loop. Therefore we may just get max_vf
using VF of original loop. */
if (LOOP_VINFO_EPILOGUE_P (loop_vinfo))
*max_vf = LOOP_VINFO_ORIG_MAX_VECT_FACTOR (loop_vinfo);
else
FOR_EACH_VEC_ELT (LOOP_VINFO_DDRS (loop_vinfo), i, ddr)
if (vect_analyze_data_ref_dependence (ddr, loop_vinfo, max_vf))
return false;
return true;
}
/* Function vect_slp_analyze_data_ref_dependence.
Return TRUE if there (might) exist a dependence between a memory-reference
DRA and a memory-reference DRB. When versioning for alias may check a
dependence at run-time, return FALSE. Adjust *MAX_VF according to
the data dependence. */
static bool
vect_slp_analyze_data_ref_dependence (struct data_dependence_relation *ddr)
{
struct data_reference *dra = DDR_A (ddr);
struct data_reference *drb = DDR_B (ddr);
/* We need to check dependences of statements marked as unvectorizable
as well, they still can prohibit vectorization. */
/* Independent data accesses. */
if (DDR_ARE_DEPENDENT (ddr) == chrec_known)
return false;
if (dra == drb)
return false;
/* Read-read is OK. */
if (DR_IS_READ (dra) && DR_IS_READ (drb))
return false;
/* If dra and drb are part of the same interleaving chain consider
them independent. */
if (STMT_VINFO_GROUPED_ACCESS (vinfo_for_stmt (vect_dr_stmt (dra)))
&& (DR_GROUP_FIRST_ELEMENT (vinfo_for_stmt (vect_dr_stmt (dra)))
== DR_GROUP_FIRST_ELEMENT (vinfo_for_stmt (vect_dr_stmt (drb)))))
return false;
/* Unknown data dependence. */
if (DDR_ARE_DEPENDENT (ddr) == chrec_dont_know)
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"can't determine dependence between ");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM, DR_REF (dra));
dump_printf (MSG_MISSED_OPTIMIZATION, " and ");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM, DR_REF (drb));
dump_printf (MSG_MISSED_OPTIMIZATION, "\n");
}
}
else if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"determined dependence between ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (dra));
dump_printf (MSG_NOTE, " and ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (drb));
dump_printf (MSG_NOTE, "\n");
}
return true;
}
/* Analyze dependences involved in the transform of SLP NODE. STORES
contain the vector of scalar stores of this instance if we are
disambiguating the loads. */
static bool
vect_slp_analyze_node_dependences (slp_instance instance, slp_tree node,
vec<gimple *> stores, gimple *last_store)
{
/* This walks over all stmts involved in the SLP load/store done
in NODE verifying we can sink them up to the last stmt in the
group. */
gimple *last_access = vect_find_last_scalar_stmt_in_slp (node);
for (unsigned k = 0; k < SLP_INSTANCE_GROUP_SIZE (instance); ++k)
{
gimple *access = SLP_TREE_SCALAR_STMTS (node)[k];
if (access == last_access)
continue;
data_reference *dr_a = STMT_VINFO_DATA_REF (vinfo_for_stmt (access));
ao_ref ref;
bool ref_initialized_p = false;
for (gimple_stmt_iterator gsi = gsi_for_stmt (access);
gsi_stmt (gsi) != last_access; gsi_next (&gsi))
{
gimple *stmt = gsi_stmt (gsi);
if (! gimple_vuse (stmt)
|| (DR_IS_READ (dr_a) && ! gimple_vdef (stmt)))
continue;
/* If we couldn't record a (single) data reference for this
stmt we have to resort to the alias oracle. */
data_reference *dr_b = STMT_VINFO_DATA_REF (vinfo_for_stmt (stmt));
if (!dr_b)
{
/* We are moving a store or sinking a load - this means
we cannot use TBAA for disambiguation. */
if (!ref_initialized_p)
ao_ref_init (&ref, DR_REF (dr_a));
if (stmt_may_clobber_ref_p_1 (stmt, &ref, false)
|| ref_maybe_used_by_stmt_p (stmt, &ref, false))
return false;
continue;
}
bool dependent = false;
/* If we run into a store of this same instance (we've just
marked those) then delay dependence checking until we run
into the last store because this is where it will have
been sunk to (and we verify if we can do that as well). */
if (gimple_visited_p (stmt))
{
if (stmt != last_store)
continue;
unsigned i;
gimple *store;
FOR_EACH_VEC_ELT (stores, i, store)
{
data_reference *store_dr
= STMT_VINFO_DATA_REF (vinfo_for_stmt (store));
ddr_p ddr = initialize_data_dependence_relation
(dr_a, store_dr, vNULL);
dependent = vect_slp_analyze_data_ref_dependence (ddr);
free_dependence_relation (ddr);
if (dependent)
break;
}
}
else
{
ddr_p ddr = initialize_data_dependence_relation (dr_a,
dr_b, vNULL);
dependent = vect_slp_analyze_data_ref_dependence (ddr);
free_dependence_relation (ddr);
}
if (dependent)
return false;
}
}
return true;
}
/* Function vect_analyze_data_ref_dependences.
Examine all the data references in the basic-block, and make sure there
do not exist any data dependences between them. Set *MAX_VF according to
the maximum vectorization factor the data dependences allow. */
bool
vect_slp_analyze_instance_dependence (slp_instance instance)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"=== vect_slp_analyze_instance_dependence ===\n");
/* The stores of this instance are at the root of the SLP tree. */
slp_tree store = SLP_INSTANCE_TREE (instance);
if (! STMT_VINFO_DATA_REF (vinfo_for_stmt (SLP_TREE_SCALAR_STMTS (store)[0])))
store = NULL;
/* Verify we can sink stores to the vectorized stmt insert location. */
gimple *last_store = NULL;
if (store)
{
if (! vect_slp_analyze_node_dependences (instance, store, vNULL, NULL))
return false;
/* Mark stores in this instance and remember the last one. */
last_store = vect_find_last_scalar_stmt_in_slp (store);
for (unsigned k = 0; k < SLP_INSTANCE_GROUP_SIZE (instance); ++k)
gimple_set_visited (SLP_TREE_SCALAR_STMTS (store)[k], true);
}
bool res = true;
/* Verify we can sink loads to the vectorized stmt insert location,
special-casing stores of this instance. */
slp_tree load;
unsigned int i;
FOR_EACH_VEC_ELT (SLP_INSTANCE_LOADS (instance), i, load)
if (! vect_slp_analyze_node_dependences (instance, load,
store
? SLP_TREE_SCALAR_STMTS (store)
: vNULL, last_store))
{
res = false;
break;
}
/* Unset the visited flag. */
if (store)
for (unsigned k = 0; k < SLP_INSTANCE_GROUP_SIZE (instance); ++k)
gimple_set_visited (SLP_TREE_SCALAR_STMTS (store)[k], false);
return res;
}
/* Record in VINFO the base alignment guarantee given by DRB. STMT is
the statement that contains DRB, which is useful for recording in the
dump file. */
static void
vect_record_base_alignment (vec_info *vinfo, gimple *stmt,
innermost_loop_behavior *drb)
{
bool existed;
innermost_loop_behavior *&entry
= vinfo->base_alignments.get_or_insert (drb->base_address, &existed);
if (!existed || entry->base_alignment < drb->base_alignment)
{
entry = drb;
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"recording new base alignment for ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, drb->base_address);
dump_printf (MSG_NOTE, "\n");
dump_printf_loc (MSG_NOTE, vect_location,
" alignment: %d\n", drb->base_alignment);
dump_printf_loc (MSG_NOTE, vect_location,
" misalignment: %d\n", drb->base_misalignment);
dump_printf_loc (MSG_NOTE, vect_location,
" based on: ");
dump_gimple_stmt (MSG_NOTE, TDF_SLIM, stmt, 0);
}
}
}
/* If the region we're going to vectorize is reached, all unconditional
data references occur at least once. We can therefore pool the base
alignment guarantees from each unconditional reference. Do this by
going through all the data references in VINFO and checking whether
the containing statement makes the reference unconditionally. If so,
record the alignment of the base address in VINFO so that it can be
used for all other references with the same base. */
void
vect_record_base_alignments (vec_info *vinfo)
{
loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo);
struct loop *loop = loop_vinfo ? LOOP_VINFO_LOOP (loop_vinfo) : NULL;
data_reference *dr;
unsigned int i;
FOR_EACH_VEC_ELT (vinfo->datarefs, i, dr)
{
gimple *stmt = vect_dr_stmt (dr);
if (!DR_IS_CONDITIONAL_IN_STMT (dr)
&& STMT_VINFO_VECTORIZABLE (vinfo_for_stmt (stmt)))
{
vect_record_base_alignment (vinfo, stmt, &DR_INNERMOST (dr));
/* If DR is nested in the loop that is being vectorized, we can also
record the alignment of the base wrt the outer loop. */
if (loop && nested_in_vect_loop_p (loop, stmt))
{
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
vect_record_base_alignment
(vinfo, stmt, &STMT_VINFO_DR_WRT_VEC_LOOP (stmt_info));
}
}
}
}
/* Return the target alignment for the vectorized form of DR. */
static unsigned int
vect_calculate_target_alignment (struct data_reference *dr)
{
gimple *stmt = vect_dr_stmt (dr);
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
return targetm.vectorize.preferred_vector_alignment (vectype);
}
/* Function vect_compute_data_ref_alignment
Compute the misalignment of the data reference DR.
Output:
1. If during the misalignment computation it is found that the data reference
cannot be vectorized then false is returned.
2. DR_MISALIGNMENT (DR) is defined.
FOR NOW: No analysis is actually performed. Misalignment is calculated
only for trivial cases. TODO. */
static bool
vect_compute_data_ref_alignment (struct data_reference *dr)
{
gimple *stmt = vect_dr_stmt (dr);
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
vec_base_alignments *base_alignments = &stmt_info->vinfo->base_alignments;
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
struct loop *loop = NULL;
tree ref = DR_REF (dr);
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"vect_compute_data_ref_alignment:\n");
if (loop_vinfo)
loop = LOOP_VINFO_LOOP (loop_vinfo);
/* Initialize misalignment to unknown. */
SET_DR_MISALIGNMENT (dr, DR_MISALIGNMENT_UNKNOWN);
innermost_loop_behavior *drb = vect_dr_behavior (dr);
bool step_preserves_misalignment_p;
unsigned HOST_WIDE_INT vector_alignment
= vect_calculate_target_alignment (dr) / BITS_PER_UNIT;
DR_TARGET_ALIGNMENT (dr) = vector_alignment;
/* No step for BB vectorization. */
if (!loop)
{
gcc_assert (integer_zerop (drb->step));
step_preserves_misalignment_p = true;
}
/* In case the dataref is in an inner-loop of the loop that is being
vectorized (LOOP), we use the base and misalignment information
relative to the outer-loop (LOOP). This is ok only if the misalignment
stays the same throughout the execution of the inner-loop, which is why
we have to check that the stride of the dataref in the inner-loop evenly
divides by the vector alignment. */
else if (nested_in_vect_loop_p (loop, stmt))
{
step_preserves_misalignment_p
= (DR_STEP_ALIGNMENT (dr) % vector_alignment) == 0;
if (dump_enabled_p ())
{
if (step_preserves_misalignment_p)
dump_printf_loc (MSG_NOTE, vect_location,
"inner step divides the vector alignment.\n");
else
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"inner step doesn't divide the vector"
" alignment.\n");
}
}
/* Similarly we can only use base and misalignment information relative to
an innermost loop if the misalignment stays the same throughout the
execution of the loop. As above, this is the case if the stride of
the dataref evenly divides by the alignment. */
else
{
poly_uint64 vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo);
step_preserves_misalignment_p
= multiple_p (DR_STEP_ALIGNMENT (dr) * vf, vector_alignment);
if (!step_preserves_misalignment_p && dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"step doesn't divide the vector alignment.\n");
}
unsigned int base_alignment = drb->base_alignment;
unsigned int base_misalignment = drb->base_misalignment;
/* Calculate the maximum of the pooled base address alignment and the
alignment that we can compute for DR itself. */
innermost_loop_behavior **entry = base_alignments->get (drb->base_address);
if (entry && base_alignment < (*entry)->base_alignment)
{
base_alignment = (*entry)->base_alignment;
base_misalignment = (*entry)->base_misalignment;
}
if (drb->offset_alignment < vector_alignment
|| !step_preserves_misalignment_p
/* We need to know whether the step wrt the vectorized loop is
negative when computing the starting misalignment below. */
|| TREE_CODE (drb->step) != INTEGER_CST)
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"Unknown alignment for access: ");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM, ref);
dump_printf (MSG_MISSED_OPTIMIZATION, "\n");
}
return true;
}
if (base_alignment < vector_alignment)
{
unsigned int max_alignment;
tree base = get_base_for_alignment (drb->base_address, &max_alignment);
if (max_alignment < vector_alignment
|| !vect_can_force_dr_alignment_p (base,
vector_alignment * BITS_PER_UNIT))
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"can't force alignment of ref: ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, ref);
dump_printf (MSG_NOTE, "\n");
}
return true;
}
/* Force the alignment of the decl.
NOTE: This is the only change to the code we make during
the analysis phase, before deciding to vectorize the loop. */
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location, "force alignment of ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, ref);
dump_printf (MSG_NOTE, "\n");
}
DR_VECT_AUX (dr)->base_decl = base;
DR_VECT_AUX (dr)->base_misaligned = true;
base_misalignment = 0;
}
poly_int64 misalignment
= base_misalignment + wi::to_poly_offset (drb->init).force_shwi ();
/* If this is a backward running DR then first access in the larger
vectype actually is N-1 elements before the address in the DR.
Adjust misalign accordingly. */
if (tree_int_cst_sgn (drb->step) < 0)
/* PLUS because STEP is negative. */
misalignment += ((TYPE_VECTOR_SUBPARTS (vectype) - 1)
* TREE_INT_CST_LOW (drb->step));
unsigned int const_misalignment;
if (!known_misalignment (misalignment, vector_alignment,
&const_misalignment))
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"Non-constant misalignment for access: ");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM, ref);
dump_printf (MSG_MISSED_OPTIMIZATION, "\n");
}
return true;
}
SET_DR_MISALIGNMENT (dr, const_misalignment);
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"misalign = %d bytes of ref ", DR_MISALIGNMENT (dr));
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM, ref);
dump_printf (MSG_MISSED_OPTIMIZATION, "\n");
}
return true;
}
/* Function vect_update_misalignment_for_peel.
Sets DR's misalignment
- to 0 if it has the same alignment as DR_PEEL,
- to the misalignment computed using NPEEL if DR's salignment is known,
- to -1 (unknown) otherwise.
DR - the data reference whose misalignment is to be adjusted.
DR_PEEL - the data reference whose misalignment is being made
zero in the vector loop by the peel.
NPEEL - the number of iterations in the peel loop if the misalignment
of DR_PEEL is known at compile time. */
static void
vect_update_misalignment_for_peel (struct data_reference *dr,
struct data_reference *dr_peel, int npeel)
{
unsigned int i;
vec<dr_p> same_aligned_drs;
struct data_reference *current_dr;
int dr_size = vect_get_scalar_dr_size (dr);
int dr_peel_size = vect_get_scalar_dr_size (dr_peel);
stmt_vec_info stmt_info = vinfo_for_stmt (vect_dr_stmt (dr));
stmt_vec_info peel_stmt_info = vinfo_for_stmt (vect_dr_stmt (dr_peel));
/* For interleaved data accesses the step in the loop must be multiplied by
the size of the interleaving group. */
if (STMT_VINFO_GROUPED_ACCESS (stmt_info))
dr_size *= DR_GROUP_SIZE (vinfo_for_stmt (DR_GROUP_FIRST_ELEMENT (stmt_info)));
if (STMT_VINFO_GROUPED_ACCESS (peel_stmt_info))
dr_peel_size *= DR_GROUP_SIZE (peel_stmt_info);
/* It can be assumed that the data refs with the same alignment as dr_peel
are aligned in the vector loop. */
same_aligned_drs
= STMT_VINFO_SAME_ALIGN_REFS (vinfo_for_stmt (vect_dr_stmt (dr_peel)));
FOR_EACH_VEC_ELT (same_aligned_drs, i, current_dr)
{
if (current_dr != dr)
continue;
gcc_assert (!known_alignment_for_access_p (dr)
|| !known_alignment_for_access_p (dr_peel)
|| (DR_MISALIGNMENT (dr) / dr_size
== DR_MISALIGNMENT (dr_peel) / dr_peel_size));
SET_DR_MISALIGNMENT (dr, 0);
return;
}
if (known_alignment_for_access_p (dr)
&& known_alignment_for_access_p (dr_peel))
{
bool negative = tree_int_cst_compare (DR_STEP (dr), size_zero_node) < 0;
int misal = DR_MISALIGNMENT (dr);
misal += negative ? -npeel * dr_size : npeel * dr_size;
misal &= DR_TARGET_ALIGNMENT (dr) - 1;
SET_DR_MISALIGNMENT (dr, misal);
return;
}
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location, "Setting misalignment " \
"to unknown (-1).\n");
SET_DR_MISALIGNMENT (dr, DR_MISALIGNMENT_UNKNOWN);
}
/* Function verify_data_ref_alignment
Return TRUE if DR can be handled with respect to alignment. */
static bool
verify_data_ref_alignment (data_reference_p dr)
{
enum dr_alignment_support supportable_dr_alignment
= vect_supportable_dr_alignment (dr, false);
if (!supportable_dr_alignment)
{
if (dump_enabled_p ())
{
if (DR_IS_READ (dr))
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: unsupported unaligned load.");
else
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: unsupported unaligned "
"store.");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_SLIM,
DR_REF (dr));
dump_printf (MSG_MISSED_OPTIMIZATION, "\n");
}
return false;
}
if (supportable_dr_alignment != dr_aligned && dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"Vectorizing an unaligned access.\n");
return true;
}
/* Function vect_verify_datarefs_alignment
Return TRUE if all data references in the loop can be
handled with respect to alignment. */
bool
vect_verify_datarefs_alignment (loop_vec_info vinfo)
{
vec<data_reference_p> datarefs = vinfo->datarefs;
struct data_reference *dr;
unsigned int i;
FOR_EACH_VEC_ELT (datarefs, i, dr)
{
gimple *stmt = vect_dr_stmt (dr);
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
if (!STMT_VINFO_RELEVANT_P (stmt_info))
continue;
/* For interleaving, only the alignment of the first access matters. */
if (STMT_VINFO_GROUPED_ACCESS (stmt_info)
&& DR_GROUP_FIRST_ELEMENT (stmt_info) != stmt)
continue;
/* Strided accesses perform only component accesses, alignment is
irrelevant for them. */
if (STMT_VINFO_STRIDED_P (stmt_info)
&& !STMT_VINFO_GROUPED_ACCESS (stmt_info))
continue;
if (! verify_data_ref_alignment (dr))
return false;
}
return true;
}
/* Given an memory reference EXP return whether its alignment is less
than its size. */
static bool
not_size_aligned (tree exp)
{
if (!tree_fits_uhwi_p (TYPE_SIZE (TREE_TYPE (exp))))
return true;
return (tree_to_uhwi (TYPE_SIZE (TREE_TYPE (exp)))
> get_object_alignment (exp));
}
/* Function vector_alignment_reachable_p
Return true if vector alignment for DR is reachable by peeling
a few loop iterations. Return false otherwise. */
static bool
vector_alignment_reachable_p (struct data_reference *dr)
{
gimple *stmt = vect_dr_stmt (dr);
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
if (STMT_VINFO_GROUPED_ACCESS (stmt_info))
{
/* For interleaved access we peel only if number of iterations in
the prolog loop ({VF - misalignment}), is a multiple of the
number of the interleaved accesses. */
int elem_size, mis_in_elements;
/* FORNOW: handle only known alignment. */
if (!known_alignment_for_access_p (dr))
return false;
poly_uint64 nelements = TYPE_VECTOR_SUBPARTS (vectype);
poly_uint64 vector_size = GET_MODE_SIZE (TYPE_MODE (vectype));
elem_size = vector_element_size (vector_size, nelements);
mis_in_elements = DR_MISALIGNMENT (dr) / elem_size;
if (!multiple_p (nelements - mis_in_elements, DR_GROUP_SIZE (stmt_info)))
return false;
}
/* If misalignment is known at the compile time then allow peeling
only if natural alignment is reachable through peeling. */
if (known_alignment_for_access_p (dr) && !aligned_access_p (dr))
{
HOST_WIDE_INT elmsize =
int_cst_value (TYPE_SIZE_UNIT (TREE_TYPE (vectype)));
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"data size =" HOST_WIDE_INT_PRINT_DEC, elmsize);
dump_printf (MSG_NOTE,
". misalignment = %d.\n", DR_MISALIGNMENT (dr));
}
if (DR_MISALIGNMENT (dr) % elmsize)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"data size does not divide the misalignment.\n");
return false;
}
}
if (!known_alignment_for_access_p (dr))
{
tree type = TREE_TYPE (DR_REF (dr));
bool is_packed = not_size_aligned (DR_REF (dr));
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"Unknown misalignment, %snaturally aligned\n",
is_packed ? "not " : "");
return targetm.vectorize.vector_alignment_reachable (type, is_packed);
}
return true;
}
/* Calculate the cost of the memory access represented by DR. */
static void
vect_get_data_access_cost (struct data_reference *dr,
unsigned int *inside_cost,
unsigned int *outside_cost,
stmt_vector_for_cost *body_cost_vec,
stmt_vector_for_cost *prologue_cost_vec)
{
gimple *stmt = vect_dr_stmt (dr);
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
int ncopies;
if (PURE_SLP_STMT (stmt_info))
ncopies = 1;
else
ncopies = vect_get_num_copies (loop_vinfo, STMT_VINFO_VECTYPE (stmt_info));
if (DR_IS_READ (dr))
vect_get_load_cost (stmt_info, ncopies, true, inside_cost, outside_cost,
prologue_cost_vec, body_cost_vec, false);
else
vect_get_store_cost (stmt_info, ncopies, inside_cost, body_cost_vec);
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"vect_get_data_access_cost: inside_cost = %d, "
"outside_cost = %d.\n", *inside_cost, *outside_cost);
}
typedef struct _vect_peel_info
{
struct data_reference *dr;
int npeel;
unsigned int count;
} *vect_peel_info;
typedef struct _vect_peel_extended_info
{
struct _vect_peel_info peel_info;
unsigned int inside_cost;
unsigned int outside_cost;
} *vect_peel_extended_info;
/* Peeling hashtable helpers. */
struct peel_info_hasher : free_ptr_hash <_vect_peel_info>
{
static inline hashval_t hash (const _vect_peel_info *);
static inline bool equal (const _vect_peel_info *, const _vect_peel_info *);
};
inline hashval_t
peel_info_hasher::hash (const _vect_peel_info *peel_info)
{
return (hashval_t) peel_info->npeel;
}
inline bool
peel_info_hasher::equal (const _vect_peel_info *a, const _vect_peel_info *b)
{
return (a->npeel == b->npeel);
}
/* Insert DR into peeling hash table with NPEEL as key. */
static void
vect_peeling_hash_insert (hash_table<peel_info_hasher> *peeling_htab,
loop_vec_info loop_vinfo, struct data_reference *dr,
int npeel)
{
struct _vect_peel_info elem, *slot;
_vect_peel_info **new_slot;
bool supportable_dr_alignment = vect_supportable_dr_alignment (dr, true);
elem.npeel = npeel;
slot = peeling_htab->find (&elem);
if (slot)
slot->count++;
else
{
slot = XNEW (struct _vect_peel_info);
slot->npeel = npeel;
slot->dr = dr;
slot->count = 1;
new_slot = peeling_htab->find_slot (slot, INSERT);
*new_slot = slot;
}
if (!supportable_dr_alignment
&& unlimited_cost_model (LOOP_VINFO_LOOP (loop_vinfo)))
slot->count += VECT_MAX_COST;
}
/* Traverse peeling hash table to find peeling option that aligns maximum
number of data accesses. */
int
vect_peeling_hash_get_most_frequent (_vect_peel_info **slot,
_vect_peel_extended_info *max)
{
vect_peel_info elem = *slot;
if (elem->count > max->peel_info.count
|| (elem->count == max->peel_info.count
&& max->peel_info.npeel > elem->npeel))
{
max->peel_info.npeel = elem->npeel;
max->peel_info.count = elem->count;
max->peel_info.dr = elem->dr;
}
return 1;
}
/* Get the costs of peeling NPEEL iterations checking data access costs
for all data refs. If UNKNOWN_MISALIGNMENT is true, we assume DR0's
misalignment will be zero after peeling. */
static void
vect_get_peeling_costs_all_drs (vec<data_reference_p> datarefs,
struct data_reference *dr0,
unsigned int *inside_cost,
unsigned int *outside_cost,
stmt_vector_for_cost *body_cost_vec,
stmt_vector_for_cost *prologue_cost_vec,
unsigned int npeel,
bool unknown_misalignment)
{
unsigned i;
data_reference *dr;
FOR_EACH_VEC_ELT (datarefs, i, dr)
{
gimple *stmt = vect_dr_stmt (dr);
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
if (!STMT_VINFO_RELEVANT_P (stmt_info))
continue;
/* For interleaving, only the alignment of the first access
matters. */
if (STMT_VINFO_GROUPED_ACCESS (stmt_info)
&& DR_GROUP_FIRST_ELEMENT (stmt_info) != stmt)
continue;
/* Strided accesses perform only component accesses, alignment is
irrelevant for them. */
if (STMT_VINFO_STRIDED_P (stmt_info)
&& !STMT_VINFO_GROUPED_ACCESS (stmt_info))
continue;
int save_misalignment;
save_misalignment = DR_MISALIGNMENT (dr);
if (npeel == 0)
;
else if (unknown_misalignment && dr == dr0)
SET_DR_MISALIGNMENT (dr, 0);
else
vect_update_misalignment_for_peel (dr, dr0, npeel);
vect_get_data_access_cost (dr, inside_cost, outside_cost,
body_cost_vec, prologue_cost_vec);
SET_DR_MISALIGNMENT (dr, save_misalignment);
}
}
/* Traverse peeling hash table and calculate cost for each peeling option.
