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
|
//===- InstructionCombining.cpp - Combine multiple instructions -----------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// InstructionCombining - Combine instructions to form fewer, simple
// instructions. This pass does not modify the CFG. This pass is where
// algebraic simplification happens.
//
// This pass combines things like:
// %Y = add i32 %X, 1
// %Z = add i32 %Y, 1
// into:
// %Z = add i32 %X, 2
//
// This is a simple worklist driven algorithm.
//
// This pass guarantees that the following canonicalizations are performed on
// the program:
// 1. If a binary operator has a constant operand, it is moved to the RHS
// 2. Bitwise operators with constant operands are always grouped so that
// shifts are performed first, then or's, then and's, then xor's.
// 3. Compare instructions are converted from <,>,<=,>= to ==,!= if possible
// 4. All cmp instructions on boolean values are replaced with logical ops
// 5. add X, X is represented as (X*2) => (X << 1)
// 6. Multiplies with a power-of-two constant argument are transformed into
// shifts.
// ... etc.
//
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "instcombine"
#include "llvm/Transforms/Scalar.h"
#include "InstCombine.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/LLVMContext.h"
#include "llvm/DerivedTypes.h"
#include "llvm/GlobalVariable.h"
#include "llvm/Operator.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/MemoryBuiltins.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/PatternMatch.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/STLExtras.h"
#include <algorithm>
#include <climits>
using namespace llvm;
using namespace llvm::PatternMatch;
STATISTIC(NumCombined , "Number of insts combined");
STATISTIC(NumConstProp, "Number of constant folds");
STATISTIC(NumDeadInst , "Number of dead inst eliminated");
STATISTIC(NumSunkInst , "Number of instructions sunk");
char InstCombiner::ID = 0;
static RegisterPass<InstCombiner>
X("instcombine", "Combine redundant instructions");
void InstCombiner::getAnalysisUsage(AnalysisUsage &AU) const {
AU.addPreservedID(LCSSAID);
AU.setPreservesCFG();
}
/// ShouldChangeType - Return true if it is desirable to convert a computation
/// from 'From' to 'To'. We don't want to convert from a legal to an illegal
/// type for example, or from a smaller to a larger illegal type.
bool InstCombiner::ShouldChangeType(const Type *From, const Type *To) const {
assert(isa<IntegerType>(From) && isa<IntegerType>(To));
// If we don't have TD, we don't know if the source/dest are legal.
if (!TD) return false;
unsigned FromWidth = From->getPrimitiveSizeInBits();
unsigned ToWidth = To->getPrimitiveSizeInBits();
bool FromLegal = TD->isLegalInteger(FromWidth);
bool ToLegal = TD->isLegalInteger(ToWidth);
// If this is a legal integer from type, and the result would be an illegal
// type, don't do the transformation.
if (FromLegal && !ToLegal)
return false;
// Otherwise, if both are illegal, do not increase the size of the result. We
// do allow things like i160 -> i64, but not i64 -> i160.
if (!FromLegal && !ToLegal && ToWidth > FromWidth)
return false;
return true;
}
/// getBitCastOperand - If the specified operand is a CastInst, a constant
/// expression bitcast, or a GetElementPtrInst with all zero indices, return the
/// operand value, otherwise return null.
// FIXME: Value::stripPointerCasts
static Value *getBitCastOperand(Value *V) {
if (Operator *O = dyn_cast<Operator>(V)) {
if (O->getOpcode() == Instruction::BitCast)
return O->getOperand(0);
if (GEPOperator *GEP = dyn_cast<GEPOperator>(V))
if (GEP->hasAllZeroIndices())
return GEP->getPointerOperand();
}
return 0;
}
// SimplifyCommutative - This performs a few simplifications for commutative
// operators:
//
// 1. Order operands such that they are listed from right (least complex) to
// left (most complex). This puts constants before unary operators before
// binary operators.
//
// 2. Transform: (op (op V, C1), C2) ==> (op V, (op C1, C2))
// 3. Transform: (op (op V1, C1), (op V2, C2)) ==> (op (op V1, V2), (op C1,C2))
//
bool InstCombiner::SimplifyCommutative(BinaryOperator &I) {
bool Changed = false;
if (getComplexity(I.getOperand(0)) < getComplexity(I.getOperand(1)))
Changed = !I.swapOperands();
if (!I.isAssociative()) return Changed;
Instruction::BinaryOps Opcode = I.getOpcode();
if (BinaryOperator *Op = dyn_cast<BinaryOperator>(I.getOperand(0)))
if (Op->getOpcode() == Opcode && isa<Constant>(Op->getOperand(1))) {
if (isa<Constant>(I.getOperand(1))) {
Constant *Folded = ConstantExpr::get(I.getOpcode(),
cast<Constant>(I.getOperand(1)),
cast<Constant>(Op->getOperand(1)));
I.setOperand(0, Op->getOperand(0));
I.setOperand(1, Folded);
return true;
}
if (BinaryOperator *Op1 = dyn_cast<BinaryOperator>(I.getOperand(1)))
if (Op1->getOpcode() == Opcode && isa<Constant>(Op1->getOperand(1)) &&
Op->hasOneUse() && Op1->hasOneUse()) {
Constant *C1 = cast<Constant>(Op->getOperand(1));
Constant *C2 = cast<Constant>(Op1->getOperand(1));
// Fold (op (op V1, C1), (op V2, C2)) ==> (op (op V1, V2), (op C1,C2))
Constant *Folded = ConstantExpr::get(I.getOpcode(), C1, C2);
Instruction *New = BinaryOperator::Create(Opcode, Op->getOperand(0),
Op1->getOperand(0),
Op1->getName(), &I);
Worklist.Add(New);
I.setOperand(0, New);
I.setOperand(1, Folded);
return true;
}
}
return Changed;
}
// dyn_castNegVal - Given a 'sub' instruction, return the RHS of the instruction
// if the LHS is a constant zero (which is the 'negate' form).
//
Value *InstCombiner::dyn_castNegVal(Value *V) const {
if (BinaryOperator::isNeg(V))
return BinaryOperator::getNegArgument(V);
// Constants can be considered to be negated values if they can be folded.
if (ConstantInt *C = dyn_cast<ConstantInt>(V))
return ConstantExpr::getNeg(C);
if (ConstantVector *C = dyn_cast<ConstantVector>(V))
if (C->getType()->getElementType()->isInteger())
return ConstantExpr::getNeg(C);
return 0;
}
// dyn_castFNegVal - Given a 'fsub' instruction, return the RHS of the
// instruction if the LHS is a constant negative zero (which is the 'negate'
// form).
//
Value *InstCombiner::dyn_castFNegVal(Value *V) const {
if (BinaryOperator::isFNeg(V))
return BinaryOperator::getFNegArgument(V);
// Constants can be considered to be negated values if they can be folded.
if (ConstantFP *C = dyn_cast<ConstantFP>(V))
return ConstantExpr::getFNeg(C);
if (ConstantVector *C = dyn_cast<ConstantVector>(V))
if (C->getType()->getElementType()->isFloatingPoint())
return ConstantExpr::getFNeg(C);
return 0;
}
/// isFreeToInvert - Return true if the specified value is free to invert (apply
/// ~ to). This happens in cases where the ~ can be eliminated.
static inline bool isFreeToInvert(Value *V) {
// ~(~(X)) -> X.
if (BinaryOperator::isNot(V))
return true;
// Constants can be considered to be not'ed values.
if (isa<ConstantInt>(V))
return true;
// Compares can be inverted if they have a single use.
if (CmpInst *CI = dyn_cast<CmpInst>(V))
return CI->hasOneUse();
return false;
}
static inline Value *dyn_castNotVal(Value *V) {
// If this is not(not(x)) don't return that this is a not: we want the two
// not's to be folded first.
if (BinaryOperator::isNot(V)) {
Value *Operand = BinaryOperator::getNotArgument(V);
if (!isFreeToInvert(Operand))
return Operand;
}
// Constants can be considered to be not'ed values...
if (ConstantInt *C = dyn_cast<ConstantInt>(V))
return ConstantInt::get(C->getType(), ~C->getValue());
return 0;
}
/// AddOne - Add one to a ConstantInt.
static Constant *AddOne(Constant *C) {
return ConstantExpr::getAdd(C, ConstantInt::get(C->getType(), 1));
}
/// SubOne - Subtract one from a ConstantInt.
static Constant *SubOne(ConstantInt *C) {
return ConstantInt::get(C->getContext(), C->getValue()-1);
}
static Value *FoldOperationIntoSelectOperand(Instruction &I, Value *SO,
InstCombiner *IC) {
if (CastInst *CI = dyn_cast<CastInst>(&I))
return IC->Builder->CreateCast(CI->getOpcode(), SO, I.getType());
// Figure out if the constant is the left or the right argument.
bool ConstIsRHS = isa<Constant>(I.getOperand(1));
Constant *ConstOperand = cast<Constant>(I.getOperand(ConstIsRHS));
if (Constant *SOC = dyn_cast<Constant>(SO)) {
if (ConstIsRHS)
return ConstantExpr::get(I.getOpcode(), SOC, ConstOperand);
return ConstantExpr::get(I.getOpcode(), ConstOperand, SOC);
}
Value *Op0 = SO, *Op1 = ConstOperand;
if (!ConstIsRHS)
std::swap(Op0, Op1);
if (BinaryOperator *BO = dyn_cast<BinaryOperator>(&I))
return IC->Builder->CreateBinOp(BO->getOpcode(), Op0, Op1,
SO->getName()+".op");
if (ICmpInst *CI = dyn_cast<ICmpInst>(&I))
return IC->Builder->CreateICmp(CI->getPredicate(), Op0, Op1,
SO->getName()+".cmp");
if (FCmpInst *CI = dyn_cast<FCmpInst>(&I))
return IC->Builder->CreateICmp(CI->getPredicate(), Op0, Op1,
SO->getName()+".cmp");
llvm_unreachable("Unknown binary instruction type!");
}
// FoldOpIntoSelect - Given an instruction with a select as one operand and a
// constant as the other operand, try to fold the binary operator into the
// select arguments. This also works for Cast instructions, which obviously do
// not have a second operand.
Instruction *InstCombiner::FoldOpIntoSelect(Instruction &Op, SelectInst *SI) {
// Don't modify shared select instructions
if (!SI->hasOneUse()) return 0;
Value *TV = SI->getOperand(1);
Value *FV = SI->getOperand(2);
if (isa<Constant>(TV) || isa<Constant>(FV)) {
// Bool selects with constant operands can be folded to logical ops.
if (SI->getType() == Type::getInt1Ty(SI->getContext())) return 0;
Value *SelectTrueVal = FoldOperationIntoSelectOperand(Op, TV, this);
Value *SelectFalseVal = FoldOperationIntoSelectOperand(Op, FV, this);
return SelectInst::Create(SI->getCondition(), SelectTrueVal,
SelectFalseVal);
}
return 0;
}
/// FoldOpIntoPhi - Given a binary operator, cast instruction, or select which
/// has a PHI node as operand #0, see if we can fold the instruction into the
/// PHI (which is only possible if all operands to the PHI are constants).
///
/// If AllowAggressive is true, FoldOpIntoPhi will allow certain transforms
/// that would normally be unprofitable because they strongly encourage jump
/// threading.
Instruction *InstCombiner::FoldOpIntoPhi(Instruction &I,
bool AllowAggressive) {
AllowAggressive = false;
PHINode *PN = cast<PHINode>(I.getOperand(0));
unsigned NumPHIValues = PN->getNumIncomingValues();
if (NumPHIValues == 0 ||
// We normally only transform phis with a single use, unless we're trying
// hard to make jump threading happen.
(!PN->hasOneUse() && !AllowAggressive))
return 0;
// Check to see if all of the operands of the PHI are simple constants
// (constantint/constantfp/undef). If there is one non-constant value,
// remember the BB it is in. If there is more than one or if *it* is a PHI,
// bail out. We don't do arbitrary constant expressions here because moving
// their computation can be expensive without a cost model.
BasicBlock *NonConstBB = 0;
for (unsigned i = 0; i != NumPHIValues; ++i)
if (!isa<Constant>(PN->getIncomingValue(i)) ||
isa<ConstantExpr>(PN->getIncomingValue(i))) {
if (NonConstBB) return 0; // More than one non-const value.
if (isa<PHINode>(PN->getIncomingValue(i))) return 0; // Itself a phi.
NonConstBB = PN->getIncomingBlock(i);
// If the incoming non-constant value is in I's block, we have an infinite
// loop.
if (NonConstBB == I.getParent())
return 0;
}
// If there is exactly one non-constant value, we can insert a copy of the
// operation in that block. However, if this is a critical edge, we would be
// inserting the computation one some other paths (e.g. inside a loop). Only
// do this if the pred block is unconditionally branching into the phi block.
if (NonConstBB != 0 && !AllowAggressive) {
BranchInst *BI = dyn_cast<BranchInst>(NonConstBB->getTerminator());
if (!BI || !BI->isUnconditional()) return 0;
}
// Okay, we can do the transformation: create the new PHI node.
PHINode *NewPN = PHINode::Create(I.getType(), "");
NewPN->reserveOperandSpace(PN->getNumOperands()/2);
InsertNewInstBefore(NewPN, *PN);
NewPN->takeName(PN);
// Next, add all of the operands to the PHI.
if (SelectInst *SI = dyn_cast<SelectInst>(&I)) {
// We only currently try to fold the condition of a select when it is a phi,
// not the true/false values.
Value *TrueV = SI->getTrueValue();
Value *FalseV = SI->getFalseValue();
BasicBlock *PhiTransBB = PN->getParent();
for (unsigned i = 0; i != NumPHIValues; ++i) {
BasicBlock *ThisBB = PN->getIncomingBlock(i);
Value *TrueVInPred = TrueV->DoPHITranslation(PhiTransBB, ThisBB);
Value *FalseVInPred = FalseV->DoPHITranslation(PhiTransBB, ThisBB);
Value *InV = 0;
if (Constant *InC = dyn_cast<Constant>(PN->getIncomingValue(i))) {
InV = InC->isNullValue() ? FalseVInPred : TrueVInPred;
} else {
assert(PN->getIncomingBlock(i) == NonConstBB);
InV = SelectInst::Create(PN->getIncomingValue(i), TrueVInPred,
FalseVInPred,
"phitmp", NonConstBB->getTerminator());
Worklist.Add(cast<Instruction>(InV));
}
NewPN->addIncoming(InV, ThisBB);
}
} else if (I.getNumOperands() == 2) {
Constant *C = cast<Constant>(I.getOperand(1));
for (unsigned i = 0; i != NumPHIValues; ++i) {
Value *InV = 0;
if (Constant *InC = dyn_cast<Constant>(PN->getIncomingValue(i))) {
if (CmpInst *CI = dyn_cast<CmpInst>(&I))
InV = ConstantExpr::getCompare(CI->getPredicate(), InC, C);
else
InV = ConstantExpr::get(I.getOpcode(), InC, C);
} else {
assert(PN->getIncomingBlock(i) == NonConstBB);
if (BinaryOperator *BO = dyn_cast<BinaryOperator>(&I))
InV = BinaryOperator::Create(BO->getOpcode(),
PN->getIncomingValue(i), C, "phitmp",
NonConstBB->getTerminator());
else if (CmpInst *CI = dyn_cast<CmpInst>(&I))
InV = CmpInst::Create(CI->getOpcode(),
CI->getPredicate(),
PN->getIncomingValue(i), C, "phitmp",
NonConstBB->getTerminator());
else
llvm_unreachable("Unknown binop!");
Worklist.Add(cast<Instruction>(InV));
}
NewPN->addIncoming(InV, PN->getIncomingBlock(i));
}
} else {
CastInst *CI = cast<CastInst>(&I);
const Type *RetTy = CI->getType();
for (unsigned i = 0; i != NumPHIValues; ++i) {
Value *InV;
if (Constant *InC = dyn_cast<Constant>(PN->getIncomingValue(i))) {
InV = ConstantExpr::getCast(CI->getOpcode(), InC, RetTy);
} else {
assert(PN->getIncomingBlock(i) == NonConstBB);
InV = CastInst::Create(CI->getOpcode(), PN->getIncomingValue(i),
I.getType(), "phitmp",
NonConstBB->getTerminator());
Worklist.Add(cast<Instruction>(InV));
}
NewPN->addIncoming(InV, PN->getIncomingBlock(i));
}
}
return ReplaceInstUsesWith(I, NewPN);
}
/// getICmpCode - Encode a icmp predicate into a three bit mask. These bits
/// are carefully arranged to allow folding of expressions such as:
///
/// (A < B) | (A > B) --> (A != B)
///
/// Note that this is only valid if the first and second predicates have the
/// same sign. Is illegal to do: (A u< B) | (A s> B)
///
/// Three bits are used to represent the condition, as follows:
/// 0 A > B
/// 1 A == B
/// 2 A < B
///
/// <=> Value Definition
/// 000 0 Always false
/// 001 1 A > B
/// 010 2 A == B
/// 011 3 A >= B
/// 100 4 A < B
/// 101 5 A != B
/// 110 6 A <= B
/// 111 7 Always true
///
static unsigned getICmpCode(const ICmpInst *ICI) {
switch (ICI->getPredicate()) {
// False -> 0
case ICmpInst::ICMP_UGT: return 1; // 001
case ICmpInst::ICMP_SGT: return 1; // 001
case ICmpInst::ICMP_EQ: return 2; // 010
case ICmpInst::ICMP_UGE: return 3; // 011
case ICmpInst::ICMP_SGE: return 3; // 011
case ICmpInst::ICMP_ULT: return 4; // 100
case ICmpInst::ICMP_SLT: return 4; // 100
case ICmpInst::ICMP_NE: return 5; // 101
case ICmpInst::ICMP_ULE: return 6; // 110
case ICmpInst::ICMP_SLE: return 6; // 110
// True -> 7
default:
llvm_unreachable("Invalid ICmp predicate!");
return 0;
}
}
/// getFCmpCode - Similar to getICmpCode but for FCmpInst. This encodes a fcmp
/// predicate into a three bit mask. It also returns whether it is an ordered
/// predicate by reference.
static unsigned getFCmpCode(FCmpInst::Predicate CC, bool &isOrdered) {
isOrdered = false;
switch (CC) {
case FCmpInst::FCMP_ORD: isOrdered = true; return 0; // 000
case FCmpInst::FCMP_UNO: return 0; // 000
case FCmpInst::FCMP_OGT: isOrdered = true; return 1; // 001
case FCmpInst::FCMP_UGT: return 1; // 001
case FCmpInst::FCMP_OEQ: isOrdered = true; return 2; // 010
case FCmpInst::FCMP_UEQ: return 2; // 010
case FCmpInst::FCMP_OGE: isOrdered = true; return 3; // 011
case FCmpInst::FCMP_UGE: return 3; // 011
case FCmpInst::FCMP_OLT: isOrdered = true; return 4; // 100
case FCmpInst::FCMP_ULT: return 4; // 100
case FCmpInst::FCMP_ONE: isOrdered = true; return 5; // 101
case FCmpInst::FCMP_UNE: return 5; // 101
case FCmpInst::FCMP_OLE: isOrdered = true; return 6; // 110
case FCmpInst::FCMP_ULE: return 6; // 110
// True -> 7
default:
// Not expecting FCMP_FALSE and FCMP_TRUE;
llvm_unreachable("Unexpected FCmp predicate!");
return 0;
}
}
/// getICmpValue - This is the complement of getICmpCode, which turns an
/// opcode and two operands into either a constant true or false, or a brand
/// new ICmp instruction. The sign is passed in to determine which kind
/// of predicate to use in the new icmp instruction.
static Value *getICmpValue(bool Sign, unsigned Code, Value *LHS, Value *RHS) {
switch (Code) {
default: assert(0 && "Illegal ICmp code!");
case 0:
return ConstantInt::getFalse(LHS->getContext());
case 1:
if (Sign)
return new ICmpInst(ICmpInst::ICMP_SGT, LHS, RHS);
return new ICmpInst(ICmpInst::ICMP_UGT, LHS, RHS);
case 2:
return new ICmpInst(ICmpInst::ICMP_EQ, LHS, RHS);
case 3:
if (Sign)
return new ICmpInst(ICmpInst::ICMP_SGE, LHS, RHS);
return new ICmpInst(ICmpInst::ICMP_UGE, LHS, RHS);
case 4:
if (Sign)
return new ICmpInst(ICmpInst::ICMP_SLT, LHS, RHS);
return new ICmpInst(ICmpInst::ICMP_ULT, LHS, RHS);
case 5:
return new ICmpInst(ICmpInst::ICMP_NE, LHS, RHS);
case 6:
if (Sign)
return new ICmpInst(ICmpInst::ICMP_SLE, LHS, RHS);
return new ICmpInst(ICmpInst::ICMP_ULE, LHS, RHS);
case 7:
return ConstantInt::getTrue(LHS->getContext());
}
}
/// getFCmpValue - This is the complement of getFCmpCode, which turns an
/// opcode and two operands into either a FCmp instruction. isordered is passed
/// in to determine which kind of predicate to use in the new fcmp instruction.
static Value *getFCmpValue(bool isordered, unsigned code,
Value *LHS, Value *RHS) {
switch (code) {
default: llvm_unreachable("Illegal FCmp code!");
case 0:
if (isordered)
return new FCmpInst(FCmpInst::FCMP_ORD, LHS, RHS);
else
return new FCmpInst(FCmpInst::FCMP_UNO, LHS, RHS);
case 1:
if (isordered)
return new FCmpInst(FCmpInst::FCMP_OGT, LHS, RHS);
else
return new FCmpInst(FCmpInst::FCMP_UGT, LHS, RHS);
case 2:
if (isordered)
return new FCmpInst(FCmpInst::FCMP_OEQ, LHS, RHS);
else
return new FCmpInst(FCmpInst::FCMP_UEQ, LHS, RHS);
case 3:
if (isordered)
return new FCmpInst(FCmpInst::FCMP_OGE, LHS, RHS);
else
return new FCmpInst(FCmpInst::FCMP_UGE, LHS, RHS);
case 4:
if (isordered)
return new FCmpInst(FCmpInst::FCMP_OLT, LHS, RHS);
else
return new FCmpInst(FCmpInst::FCMP_ULT, LHS, RHS);
case 5:
if (isordered)
return new FCmpInst(FCmpInst::FCMP_ONE, LHS, RHS);
else
return new FCmpInst(FCmpInst::FCMP_UNE, LHS, RHS);
case 6:
if (isordered)
return new FCmpInst(FCmpInst::FCMP_OLE, LHS, RHS);
else
return new FCmpInst(FCmpInst::FCMP_ULE, LHS, RHS);
case 7: return ConstantInt::getTrue(LHS->getContext());
}
}
/// PredicatesFoldable - Return true if both predicates match sign or if at
/// least one of them is an equality comparison (which is signless).
static bool PredicatesFoldable(ICmpInst::Predicate p1, ICmpInst::Predicate p2) {
return (CmpInst::isSigned(p1) == CmpInst::isSigned(p2)) ||
(CmpInst::isSigned(p1) && ICmpInst::isEquality(p2)) ||
(CmpInst::isSigned(p2) && ICmpInst::isEquality(p1));
}
// OptAndOp - This handles expressions of the form ((val OP C1) & C2). Where
// the Op parameter is 'OP', OpRHS is 'C1', and AndRHS is 'C2'. Op is
// guaranteed to be a binary operator.
Instruction *InstCombiner::OptAndOp(Instruction *Op,
ConstantInt *OpRHS,
ConstantInt *AndRHS,
BinaryOperator &TheAnd) {
Value *X = Op->getOperand(0);
Constant *Together = 0;
if (!Op->isShift())
Together = ConstantExpr::getAnd(AndRHS, OpRHS);
switch (Op->getOpcode()) {
case Instruction::Xor:
if (Op->hasOneUse()) {
// (X ^ C1) & C2 --> (X & C2) ^ (C1&C2)
Value *And = Builder->CreateAnd(X, AndRHS);
And->takeName(Op);
return BinaryOperator::CreateXor(And, Together);
}
break;
case Instruction::Or:
if (Together == AndRHS) // (X | C) & C --> C
return ReplaceInstUsesWith(TheAnd, AndRHS);
if (Op->hasOneUse() && Together != OpRHS) {
// (X | C1) & C2 --> (X | (C1&C2)) & C2
Value *Or = Builder->CreateOr(X, Together);
Or->takeName(Op);
return BinaryOperator::CreateAnd(Or, AndRHS);
}
break;
case Instruction::Add:
if (Op->hasOneUse()) {
// Adding a one to a single bit bit-field should be turned into an XOR
// of the bit. First thing to check is to see if this AND is with a
// single bit constant.
const APInt &AndRHSV = cast<ConstantInt>(AndRHS)->getValue();
// If there is only one bit set.
if (AndRHSV.isPowerOf2()) {
// Ok, at this point, we know that we are masking the result of the
// ADD down to exactly one bit. If the constant we are adding has
// no bits set below this bit, then we can eliminate the ADD.
const APInt& AddRHS = cast<ConstantInt>(OpRHS)->getValue();
// Check to see if any bits below the one bit set in AndRHSV are set.
if ((AddRHS & (AndRHSV-1)) == 0) {
// If not, the only thing that can effect the output of the AND is
// the bit specified by AndRHSV. If that bit is set, the effect of
// the XOR is to toggle the bit. If it is clear, then the ADD has
// no effect.
if ((AddRHS & AndRHSV) == 0) { // Bit is not set, noop
TheAnd.setOperand(0, X);
return &TheAnd;
} else {
// Pull the XOR out of the AND.
Value *NewAnd = Builder->CreateAnd(X, AndRHS);
NewAnd->takeName(Op);
return BinaryOperator::CreateXor(NewAnd, AndRHS);
}
}
}
}
break;
case Instruction::Shl: {
// We know that the AND will not produce any of the bits shifted in, so if
// the anded constant includes them, clear them now!
//
uint32_t BitWidth = AndRHS->getType()->getBitWidth();
uint32_t OpRHSVal = OpRHS->getLimitedValue(BitWidth);
APInt ShlMask(APInt::getHighBitsSet(BitWidth, BitWidth-OpRHSVal));
ConstantInt *CI = ConstantInt::get(AndRHS->getContext(),
AndRHS->getValue() & ShlMask);
if (CI->getValue() == ShlMask) {
// Masking out bits that the shift already masks
return ReplaceInstUsesWith(TheAnd, Op); // No need for the and.
} else if (CI != AndRHS) { // Reducing bits set in and.
TheAnd.setOperand(1, CI);
return &TheAnd;
}
break;
}
case Instruction::LShr: {
// We know that the AND will not produce any of the bits shifted in, so if
// the anded constant includes them, clear them now! This only applies to
// unsigned shifts, because a signed shr may bring in set bits!
//
uint32_t BitWidth = AndRHS->getType()->getBitWidth();
uint32_t OpRHSVal = OpRHS->getLimitedValue(BitWidth);
APInt ShrMask(APInt::getLowBitsSet(BitWidth, BitWidth - OpRHSVal));
ConstantInt *CI = ConstantInt::get(Op->getContext(),
AndRHS->getValue() & ShrMask);
if (CI->getValue() == ShrMask) {
// Masking out bits that the shift already masks.
return ReplaceInstUsesWith(TheAnd, Op);
} else if (CI != AndRHS) {
TheAnd.setOperand(1, CI); // Reduce bits set in and cst.
return &TheAnd;
}
break;
}
case Instruction::AShr:
// Signed shr.
// See if this is shifting in some sign extension, then masking it out
// with an and.
if (Op->hasOneUse()) {
uint32_t BitWidth = AndRHS->getType()->getBitWidth();
uint32_t OpRHSVal = OpRHS->getLimitedValue(BitWidth);
APInt ShrMask(APInt::getLowBitsSet(BitWidth, BitWidth - OpRHSVal));
Constant *C = ConstantInt::get(Op->getContext(),
AndRHS->getValue() & ShrMask);
if (C == AndRHS) { // Masking out bits shifted in.
// (Val ashr C1) & C2 -> (Val lshr C1) & C2
// Make the argument unsigned.
Value *ShVal = Op->getOperand(0);
ShVal = Builder->CreateLShr(ShVal, OpRHS, Op->getName());
return BinaryOperator::CreateAnd(ShVal, AndRHS, TheAnd.getName());
}
}
break;
}
return 0;
}
/// InsertRangeTest - Emit a computation of: (V >= Lo && V < Hi) if Inside is
/// true, otherwise (V < Lo || V >= Hi). In pratice, we emit the more efficient
/// (V-Lo) <u Hi-Lo. This method expects that Lo <= Hi. isSigned indicates
/// whether to treat the V, Lo and HI as signed or not. IB is the location to
/// insert new instructions.
Instruction *InstCombiner::InsertRangeTest(Value *V, Constant *Lo, Constant *Hi,
bool isSigned, bool Inside,
Instruction &IB) {
assert(cast<ConstantInt>(ConstantExpr::getICmp((isSigned ?
ICmpInst::ICMP_SLE:ICmpInst::ICMP_ULE), Lo, Hi))->getZExtValue() &&
"Lo is not <= Hi in range emission code!");
if (Inside) {
if (Lo == Hi) // Trivially false.
return new ICmpInst(ICmpInst::ICMP_NE, V, V);
// V >= Min && V < Hi --> V < Hi
if (cast<ConstantInt>(Lo)->isMinValue(isSigned)) {
ICmpInst::Predicate pred = (isSigned ?
ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT);
return new ICmpInst(pred, V, Hi);
}
// Emit V-Lo <u Hi-Lo
Constant *NegLo = ConstantExpr::getNeg(Lo);
Value *Add = Builder->CreateAdd(V, NegLo, V->getName()+".off");
Constant *UpperBound = ConstantExpr::getAdd(NegLo, Hi);
return new ICmpInst(ICmpInst::ICMP_ULT, Add, UpperBound);
}
if (Lo == Hi) // Trivially true.
return new ICmpInst(ICmpInst::ICMP_EQ, V, V);
// V < Min || V >= Hi -> V > Hi-1
Hi = SubOne(cast<ConstantInt>(Hi));
if (cast<ConstantInt>(Lo)->isMinValue(isSigned)) {
ICmpInst::Predicate pred = (isSigned ?
ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT);
return new ICmpInst(pred, V, Hi);
}
// Emit V-Lo >u Hi-1-Lo
// Note that Hi has already had one subtracted from it, above.
ConstantInt *NegLo = cast<ConstantInt>(ConstantExpr::getNeg(Lo));
Value *Add = Builder->CreateAdd(V, NegLo, V->getName()+".off");
Constant *LowerBound = ConstantExpr::getAdd(NegLo, Hi);
return new ICmpInst(ICmpInst::ICMP_UGT, Add, LowerBound);
}
// isRunOfOnes - Returns true iff Val consists of one contiguous run of 1s with
// any number of 0s on either side. The 1s are allowed to wrap from LSB to
// MSB, so 0x000FFF0, 0x0000FFFF, and 0xFF0000FF are all runs. 0x0F0F0000 is
// not, since all 1s are not contiguous.
static bool isRunOfOnes(ConstantInt *Val, uint32_t &MB, uint32_t &ME) {
const APInt& V = Val->getValue();
uint32_t BitWidth = Val->getType()->getBitWidth();
if (!APIntOps::isShiftedMask(BitWidth, V)) return false;
// look for the first zero bit after the run of ones
MB = BitWidth - ((V - 1) ^ V).countLeadingZeros();
// look for the first non-zero bit
ME = V.getActiveBits();
return true;
}
/// FoldLogicalPlusAnd - This is part of an expression (LHS +/- RHS) & Mask,
/// where isSub determines whether the operator is a sub. If we can fold one of
/// the following xforms:
///
/// ((A & N) +/- B) & Mask -> (A +/- B) & Mask iff N&Mask == Mask
/// ((A | N) +/- B) & Mask -> (A +/- B) & Mask iff N&Mask == 0
/// ((A ^ N) +/- B) & Mask -> (A +/- B) & Mask iff N&Mask == 0
///
/// return (A +/- B).
///
Value *InstCombiner::FoldLogicalPlusAnd(Value *LHS, Value *RHS,
ConstantInt *Mask, bool isSub,
Instruction &I) {
Instruction *LHSI = dyn_cast<Instruction>(LHS);
if (!LHSI || LHSI->getNumOperands() != 2 ||
!isa<ConstantInt>(LHSI->getOperand(1))) return 0;
ConstantInt *N = cast<ConstantInt>(LHSI->getOperand(1));
switch (LHSI->getOpcode()) {
default: return 0;
case Instruction::And:
if (ConstantExpr::getAnd(N, Mask) == Mask) {
// If the AndRHS is a power of two minus one (0+1+), this is simple.
if ((Mask->getValue().countLeadingZeros() +
Mask->getValue().countPopulation()) ==
Mask->getValue().getBitWidth())
break;
// Otherwise, if Mask is 0+1+0+, and if B is known to have the low 0+
// part, we don't need any explicit masks to take them out of A. If that
// is all N is, ignore it.
uint32_t MB = 0, ME = 0;
if (isRunOfOnes(Mask, MB, ME)) { // begin/end bit of run, inclusive
uint32_t BitWidth = cast<IntegerType>(RHS->getType())->getBitWidth();
APInt Mask(APInt::getLowBitsSet(BitWidth, MB-1));
if (MaskedValueIsZero(RHS, Mask))
break;
}
}
return 0;
case Instruction::Or:
case Instruction::Xor:
// If the AndRHS is a power of two minus one (0+1+), and N&Mask == 0
if ((Mask->getValue().countLeadingZeros() +
Mask->getValue().countPopulation()) == Mask->getValue().getBitWidth()
&& ConstantExpr::getAnd(N, Mask)->isNullValue())
break;
return 0;
}
if (isSub)
return Builder->CreateSub(LHSI->getOperand(0), RHS, "fold");
return Builder->CreateAdd(LHSI->getOperand(0), RHS, "fold");
}
/// FoldAndOfICmps - Fold (icmp)&(icmp) if possible.
Instruction *InstCombiner::FoldAndOfICmps(Instruction &I,
ICmpInst *LHS, ICmpInst *RHS) {
ICmpInst::Predicate LHSCC = LHS->getPredicate(), RHSCC = RHS->getPredicate();
// (icmp1 A, B) & (icmp2 A, B) --> (icmp3 A, B)
if (PredicatesFoldable(LHSCC, RHSCC)) {
if (LHS->getOperand(0) == RHS->getOperand(1) &&
LHS->getOperand(1) == RHS->getOperand(0))
LHS->swapOperands();
if (LHS->getOperand(0) == RHS->getOperand(0) &&
LHS->getOperand(1) == RHS->getOperand(1)) {
Value *Op0 = LHS->getOperand(0), *Op1 = LHS->getOperand(1);
unsigned Code = getICmpCode(LHS) & getICmpCode(RHS);
bool isSigned = LHS->isSigned() || RHS->isSigned();
Value *RV = getICmpValue(isSigned, Code, Op0, Op1);
if (Instruction *I = dyn_cast<Instruction>(RV))
return I;
// Otherwise, it's a constant boolean value.
return ReplaceInstUsesWith(I, RV);
}
}
// This only handles icmp of constants: (icmp1 A, C1) & (icmp2 B, C2).
Value *Val = LHS->getOperand(0), *Val2 = RHS->getOperand(0);
ConstantInt *LHSCst = dyn_cast<ConstantInt>(LHS->getOperand(1));
ConstantInt *RHSCst = dyn_cast<ConstantInt>(RHS->getOperand(1));
if (LHSCst == 0 || RHSCst == 0) return 0;
if (LHSCst == RHSCst && LHSCC == RHSCC) {
// (icmp ult A, C) & (icmp ult B, C) --> (icmp ult (A|B), C)
// where C is a power of 2
if (LHSCC == ICmpInst::ICMP_ULT &&
LHSCst->getValue().isPowerOf2()) {
Value *NewOr = Builder->CreateOr(Val, Val2);
return new ICmpInst(LHSCC, NewOr, LHSCst);
}
// (icmp eq A, 0) & (icmp eq B, 0) --> (icmp eq (A|B), 0)
if (LHSCC == ICmpInst::ICMP_EQ && LHSCst->isZero()) {
Value *NewOr = Builder->CreateOr(Val, Val2);
return new ICmpInst(LHSCC, NewOr, LHSCst);
}
}
// From here on, we only handle:
// (icmp1 A, C1) & (icmp2 A, C2) --> something simpler.
if (Val != Val2) return 0;
// ICMP_[US][GL]E X, CST is folded to ICMP_[US][GL]T elsewhere.
if (LHSCC == ICmpInst::ICMP_UGE || LHSCC == ICmpInst::ICMP_ULE ||
RHSCC == ICmpInst::ICMP_UGE || RHSCC == ICmpInst::ICMP_ULE ||
LHSCC == ICmpInst::ICMP_SGE || LHSCC == ICmpInst::ICMP_SLE ||
RHSCC == ICmpInst::ICMP_SGE || RHSCC == ICmpInst::ICMP_SLE)
return 0;
// We can't fold (ugt x, C) & (sgt x, C2).
if (!PredicatesFoldable(LHSCC, RHSCC))
return 0;
// Ensure that the larger constant is on the RHS.