Find the one with the lowest cost. */
int
vect_peeling_hash_get_lowest_cost (_vect_peel_info **slot,
_vect_peel_extended_info *min)
{
vect_peel_info elem = *slot;
int dummy;
unsigned int inside_cost = 0, outside_cost = 0;
gimple *stmt = vect_dr_stmt (elem->dr);
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
stmt_vector_for_cost prologue_cost_vec, body_cost_vec,
epilogue_cost_vec;
prologue_cost_vec.create (2);
body_cost_vec.create (2);
epilogue_cost_vec.create (2);
vect_get_peeling_costs_all_drs (LOOP_VINFO_DATAREFS (loop_vinfo),
elem->dr, &inside_cost, &outside_cost,
&body_cost_vec, &prologue_cost_vec,
elem->npeel, false);
body_cost_vec.release ();
outside_cost += vect_get_known_peeling_cost
(loop_vinfo, elem->npeel, &dummy,
&LOOP_VINFO_SCALAR_ITERATION_COST (loop_vinfo),
&prologue_cost_vec, &epilogue_cost_vec);
/* Prologue and epilogue costs are added to the target model later.
These costs depend only on the scalar iteration cost, the
number of peeling iterations finally chosen, and the number of
misaligned statements. So discard the information found here. */
prologue_cost_vec.release ();
epilogue_cost_vec.release ();
if (inside_cost < min->inside_cost
|| (inside_cost == min->inside_cost
&& outside_cost < min->outside_cost))
{
min->inside_cost = inside_cost;
min->outside_cost = outside_cost;
min->peel_info.dr = elem->dr;
min->peel_info.npeel = elem->npeel;
min->peel_info.count = elem->count;
}
return 1;
}
/* Choose best peeling option by traversing peeling hash table and either
choosing an option with the lowest cost (if cost model is enabled) or the
option that aligns as many accesses as possible. */
static struct _vect_peel_extended_info
vect_peeling_hash_choose_best_peeling (hash_table<peel_info_hasher> *peeling_htab,
loop_vec_info loop_vinfo)
{
struct _vect_peel_extended_info res;
res.peel_info.dr = NULL;
if (!unlimited_cost_model (LOOP_VINFO_LOOP (loop_vinfo)))
{
res.inside_cost = INT_MAX;
res.outside_cost = INT_MAX;
peeling_htab->traverse <_vect_peel_extended_info *,
vect_peeling_hash_get_lowest_cost> (&res);
}
else
{
res.peel_info.count = 0;
peeling_htab->traverse <_vect_peel_extended_info *,
vect_peeling_hash_get_most_frequent> (&res);
res.inside_cost = 0;
res.outside_cost = 0;
}
return res;
}
/* Return true if the new peeling NPEEL is supported. */
static bool
vect_peeling_supportable (loop_vec_info loop_vinfo, struct data_reference *dr0,
unsigned npeel)
{
unsigned i;
struct data_reference *dr = NULL;
vec<data_reference_p> datarefs = LOOP_VINFO_DATAREFS (loop_vinfo);
gimple *stmt;
stmt_vec_info stmt_info;
enum dr_alignment_support supportable_dr_alignment;
/* Ensure that all data refs can be vectorized after the peel. */
FOR_EACH_VEC_ELT (datarefs, i, dr)
{
int save_misalignment;
if (dr == dr0)
continue;
stmt = vect_dr_stmt (dr);
stmt_info = vinfo_for_stmt (stmt);
/* For interleaving, only the alignment of the first access
matters. */
if (STMT_VINFO_GROUPED_ACCESS (stmt_info)
&& DR_GROUP_FIRST_ELEMENT (stmt_info) != stmt)
continue;
/* Strided accesses perform only component accesses, alignment is
irrelevant for them. */
if (STMT_VINFO_STRIDED_P (stmt_info)
&& !STMT_VINFO_GROUPED_ACCESS (stmt_info))
continue;
save_misalignment = DR_MISALIGNMENT (dr);
vect_update_misalignment_for_peel (dr, dr0, npeel);
supportable_dr_alignment = vect_supportable_dr_alignment (dr, false);
SET_DR_MISALIGNMENT (dr, save_misalignment);
if (!supportable_dr_alignment)
return false;
}
return true;
}
/* Function vect_enhance_data_refs_alignment
This pass will use loop versioning and loop peeling in order to enhance
the alignment of data references in the loop.
FOR NOW: we assume that whatever versioning/peeling takes place, only the
original loop is to be vectorized. Any other loops that are created by
the transformations performed in this pass - are not supposed to be
vectorized. This restriction will be relaxed.
This pass will require a cost model to guide it whether to apply peeling
or versioning or a combination of the two. For example, the scheme that
intel uses when given a loop with several memory accesses, is as follows:
choose one memory access ('p') which alignment you want to force by doing
peeling. Then, either (1) generate a loop in which 'p' is aligned and all
other accesses are not necessarily aligned, or (2) use loop versioning to
generate one loop in which all accesses are aligned, and another loop in
which only 'p' is necessarily aligned.
("Automatic Intra-Register Vectorization for the Intel Architecture",
Aart J.C. Bik, Milind Girkar, Paul M. Grey and Ximmin Tian, International
Journal of Parallel Programming, Vol. 30, No. 2, April 2002.)
Devising a cost model is the most critical aspect of this work. It will
guide us on which access to peel for, whether to use loop versioning, how
many versions to create, etc. The cost model will probably consist of
generic considerations as well as target specific considerations (on
powerpc for example, misaligned stores are more painful than misaligned
loads).
Here are the general steps involved in alignment enhancements:
-- original loop, before alignment analysis:
for (i=0; i<N; i++){
x = q[i]; # DR_MISALIGNMENT(q) = unknown
p[i] = y; # DR_MISALIGNMENT(p) = unknown
}
-- After vect_compute_data_refs_alignment:
for (i=0; i<N; i++){
x = q[i]; # DR_MISALIGNMENT(q) = 3
p[i] = y; # DR_MISALIGNMENT(p) = unknown
}
-- Possibility 1: we do loop versioning:
if (p is aligned) {
for (i=0; i<N; i++){ # loop 1A
x = q[i]; # DR_MISALIGNMENT(q) = 3
p[i] = y; # DR_MISALIGNMENT(p) = 0
}
}
else {
for (i=0; i<N; i++){ # loop 1B
x = q[i]; # DR_MISALIGNMENT(q) = 3
p[i] = y; # DR_MISALIGNMENT(p) = unaligned
}
}
-- Possibility 2: we do loop peeling:
for (i = 0; i < 3; i++){ # (scalar loop, not to be vectorized).
x = q[i];
p[i] = y;
}
for (i = 3; i < N; i++){ # loop 2A
x = q[i]; # DR_MISALIGNMENT(q) = 0
p[i] = y; # DR_MISALIGNMENT(p) = unknown
}
-- Possibility 3: combination of loop peeling and versioning:
for (i = 0; i < 3; i++){ # (scalar loop, not to be vectorized).
x = q[i];
p[i] = y;
}
if (p is aligned) {
for (i = 3; i<N; i++){ # loop 3A
x = q[i]; # DR_MISALIGNMENT(q) = 0
p[i] = y; # DR_MISALIGNMENT(p) = 0
}
}
else {
for (i = 3; i<N; i++){ # loop 3B
x = q[i]; # DR_MISALIGNMENT(q) = 0
p[i] = y; # DR_MISALIGNMENT(p) = unaligned
}
}
These loops are later passed to loop_transform to be vectorized. The
vectorizer will use the alignment information to guide the transformation
(whether to generate regular loads/stores, or with special handling for
misalignment). */
bool
vect_enhance_data_refs_alignment (loop_vec_info loop_vinfo)
{
vec<data_reference_p> datarefs = LOOP_VINFO_DATAREFS (loop_vinfo);
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
enum dr_alignment_support supportable_dr_alignment;
struct data_reference *dr0 = NULL, *first_store = NULL;
struct data_reference *dr;
unsigned int i, j;
bool do_peeling = false;
bool do_versioning = false;
bool stat;
gimple *stmt;
stmt_vec_info stmt_info;
unsigned int npeel = 0;
bool one_misalignment_known = false;
bool one_misalignment_unknown = false;
bool one_dr_unsupportable = false;
struct data_reference *unsupportable_dr = NULL;
poly_uint64 vf = LOOP_VINFO_VECT_FACTOR (loop_vinfo);
unsigned possible_npeel_number = 1;
tree vectype;
unsigned int mis, same_align_drs_max = 0;
hash_table<peel_info_hasher> peeling_htab (1);
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"=== vect_enhance_data_refs_alignment ===\n");
/* Reset data so we can safely be called multiple times. */
LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo).truncate (0);
LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo) = 0;
/* While cost model enhancements are expected in the future, the high level
view of the code at this time is as follows:
A) If there is a misaligned access then see if peeling to align
this access can make all data references satisfy
vect_supportable_dr_alignment. If so, update data structures
as needed and return true.
B) If peeling wasn't possible and there is a data reference with an
unknown misalignment that does not satisfy vect_supportable_dr_alignment
then see if loop versioning checks can be used to make all data
references satisfy vect_supportable_dr_alignment. If so, update
data structures as needed and return true.
C) If neither peeling nor versioning were successful then return false if
any data reference does not satisfy vect_supportable_dr_alignment.
D) Return true (all data references satisfy vect_supportable_dr_alignment).
Note, Possibility 3 above (which is peeling and versioning together) is not
being done at this time. */
/* (1) Peeling to force alignment. */
/* (1.1) Decide whether to perform peeling, and how many iterations to peel:
Considerations:
+ How many accesses will become aligned due to the peeling
- How many accesses will become unaligned due to the peeling,
and the cost of misaligned accesses.
- The cost of peeling (the extra runtime checks, the increase
in code size). */
FOR_EACH_VEC_ELT (datarefs, i, dr)
{
stmt = vect_dr_stmt (dr);
stmt_info = vinfo_for_stmt (stmt);
if (!STMT_VINFO_RELEVANT_P (stmt_info))
continue;
/* For interleaving, only the alignment of the first access
matters. */
if (STMT_VINFO_GROUPED_ACCESS (stmt_info)
&& DR_GROUP_FIRST_ELEMENT (stmt_info) != stmt)
continue;
/* For invariant accesses there is nothing to enhance. */
if (integer_zerop (DR_STEP (dr)))
continue;
/* Strided accesses perform only component accesses, alignment is
irrelevant for them. */
if (STMT_VINFO_STRIDED_P (stmt_info)
&& !STMT_VINFO_GROUPED_ACCESS (stmt_info))
continue;
supportable_dr_alignment = vect_supportable_dr_alignment (dr, true);
do_peeling = vector_alignment_reachable_p (dr);
if (do_peeling)
{
if (known_alignment_for_access_p (dr))
{
unsigned int npeel_tmp = 0;
bool negative = tree_int_cst_compare (DR_STEP (dr),
size_zero_node) < 0;
vectype = STMT_VINFO_VECTYPE (stmt_info);
unsigned int target_align = DR_TARGET_ALIGNMENT (dr);
unsigned int dr_size = vect_get_scalar_dr_size (dr);
mis = (negative ? DR_MISALIGNMENT (dr) : -DR_MISALIGNMENT (dr));
if (DR_MISALIGNMENT (dr) != 0)
npeel_tmp = (mis & (target_align - 1)) / dr_size;
/* For multiple types, it is possible that the bigger type access
will have more than one peeling option. E.g., a loop with two
types: one of size (vector size / 4), and the other one of
size (vector size / 8). Vectorization factor will 8. If both
accesses are misaligned by 3, the first one needs one scalar
iteration to be aligned, and the second one needs 5. But the
first one will be aligned also by peeling 5 scalar
iterations, and in that case both accesses will be aligned.
Hence, except for the immediate peeling amount, we also want
to try to add full vector size, while we don't exceed
vectorization factor.
We do this automatically for cost model, since we calculate
cost for every peeling option. */
if (unlimited_cost_model (LOOP_VINFO_LOOP (loop_vinfo)))
{
poly_uint64 nscalars = (STMT_SLP_TYPE (stmt_info)
? vf * DR_GROUP_SIZE (stmt_info) : vf);
possible_npeel_number
= vect_get_num_vectors (nscalars, vectype);
/* NPEEL_TMP is 0 when there is no misalignment, but also
allow peeling NELEMENTS. */
if (DR_MISALIGNMENT (dr) == 0)
possible_npeel_number++;
}
/* Save info about DR in the hash table. Also include peeling
amounts according to the explanation above. */
for (j = 0; j < possible_npeel_number; j++)
{
vect_peeling_hash_insert (&peeling_htab, loop_vinfo,
dr, npeel_tmp);
npeel_tmp += target_align / dr_size;
}
one_misalignment_known = true;
}
else
{
/* If we don't know any misalignment values, we prefer
peeling for data-ref that has the maximum number of data-refs
with the same alignment, unless the target prefers to align
stores over load. */
unsigned same_align_drs
= STMT_VINFO_SAME_ALIGN_REFS (stmt_info).length ();
if (!dr0
|| same_align_drs_max < same_align_drs)
{
same_align_drs_max = same_align_drs;
dr0 = dr;
}
/* For data-refs with the same number of related
accesses prefer the one where the misalign
computation will be invariant in the outermost loop. */
else if (same_align_drs_max == same_align_drs)
{
struct loop *ivloop0, *ivloop;
ivloop0 = outermost_invariant_loop_for_expr
(loop, DR_BASE_ADDRESS (dr0));
ivloop = outermost_invariant_loop_for_expr
(loop, DR_BASE_ADDRESS (dr));
if ((ivloop && !ivloop0)
|| (ivloop && ivloop0
&& flow_loop_nested_p (ivloop, ivloop0)))
dr0 = dr;
}
one_misalignment_unknown = true;
/* Check for data refs with unsupportable alignment that
can be peeled. */
if (!supportable_dr_alignment)
{
one_dr_unsupportable = true;
unsupportable_dr = dr;
}
if (!first_store && DR_IS_WRITE (dr))
first_store = dr;
}
}
else
{
if (!aligned_access_p (dr))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"vector alignment may not be reachable\n");
break;
}
}
}
/* Check if we can possibly peel the loop. */
if (!vect_can_advance_ivs_p (loop_vinfo)
|| !slpeel_can_duplicate_loop_p (loop, single_exit (loop))
|| loop->inner)
do_peeling = false;
struct _vect_peel_extended_info peel_for_known_alignment;
struct _vect_peel_extended_info peel_for_unknown_alignment;
struct _vect_peel_extended_info best_peel;
peel_for_unknown_alignment.inside_cost = INT_MAX;
peel_for_unknown_alignment.outside_cost = INT_MAX;
peel_for_unknown_alignment.peel_info.count = 0;
if (do_peeling
&& one_misalignment_unknown)
{
/* Check if the target requires to prefer stores over loads, i.e., if
misaligned stores are more expensive than misaligned loads (taking
drs with same alignment into account). */
unsigned int load_inside_cost = 0;
unsigned int load_outside_cost = 0;
unsigned int store_inside_cost = 0;
unsigned int store_outside_cost = 0;
unsigned int estimated_npeels = vect_vf_for_cost (loop_vinfo) / 2;
stmt_vector_for_cost dummy;
dummy.create (2);
vect_get_peeling_costs_all_drs (datarefs, dr0,
&load_inside_cost,
&load_outside_cost,
&dummy, &dummy, estimated_npeels, true);
dummy.release ();
if (first_store)
{
dummy.create (2);
vect_get_peeling_costs_all_drs (datarefs, first_store,
&store_inside_cost,
&store_outside_cost,
&dummy, &dummy,
estimated_npeels, true);
dummy.release ();
}
else
{
store_inside_cost = INT_MAX;
store_outside_cost = INT_MAX;
}
if (load_inside_cost > store_inside_cost
|| (load_inside_cost == store_inside_cost
&& load_outside_cost > store_outside_cost))
{
dr0 = first_store;
peel_for_unknown_alignment.inside_cost = store_inside_cost;
peel_for_unknown_alignment.outside_cost = store_outside_cost;
}
else
{
peel_for_unknown_alignment.inside_cost = load_inside_cost;
peel_for_unknown_alignment.outside_cost = load_outside_cost;
}
stmt_vector_for_cost prologue_cost_vec, epilogue_cost_vec;
prologue_cost_vec.create (2);
epilogue_cost_vec.create (2);
int dummy2;
peel_for_unknown_alignment.outside_cost += vect_get_known_peeling_cost
(loop_vinfo, estimated_npeels, &dummy2,
&LOOP_VINFO_SCALAR_ITERATION_COST (loop_vinfo),
&prologue_cost_vec, &epilogue_cost_vec);
prologue_cost_vec.release ();
epilogue_cost_vec.release ();
peel_for_unknown_alignment.peel_info.count = 1
+ STMT_VINFO_SAME_ALIGN_REFS
(vinfo_for_stmt (vect_dr_stmt (dr0))).length ();
}
peel_for_unknown_alignment.peel_info.npeel = 0;
peel_for_unknown_alignment.peel_info.dr = dr0;
best_peel = peel_for_unknown_alignment;
peel_for_known_alignment.inside_cost = INT_MAX;
peel_for_known_alignment.outside_cost = INT_MAX;
peel_for_known_alignment.peel_info.count = 0;
peel_for_known_alignment.peel_info.dr = NULL;
if (do_peeling && one_misalignment_known)
{
/* Peeling is possible, but there is no data access that is not supported
unless aligned. So we try to choose the best possible peeling from
the hash table. */
peel_for_known_alignment = vect_peeling_hash_choose_best_peeling
(&peeling_htab, loop_vinfo);
}
/* Compare costs of peeling for known and unknown alignment. */
if (peel_for_known_alignment.peel_info.dr != NULL
&& peel_for_unknown_alignment.inside_cost
>= peel_for_known_alignment.inside_cost)
{
best_peel = peel_for_known_alignment;
/* If the best peeling for known alignment has NPEEL == 0, perform no
peeling at all except if there is an unsupportable dr that we can
align. */
if (best_peel.peel_info.npeel == 0 && !one_dr_unsupportable)
do_peeling = false;
}
/* If there is an unsupportable data ref, prefer this over all choices so far
since we'd have to discard a chosen peeling except when it accidentally
aligned the unsupportable data ref. */
if (one_dr_unsupportable)
dr0 = unsupportable_dr;
else if (do_peeling)
{
/* Calculate the penalty for no peeling, i.e. leaving everything as-is.
TODO: Use nopeel_outside_cost or get rid of it? */
unsigned nopeel_inside_cost = 0;
unsigned nopeel_outside_cost = 0;
stmt_vector_for_cost dummy;
dummy.create (2);
vect_get_peeling_costs_all_drs (datarefs, NULL, &nopeel_inside_cost,
&nopeel_outside_cost, &dummy, &dummy,
0, false);
dummy.release ();
/* Add epilogue costs. As we do not peel for alignment here, no prologue
costs will be recorded. */
stmt_vector_for_cost prologue_cost_vec, epilogue_cost_vec;
prologue_cost_vec.create (2);
epilogue_cost_vec.create (2);
int dummy2;
nopeel_outside_cost += vect_get_known_peeling_cost
(loop_vinfo, 0, &dummy2,
&LOOP_VINFO_SCALAR_ITERATION_COST (loop_vinfo),
&prologue_cost_vec, &epilogue_cost_vec);
prologue_cost_vec.release ();
epilogue_cost_vec.release ();
npeel = best_peel.peel_info.npeel;
dr0 = best_peel.peel_info.dr;
/* If no peeling is not more expensive than the best peeling we
have so far, don't perform any peeling. */
if (nopeel_inside_cost <= best_peel.inside_cost)
do_peeling = false;
}
if (do_peeling)
{
stmt = vect_dr_stmt (dr0);
stmt_info = vinfo_for_stmt (stmt);
vectype = STMT_VINFO_VECTYPE (stmt_info);
if (known_alignment_for_access_p (dr0))
{
bool negative = tree_int_cst_compare (DR_STEP (dr0),
size_zero_node) < 0;
if (!npeel)
{
/* Since it's known at compile time, compute the number of
iterations in the peeled loop (the peeling factor) for use in
updating DR_MISALIGNMENT values. The peeling factor is the
vectorization factor minus the misalignment as an element
count. */
mis = negative ? DR_MISALIGNMENT (dr0) : -DR_MISALIGNMENT (dr0);
unsigned int target_align = DR_TARGET_ALIGNMENT (dr0);
npeel = ((mis & (target_align - 1))
/ vect_get_scalar_dr_size (dr0));
}
/* For interleaved data access every iteration accesses all the
members of the group, therefore we divide the number of iterations
by the group size. */
stmt_info = vinfo_for_stmt (vect_dr_stmt (dr0));
if (STMT_VINFO_GROUPED_ACCESS (stmt_info))
npeel /= DR_GROUP_SIZE (stmt_info);
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"Try peeling by %d\n", npeel);
}
/* Ensure that all datarefs can be vectorized after the peel. */
if (!vect_peeling_supportable (loop_vinfo, dr0, npeel))
do_peeling = false;
/* Check if all datarefs are supportable and log. */
if (do_peeling && known_alignment_for_access_p (dr0) && npeel == 0)
{
stat = vect_verify_datarefs_alignment (loop_vinfo);
if (!stat)
do_peeling = false;
else
return stat;
}
/* Cost model #1 - honor --param vect-max-peeling-for-alignment. */
if (do_peeling)
{
unsigned max_allowed_peel
= PARAM_VALUE (PARAM_VECT_MAX_PEELING_FOR_ALIGNMENT);
if (max_allowed_peel != (unsigned)-1)
{
unsigned max_peel = npeel;
if (max_peel == 0)
{
unsigned int target_align = DR_TARGET_ALIGNMENT (dr0);
max_peel = target_align / vect_get_scalar_dr_size (dr0) - 1;
}
if (max_peel > max_allowed_peel)
{
do_peeling = false;
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"Disable peeling, max peels reached: %d\n", max_peel);
}
}
}
/* Cost model #2 - if peeling may result in a remaining loop not
iterating enough to be vectorized then do not peel. Since this
is a cost heuristic rather than a correctness decision, use the
most likely runtime value for variable vectorization factors. */
if (do_peeling
&& LOOP_VINFO_NITERS_KNOWN_P (loop_vinfo))
{
unsigned int assumed_vf = vect_vf_for_cost (loop_vinfo);
unsigned int max_peel = npeel == 0 ? assumed_vf - 1 : npeel;
if ((unsigned HOST_WIDE_INT) LOOP_VINFO_INT_NITERS (loop_vinfo)
< assumed_vf + max_peel)
do_peeling = false;
}
if (do_peeling)
{
/* (1.2) Update the DR_MISALIGNMENT of each data reference DR_i.