bool ShouldSwap;
if (CmpInst::isSigned(LHSCC) ||
(ICmpInst::isEquality(LHSCC) &&
CmpInst::isSigned(RHSCC)))
ShouldSwap = LHSCst->getValue().sgt(RHSCst->getValue());
else
ShouldSwap = LHSCst->getValue().ugt(RHSCst->getValue());
if (ShouldSwap) {
std::swap(LHS, RHS);
std::swap(LHSCst, RHSCst);
std::swap(LHSCC, RHSCC);
}
// At this point, we know we have have two icmp instructions
// comparing a value against two constants and and'ing the result
// together. Because of the above check, we know that we only have
// icmp eq, icmp ne, icmp [su]lt, and icmp [SU]gt here. We also know
// (from the icmp folding check above), that the two constants
// are not equal and that the larger constant is on the RHS
assert(LHSCst != RHSCst && "Compares not folded above?");
switch (LHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_EQ:
switch (RHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_EQ: // (X == 13 & X == 15) -> false
case ICmpInst::ICMP_UGT: // (X == 13 & X > 15) -> false
case ICmpInst::ICMP_SGT: // (X == 13 & X > 15) -> false
return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getContext()));
case ICmpInst::ICMP_NE: // (X == 13 & X != 15) -> X == 13
case ICmpInst::ICMP_ULT: // (X == 13 & X < 15) -> X == 13
case ICmpInst::ICMP_SLT: // (X == 13 & X < 15) -> X == 13
return ReplaceInstUsesWith(I, LHS);
}
case ICmpInst::ICMP_NE:
switch (RHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_ULT:
if (LHSCst == SubOne(RHSCst)) // (X != 13 & X u< 14) -> X < 13
return new ICmpInst(ICmpInst::ICMP_ULT, Val, LHSCst);
break; // (X != 13 & X u< 15) -> no change
case ICmpInst::ICMP_SLT:
if (LHSCst == SubOne(RHSCst)) // (X != 13 & X s< 14) -> X < 13
return new ICmpInst(ICmpInst::ICMP_SLT, Val, LHSCst);
break; // (X != 13 & X s< 15) -> no change
case ICmpInst::ICMP_EQ: // (X != 13 & X == 15) -> X == 15
case ICmpInst::ICMP_UGT: // (X != 13 & X u> 15) -> X u> 15
case ICmpInst::ICMP_SGT: // (X != 13 & X s> 15) -> X s> 15
return ReplaceInstUsesWith(I, RHS);
case ICmpInst::ICMP_NE:
if (LHSCst == SubOne(RHSCst)){// (X != 13 & X != 14) -> X-13 >u 1
Constant *AddCST = ConstantExpr::getNeg(LHSCst);
Value *Add = Builder->CreateAdd(Val, AddCST, Val->getName()+".off");
return new ICmpInst(ICmpInst::ICMP_UGT, Add,
ConstantInt::get(Add->getType(), 1));
}
break; // (X != 13 & X != 15) -> no change
}
break;
case ICmpInst::ICMP_ULT:
switch (RHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_EQ: // (X u< 13 & X == 15) -> false
case ICmpInst::ICMP_UGT: // (X u< 13 & X u> 15) -> false
return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getContext()));
case ICmpInst::ICMP_SGT: // (X u< 13 & X s> 15) -> no change
break;
case ICmpInst::ICMP_NE: // (X u< 13 & X != 15) -> X u< 13
case ICmpInst::ICMP_ULT: // (X u< 13 & X u< 15) -> X u< 13
return ReplaceInstUsesWith(I, LHS);
case ICmpInst::ICMP_SLT: // (X u< 13 & X s< 15) -> no change
break;
}
break;
case ICmpInst::ICMP_SLT:
switch (RHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_EQ: // (X s< 13 & X == 15) -> false
case ICmpInst::ICMP_SGT: // (X s< 13 & X s> 15) -> false
return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getContext()));
case ICmpInst::ICMP_UGT: // (X s< 13 & X u> 15) -> no change
break;
case ICmpInst::ICMP_NE: // (X s< 13 & X != 15) -> X < 13
case ICmpInst::ICMP_SLT: // (X s< 13 & X s< 15) -> X < 13
return ReplaceInstUsesWith(I, LHS);
case ICmpInst::ICMP_ULT: // (X s< 13 & X u< 15) -> no change
break;
}
break;
case ICmpInst::ICMP_UGT:
switch (RHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_EQ: // (X u> 13 & X == 15) -> X == 15
case ICmpInst::ICMP_UGT: // (X u> 13 & X u> 15) -> X u> 15
return ReplaceInstUsesWith(I, RHS);
case ICmpInst::ICMP_SGT: // (X u> 13 & X s> 15) -> no change
break;
case ICmpInst::ICMP_NE:
if (RHSCst == AddOne(LHSCst)) // (X u> 13 & X != 14) -> X u> 14
return new ICmpInst(LHSCC, Val, RHSCst);
break; // (X u> 13 & X != 15) -> no change
case ICmpInst::ICMP_ULT: // (X u> 13 & X u< 15) -> (X-14) <u 1
return InsertRangeTest(Val, AddOne(LHSCst),
RHSCst, false, true, I);
case ICmpInst::ICMP_SLT: // (X u> 13 & X s< 15) -> no change
break;
}
break;
case ICmpInst::ICMP_SGT:
switch (RHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_EQ: // (X s> 13 & X == 15) -> X == 15
case ICmpInst::ICMP_SGT: // (X s> 13 & X s> 15) -> X s> 15
return ReplaceInstUsesWith(I, RHS);
case ICmpInst::ICMP_UGT: // (X s> 13 & X u> 15) -> no change
break;
case ICmpInst::ICMP_NE:
if (RHSCst == AddOne(LHSCst)) // (X s> 13 & X != 14) -> X s> 14
return new ICmpInst(LHSCC, Val, RHSCst);
break; // (X s> 13 & X != 15) -> no change
case ICmpInst::ICMP_SLT: // (X s> 13 & X s< 15) -> (X-14) s< 1
return InsertRangeTest(Val, AddOne(LHSCst),
RHSCst, true, true, I);
case ICmpInst::ICMP_ULT: // (X s> 13 & X u< 15) -> no change
break;
}
break;
}
return 0;
}
Instruction *InstCombiner::FoldAndOfFCmps(Instruction &I, FCmpInst *LHS,
FCmpInst *RHS) {
if (LHS->getPredicate() == FCmpInst::FCMP_ORD &&
RHS->getPredicate() == FCmpInst::FCMP_ORD) {
// (fcmp ord x, c) & (fcmp ord y, c) -> (fcmp ord x, y)
if (ConstantFP *LHSC = dyn_cast<ConstantFP>(LHS->getOperand(1)))
if (ConstantFP *RHSC = dyn_cast<ConstantFP>(RHS->getOperand(1))) {
// If either of the constants are nans, then the whole thing returns
// false.
if (LHSC->getValueAPF().isNaN() || RHSC->getValueAPF().isNaN())
return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getContext()));
return new FCmpInst(FCmpInst::FCMP_ORD,
LHS->getOperand(0), RHS->getOperand(0));
}
// Handle vector zeros. This occurs because the canonical form of
// "fcmp ord x,x" is "fcmp ord x, 0".
if (isa<ConstantAggregateZero>(LHS->getOperand(1)) &&
isa<ConstantAggregateZero>(RHS->getOperand(1)))
return new FCmpInst(FCmpInst::FCMP_ORD,
LHS->getOperand(0), RHS->getOperand(0));
return 0;
}
Value *Op0LHS = LHS->getOperand(0), *Op0RHS = LHS->getOperand(1);
Value *Op1LHS = RHS->getOperand(0), *Op1RHS = RHS->getOperand(1);
FCmpInst::Predicate Op0CC = LHS->getPredicate(), Op1CC = RHS->getPredicate();
if (Op0LHS == Op1RHS && Op0RHS == Op1LHS) {
// Swap RHS operands to match LHS.
Op1CC = FCmpInst::getSwappedPredicate(Op1CC);
std::swap(Op1LHS, Op1RHS);
}
if (Op0LHS == Op1LHS && Op0RHS == Op1RHS) {
// Simplify (fcmp cc0 x, y) & (fcmp cc1 x, y).
if (Op0CC == Op1CC)
return new FCmpInst((FCmpInst::Predicate)Op0CC, Op0LHS, Op0RHS);
if (Op0CC == FCmpInst::FCMP_FALSE || Op1CC == FCmpInst::FCMP_FALSE)
return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getContext()));
if (Op0CC == FCmpInst::FCMP_TRUE)
return ReplaceInstUsesWith(I, RHS);
if (Op1CC == FCmpInst::FCMP_TRUE)
return ReplaceInstUsesWith(I, LHS);
bool Op0Ordered;
bool Op1Ordered;
unsigned Op0Pred = getFCmpCode(Op0CC, Op0Ordered);
unsigned Op1Pred = getFCmpCode(Op1CC, Op1Ordered);
if (Op1Pred == 0) {
std::swap(LHS, RHS);
std::swap(Op0Pred, Op1Pred);
std::swap(Op0Ordered, Op1Ordered);
}
if (Op0Pred == 0) {
// uno && ueq -> uno && (uno || eq) -> ueq
// ord && olt -> ord && (ord && lt) -> olt
if (Op0Ordered == Op1Ordered)
return ReplaceInstUsesWith(I, RHS);
// uno && oeq -> uno && (ord && eq) -> false
// uno && ord -> false
if (!Op0Ordered)
return ReplaceInstUsesWith(I, ConstantInt::getFalse(I.getContext()));
// ord && ueq -> ord && (uno || eq) -> oeq
return cast<Instruction>(getFCmpValue(true, Op1Pred, Op0LHS, Op0RHS));
}
}
return 0;
}
Instruction *InstCombiner::visitAnd(BinaryOperator &I) {
bool Changed = SimplifyCommutative(I);
Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
if (Value *V = SimplifyAndInst(Op0, Op1, TD))
return ReplaceInstUsesWith(I, V);
// See if we can simplify any instructions used by the instruction whose sole
// purpose is to compute bits we don't care about.
if (SimplifyDemandedInstructionBits(I))
return &I;
if (ConstantInt *AndRHS = dyn_cast<ConstantInt>(Op1)) {
const APInt &AndRHSMask = AndRHS->getValue();
APInt NotAndRHS(~AndRHSMask);
// Optimize a variety of ((val OP C1) & C2) combinations...
if (BinaryOperator *Op0I = dyn_cast<BinaryOperator>(Op0)) {
Value *Op0LHS = Op0I->getOperand(0);
Value *Op0RHS = Op0I->getOperand(1);
switch (Op0I->getOpcode()) {
default: break;
case Instruction::Xor:
case Instruction::Or:
// If the mask is only needed on one incoming arm, push it up.
if (!Op0I->hasOneUse()) break;
if (MaskedValueIsZero(Op0LHS, NotAndRHS)) {
// Not masking anything out for the LHS, move to RHS.
Value *NewRHS = Builder->CreateAnd(Op0RHS, AndRHS,
Op0RHS->getName()+".masked");
return BinaryOperator::Create(Op0I->getOpcode(), Op0LHS, NewRHS);
}
if (!isa<Constant>(Op0RHS) &&
MaskedValueIsZero(Op0RHS, NotAndRHS)) {
// Not masking anything out for the RHS, move to LHS.
Value *NewLHS = Builder->CreateAnd(Op0LHS, AndRHS,
Op0LHS->getName()+".masked");
return BinaryOperator::Create(Op0I->getOpcode(), NewLHS, Op0RHS);
}
break;
case Instruction::Add:
// ((A & N) + B) & AndRHS -> (A + B) & AndRHS iff N&AndRHS == AndRHS.
// ((A | N) + B) & AndRHS -> (A + B) & AndRHS iff N&AndRHS == 0
// ((A ^ N) + B) & AndRHS -> (A + B) & AndRHS iff N&AndRHS == 0
if (Value *V = FoldLogicalPlusAnd(Op0LHS, Op0RHS, AndRHS, false, I))
return BinaryOperator::CreateAnd(V, AndRHS);
if (Value *V = FoldLogicalPlusAnd(Op0RHS, Op0LHS, AndRHS, false, I))
return BinaryOperator::CreateAnd(V, AndRHS); // Add commutes
break;
case Instruction::Sub:
// ((A & N) - B) & AndRHS -> (A - B) & AndRHS iff N&AndRHS == AndRHS.
// ((A | N) - B) & AndRHS -> (A - B) & AndRHS iff N&AndRHS == 0
// ((A ^ N) - B) & AndRHS -> (A - B) & AndRHS iff N&AndRHS == 0
if (Value *V = FoldLogicalPlusAnd(Op0LHS, Op0RHS, AndRHS, true, I))
return BinaryOperator::CreateAnd(V, AndRHS);
// (A - N) & AndRHS -> -N & AndRHS iff A&AndRHS==0 and AndRHS
// has 1's for all bits that the subtraction with A might affect.
if (Op0I->hasOneUse()) {
uint32_t BitWidth = AndRHSMask.getBitWidth();
uint32_t Zeros = AndRHSMask.countLeadingZeros();
APInt Mask = APInt::getLowBitsSet(BitWidth, BitWidth - Zeros);
ConstantInt *A = dyn_cast<ConstantInt>(Op0LHS);
if (!(A && A->isZero()) && // avoid infinite recursion.
MaskedValueIsZero(Op0LHS, Mask)) {
Value *NewNeg = Builder->CreateNeg(Op0RHS);
return BinaryOperator::CreateAnd(NewNeg, AndRHS);
}
}
break;
case Instruction::Shl:
case Instruction::LShr:
// (1 << x) & 1 --> zext(x == 0)
// (1 >> x) & 1 --> zext(x == 0)
if (AndRHSMask == 1 && Op0LHS == AndRHS) {
Value *NewICmp =
Builder->CreateICmpEQ(Op0RHS, Constant::getNullValue(I.getType()));
return new ZExtInst(NewICmp, I.getType());
}
break;
}
if (ConstantInt *Op0CI = dyn_cast<ConstantInt>(Op0I->getOperand(1)))
if (Instruction *Res = OptAndOp(Op0I, Op0CI, AndRHS, I))
return Res;
} else if (CastInst *CI = dyn_cast<CastInst>(Op0)) {
// If this is an integer truncation or change from signed-to-unsigned, and
// if the source is an and/or with immediate, transform it. This
// frequently occurs for bitfield accesses.
if (Instruction *CastOp = dyn_cast<Instruction>(CI->getOperand(0))) {
if ((isa<TruncInst>(CI) || isa<BitCastInst>(CI)) &&
CastOp->getNumOperands() == 2)
if (ConstantInt *AndCI =dyn_cast<ConstantInt>(CastOp->getOperand(1))){
if (CastOp->getOpcode() == Instruction::And) {
// Change: and (cast (and X, C1) to T), C2
// into : and (cast X to T), trunc_or_bitcast(C1)&C2
// This will fold the two constants together, which may allow
// other simplifications.
Value *NewCast = Builder->CreateTruncOrBitCast(
CastOp->getOperand(0), I.getType(),
CastOp->getName()+".shrunk");
// trunc_or_bitcast(C1)&C2
Constant *C3 = ConstantExpr::getTruncOrBitCast(AndCI,I.getType());
C3 = ConstantExpr::getAnd(C3, AndRHS);
return BinaryOperator::CreateAnd(NewCast, C3);
} else if (CastOp->getOpcode() == Instruction::Or) {
// Change: and (cast (or X, C1) to T), C2
// into : trunc(C1)&C2 iff trunc(C1)&C2 == C2
Constant *C3 = ConstantExpr::getTruncOrBitCast(AndCI,I.getType());
if (ConstantExpr::getAnd(C3, AndRHS) == AndRHS)
// trunc(C1)&C2
return ReplaceInstUsesWith(I, AndRHS);
}
}
}
}
// Try to fold constant and into select arguments.
if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
if (Instruction *R = FoldOpIntoSelect(I, SI))
return R;
if (isa<PHINode>(Op0))
if (Instruction *NV = FoldOpIntoPhi(I))
return NV;
}
// (~A & ~B) == (~(A | B)) - De Morgan's Law
if (Value *Op0NotVal = dyn_castNotVal(Op0))
if (Value *Op1NotVal = dyn_castNotVal(Op1))
if (Op0->hasOneUse() && Op1->hasOneUse()) {
Value *Or = Builder->CreateOr(Op0NotVal, Op1NotVal,
I.getName()+".demorgan");
return BinaryOperator::CreateNot(Or);
}
{
Value *A = 0, *B = 0, *C = 0, *D = 0;
// (A|B) & ~(A&B) -> A^B
if (match(Op0, m_Or(m_Value(A), m_Value(B))) &&
match(Op1, m_Not(m_And(m_Value(C), m_Value(D)))) &&
((A == C && B == D) || (A == D && B == C)))
return BinaryOperator::CreateXor(A, B);
// ~(A&B) & (A|B) -> A^B
if (match(Op1, m_Or(m_Value(A), m_Value(B))) &&
match(Op0, m_Not(m_And(m_Value(C), m_Value(D)))) &&
((A == C && B == D) || (A == D && B == C)))
return BinaryOperator::CreateXor(A, B);
if (Op0->hasOneUse() &&
match(Op0, m_Xor(m_Value(A), m_Value(B)))) {
if (A == Op1) { // (A^B)&A -> A&(A^B)
I.swapOperands(); // Simplify below
std::swap(Op0, Op1);
} else if (B == Op1) { // (A^B)&B -> B&(B^A)
cast<BinaryOperator>(Op0)->swapOperands();
I.swapOperands(); // Simplify below
std::swap(Op0, Op1);
}
}
if (Op1->hasOneUse() &&
match(Op1, m_Xor(m_Value(A), m_Value(B)))) {
if (B == Op0) { // B&(A^B) -> B&(B^A)
cast<BinaryOperator>(Op1)->swapOperands();
std::swap(A, B);
}
if (A == Op0) // A&(A^B) -> A & ~B
return BinaryOperator::CreateAnd(A, Builder->CreateNot(B, "tmp"));
}
// (A&((~A)|B)) -> A&B
if (match(Op0, m_Or(m_Not(m_Specific(Op1)), m_Value(A))) ||
match(Op0, m_Or(m_Value(A), m_Not(m_Specific(Op1)))))
return BinaryOperator::CreateAnd(A, Op1);
if (match(Op1, m_Or(m_Not(m_Specific(Op0)), m_Value(A))) ||
match(Op1, m_Or(m_Value(A), m_Not(m_Specific(Op0)))))
return BinaryOperator::CreateAnd(A, Op0);
}
if (ICmpInst *RHS = dyn_cast<ICmpInst>(Op1))
if (ICmpInst *LHS = dyn_cast<ICmpInst>(Op0))
if (Instruction *Res = FoldAndOfICmps(I, LHS, RHS))
return Res;
// fold (and (cast A), (cast B)) -> (cast (and A, B))
if (CastInst *Op0C = dyn_cast<CastInst>(Op0))
if (CastInst *Op1C = dyn_cast<CastInst>(Op1))
if (Op0C->getOpcode() == Op1C->getOpcode()) { // same cast kind ?
const Type *SrcTy = Op0C->getOperand(0)->getType();
if (SrcTy == Op1C->getOperand(0)->getType() &&
SrcTy->isIntOrIntVector() &&
// Only do this if the casts both really cause code to be generated.
ValueRequiresCast(Op0C->getOpcode(), Op0C->getOperand(0),
I.getType()) &&
ValueRequiresCast(Op1C->getOpcode(), Op1C->getOperand(0),
I.getType())) {
Value *NewOp = Builder->CreateAnd(Op0C->getOperand(0),
Op1C->getOperand(0), I.getName());
return CastInst::Create(Op0C->getOpcode(), NewOp, I.getType());
}
}
// (X >> Z) & (Y >> Z) -> (X&Y) >> Z for all shifts.
if (BinaryOperator *SI1 = dyn_cast<BinaryOperator>(Op1)) {
if (BinaryOperator *SI0 = dyn_cast<BinaryOperator>(Op0))
if (SI0->isShift() && SI0->getOpcode() == SI1->getOpcode() &&
SI0->getOperand(1) == SI1->getOperand(1) &&
(SI0->hasOneUse() || SI1->hasOneUse())) {
Value *NewOp =
Builder->CreateAnd(SI0->getOperand(0), SI1->getOperand(0),
SI0->getName());
return BinaryOperator::Create(SI1->getOpcode(), NewOp,
SI1->getOperand(1));
}
}
// If and'ing two fcmp, try combine them into one.
if (FCmpInst *LHS = dyn_cast<FCmpInst>(I.getOperand(0))) {
if (FCmpInst *RHS = dyn_cast<FCmpInst>(I.getOperand(1)))
if (Instruction *Res = FoldAndOfFCmps(I, LHS, RHS))
return Res;
}
return Changed ? &I : 0;
}
/// CollectBSwapParts - Analyze the specified subexpression and see if it is
/// capable of providing pieces of a bswap. The subexpression provides pieces
/// of a bswap if it is proven that each of the non-zero bytes in the output of
/// the expression came from the corresponding "byte swapped" byte in some other
/// value. For example, if the current subexpression is "(shl i32 %X, 24)" then
/// we know that the expression deposits the low byte of %X into the high byte
/// of the bswap result and that all other bytes are zero. This expression is
/// accepted, the high byte of ByteValues is set to X to indicate a correct
/// match.
///
/// This function returns true if the match was unsuccessful and false if so.
/// On entry to the function the "OverallLeftShift" is a signed integer value
/// indicating the number of bytes that the subexpression is later shifted. For
/// example, if the expression is later right shifted by 16 bits, the
/// OverallLeftShift value would be -2 on entry. This is used to specify which
/// byte of ByteValues is actually being set.
///
/// Similarly, ByteMask is a bitmask where a bit is clear if its corresponding
/// byte is masked to zero by a user. For example, in (X & 255), X will be
/// processed with a bytemask of 1. Because bytemask is 32-bits, this limits
/// this function to working on up to 32-byte (256 bit) values. ByteMask is
/// always in the local (OverallLeftShift) coordinate space.
///
static bool CollectBSwapParts(Value *V, int OverallLeftShift, uint32_t ByteMask,
SmallVector<Value*, 8> &ByteValues) {
if (Instruction *I = dyn_cast<Instruction>(V)) {
// If this is an or instruction, it may be an inner node of the bswap.
if (I->getOpcode() == Instruction::Or) {
return CollectBSwapParts(I->getOperand(0), OverallLeftShift, ByteMask,
ByteValues) ||
CollectBSwapParts(I->getOperand(1), OverallLeftShift, ByteMask,
ByteValues);
}
// If this is a logical shift by a constant multiple of 8, recurse with
// OverallLeftShift and ByteMask adjusted.
if (I->isLogicalShift() && isa<ConstantInt>(I->getOperand(1))) {
unsigned ShAmt =
cast<ConstantInt>(I->getOperand(1))->getLimitedValue(~0U);
// Ensure the shift amount is defined and of a byte value.
if ((ShAmt & 7) || (ShAmt > 8*ByteValues.size()))
return true;
unsigned ByteShift = ShAmt >> 3;
if (I->getOpcode() == Instruction::Shl) {
// X << 2 -> collect(X, +2)
OverallLeftShift += ByteShift;
ByteMask >>= ByteShift;
} else {
// X >>u 2 -> collect(X, -2)
OverallLeftShift -= ByteShift;
ByteMask <<= ByteShift;
ByteMask &= (~0U >> (32-ByteValues.size()));
}
if (OverallLeftShift >= (int)ByteValues.size()) return true;
if (OverallLeftShift <= -(int)ByteValues.size()) return true;
return CollectBSwapParts(I->getOperand(0), OverallLeftShift, ByteMask,
ByteValues);
}
// If this is a logical 'and' with a mask that clears bytes, clear the
// corresponding bytes in ByteMask.
if (I->getOpcode() == Instruction::And &&
isa<ConstantInt>(I->getOperand(1))) {
// Scan every byte of the and mask, seeing if the byte is either 0 or 255.
unsigned NumBytes = ByteValues.size();
APInt Byte(I->getType()->getPrimitiveSizeInBits(), 255);
const APInt &AndMask = cast<ConstantInt>(I->getOperand(1))->getValue();
for (unsigned i = 0; i != NumBytes; ++i, Byte <<= 8) {
// If this byte is masked out by a later operation, we don't care what
// the and mask is.
if ((ByteMask & (1 << i)) == 0)
continue;
// If the AndMask is all zeros for this byte, clear the bit.
APInt MaskB = AndMask & Byte;
if (MaskB == 0) {
ByteMask &= ~(1U << i);
continue;
}
// If the AndMask is not all ones for this byte, it's not a bytezap.
if (MaskB != Byte)
return true;
// Otherwise, this byte is kept.
}
return CollectBSwapParts(I->getOperand(0), OverallLeftShift, ByteMask,
ByteValues);
}
}
// Okay, we got to something that isn't a shift, 'or' or 'and'. This must be
// the input value to the bswap. Some observations: 1) if more than one byte
// is demanded from this input, then it could not be successfully assembled
// into a byteswap. At least one of the two bytes would not be aligned with
// their ultimate destination.
if (!isPowerOf2_32(ByteMask)) return true;
unsigned InputByteNo = CountTrailingZeros_32(ByteMask);
// 2) The input and ultimate destinations must line up: if byte 3 of an i32
// is demanded, it needs to go into byte 0 of the result. This means that the
// byte needs to be shifted until it lands in the right byte bucket. The
// shift amount depends on the position: if the byte is coming from the high
// part of the value (e.g. byte 3) then it must be shifted right. If from the
// low part, it must be shifted left.
unsigned DestByteNo = InputByteNo + OverallLeftShift;
if (InputByteNo < ByteValues.size()/2) {
if (ByteValues.size()-1-DestByteNo != InputByteNo)
return true;
} else {
if (ByteValues.size()-1-DestByteNo != InputByteNo)
return true;
}
// If the destination byte value is already defined, the values are or'd
// together, which isn't a bswap (unless it's an or of the same bits).
if (ByteValues[DestByteNo] && ByteValues[DestByteNo] != V)
return true;
ByteValues[DestByteNo] = V;
return false;
}
/// MatchBSwap - Given an OR instruction, check to see if this is a bswap idiom.
/// If so, insert the new bswap intrinsic and return it.
Instruction *InstCombiner::MatchBSwap(BinaryOperator &I) {
const IntegerType *ITy = dyn_cast<IntegerType>(I.getType());
if (!ITy || ITy->getBitWidth() % 16 ||
// ByteMask only allows up to 32-byte values.
ITy->getBitWidth() > 32*8)
return 0; // Can only bswap pairs of bytes. Can't do vectors.
/// ByteValues - For each byte of the result, we keep track of which value
/// defines each byte.
SmallVector<Value*, 8> ByteValues;
ByteValues.resize(ITy->getBitWidth()/8);
// Try to find all the pieces corresponding to the bswap.
uint32_t ByteMask = ~0U >> (32-ByteValues.size());
if (CollectBSwapParts(&I, 0, ByteMask, ByteValues))
return 0;
// Check to see if all of the bytes come from the same value.
Value *V = ByteValues[0];
if (V == 0) return 0; // Didn't find a byte? Must be zero.
// Check to make sure that all of the bytes come from the same value.
for (unsigned i = 1, e = ByteValues.size(); i != e; ++i)
if (ByteValues[i] != V)
return 0;
const Type *Tys[] = { ITy };
Module *M = I.getParent()->getParent()->getParent();
Function *F = Intrinsic::getDeclaration(M, Intrinsic::bswap, Tys, 1);
return CallInst::Create(F, V);
}
/// MatchSelectFromAndOr - We have an expression of the form (A&C)|(B&D). Check
/// If A is (cond?-1:0) and either B or D is ~(cond?-1,0) or (cond?0,-1), then
/// we can simplify this expression to "cond ? C : D or B".
static Instruction *MatchSelectFromAndOr(Value *A, Value *B,
Value *C, Value *D) {
// If A is not a select of -1/0, this cannot match.
Value *Cond = 0;
if (!match(A, m_SelectCst<-1, 0>(m_Value(Cond))))
return 0;
// ((cond?-1:0)&C) | (B&(cond?0:-1)) -> cond ? C : B.
if (match(D, m_SelectCst<0, -1>(m_Specific(Cond))))
return SelectInst::Create(Cond, C, B);
if (match(D, m_Not(m_SelectCst<-1, 0>(m_Specific(Cond)))))
return SelectInst::Create(Cond, C, B);
// ((cond?-1:0)&C) | ((cond?0:-1)&D) -> cond ? C : D.
if (match(B, m_SelectCst<0, -1>(m_Specific(Cond))))
return SelectInst::Create(Cond, C, D);
if (match(B, m_Not(m_SelectCst<-1, 0>(m_Specific(Cond)))))
return SelectInst::Create(Cond, C, D);
return 0;
}
/// FoldOrOfICmps - Fold (icmp)|(icmp) if possible.
Instruction *InstCombiner::FoldOrOfICmps(Instruction &I,
ICmpInst *LHS, ICmpInst *RHS) {
ICmpInst::Predicate LHSCC = LHS->getPredicate(), RHSCC = RHS->getPredicate();
// (icmp1 A, B) | (icmp2 A, B) --> (icmp3 A, B)
if (PredicatesFoldable(LHSCC, RHSCC)) {
if (LHS->getOperand(0) == RHS->getOperand(1) &&
LHS->getOperand(1) == RHS->getOperand(0))
LHS->swapOperands();
if (LHS->getOperand(0) == RHS->getOperand(0) &&
LHS->getOperand(1) == RHS->getOperand(1)) {
Value *Op0 = LHS->getOperand(0), *Op1 = LHS->getOperand(1);
unsigned Code = getICmpCode(LHS) | getICmpCode(RHS);
bool isSigned = LHS->isSigned() || RHS->isSigned();
Value *RV = getICmpValue(isSigned, Code, Op0, Op1);
if (Instruction *I = dyn_cast<Instruction>(RV))
return I;
// Otherwise, it's a constant boolean value.
return ReplaceInstUsesWith(I, RV);
}
}
// This only handles icmp of constants: (icmp1 A, C1) | (icmp2 B, C2).
Value *Val = LHS->getOperand(0), *Val2 = RHS->getOperand(0);
ConstantInt *LHSCst = dyn_cast<ConstantInt>(LHS->getOperand(1));
ConstantInt *RHSCst = dyn_cast<ConstantInt>(RHS->getOperand(1));
if (LHSCst == 0 || RHSCst == 0) return 0;
// (icmp ne A, 0) | (icmp ne B, 0) --> (icmp ne (A|B), 0)
if (LHSCst == RHSCst && LHSCC == RHSCC &&
LHSCC == ICmpInst::ICMP_NE && LHSCst->isZero()) {
Value *NewOr = Builder->CreateOr(Val, Val2);
return new ICmpInst(LHSCC, NewOr, LHSCst);
}
// From here on, we only handle:
// (icmp1 A, C1) | (icmp2 A, C2) --> something simpler.
if (Val != Val2) return 0;
// ICMP_[US][GL]E X, CST is folded to ICMP_[US][GL]T elsewhere.
if (LHSCC == ICmpInst::ICMP_UGE || LHSCC == ICmpInst::ICMP_ULE ||
RHSCC == ICmpInst::ICMP_UGE || RHSCC == ICmpInst::ICMP_ULE ||
LHSCC == ICmpInst::ICMP_SGE || LHSCC == ICmpInst::ICMP_SLE ||
RHSCC == ICmpInst::ICMP_SGE || RHSCC == ICmpInst::ICMP_SLE)
return 0;
// We can't fold (ugt x, C) | (sgt x, C2).
if (!PredicatesFoldable(LHSCC, RHSCC))
return 0;
// Ensure that the larger constant is on the RHS.
bool ShouldSwap;
if (CmpInst::isSigned(LHSCC) ||
(ICmpInst::isEquality(LHSCC) &&
CmpInst::isSigned(RHSCC)))
ShouldSwap = LHSCst->getValue().sgt(RHSCst->getValue());
else
ShouldSwap = LHSCst->getValue().ugt(RHSCst->getValue());
if (ShouldSwap) {
std::swap(LHS, RHS);
std::swap(LHSCst, RHSCst);
std::swap(LHSCC, RHSCC);
}
// At this point, we know we have have two icmp instructions
// comparing a value against two constants and or'ing the result
// together. Because of the above check, we know that we only have
// ICMP_EQ, ICMP_NE, ICMP_LT, and ICMP_GT here. We also know (from the
// icmp folding check above), that the two constants are not
// equal.
assert(LHSCst != RHSCst && "Compares not folded above?");
switch (LHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_EQ:
switch (RHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_EQ:
if (LHSCst == SubOne(RHSCst)) {
// (X == 13 | X == 14) -> X-13 <u 2
Constant *AddCST = ConstantExpr::getNeg(LHSCst);
Value *Add = Builder->CreateAdd(Val, AddCST, Val->getName()+".off");
AddCST = ConstantExpr::getSub(AddOne(RHSCst), LHSCst);
return new ICmpInst(ICmpInst::ICMP_ULT, Add, AddCST);
}
break; // (X == 13 | X == 15) -> no change
case ICmpInst::ICMP_UGT: // (X == 13 | X u> 14) -> no change
case ICmpInst::ICMP_SGT: // (X == 13 | X s> 14) -> no change
break;
case ICmpInst::ICMP_NE: // (X == 13 | X != 15) -> X != 15
case ICmpInst::ICMP_ULT: // (X == 13 | X u< 15) -> X u< 15
case ICmpInst::ICMP_SLT: // (X == 13 | X s< 15) -> X s< 15
return ReplaceInstUsesWith(I, RHS);
}
break;
case ICmpInst::ICMP_NE:
switch (RHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_EQ: // (X != 13 | X == 15) -> X != 13
case ICmpInst::ICMP_UGT: // (X != 13 | X u> 15) -> X != 13
case ICmpInst::ICMP_SGT: // (X != 13 | X s> 15) -> X != 13
return ReplaceInstUsesWith(I, LHS);
case ICmpInst::ICMP_NE: // (X != 13 | X != 15) -> true
case ICmpInst::ICMP_ULT: // (X != 13 | X u< 15) -> true
case ICmpInst::ICMP_SLT: // (X != 13 | X s< 15) -> true
return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getContext()));
}
break;
case ICmpInst::ICMP_ULT:
switch (RHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_EQ: // (X u< 13 | X == 14) -> no change
break;
case ICmpInst::ICMP_UGT: // (X u< 13 | X u> 15) -> (X-13) u> 2
// If RHSCst is [us]MAXINT, it is always false. Not handling
// this can cause overflow.
if (RHSCst->isMaxValue(false))
return ReplaceInstUsesWith(I, LHS);
return InsertRangeTest(Val, LHSCst, AddOne(RHSCst),
false, false, I);
case ICmpInst::ICMP_SGT: // (X u< 13 | X s> 15) -> no change
break;
case ICmpInst::ICMP_NE: // (X u< 13 | X != 15) -> X != 15
case ICmpInst::ICMP_ULT: // (X u< 13 | X u< 15) -> X u< 15
return ReplaceInstUsesWith(I, RHS);
case ICmpInst::ICMP_SLT: // (X u< 13 | X s< 15) -> no change
break;
}
break;
case ICmpInst::ICMP_SLT:
switch (RHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_EQ: // (X s< 13 | X == 14) -> no change
break;
case ICmpInst::ICMP_SGT: // (X s< 13 | X s> 15) -> (X-13) s> 2
// If RHSCst is [us]MAXINT, it is always false. Not handling
// this can cause overflow.
if (RHSCst->isMaxValue(true))
return ReplaceInstUsesWith(I, LHS);
return InsertRangeTest(Val, LHSCst, AddOne(RHSCst),
true, false, I);
case ICmpInst::ICMP_UGT: // (X s< 13 | X u> 15) -> no change
break;
case ICmpInst::ICMP_NE: // (X s< 13 | X != 15) -> X != 15
case ICmpInst::ICMP_SLT: // (X s< 13 | X s< 15) -> X s< 15
return ReplaceInstUsesWith(I, RHS);
case ICmpInst::ICMP_ULT: // (X s< 13 | X u< 15) -> no change
break;
}
break;
case ICmpInst::ICMP_UGT:
switch (RHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_EQ: // (X u> 13 | X == 15) -> X u> 13
case ICmpInst::ICMP_UGT: // (X u> 13 | X u> 15) -> X u> 13
return ReplaceInstUsesWith(I, LHS);
case ICmpInst::ICMP_SGT: // (X u> 13 | X s> 15) -> no change
break;
case ICmpInst::ICMP_NE: // (X u> 13 | X != 15) -> true
case ICmpInst::ICMP_ULT: // (X u> 13 | X u< 15) -> true
return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getContext()));
case ICmpInst::ICMP_SLT: // (X u> 13 | X s< 15) -> no change
break;
}
break;
case ICmpInst::ICMP_SGT:
switch (RHSCC) {
default: llvm_unreachable("Unknown integer condition code!");
case ICmpInst::ICMP_EQ: // (X s> 13 | X == 15) -> X > 13
case ICmpInst::ICMP_SGT: // (X s> 13 | X s> 15) -> X > 13
return ReplaceInstUsesWith(I, LHS);
case ICmpInst::ICMP_UGT: // (X s> 13 | X u> 15) -> no change
break;
case ICmpInst::ICMP_NE: // (X s> 13 | X != 15) -> true
case ICmpInst::ICMP_SLT: // (X s> 13 | X s< 15) -> true
return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getContext()));
case ICmpInst::ICMP_ULT: // (X s> 13 | X u< 15) -> no change
break;
}
break;
}
return 0;
}
Instruction *InstCombiner::FoldOrOfFCmps(Instruction &I, FCmpInst *LHS,
FCmpInst *RHS) {
if (LHS->getPredicate() == FCmpInst::FCMP_UNO &&
RHS->getPredicate() == FCmpInst::FCMP_UNO &&
LHS->getOperand(0)->getType() == RHS->getOperand(0)->getType()) {
if (ConstantFP *LHSC = dyn_cast<ConstantFP>(LHS->getOperand(1)))
if (ConstantFP *RHSC = dyn_cast<ConstantFP>(RHS->getOperand(1))) {
// If either of the constants are nans, then the whole thing returns
// true.
if (LHSC->getValueAPF().isNaN() || RHSC->getValueAPF().isNaN())
return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getContext()));
// Otherwise, no need to compare the two constants, compare the
// rest.
return new FCmpInst(FCmpInst::FCMP_UNO,
LHS->getOperand(0), RHS->getOperand(0));
}
// Handle vector zeros. This occurs because the canonical form of
// "fcmp uno x,x" is "fcmp uno x, 0".
if (isa<ConstantAggregateZero>(LHS->getOperand(1)) &&
isa<ConstantAggregateZero>(RHS->getOperand(1)))
return new FCmpInst(FCmpInst::FCMP_UNO,
LHS->getOperand(0), RHS->getOperand(0));
return 0;
}
Value *Op0LHS = LHS->getOperand(0), *Op0RHS = LHS->getOperand(1);
Value *Op1LHS = RHS->getOperand(0), *Op1RHS = RHS->getOperand(1);
FCmpInst::Predicate Op0CC = LHS->getPredicate(), Op1CC = RHS->getPredicate();
if (Op0LHS == Op1RHS && Op0RHS == Op1LHS) {
// Swap RHS operands to match LHS.
Op1CC = FCmpInst::getSwappedPredicate(Op1CC);
std::swap(Op1LHS, Op1RHS);
}
if (Op0LHS == Op1LHS && Op0RHS == Op1RHS) {
// Simplify (fcmp cc0 x, y) | (fcmp cc1 x, y).
if (Op0CC == Op1CC)
return new FCmpInst((FCmpInst::Predicate)Op0CC,
Op0LHS, Op0RHS);
if (Op0CC == FCmpInst::FCMP_TRUE || Op1CC == FCmpInst::FCMP_TRUE)
return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getContext()));
if (Op0CC == FCmpInst::FCMP_FALSE)
return ReplaceInstUsesWith(I, RHS);
if (Op1CC == FCmpInst::FCMP_FALSE)
return ReplaceInstUsesWith(I, LHS);
bool Op0Ordered;
bool Op1Ordered;
unsigned Op0Pred = getFCmpCode(Op0CC, Op0Ordered);
unsigned Op1Pred = getFCmpCode(Op1CC, Op1Ordered);
if (Op0Ordered == Op1Ordered) {
// If both are ordered or unordered, return a new fcmp with
// or'ed predicates.
Value *RV = getFCmpValue(Op0Ordered, Op0Pred|Op1Pred, Op0LHS, Op0RHS);
if (Instruction *I = dyn_cast<Instruction>(RV))
return I;
// Otherwise, it's a constant boolean value...
return ReplaceInstUsesWith(I, RV);
}
}
return 0;
}
/// FoldOrWithConstants - This helper function folds:
///
/// ((A | B) & C1) | (B & C2)
///
/// into:
///
/// (A & C1) | B
///
/// when the XOR of the two constants is "all ones" (-1).