If the misalignment of DR_i is identical to that of dr0 then set
DR_MISALIGNMENT (DR_i) to zero. If the misalignment of DR_i and
dr0 are known at compile time then increment DR_MISALIGNMENT (DR_i)
by the peeling factor times the element size of DR_i (MOD the
vectorization factor times the size). Otherwise, the
misalignment of DR_i must be set to unknown. */
FOR_EACH_VEC_ELT (datarefs, i, dr)
if (dr != dr0)
{
/* Strided accesses perform only component accesses, alignment
is irrelevant for them. */
stmt_info = vinfo_for_stmt (vect_dr_stmt (dr));
if (STMT_VINFO_STRIDED_P (stmt_info)
&& !STMT_VINFO_GROUPED_ACCESS (stmt_info))
continue;
vect_update_misalignment_for_peel (dr, dr0, npeel);
}
LOOP_VINFO_UNALIGNED_DR (loop_vinfo) = dr0;
if (npeel)
LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo) = npeel;
else
LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo)
= DR_MISALIGNMENT (dr0);
SET_DR_MISALIGNMENT (dr0, 0);
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"Alignment of access forced using peeling.\n");
dump_printf_loc (MSG_NOTE, vect_location,
"Peeling for alignment will be applied.\n");
}
/* The inside-loop cost will be accounted for in vectorizable_load
and vectorizable_store correctly with adjusted alignments.
Drop the body_cst_vec on the floor here. */
stat = vect_verify_datarefs_alignment (loop_vinfo);
gcc_assert (stat);
return stat;
}
}
/* (2) Versioning to force alignment. */
/* Try versioning if:
1) optimize loop for speed
2) there is at least one unsupported misaligned data ref with an unknown
misalignment, and
3) all misaligned data refs with a known misalignment are supported, and
4) the number of runtime alignment checks is within reason. */
do_versioning =
optimize_loop_nest_for_speed_p (loop)
&& (!loop->inner); /* FORNOW */
if (do_versioning)
{
FOR_EACH_VEC_ELT (datarefs, i, dr)
{
stmt = vect_dr_stmt (dr);
stmt_info = vinfo_for_stmt (stmt);
/* For interleaving, only the alignment of the first access
matters. */
if (aligned_access_p (dr)
|| (STMT_VINFO_GROUPED_ACCESS (stmt_info)
&& DR_GROUP_FIRST_ELEMENT (stmt_info) != stmt))
continue;
if (STMT_VINFO_STRIDED_P (stmt_info))
{
/* Strided loads perform only component accesses, alignment is
irrelevant for them. */
if (!STMT_VINFO_GROUPED_ACCESS (stmt_info))
continue;
do_versioning = false;
break;
}
supportable_dr_alignment = vect_supportable_dr_alignment (dr, false);
if (!supportable_dr_alignment)
{
gimple *stmt;
int mask;
tree vectype;
if (known_alignment_for_access_p (dr)
|| LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo).length ()
>= (unsigned) PARAM_VALUE (PARAM_VECT_MAX_VERSION_FOR_ALIGNMENT_CHECKS))
{
do_versioning = false;
break;
}
stmt = vect_dr_stmt (dr);
vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt));
gcc_assert (vectype);
/* At present we don't support versioning for alignment
with variable VF, since there's no guarantee that the
VF is a power of two. We could relax this if we added
a way of enforcing a power-of-two size. */
unsigned HOST_WIDE_INT size;
if (!GET_MODE_SIZE (TYPE_MODE (vectype)).is_constant (&size))
{
do_versioning = false;
break;
}
/* The rightmost bits of an aligned address must be zeros.
Construct the mask needed for this test. For example,
GET_MODE_SIZE for the vector mode V4SI is 16 bytes so the
mask must be 15 = 0xf. */
mask = size - 1;
/* FORNOW: use the same mask to test all potentially unaligned
references in the loop. The vectorizer currently supports
a single vector size, see the reference to
GET_MODE_NUNITS (TYPE_MODE (vectype)) where the
vectorization factor is computed. */
gcc_assert (!LOOP_VINFO_PTR_MASK (loop_vinfo)
|| LOOP_VINFO_PTR_MASK (loop_vinfo) == mask);
LOOP_VINFO_PTR_MASK (loop_vinfo) = mask;
LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo).safe_push (
vect_dr_stmt (dr));
}
}
/* Versioning requires at least one misaligned data reference. */
if (!LOOP_REQUIRES_VERSIONING_FOR_ALIGNMENT (loop_vinfo))
do_versioning = false;
else if (!do_versioning)
LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo).truncate (0);
}
if (do_versioning)
{
vec<gimple *> may_misalign_stmts
= LOOP_VINFO_MAY_MISALIGN_STMTS (loop_vinfo);
gimple *stmt;
/* It can now be assumed that the data references in the statements
in LOOP_VINFO_MAY_MISALIGN_STMTS will be aligned in the version
of the loop being vectorized. */
FOR_EACH_VEC_ELT (may_misalign_stmts, i, stmt)
{
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
dr = STMT_VINFO_DATA_REF (stmt_info);
SET_DR_MISALIGNMENT (dr, 0);
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"Alignment of access forced using versioning.\n");
}
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"Versioning for alignment will be applied.\n");
/* Peeling and versioning can't be done together at this time. */
gcc_assert (! (do_peeling && do_versioning));
stat = vect_verify_datarefs_alignment (loop_vinfo);
gcc_assert (stat);
return stat;
}
/* This point is reached if neither peeling nor versioning is being done. */
gcc_assert (! (do_peeling || do_versioning));
stat = vect_verify_datarefs_alignment (loop_vinfo);
return stat;
}
/* Function vect_find_same_alignment_drs.
Update group and alignment relations according to the chosen
vectorization factor. */
static void
vect_find_same_alignment_drs (struct data_dependence_relation *ddr)
{
struct data_reference *dra = DDR_A (ddr);
struct data_reference *drb = DDR_B (ddr);
stmt_vec_info stmtinfo_a = vinfo_for_stmt (vect_dr_stmt (dra));
stmt_vec_info stmtinfo_b = vinfo_for_stmt (vect_dr_stmt (drb));
if (DDR_ARE_DEPENDENT (ddr) == chrec_known)
return;
if (dra == drb)
return;
if (!operand_equal_p (DR_BASE_ADDRESS (dra), DR_BASE_ADDRESS (drb), 0)
|| !operand_equal_p (DR_OFFSET (dra), DR_OFFSET (drb), 0)
|| !operand_equal_p (DR_STEP (dra), DR_STEP (drb), 0))
return;
/* Two references with distance zero have the same alignment. */
poly_offset_int diff = (wi::to_poly_offset (DR_INIT (dra))
- wi::to_poly_offset (DR_INIT (drb)));
if (maybe_ne (diff, 0))
{
/* Get the wider of the two alignments. */
unsigned int align_a = (vect_calculate_target_alignment (dra)
/ BITS_PER_UNIT);
unsigned int align_b = (vect_calculate_target_alignment (drb)
/ BITS_PER_UNIT);
unsigned int max_align = MAX (align_a, align_b);
/* Require the gap to be a multiple of the larger vector alignment. */
if (!multiple_p (diff, max_align))
return;
}
STMT_VINFO_SAME_ALIGN_REFS (stmtinfo_a).safe_push (drb);
STMT_VINFO_SAME_ALIGN_REFS (stmtinfo_b).safe_push (dra);
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"accesses have the same alignment: ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (dra));
dump_printf (MSG_NOTE, " and ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (drb));
dump_printf (MSG_NOTE, "\n");
}
}
/* Function vect_analyze_data_refs_alignment
Analyze the alignment of the data-references in the loop.
Return FALSE if a data reference is found that cannot be vectorized. */
bool
vect_analyze_data_refs_alignment (loop_vec_info vinfo)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"=== vect_analyze_data_refs_alignment ===\n");
/* Mark groups of data references with same alignment using
data dependence information. */
vec<ddr_p> ddrs = vinfo->ddrs;
struct data_dependence_relation *ddr;
unsigned int i;
FOR_EACH_VEC_ELT (ddrs, i, ddr)
vect_find_same_alignment_drs (ddr);
vec<data_reference_p> datarefs = vinfo->datarefs;
struct data_reference *dr;
vect_record_base_alignments (vinfo);
FOR_EACH_VEC_ELT (datarefs, i, dr)
{
stmt_vec_info stmt_info = vinfo_for_stmt (vect_dr_stmt (dr));
if (STMT_VINFO_VECTORIZABLE (stmt_info)
&& !vect_compute_data_ref_alignment (dr))
{
/* Strided accesses perform only component accesses, misalignment
information is irrelevant for them. */
if (STMT_VINFO_STRIDED_P (stmt_info)
&& !STMT_VINFO_GROUPED_ACCESS (stmt_info))
continue;
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: can't calculate alignment "
"for data ref.\n");
return false;
}
}
return true;
}
/* Analyze alignment of DRs of stmts in NODE. */
static bool
vect_slp_analyze_and_verify_node_alignment (slp_tree node)
{
/* We vectorize from the first scalar stmt in the node unless
the node is permuted in which case we start from the first
element in the group. */
gimple *first_stmt = SLP_TREE_SCALAR_STMTS (node)[0];
data_reference_p first_dr = STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt));
if (SLP_TREE_LOAD_PERMUTATION (node).exists ())
first_stmt = DR_GROUP_FIRST_ELEMENT (vinfo_for_stmt (first_stmt));
data_reference_p dr = STMT_VINFO_DATA_REF (vinfo_for_stmt (first_stmt));
if (! vect_compute_data_ref_alignment (dr)
/* For creating the data-ref pointer we need alignment of the
first element anyway. */
|| (dr != first_dr
&& ! vect_compute_data_ref_alignment (first_dr))
|| ! verify_data_ref_alignment (dr))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: bad data alignment in basic "
"block.\n");
return false;
}
return true;
}
/* Function vect_slp_analyze_instance_alignment
Analyze the alignment of the data-references in the SLP instance.
Return FALSE if a data reference is found that cannot be vectorized. */
bool
vect_slp_analyze_and_verify_instance_alignment (slp_instance instance)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"=== vect_slp_analyze_and_verify_instance_alignment ===\n");
slp_tree node;
unsigned i;
FOR_EACH_VEC_ELT (SLP_INSTANCE_LOADS (instance), i, node)
if (! vect_slp_analyze_and_verify_node_alignment (node))
return false;
node = SLP_INSTANCE_TREE (instance);
if (STMT_VINFO_DATA_REF (vinfo_for_stmt (SLP_TREE_SCALAR_STMTS (node)[0]))
&& ! vect_slp_analyze_and_verify_node_alignment
(SLP_INSTANCE_TREE (instance)))
return false;
return true;
}
/* Analyze groups of accesses: check that DR belongs to a group of
accesses of legal size, step, etc. Detect gaps, single element
interleaving, and other special cases. Set grouped access info.
Collect groups of strided stores for further use in SLP analysis.
Worker for vect_analyze_group_access. */
static bool
vect_analyze_group_access_1 (struct data_reference *dr)
{
tree step = DR_STEP (dr);
tree scalar_type = TREE_TYPE (DR_REF (dr));
HOST_WIDE_INT type_size = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (scalar_type));
gimple *stmt = vect_dr_stmt (dr);
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
bb_vec_info bb_vinfo = STMT_VINFO_BB_VINFO (stmt_info);
HOST_WIDE_INT dr_step = -1;
HOST_WIDE_INT groupsize, last_accessed_element = 1;
bool slp_impossible = false;
/* For interleaving, GROUPSIZE is STEP counted in elements, i.e., the
size of the interleaving group (including gaps). */
if (tree_fits_shwi_p (step))
{
dr_step = tree_to_shwi (step);
/* Check that STEP is a multiple of type size. Otherwise there is
a non-element-sized gap at the end of the group which we
cannot represent in DR_GROUP_GAP or DR_GROUP_SIZE.
??? As we can handle non-constant step fine here we should
simply remove uses of DR_GROUP_GAP between the last and first
element and instead rely on DR_STEP. DR_GROUP_SIZE then would
simply not include that gap. */
if ((dr_step % type_size) != 0)
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"Step ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, step);
dump_printf (MSG_NOTE,
" is not a multiple of the element size for ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (dr));
dump_printf (MSG_NOTE, "\n");
}
return false;
}
groupsize = absu_hwi (dr_step) / type_size;
}
else
groupsize = 0;
/* Not consecutive access is possible only if it is a part of interleaving. */
if (!DR_GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt)))
{
/* Check if it this DR is a part of interleaving, and is a single
element of the group that is accessed in the loop. */
/* Gaps are supported only for loads. STEP must be a multiple of the type
size. */
if (DR_IS_READ (dr)
&& (dr_step % type_size) == 0
&& groupsize > 0)
{
DR_GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt)) = stmt;
DR_GROUP_SIZE (vinfo_for_stmt (stmt)) = groupsize;
DR_GROUP_GAP (stmt_info) = groupsize - 1;
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"Detected single element interleaving ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (dr));
dump_printf (MSG_NOTE, " step ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, step);
dump_printf (MSG_NOTE, "\n");
}
return true;
}
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not consecutive access ");
dump_gimple_stmt (MSG_MISSED_OPTIMIZATION, TDF_SLIM, stmt, 0);
}
if (bb_vinfo)
{
/* Mark the statement as unvectorizable. */
STMT_VINFO_VECTORIZABLE (vinfo_for_stmt (vect_dr_stmt (dr))) = false;
return true;
}
dump_printf_loc (MSG_NOTE, vect_location, "using strided accesses\n");
STMT_VINFO_STRIDED_P (stmt_info) = true;
return true;
}
if (DR_GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt)) == stmt)
{
/* First stmt in the interleaving chain. Check the chain. */
gimple *next = DR_GROUP_NEXT_ELEMENT (vinfo_for_stmt (stmt));
struct data_reference *data_ref = dr;
unsigned int count = 1;
tree prev_init = DR_INIT (data_ref);
gimple *prev = stmt;
HOST_WIDE_INT diff, gaps = 0;
/* By construction, all group members have INTEGER_CST DR_INITs. */
while (next)
{
/* Skip same data-refs. In case that two or more stmts share
data-ref (supported only for loads), we vectorize only the first
stmt, and the rest get their vectorized loads from the first
one. */
if (!tree_int_cst_compare (DR_INIT (data_ref),
DR_INIT (STMT_VINFO_DATA_REF (
vinfo_for_stmt (next)))))
{
if (DR_IS_WRITE (data_ref))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"Two store stmts share the same dr.\n");
return false;
}
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"Two or more load stmts share the same dr.\n");
/* For load use the same data-ref load. */
DR_GROUP_SAME_DR_STMT (vinfo_for_stmt (next)) = prev;
prev = next;
next = DR_GROUP_NEXT_ELEMENT (vinfo_for_stmt (next));
continue;
}
prev = next;
data_ref = STMT_VINFO_DATA_REF (vinfo_for_stmt (next));
/* All group members have the same STEP by construction. */
gcc_checking_assert (operand_equal_p (DR_STEP (data_ref), step, 0));
/* Check that the distance between two accesses is equal to the type
size. Otherwise, we have gaps. */
diff = (TREE_INT_CST_LOW (DR_INIT (data_ref))
- TREE_INT_CST_LOW (prev_init)) / type_size;
if (diff != 1)
{
/* FORNOW: SLP of accesses with gaps is not supported. */
slp_impossible = true;
if (DR_IS_WRITE (data_ref))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"interleaved store with gaps\n");
return false;
}
gaps += diff - 1;
}
last_accessed_element += diff;
/* Store the gap from the previous member of the group. If there is no
gap in the access, DR_GROUP_GAP is always 1. */
DR_GROUP_GAP (vinfo_for_stmt (next)) = diff;
prev_init = DR_INIT (data_ref);
next = DR_GROUP_NEXT_ELEMENT (vinfo_for_stmt (next));
/* Count the number of data-refs in the chain. */
count++;
}
if (groupsize == 0)
groupsize = count + gaps;
/* This could be UINT_MAX but as we are generating code in a very
inefficient way we have to cap earlier. See PR78699 for example. */
if (groupsize > 4096)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"group is too large\n");
return false;
}
/* Check that the size of the interleaving is equal to count for stores,
i.e., that there are no gaps. */
if (groupsize != count
&& !DR_IS_READ (dr))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"interleaved store with gaps\n");
return false;
}
/* If there is a gap after the last load in the group it is the
difference between the groupsize and the last accessed
element.
When there is no gap, this difference should be 0. */
DR_GROUP_GAP (vinfo_for_stmt (stmt)) = groupsize - last_accessed_element;
DR_GROUP_SIZE (vinfo_for_stmt (stmt)) = groupsize;
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"Detected interleaving ");
if (DR_IS_READ (dr))
dump_printf (MSG_NOTE, "load ");
else
dump_printf (MSG_NOTE, "store ");
dump_printf (MSG_NOTE, "of size %u starting with ",
(unsigned)groupsize);
dump_gimple_stmt (MSG_NOTE, TDF_SLIM, stmt, 0);
if (DR_GROUP_GAP (vinfo_for_stmt (stmt)) != 0)
dump_printf_loc (MSG_NOTE, vect_location,
"There is a gap of %u elements after the group\n",
DR_GROUP_GAP (vinfo_for_stmt (stmt)));
}
/* SLP: create an SLP data structure for every interleaving group of
stores for further analysis in vect_analyse_slp. */
if (DR_IS_WRITE (dr) && !slp_impossible)
{
if (loop_vinfo)
LOOP_VINFO_GROUPED_STORES (loop_vinfo).safe_push (stmt);
if (bb_vinfo)
BB_VINFO_GROUPED_STORES (bb_vinfo).safe_push (stmt);
}
}
return true;
}
/* Analyze groups of accesses: check that DR belongs to a group of
accesses of legal size, step, etc. Detect gaps, single element
interleaving, and other special cases. Set grouped access info.
Collect groups of strided stores for further use in SLP analysis. */
static bool
vect_analyze_group_access (struct data_reference *dr)
{
if (!vect_analyze_group_access_1 (dr))
{
/* Dissolve the group if present. */
gimple *next;
gimple *stmt = DR_GROUP_FIRST_ELEMENT (vinfo_for_stmt (vect_dr_stmt (dr)));
while (stmt)
{
stmt_vec_info vinfo = vinfo_for_stmt (stmt);
next = DR_GROUP_NEXT_ELEMENT (vinfo);
DR_GROUP_FIRST_ELEMENT (vinfo) = NULL;
DR_GROUP_NEXT_ELEMENT (vinfo) = NULL;
stmt = next;
}
return false;
}
return true;
}
/* Analyze the access pattern of the data-reference DR.
In case of non-consecutive accesses call vect_analyze_group_access() to
analyze groups of accesses. */
static bool
vect_analyze_data_ref_access (struct data_reference *dr)
{
tree step = DR_STEP (dr);
tree scalar_type = TREE_TYPE (DR_REF (dr));
gimple *stmt = vect_dr_stmt (dr);
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
struct loop *loop = NULL;
if (STMT_VINFO_GATHER_SCATTER_P (stmt_info))
return true;
if (loop_vinfo)
loop = LOOP_VINFO_LOOP (loop_vinfo);
if (loop_vinfo && !step)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"bad data-ref access in loop\n");
return false;
}
/* Allow loads with zero step in inner-loop vectorization. */
if (loop_vinfo && integer_zerop (step))
{
DR_GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt)) = NULL;
if (!nested_in_vect_loop_p (loop, stmt))
return DR_IS_READ (dr);
/* Allow references with zero step for outer loops marked
with pragma omp simd only - it guarantees absence of
loop-carried dependencies between inner loop iterations. */
if (loop->safelen < 2)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"zero step in inner loop of nest\n");
return false;
}
}
if (loop && nested_in_vect_loop_p (loop, stmt))
{
/* Interleaved accesses are not yet supported within outer-loop
vectorization for references in the inner-loop. */
DR_GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt)) = NULL;
/* For the rest of the analysis we use the outer-loop step. */
step = STMT_VINFO_DR_STEP (stmt_info);
if (integer_zerop (step))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"zero step in outer loop.\n");
return DR_IS_READ (dr);
}
}
/* Consecutive? */
if (TREE_CODE (step) == INTEGER_CST)
{
HOST_WIDE_INT dr_step = TREE_INT_CST_LOW (step);
if (!tree_int_cst_compare (step, TYPE_SIZE_UNIT (scalar_type))
|| (dr_step < 0
&& !compare_tree_int (TYPE_SIZE_UNIT (scalar_type), -dr_step)))
{
/* Mark that it is not interleaving. */
DR_GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt)) = NULL;
return true;
}
}
if (loop && nested_in_vect_loop_p (loop, stmt))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"grouped access in outer loop.\n");
return false;
}
/* Assume this is a DR handled by non-constant strided load case. */
if (TREE_CODE (step) != INTEGER_CST)
return (STMT_VINFO_STRIDED_P (stmt_info)
&& (!STMT_VINFO_GROUPED_ACCESS (stmt_info)
|| vect_analyze_group_access (dr)));
/* Not consecutive access - check if it's a part of interleaving group. */
return vect_analyze_group_access (dr);
}
/* Compare two data-references DRA and DRB to group them into chunks
suitable for grouping. */
static int
dr_group_sort_cmp (const void *dra_, const void *drb_)
{
data_reference_p dra = *(data_reference_p *)const_cast<void *>(dra_);
data_reference_p drb = *(data_reference_p *)const_cast<void *>(drb_);
int cmp;
/* Stabilize sort. */
if (dra == drb)
return 0;
/* DRs in different loops never belong to the same group. */
loop_p loopa = gimple_bb (DR_STMT (dra))->loop_father;
loop_p loopb = gimple_bb (DR_STMT (drb))->loop_father;
if (loopa != loopb)
return loopa->num < loopb->num ? -1 : 1;
/* Ordering of DRs according to base. */
cmp = data_ref_compare_tree (DR_BASE_ADDRESS (dra),
DR_BASE_ADDRESS (drb));
if (cmp != 0)
return cmp;
/* And according to DR_OFFSET. */
cmp = data_ref_compare_tree (DR_OFFSET (dra), DR_OFFSET (drb));
if (cmp != 0)
return cmp;
/* Put reads before writes. */
if (DR_IS_READ (dra) != DR_IS_READ (drb))
return DR_IS_READ (dra) ? -1 : 1;
/* Then sort after access size. */
cmp = data_ref_compare_tree (TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (dra))),
TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (drb))));
if (cmp != 0)
return cmp;
/* And after step. */
cmp = data_ref_compare_tree (DR_STEP (dra), DR_STEP (drb));
if (cmp != 0)
return cmp;
/* Then sort after DR_INIT. In case of identical DRs sort after stmt UID. */
cmp = data_ref_compare_tree (DR_INIT (dra), DR_INIT (drb));
if (cmp == 0)
return gimple_uid (DR_STMT (dra)) < gimple_uid (DR_STMT (drb)) ? -1 : 1;
return cmp;
}
/* If OP is the result of a conversion, return the unconverted value,
otherwise return null. */
static tree
strip_conversion (tree op)
{
if (TREE_CODE (op) != SSA_NAME)
return NULL_TREE;
gimple *stmt = SSA_NAME_DEF_STMT (op);
if (!is_gimple_assign (stmt)
|| !CONVERT_EXPR_CODE_P (gimple_assign_rhs_code (stmt)))
return NULL_TREE;
return gimple_assign_rhs1 (stmt);
}
/* Return true if vectorizable_* routines can handle statements STMT1
and STMT2 being in a single group. */
static bool
can_group_stmts_p (gimple *stmt1, gimple *stmt2)
{
if (gimple_assign_single_p (stmt1))
return gimple_assign_single_p (stmt2);
if (is_gimple_call (stmt1) && gimple_call_internal_p (stmt1))
{
/* Check for two masked loads or two masked stores. */
if (!is_gimple_call (stmt2) || !gimple_call_internal_p (stmt2))
return false;
internal_fn ifn = gimple_call_internal_fn (stmt1);
if (ifn != IFN_MASK_LOAD && ifn != IFN_MASK_STORE)
return false;
if (ifn != gimple_call_internal_fn (stmt2))
return false;
/* Check that the masks are the same. Cope with casts of masks,
like those created by build_mask_conversion. */
tree mask1 = gimple_call_arg (stmt1, 2);
tree mask2 = gimple_call_arg (stmt2, 2);
if (!operand_equal_p (mask1, mask2, 0))
{
mask1 = strip_conversion (mask1);
if (!mask1)
return false;
mask2 = strip_conversion (mask2);
if (!mask2)
return false;
if (!operand_equal_p (mask1, mask2, 0))
return false;
}
return true;
}
return false;
}
/* Function vect_analyze_data_ref_accesses.
Analyze the access pattern of all the data references in the loop.
FORNOW: the only access pattern that is considered vectorizable is a
simple step 1 (consecutive) access.