Instruction *InstCombiner::FoldOrWithConstants(BinaryOperator &I, Value *Op,
Value *A, Value *B, Value *C) {
ConstantInt *CI1 = dyn_cast<ConstantInt>(C);
if (!CI1) return 0;
Value *V1 = 0;
ConstantInt *CI2 = 0;
if (!match(Op, m_And(m_Value(V1), m_ConstantInt(CI2)))) return 0;
APInt Xor = CI1->getValue() ^ CI2->getValue();
if (!Xor.isAllOnesValue()) return 0;
if (V1 == A || V1 == B) {
Value *NewOp = Builder->CreateAnd((V1 == A) ? B : A, CI1);
return BinaryOperator::CreateOr(NewOp, V1);
}
return 0;
}
Instruction *InstCombiner::visitOr(BinaryOperator &I) {
bool Changed = SimplifyCommutative(I);
Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
if (Value *V = SimplifyOrInst(Op0, Op1, TD))
return ReplaceInstUsesWith(I, V);
// See if we can simplify any instructions used by the instruction whose sole
// purpose is to compute bits we don't care about.
if (SimplifyDemandedInstructionBits(I))
return &I;
if (ConstantInt *RHS = dyn_cast<ConstantInt>(Op1)) {
ConstantInt *C1 = 0; Value *X = 0;
// (X & C1) | C2 --> (X | C2) & (C1|C2)
if (match(Op0, m_And(m_Value(X), m_ConstantInt(C1))) &&
Op0->hasOneUse()) {
Value *Or = Builder->CreateOr(X, RHS);
Or->takeName(Op0);
return BinaryOperator::CreateAnd(Or,
ConstantInt::get(I.getContext(),
RHS->getValue() | C1->getValue()));
}
// (X ^ C1) | C2 --> (X | C2) ^ (C1&~C2)
if (match(Op0, m_Xor(m_Value(X), m_ConstantInt(C1))) &&
Op0->hasOneUse()) {
Value *Or = Builder->CreateOr(X, RHS);
Or->takeName(Op0);
return BinaryOperator::CreateXor(Or,
ConstantInt::get(I.getContext(),
C1->getValue() & ~RHS->getValue()));
}
// Try to fold constant and into select arguments.
if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
if (Instruction *R = FoldOpIntoSelect(I, SI))
return R;
if (isa<PHINode>(Op0))
if (Instruction *NV = FoldOpIntoPhi(I))
return NV;
}
Value *A = 0, *B = 0;
ConstantInt *C1 = 0, *C2 = 0;
// (A | B) | C and A | (B | C) -> bswap if possible.
// (A >> B) | (C << D) and (A << B) | (B >> C) -> bswap if possible.
if (match(Op0, m_Or(m_Value(), m_Value())) ||
match(Op1, m_Or(m_Value(), m_Value())) ||
(match(Op0, m_Shift(m_Value(), m_Value())) &&
match(Op1, m_Shift(m_Value(), m_Value())))) {
if (Instruction *BSwap = MatchBSwap(I))
return BSwap;
}
// (X^C)|Y -> (X|Y)^C iff Y&C == 0
if (Op0->hasOneUse() &&
match(Op0, m_Xor(m_Value(A), m_ConstantInt(C1))) &&
MaskedValueIsZero(Op1, C1->getValue())) {
Value *NOr = Builder->CreateOr(A, Op1);
NOr->takeName(Op0);
return BinaryOperator::CreateXor(NOr, C1);
}
// Y|(X^C) -> (X|Y)^C iff Y&C == 0
if (Op1->hasOneUse() &&
match(Op1, m_Xor(m_Value(A), m_ConstantInt(C1))) &&
MaskedValueIsZero(Op0, C1->getValue())) {
Value *NOr = Builder->CreateOr(A, Op0);
NOr->takeName(Op0);
return BinaryOperator::CreateXor(NOr, C1);
}
// (A & C)|(B & D)
Value *C = 0, *D = 0;
if (match(Op0, m_And(m_Value(A), m_Value(C))) &&
match(Op1, m_And(m_Value(B), m_Value(D)))) {
Value *V1 = 0, *V2 = 0, *V3 = 0;
C1 = dyn_cast<ConstantInt>(C);
C2 = dyn_cast<ConstantInt>(D);
if (C1 && C2) { // (A & C1)|(B & C2)
// If we have: ((V + N) & C1) | (V & C2)
// .. and C2 = ~C1 and C2 is 0+1+ and (N & C2) == 0
// replace with V+N.
if (C1->getValue() == ~C2->getValue()) {
if ((C2->getValue() & (C2->getValue()+1)) == 0 && // C2 == 0+1+
match(A, m_Add(m_Value(V1), m_Value(V2)))) {
// Add commutes, try both ways.
if (V1 == B && MaskedValueIsZero(V2, C2->getValue()))
return ReplaceInstUsesWith(I, A);
if (V2 == B && MaskedValueIsZero(V1, C2->getValue()))
return ReplaceInstUsesWith(I, A);
}
// Or commutes, try both ways.
if ((C1->getValue() & (C1->getValue()+1)) == 0 &&
match(B, m_Add(m_Value(V1), m_Value(V2)))) {
// Add commutes, try both ways.
if (V1 == A && MaskedValueIsZero(V2, C1->getValue()))
return ReplaceInstUsesWith(I, B);
if (V2 == A && MaskedValueIsZero(V1, C1->getValue()))
return ReplaceInstUsesWith(I, B);
}
}
// ((V | N) & C1) | (V & C2) --> (V|N) & (C1|C2)
// iff (C1&C2) == 0 and (N&~C1) == 0
if ((C1->getValue() & C2->getValue()) == 0) {
if (match(A, m_Or(m_Value(V1), m_Value(V2))) &&
((V1 == B && MaskedValueIsZero(V2, ~C1->getValue())) || // (V|N)
(V2 == B && MaskedValueIsZero(V1, ~C1->getValue())))) // (N|V)
return BinaryOperator::CreateAnd(A,
ConstantInt::get(A->getContext(),
C1->getValue()|C2->getValue()));
// Or commutes, try both ways.
if (match(B, m_Or(m_Value(V1), m_Value(V2))) &&
((V1 == A && MaskedValueIsZero(V2, ~C2->getValue())) || // (V|N)
(V2 == A && MaskedValueIsZero(V1, ~C2->getValue())))) // (N|V)
return BinaryOperator::CreateAnd(B,
ConstantInt::get(B->getContext(),
C1->getValue()|C2->getValue()));
}
}
// Check to see if we have any common things being and'ed. If so, find the
// terms for V1 & (V2|V3).
if (Op0->hasOneUse() || Op1->hasOneUse()) {
V1 = 0;
if (A == B) // (A & C)|(A & D) == A & (C|D)
V1 = A, V2 = C, V3 = D;
else if (A == D) // (A & C)|(B & A) == A & (B|C)
V1 = A, V2 = B, V3 = C;
else if (C == B) // (A & C)|(C & D) == C & (A|D)
V1 = C, V2 = A, V3 = D;
else if (C == D) // (A & C)|(B & C) == C & (A|B)
V1 = C, V2 = A, V3 = B;
if (V1) {
Value *Or = Builder->CreateOr(V2, V3, "tmp");
return BinaryOperator::CreateAnd(V1, Or);
}
}
// (A & (C0?-1:0)) | (B & ~(C0?-1:0)) -> C0 ? A : B, and commuted variants
if (Instruction *Match = MatchSelectFromAndOr(A, B, C, D))
return Match;
if (Instruction *Match = MatchSelectFromAndOr(B, A, D, C))
return Match;
if (Instruction *Match = MatchSelectFromAndOr(C, B, A, D))
return Match;
if (Instruction *Match = MatchSelectFromAndOr(D, A, B, C))
return Match;
// ((A&~B)|(~A&B)) -> A^B
if ((match(C, m_Not(m_Specific(D))) &&
match(B, m_Not(m_Specific(A)))))
return BinaryOperator::CreateXor(A, D);
// ((~B&A)|(~A&B)) -> A^B
if ((match(A, m_Not(m_Specific(D))) &&
match(B, m_Not(m_Specific(C)))))
return BinaryOperator::CreateXor(C, D);
// ((A&~B)|(B&~A)) -> A^B
if ((match(C, m_Not(m_Specific(B))) &&
match(D, m_Not(m_Specific(A)))))
return BinaryOperator::CreateXor(A, B);
// ((~B&A)|(B&~A)) -> A^B
if ((match(A, m_Not(m_Specific(B))) &&
match(D, m_Not(m_Specific(C)))))
return BinaryOperator::CreateXor(C, B);
}
// (X >> Z) | (Y >> Z) -> (X|Y) >> Z for all shifts.
if (BinaryOperator *SI1 = dyn_cast<BinaryOperator>(Op1)) {
if (BinaryOperator *SI0 = dyn_cast<BinaryOperator>(Op0))
if (SI0->isShift() && SI0->getOpcode() == SI1->getOpcode() &&
SI0->getOperand(1) == SI1->getOperand(1) &&
(SI0->hasOneUse() || SI1->hasOneUse())) {
Value *NewOp = Builder->CreateOr(SI0->getOperand(0), SI1->getOperand(0),
SI0->getName());
return BinaryOperator::Create(SI1->getOpcode(), NewOp,
SI1->getOperand(1));
}
}
// ((A|B)&1)|(B&-2) -> (A&1) | B
if (match(Op0, m_And(m_Or(m_Value(A), m_Value(B)), m_Value(C))) ||
match(Op0, m_And(m_Value(C), m_Or(m_Value(A), m_Value(B))))) {
Instruction *Ret = FoldOrWithConstants(I, Op1, A, B, C);
if (Ret) return Ret;
}
// (B&-2)|((A|B)&1) -> (A&1) | B
if (match(Op1, m_And(m_Or(m_Value(A), m_Value(B)), m_Value(C))) ||
match(Op1, m_And(m_Value(C), m_Or(m_Value(A), m_Value(B))))) {
Instruction *Ret = FoldOrWithConstants(I, Op0, A, B, C);
if (Ret) return Ret;
}
// (~A | ~B) == (~(A & B)) - De Morgan's Law
if (Value *Op0NotVal = dyn_castNotVal(Op0))
if (Value *Op1NotVal = dyn_castNotVal(Op1))
if (Op0->hasOneUse() && Op1->hasOneUse()) {
Value *And = Builder->CreateAnd(Op0NotVal, Op1NotVal,
I.getName()+".demorgan");
return BinaryOperator::CreateNot(And);
}
if (ICmpInst *RHS = dyn_cast<ICmpInst>(I.getOperand(1)))
if (ICmpInst *LHS = dyn_cast<ICmpInst>(I.getOperand(0)))
if (Instruction *Res = FoldOrOfICmps(I, LHS, RHS))
return Res;
// fold (or (cast A), (cast B)) -> (cast (or A, B))
if (CastInst *Op0C = dyn_cast<CastInst>(Op0)) {
if (CastInst *Op1C = dyn_cast<CastInst>(Op1))
if (Op0C->getOpcode() == Op1C->getOpcode()) {// same cast kind ?
if (!isa<ICmpInst>(Op0C->getOperand(0)) ||
!isa<ICmpInst>(Op1C->getOperand(0))) {
const Type *SrcTy = Op0C->getOperand(0)->getType();
if (SrcTy == Op1C->getOperand(0)->getType() &&
SrcTy->isIntOrIntVector() &&
// Only do this if the casts both really cause code to be
// generated.
ValueRequiresCast(Op0C->getOpcode(), Op0C->getOperand(0),
I.getType()) &&
ValueRequiresCast(Op1C->getOpcode(), Op1C->getOperand(0),
I.getType())) {
Value *NewOp = Builder->CreateOr(Op0C->getOperand(0),
Op1C->getOperand(0), I.getName());
return CastInst::Create(Op0C->getOpcode(), NewOp, I.getType());
}
}
}
}
// (fcmp uno x, c) | (fcmp uno y, c) -> (fcmp uno x, y)
if (FCmpInst *LHS = dyn_cast<FCmpInst>(I.getOperand(0))) {
if (FCmpInst *RHS = dyn_cast<FCmpInst>(I.getOperand(1)))
if (Instruction *Res = FoldOrOfFCmps(I, LHS, RHS))
return Res;
}
return Changed ? &I : 0;
}
Instruction *InstCombiner::visitXor(BinaryOperator &I) {
bool Changed = SimplifyCommutative(I);
Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
if (isa<UndefValue>(Op1)) {
if (isa<UndefValue>(Op0))
// Handle undef ^ undef -> 0 special case. This is a common
// idiom (misuse).
return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
return ReplaceInstUsesWith(I, Op1); // X ^ undef -> undef
}
// xor X, X = 0
if (Op0 == Op1)
return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
// See if we can simplify any instructions used by the instruction whose sole
// purpose is to compute bits we don't care about.
if (SimplifyDemandedInstructionBits(I))
return &I;
if (isa<VectorType>(I.getType()))
if (isa<ConstantAggregateZero>(Op1))
return ReplaceInstUsesWith(I, Op0); // X ^ <0,0> -> X
// Is this a ~ operation?
if (Value *NotOp = dyn_castNotVal(&I)) {
if (BinaryOperator *Op0I = dyn_cast<BinaryOperator>(NotOp)) {
if (Op0I->getOpcode() == Instruction::And ||
Op0I->getOpcode() == Instruction::Or) {
// ~(~X & Y) --> (X | ~Y) - De Morgan's Law
// ~(~X | Y) === (X & ~Y) - De Morgan's Law
if (dyn_castNotVal(Op0I->getOperand(1)))
Op0I->swapOperands();
if (Value *Op0NotVal = dyn_castNotVal(Op0I->getOperand(0))) {
Value *NotY =
Builder->CreateNot(Op0I->getOperand(1),
Op0I->getOperand(1)->getName()+".not");
if (Op0I->getOpcode() == Instruction::And)
return BinaryOperator::CreateOr(Op0NotVal, NotY);
return BinaryOperator::CreateAnd(Op0NotVal, NotY);
}
// ~(X & Y) --> (~X | ~Y) - De Morgan's Law
// ~(X | Y) === (~X & ~Y) - De Morgan's Law
if (isFreeToInvert(Op0I->getOperand(0)) &&
isFreeToInvert(Op0I->getOperand(1))) {
Value *NotX =
Builder->CreateNot(Op0I->getOperand(0), "notlhs");
Value *NotY =
Builder->CreateNot(Op0I->getOperand(1), "notrhs");
if (Op0I->getOpcode() == Instruction::And)
return BinaryOperator::CreateOr(NotX, NotY);
return BinaryOperator::CreateAnd(NotX, NotY);
}
}
}
}
if (ConstantInt *RHS = dyn_cast<ConstantInt>(Op1)) {
if (RHS->isOne() && Op0->hasOneUse()) {
// xor (cmp A, B), true = not (cmp A, B) = !cmp A, B
if (ICmpInst *ICI = dyn_cast<ICmpInst>(Op0))
return new ICmpInst(ICI->getInversePredicate(),
ICI->getOperand(0), ICI->getOperand(1));
if (FCmpInst *FCI = dyn_cast<FCmpInst>(Op0))
return new FCmpInst(FCI->getInversePredicate(),
FCI->getOperand(0), FCI->getOperand(1));
}
// fold (xor(zext(cmp)), 1) and (xor(sext(cmp)), -1) to ext(!cmp).
if (CastInst *Op0C = dyn_cast<CastInst>(Op0)) {
if (CmpInst *CI = dyn_cast<CmpInst>(Op0C->getOperand(0))) {
if (CI->hasOneUse() && Op0C->hasOneUse()) {
Instruction::CastOps Opcode = Op0C->getOpcode();
if ((Opcode == Instruction::ZExt || Opcode == Instruction::SExt) &&
(RHS == ConstantExpr::getCast(Opcode,
ConstantInt::getTrue(I.getContext()),
Op0C->getDestTy()))) {
CI->setPredicate(CI->getInversePredicate());
return CastInst::Create(Opcode, CI, Op0C->getType());
}
}
}
}
if (BinaryOperator *Op0I = dyn_cast<BinaryOperator>(Op0)) {
// ~(c-X) == X-c-1 == X+(-c-1)
if (Op0I->getOpcode() == Instruction::Sub && RHS->isAllOnesValue())
if (Constant *Op0I0C = dyn_cast<Constant>(Op0I->getOperand(0))) {
Constant *NegOp0I0C = ConstantExpr::getNeg(Op0I0C);
Constant *ConstantRHS = ConstantExpr::getSub(NegOp0I0C,
ConstantInt::get(I.getType(), 1));
return BinaryOperator::CreateAdd(Op0I->getOperand(1), ConstantRHS);
}
if (ConstantInt *Op0CI = dyn_cast<ConstantInt>(Op0I->getOperand(1))) {
if (Op0I->getOpcode() == Instruction::Add) {
// ~(X-c) --> (-c-1)-X
if (RHS->isAllOnesValue()) {
Constant *NegOp0CI = ConstantExpr::getNeg(Op0CI);
return BinaryOperator::CreateSub(
ConstantExpr::getSub(NegOp0CI,
ConstantInt::get(I.getType(), 1)),
Op0I->getOperand(0));
} else if (RHS->getValue().isSignBit()) {
// (X + C) ^ signbit -> (X + C + signbit)
Constant *C = ConstantInt::get(I.getContext(),
RHS->getValue() + Op0CI->getValue());
return BinaryOperator::CreateAdd(Op0I->getOperand(0), C);
}
} else if (Op0I->getOpcode() == Instruction::Or) {
// (X|C1)^C2 -> X^(C1|C2) iff X&~C1 == 0
if (MaskedValueIsZero(Op0I->getOperand(0), Op0CI->getValue())) {
Constant *NewRHS = ConstantExpr::getOr(Op0CI, RHS);
// Anything in both C1 and C2 is known to be zero, remove it from
// NewRHS.
Constant *CommonBits = ConstantExpr::getAnd(Op0CI, RHS);
NewRHS = ConstantExpr::getAnd(NewRHS,
ConstantExpr::getNot(CommonBits));
Worklist.Add(Op0I);
I.setOperand(0, Op0I->getOperand(0));
I.setOperand(1, NewRHS);
return &I;
}
}
}
}
// Try to fold constant and into select arguments.
if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
if (Instruction *R = FoldOpIntoSelect(I, SI))
return R;
if (isa<PHINode>(Op0))
if (Instruction *NV = FoldOpIntoPhi(I))
return NV;
}
if (Value *X = dyn_castNotVal(Op0)) // ~A ^ A == -1
if (X == Op1)
return ReplaceInstUsesWith(I, Constant::getAllOnesValue(I.getType()));
if (Value *X = dyn_castNotVal(Op1)) // A ^ ~A == -1
if (X == Op0)
return ReplaceInstUsesWith(I, Constant::getAllOnesValue(I.getType()));
BinaryOperator *Op1I = dyn_cast<BinaryOperator>(Op1);
if (Op1I) {
Value *A, *B;
if (match(Op1I, m_Or(m_Value(A), m_Value(B)))) {
if (A == Op0) { // B^(B|A) == (A|B)^B
Op1I->swapOperands();
I.swapOperands();
std::swap(Op0, Op1);
} else if (B == Op0) { // B^(A|B) == (A|B)^B
I.swapOperands(); // Simplified below.
std::swap(Op0, Op1);
}
} else if (match(Op1I, m_Xor(m_Specific(Op0), m_Value(B)))) {
return ReplaceInstUsesWith(I, B); // A^(A^B) == B
} else if (match(Op1I, m_Xor(m_Value(A), m_Specific(Op0)))) {
return ReplaceInstUsesWith(I, A); // A^(B^A) == B
} else if (match(Op1I, m_And(m_Value(A), m_Value(B))) &&
Op1I->hasOneUse()){
if (A == Op0) { // A^(A&B) -> A^(B&A)
Op1I->swapOperands();
std::swap(A, B);
}
if (B == Op0) { // A^(B&A) -> (B&A)^A
I.swapOperands(); // Simplified below.
std::swap(Op0, Op1);
}
}
}
BinaryOperator *Op0I = dyn_cast<BinaryOperator>(Op0);
if (Op0I) {
Value *A, *B;
if (match(Op0I, m_Or(m_Value(A), m_Value(B))) &&
Op0I->hasOneUse()) {
if (A == Op1) // (B|A)^B == (A|B)^B
std::swap(A, B);
if (B == Op1) // (A|B)^B == A & ~B
return BinaryOperator::CreateAnd(A, Builder->CreateNot(Op1, "tmp"));
} else if (match(Op0I, m_Xor(m_Specific(Op1), m_Value(B)))) {
return ReplaceInstUsesWith(I, B); // (A^B)^A == B
} else if (match(Op0I, m_Xor(m_Value(A), m_Specific(Op1)))) {
return ReplaceInstUsesWith(I, A); // (B^A)^A == B
} else if (match(Op0I, m_And(m_Value(A), m_Value(B))) &&
Op0I->hasOneUse()){
if (A == Op1) // (A&B)^A -> (B&A)^A
std::swap(A, B);
if (B == Op1 && // (B&A)^A == ~B & A
!isa<ConstantInt>(Op1)) { // Canonical form is (B&C)^C
return BinaryOperator::CreateAnd(Builder->CreateNot(A, "tmp"), Op1);
}
}
}
// (X >> Z) ^ (Y >> Z) -> (X^Y) >> Z for all shifts.
if (Op0I && Op1I && Op0I->isShift() &&
Op0I->getOpcode() == Op1I->getOpcode() &&
Op0I->getOperand(1) == Op1I->getOperand(1) &&
(Op1I->hasOneUse() || Op1I->hasOneUse())) {
Value *NewOp =
Builder->CreateXor(Op0I->getOperand(0), Op1I->getOperand(0),
Op0I->getName());
return BinaryOperator::Create(Op1I->getOpcode(), NewOp,
Op1I->getOperand(1));
}
if (Op0I && Op1I) {
Value *A, *B, *C, *D;
// (A & B)^(A | B) -> A ^ B
if (match(Op0I, m_And(m_Value(A), m_Value(B))) &&
match(Op1I, m_Or(m_Value(C), m_Value(D)))) {
if ((A == C && B == D) || (A == D && B == C))
return BinaryOperator::CreateXor(A, B);
}
// (A | B)^(A & B) -> A ^ B
if (match(Op0I, m_Or(m_Value(A), m_Value(B))) &&
match(Op1I, m_And(m_Value(C), m_Value(D)))) {
if ((A == C && B == D) || (A == D && B == C))
return BinaryOperator::CreateXor(A, B);
}
// (A & B)^(C & D)
if ((Op0I->hasOneUse() || Op1I->hasOneUse()) &&
match(Op0I, m_And(m_Value(A), m_Value(B))) &&
match(Op1I, m_And(m_Value(C), m_Value(D)))) {
// (X & Y)^(X & Y) -> (Y^Z) & X
Value *X = 0, *Y = 0, *Z = 0;
if (A == C)
X = A, Y = B, Z = D;
else if (A == D)
X = A, Y = B, Z = C;
else if (B == C)
X = B, Y = A, Z = D;
else if (B == D)
X = B, Y = A, Z = C;
if (X) {
Value *NewOp = Builder->CreateXor(Y, Z, Op0->getName());
return BinaryOperator::CreateAnd(NewOp, X);
}
}
}
// (icmp1 A, B) ^ (icmp2 A, B) --> (icmp3 A, B)
if (ICmpInst *RHS = dyn_cast<ICmpInst>(I.getOperand(1)))
if (ICmpInst *LHS = dyn_cast<ICmpInst>(I.getOperand(0)))
if (PredicatesFoldable(LHS->getPredicate(), RHS->getPredicate())) {
if (LHS->getOperand(0) == RHS->getOperand(1) &&
LHS->getOperand(1) == RHS->getOperand(0))
LHS->swapOperands();
if (LHS->getOperand(0) == RHS->getOperand(0) &&
LHS->getOperand(1) == RHS->getOperand(1)) {
Value *Op0 = LHS->getOperand(0), *Op1 = LHS->getOperand(1);
unsigned Code = getICmpCode(LHS) ^ getICmpCode(RHS);
bool isSigned = LHS->isSigned() || RHS->isSigned();
Value *RV = getICmpValue(isSigned, Code, Op0, Op1);
if (Instruction *I = dyn_cast<Instruction>(RV))
return I;
// Otherwise, it's a constant boolean value.
return ReplaceInstUsesWith(I, RV);
}
}
// fold (xor (cast A), (cast B)) -> (cast (xor A, B))
if (CastInst *Op0C = dyn_cast<CastInst>(Op0)) {
if (CastInst *Op1C = dyn_cast<CastInst>(Op1))
if (Op0C->getOpcode() == Op1C->getOpcode()) { // same cast kind?
const Type *SrcTy = Op0C->getOperand(0)->getType();
if (SrcTy == Op1C->getOperand(0)->getType() && SrcTy->isInteger() &&
// Only do this if the casts both really cause code to be generated.
ValueRequiresCast(Op0C->getOpcode(), Op0C->getOperand(0),
I.getType()) &&
ValueRequiresCast(Op1C->getOpcode(), Op1C->getOperand(0),
I.getType())) {
Value *NewOp = Builder->CreateXor(Op0C->getOperand(0),
Op1C->getOperand(0), I.getName());
return CastInst::Create(Op0C->getOpcode(), NewOp, I.getType());
}
}
}
return Changed ? &I : 0;
}
Instruction *InstCombiner::visitShl(BinaryOperator &I) {
return commonShiftTransforms(I);
}
Instruction *InstCombiner::visitLShr(BinaryOperator &I) {
return commonShiftTransforms(I);
}
Instruction *InstCombiner::visitAShr(BinaryOperator &I) {
if (Instruction *R = commonShiftTransforms(I))
return R;
Value *Op0 = I.getOperand(0);
// ashr int -1, X = -1 (for any arithmetic shift rights of ~0)
if (ConstantInt *CSI = dyn_cast<ConstantInt>(Op0))
if (CSI->isAllOnesValue())
return ReplaceInstUsesWith(I, CSI);
// See if we can turn a signed shr into an unsigned shr.
if (MaskedValueIsZero(Op0,
APInt::getSignBit(I.getType()->getScalarSizeInBits())))
return BinaryOperator::CreateLShr(Op0, I.getOperand(1));
// Arithmetic shifting an all-sign-bit value is a no-op.
unsigned NumSignBits = ComputeNumSignBits(Op0);
if (NumSignBits == Op0->getType()->getScalarSizeInBits())
return ReplaceInstUsesWith(I, Op0);
return 0;
}
Instruction *InstCombiner::commonShiftTransforms(BinaryOperator &I) {
assert(I.getOperand(1)->getType() == I.getOperand(0)->getType());
Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
// shl X, 0 == X and shr X, 0 == X
// shl 0, X == 0 and shr 0, X == 0
if (Op1 == Constant::getNullValue(Op1->getType()) ||
Op0 == Constant::getNullValue(Op0->getType()))
return ReplaceInstUsesWith(I, Op0);
if (isa<UndefValue>(Op0)) {
if (I.getOpcode() == Instruction::AShr) // undef >>s X -> undef
return ReplaceInstUsesWith(I, Op0);
else // undef << X -> 0, undef >>u X -> 0
return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
}
if (isa<UndefValue>(Op1)) {
if (I.getOpcode() == Instruction::AShr) // X >>s undef -> X
return ReplaceInstUsesWith(I, Op0);
else // X << undef, X >>u undef -> 0
return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
}
// See if we can fold away this shift.
if (SimplifyDemandedInstructionBits(I))
return &I;
// Try to fold constant and into select arguments.
if (isa<Constant>(Op0))
if (SelectInst *SI = dyn_cast<SelectInst>(Op1))
if (Instruction *R = FoldOpIntoSelect(I, SI))
return R;
if (ConstantInt *CUI = dyn_cast<ConstantInt>(Op1))
if (Instruction *Res = FoldShiftByConstant(Op0, CUI, I))
return Res;
return 0;
}
Instruction *InstCombiner::FoldShiftByConstant(Value *Op0, ConstantInt *Op1,
BinaryOperator &I) {
bool isLeftShift = I.getOpcode() == Instruction::Shl;
// See if we can simplify any instructions used by the instruction whose sole
// purpose is to compute bits we don't care about.
uint32_t TypeBits = Op0->getType()->getScalarSizeInBits();
// shl i32 X, 32 = 0 and srl i8 Y, 9 = 0, ... just don't eliminate
// a signed shift.
//
if (Op1->uge(TypeBits)) {
if (I.getOpcode() != Instruction::AShr)
return ReplaceInstUsesWith(I, Constant::getNullValue(Op0->getType()));
else {
I.setOperand(1, ConstantInt::get(I.getType(), TypeBits-1));
return &I;
}
}
// ((X*C1) << C2) == (X * (C1 << C2))
if (BinaryOperator *BO = dyn_cast<BinaryOperator>(Op0))
if (BO->getOpcode() == Instruction::Mul && isLeftShift)
if (Constant *BOOp = dyn_cast<Constant>(BO->getOperand(1)))
return BinaryOperator::CreateMul(BO->getOperand(0),
ConstantExpr::getShl(BOOp, Op1));
// Try to fold constant and into select arguments.
if (SelectInst *SI = dyn_cast<SelectInst>(Op0))
if (Instruction *R = FoldOpIntoSelect(I, SI))
return R;
if (isa<PHINode>(Op0))
if (Instruction *NV = FoldOpIntoPhi(I))
return NV;
// Fold shift2(trunc(shift1(x,c1)), c2) -> trunc(shift2(shift1(x,c1),c2))
if (TruncInst *TI = dyn_cast<TruncInst>(Op0)) {
Instruction *TrOp = dyn_cast<Instruction>(TI->getOperand(0));
// If 'shift2' is an ashr, we would have to get the sign bit into a funny
// place. Don't try to do this transformation in this case. Also, we
// require that the input operand is a shift-by-constant so that we have
// confidence that the shifts will get folded together. We could do this
// xform in more cases, but it is unlikely to be profitable.
if (TrOp && I.isLogicalShift() && TrOp->isShift() &&
isa<ConstantInt>(TrOp->getOperand(1))) {
// Okay, we'll do this xform. Make the shift of shift.
Constant *ShAmt = ConstantExpr::getZExt(Op1, TrOp->getType());
// (shift2 (shift1 & 0x00FF), c2)
Value *NSh = Builder->CreateBinOp(I.getOpcode(), TrOp, ShAmt,I.getName());
// For logical shifts, the truncation has the effect of making the high
// part of the register be zeros. Emulate this by inserting an AND to
// clear the top bits as needed. This 'and' will usually be zapped by
// other xforms later if dead.
unsigned SrcSize = TrOp->getType()->getScalarSizeInBits();
unsigned DstSize = TI->getType()->getScalarSizeInBits();
APInt MaskV(APInt::getLowBitsSet(SrcSize, DstSize));
// The mask we constructed says what the trunc would do if occurring
// between the shifts. We want to know the effect *after* the second
// shift. We know that it is a logical shift by a constant, so adjust the
// mask as appropriate.
if (I.getOpcode() == Instruction::Shl)
MaskV <<= Op1->getZExtValue();
else {
assert(I.getOpcode() == Instruction::LShr && "Unknown logical shift");
MaskV = MaskV.lshr(Op1->getZExtValue());
}
// shift1 & 0x00FF
Value *And = Builder->CreateAnd(NSh,
ConstantInt::get(I.getContext(), MaskV),
TI->getName());
// Return the value truncated to the interesting size.
return new TruncInst(And, I.getType());
}
}
if (Op0->hasOneUse()) {
if (BinaryOperator *Op0BO = dyn_cast<BinaryOperator>(Op0)) {
// Turn ((X >> C) + Y) << C -> (X + (Y << C)) & (~0 << C)
Value *V1, *V2;
ConstantInt *CC;
switch (Op0BO->getOpcode()) {
default: break;
case Instruction::Add:
case Instruction::And:
case Instruction::Or:
case Instruction::Xor: {
// These operators commute.
// Turn (Y + (X >> C)) << C -> (X + (Y << C)) & (~0 << C)
if (isLeftShift && Op0BO->getOperand(1)->hasOneUse() &&
match(Op0BO->getOperand(1), m_Shr(m_Value(V1),
m_Specific(Op1)))) {
Value *YS = // (Y << C)
Builder->CreateShl(Op0BO->getOperand(0), Op1, Op0BO->getName());
// (X + (Y << C))
Value *X = Builder->CreateBinOp(Op0BO->getOpcode(), YS, V1,
Op0BO->getOperand(1)->getName());
uint32_t Op1Val = Op1->getLimitedValue(TypeBits);
return BinaryOperator::CreateAnd(X, ConstantInt::get(I.getContext(),
APInt::getHighBitsSet(TypeBits, TypeBits-Op1Val)));
}
// Turn (Y + ((X >> C) & CC)) << C -> ((X & (CC << C)) + (Y << C))
Value *Op0BOOp1 = Op0BO->getOperand(1);
if (isLeftShift && Op0BOOp1->hasOneUse() &&
match(Op0BOOp1,
m_And(m_Shr(m_Value(V1), m_Specific(Op1)),
m_ConstantInt(CC))) &&
cast<BinaryOperator>(Op0BOOp1)->getOperand(0)->hasOneUse()) {
Value *YS = // (Y << C)
Builder->CreateShl(Op0BO->getOperand(0), Op1,
Op0BO->getName());
// X & (CC << C)
Value *XM = Builder->CreateAnd(V1, ConstantExpr::getShl(CC, Op1),
V1->getName()+".mask");
return BinaryOperator::Create(Op0BO->getOpcode(), YS, XM);
}
}
// FALL THROUGH.
case Instruction::Sub: {
// Turn ((X >> C) + Y) << C -> (X + (Y << C)) & (~0 << C)
if (isLeftShift && Op0BO->getOperand(0)->hasOneUse() &&
match(Op0BO->getOperand(0), m_Shr(m_Value(V1),
m_Specific(Op1)))) {
Value *YS = // (Y << C)
Builder->CreateShl(Op0BO->getOperand(1), Op1, Op0BO->getName());
// (X + (Y << C))
Value *X = Builder->CreateBinOp(Op0BO->getOpcode(), V1, YS,
Op0BO->getOperand(0)->getName());
uint32_t Op1Val = Op1->getLimitedValue(TypeBits);
return BinaryOperator::CreateAnd(X, ConstantInt::get(I.getContext(),
APInt::getHighBitsSet(TypeBits, TypeBits-Op1Val)));
}
// Turn (((X >> C)&CC) + Y) << C -> (X + (Y << C)) & (CC << C)
if (isLeftShift && Op0BO->getOperand(0)->hasOneUse() &&
match(Op0BO->getOperand(0),
m_And(m_Shr(m_Value(V1), m_Value(V2)),
m_ConstantInt(CC))) && V2 == Op1 &&
cast<BinaryOperator>(Op0BO->getOperand(0))
->getOperand(0)->hasOneUse()) {
Value *YS = // (Y << C)
Builder->CreateShl(Op0BO->getOperand(1), Op1, Op0BO->getName());
// X & (CC << C)
Value *XM = Builder->CreateAnd(V1, ConstantExpr::getShl(CC, Op1),
V1->getName()+".mask");
return BinaryOperator::Create(Op0BO->getOpcode(), XM, YS);
}
break;
}
}
// If the operand is an bitwise operator with a constant RHS, and the
// shift is the only use, we can pull it out of the shift.
if (ConstantInt *Op0C = dyn_cast<ConstantInt>(Op0BO->getOperand(1))) {
bool isValid = true; // Valid only for And, Or, Xor
bool highBitSet = false; // Transform if high bit of constant set?
switch (Op0BO->getOpcode()) {
default: isValid = false; break; // Do not perform transform!
case Instruction::Add:
isValid = isLeftShift;
break;
case Instruction::Or:
case Instruction::Xor:
highBitSet = false;
break;
case Instruction::And:
highBitSet = true;
break;
}
// If this is a signed shift right, and the high bit is modified
// by the logical operation, do not perform the transformation.
// The highBitSet boolean indicates the value of the high bit of
// the constant which would cause it to be modified for this
// operation.
//
if (isValid && I.getOpcode() == Instruction::AShr)
isValid = Op0C->getValue()[TypeBits-1] == highBitSet;
if (isValid) {
Constant *NewRHS = ConstantExpr::get(I.getOpcode(), Op0C, Op1);
Value *NewShift =
Builder->CreateBinOp(I.getOpcode(), Op0BO->getOperand(0), Op1);
NewShift->takeName(Op0BO);
return BinaryOperator::Create(Op0BO->getOpcode(), NewShift,
NewRHS);
}
}
}
}
// Find out if this is a shift of a shift by a constant.
BinaryOperator *ShiftOp = dyn_cast<BinaryOperator>(Op0);
if (ShiftOp && !ShiftOp->isShift())
ShiftOp = 0;
if (ShiftOp && isa<ConstantInt>(ShiftOp->getOperand(1))) {
ConstantInt *ShiftAmt1C = cast<ConstantInt>(ShiftOp->getOperand(1));
uint32_t ShiftAmt1 = ShiftAmt1C->getLimitedValue(TypeBits);
uint32_t ShiftAmt2 = Op1->getLimitedValue(TypeBits);
assert(ShiftAmt2 != 0 && "Should have been simplified earlier");
if (ShiftAmt1 == 0) return 0; // Will be simplified in the future.
Value *X = ShiftOp->getOperand(0);
uint32_t AmtSum = ShiftAmt1+ShiftAmt2; // Fold into one big shift.
const IntegerType *Ty = cast<IntegerType>(I.getType());
// Check for (X << c1) << c2 and (X >> c1) >> c2
if (I.getOpcode() == ShiftOp->getOpcode()) {
// If this is oversized composite shift, then unsigned shifts get 0, ashr
// saturates.
if (AmtSum >= TypeBits) {
if (I.getOpcode() != Instruction::AShr)
return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
AmtSum = TypeBits-1; // Saturate to 31 for i32 ashr.
}
return BinaryOperator::Create(I.getOpcode(), X,
ConstantInt::get(Ty, AmtSum));
}
if (ShiftOp->getOpcode() == Instruction::LShr &&
I.getOpcode() == Instruction::AShr) {
if (AmtSum >= TypeBits)
return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
// ((X >>u C1) >>s C2) -> (X >>u (C1+C2)) since C1 != 0.
return BinaryOperator::CreateLShr(X, ConstantInt::get(Ty, AmtSum));
}
if (ShiftOp->getOpcode() == Instruction::AShr &&
I.getOpcode() == Instruction::LShr) {
// ((X >>s C1) >>u C2) -> ((X >>s (C1+C2)) & mask) since C1 != 0.
if (AmtSum >= TypeBits)
AmtSum = TypeBits-1;
Value *Shift = Builder->CreateAShr(X, ConstantInt::get(Ty, AmtSum));
APInt Mask(APInt::getLowBitsSet(TypeBits, TypeBits - ShiftAmt2));
return BinaryOperator::CreateAnd(Shift,
ConstantInt::get(I.getContext(), Mask));
}
// Okay, if we get here, one shift must be left, and the other shift must be
// right. See if the amounts are equal.
if (ShiftAmt1 == ShiftAmt2) {
// If we have ((X >>? C) << C), turn this into X & (-1 << C).
if (I.getOpcode() == Instruction::Shl) {
APInt Mask(APInt::getHighBitsSet(TypeBits, TypeBits - ShiftAmt1));
return BinaryOperator::CreateAnd(X,
ConstantInt::get(I.getContext(),Mask));
}
// If we have ((X << C) >>u C), turn this into X & (-1 >>u C).
if (I.getOpcode() == Instruction::LShr) {
APInt Mask(APInt::getLowBitsSet(TypeBits, TypeBits - ShiftAmt1));
return BinaryOperator::CreateAnd(X,
ConstantInt::get(I.getContext(), Mask));
}
// We can simplify ((X << C) >>s C) into a trunc + sext.