FORNOW: handle only arrays and pointer accesses. */
bool
vect_analyze_data_ref_accesses (vec_info *vinfo)
{
unsigned int i;
vec<data_reference_p> datarefs = vinfo->datarefs;
struct data_reference *dr;
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"=== vect_analyze_data_ref_accesses ===\n");
if (datarefs.is_empty ())
return true;
/* Sort the array of datarefs to make building the interleaving chains
linear. Don't modify the original vector's order, it is needed for
determining what dependencies are reversed. */
vec<data_reference_p> datarefs_copy = datarefs.copy ();
datarefs_copy.qsort (dr_group_sort_cmp);
/* Build the interleaving chains. */
for (i = 0; i < datarefs_copy.length () - 1;)
{
data_reference_p dra = datarefs_copy[i];
stmt_vec_info stmtinfo_a = vinfo_for_stmt (vect_dr_stmt (dra));
stmt_vec_info lastinfo = NULL;
if (!STMT_VINFO_VECTORIZABLE (stmtinfo_a)
|| STMT_VINFO_GATHER_SCATTER_P (stmtinfo_a))
{
++i;
continue;
}
for (i = i + 1; i < datarefs_copy.length (); ++i)
{
data_reference_p drb = datarefs_copy[i];
stmt_vec_info stmtinfo_b = vinfo_for_stmt (vect_dr_stmt (drb));
if (!STMT_VINFO_VECTORIZABLE (stmtinfo_b)
|| STMT_VINFO_GATHER_SCATTER_P (stmtinfo_b))
break;
/* ??? Imperfect sorting (non-compatible types, non-modulo
accesses, same accesses) can lead to a group to be artificially
split here as we don't just skip over those. If it really
matters we can push those to a worklist and re-iterate
over them. The we can just skip ahead to the next DR here. */
/* DRs in a different loop should not be put into the same
interleaving group. */
if (gimple_bb (DR_STMT (dra))->loop_father
!= gimple_bb (DR_STMT (drb))->loop_father)
break;
/* Check that the data-refs have same first location (except init)
and they are both either store or load (not load and store,
not masked loads or stores). */
if (DR_IS_READ (dra) != DR_IS_READ (drb)
|| data_ref_compare_tree (DR_BASE_ADDRESS (dra),
DR_BASE_ADDRESS (drb)) != 0
|| data_ref_compare_tree (DR_OFFSET (dra), DR_OFFSET (drb)) != 0
|| !can_group_stmts_p (vect_dr_stmt (dra), vect_dr_stmt (drb)))
break;
/* Check that the data-refs have the same constant size. */
tree sza = TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (dra)));
tree szb = TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (drb)));
if (!tree_fits_uhwi_p (sza)
|| !tree_fits_uhwi_p (szb)
|| !tree_int_cst_equal (sza, szb))
break;
/* Check that the data-refs have the same step. */
if (data_ref_compare_tree (DR_STEP (dra), DR_STEP (drb)) != 0)
break;
/* Check the types are compatible.
??? We don't distinguish this during sorting. */
if (!types_compatible_p (TREE_TYPE (DR_REF (dra)),
TREE_TYPE (DR_REF (drb))))
break;
/* Check that the DR_INITs are compile-time constants. */
if (TREE_CODE (DR_INIT (dra)) != INTEGER_CST
|| TREE_CODE (DR_INIT (drb)) != INTEGER_CST)
break;
/* Sorting has ensured that DR_INIT (dra) <= DR_INIT (drb). */
HOST_WIDE_INT init_a = TREE_INT_CST_LOW (DR_INIT (dra));
HOST_WIDE_INT init_b = TREE_INT_CST_LOW (DR_INIT (drb));
HOST_WIDE_INT init_prev
= TREE_INT_CST_LOW (DR_INIT (datarefs_copy[i-1]));
gcc_assert (init_a <= init_b
&& init_a <= init_prev
&& init_prev <= init_b);
/* Do not place the same access in the interleaving chain twice. */
if (init_b == init_prev)
{
gcc_assert (gimple_uid (DR_STMT (datarefs_copy[i-1]))
< gimple_uid (DR_STMT (drb)));
/* ??? For now we simply "drop" the later reference which is
otherwise the same rather than finishing off this group.
In the end we'd want to re-process duplicates forming
multiple groups from the refs, likely by just collecting
all candidates (including duplicates and split points
below) in a vector and then process them together. */
continue;
}
/* If init_b == init_a + the size of the type * k, we have an
interleaving, and DRA is accessed before DRB. */
HOST_WIDE_INT type_size_a = tree_to_uhwi (sza);
if (type_size_a == 0
|| (init_b - init_a) % type_size_a != 0)
break;
/* If we have a store, the accesses are adjacent. This splits
groups into chunks we support (we don't support vectorization
of stores with gaps). */
if (!DR_IS_READ (dra) && init_b - init_prev != type_size_a)
break;
/* If the step (if not zero or non-constant) is greater than the
difference between data-refs' inits this splits groups into
suitable sizes. */
if (tree_fits_shwi_p (DR_STEP (dra)))
{
HOST_WIDE_INT step = tree_to_shwi (DR_STEP (dra));
if (step != 0 && step <= (init_b - init_a))
break;
}
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"Detected interleaving ");
if (DR_IS_READ (dra))
dump_printf (MSG_NOTE, "load ");
else
dump_printf (MSG_NOTE, "store ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (dra));
dump_printf (MSG_NOTE, " and ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (drb));
dump_printf (MSG_NOTE, "\n");
}
/* Link the found element into the group list. */
if (!DR_GROUP_FIRST_ELEMENT (stmtinfo_a))
{
DR_GROUP_FIRST_ELEMENT (stmtinfo_a) = vect_dr_stmt (dra);
lastinfo = stmtinfo_a;
}
DR_GROUP_FIRST_ELEMENT (stmtinfo_b) = vect_dr_stmt (dra);
DR_GROUP_NEXT_ELEMENT (lastinfo) = vect_dr_stmt (drb);
lastinfo = stmtinfo_b;
}
}
FOR_EACH_VEC_ELT (datarefs_copy, i, dr)
if (STMT_VINFO_VECTORIZABLE (vinfo_for_stmt (vect_dr_stmt (dr)))
&& !vect_analyze_data_ref_access (dr))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: complicated access pattern.\n");
if (is_a <bb_vec_info> (vinfo))
{
/* Mark the statement as not vectorizable. */
STMT_VINFO_VECTORIZABLE (vinfo_for_stmt (vect_dr_stmt (dr))) = false;
continue;
}
else
{
datarefs_copy.release ();
return false;
}
}
datarefs_copy.release ();
return true;
}
/* Function vect_vfa_segment_size.
Input:
DR: The data reference.
LENGTH_FACTOR: segment length to consider.
Return a value suitable for the dr_with_seg_len::seg_len field.
This is the "distance travelled" by the pointer from the first
iteration in the segment to the last. Note that it does not include
the size of the access; in effect it only describes the first byte. */
static tree
vect_vfa_segment_size (struct data_reference *dr, tree length_factor)
{
length_factor = size_binop (MINUS_EXPR,
fold_convert (sizetype, length_factor),
size_one_node);
return size_binop (MULT_EXPR, fold_convert (sizetype, DR_STEP (dr)),
length_factor);
}
/* Return a value that, when added to abs (vect_vfa_segment_size (dr)),
gives the worst-case number of bytes covered by the segment. */
static unsigned HOST_WIDE_INT
vect_vfa_access_size (data_reference *dr)
{
stmt_vec_info stmt_vinfo = vinfo_for_stmt (vect_dr_stmt (dr));
tree ref_type = TREE_TYPE (DR_REF (dr));
unsigned HOST_WIDE_INT ref_size = tree_to_uhwi (TYPE_SIZE_UNIT (ref_type));
unsigned HOST_WIDE_INT access_size = ref_size;
if (DR_GROUP_FIRST_ELEMENT (stmt_vinfo))
{
gcc_assert (DR_GROUP_FIRST_ELEMENT (stmt_vinfo) == vect_dr_stmt (dr));
access_size *= DR_GROUP_SIZE (stmt_vinfo) - DR_GROUP_GAP (stmt_vinfo);
}
if (STMT_VINFO_VEC_STMT (stmt_vinfo)
&& (vect_supportable_dr_alignment (dr, false)
== dr_explicit_realign_optimized))
{
/* We might access a full vector's worth. */
tree vectype = STMT_VINFO_VECTYPE (stmt_vinfo);
access_size += tree_to_uhwi (TYPE_SIZE_UNIT (vectype)) - ref_size;
}
return access_size;
}
/* Get the minimum alignment for all the scalar accesses that DR describes. */
static unsigned int
vect_vfa_align (const data_reference *dr)
{
return TYPE_ALIGN_UNIT (TREE_TYPE (DR_REF (dr)));
}
/* Function vect_no_alias_p.
Given data references A and B with equal base and offset, see whether
the alias relation can be decided at compilation time. Return 1 if
it can and the references alias, 0 if it can and the references do
not alias, and -1 if we cannot decide at compile time. SEGMENT_LENGTH_A,
SEGMENT_LENGTH_B, ACCESS_SIZE_A and ACCESS_SIZE_B are the equivalent
of dr_with_seg_len::{seg_len,access_size} for A and B. */
static int
vect_compile_time_alias (struct data_reference *a, struct data_reference *b,
tree segment_length_a, tree segment_length_b,
unsigned HOST_WIDE_INT access_size_a,
unsigned HOST_WIDE_INT access_size_b)
{
poly_offset_int offset_a = wi::to_poly_offset (DR_INIT (a));
poly_offset_int offset_b = wi::to_poly_offset (DR_INIT (b));
poly_uint64 const_length_a;
poly_uint64 const_length_b;
/* For negative step, we need to adjust address range by TYPE_SIZE_UNIT
bytes, e.g., int a[3] -> a[1] range is [a+4, a+16) instead of
[a, a+12) */
if (tree_int_cst_compare (DR_STEP (a), size_zero_node) < 0)
{
const_length_a = (-wi::to_poly_wide (segment_length_a)).force_uhwi ();
offset_a = (offset_a + access_size_a) - const_length_a;
}
else
const_length_a = tree_to_poly_uint64 (segment_length_a);
if (tree_int_cst_compare (DR_STEP (b), size_zero_node) < 0)
{
const_length_b = (-wi::to_poly_wide (segment_length_b)).force_uhwi ();
offset_b = (offset_b + access_size_b) - const_length_b;
}
else
const_length_b = tree_to_poly_uint64 (segment_length_b);
const_length_a += access_size_a;
const_length_b += access_size_b;
if (ranges_known_overlap_p (offset_a, const_length_a,
offset_b, const_length_b))
return 1;
if (!ranges_maybe_overlap_p (offset_a, const_length_a,
offset_b, const_length_b))
return 0;
return -1;
}
/* Return true if the minimum nonzero dependence distance for loop LOOP_DEPTH
in DDR is >= VF. */
static bool
dependence_distance_ge_vf (data_dependence_relation *ddr,
unsigned int loop_depth, poly_uint64 vf)
{
if (DDR_ARE_DEPENDENT (ddr) != NULL_TREE
|| DDR_NUM_DIST_VECTS (ddr) == 0)
return false;
/* If the dependence is exact, we should have limited the VF instead. */
gcc_checking_assert (DDR_COULD_BE_INDEPENDENT_P (ddr));
unsigned int i;
lambda_vector dist_v;
FOR_EACH_VEC_ELT (DDR_DIST_VECTS (ddr), i, dist_v)
{
HOST_WIDE_INT dist = dist_v[loop_depth];
if (dist != 0
&& !(dist > 0 && DDR_REVERSED_P (ddr))
&& maybe_lt ((unsigned HOST_WIDE_INT) abs_hwi (dist), vf))
return false;
}
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"dependence distance between ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (DDR_A (ddr)));
dump_printf (MSG_NOTE, " and ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (DDR_B (ddr)));
dump_printf (MSG_NOTE, " is >= VF\n");
}
return true;
}
/* Dump LOWER_BOUND using flags DUMP_KIND. Dumps are known to be enabled. */
static void
dump_lower_bound (int dump_kind, const vec_lower_bound &lower_bound)
{
dump_printf (dump_kind, "%s (", lower_bound.unsigned_p ? "unsigned" : "abs");
dump_generic_expr (dump_kind, TDF_SLIM, lower_bound.expr);
dump_printf (dump_kind, ") >= ");
dump_dec (dump_kind, lower_bound.min_value);
}
/* Record that the vectorized loop requires the vec_lower_bound described
by EXPR, UNSIGNED_P and MIN_VALUE. */
static void
vect_check_lower_bound (loop_vec_info loop_vinfo, tree expr, bool unsigned_p,
poly_uint64 min_value)
{
vec<vec_lower_bound> lower_bounds = LOOP_VINFO_LOWER_BOUNDS (loop_vinfo);
for (unsigned int i = 0; i < lower_bounds.length (); ++i)
if (operand_equal_p (lower_bounds[i].expr, expr, 0))
{
unsigned_p &= lower_bounds[i].unsigned_p;
min_value = upper_bound (lower_bounds[i].min_value, min_value);
if (lower_bounds[i].unsigned_p != unsigned_p
|| maybe_lt (lower_bounds[i].min_value, min_value))
{
lower_bounds[i].unsigned_p = unsigned_p;
lower_bounds[i].min_value = min_value;
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"updating run-time check to ");
dump_lower_bound (MSG_NOTE, lower_bounds[i]);
dump_printf (MSG_NOTE, "\n");
}
}
return;
}
vec_lower_bound lower_bound (expr, unsigned_p, min_value);
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location, "need a run-time check that ");
dump_lower_bound (MSG_NOTE, lower_bound);
dump_printf (MSG_NOTE, "\n");
}
LOOP_VINFO_LOWER_BOUNDS (loop_vinfo).safe_push (lower_bound);
}
/* Return true if it's unlikely that the step of the vectorized form of DR
will span fewer than GAP bytes. */
static bool
vect_small_gap_p (loop_vec_info loop_vinfo, data_reference *dr, poly_int64 gap)
{
stmt_vec_info stmt_info = vinfo_for_stmt (vect_dr_stmt (dr));
HOST_WIDE_INT count
= estimated_poly_value (LOOP_VINFO_VECT_FACTOR (loop_vinfo));
if (DR_GROUP_FIRST_ELEMENT (stmt_info))
count *= DR_GROUP_SIZE (vinfo_for_stmt (DR_GROUP_FIRST_ELEMENT (stmt_info)));
return estimated_poly_value (gap) <= count * vect_get_scalar_dr_size (dr);
}
/* Return true if we know that there is no alias between DR_A and DR_B
when abs (DR_STEP (DR_A)) >= N for some N. When returning true, set
*LOWER_BOUND_OUT to this N. */
static bool
vectorizable_with_step_bound_p (data_reference *dr_a, data_reference *dr_b,
poly_uint64 *lower_bound_out)
{
/* Check that there is a constant gap of known sign between DR_A
and DR_B. */
poly_int64 init_a, init_b;
if (!operand_equal_p (DR_BASE_ADDRESS (dr_a), DR_BASE_ADDRESS (dr_b), 0)
|| !operand_equal_p (DR_OFFSET (dr_a), DR_OFFSET (dr_b), 0)
|| !operand_equal_p (DR_STEP (dr_a), DR_STEP (dr_b), 0)
|| !poly_int_tree_p (DR_INIT (dr_a), &init_a)
|| !poly_int_tree_p (DR_INIT (dr_b), &init_b)
|| !ordered_p (init_a, init_b))
return false;
/* Sort DR_A and DR_B by the address they access. */
if (maybe_lt (init_b, init_a))
{
std::swap (init_a, init_b);
std::swap (dr_a, dr_b);
}
/* If the two accesses could be dependent within a scalar iteration,
make sure that we'd retain their order. */
if (maybe_gt (init_a + vect_get_scalar_dr_size (dr_a), init_b)
&& !vect_preserves_scalar_order_p (vect_dr_stmt (dr_a),
vect_dr_stmt (dr_b)))
return false;
/* There is no alias if abs (DR_STEP) is greater than or equal to
the bytes spanned by the combination of the two accesses. */
*lower_bound_out = init_b + vect_get_scalar_dr_size (dr_b) - init_a;
return true;
}
/* Function vect_prune_runtime_alias_test_list.
Prune a list of ddrs to be tested at run-time by versioning for alias.
Merge several alias checks into one if possible.
Return FALSE if resulting list of ddrs is longer then allowed by
PARAM_VECT_MAX_VERSION_FOR_ALIAS_CHECKS, otherwise return TRUE. */
bool
vect_prune_runtime_alias_test_list (loop_vec_info loop_vinfo)
{
typedef pair_hash <tree_operand_hash, tree_operand_hash> tree_pair_hash;
hash_set <tree_pair_hash> compared_objects;
vec<ddr_p> may_alias_ddrs = LOOP_VINFO_MAY_ALIAS_DDRS (loop_vinfo);
vec<dr_with_seg_len_pair_t> &comp_alias_ddrs
= LOOP_VINFO_COMP_ALIAS_DDRS (loop_vinfo);
vec<vec_object_pair> &check_unequal_addrs
= LOOP_VINFO_CHECK_UNEQUAL_ADDRS (loop_vinfo);
poly_uint64 vect_factor = LOOP_VINFO_VECT_FACTOR (loop_vinfo);
tree scalar_loop_iters = LOOP_VINFO_NITERS (loop_vinfo);
ddr_p ddr;
unsigned int i;
tree length_factor;
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"=== vect_prune_runtime_alias_test_list ===\n");
/* Step values are irrelevant for aliasing if the number of vector
iterations is equal to the number of scalar iterations (which can
happen for fully-SLP loops). */
bool ignore_step_p = known_eq (LOOP_VINFO_VECT_FACTOR (loop_vinfo), 1U);
if (!ignore_step_p)
{
/* Convert the checks for nonzero steps into bound tests. */
tree value;
FOR_EACH_VEC_ELT (LOOP_VINFO_CHECK_NONZERO (loop_vinfo), i, value)
vect_check_lower_bound (loop_vinfo, value, true, 1);
}
if (may_alias_ddrs.is_empty ())
return true;
comp_alias_ddrs.create (may_alias_ddrs.length ());
unsigned int loop_depth
= index_in_loop_nest (LOOP_VINFO_LOOP (loop_vinfo)->num,
LOOP_VINFO_LOOP_NEST (loop_vinfo));
/* First, we collect all data ref pairs for aliasing checks. */
FOR_EACH_VEC_ELT (may_alias_ddrs, i, ddr)
{
int comp_res;
poly_uint64 lower_bound;
struct data_reference *dr_a, *dr_b;
gimple *dr_group_first_a, *dr_group_first_b;
tree segment_length_a, segment_length_b;
unsigned HOST_WIDE_INT access_size_a, access_size_b;
unsigned int align_a, align_b;
gimple *stmt_a, *stmt_b;
/* Ignore the alias if the VF we chose ended up being no greater
than the dependence distance. */
if (dependence_distance_ge_vf (ddr, loop_depth, vect_factor))
continue;
if (DDR_OBJECT_A (ddr))
{
vec_object_pair new_pair (DDR_OBJECT_A (ddr), DDR_OBJECT_B (ddr));
if (!compared_objects.add (new_pair))
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location, "checking that ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, new_pair.first);
dump_printf (MSG_NOTE, " and ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, new_pair.second);
dump_printf (MSG_NOTE, " have different addresses\n");
}
LOOP_VINFO_CHECK_UNEQUAL_ADDRS (loop_vinfo).safe_push (new_pair);
}
continue;
}
dr_a = DDR_A (ddr);
stmt_a = vect_dr_stmt (DDR_A (ddr));
dr_b = DDR_B (ddr);
stmt_b = vect_dr_stmt (DDR_B (ddr));
/* Skip the pair if inter-iteration dependencies are irrelevant
and intra-iteration dependencies are guaranteed to be honored. */
if (ignore_step_p
&& (vect_preserves_scalar_order_p (stmt_a, stmt_b)
|| vectorizable_with_step_bound_p (dr_a, dr_b, &lower_bound)))
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"no need for alias check between ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (dr_a));
dump_printf (MSG_NOTE, " and ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (dr_b));
dump_printf (MSG_NOTE, " when VF is 1\n");
}
continue;
}
/* See whether we can handle the alias using a bounds check on
the step, and whether that's likely to be the best approach.
(It might not be, for example, if the minimum step is much larger
than the number of bytes handled by one vector iteration.) */
if (!ignore_step_p
&& TREE_CODE (DR_STEP (dr_a)) != INTEGER_CST
&& vectorizable_with_step_bound_p (dr_a, dr_b, &lower_bound)
&& (vect_small_gap_p (loop_vinfo, dr_a, lower_bound)
|| vect_small_gap_p (loop_vinfo, dr_b, lower_bound)))
{
bool unsigned_p = dr_known_forward_stride_p (dr_a);
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location, "no alias between ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (dr_a));
dump_printf (MSG_NOTE, " and ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (dr_b));
dump_printf (MSG_NOTE, " when the step ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_STEP (dr_a));
dump_printf (MSG_NOTE, " is outside ");
if (unsigned_p)
dump_printf (MSG_NOTE, "[0");
else
{
dump_printf (MSG_NOTE, "(");
dump_dec (MSG_NOTE, poly_int64 (-lower_bound));
}
dump_printf (MSG_NOTE, ", ");
dump_dec (MSG_NOTE, lower_bound);
dump_printf (MSG_NOTE, ")\n");
}
vect_check_lower_bound (loop_vinfo, DR_STEP (dr_a), unsigned_p,
lower_bound);
continue;
}
dr_group_first_a = DR_GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt_a));
if (dr_group_first_a)
{
stmt_a = dr_group_first_a;
dr_a = STMT_VINFO_DATA_REF (vinfo_for_stmt (stmt_a));
}
dr_group_first_b = DR_GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt_b));
if (dr_group_first_b)
{
stmt_b = dr_group_first_b;
dr_b = STMT_VINFO_DATA_REF (vinfo_for_stmt (stmt_b));
}
if (ignore_step_p)
{
segment_length_a = size_zero_node;
segment_length_b = size_zero_node;
}
else
{
if (!operand_equal_p (DR_STEP (dr_a), DR_STEP (dr_b), 0))
length_factor = scalar_loop_iters;
else
length_factor = size_int (vect_factor);
segment_length_a = vect_vfa_segment_size (dr_a, length_factor);
segment_length_b = vect_vfa_segment_size (dr_b, length_factor);
}
access_size_a = vect_vfa_access_size (dr_a);
access_size_b = vect_vfa_access_size (dr_b);
align_a = vect_vfa_align (dr_a);
align_b = vect_vfa_align (dr_b);
comp_res = data_ref_compare_tree (DR_BASE_ADDRESS (dr_a),
DR_BASE_ADDRESS (dr_b));
if (comp_res == 0)
comp_res = data_ref_compare_tree (DR_OFFSET (dr_a),
DR_OFFSET (dr_b));
/* See whether the alias is known at compilation time. */
if (comp_res == 0
&& TREE_CODE (DR_STEP (dr_a)) == INTEGER_CST
&& TREE_CODE (DR_STEP (dr_b)) == INTEGER_CST
&& poly_int_tree_p (segment_length_a)
&& poly_int_tree_p (segment_length_b))
{
int res = vect_compile_time_alias (dr_a, dr_b,
segment_length_a,
segment_length_b,
access_size_a,
access_size_b);
if (res >= 0 && dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"can tell at compile time that ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (dr_a));
dump_printf (MSG_NOTE, " and ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_REF (dr_b));
if (res == 0)
dump_printf (MSG_NOTE, " do not alias\n");
else
dump_printf (MSG_NOTE, " alias\n");
}
if (res == 0)
continue;
if (res == 1)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"not vectorized: compilation time alias.\n");
return false;
}
}
dr_with_seg_len_pair_t dr_with_seg_len_pair
(dr_with_seg_len (dr_a, segment_length_a, access_size_a, align_a),
dr_with_seg_len (dr_b, segment_length_b, access_size_b, align_b));
/* Canonicalize pairs by sorting the two DR members. */
if (comp_res > 0)
std::swap (dr_with_seg_len_pair.first, dr_with_seg_len_pair.second);
comp_alias_ddrs.safe_push (dr_with_seg_len_pair);
}
prune_runtime_alias_test_list (&comp_alias_ddrs, vect_factor);
unsigned int count = (comp_alias_ddrs.length ()
+ check_unequal_addrs.length ());
dump_printf_loc (MSG_NOTE, vect_location,
"improved number of alias checks from %d to %d\n",
may_alias_ddrs.length (), count);
if ((int) count > PARAM_VALUE (PARAM_VECT_MAX_VERSION_FOR_ALIAS_CHECKS))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"number of versioning for alias "
"run-time tests exceeds %d "
"(--param vect-max-version-for-alias-checks)\n",
PARAM_VALUE (PARAM_VECT_MAX_VERSION_FOR_ALIAS_CHECKS));
return false;
}
return true;
}
/* Check whether we can use an internal function for a gather load
or scatter store. READ_P is true for loads and false for stores.
MASKED_P is true if the load or store is conditional. MEMORY_TYPE is
the type of the memory elements being loaded or stored. OFFSET_BITS
is the number of bits in each scalar offset and OFFSET_SIGN is the
sign of the offset. SCALE is the amount by which the offset should
be multiplied *after* it has been converted to address width.
Return true if the function is supported, storing the function
id in *IFN_OUT and the type of a vector element in *ELEMENT_TYPE_OUT. */
bool
vect_gather_scatter_fn_p (bool read_p, bool masked_p, tree vectype,
tree memory_type, unsigned int offset_bits,
signop offset_sign, int scale,
internal_fn *ifn_out, tree *element_type_out)
{
unsigned int memory_bits = tree_to_uhwi (TYPE_SIZE (memory_type));
unsigned int element_bits = tree_to_uhwi (TYPE_SIZE (TREE_TYPE (vectype)));
if (offset_bits > element_bits)
/* Internal functions require the offset to be the same width as
the vector elements. We can extend narrower offsets, but it isn't
safe to truncate wider offsets. */
return false;
if (element_bits != memory_bits)
/* For now the vector elements must be the same width as the
memory elements. */
return false;
/* Work out which function we need. */
internal_fn ifn;
if (read_p)
ifn = masked_p ? IFN_MASK_GATHER_LOAD : IFN_GATHER_LOAD;
else
ifn = masked_p ? IFN_MASK_SCATTER_STORE : IFN_SCATTER_STORE;
/* Test whether the target supports this combination. */
if (!internal_gather_scatter_fn_supported_p (ifn, vectype, memory_type,
offset_sign, scale))
return false;
*ifn_out = ifn;
*element_type_out = TREE_TYPE (vectype);
return true;
}
/* CALL is a call to an internal gather load or scatter store function.
Describe the operation in INFO. */
static void
vect_describe_gather_scatter_call (gcall *call, gather_scatter_info *info)
{
stmt_vec_info stmt_info = vinfo_for_stmt (call);
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
info->ifn = gimple_call_internal_fn (call);
info->decl = NULL_TREE;
info->base = gimple_call_arg (call, 0);
info->offset = gimple_call_arg (call, 1);
info->offset_dt = vect_unknown_def_type;
info->offset_vectype = NULL_TREE;
info->scale = TREE_INT_CST_LOW (gimple_call_arg (call, 2));
info->element_type = TREE_TYPE (vectype);
info->memory_type = TREE_TYPE (DR_REF (dr));
}
/* Return true if a non-affine read or write in STMT is suitable for a
gather load or scatter store. Describe the operation in *INFO if so. */
bool
vect_check_gather_scatter (gimple *stmt, loop_vec_info loop_vinfo,
gather_scatter_info *info)
{
HOST_WIDE_INT scale = 1;
poly_int64 pbitpos, pbitsize;
struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
tree offtype = NULL_TREE;
tree decl = NULL_TREE, base, off;
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
tree memory_type = TREE_TYPE (DR_REF (dr));
machine_mode pmode;
int punsignedp, reversep, pvolatilep = 0;
internal_fn ifn;
tree element_type;
bool masked_p = false;
/* See whether this is already a call to a gather/scatter internal function.
If not, see whether it's a masked load or store. */
gcall *call = dyn_cast <gcall *> (stmt);
if (call && gimple_call_internal_p (call))
{
ifn = gimple_call_internal_fn (stmt);
if (internal_gather_scatter_fn_p (ifn))
{
vect_describe_gather_scatter_call (call, info);
return true;
}
masked_p = (ifn == IFN_MASK_LOAD || ifn == IFN_MASK_STORE);
}
/* True if we should aim to use internal functions rather than
built-in functions. */
bool use_ifn_p = (DR_IS_READ (dr)
? supports_vec_gather_load_p ()
: supports_vec_scatter_store_p ());
base = DR_REF (dr);
/* For masked loads/stores, DR_REF (dr) is an artificial MEM_REF,
see if we can use the def stmt of the address. */
if (masked_p
&& TREE_CODE (base) == MEM_REF
&& TREE_CODE (TREE_OPERAND (base, 0)) == SSA_NAME
&& integer_zerop (TREE_OPERAND (base, 1))
&& !expr_invariant_in_loop_p (loop, TREE_OPERAND (base, 0)))
{
gimple *def_stmt = SSA_NAME_DEF_STMT (TREE_OPERAND (base, 0));
if (is_gimple_assign (def_stmt)
&& gimple_assign_rhs_code (def_stmt) == ADDR_EXPR)
base = TREE_OPERAND (gimple_assign_rhs1 (def_stmt), 0);
}
/* The gather and scatter builtins need address of the form
loop_invariant + vector * {1, 2, 4, 8}
or
loop_invariant + sign_extend (vector) * { 1, 2, 4, 8 }.
Unfortunately DR_BASE_ADDRESS/DR_OFFSET can be a mixture
of loop invariants/SSA_NAMEs defined in the loop, with casts,
multiplications and additions in it. To get a vector, we need
a single SSA_NAME that will be defined in the loop and will
contain everything that is not loop invariant and that can be
vectorized. The following code attempts to find such a preexistng
SSA_NAME OFF and put the loop invariants into a tree BASE
that can be gimplified before the loop. */
base = get_inner_reference (base, &pbitsize, &pbitpos, &off, &pmode,
&punsignedp, &reversep, &pvolatilep);
gcc_assert (base && !reversep);
poly_int64 pbytepos = exact_div (pbitpos, BITS_PER_UNIT);
if (TREE_CODE (base) == MEM_REF)
{
if (!integer_zerop (TREE_OPERAND (base, 1)))
{
if (off == NULL_TREE)
off = wide_int_to_tree (sizetype, mem_ref_offset (base));
else
off = size_binop (PLUS_EXPR, off,
fold_convert (sizetype, TREE_OPERAND (base, 1)));
}
base = TREE_OPERAND (base, 0);
}
else
base = build_fold_addr_expr (base);
if (off == NULL_TREE)
off = size_zero_node;
/* If base is not loop invariant, either off is 0, then we start with just
the constant offset in the loop invariant BASE and continue with base
as OFF, otherwise give up.
We could handle that case by gimplifying the addition of base + off
into some SSA_NAME and use that as off, but for now punt. */
if (!expr_invariant_in_loop_p (loop, base))
{
if (!integer_zerop (off))
return false;
off = base;
base = size_int (pbytepos);
}
/* Otherwise put base + constant offset into the loop invariant BASE
and continue with OFF. */
else
{
base = fold_convert (sizetype, base);
base = size_binop (PLUS_EXPR, base, size_int (pbytepos));
}
/* OFF at this point may be either a SSA_NAME or some tree expression
from get_inner_reference. Try to peel off loop invariants from it
into BASE as long as possible. */
STRIP_NOPS (off);
while (offtype == NULL_TREE)
{
enum tree_code code;
tree op0, op1, add = NULL_TREE;
if (TREE_CODE (off) == SSA_NAME)
{
gimple *def_stmt = SSA_NAME_DEF_STMT (off);
if (expr_invariant_in_loop_p (loop, off))
return false;
if (gimple_code (def_stmt) != GIMPLE_ASSIGN)
break;
op0 = gimple_assign_rhs1 (def_stmt);
code = gimple_assign_rhs_code (def_stmt);
op1 = gimple_assign_rhs2 (def_stmt);
}
else
{
if (get_gimple_rhs_class (TREE_CODE (off)) == GIMPLE_TERNARY_RHS)
return false;
code = TREE_CODE (off);
extract_ops_from_tree (off, &code, &op0, &op1);
}
switch (code)
{
case POINTER_PLUS_EXPR:
case PLUS_EXPR:
if (expr_invariant_in_loop_p (loop, op0))
{
add = op0;
off = op1;
do_add:
add = fold_convert (sizetype, add);
if (scale != 1)
add = size_binop (MULT_EXPR, add, size_int (scale));
base = size_binop (PLUS_EXPR, base, add);
continue;
}
if (expr_invariant_in_loop_p (loop, op1))
{
add = op1;
off = op0;
goto do_add;
}
break;
case MINUS_EXPR:
if (expr_invariant_in_loop_p (loop, op1))
{
add = fold_convert (sizetype, op1);
add = size_binop (MINUS_EXPR, size_zero_node, add);
off = op0;
goto do_add;
}
break;
case MULT_EXPR:
if (scale == 1 && tree_fits_shwi_p (op1))
{
int new_scale = tree_to_shwi (op1);
/* Only treat this as a scaling operation if the target
supports it. */
if (use_ifn_p
&& !vect_gather_scatter_fn_p (DR_IS_READ (dr), masked_p,
vectype, memory_type, 1,
TYPE_SIGN (TREE_TYPE (op0)),
new_scale, &ifn,
&element_type))
break;
scale = new_scale;
off = op0;
continue;
}
break;
case SSA_NAME:
off = op0;
continue;
CASE_CONVERT:
if (!POINTER_TYPE_P (TREE_TYPE (op0))
&& !INTEGRAL_TYPE_P (TREE_TYPE (op0)))
break;
if (TYPE_PRECISION (TREE_TYPE (op0))
== TYPE_PRECISION (TREE_TYPE (off)))
{
off = op0;
continue;
}
/* The internal functions need the offset to be the same width
as the elements of VECTYPE. Don't include operations that
cast the offset from that width to a different width. */
if (use_ifn_p
&& (int_size_in_bytes (TREE_TYPE (vectype))
== int_size_in_bytes (TREE_TYPE (off))))
break;
if (TYPE_PRECISION (TREE_TYPE (op0))
< TYPE_PRECISION (TREE_TYPE (off)))
{
off = op0;
offtype = TREE_TYPE (off);
STRIP_NOPS (off);
continue;
}
break;
default:
break;
}
break;
}
/* If at the end OFF still isn't a SSA_NAME or isn't
defined in the loop, punt. */
if (TREE_CODE (off) != SSA_NAME
|| expr_invariant_in_loop_p (loop, off))
return false;
if (offtype == NULL_TREE)
offtype = TREE_TYPE (off);
if (use_ifn_p)
{
if (!vect_gather_scatter_fn_p (DR_IS_READ (dr), masked_p, vectype,
memory_type, TYPE_PRECISION (offtype),
TYPE_SIGN (offtype), scale, &ifn,
&element_type))
return false;
}
else
{
if (DR_IS_READ (dr))
{
if (targetm.vectorize.builtin_gather)
decl = targetm.vectorize.builtin_gather (vectype, offtype, scale);
}
else
{
if (targetm.vectorize.builtin_scatter)
decl = targetm.vectorize.builtin_scatter (vectype, offtype, scale);
}
if (!decl)
return false;
ifn = IFN_LAST;
element_type = TREE_TYPE (vectype);
}
info->ifn = ifn;
info->decl = decl;
info->base = base;
info->offset = off;
info->offset_dt = vect_unknown_def_type;
info->offset_vectype = NULL_TREE;
info->scale = scale;
info->element_type = element_type;
info->memory_type = memory_type;
return true;
}
/* Find the data references in STMT, analyze them with respect to LOOP and
append them to DATAREFS. Return false if datarefs in this stmt cannot
be handled. */
bool
vect_find_stmt_data_reference (loop_p loop, gimple *stmt,
vec<data_reference_p> *datarefs)
{
/* We can ignore clobbers for dataref analysis - they are removed during
loop vectorization and BB vectorization checks dependences with a
stmt walk. */
if (gimple_clobber_p (stmt))
return true;
if (gimple_has_volatile_ops (stmt))
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: volatile type ");
dump_gimple_stmt (MSG_MISSED_OPTIMIZATION, TDF_SLIM, stmt, 0);
}
return false;
}
if (stmt_can_throw_internal (stmt))
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: statement can throw an "
"exception ");
dump_gimple_stmt (MSG_MISSED_OPTIMIZATION, TDF_SLIM, stmt, 0);
}
return false;
}
auto_vec<data_reference_p, 2> refs;
if (!find_data_references_in_stmt (loop, stmt, &refs))
return false;
if (refs.is_empty ())
return true;
if (refs.length () > 1)
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: more than one data ref "
"in stmt: ");
dump_gimple_stmt (MSG_MISSED_OPTIMIZATION, TDF_SLIM, stmt, 0);
}
return false;
}
if (gcall *call = dyn_cast <gcall *> (stmt))
if (!gimple_call_internal_p (call)
|| (gimple_call_internal_fn (call) != IFN_MASK_LOAD
&& gimple_call_internal_fn (call) != IFN_MASK_STORE))
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: dr in a call ");
dump_gimple_stmt (MSG_MISSED_OPTIMIZATION, TDF_SLIM, stmt, 0);
}
return false;
}
data_reference_p dr = refs.pop ();
if (TREE_CODE (DR_REF (dr)) == COMPONENT_REF
&& DECL_BIT_FIELD (TREE_OPERAND (DR_REF (dr), 1)))
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: statement is bitfield "
"access ");
dump_gimple_stmt (MSG_MISSED_OPTIMIZATION, TDF_SLIM, stmt, 0);
}
return false;
}
if (DR_BASE_ADDRESS (dr)
&& TREE_CODE (DR_BASE_ADDRESS (dr)) == INTEGER_CST)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: base addr of dr is a "
"constant\n");
return false;
}
datarefs->safe_push (dr);
return true;
}
/* Function vect_analyze_data_refs.
Find all the data references in the loop or basic block.
The general structure of the analysis of data refs in the vectorizer is as
follows:
1- vect_analyze_data_refs(loop/bb): call
compute_data_dependences_for_loop/bb to find and analyze all data-refs
in the loop/bb and their dependences.
2- vect_analyze_dependences(): apply dependence testing using ddrs.
3- vect_analyze_drs_alignment(): check that ref_stmt.alignment is ok.
4- vect_analyze_drs_access(): check that ref_stmt.step is ok.
*/
bool
vect_analyze_data_refs (vec_info *vinfo, poly_uint64 *min_vf)
{
struct loop *loop = NULL;
unsigned int i;
struct data_reference *dr;
tree scalar_type;
if (dump_enabled_p ())
dump_printf_loc (MSG_NOTE, vect_location,
"=== vect_analyze_data_refs ===\n");
if (loop_vec_info loop_vinfo = dyn_cast <loop_vec_info> (vinfo))
loop = LOOP_VINFO_LOOP (loop_vinfo);
/* Go through the data-refs, check that the analysis succeeded. Update
pointer from stmt_vec_info struct to DR and vectype. */
vec<data_reference_p> datarefs = vinfo->datarefs;
FOR_EACH_VEC_ELT (datarefs, i, dr)
{
gimple *stmt;
stmt_vec_info stmt_info;
enum { SG_NONE, GATHER, SCATTER } gatherscatter = SG_NONE;
bool simd_lane_access = false;
poly_uint64 vf;
gcc_assert (DR_REF (dr));
stmt = vect_dr_stmt (dr);
stmt_info = vinfo_for_stmt (stmt);
/* Check that analysis of the data-ref succeeded. */
if (!DR_BASE_ADDRESS (dr) || !DR_OFFSET (dr) || !DR_INIT (dr)
|| !DR_STEP (dr))
{
bool maybe_gather
= DR_IS_READ (dr)
&& !TREE_THIS_VOLATILE (DR_REF (dr))
&& (targetm.vectorize.builtin_gather != NULL
|| supports_vec_gather_load_p ());
bool maybe_scatter
= DR_IS_WRITE (dr)
&& !TREE_THIS_VOLATILE (DR_REF (dr))
&& (targetm.vectorize.builtin_scatter != NULL
|| supports_vec_scatter_store_p ());
bool maybe_simd_lane_access
= is_a <loop_vec_info> (vinfo) && loop->simduid;
/* If target supports vector gather loads or scatter stores, or if
this might be a SIMD lane access, see if they can't be used. */
if (is_a <loop_vec_info> (vinfo)
&& (maybe_gather || maybe_scatter || maybe_simd_lane_access)
&& !nested_in_vect_loop_p (loop, stmt))
{
struct data_reference *newdr
= create_data_ref (NULL, loop_containing_stmt (stmt),
DR_REF (dr), stmt, !maybe_scatter,
DR_IS_CONDITIONAL_IN_STMT (dr));
gcc_assert (newdr != NULL && DR_REF (newdr));
if (DR_BASE_ADDRESS (newdr)
&& DR_OFFSET (newdr)
&& DR_INIT (newdr)
&& DR_STEP (newdr)
&& integer_zerop (DR_STEP (newdr)))
{
if (maybe_simd_lane_access)
{
tree off = DR_OFFSET (newdr);
STRIP_NOPS (off);
if (TREE_CODE (DR_INIT (newdr)) == INTEGER_CST
&& TREE_CODE (off) == MULT_EXPR
&& tree_fits_uhwi_p (TREE_OPERAND (off, 1)))
{
tree step = TREE_OPERAND (off, 1);
off = TREE_OPERAND (off, 0);
STRIP_NOPS (off);
if (CONVERT_EXPR_P (off)
&& TYPE_PRECISION (TREE_TYPE (TREE_OPERAND (off,
0)))
< TYPE_PRECISION (TREE_TYPE (off)))
off = TREE_OPERAND (off, 0);
if (TREE_CODE (off) == SSA_NAME)
{
gimple *def = SSA_NAME_DEF_STMT (off);
tree reft = TREE_TYPE (DR_REF (newdr));
if (is_gimple_call (def)
&& gimple_call_internal_p (def)
&& (gimple_call_internal_fn (def)
== IFN_GOMP_SIMD_LANE))
{
tree arg = gimple_call_arg (def, 0);
gcc_assert (TREE_CODE (arg) == SSA_NAME);
arg = SSA_NAME_VAR (arg);
if (arg == loop->simduid
/* For now. */
&& tree_int_cst_equal
(TYPE_SIZE_UNIT (reft),
step))
{
DR_OFFSET (newdr) = ssize_int (0);
DR_STEP (newdr) = step;
DR_OFFSET_ALIGNMENT (newdr)
= BIGGEST_ALIGNMENT;
DR_STEP_ALIGNMENT (newdr)
= highest_pow2_factor (step);
dr = newdr;
simd_lane_access = true;
}
}
}
}
}
if (!simd_lane_access && (maybe_gather || maybe_scatter))
{
dr = newdr;
if (maybe_gather)
gatherscatter = GATHER;
else
gatherscatter = SCATTER;
}
}
if (gatherscatter == SG_NONE && !simd_lane_access)
free_data_ref (newdr);
}
if (gatherscatter == SG_NONE && !simd_lane_access)
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: data ref analysis "
"failed ");
dump_gimple_stmt (MSG_MISSED_OPTIMIZATION, TDF_SLIM, stmt, 0);
}
if (is_a <bb_vec_info> (vinfo))
{
/* In BB vectorization the ref can still participate
in dependence analysis, we just can't vectorize it. */
STMT_VINFO_VECTORIZABLE (stmt_info) = false;
continue;
}
return false;
}
}
if (TREE_CODE (DR_BASE_ADDRESS (dr)) == ADDR_EXPR
&& VAR_P (TREE_OPERAND (DR_BASE_ADDRESS (dr), 0))
&& DECL_NONALIASED (TREE_OPERAND (DR_BASE_ADDRESS (dr), 0)))
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: base object not addressable "
"for stmt: ");
dump_gimple_stmt (MSG_MISSED_OPTIMIZATION, TDF_SLIM, stmt, 0);
}
if (is_a <bb_vec_info> (vinfo))
{
/* In BB vectorization the ref can still participate
in dependence analysis, we just can't vectorize it. */
STMT_VINFO_VECTORIZABLE (stmt_info) = false;
continue;
}
return false;
}
if (is_a <loop_vec_info> (vinfo)
&& TREE_CODE (DR_STEP (dr)) != INTEGER_CST)
{
if (nested_in_vect_loop_p (loop, stmt))
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: not suitable for strided "
"load ");
dump_gimple_stmt (MSG_MISSED_OPTIMIZATION, TDF_SLIM, stmt, 0);
}
return false;
}
STMT_VINFO_STRIDED_P (stmt_info) = true;
}
/* Update DR field in stmt_vec_info struct. */
/* If the dataref is in an inner-loop of the loop that is considered for
for vectorization, we also want to analyze the access relative to
the outer-loop (DR contains information only relative to the
inner-most enclosing loop). We do that by building a reference to the
first location accessed by the inner-loop, and analyze it relative to
the outer-loop. */
if (loop && nested_in_vect_loop_p (loop, stmt))
{
/* Build a reference to the first location accessed by the
inner loop: *(BASE + INIT + OFFSET). By construction,
this address must be invariant in the inner loop, so we
can consider it as being used in the outer loop. */
tree base = unshare_expr (DR_BASE_ADDRESS (dr));
tree offset = unshare_expr (DR_OFFSET (dr));
tree init = unshare_expr (DR_INIT (dr));
tree init_offset = fold_build2 (PLUS_EXPR, TREE_TYPE (offset),
init, offset);
tree init_addr = fold_build_pointer_plus (base, init_offset);
tree init_ref = build_fold_indirect_ref (init_addr);
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"analyze in outer loop: ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, init_ref);
dump_printf (MSG_NOTE, "\n");
}
if (!dr_analyze_innermost (&STMT_VINFO_DR_WRT_VEC_LOOP (stmt_info),
init_ref, loop))
/* dr_analyze_innermost already explained the failure. */
return false;
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"\touter base_address: ");
dump_generic_expr (MSG_NOTE, TDF_SLIM,
STMT_VINFO_DR_BASE_ADDRESS (stmt_info));
dump_printf (MSG_NOTE, "\n\touter offset from base address: ");
dump_generic_expr (MSG_NOTE, TDF_SLIM,
STMT_VINFO_DR_OFFSET (stmt_info));
dump_printf (MSG_NOTE,
"\n\touter constant offset from base address: ");
dump_generic_expr (MSG_NOTE, TDF_SLIM,
STMT_VINFO_DR_INIT (stmt_info));
dump_printf (MSG_NOTE, "\n\touter step: ");
dump_generic_expr (MSG_NOTE, TDF_SLIM,
STMT_VINFO_DR_STEP (stmt_info));
dump_printf (MSG_NOTE, "\n\touter base alignment: %d\n",
STMT_VINFO_DR_BASE_ALIGNMENT (stmt_info));
dump_printf (MSG_NOTE, "\n\touter base misalignment: %d\n",
STMT_VINFO_DR_BASE_MISALIGNMENT (stmt_info));
dump_printf (MSG_NOTE, "\n\touter offset alignment: %d\n",
STMT_VINFO_DR_OFFSET_ALIGNMENT (stmt_info));
dump_printf (MSG_NOTE, "\n\touter step alignment: %d\n",
STMT_VINFO_DR_STEP_ALIGNMENT (stmt_info));
}
}
gcc_assert (!STMT_VINFO_DATA_REF (stmt_info));
STMT_VINFO_DATA_REF (stmt_info) = dr;
if (simd_lane_access)
{
STMT_VINFO_SIMD_LANE_ACCESS_P (stmt_info) = true;
free_data_ref (datarefs[i]);
datarefs[i] = dr;
}
/* Set vectype for STMT. */
scalar_type = TREE_TYPE (DR_REF (dr));
STMT_VINFO_VECTYPE (stmt_info)
= get_vectype_for_scalar_type (scalar_type);
if (!STMT_VINFO_VECTYPE (stmt_info))
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"not vectorized: no vectype for stmt: ");
dump_gimple_stmt (MSG_MISSED_OPTIMIZATION, TDF_SLIM, stmt, 0);
dump_printf (MSG_MISSED_OPTIMIZATION, " scalar_type: ");
dump_generic_expr (MSG_MISSED_OPTIMIZATION, TDF_DETAILS,
scalar_type);
dump_printf (MSG_MISSED_OPTIMIZATION, "\n");
}
if (is_a <bb_vec_info> (vinfo))
{
/* No vector type is fine, the ref can still participate
in dependence analysis, we just can't vectorize it. */
STMT_VINFO_VECTORIZABLE (stmt_info) = false;
continue;
}
if (gatherscatter != SG_NONE || simd_lane_access)
{
STMT_VINFO_DATA_REF (stmt_info) = NULL;
if (gatherscatter != SG_NONE)
free_data_ref (dr);
}
return false;
}
else
{
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location,
"got vectype for stmt: ");
dump_gimple_stmt (MSG_NOTE, TDF_SLIM, stmt, 0);
dump_generic_expr (MSG_NOTE, TDF_SLIM,
STMT_VINFO_VECTYPE (stmt_info));
dump_printf (MSG_NOTE, "\n");
}
}
/* Adjust the minimal vectorization factor according to the
vector type. */
vf = TYPE_VECTOR_SUBPARTS (STMT_VINFO_VECTYPE (stmt_info));
*min_vf = upper_bound (*min_vf, vf);
if (gatherscatter != SG_NONE)
{
gather_scatter_info gs_info;
if (!vect_check_gather_scatter (stmt, as_a <loop_vec_info> (vinfo),
&gs_info)
|| !get_vectype_for_scalar_type (TREE_TYPE (gs_info.offset)))
{
STMT_VINFO_DATA_REF (stmt_info) = NULL;
free_data_ref (dr);
if (dump_enabled_p ())
{
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
(gatherscatter == GATHER) ?