// NOTE: we could do this for any C, but that would make 'unusual' integer
// types. For now, just stick to ones well-supported by the code
// generators.
const Type *SExtType = 0;
switch (Ty->getBitWidth() - ShiftAmt1) {
case 1 :
case 8 :
case 16 :
case 32 :
case 64 :
case 128:
SExtType = IntegerType::get(I.getContext(),
Ty->getBitWidth() - ShiftAmt1);
break;
default: break;
}
if (SExtType)
return new SExtInst(Builder->CreateTrunc(X, SExtType, "sext"), Ty);
// Otherwise, we can't handle it yet.
} else if (ShiftAmt1 < ShiftAmt2) {
uint32_t ShiftDiff = ShiftAmt2-ShiftAmt1;
// (X >>? C1) << C2 --> X << (C2-C1) & (-1 << C2)
if (I.getOpcode() == Instruction::Shl) {
assert(ShiftOp->getOpcode() == Instruction::LShr ||
ShiftOp->getOpcode() == Instruction::AShr);
Value *Shift = Builder->CreateShl(X, ConstantInt::get(Ty, ShiftDiff));
APInt Mask(APInt::getHighBitsSet(TypeBits, TypeBits - ShiftAmt2));
return BinaryOperator::CreateAnd(Shift,
ConstantInt::get(I.getContext(),Mask));
}
// (X << C1) >>u C2 --> X >>u (C2-C1) & (-1 >> C2)
if (I.getOpcode() == Instruction::LShr) {
assert(ShiftOp->getOpcode() == Instruction::Shl);
Value *Shift = Builder->CreateLShr(X, ConstantInt::get(Ty, ShiftDiff));
APInt Mask(APInt::getLowBitsSet(TypeBits, TypeBits - ShiftAmt2));
return BinaryOperator::CreateAnd(Shift,
ConstantInt::get(I.getContext(),Mask));
}
// We can't handle (X << C1) >>s C2, it shifts arbitrary bits in.
} else {
assert(ShiftAmt2 < ShiftAmt1);
uint32_t ShiftDiff = ShiftAmt1-ShiftAmt2;
// (X >>? C1) << C2 --> X >>? (C1-C2) & (-1 << C2)
if (I.getOpcode() == Instruction::Shl) {
assert(ShiftOp->getOpcode() == Instruction::LShr ||
ShiftOp->getOpcode() == Instruction::AShr);
Value *Shift = Builder->CreateBinOp(ShiftOp->getOpcode(), X,
ConstantInt::get(Ty, ShiftDiff));
APInt Mask(APInt::getHighBitsSet(TypeBits, TypeBits - ShiftAmt2));
return BinaryOperator::CreateAnd(Shift,
ConstantInt::get(I.getContext(),Mask));
}
// (X << C1) >>u C2 --> X << (C1-C2) & (-1 >> C2)
if (I.getOpcode() == Instruction::LShr) {
assert(ShiftOp->getOpcode() == Instruction::Shl);
Value *Shift = Builder->CreateShl(X, ConstantInt::get(Ty, ShiftDiff));
APInt Mask(APInt::getLowBitsSet(TypeBits, TypeBits - ShiftAmt2));
return BinaryOperator::CreateAnd(Shift,
ConstantInt::get(I.getContext(),Mask));
}
// We can't handle (X << C1) >>a C2, it shifts arbitrary bits in.
}
}
return 0;
}
/// FindElementAtOffset - Given a type and a constant offset, determine whether
/// or not there is a sequence of GEP indices into the type that will land us at
/// the specified offset. If so, fill them into NewIndices and return the
/// resultant element type, otherwise return null.
const Type *InstCombiner::FindElementAtOffset(const Type *Ty, int64_t Offset,
SmallVectorImpl<Value*> &NewIndices) {
if (!TD) return 0;
if (!Ty->isSized()) return 0;
// Start with the index over the outer type. Note that the type size
// might be zero (even if the offset isn't zero) if the indexed type
// is something like [0 x {int, int}]
const Type *IntPtrTy = TD->getIntPtrType(Ty->getContext());
int64_t FirstIdx = 0;
if (int64_t TySize = TD->getTypeAllocSize(Ty)) {
FirstIdx = Offset/TySize;
Offset -= FirstIdx*TySize;
// Handle hosts where % returns negative instead of values [0..TySize).
if (Offset < 0) {
--FirstIdx;
Offset += TySize;
assert(Offset >= 0);
}
assert((uint64_t)Offset < (uint64_t)TySize && "Out of range offset");
}
NewIndices.push_back(ConstantInt::get(IntPtrTy, FirstIdx));
// Index into the types. If we fail, set OrigBase to null.
while (Offset) {
// Indexing into tail padding between struct/array elements.
if (uint64_t(Offset*8) >= TD->getTypeSizeInBits(Ty))
return 0;
if (const StructType *STy = dyn_cast<StructType>(Ty)) {
const StructLayout *SL = TD->getStructLayout(STy);
assert(Offset < (int64_t)SL->getSizeInBytes() &&
"Offset must stay within the indexed type");
unsigned Elt = SL->getElementContainingOffset(Offset);
NewIndices.push_back(ConstantInt::get(Type::getInt32Ty(Ty->getContext()),
Elt));
Offset -= SL->getElementOffset(Elt);
Ty = STy->getElementType(Elt);
} else if (const ArrayType *AT = dyn_cast<ArrayType>(Ty)) {
uint64_t EltSize = TD->getTypeAllocSize(AT->getElementType());
assert(EltSize && "Cannot index into a zero-sized array");
NewIndices.push_back(ConstantInt::get(IntPtrTy,Offset/EltSize));
Offset %= EltSize;
Ty = AT->getElementType();
} else {
// Otherwise, we can't index into the middle of this atomic type, bail.
return 0;
}
}
return Ty;
}
Instruction *InstCombiner::visitGetElementPtrInst(GetElementPtrInst &GEP) {
SmallVector<Value*, 8> Ops(GEP.op_begin(), GEP.op_end());
if (Value *V = SimplifyGEPInst(&Ops[0], Ops.size(), TD))
return ReplaceInstUsesWith(GEP, V);
Value *PtrOp = GEP.getOperand(0);
if (isa<UndefValue>(GEP.getOperand(0)))
return ReplaceInstUsesWith(GEP, UndefValue::get(GEP.getType()));
// Eliminate unneeded casts for indices.
if (TD) {
bool MadeChange = false;
unsigned PtrSize = TD->getPointerSizeInBits();
gep_type_iterator GTI = gep_type_begin(GEP);
for (User::op_iterator I = GEP.op_begin() + 1, E = GEP.op_end();
I != E; ++I, ++GTI) {
if (!isa<SequentialType>(*GTI)) continue;
// If we are using a wider index than needed for this platform, shrink it
// to what we need. If narrower, sign-extend it to what we need. This
// explicit cast can make subsequent optimizations more obvious.
unsigned OpBits = cast<IntegerType>((*I)->getType())->getBitWidth();
if (OpBits == PtrSize)
continue;
*I = Builder->CreateIntCast(*I, TD->getIntPtrType(GEP.getContext()),true);
MadeChange = true;
}
if (MadeChange) return &GEP;
}
// Combine Indices - If the source pointer to this getelementptr instruction
// is a getelementptr instruction, combine the indices of the two
// getelementptr instructions into a single instruction.
//
if (GEPOperator *Src = dyn_cast<GEPOperator>(PtrOp)) {
// Note that if our source is a gep chain itself that we wait for that
// chain to be resolved before we perform this transformation. This
// avoids us creating a TON of code in some cases.
//
if (GetElementPtrInst *SrcGEP =
dyn_cast<GetElementPtrInst>(Src->getOperand(0)))
if (SrcGEP->getNumOperands() == 2)
return 0; // Wait until our source is folded to completion.
SmallVector<Value*, 8> Indices;
// Find out whether the last index in the source GEP is a sequential idx.
bool EndsWithSequential = false;
for (gep_type_iterator I = gep_type_begin(*Src), E = gep_type_end(*Src);
I != E; ++I)
EndsWithSequential = !isa<StructType>(*I);
// Can we combine the two pointer arithmetics offsets?
if (EndsWithSequential) {
// Replace: gep (gep %P, long B), long A, ...
// With: T = long A+B; gep %P, T, ...
//
Value *Sum;
Value *SO1 = Src->getOperand(Src->getNumOperands()-1);
Value *GO1 = GEP.getOperand(1);
if (SO1 == Constant::getNullValue(SO1->getType())) {
Sum = GO1;
} else if (GO1 == Constant::getNullValue(GO1->getType())) {
Sum = SO1;
} else {
// If they aren't the same type, then the input hasn't been processed
// by the loop above yet (which canonicalizes sequential index types to
// intptr_t). Just avoid transforming this until the input has been
// normalized.
if (SO1->getType() != GO1->getType())
return 0;
Sum = Builder->CreateAdd(SO1, GO1, PtrOp->getName()+".sum");
}
// Update the GEP in place if possible.
if (Src->getNumOperands() == 2) {
GEP.setOperand(0, Src->getOperand(0));
GEP.setOperand(1, Sum);
return &GEP;
}
Indices.append(Src->op_begin()+1, Src->op_end()-1);
Indices.push_back(Sum);
Indices.append(GEP.op_begin()+2, GEP.op_end());
} else if (isa<Constant>(*GEP.idx_begin()) &&
cast<Constant>(*GEP.idx_begin())->isNullValue() &&
Src->getNumOperands() != 1) {
// Otherwise we can do the fold if the first index of the GEP is a zero
Indices.append(Src->op_begin()+1, Src->op_end());
Indices.append(GEP.idx_begin()+1, GEP.idx_end());
}
if (!Indices.empty())
return (cast<GEPOperator>(&GEP)->isInBounds() &&
Src->isInBounds()) ?
GetElementPtrInst::CreateInBounds(Src->getOperand(0), Indices.begin(),
Indices.end(), GEP.getName()) :
GetElementPtrInst::Create(Src->getOperand(0), Indices.begin(),
Indices.end(), GEP.getName());
}
// Handle gep(bitcast x) and gep(gep x, 0, 0, 0).
if (Value *X = getBitCastOperand(PtrOp)) {
assert(isa<PointerType>(X->getType()) && "Must be cast from pointer");
// If the input bitcast is actually "bitcast(bitcast(x))", then we don't
// want to change the gep until the bitcasts are eliminated.
if (getBitCastOperand(X)) {
Worklist.AddValue(PtrOp);
return 0;
}
bool HasZeroPointerIndex = false;
if (ConstantInt *C = dyn_cast<ConstantInt>(GEP.getOperand(1)))
HasZeroPointerIndex = C->isZero();
// Transform: GEP (bitcast [10 x i8]* X to [0 x i8]*), i32 0, ...
// into : GEP [10 x i8]* X, i32 0, ...
//
// Likewise, transform: GEP (bitcast i8* X to [0 x i8]*), i32 0, ...
// into : GEP i8* X, ...
//
// This occurs when the program declares an array extern like "int X[];"
if (HasZeroPointerIndex) {
const PointerType *CPTy = cast<PointerType>(PtrOp->getType());
const PointerType *XTy = cast<PointerType>(X->getType());
if (const ArrayType *CATy =
dyn_cast<ArrayType>(CPTy->getElementType())) {
// GEP (bitcast i8* X to [0 x i8]*), i32 0, ... ?
if (CATy->getElementType() == XTy->getElementType()) {
// -> GEP i8* X, ...
SmallVector<Value*, 8> Indices(GEP.idx_begin()+1, GEP.idx_end());
return cast<GEPOperator>(&GEP)->isInBounds() ?
GetElementPtrInst::CreateInBounds(X, Indices.begin(), Indices.end(),
GEP.getName()) :
GetElementPtrInst::Create(X, Indices.begin(), Indices.end(),
GEP.getName());
}
if (const ArrayType *XATy = dyn_cast<ArrayType>(XTy->getElementType())){
// GEP (bitcast [10 x i8]* X to [0 x i8]*), i32 0, ... ?
if (CATy->getElementType() == XATy->getElementType()) {
// -> GEP [10 x i8]* X, i32 0, ...
// At this point, we know that the cast source type is a pointer
// to an array of the same type as the destination pointer
// array. Because the array type is never stepped over (there
// is a leading zero) we can fold the cast into this GEP.
GEP.setOperand(0, X);
return &GEP;
}
}
}
} else if (GEP.getNumOperands() == 2) {
// Transform things like:
// %t = getelementptr i32* bitcast ([2 x i32]* %str to i32*), i32 %V
// into: %t1 = getelementptr [2 x i32]* %str, i32 0, i32 %V; bitcast
const Type *SrcElTy = cast<PointerType>(X->getType())->getElementType();
const Type *ResElTy=cast<PointerType>(PtrOp->getType())->getElementType();
if (TD && isa<ArrayType>(SrcElTy) &&
TD->getTypeAllocSize(cast<ArrayType>(SrcElTy)->getElementType()) ==
TD->getTypeAllocSize(ResElTy)) {
Value *Idx[2];
Idx[0] = Constant::getNullValue(Type::getInt32Ty(GEP.getContext()));
Idx[1] = GEP.getOperand(1);
Value *NewGEP = cast<GEPOperator>(&GEP)->isInBounds() ?
Builder->CreateInBoundsGEP(X, Idx, Idx + 2, GEP.getName()) :
Builder->CreateGEP(X, Idx, Idx + 2, GEP.getName());
// V and GEP are both pointer types --> BitCast
return new BitCastInst(NewGEP, GEP.getType());
}
// Transform things like:
// getelementptr i8* bitcast ([100 x double]* X to i8*), i32 %tmp
// (where tmp = 8*tmp2) into:
// getelementptr [100 x double]* %arr, i32 0, i32 %tmp2; bitcast
if (TD && isa<ArrayType>(SrcElTy) &&
ResElTy == Type::getInt8Ty(GEP.getContext())) {
uint64_t ArrayEltSize =
TD->getTypeAllocSize(cast<ArrayType>(SrcElTy)->getElementType());
// Check to see if "tmp" is a scale by a multiple of ArrayEltSize. We
// allow either a mul, shift, or constant here.
Value *NewIdx = 0;
ConstantInt *Scale = 0;
if (ArrayEltSize == 1) {
NewIdx = GEP.getOperand(1);
Scale = ConstantInt::get(cast<IntegerType>(NewIdx->getType()), 1);
} else if (ConstantInt *CI = dyn_cast<ConstantInt>(GEP.getOperand(1))) {
NewIdx = ConstantInt::get(CI->getType(), 1);
Scale = CI;
} else if (Instruction *Inst =dyn_cast<Instruction>(GEP.getOperand(1))){
if (Inst->getOpcode() == Instruction::Shl &&
isa<ConstantInt>(Inst->getOperand(1))) {
ConstantInt *ShAmt = cast<ConstantInt>(Inst->getOperand(1));
uint32_t ShAmtVal = ShAmt->getLimitedValue(64);
Scale = ConstantInt::get(cast<IntegerType>(Inst->getType()),
1ULL << ShAmtVal);
NewIdx = Inst->getOperand(0);
} else if (Inst->getOpcode() == Instruction::Mul &&
isa<ConstantInt>(Inst->getOperand(1))) {
Scale = cast<ConstantInt>(Inst->getOperand(1));
NewIdx = Inst->getOperand(0);
}
}
// If the index will be to exactly the right offset with the scale taken
// out, perform the transformation. Note, we don't know whether Scale is
// signed or not. We'll use unsigned version of division/modulo
// operation after making sure Scale doesn't have the sign bit set.
if (ArrayEltSize && Scale && Scale->getSExtValue() >= 0LL &&
Scale->getZExtValue() % ArrayEltSize == 0) {
Scale = ConstantInt::get(Scale->getType(),
Scale->getZExtValue() / ArrayEltSize);
if (Scale->getZExtValue() != 1) {
Constant *C = ConstantExpr::getIntegerCast(Scale, NewIdx->getType(),
false /*ZExt*/);
NewIdx = Builder->CreateMul(NewIdx, C, "idxscale");
}
// Insert the new GEP instruction.
Value *Idx[2];
Idx[0] = Constant::getNullValue(Type::getInt32Ty(GEP.getContext()));
Idx[1] = NewIdx;
Value *NewGEP = cast<GEPOperator>(&GEP)->isInBounds() ?
Builder->CreateInBoundsGEP(X, Idx, Idx + 2, GEP.getName()) :
Builder->CreateGEP(X, Idx, Idx + 2, GEP.getName());
// The NewGEP must be pointer typed, so must the old one -> BitCast
return new BitCastInst(NewGEP, GEP.getType());
}
}
}
}
/// See if we can simplify:
/// X = bitcast A* to B*
/// Y = gep X, <...constant indices...>
/// into a gep of the original struct. This is important for SROA and alias
/// analysis of unions. If "A" is also a bitcast, wait for A/X to be merged.
if (BitCastInst *BCI = dyn_cast<BitCastInst>(PtrOp)) {
if (TD &&
!isa<BitCastInst>(BCI->getOperand(0)) && GEP.hasAllConstantIndices()) {
// Determine how much the GEP moves the pointer. We are guaranteed to get
// a constant back from EmitGEPOffset.
ConstantInt *OffsetV = cast<ConstantInt>(EmitGEPOffset(&GEP));
int64_t Offset = OffsetV->getSExtValue();
// If this GEP instruction doesn't move the pointer, just replace the GEP
// with a bitcast of the real input to the dest type.
if (Offset == 0) {
// If the bitcast is of an allocation, and the allocation will be
// converted to match the type of the cast, don't touch this.
if (isa<AllocaInst>(BCI->getOperand(0)) ||
isMalloc(BCI->getOperand(0))) {
// See if the bitcast simplifies, if so, don't nuke this GEP yet.
if (Instruction *I = visitBitCast(*BCI)) {
if (I != BCI) {
I->takeName(BCI);
BCI->getParent()->getInstList().insert(BCI, I);
ReplaceInstUsesWith(*BCI, I);
}
return &GEP;
}
}
return new BitCastInst(BCI->getOperand(0), GEP.getType());
}
// Otherwise, if the offset is non-zero, we need to find out if there is a
// field at Offset in 'A's type. If so, we can pull the cast through the
// GEP.
SmallVector<Value*, 8> NewIndices;
const Type *InTy =
cast<PointerType>(BCI->getOperand(0)->getType())->getElementType();
if (FindElementAtOffset(InTy, Offset, NewIndices)) {
Value *NGEP = cast<GEPOperator>(&GEP)->isInBounds() ?
Builder->CreateInBoundsGEP(BCI->getOperand(0), NewIndices.begin(),
NewIndices.end()) :
Builder->CreateGEP(BCI->getOperand(0), NewIndices.begin(),
NewIndices.end());
if (NGEP->getType() == GEP.getType())
return ReplaceInstUsesWith(GEP, NGEP);
NGEP->takeName(&GEP);
return new BitCastInst(NGEP, GEP.getType());
}
}
}
return 0;
}
Instruction *InstCombiner::visitFree(Instruction &FI) {
Value *Op = FI.getOperand(1);
// free undef -> unreachable.
if (isa<UndefValue>(Op)) {
// Insert a new store to null because we cannot modify the CFG here.
new StoreInst(ConstantInt::getTrue(FI.getContext()),
UndefValue::get(Type::getInt1PtrTy(FI.getContext())), &FI);
return EraseInstFromFunction(FI);
}
// If we have 'free null' delete the instruction. This can happen in stl code
// when lots of inlining happens.
if (isa<ConstantPointerNull>(Op))
return EraseInstFromFunction(FI);
// If we have a malloc call whose only use is a free call, delete both.
if (isMalloc(Op)) {
if (CallInst* CI = extractMallocCallFromBitCast(Op)) {
if (Op->hasOneUse() && CI->hasOneUse()) {
EraseInstFromFunction(FI);
EraseInstFromFunction(*CI);
return EraseInstFromFunction(*cast<Instruction>(Op));
}
} else {
// Op is a call to malloc
if (Op->hasOneUse()) {
EraseInstFromFunction(FI);
return EraseInstFromFunction(*cast<Instruction>(Op));
}
}
}
return 0;
}
Instruction *InstCombiner::visitBranchInst(BranchInst &BI) {
// Change br (not X), label True, label False to: br X, label False, True
Value *X = 0;
BasicBlock *TrueDest;
BasicBlock *FalseDest;
if (match(&BI, m_Br(m_Not(m_Value(X)), TrueDest, FalseDest)) &&
!isa<Constant>(X)) {
// Swap Destinations and condition...