"not vectorized: not suitable for gather "
"load " :
"not vectorized: not suitable for scatter "
"store ");
dump_gimple_stmt (MSG_MISSED_OPTIMIZATION, TDF_SLIM, stmt, 0);
}
return false;
}
free_data_ref (datarefs[i]);
datarefs[i] = dr;
STMT_VINFO_GATHER_SCATTER_P (stmt_info) = gatherscatter;
}
}
/* We used to stop processing and prune the list here. Verify we no
longer need to. */
gcc_assert (i == datarefs.length ());
return true;
}
/* Function vect_get_new_vect_var.
Returns a name for a new variable. The current naming scheme appends the
prefix "vect_" or "vect_p" (depending on the value of VAR_KIND) to
the name of vectorizer generated variables, and appends that to NAME if
provided. */
tree
vect_get_new_vect_var (tree type, enum vect_var_kind var_kind, const char *name)
{
const char *prefix;
tree new_vect_var;
switch (var_kind)
{
case vect_simple_var:
prefix = "vect";
break;
case vect_scalar_var:
prefix = "stmp";
break;
case vect_mask_var:
prefix = "mask";
break;
case vect_pointer_var:
prefix = "vectp";
break;
default:
gcc_unreachable ();
}
if (name)
{
char* tmp = concat (prefix, "_", name, NULL);
new_vect_var = create_tmp_reg (type, tmp);
free (tmp);
}
else
new_vect_var = create_tmp_reg (type, prefix);
return new_vect_var;
}
/* Like vect_get_new_vect_var but return an SSA name. */
tree
vect_get_new_ssa_name (tree type, enum vect_var_kind var_kind, const char *name)
{
const char *prefix;
tree new_vect_var;
switch (var_kind)
{
case vect_simple_var:
prefix = "vect";
break;
case vect_scalar_var:
prefix = "stmp";
break;
case vect_pointer_var:
prefix = "vectp";
break;
default:
gcc_unreachable ();
}
if (name)
{
char* tmp = concat (prefix, "_", name, NULL);
new_vect_var = make_temp_ssa_name (type, NULL, tmp);
free (tmp);
}
else
new_vect_var = make_temp_ssa_name (type, NULL, prefix);
return new_vect_var;
}
/* Duplicate ptr info and set alignment/misaligment on NAME from DR. */
static void
vect_duplicate_ssa_name_ptr_info (tree name, data_reference *dr)
{
duplicate_ssa_name_ptr_info (name, DR_PTR_INFO (dr));
int misalign = DR_MISALIGNMENT (dr);
if (misalign == DR_MISALIGNMENT_UNKNOWN)
mark_ptr_info_alignment_unknown (SSA_NAME_PTR_INFO (name));
else
set_ptr_info_alignment (SSA_NAME_PTR_INFO (name),
DR_TARGET_ALIGNMENT (dr), misalign);
}
/* Function vect_create_addr_base_for_vector_ref.
Create an expression that computes the address of the first memory location
that will be accessed for a data reference.
Input:
STMT: The statement containing the data reference.
NEW_STMT_LIST: Must be initialized to NULL_TREE or a statement list.
OFFSET: Optional. If supplied, it is be added to the initial address.
LOOP: Specify relative to which loop-nest should the address be computed.
For example, when the dataref is in an inner-loop nested in an
outer-loop that is now being vectorized, LOOP can be either the
outer-loop, or the inner-loop. The first memory location accessed
by the following dataref ('in' points to short):
for (i=0; i<N; i++)
for (j=0; j<M; j++)
s += in[i+j]
is as follows:
if LOOP=i_loop: &in (relative to i_loop)
if LOOP=j_loop: &in+i*2B (relative to j_loop)
BYTE_OFFSET: Optional, defaulted to NULL. If supplied, it is added to the
initial address. Unlike OFFSET, which is number of elements to
be added, BYTE_OFFSET is measured in bytes.
Output:
1. Return an SSA_NAME whose value is the address of the memory location of
the first vector of the data reference.
2. If new_stmt_list is not NULL_TREE after return then the caller must insert
these statement(s) which define the returned SSA_NAME.
FORNOW: We are only handling array accesses with step 1. */
tree
vect_create_addr_base_for_vector_ref (gimple *stmt,
gimple_seq *new_stmt_list,
tree offset,
tree byte_offset)
{
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
const char *base_name;
tree addr_base;
tree dest;
gimple_seq seq = NULL;
tree vect_ptr_type;
tree step = TYPE_SIZE_UNIT (TREE_TYPE (DR_REF (dr)));
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
innermost_loop_behavior *drb = vect_dr_behavior (dr);
tree data_ref_base = unshare_expr (drb->base_address);
tree base_offset = unshare_expr (drb->offset);
tree init = unshare_expr (drb->init);
if (loop_vinfo)
base_name = get_name (data_ref_base);
else
{
base_offset = ssize_int (0);
init = ssize_int (0);
base_name = get_name (DR_REF (dr));
}
/* Create base_offset */
base_offset = size_binop (PLUS_EXPR,
fold_convert (sizetype, base_offset),
fold_convert (sizetype, init));
if (offset)
{
offset = fold_build2 (MULT_EXPR, sizetype,
fold_convert (sizetype, offset), step);
base_offset = fold_build2 (PLUS_EXPR, sizetype,
base_offset, offset);
}
if (byte_offset)
{
byte_offset = fold_convert (sizetype, byte_offset);
base_offset = fold_build2 (PLUS_EXPR, sizetype,
base_offset, byte_offset);
}
/* base + base_offset */
if (loop_vinfo)
addr_base = fold_build_pointer_plus (data_ref_base, base_offset);
else
{
addr_base = build1 (ADDR_EXPR,
build_pointer_type (TREE_TYPE (DR_REF (dr))),
unshare_expr (DR_REF (dr)));
}
vect_ptr_type = build_pointer_type (STMT_VINFO_VECTYPE (stmt_info));
dest = vect_get_new_vect_var (vect_ptr_type, vect_pointer_var, base_name);
addr_base = force_gimple_operand (addr_base, &seq, true, dest);
gimple_seq_add_seq (new_stmt_list, seq);
if (DR_PTR_INFO (dr)
&& TREE_CODE (addr_base) == SSA_NAME
&& !SSA_NAME_PTR_INFO (addr_base))
{
vect_duplicate_ssa_name_ptr_info (addr_base, dr);
if (offset || byte_offset)
mark_ptr_info_alignment_unknown (SSA_NAME_PTR_INFO (addr_base));
}
if (dump_enabled_p ())
{
dump_printf_loc (MSG_NOTE, vect_location, "created ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, addr_base);
dump_printf (MSG_NOTE, "\n");
}
return addr_base;
}
/* Function vect_create_data_ref_ptr.
Create a new pointer-to-AGGR_TYPE variable (ap), that points to the first
location accessed in the loop by STMT, along with the def-use update
chain to appropriately advance the pointer through the loop iterations.
Also set aliasing information for the pointer. This pointer is used by
the callers to this function to create a memory reference expression for
vector load/store access.
Input:
1. STMT: a stmt that references memory. Expected to be of the form
GIMPLE_ASSIGN <name, data-ref> or
GIMPLE_ASSIGN <data-ref, name>.
2. AGGR_TYPE: the type of the reference, which should be either a vector
or an array.
3. AT_LOOP: the loop where the vector memref is to be created.
4. OFFSET (optional): an offset to be added to the initial address accessed
by the data-ref in STMT.
5. BSI: location where the new stmts are to be placed if there is no loop
6. ONLY_INIT: indicate if ap is to be updated in the loop, or remain
pointing to the initial address.
7. BYTE_OFFSET (optional, defaults to NULL): a byte offset to be added
to the initial address accessed by the data-ref in STMT. This is
similar to OFFSET, but OFFSET is counted in elements, while BYTE_OFFSET
in bytes.
8. IV_STEP (optional, defaults to NULL): the amount that should be added
to the IV during each iteration of the loop. NULL says to move
by one copy of AGGR_TYPE up or down, depending on the step of the
data reference.
Output:
1. Declare a new ptr to vector_type, and have it point to the base of the
data reference (initial addressed accessed by the data reference).
For example, for vector of type V8HI, the following code is generated:
v8hi *ap;
ap = (v8hi *)initial_address;
if OFFSET is not supplied:
initial_address = &a[init];
if OFFSET is supplied:
initial_address = &a[init + OFFSET];
if BYTE_OFFSET is supplied:
initial_address = &a[init] + BYTE_OFFSET;
Return the initial_address in INITIAL_ADDRESS.
2. If ONLY_INIT is true, just return the initial pointer. Otherwise, also
update the pointer in each iteration of the loop.
Return the increment stmt that updates the pointer in PTR_INCR.
3. Set INV_P to true if the access pattern of the data reference in the
vectorized loop is invariant. Set it to false otherwise.
4. Return the pointer. */
tree
vect_create_data_ref_ptr (gimple *stmt, tree aggr_type, struct loop *at_loop,
tree offset, tree *initial_address,
gimple_stmt_iterator *gsi, gimple **ptr_incr,
bool only_init, bool *inv_p, tree byte_offset,
tree iv_step)
{
const char *base_name;
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
struct loop *loop = NULL;
bool nested_in_vect_loop = false;
struct loop *containing_loop = NULL;
tree aggr_ptr_type;
tree aggr_ptr;
tree new_temp;
gimple_seq new_stmt_list = NULL;
edge pe = NULL;
basic_block new_bb;
tree aggr_ptr_init;
struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
tree aptr;
gimple_stmt_iterator incr_gsi;
bool insert_after;
tree indx_before_incr, indx_after_incr;
gimple *incr;
tree step;
bb_vec_info bb_vinfo = STMT_VINFO_BB_VINFO (stmt_info);
gcc_assert (iv_step != NULL_TREE
|| TREE_CODE (aggr_type) == ARRAY_TYPE
|| TREE_CODE (aggr_type) == VECTOR_TYPE);
if (loop_vinfo)
{
loop = LOOP_VINFO_LOOP (loop_vinfo);
nested_in_vect_loop = nested_in_vect_loop_p (loop, stmt);
containing_loop = (gimple_bb (stmt))->loop_father;
pe = loop_preheader_edge (loop);
}
else
{
gcc_assert (bb_vinfo);
only_init = true;
*ptr_incr = NULL;
}
/* Check the step (evolution) of the load in LOOP, and record
whether it's invariant. */
step = vect_dr_behavior (dr)->step;
if (integer_zerop (step))
*inv_p = true;
else
*inv_p = false;
/* Create an expression for the first address accessed by this load
in LOOP. */
base_name = get_name (DR_BASE_ADDRESS (dr));
if (dump_enabled_p ())
{
tree dr_base_type = TREE_TYPE (DR_BASE_OBJECT (dr));
dump_printf_loc (MSG_NOTE, vect_location,
"create %s-pointer variable to type: ",
get_tree_code_name (TREE_CODE (aggr_type)));
dump_generic_expr (MSG_NOTE, TDF_SLIM, aggr_type);
if (TREE_CODE (dr_base_type) == ARRAY_TYPE)
dump_printf (MSG_NOTE, " vectorizing an array ref: ");
else if (TREE_CODE (dr_base_type) == VECTOR_TYPE)
dump_printf (MSG_NOTE, " vectorizing a vector ref: ");
else if (TREE_CODE (dr_base_type) == RECORD_TYPE)
dump_printf (MSG_NOTE, " vectorizing a record based array ref: ");
else
dump_printf (MSG_NOTE, " vectorizing a pointer ref: ");
dump_generic_expr (MSG_NOTE, TDF_SLIM, DR_BASE_OBJECT (dr));
dump_printf (MSG_NOTE, "\n");
}
/* (1) Create the new aggregate-pointer variable.
Vector and array types inherit the alias set of their component
type by default so we need to use a ref-all pointer if the data
reference does not conflict with the created aggregated data
reference because it is not addressable. */
bool need_ref_all = false;
if (!alias_sets_conflict_p (get_alias_set (aggr_type),
get_alias_set (DR_REF (dr))))
need_ref_all = true;
/* Likewise for any of the data references in the stmt group. */
else if (DR_GROUP_SIZE (stmt_info) > 1)
{
gimple *orig_stmt = DR_GROUP_FIRST_ELEMENT (stmt_info);
do
{
stmt_vec_info sinfo = vinfo_for_stmt (orig_stmt);
struct data_reference *sdr = STMT_VINFO_DATA_REF (sinfo);
if (!alias_sets_conflict_p (get_alias_set (aggr_type),
get_alias_set (DR_REF (sdr))))
{
need_ref_all = true;
break;
}
orig_stmt = DR_GROUP_NEXT_ELEMENT (sinfo);
}
while (orig_stmt);
}
aggr_ptr_type = build_pointer_type_for_mode (aggr_type, ptr_mode,
need_ref_all);
aggr_ptr = vect_get_new_vect_var (aggr_ptr_type, vect_pointer_var, base_name);
/* Note: If the dataref is in an inner-loop nested in LOOP, and we are
vectorizing LOOP (i.e., outer-loop vectorization), we need to create two
def-use update cycles for the pointer: one relative to the outer-loop
(LOOP), which is what steps (3) and (4) below do. The other is relative
to the inner-loop (which is the inner-most loop containing the dataref),
and this is done be step (5) below.
When vectorizing inner-most loops, the vectorized loop (LOOP) is also the
inner-most loop, and so steps (3),(4) work the same, and step (5) is
redundant. Steps (3),(4) create the following:
vp0 = &base_addr;
LOOP: vp1 = phi(vp0,vp2)
...
...
vp2 = vp1 + step
goto LOOP
If there is an inner-loop nested in loop, then step (5) will also be
applied, and an additional update in the inner-loop will be created:
vp0 = &base_addr;
LOOP: vp1 = phi(vp0,vp2)
...
inner: vp3 = phi(vp1,vp4)
vp4 = vp3 + inner_step
if () goto inner
...
vp2 = vp1 + step
if () goto LOOP */
/* (2) Calculate the initial address of the aggregate-pointer, and set
the aggregate-pointer to point to it before the loop. */
/* Create: (&(base[init_val+offset]+byte_offset) in the loop preheader. */
new_temp = vect_create_addr_base_for_vector_ref (stmt, &new_stmt_list,
offset, byte_offset);
if (new_stmt_list)
{
if (pe)
{
new_bb = gsi_insert_seq_on_edge_immediate (pe, new_stmt_list);
gcc_assert (!new_bb);
}
else
gsi_insert_seq_before (gsi, new_stmt_list, GSI_SAME_STMT);
}
*initial_address = new_temp;
aggr_ptr_init = new_temp;
/* (3) Handle the updating of the aggregate-pointer inside the loop.
This is needed when ONLY_INIT is false, and also when AT_LOOP is the
inner-loop nested in LOOP (during outer-loop vectorization). */
/* No update in loop is required. */
if (only_init && (!loop_vinfo || at_loop == loop))
aptr = aggr_ptr_init;
else
{
if (iv_step == NULL_TREE)
{
/* The step of the aggregate pointer is the type size. */
iv_step = TYPE_SIZE_UNIT (aggr_type);
/* One exception to the above is when the scalar step of the load in
LOOP is zero. In this case the step here is also zero. */
if (*inv_p)
iv_step = size_zero_node;
else if (tree_int_cst_sgn (step) == -1)
iv_step = fold_build1 (NEGATE_EXPR, TREE_TYPE (iv_step), iv_step);
}
standard_iv_increment_position (loop, &incr_gsi, &insert_after);
create_iv (aggr_ptr_init,
fold_convert (aggr_ptr_type, iv_step),
aggr_ptr, loop, &incr_gsi, insert_after,
&indx_before_incr, &indx_after_incr);
incr = gsi_stmt (incr_gsi);
set_vinfo_for_stmt (incr, new_stmt_vec_info (incr, loop_vinfo));
/* Copy the points-to information if it exists. */
if (DR_PTR_INFO (dr))
{
vect_duplicate_ssa_name_ptr_info (indx_before_incr, dr);
vect_duplicate_ssa_name_ptr_info (indx_after_incr, dr);
}
if (ptr_incr)
*ptr_incr = incr;
aptr = indx_before_incr;
}
if (!nested_in_vect_loop || only_init)
return aptr;
/* (4) Handle the updating of the aggregate-pointer inside the inner-loop
nested in LOOP, if exists. */
gcc_assert (nested_in_vect_loop);
if (!only_init)
{
standard_iv_increment_position (containing_loop, &incr_gsi,
&insert_after);
create_iv (aptr, fold_convert (aggr_ptr_type, DR_STEP (dr)), aggr_ptr,
containing_loop, &incr_gsi, insert_after, &indx_before_incr,
&indx_after_incr);
incr = gsi_stmt (incr_gsi);
set_vinfo_for_stmt (incr, new_stmt_vec_info (incr, loop_vinfo));
/* Copy the points-to information if it exists. */
if (DR_PTR_INFO (dr))
{
vect_duplicate_ssa_name_ptr_info (indx_before_incr, dr);
vect_duplicate_ssa_name_ptr_info (indx_after_incr, dr);
}
if (ptr_incr)
*ptr_incr = incr;
return indx_before_incr;
}
else
gcc_unreachable ();
}
/* Function bump_vector_ptr
Increment a pointer (to a vector type) by vector-size. If requested,
i.e. if PTR-INCR is given, then also connect the new increment stmt
to the existing def-use update-chain of the pointer, by modifying
the PTR_INCR as illustrated below:
The pointer def-use update-chain before this function:
DATAREF_PTR = phi (p_0, p_2)
....
PTR_INCR: p_2 = DATAREF_PTR + step
The pointer def-use update-chain after this function:
DATAREF_PTR = phi (p_0, p_2)
....
NEW_DATAREF_PTR = DATAREF_PTR + BUMP
....
PTR_INCR: p_2 = NEW_DATAREF_PTR + step
Input:
DATAREF_PTR - ssa_name of a pointer (to vector type) that is being updated
in the loop.
PTR_INCR - optional. The stmt that updates the pointer in each iteration of
the loop. The increment amount across iterations is expected
to be vector_size.
BSI - location where the new update stmt is to be placed.
STMT - the original scalar memory-access stmt that is being vectorized.
BUMP - optional. The offset by which to bump the pointer. If not given,
the offset is assumed to be vector_size.
Output: Return NEW_DATAREF_PTR as illustrated above.
*/
tree
bump_vector_ptr (tree dataref_ptr, gimple *ptr_incr, gimple_stmt_iterator *gsi,
gimple *stmt, tree bump)
{
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
tree update = TYPE_SIZE_UNIT (vectype);
gassign *incr_stmt;
ssa_op_iter iter;
use_operand_p use_p;
tree new_dataref_ptr;
if (bump)
update = bump;
if (TREE_CODE (dataref_ptr) == SSA_NAME)
new_dataref_ptr = copy_ssa_name (dataref_ptr);
else
new_dataref_ptr = make_ssa_name (TREE_TYPE (dataref_ptr));
incr_stmt = gimple_build_assign (new_dataref_ptr, POINTER_PLUS_EXPR,
dataref_ptr, update);
vect_finish_stmt_generation (stmt, incr_stmt, gsi);
/* Copy the points-to information if it exists. */
if (DR_PTR_INFO (dr))
{
duplicate_ssa_name_ptr_info (new_dataref_ptr, DR_PTR_INFO (dr));
mark_ptr_info_alignment_unknown (SSA_NAME_PTR_INFO (new_dataref_ptr));
}
if (!ptr_incr)
return new_dataref_ptr;
/* Update the vector-pointer's cross-iteration increment. */
FOR_EACH_SSA_USE_OPERAND (use_p, ptr_incr, iter, SSA_OP_USE)
{
tree use = USE_FROM_PTR (use_p);
if (use == dataref_ptr)
SET_USE (use_p, new_dataref_ptr);
else
gcc_assert (operand_equal_p (use, update, 0));
}
return new_dataref_ptr;
}
/* Copy memory reference info such as base/clique from the SRC reference
to the DEST MEM_REF. */
void
vect_copy_ref_info (tree dest, tree src)
{
if (TREE_CODE (dest) != MEM_REF)
return;
tree src_base = src;
while (handled_component_p (src_base))
src_base = TREE_OPERAND (src_base, 0);
if (TREE_CODE (src_base) != MEM_REF
&& TREE_CODE (src_base) != TARGET_MEM_REF)
return;
MR_DEPENDENCE_CLIQUE (dest) = MR_DEPENDENCE_CLIQUE (src_base);
MR_DEPENDENCE_BASE (dest) = MR_DEPENDENCE_BASE (src_base);
}
/* Function vect_create_destination_var.
Create a new temporary of type VECTYPE. */
tree
vect_create_destination_var (tree scalar_dest, tree vectype)
{
tree vec_dest;
const char *name;
char *new_name;
tree type;
enum vect_var_kind kind;
kind = vectype
? VECTOR_BOOLEAN_TYPE_P (vectype)
? vect_mask_var
: vect_simple_var
: vect_scalar_var;
type = vectype ? vectype : TREE_TYPE (scalar_dest);
gcc_assert (TREE_CODE (scalar_dest) == SSA_NAME);
name = get_name (scalar_dest);
if (name)
new_name = xasprintf ("%s_%u", name, SSA_NAME_VERSION (scalar_dest));
else
new_name = xasprintf ("_%u", SSA_NAME_VERSION (scalar_dest));
vec_dest = vect_get_new_vect_var (type, kind, new_name);
free (new_name);
return vec_dest;
}
/* Function vect_grouped_store_supported.
Returns TRUE if interleave high and interleave low permutations
are supported, and FALSE otherwise. */
bool
vect_grouped_store_supported (tree vectype, unsigned HOST_WIDE_INT count)
{
machine_mode mode = TYPE_MODE (vectype);
/* vect_permute_store_chain requires the group size to be equal to 3 or
be a power of two. */
if (count != 3 && exact_log2 (count) == -1)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"the size of the group of accesses"
" is not a power of 2 or not eqaul to 3\n");
return false;
}
/* Check that the permutation is supported. */
if (VECTOR_MODE_P (mode))
{
unsigned int i;
if (count == 3)
{
unsigned int j0 = 0, j1 = 0, j2 = 0;
unsigned int i, j;
unsigned int nelt;
if (!GET_MODE_NUNITS (mode).is_constant (&nelt))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"cannot handle groups of 3 stores for"
" variable-length vectors\n");
return false;
}
vec_perm_builder sel (nelt, nelt, 1);
sel.quick_grow (nelt);
vec_perm_indices indices;
for (j = 0; j < 3; j++)
{
int nelt0 = ((3 - j) * nelt) % 3;
int nelt1 = ((3 - j) * nelt + 1) % 3;
int nelt2 = ((3 - j) * nelt + 2) % 3;
for (i = 0; i < nelt; i++)
{
if (3 * i + nelt0 < nelt)
sel[3 * i + nelt0] = j0++;
if (3 * i + nelt1 < nelt)
sel[3 * i + nelt1] = nelt + j1++;
if (3 * i + nelt2 < nelt)
sel[3 * i + nelt2] = 0;
}
indices.new_vector (sel, 2, nelt);
if (!can_vec_perm_const_p (mode, indices))
{
if (dump_enabled_p ())
dump_printf (MSG_MISSED_OPTIMIZATION,
"permutation op not supported by target.\n");
return false;
}
for (i = 0; i < nelt; i++)
{
if (3 * i + nelt0 < nelt)
sel[3 * i + nelt0] = 3 * i + nelt0;
if (3 * i + nelt1 < nelt)
sel[3 * i + nelt1] = 3 * i + nelt1;
if (3 * i + nelt2 < nelt)
sel[3 * i + nelt2] = nelt + j2++;
}
indices.new_vector (sel, 2, nelt);
if (!can_vec_perm_const_p (mode, indices))
{
if (dump_enabled_p ())
dump_printf (MSG_MISSED_OPTIMIZATION,
"permutation op not supported by target.\n");
return false;
}
}
return true;
}
else
{
/* If length is not equal to 3 then only power of 2 is supported. */
gcc_assert (pow2p_hwi (count));
poly_uint64 nelt = GET_MODE_NUNITS (mode);
/* The encoding has 2 interleaved stepped patterns. */
vec_perm_builder sel (nelt, 2, 3);
sel.quick_grow (6);
for (i = 0; i < 3; i++)
{
sel[i * 2] = i;
sel[i * 2 + 1] = i + nelt;
}
vec_perm_indices indices (sel, 2, nelt);
if (can_vec_perm_const_p (mode, indices))
{
for (i = 0; i < 6; i++)
sel[i] += exact_div (nelt, 2);
indices.new_vector (sel, 2, nelt);
if (can_vec_perm_const_p (mode, indices))
return true;
}
}
}
if (dump_enabled_p ())
dump_printf (MSG_MISSED_OPTIMIZATION,
"permutaion op not supported by target.\n");
return false;
}
/* Return TRUE if vec_{mask_}store_lanes is available for COUNT vectors of
type VECTYPE. MASKED_P says whether the masked form is needed. */
bool
vect_store_lanes_supported (tree vectype, unsigned HOST_WIDE_INT count,
bool masked_p)
{
if (masked_p)
return vect_lanes_optab_supported_p ("vec_mask_store_lanes",
vec_mask_store_lanes_optab,
vectype, count);
else
return vect_lanes_optab_supported_p ("vec_store_lanes",
vec_store_lanes_optab,
vectype, count);
}
/* Function vect_permute_store_chain.
Given a chain of interleaved stores in DR_CHAIN of LENGTH that must be
a power of 2 or equal to 3, generate interleave_high/low stmts to reorder
the data correctly for the stores. Return the final references for stores
in RESULT_CHAIN.
E.g., LENGTH is 4 and the scalar type is short, i.e., VF is 8.
The input is 4 vectors each containing 8 elements. We assign a number to
each element, the input sequence is:
1st vec: 0 1 2 3 4 5 6 7
2nd vec: 8 9 10 11 12 13 14 15
3rd vec: 16 17 18 19 20 21 22 23
4th vec: 24 25 26 27 28 29 30 31
The output sequence should be:
1st vec: 0 8 16 24 1 9 17 25
2nd vec: 2 10 18 26 3 11 19 27
3rd vec: 4 12 20 28 5 13 21 30
4th vec: 6 14 22 30 7 15 23 31
i.e., we interleave the contents of the four vectors in their order.
We use interleave_high/low instructions to create such output. The input of
each interleave_high/low operation is two vectors:
1st vec 2nd vec
0 1 2 3 4 5 6 7
the even elements of the result vector are obtained left-to-right from the
high/low elements of the first vector. The odd elements of the result are
obtained left-to-right from the high/low elements of the second vector.
The output of interleave_high will be: 0 4 1 5
and of interleave_low: 2 6 3 7
The permutation is done in log LENGTH stages. In each stage interleave_high
and interleave_low stmts are created for each pair of vectors in DR_CHAIN,
where the first argument is taken from the first half of DR_CHAIN and the
second argument from it's second half.
In our example,
I1: interleave_high (1st vec, 3rd vec)
I2: interleave_low (1st vec, 3rd vec)
I3: interleave_high (2nd vec, 4th vec)
I4: interleave_low (2nd vec, 4th vec)
The output for the first stage is:
I1: 0 16 1 17 2 18 3 19
I2: 4 20 5 21 6 22 7 23
I3: 8 24 9 25 10 26 11 27
I4: 12 28 13 29 14 30 15 31
The output of the second stage, i.e. the final result is:
I1: 0 8 16 24 1 9 17 25
I2: 2 10 18 26 3 11 19 27
I3: 4 12 20 28 5 13 21 30
I4: 6 14 22 30 7 15 23 31. */
void
vect_permute_store_chain (vec<tree> dr_chain,
unsigned int length,
gimple *stmt,
gimple_stmt_iterator *gsi,
vec<tree> *result_chain)
{
tree vect1, vect2, high, low;
gimple *perm_stmt;
tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt));
tree perm_mask_low, perm_mask_high;
tree data_ref;
tree perm3_mask_low, perm3_mask_high;
unsigned int i, j, n, log_length = exact_log2 (length);
result_chain->quick_grow (length);
memcpy (result_chain->address (), dr_chain.address (),
length * sizeof (tree));
if (length == 3)
{
/* vect_grouped_store_supported ensures that this is constant. */
unsigned int nelt = TYPE_VECTOR_SUBPARTS (vectype).to_constant ();
unsigned int j0 = 0, j1 = 0, j2 = 0;
vec_perm_builder sel (nelt, nelt, 1);
sel.quick_grow (nelt);
vec_perm_indices indices;
for (j = 0; j < 3; j++)
{
int nelt0 = ((3 - j) * nelt) % 3;
int nelt1 = ((3 - j) * nelt + 1) % 3;
int nelt2 = ((3 - j) * nelt + 2) % 3;
for (i = 0; i < nelt; i++)
{
if (3 * i + nelt0 < nelt)
sel[3 * i + nelt0] = j0++;
if (3 * i + nelt1 < nelt)
sel[3 * i + nelt1] = nelt + j1++;
if (3 * i + nelt2 < nelt)
sel[3 * i + nelt2] = 0;
}
indices.new_vector (sel, 2, nelt);
perm3_mask_low = vect_gen_perm_mask_checked (vectype, indices);
for (i = 0; i < nelt; i++)
{
if (3 * i + nelt0 < nelt)
sel[3 * i + nelt0] = 3 * i + nelt0;
if (3 * i + nelt1 < nelt)
sel[3 * i + nelt1] = 3 * i + nelt1;
if (3 * i + nelt2 < nelt)
sel[3 * i + nelt2] = nelt + j2++;
}
indices.new_vector (sel, 2, nelt);
perm3_mask_high = vect_gen_perm_mask_checked (vectype, indices);
vect1 = dr_chain[0];
vect2 = dr_chain[1];
/* Create interleaving stmt:
low = VEC_PERM_EXPR <vect1, vect2,
{j, nelt, *, j + 1, nelt + j + 1, *,
j + 2, nelt + j + 2, *, ...}> */
data_ref = make_temp_ssa_name (vectype, NULL, "vect_shuffle3_low");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR, vect1,
vect2, perm3_mask_low);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
vect1 = data_ref;
vect2 = dr_chain[2];
/* Create interleaving stmt:
low = VEC_PERM_EXPR <vect1, vect2,
{0, 1, nelt + j, 3, 4, nelt + j + 1,
6, 7, nelt + j + 2, ...}> */
data_ref = make_temp_ssa_name (vectype, NULL, "vect_shuffle3_high");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR, vect1,
vect2, perm3_mask_high);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
(*result_chain)[j] = data_ref;
}
}
else
{
/* If length is not equal to 3 then only power of 2 is supported. */
gcc_assert (pow2p_hwi (length));
/* The encoding has 2 interleaved stepped patterns. */
poly_uint64 nelt = TYPE_VECTOR_SUBPARTS (vectype);
vec_perm_builder sel (nelt, 2, 3);
sel.quick_grow (6);
for (i = 0; i < 3; i++)
{
sel[i * 2] = i;
sel[i * 2 + 1] = i + nelt;
}
vec_perm_indices indices (sel, 2, nelt);
perm_mask_high = vect_gen_perm_mask_checked (vectype, indices);
for (i = 0; i < 6; i++)
sel[i] += exact_div (nelt, 2);
indices.new_vector (sel, 2, nelt);
perm_mask_low = vect_gen_perm_mask_checked (vectype, indices);
for (i = 0, n = log_length; i < n; i++)
{
for (j = 0; j < length/2; j++)
{
vect1 = dr_chain[j];
vect2 = dr_chain[j+length/2];
/* Create interleaving stmt:
high = VEC_PERM_EXPR <vect1, vect2, {0, nelt, 1, nelt+1,
...}> */
high = make_temp_ssa_name (vectype, NULL, "vect_inter_high");
perm_stmt = gimple_build_assign (high, VEC_PERM_EXPR, vect1,
vect2, perm_mask_high);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
(*result_chain)[2*j] = high;
/* Create interleaving stmt:
low = VEC_PERM_EXPR <vect1, vect2,
{nelt/2, nelt*3/2, nelt/2+1, nelt*3/2+1,
...}> */
low = make_temp_ssa_name (vectype, NULL, "vect_inter_low");
perm_stmt = gimple_build_assign (low, VEC_PERM_EXPR, vect1,
vect2, perm_mask_low);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
(*result_chain)[2*j+1] = low;
}
memcpy (dr_chain.address (), result_chain->address (),
length * sizeof (tree));
}
}
}
/* Function vect_setup_realignment
This function is called when vectorizing an unaligned load using
the dr_explicit_realign[_optimized] scheme.
This function generates the following code at the loop prolog:
p = initial_addr;
x msq_init = *(floor(p)); # prolog load
realignment_token = call target_builtin;
loop:
x msq = phi (msq_init, ---)
The stmts marked with x are generated only for the case of
dr_explicit_realign_optimized.
The code above sets up a new (vector) pointer, pointing to the first
location accessed by STMT, and a "floor-aligned" load using that pointer.
It also generates code to compute the "realignment-token" (if the relevant
target hook was defined), and creates a phi-node at the loop-header bb
whose arguments are the result of the prolog-load (created by this
function) and the result of a load that takes place in the loop (to be
created by the caller to this function).
For the case of dr_explicit_realign_optimized:
The caller to this function uses the phi-result (msq) to create the
realignment code inside the loop, and sets up the missing phi argument,
as follows:
loop:
msq = phi (msq_init, lsq)
lsq = *(floor(p')); # load in loop
result = realign_load (msq, lsq, realignment_token);
For the case of dr_explicit_realign:
loop:
msq = *(floor(p)); # load in loop
p' = p + (VS-1);
lsq = *(floor(p')); # load in loop
result = realign_load (msq, lsq, realignment_token);
Input:
STMT - (scalar) load stmt to be vectorized. This load accesses
a memory location that may be unaligned.
BSI - place where new code is to be inserted.
ALIGNMENT_SUPPORT_SCHEME - which of the two misalignment handling schemes
is used.
Output:
REALIGNMENT_TOKEN - the result of a call to the builtin_mask_for_load
target hook, if defined.
Return value - the result of the loop-header phi node. */
tree
vect_setup_realignment (gimple *stmt, gimple_stmt_iterator *gsi,
tree *realignment_token,
enum dr_alignment_support alignment_support_scheme,
tree init_addr,
struct loop **at_loop)
{
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
struct data_reference *dr = STMT_VINFO_DATA_REF (stmt_info);
struct loop *loop = NULL;
edge pe = NULL;
tree scalar_dest = gimple_assign_lhs (stmt);
tree vec_dest;
gimple *inc;
tree ptr;
tree data_ref;
basic_block new_bb;
tree msq_init = NULL_TREE;
tree new_temp;
gphi *phi_stmt;
tree msq = NULL_TREE;
gimple_seq stmts = NULL;
bool inv_p;
bool compute_in_loop = false;
bool nested_in_vect_loop = false;
struct loop *containing_loop = (gimple_bb (stmt))->loop_father;
struct loop *loop_for_initial_load = NULL;
if (loop_vinfo)
{
loop = LOOP_VINFO_LOOP (loop_vinfo);
nested_in_vect_loop = nested_in_vect_loop_p (loop, stmt);
}
gcc_assert (alignment_support_scheme == dr_explicit_realign
|| alignment_support_scheme == dr_explicit_realign_optimized);
/* We need to generate three things:
1. the misalignment computation
2. the extra vector load (for the optimized realignment scheme).
3. the phi node for the two vectors from which the realignment is
done (for the optimized realignment scheme). */
/* 1. Determine where to generate the misalignment computation.
If INIT_ADDR is NULL_TREE, this indicates that the misalignment
calculation will be generated by this function, outside the loop (in the
preheader). Otherwise, INIT_ADDR had already been computed for us by the
caller, inside the loop.
Background: If the misalignment remains fixed throughout the iterations of
the loop, then both realignment schemes are applicable, and also the
misalignment computation can be done outside LOOP. This is because we are
vectorizing LOOP, and so the memory accesses in LOOP advance in steps that
are a multiple of VS (the Vector Size), and therefore the misalignment in
different vectorized LOOP iterations is always the same.
The problem arises only if the memory access is in an inner-loop nested
inside LOOP, which is now being vectorized using outer-loop vectorization.
This is the only case when the misalignment of the memory access may not
remain fixed throughout the iterations of the inner-loop (as explained in
detail in vect_supportable_dr_alignment). In this case, not only is the
optimized realignment scheme not applicable, but also the misalignment
computation (and generation of the realignment token that is passed to
REALIGN_LOAD) have to be done inside the loop.
In short, INIT_ADDR indicates whether we are in a COMPUTE_IN_LOOP mode
or not, which in turn determines if the misalignment is computed inside
the inner-loop, or outside LOOP. */
if (init_addr != NULL_TREE || !loop_vinfo)
{
compute_in_loop = true;
gcc_assert (alignment_support_scheme == dr_explicit_realign);
}
/* 2. Determine where to generate the extra vector load.
For the optimized realignment scheme, instead of generating two vector
loads in each iteration, we generate a single extra vector load in the
preheader of the loop, and in each iteration reuse the result of the
vector load from the previous iteration. In case the memory access is in
an inner-loop nested inside LOOP, which is now being vectorized using
outer-loop vectorization, we need to determine whether this initial vector
load should be generated at the preheader of the inner-loop, or can be
generated at the preheader of LOOP. If the memory access has no evolution
in LOOP, it can be generated in the preheader of LOOP. Otherwise, it has
to be generated inside LOOP (in the preheader of the inner-loop). */
if (nested_in_vect_loop)
{
tree outerloop_step = STMT_VINFO_DR_STEP (stmt_info);
bool invariant_in_outerloop =
(tree_int_cst_compare (outerloop_step, size_zero_node) == 0);
loop_for_initial_load = (invariant_in_outerloop ? loop : loop->inner);
}
else
loop_for_initial_load = loop;
if (at_loop)
*at_loop = loop_for_initial_load;
if (loop_for_initial_load)
pe = loop_preheader_edge (loop_for_initial_load);
/* 3. For the case of the optimized realignment, create the first vector
load at the loop preheader. */
if (alignment_support_scheme == dr_explicit_realign_optimized)
{
/* Create msq_init = *(floor(p1)) in the loop preheader */
gassign *new_stmt;
gcc_assert (!compute_in_loop);
vec_dest = vect_create_destination_var (scalar_dest, vectype);
ptr = vect_create_data_ref_ptr (stmt, vectype, loop_for_initial_load,
NULL_TREE, &init_addr, NULL, &inc,
true, &inv_p);
if (TREE_CODE (ptr) == SSA_NAME)
new_temp = copy_ssa_name (ptr);
else
new_temp = make_ssa_name (TREE_TYPE (ptr));
unsigned int align = DR_TARGET_ALIGNMENT (dr);
new_stmt = gimple_build_assign
(new_temp, BIT_AND_EXPR, ptr,
build_int_cst (TREE_TYPE (ptr), -(HOST_WIDE_INT) align));
new_bb = gsi_insert_on_edge_immediate (pe, new_stmt);
gcc_assert (!new_bb);
data_ref
= build2 (MEM_REF, TREE_TYPE (vec_dest), new_temp,
build_int_cst (reference_alias_ptr_type (DR_REF (dr)), 0));
vect_copy_ref_info (data_ref, DR_REF (dr));
new_stmt = gimple_build_assign (vec_dest, data_ref);
new_temp = make_ssa_name (vec_dest, new_stmt);
gimple_assign_set_lhs (new_stmt, new_temp);
if (pe)
{
new_bb = gsi_insert_on_edge_immediate (pe, new_stmt);
gcc_assert (!new_bb);
}
else
gsi_insert_before (gsi, new_stmt, GSI_SAME_STMT);
msq_init = gimple_assign_lhs (new_stmt);
}
/* 4. Create realignment token using a target builtin, if available.
It is done either inside the containing loop, or before LOOP (as
determined above). */
if (targetm.vectorize.builtin_mask_for_load)
{
gcall *new_stmt;
tree builtin_decl;
/* Compute INIT_ADDR - the initial addressed accessed by this memref. */
if (!init_addr)
{
/* Generate the INIT_ADDR computation outside LOOP. */
init_addr = vect_create_addr_base_for_vector_ref (stmt, &stmts,
NULL_TREE);
if (loop)
{
pe = loop_preheader_edge (loop);
new_bb = gsi_insert_seq_on_edge_immediate (pe, stmts);
gcc_assert (!new_bb);
}
else
gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT);
}
builtin_decl = targetm.vectorize.builtin_mask_for_load ();
new_stmt = gimple_build_call (builtin_decl, 1, init_addr);
vec_dest =
vect_create_destination_var (scalar_dest,
gimple_call_return_type (new_stmt));
new_temp = make_ssa_name (vec_dest, new_stmt);
gimple_call_set_lhs (new_stmt, new_temp);
if (compute_in_loop)
gsi_insert_before (gsi, new_stmt, GSI_SAME_STMT);
else
{
/* Generate the misalignment computation outside LOOP. */
pe = loop_preheader_edge (loop);
new_bb = gsi_insert_on_edge_immediate (pe, new_stmt);
gcc_assert (!new_bb);
}
*realignment_token = gimple_call_lhs (new_stmt);
/* The result of the CALL_EXPR to this builtin is determined from
the value of the parameter and no global variables are touched
which makes the builtin a "const" function. Requiring the
builtin to have the "const" attribute makes it unnecessary
to call mark_call_clobbered. */
gcc_assert (TREE_READONLY (builtin_decl));
}
if (alignment_support_scheme == dr_explicit_realign)
return msq;
gcc_assert (!compute_in_loop);
gcc_assert (alignment_support_scheme == dr_explicit_realign_optimized);
/* 5. Create msq = phi <msq_init, lsq> in loop */
pe = loop_preheader_edge (containing_loop);
vec_dest = vect_create_destination_var (scalar_dest, vectype);
msq = make_ssa_name (vec_dest);
phi_stmt = create_phi_node (msq, containing_loop->header);
add_phi_arg (phi_stmt, msq_init, pe, UNKNOWN_LOCATION);
return msq;
}
/* Function vect_grouped_load_supported.
COUNT is the size of the load group (the number of statements plus the
number of gaps). SINGLE_ELEMENT_P is true if there is actually
only one statement, with a gap of COUNT - 1.
Returns true if a suitable permute exists. */
bool
vect_grouped_load_supported (tree vectype, bool single_element_p,
unsigned HOST_WIDE_INT count)
{
machine_mode mode = TYPE_MODE (vectype);
/* If this is single-element interleaving with an element distance
that leaves unused vector loads around punt - we at least create
very sub-optimal code in that case (and blow up memory,
see PR65518). */
if (single_element_p && maybe_gt (count, TYPE_VECTOR_SUBPARTS (vectype)))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"single-element interleaving not supported "
"for not adjacent vector loads\n");
return false;
}
/* vect_permute_load_chain requires the group size to be equal to 3 or
be a power of two. */
if (count != 3 && exact_log2 (count) == -1)
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"the size of the group of accesses"
" is not a power of 2 or not equal to 3\n");
return false;
}
/* Check that the permutation is supported. */
if (VECTOR_MODE_P (mode))
{
unsigned int i, j;
if (count == 3)
{
unsigned int nelt;
if (!GET_MODE_NUNITS (mode).is_constant (&nelt))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"cannot handle groups of 3 loads for"
" variable-length vectors\n");
return false;
}
vec_perm_builder sel (nelt, nelt, 1);
sel.quick_grow (nelt);
vec_perm_indices indices;
unsigned int k;
for (k = 0; k < 3; k++)
{
for (i = 0; i < nelt; i++)
if (3 * i + k < 2 * nelt)
sel[i] = 3 * i + k;
else
sel[i] = 0;
indices.new_vector (sel, 2, nelt);
if (!can_vec_perm_const_p (mode, indices))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"shuffle of 3 loads is not supported by"
" target\n");
return false;
}
for (i = 0, j = 0; i < nelt; i++)
if (3 * i + k < 2 * nelt)
sel[i] = i;
else
sel[i] = nelt + ((nelt + k) % 3) + 3 * (j++);
indices.new_vector (sel, 2, nelt);
if (!can_vec_perm_const_p (mode, indices))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"shuffle of 3 loads is not supported by"
" target\n");
return false;
}
}
return true;
}
else
{
/* If length is not equal to 3 then only power of 2 is supported. */
gcc_assert (pow2p_hwi (count));
poly_uint64 nelt = GET_MODE_NUNITS (mode);
/* The encoding has a single stepped pattern. */
vec_perm_builder sel (nelt, 1, 3);
sel.quick_grow (3);
for (i = 0; i < 3; i++)
sel[i] = i * 2;
vec_perm_indices indices (sel, 2, nelt);
if (can_vec_perm_const_p (mode, indices))
{
for (i = 0; i < 3; i++)
sel[i] = i * 2 + 1;
indices.new_vector (sel, 2, nelt);
if (can_vec_perm_const_p (mode, indices))
return true;
}
}
}
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"extract even/odd not supported by target\n");
return false;
}
/* Return TRUE if vec_{masked_}load_lanes is available for COUNT vectors of
type VECTYPE. MASKED_P says whether the masked form is needed. */
bool
vect_load_lanes_supported (tree vectype, unsigned HOST_WIDE_INT count,
bool masked_p)
{
if (masked_p)
return vect_lanes_optab_supported_p ("vec_mask_load_lanes",
vec_mask_load_lanes_optab,
vectype, count);
else
return vect_lanes_optab_supported_p ("vec_load_lanes",
vec_load_lanes_optab,
vectype, count);
}
/* Function vect_permute_load_chain.
Given a chain of interleaved loads in DR_CHAIN of LENGTH that must be
a power of 2 or equal to 3, generate extract_even/odd stmts to reorder
the input data correctly. Return the final references for loads in
RESULT_CHAIN.
E.g., LENGTH is 4 and the scalar type is short, i.e., VF is 8.
The input is 4 vectors each containing 8 elements. We assign a number to each
element, the input sequence is:
1st vec: 0 1 2 3 4 5 6 7
2nd vec: 8 9 10 11 12 13 14 15
3rd vec: 16 17 18 19 20 21 22 23
4th vec: 24 25 26 27 28 29 30 31
The output sequence should be:
1st vec: 0 4 8 12 16 20 24 28
2nd vec: 1 5 9 13 17 21 25 29
3rd vec: 2 6 10 14 18 22 26 30
4th vec: 3 7 11 15 19 23 27 31
i.e., the first output vector should contain the first elements of each
interleaving group, etc.
We use extract_even/odd instructions to create such output. The input of
each extract_even/odd operation is two vectors
1st vec 2nd vec
0 1 2 3 4 5 6 7
and the output is the vector of extracted even/odd elements. The output of
extract_even will be: 0 2 4 6
and of extract_odd: 1 3 5 7
The permutation is done in log LENGTH stages. In each stage extract_even
and extract_odd stmts are created for each pair of vectors in DR_CHAIN in
their order. In our example,
E1: extract_even (1st vec, 2nd vec)
E2: extract_odd (1st vec, 2nd vec)
E3: extract_even (3rd vec, 4th vec)
E4: extract_odd (3rd vec, 4th vec)
The output for the first stage will be:
E1: 0 2 4 6 8 10 12 14
E2: 1 3 5 7 9 11 13 15
E3: 16 18 20 22 24 26 28 30
E4: 17 19 21 23 25 27 29 31
In order to proceed and create the correct sequence for the next stage (or
for the correct output, if the second stage is the last one, as in our
example), we first put the output of extract_even operation and then the
output of extract_odd in RESULT_CHAIN (which is then copied to DR_CHAIN).