BI.setCondition(X);
BI.setSuccessor(0, FalseDest);
BI.setSuccessor(1, TrueDest);
return &BI;
}
// Cannonicalize fcmp_one -> fcmp_oeq
FCmpInst::Predicate FPred; Value *Y;
if (match(&BI, m_Br(m_FCmp(FPred, m_Value(X), m_Value(Y)),
TrueDest, FalseDest)) &&
BI.getCondition()->hasOneUse())
if (FPred == FCmpInst::FCMP_ONE || FPred == FCmpInst::FCMP_OLE ||
FPred == FCmpInst::FCMP_OGE) {
FCmpInst *Cond = cast<FCmpInst>(BI.getCondition());
Cond->setPredicate(FCmpInst::getInversePredicate(FPred));
// Swap Destinations and condition.
BI.setSuccessor(0, FalseDest);
BI.setSuccessor(1, TrueDest);
Worklist.Add(Cond);
return &BI;
}
// Cannonicalize icmp_ne -> icmp_eq
ICmpInst::Predicate IPred;
if (match(&BI, m_Br(m_ICmp(IPred, m_Value(X), m_Value(Y)),
TrueDest, FalseDest)) &&
BI.getCondition()->hasOneUse())
if (IPred == ICmpInst::ICMP_NE || IPred == ICmpInst::ICMP_ULE ||
IPred == ICmpInst::ICMP_SLE || IPred == ICmpInst::ICMP_UGE ||
IPred == ICmpInst::ICMP_SGE) {
ICmpInst *Cond = cast<ICmpInst>(BI.getCondition());
Cond->setPredicate(ICmpInst::getInversePredicate(IPred));
// Swap Destinations and condition.
BI.setSuccessor(0, FalseDest);
BI.setSuccessor(1, TrueDest);
Worklist.Add(Cond);
return &BI;
}
return 0;
}
Instruction *InstCombiner::visitSwitchInst(SwitchInst &SI) {
Value *Cond = SI.getCondition();
if (Instruction *I = dyn_cast<Instruction>(Cond)) {
if (I->getOpcode() == Instruction::Add)
if (ConstantInt *AddRHS = dyn_cast<ConstantInt>(I->getOperand(1))) {
// change 'switch (X+4) case 1:' into 'switch (X) case -3'
for (unsigned i = 2, e = SI.getNumOperands(); i != e; i += 2)
SI.setOperand(i,
ConstantExpr::getSub(cast<Constant>(SI.getOperand(i)),
AddRHS));
SI.setOperand(0, I->getOperand(0));
Worklist.Add(I);
return &SI;
}
}
return 0;
}
Instruction *InstCombiner::visitExtractValueInst(ExtractValueInst &EV) {
Value *Agg = EV.getAggregateOperand();
if (!EV.hasIndices())
return ReplaceInstUsesWith(EV, Agg);
if (Constant *C = dyn_cast<Constant>(Agg)) {
if (isa<UndefValue>(C))
return ReplaceInstUsesWith(EV, UndefValue::get(EV.getType()));
if (isa<ConstantAggregateZero>(C))
return ReplaceInstUsesWith(EV, Constant::getNullValue(EV.getType()));
if (isa<ConstantArray>(C) || isa<ConstantStruct>(C)) {
// Extract the element indexed by the first index out of the constant
Value *V = C->getOperand(*EV.idx_begin());
if (EV.getNumIndices() > 1)
// Extract the remaining indices out of the constant indexed by the
// first index
return ExtractValueInst::Create(V, EV.idx_begin() + 1, EV.idx_end());
else
return ReplaceInstUsesWith(EV, V);
}
return 0; // Can't handle other constants
}
if (InsertValueInst *IV = dyn_cast<InsertValueInst>(Agg)) {
// We're extracting from an insertvalue instruction, compare the indices
const unsigned *exti, *exte, *insi, *inse;
for (exti = EV.idx_begin(), insi = IV->idx_begin(),
exte = EV.idx_end(), inse = IV->idx_end();
exti != exte && insi != inse;
++exti, ++insi) {
if (*insi != *exti)
// The insert and extract both reference distinctly different elements.
// This means the extract is not influenced by the insert, and we can
// replace the aggregate operand of the extract with the aggregate
// operand of the insert. i.e., replace
// %I = insertvalue { i32, { i32 } } %A, { i32 } { i32 42 }, 1
// %E = extractvalue { i32, { i32 } } %I, 0
// with
// %E = extractvalue { i32, { i32 } } %A, 0
return ExtractValueInst::Create(IV->getAggregateOperand(),
EV.idx_begin(), EV.idx_end());
}
if (exti == exte && insi == inse)
// Both iterators are at the end: Index lists are identical. Replace
// %B = insertvalue { i32, { i32 } } %A, i32 42, 1, 0
// %C = extractvalue { i32, { i32 } } %B, 1, 0
// with "i32 42"
return ReplaceInstUsesWith(EV, IV->getInsertedValueOperand());
if (exti == exte) {
// The extract list is a prefix of the insert list. i.e. replace
// %I = insertvalue { i32, { i32 } } %A, i32 42, 1, 0
// %E = extractvalue { i32, { i32 } } %I, 1
// with
// %X = extractvalue { i32, { i32 } } %A, 1
// %E = insertvalue { i32 } %X, i32 42, 0
// by switching the order of the insert and extract (though the
// insertvalue should be left in, since it may have other uses).
Value *NewEV = Builder->CreateExtractValue(IV->getAggregateOperand(),
EV.idx_begin(), EV.idx_end());
return InsertValueInst::Create(NewEV, IV->getInsertedValueOperand(),
insi, inse);
}
if (insi == inse)
// The insert list is a prefix of the extract list
// We can simply remove the common indices from the extract and make it
// operate on the inserted value instead of the insertvalue result.
// i.e., replace
// %I = insertvalue { i32, { i32 } } %A, { i32 } { i32 42 }, 1
// %E = extractvalue { i32, { i32 } } %I, 1, 0
// with
// %E extractvalue { i32 } { i32 42 }, 0
return ExtractValueInst::Create(IV->getInsertedValueOperand(),
exti, exte);
}
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(Agg)) {
// We're extracting from an intrinsic, see if we're the only user, which
// allows us to simplify multiple result intrinsics to simpler things that
// just get one value..
if (II->hasOneUse()) {
// Check if we're grabbing the overflow bit or the result of a 'with
// overflow' intrinsic. If it's the latter we can remove the intrinsic
// and replace it with a traditional binary instruction.
switch (II->getIntrinsicID()) {
case Intrinsic::uadd_with_overflow:
case Intrinsic::sadd_with_overflow:
if (*EV.idx_begin() == 0) { // Normal result.
Value *LHS = II->getOperand(1), *RHS = II->getOperand(2);
II->replaceAllUsesWith(UndefValue::get(II->getType()));
EraseInstFromFunction(*II);
return BinaryOperator::CreateAdd(LHS, RHS);
}
break;
case Intrinsic::usub_with_overflow:
case Intrinsic::ssub_with_overflow:
if (*EV.idx_begin() == 0) { // Normal result.
Value *LHS = II->getOperand(1), *RHS = II->getOperand(2);
II->replaceAllUsesWith(UndefValue::get(II->getType()));
EraseInstFromFunction(*II);
return BinaryOperator::CreateSub(LHS, RHS);
}
break;
case Intrinsic::umul_with_overflow:
case Intrinsic::smul_with_overflow:
if (*EV.idx_begin() == 0) { // Normal result.
Value *LHS = II->getOperand(1), *RHS = II->getOperand(2);
II->replaceAllUsesWith(UndefValue::get(II->getType()));
EraseInstFromFunction(*II);
return BinaryOperator::CreateMul(LHS, RHS);
}
break;
default:
break;
}
}
}
// Can't simplify extracts from other values. Note that nested extracts are
// already simplified implicitely by the above (extract ( extract (insert) )
// will be translated into extract ( insert ( extract ) ) first and then just
// the value inserted, if appropriate).
return 0;
}
/// TryToSinkInstruction - Try to move the specified instruction from its
/// current block into the beginning of DestBlock, which can only happen if it's
/// safe to move the instruction past all of the instructions between it and the
/// end of its block.
static bool TryToSinkInstruction(Instruction *I, BasicBlock *DestBlock) {
assert(I->hasOneUse() && "Invariants didn't hold!");
// Cannot move control-flow-involving, volatile loads, vaarg, etc.
if (isa<PHINode>(I) || I->mayHaveSideEffects() || isa<TerminatorInst>(I))
return false;
// Do not sink alloca instructions out of the entry block.
if (isa<AllocaInst>(I) && I->getParent() ==
&DestBlock->getParent()->getEntryBlock())
return false;
// We can only sink load instructions if there is nothing between the load and
// the end of block that could change the value.
if (I->mayReadFromMemory()) {
for (BasicBlock::iterator Scan = I, E = I->getParent()->end();
Scan != E; ++Scan)
if (Scan->mayWriteToMemory())
return false;
}
BasicBlock::iterator InsertPos = DestBlock->getFirstNonPHI();
I->moveBefore(InsertPos);
++NumSunkInst;
return true;
}
/// AddReachableCodeToWorklist - Walk the function in depth-first order, adding
/// all reachable code to the worklist.
///
/// This has a couple of tricks to make the code faster and more powerful. In
/// particular, we constant fold and DCE instructions as we go, to avoid adding
/// them to the worklist (this significantly speeds up instcombine on code where
/// many instructions are dead or constant). Additionally, if we find a branch
/// whose condition is a known constant, we only visit the reachable successors.
///
static bool AddReachableCodeToWorklist(BasicBlock *BB,
SmallPtrSet<BasicBlock*, 64> &Visited,
InstCombiner &IC,
const TargetData *TD) {
bool MadeIRChange = false;
SmallVector<BasicBlock*, 256> Worklist;
Worklist.push_back(BB);
std::vector<Instruction*> InstrsForInstCombineWorklist;
InstrsForInstCombineWorklist.reserve(128);
SmallPtrSet<ConstantExpr*, 64> FoldedConstants;
while (!Worklist.empty()) {
BB = Worklist.back();
Worklist.pop_back();
// We have now visited this block! If we've already been here, ignore it.
if (!Visited.insert(BB)) continue;
for (BasicBlock::iterator BBI = BB->begin(), E = BB->end(); BBI != E; ) {
Instruction *Inst = BBI++;
// DCE instruction if trivially dead.
if (isInstructionTriviallyDead(Inst)) {
++NumDeadInst;
DEBUG(errs() << "IC: DCE: " << *Inst << '\n');
Inst->eraseFromParent();
continue;
}
// ConstantProp instruction if trivially constant.
if (!Inst->use_empty() && isa<Constant>(Inst->getOperand(0)))
if (Constant *C = ConstantFoldInstruction(Inst, TD)) {
DEBUG(errs() << "IC: ConstFold to: " << *C << " from: "
<< *Inst << '\n');
Inst->replaceAllUsesWith(C);
++NumConstProp;
Inst->eraseFromParent();
continue;
}
if (TD) {
// See if we can constant fold its operands.
for (User::op_iterator i = Inst->op_begin(), e = Inst->op_end();
i != e; ++i) {
ConstantExpr *CE = dyn_cast<ConstantExpr>(i);
if (CE == 0) continue;
// If we already folded this constant, don't try again.
if (!FoldedConstants.insert(CE))
continue;
Constant *NewC = ConstantFoldConstantExpression(CE, TD);
if (NewC && NewC != CE) {
*i = NewC;
MadeIRChange = true;
}
}
}
InstrsForInstCombineWorklist.push_back(Inst);
}
// Recursively visit successors. If this is a branch or switch on a
// constant, only visit the reachable successor.
TerminatorInst *TI = BB->getTerminator();
if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
if (BI->isConditional() && isa<ConstantInt>(BI->getCondition())) {
bool CondVal = cast<ConstantInt>(BI->getCondition())->getZExtValue();
BasicBlock *ReachableBB = BI->getSuccessor(!CondVal);
Worklist.push_back(ReachableBB);
continue;
}
} else if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
if (ConstantInt *Cond = dyn_cast<ConstantInt>(SI->getCondition())) {
// See if this is an explicit destination.
for (unsigned i = 1, e = SI->getNumSuccessors(); i != e; ++i)
if (SI->getCaseValue(i) == Cond) {
BasicBlock *ReachableBB = SI->getSuccessor(i);
Worklist.push_back(ReachableBB);
continue;
}
// Otherwise it is the default destination.
Worklist.push_back(SI->getSuccessor(0));
continue;
}
}
for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
Worklist.push_back(TI->getSuccessor(i));
}
// Once we've found all of the instructions to add to instcombine's worklist,
// add them in reverse order. This way instcombine will visit from the top
// of the function down. This jives well with the way that it adds all uses
// of instructions to the worklist after doing a transformation, thus avoiding
// some N^2 behavior in pathological cases.
IC.Worklist.AddInitialGroup(&InstrsForInstCombineWorklist[0],
InstrsForInstCombineWorklist.size());
return MadeIRChange;
}
bool InstCombiner::DoOneIteration(Function &F, unsigned Iteration) {
MadeIRChange = false;
DEBUG(errs() << "\n\nINSTCOMBINE ITERATION #" << Iteration << " on "
<< F.getNameStr() << "\n");
{
// Do a depth-first traversal of the function, populate the worklist with
// the reachable instructions. Ignore blocks that are not reachable. Keep
// track of which blocks we visit.
SmallPtrSet<BasicBlock*, 64> Visited;
MadeIRChange |= AddReachableCodeToWorklist(F.begin(), Visited, *this, TD);
// Do a quick scan over the function. If we find any blocks that are
// unreachable, remove any instructions inside of them. This prevents
// the instcombine code from having to deal with some bad special cases.
for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
if (!Visited.count(BB)) {
Instruction *Term = BB->getTerminator();
while (Term != BB->begin()) { // Remove instrs bottom-up
BasicBlock::iterator I = Term; --I;
DEBUG(errs() << "IC: DCE: " << *I << '\n');
// A debug intrinsic shouldn't force another iteration if we weren't
// going to do one without it.
if (!isa<DbgInfoIntrinsic>(I)) {
++NumDeadInst;
MadeIRChange = true;
}
// If I is not void type then replaceAllUsesWith undef.
// This allows ValueHandlers and custom metadata to adjust itself.
if (!I->getType()->isVoidTy())
I->replaceAllUsesWith(UndefValue::get(I->getType()));
I->eraseFromParent();
}
}
}
while (!Worklist.isEmpty()) {
Instruction *I = Worklist.RemoveOne();
if (I == 0) continue; // skip null values.
// Check to see if we can DCE the instruction.
if (isInstructionTriviallyDead(I)) {
DEBUG(errs() << "IC: DCE: " << *I << '\n');
EraseInstFromFunction(*I);
++NumDeadInst;
MadeIRChange = true;
continue;
}
// Instruction isn't dead, see if we can constant propagate it.
if (!I->use_empty() && isa<Constant>(I->getOperand(0)))
if (Constant *C = ConstantFoldInstruction(I, TD)) {
DEBUG(errs() << "IC: ConstFold to: " << *C << " from: " << *I << '\n');
// Add operands to the worklist.
ReplaceInstUsesWith(*I, C);
++NumConstProp;
EraseInstFromFunction(*I);
MadeIRChange = true;
continue;
}
// See if we can trivially sink this instruction to a successor basic block.
if (I->hasOneUse()) {
BasicBlock *BB = I->getParent();
Instruction *UserInst = cast<Instruction>(I->use_back());
BasicBlock *UserParent;
// Get the block the use occurs in.
if (PHINode *PN = dyn_cast<PHINode>(UserInst))
UserParent = PN->getIncomingBlock(I->use_begin().getUse());
else
UserParent = UserInst->getParent();
if (UserParent != BB) {
bool UserIsSuccessor = false;
// See if the user is one of our successors.
for (succ_iterator SI = succ_begin(BB), E = succ_end(BB); SI != E; ++SI)
if (*SI == UserParent) {
UserIsSuccessor = true;
break;
}
// If the user is one of our immediate successors, and if that successor
// only has us as a predecessors (we'd have to split the critical edge
// otherwise), we can keep going.
if (UserIsSuccessor && UserParent->getSinglePredecessor())
// Okay, the CFG is simple enough, try to sink this instruction.
MadeIRChange |= TryToSinkInstruction(I, UserParent);
}
}
// Now that we have an instruction, try combining it to simplify it.
Builder->SetInsertPoint(I->getParent(), I);
#ifndef NDEBUG
std::string OrigI;
#endif
DEBUG(raw_string_ostream SS(OrigI); I->print(SS); OrigI = SS.str(););
DEBUG(errs() << "IC: Visiting: " << OrigI << '\n');
if (Instruction *Result = visit(*I)) {
++NumCombined;
// Should we replace the old instruction with a new one?
if (Result != I) {
DEBUG(errs() << "IC: Old = " << *I << '\n'
<< " New = " << *Result << '\n');
// Everything uses the new instruction now.
I->replaceAllUsesWith(Result);
// Push the new instruction and any users onto the worklist.
Worklist.Add(Result);
Worklist.AddUsersToWorkList(*Result);
// Move the name to the new instruction first.
Result->takeName(I);
// Insert the new instruction into the basic block...
BasicBlock *InstParent = I->getParent();
BasicBlock::iterator InsertPos = I;
if (!isa<PHINode>(Result)) // If combining a PHI, don't insert
while (isa<PHINode>(InsertPos)) // middle of a block of PHIs.
++InsertPos;
InstParent->getInstList().insert(InsertPos, Result);
EraseInstFromFunction(*I);
} else {
#ifndef NDEBUG
DEBUG(errs() << "IC: Mod = " << OrigI << '\n'
<< " New = " << *I << '\n');
#endif
// If the instruction was modified, it's possible that it is now dead.
// if so, remove it.
if (isInstructionTriviallyDead(I)) {
EraseInstFromFunction(*I);
} else {
Worklist.Add(I);
Worklist.AddUsersToWorkList(*I);
}
}
MadeIRChange = true;
}
}
Worklist.Zap();
return MadeIRChange;
}
bool InstCombiner::runOnFunction(Function &F) {
MustPreserveLCSSA = mustPreserveAnalysisID(LCSSAID);
TD = getAnalysisIfAvailable<TargetData>();
/// Builder - This is an IRBuilder that automatically inserts new
/// instructions into the worklist when they are created.
IRBuilder<true, TargetFolder, InstCombineIRInserter>
TheBuilder(F.getContext(), TargetFolder(TD),
InstCombineIRInserter(Worklist));
Builder = &TheBuilder;
bool EverMadeChange = false;
// Iterate while there is work to do.
unsigned Iteration = 0;
while (DoOneIteration(F, Iteration++))
EverMadeChange = true;
Builder = 0;
return EverMadeChange;
}
FunctionPass *llvm::createInstructionCombiningPass() {
return new InstCombiner();
}
|