The input for the second stage is:
1st vec (E1): 0 2 4 6 8 10 12 14
2nd vec (E3): 16 18 20 22 24 26 28 30
3rd vec (E2): 1 3 5 7 9 11 13 15
4th vec (E4): 17 19 21 23 25 27 29 31
The output of the second stage:
E1: 0 4 8 12 16 20 24 28
E2: 2 6 10 14 18 22 26 30
E3: 1 5 9 13 17 21 25 29
E4: 3 7 11 15 19 23 27 31
And RESULT_CHAIN after reordering:
1st vec (E1): 0 4 8 12 16 20 24 28
2nd vec (E3): 1 5 9 13 17 21 25 29
3rd vec (E2): 2 6 10 14 18 22 26 30
4th vec (E4): 3 7 11 15 19 23 27 31. */
static void
vect_permute_load_chain (vec<tree> dr_chain,
unsigned int length,
gimple *stmt,
gimple_stmt_iterator *gsi,
vec<tree> *result_chain)
{
tree data_ref, first_vect, second_vect;
tree perm_mask_even, perm_mask_odd;
tree perm3_mask_low, perm3_mask_high;
gimple *perm_stmt;
tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt));
unsigned int i, j, log_length = exact_log2 (length);
result_chain->quick_grow (length);
memcpy (result_chain->address (), dr_chain.address (),
length * sizeof (tree));
if (length == 3)
{
/* vect_grouped_load_supported ensures that this is constant. */
unsigned nelt = TYPE_VECTOR_SUBPARTS (vectype).to_constant ();
unsigned int k;
vec_perm_builder sel (nelt, nelt, 1);
sel.quick_grow (nelt);
vec_perm_indices indices;
for (k = 0; k < 3; k++)
{
for (i = 0; i < nelt; i++)
if (3 * i + k < 2 * nelt)
sel[i] = 3 * i + k;
else
sel[i] = 0;
indices.new_vector (sel, 2, nelt);
perm3_mask_low = vect_gen_perm_mask_checked (vectype, indices);
for (i = 0, j = 0; i < nelt; i++)
if (3 * i + k < 2 * nelt)
sel[i] = i;
else
sel[i] = nelt + ((nelt + k) % 3) + 3 * (j++);
indices.new_vector (sel, 2, nelt);
perm3_mask_high = vect_gen_perm_mask_checked (vectype, indices);
first_vect = dr_chain[0];
second_vect = dr_chain[1];
/* Create interleaving stmt (low part of):
low = VEC_PERM_EXPR <first_vect, second_vect2, {k, 3 + k, 6 + k,
...}> */
data_ref = make_temp_ssa_name (vectype, NULL, "vect_shuffle3_low");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR, first_vect,
second_vect, perm3_mask_low);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
/* Create interleaving stmt (high part of):
high = VEC_PERM_EXPR <first_vect, second_vect2, {k, 3 + k, 6 + k,
...}> */
first_vect = data_ref;
second_vect = dr_chain[2];
data_ref = make_temp_ssa_name (vectype, NULL, "vect_shuffle3_high");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR, first_vect,
second_vect, perm3_mask_high);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
(*result_chain)[k] = data_ref;
}
}
else
{
/* If length is not equal to 3 then only power of 2 is supported. */
gcc_assert (pow2p_hwi (length));
/* The encoding has a single stepped pattern. */
poly_uint64 nelt = TYPE_VECTOR_SUBPARTS (vectype);
vec_perm_builder sel (nelt, 1, 3);
sel.quick_grow (3);
for (i = 0; i < 3; ++i)
sel[i] = i * 2;
vec_perm_indices indices (sel, 2, nelt);
perm_mask_even = vect_gen_perm_mask_checked (vectype, indices);
for (i = 0; i < 3; ++i)
sel[i] = i * 2 + 1;
indices.new_vector (sel, 2, nelt);
perm_mask_odd = vect_gen_perm_mask_checked (vectype, indices);
for (i = 0; i < log_length; i++)
{
for (j = 0; j < length; j += 2)
{
first_vect = dr_chain[j];
second_vect = dr_chain[j+1];
/* data_ref = permute_even (first_data_ref, second_data_ref); */
data_ref = make_temp_ssa_name (vectype, NULL, "vect_perm_even");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR,
first_vect, second_vect,
perm_mask_even);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
(*result_chain)[j/2] = data_ref;
/* data_ref = permute_odd (first_data_ref, second_data_ref); */
data_ref = make_temp_ssa_name (vectype, NULL, "vect_perm_odd");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR,
first_vect, second_vect,
perm_mask_odd);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
(*result_chain)[j/2+length/2] = data_ref;
}
memcpy (dr_chain.address (), result_chain->address (),
length * sizeof (tree));
}
}
}
/* Function vect_shift_permute_load_chain.
Given a chain of loads in DR_CHAIN of LENGTH 2 or 3, generate
sequence of stmts to reorder the input data accordingly.
Return the final references for loads in RESULT_CHAIN.
Return true if successed, false otherwise.
E.g., LENGTH is 3 and the scalar type is short, i.e., VF is 8.
The input is 3 vectors each containing 8 elements. We assign a
number to each element, the input sequence is:
1st vec: 0 1 2 3 4 5 6 7
2nd vec: 8 9 10 11 12 13 14 15
3rd vec: 16 17 18 19 20 21 22 23
The output sequence should be:
1st vec: 0 3 6 9 12 15 18 21
2nd vec: 1 4 7 10 13 16 19 22
3rd vec: 2 5 8 11 14 17 20 23
We use 3 shuffle instructions and 3 * 3 - 1 shifts to create such output.
First we shuffle all 3 vectors to get correct elements order:
1st vec: ( 0 3 6) ( 1 4 7) ( 2 5)
2nd vec: ( 8 11 14) ( 9 12 15) (10 13)
3rd vec: (16 19 22) (17 20 23) (18 21)
Next we unite and shift vector 3 times:
1st step:
shift right by 6 the concatenation of:
"1st vec" and "2nd vec"
( 0 3 6) ( 1 4 7) |( 2 5) _ ( 8 11 14) ( 9 12 15)| (10 13)
"2nd vec" and "3rd vec"
( 8 11 14) ( 9 12 15) |(10 13) _ (16 19 22) (17 20 23)| (18 21)
"3rd vec" and "1st vec"
(16 19 22) (17 20 23) |(18 21) _ ( 0 3 6) ( 1 4 7)| ( 2 5)
| New vectors |
So that now new vectors are:
1st vec: ( 2 5) ( 8 11 14) ( 9 12 15)
2nd vec: (10 13) (16 19 22) (17 20 23)
3rd vec: (18 21) ( 0 3 6) ( 1 4 7)
2nd step:
shift right by 5 the concatenation of:
"1st vec" and "3rd vec"
( 2 5) ( 8 11 14) |( 9 12 15) _ (18 21) ( 0 3 6)| ( 1 4 7)
"2nd vec" and "1st vec"
(10 13) (16 19 22) |(17 20 23) _ ( 2 5) ( 8 11 14)| ( 9 12 15)
"3rd vec" and "2nd vec"
(18 21) ( 0 3 6) |( 1 4 7) _ (10 13) (16 19 22)| (17 20 23)
| New vectors |
So that now new vectors are:
1st vec: ( 9 12 15) (18 21) ( 0 3 6)
2nd vec: (17 20 23) ( 2 5) ( 8 11 14)
3rd vec: ( 1 4 7) (10 13) (16 19 22) READY
3rd step:
shift right by 5 the concatenation of:
"1st vec" and "1st vec"
( 9 12 15) (18 21) |( 0 3 6) _ ( 9 12 15) (18 21)| ( 0 3 6)
shift right by 3 the concatenation of:
"2nd vec" and "2nd vec"
(17 20 23) |( 2 5) ( 8 11 14) _ (17 20 23)| ( 2 5) ( 8 11 14)
| New vectors |
So that now all vectors are READY:
1st vec: ( 0 3 6) ( 9 12 15) (18 21)
2nd vec: ( 2 5) ( 8 11 14) (17 20 23)
3rd vec: ( 1 4 7) (10 13) (16 19 22)
This algorithm is faster than one in vect_permute_load_chain if:
1. "shift of a concatination" is faster than general permutation.
This is usually so.
2. The TARGET machine can't execute vector instructions in parallel.
This is because each step of the algorithm depends on previous.
The algorithm in vect_permute_load_chain is much more parallel.
The algorithm is applicable only for LOAD CHAIN LENGTH less than VF.
*/
static bool
vect_shift_permute_load_chain (vec<tree> dr_chain,
unsigned int length,
gimple *stmt,
gimple_stmt_iterator *gsi,
vec<tree> *result_chain)
{
tree vect[3], vect_shift[3], data_ref, first_vect, second_vect;
tree perm2_mask1, perm2_mask2, perm3_mask;
tree select_mask, shift1_mask, shift2_mask, shift3_mask, shift4_mask;
gimple *perm_stmt;
tree vectype = STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt));
unsigned int i;
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
unsigned HOST_WIDE_INT nelt, vf;
if (!TYPE_VECTOR_SUBPARTS (vectype).is_constant (&nelt)
|| !LOOP_VINFO_VECT_FACTOR (loop_vinfo).is_constant (&vf))
/* Not supported for variable-length vectors. */
return false;
vec_perm_builder sel (nelt, nelt, 1);
sel.quick_grow (nelt);
result_chain->quick_grow (length);
memcpy (result_chain->address (), dr_chain.address (),
length * sizeof (tree));
if (pow2p_hwi (length) && vf > 4)
{
unsigned int j, log_length = exact_log2 (length);
for (i = 0; i < nelt / 2; ++i)
sel[i] = i * 2;
for (i = 0; i < nelt / 2; ++i)
sel[nelt / 2 + i] = i * 2 + 1;
vec_perm_indices indices (sel, 2, nelt);
if (!can_vec_perm_const_p (TYPE_MODE (vectype), indices))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"shuffle of 2 fields structure is not \
supported by target\n");
return false;
}
perm2_mask1 = vect_gen_perm_mask_checked (vectype, indices);
for (i = 0; i < nelt / 2; ++i)
sel[i] = i * 2 + 1;
for (i = 0; i < nelt / 2; ++i)
sel[nelt / 2 + i] = i * 2;
indices.new_vector (sel, 2, nelt);
if (!can_vec_perm_const_p (TYPE_MODE (vectype), indices))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"shuffle of 2 fields structure is not \
supported by target\n");
return false;
}
perm2_mask2 = vect_gen_perm_mask_checked (vectype, indices);
/* Generating permutation constant to shift all elements.
For vector length 8 it is {4 5 6 7 8 9 10 11}. */
for (i = 0; i < nelt; i++)
sel[i] = nelt / 2 + i;
indices.new_vector (sel, 2, nelt);
if (!can_vec_perm_const_p (TYPE_MODE (vectype), indices))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"shift permutation is not supported by target\n");
return false;
}
shift1_mask = vect_gen_perm_mask_checked (vectype, indices);
/* Generating permutation constant to select vector from 2.
For vector length 8 it is {0 1 2 3 12 13 14 15}. */
for (i = 0; i < nelt / 2; i++)
sel[i] = i;
for (i = nelt / 2; i < nelt; i++)
sel[i] = nelt + i;
indices.new_vector (sel, 2, nelt);
if (!can_vec_perm_const_p (TYPE_MODE (vectype), indices))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"select is not supported by target\n");
return false;
}
select_mask = vect_gen_perm_mask_checked (vectype, indices);
for (i = 0; i < log_length; i++)
{
for (j = 0; j < length; j += 2)
{
first_vect = dr_chain[j];
second_vect = dr_chain[j + 1];
data_ref = make_temp_ssa_name (vectype, NULL, "vect_shuffle2");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR,
first_vect, first_vect,
perm2_mask1);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
vect[0] = data_ref;
data_ref = make_temp_ssa_name (vectype, NULL, "vect_shuffle2");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR,
second_vect, second_vect,
perm2_mask2);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
vect[1] = data_ref;
data_ref = make_temp_ssa_name (vectype, NULL, "vect_shift");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR,
vect[0], vect[1], shift1_mask);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
(*result_chain)[j/2 + length/2] = data_ref;
data_ref = make_temp_ssa_name (vectype, NULL, "vect_select");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR,
vect[0], vect[1], select_mask);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
(*result_chain)[j/2] = data_ref;
}
memcpy (dr_chain.address (), result_chain->address (),
length * sizeof (tree));
}
return true;
}
if (length == 3 && vf > 2)
{
unsigned int k = 0, l = 0;
/* Generating permutation constant to get all elements in rigth order.
For vector length 8 it is {0 3 6 1 4 7 2 5}. */
for (i = 0; i < nelt; i++)
{
if (3 * k + (l % 3) >= nelt)
{
k = 0;
l += (3 - (nelt % 3));
}
sel[i] = 3 * k + (l % 3);
k++;
}
vec_perm_indices indices (sel, 2, nelt);
if (!can_vec_perm_const_p (TYPE_MODE (vectype), indices))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"shuffle of 3 fields structure is not \
supported by target\n");
return false;
}
perm3_mask = vect_gen_perm_mask_checked (vectype, indices);
/* Generating permutation constant to shift all elements.
For vector length 8 it is {6 7 8 9 10 11 12 13}. */
for (i = 0; i < nelt; i++)
sel[i] = 2 * (nelt / 3) + (nelt % 3) + i;
indices.new_vector (sel, 2, nelt);
if (!can_vec_perm_const_p (TYPE_MODE (vectype), indices))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"shift permutation is not supported by target\n");
return false;
}
shift1_mask = vect_gen_perm_mask_checked (vectype, indices);
/* Generating permutation constant to shift all elements.
For vector length 8 it is {5 6 7 8 9 10 11 12}. */
for (i = 0; i < nelt; i++)
sel[i] = 2 * (nelt / 3) + 1 + i;
indices.new_vector (sel, 2, nelt);
if (!can_vec_perm_const_p (TYPE_MODE (vectype), indices))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"shift permutation is not supported by target\n");
return false;
}
shift2_mask = vect_gen_perm_mask_checked (vectype, indices);
/* Generating permutation constant to shift all elements.
For vector length 8 it is {3 4 5 6 7 8 9 10}. */
for (i = 0; i < nelt; i++)
sel[i] = (nelt / 3) + (nelt % 3) / 2 + i;
indices.new_vector (sel, 2, nelt);
if (!can_vec_perm_const_p (TYPE_MODE (vectype), indices))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"shift permutation is not supported by target\n");
return false;
}
shift3_mask = vect_gen_perm_mask_checked (vectype, indices);
/* Generating permutation constant to shift all elements.
For vector length 8 it is {5 6 7 8 9 10 11 12}. */
for (i = 0; i < nelt; i++)
sel[i] = 2 * (nelt / 3) + (nelt % 3) / 2 + i;
indices.new_vector (sel, 2, nelt);
if (!can_vec_perm_const_p (TYPE_MODE (vectype), indices))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
"shift permutation is not supported by target\n");
return false;
}
shift4_mask = vect_gen_perm_mask_checked (vectype, indices);
for (k = 0; k < 3; k++)
{
data_ref = make_temp_ssa_name (vectype, NULL, "vect_shuffle3");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR,
dr_chain[k], dr_chain[k],
perm3_mask);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
vect[k] = data_ref;
}
for (k = 0; k < 3; k++)
{
data_ref = make_temp_ssa_name (vectype, NULL, "vect_shift1");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR,
vect[k % 3], vect[(k + 1) % 3],
shift1_mask);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
vect_shift[k] = data_ref;
}
for (k = 0; k < 3; k++)
{
data_ref = make_temp_ssa_name (vectype, NULL, "vect_shift2");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR,
vect_shift[(4 - k) % 3],
vect_shift[(3 - k) % 3],
shift2_mask);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
vect[k] = data_ref;
}
(*result_chain)[3 - (nelt % 3)] = vect[2];
data_ref = make_temp_ssa_name (vectype, NULL, "vect_shift3");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR, vect[0],
vect[0], shift3_mask);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
(*result_chain)[nelt % 3] = data_ref;
data_ref = make_temp_ssa_name (vectype, NULL, "vect_shift4");
perm_stmt = gimple_build_assign (data_ref, VEC_PERM_EXPR, vect[1],
vect[1], shift4_mask);
vect_finish_stmt_generation (stmt, perm_stmt, gsi);
(*result_chain)[0] = data_ref;
return true;
}
return false;
}
/* Function vect_transform_grouped_load.
Given a chain of input interleaved data-refs (in DR_CHAIN), build statements
to perform their permutation and ascribe the result vectorized statements to
the scalar statements.
*/
void
vect_transform_grouped_load (gimple *stmt, vec<tree> dr_chain, int size,
gimple_stmt_iterator *gsi)
{
machine_mode mode;
vec<tree> result_chain = vNULL;
/* DR_CHAIN contains input data-refs that are a part of the interleaving.
RESULT_CHAIN is the output of vect_permute_load_chain, it contains permuted
vectors, that are ready for vector computation. */
result_chain.create (size);
/* If reassociation width for vector type is 2 or greater target machine can
execute 2 or more vector instructions in parallel. Otherwise try to
get chain for loads group using vect_shift_permute_load_chain. */
mode = TYPE_MODE (STMT_VINFO_VECTYPE (vinfo_for_stmt (stmt)));
if (targetm.sched.reassociation_width (VEC_PERM_EXPR, mode) > 1
|| pow2p_hwi (size)
|| !vect_shift_permute_load_chain (dr_chain, size, stmt,
gsi, &result_chain))
vect_permute_load_chain (dr_chain, size, stmt, gsi, &result_chain);
vect_record_grouped_load_vectors (stmt, result_chain);
result_chain.release ();
}
/* RESULT_CHAIN contains the output of a group of grouped loads that were
generated as part of the vectorization of STMT. Assign the statement
for each vector to the associated scalar statement. */
void
vect_record_grouped_load_vectors (gimple *stmt, vec<tree> result_chain)
{
gimple *first_stmt = DR_GROUP_FIRST_ELEMENT (vinfo_for_stmt (stmt));
gimple *next_stmt, *new_stmt;
unsigned int i, gap_count;
tree tmp_data_ref;
/* Put a permuted data-ref in the VECTORIZED_STMT field.
Since we scan the chain starting from it's first node, their order
corresponds the order of data-refs in RESULT_CHAIN. */
next_stmt = first_stmt;
gap_count = 1;
FOR_EACH_VEC_ELT (result_chain, i, tmp_data_ref)
{
if (!next_stmt)
break;
/* Skip the gaps. Loads created for the gaps will be removed by dead
code elimination pass later. No need to check for the first stmt in
the group, since it always exists.
DR_GROUP_GAP is the number of steps in elements from the previous
access (if there is no gap DR_GROUP_GAP is 1). We skip loads that
correspond to the gaps. */
if (next_stmt != first_stmt
&& gap_count < DR_GROUP_GAP (vinfo_for_stmt (next_stmt)))
{
gap_count++;
continue;
}
while (next_stmt)
{
new_stmt = SSA_NAME_DEF_STMT (tmp_data_ref);
/* We assume that if VEC_STMT is not NULL, this is a case of multiple
copies, and we put the new vector statement in the first available
RELATED_STMT. */
if (!STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt)))
STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt)) = new_stmt;
else
{
if (!DR_GROUP_SAME_DR_STMT (vinfo_for_stmt (next_stmt)))
{
gimple *prev_stmt =
STMT_VINFO_VEC_STMT (vinfo_for_stmt (next_stmt));
gimple *rel_stmt =
STMT_VINFO_RELATED_STMT (vinfo_for_stmt (prev_stmt));
while (rel_stmt)
{
prev_stmt = rel_stmt;
rel_stmt =
STMT_VINFO_RELATED_STMT (vinfo_for_stmt (rel_stmt));
}
STMT_VINFO_RELATED_STMT (vinfo_for_stmt (prev_stmt)) =
new_stmt;
}
}
next_stmt = DR_GROUP_NEXT_ELEMENT (vinfo_for_stmt (next_stmt));
gap_count = 1;
/* If NEXT_STMT accesses the same DR as the previous statement,
put the same TMP_DATA_REF as its vectorized statement; otherwise
get the next data-ref from RESULT_CHAIN. */
if (!next_stmt || !DR_GROUP_SAME_DR_STMT (vinfo_for_stmt (next_stmt)))
break;
}
}
}
/* Function vect_force_dr_alignment_p.
Returns whether the alignment of a DECL can be forced to be aligned
on ALIGNMENT bit boundary. */
bool
vect_can_force_dr_alignment_p (const_tree decl, unsigned int alignment)
{
if (!VAR_P (decl))
return false;
if (decl_in_symtab_p (decl)
&& !symtab_node::get (decl)->can_increase_alignment_p ())
return false;
if (TREE_STATIC (decl))
return (alignment <= MAX_OFILE_ALIGNMENT);
else
return (alignment <= MAX_STACK_ALIGNMENT);
}
/* Return whether the data reference DR is supported with respect to its
alignment.
If CHECK_ALIGNED_ACCESSES is TRUE, check if the access is supported even
it is aligned, i.e., check if it is possible to vectorize it with different
alignment. */
enum dr_alignment_support
vect_supportable_dr_alignment (struct data_reference *dr,
bool check_aligned_accesses)
{
gimple *stmt = vect_dr_stmt (dr);
stmt_vec_info stmt_info = vinfo_for_stmt (stmt);
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
machine_mode mode = TYPE_MODE (vectype);
loop_vec_info loop_vinfo = STMT_VINFO_LOOP_VINFO (stmt_info);
struct loop *vect_loop = NULL;
bool nested_in_vect_loop = false;
if (aligned_access_p (dr) && !check_aligned_accesses)
return dr_aligned;
/* For now assume all conditional loads/stores support unaligned
access without any special code. */
if (is_gimple_call (stmt)
&& gimple_call_internal_p (stmt)
&& (gimple_call_internal_fn (stmt) == IFN_MASK_LOAD
|| gimple_call_internal_fn (stmt) == IFN_MASK_STORE))
return dr_unaligned_supported;
if (loop_vinfo)
{
vect_loop = LOOP_VINFO_LOOP (loop_vinfo);
nested_in_vect_loop = nested_in_vect_loop_p (vect_loop, stmt);
}
/* Possibly unaligned access. */
/* We can choose between using the implicit realignment scheme (generating
a misaligned_move stmt) and the explicit realignment scheme (generating
aligned loads with a REALIGN_LOAD). There are two variants to the
explicit realignment scheme: optimized, and unoptimized.
We can optimize the realignment only if the step between consecutive
vector loads is equal to the vector size. Since the vector memory
accesses advance in steps of VS (Vector Size) in the vectorized loop, it
is guaranteed that the misalignment amount remains the same throughout the
execution of the vectorized loop. Therefore, we can create the
"realignment token" (the permutation mask that is passed to REALIGN_LOAD)
at the loop preheader.
However, in the case of outer-loop vectorization, when vectorizing a
memory access in the inner-loop nested within the LOOP that is now being
vectorized, while it is guaranteed that the misalignment of the
vectorized memory access will remain the same in different outer-loop
iterations, it is *not* guaranteed that is will remain the same throughout
the execution of the inner-loop. This is because the inner-loop advances
with the original scalar step (and not in steps of VS). If the inner-loop
step happens to be a multiple of VS, then the misalignment remains fixed
and we can use the optimized realignment scheme. For example:
for (i=0; i<N; i++)
for (j=0; j<M; j++)
s += a[i+j];
When vectorizing the i-loop in the above example, the step between
consecutive vector loads is 1, and so the misalignment does not remain
fixed across the execution of the inner-loop, and the realignment cannot
be optimized (as illustrated in the following pseudo vectorized loop):
for (i=0; i<N; i+=4)
for (j=0; j<M; j++){
vs += vp[i+j]; // misalignment of &vp[i+j] is {0,1,2,3,0,1,2,3,...}
// when j is {0,1,2,3,4,5,6,7,...} respectively.
// (assuming that we start from an aligned address).
}
We therefore have to use the unoptimized realignment scheme:
for (i=0; i<N; i+=4)
for (j=k; j<M; j+=4)
vs += vp[i+j]; // misalignment of &vp[i+j] is always k (assuming
// that the misalignment of the initial address is
// 0).
The loop can then be vectorized as follows:
for (k=0; k<4; k++){
rt = get_realignment_token (&vp[k]);
for (i=0; i<N; i+=4){
v1 = vp[i+k];
for (j=k; j<M; j+=4){
v2 = vp[i+j+VS-1];
va = REALIGN_LOAD <v1,v2,rt>;
vs += va;
v1 = v2;
}
}
} */
if (DR_IS_READ (dr))
{
bool is_packed = false;
tree type = (TREE_TYPE (DR_REF (dr)));
if (optab_handler (vec_realign_load_optab, mode) != CODE_FOR_nothing
&& (!targetm.vectorize.builtin_mask_for_load
|| targetm.vectorize.builtin_mask_for_load ()))
{
tree vectype = STMT_VINFO_VECTYPE (stmt_info);
/* If we are doing SLP then the accesses need not have the
same alignment, instead it depends on the SLP group size. */
if (loop_vinfo
&& STMT_SLP_TYPE (stmt_info)
&& !multiple_p (LOOP_VINFO_VECT_FACTOR (loop_vinfo)
* DR_GROUP_SIZE (vinfo_for_stmt
(DR_GROUP_FIRST_ELEMENT (stmt_info))),
TYPE_VECTOR_SUBPARTS (vectype)))
;
else if (!loop_vinfo
|| (nested_in_vect_loop
&& maybe_ne (TREE_INT_CST_LOW (DR_STEP (dr)),
GET_MODE_SIZE (TYPE_MODE (vectype)))))
return dr_explicit_realign;
else
return dr_explicit_realign_optimized;
}
if (!known_alignment_for_access_p (dr))
is_packed = not_size_aligned (DR_REF (dr));
if (targetm.vectorize.support_vector_misalignment
(mode, type, DR_MISALIGNMENT (dr), is_packed))
/* Can't software pipeline the loads, but can at least do them. */
return dr_unaligned_supported;
}
else
{
bool is_packed = false;
tree type = (TREE_TYPE (DR_REF (dr)));
if (!known_alignment_for_access_p (dr))
is_packed = not_size_aligned (DR_REF (dr));
if (targetm.vectorize.support_vector_misalignment
(mode, type, DR_MISALIGNMENT (dr), is_packed))
return dr_unaligned_supported;
}
/* Unsupported. */
return dr_unaligned_unsupported;
}
|