1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
|
//===-- PPCISelLowering.cpp - PPC DAG Lowering Implementation -------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements the PPCISelLowering class.
//
//===----------------------------------------------------------------------===//
#include "PPCISelLowering.h"
#include "PPCMachineFunctionInfo.h"
#include "PPCPredicates.h"
#include "PPCTargetMachine.h"
#include "PPCPerfectShuffle.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/VectorExtras.h"
#include "llvm/Analysis/ScalarEvolutionExpressions.h"
#include "llvm/CodeGen/CallingConvLower.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/PseudoSourceValue.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/Constants.h"
#include "llvm/Function.h"
#include "llvm/Intrinsics.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/CommandLine.h"
using namespace llvm;
static cl::opt<bool> EnablePPCPreinc("enable-ppc-preinc",
cl::desc("enable preincrement load/store generation on PPC (experimental)"),
cl::Hidden);
PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
: TargetLowering(TM), PPCSubTarget(*TM.getSubtargetImpl()) {
setPow2DivIsCheap();
// Use _setjmp/_longjmp instead of setjmp/longjmp.
setUseUnderscoreSetJmp(true);
setUseUnderscoreLongJmp(true);
// Set up the register classes.
addRegisterClass(MVT::i32, PPC::GPRCRegisterClass);
addRegisterClass(MVT::f32, PPC::F4RCRegisterClass);
addRegisterClass(MVT::f64, PPC::F8RCRegisterClass);
// PowerPC has an i16 but no i8 (or i1) SEXTLOAD
setLoadXAction(ISD::SEXTLOAD, MVT::i1, Promote);
setLoadXAction(ISD::SEXTLOAD, MVT::i8, Expand);
setTruncStoreAction(MVT::f64, MVT::f32, Expand);
// PowerPC has pre-inc load and store's.
setIndexedLoadAction(ISD::PRE_INC, MVT::i1, Legal);
setIndexedLoadAction(ISD::PRE_INC, MVT::i8, Legal);
setIndexedLoadAction(ISD::PRE_INC, MVT::i16, Legal);
setIndexedLoadAction(ISD::PRE_INC, MVT::i32, Legal);
setIndexedLoadAction(ISD::PRE_INC, MVT::i64, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::i1, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::i8, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::i16, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::i32, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::i64, Legal);
setOperationAction(ISD::ConstantFP, MVT::f64, Expand);
setOperationAction(ISD::ConstantFP, MVT::f32, Expand);
// Shortening conversions involving ppcf128 get expanded (2 regs -> 1 reg)
setConvertAction(MVT::ppcf128, MVT::f64, Expand);
setConvertAction(MVT::ppcf128, MVT::f32, Expand);
// This is used in the ppcf128->int sequence. Note it has different semantics
// from FP_ROUND: that rounds to nearest, this rounds to zero.
setOperationAction(ISD::FP_ROUND_INREG, MVT::ppcf128, Custom);
// PowerPC has no intrinsics for these particular operations
setOperationAction(ISD::MEMMOVE, MVT::Other, Expand);
setOperationAction(ISD::MEMSET, MVT::Other, Expand);
setOperationAction(ISD::MEMCPY, MVT::Other, Expand);
// PowerPC has no SREM/UREM instructions
setOperationAction(ISD::SREM, MVT::i32, Expand);
setOperationAction(ISD::UREM, MVT::i32, Expand);
setOperationAction(ISD::SREM, MVT::i64, Expand);
setOperationAction(ISD::UREM, MVT::i64, Expand);
// Don't use SMUL_LOHI/UMUL_LOHI or SDIVREM/UDIVREM to lower SREM/UREM.
setOperationAction(ISD::UMUL_LOHI, MVT::i32, Expand);
setOperationAction(ISD::SMUL_LOHI, MVT::i32, Expand);
setOperationAction(ISD::UMUL_LOHI, MVT::i64, Expand);
setOperationAction(ISD::SMUL_LOHI, MVT::i64, Expand);
setOperationAction(ISD::UDIVREM, MVT::i32, Expand);
setOperationAction(ISD::SDIVREM, MVT::i32, Expand);
setOperationAction(ISD::UDIVREM, MVT::i64, Expand);
setOperationAction(ISD::SDIVREM, MVT::i64, Expand);
// We don't support sin/cos/sqrt/fmod/pow
setOperationAction(ISD::FSIN , MVT::f64, Expand);
setOperationAction(ISD::FCOS , MVT::f64, Expand);
setOperationAction(ISD::FREM , MVT::f64, Expand);
setOperationAction(ISD::FPOW , MVT::f64, Expand);
setOperationAction(ISD::FSIN , MVT::f32, Expand);
setOperationAction(ISD::FCOS , MVT::f32, Expand);
setOperationAction(ISD::FREM , MVT::f32, Expand);
setOperationAction(ISD::FPOW , MVT::f32, Expand);
setOperationAction(ISD::FLT_ROUNDS, MVT::i32, Custom);
// If we're enabling GP optimizations, use hardware square root
if (!TM.getSubtarget<PPCSubtarget>().hasFSQRT()) {
setOperationAction(ISD::FSQRT, MVT::f64, Expand);
setOperationAction(ISD::FSQRT, MVT::f32, Expand);
}
setOperationAction(ISD::FCOPYSIGN, MVT::f64, Expand);
setOperationAction(ISD::FCOPYSIGN, MVT::f32, Expand);
// PowerPC does not have BSWAP, CTPOP or CTTZ
setOperationAction(ISD::BSWAP, MVT::i32 , Expand);
setOperationAction(ISD::CTPOP, MVT::i32 , Expand);
setOperationAction(ISD::CTTZ , MVT::i32 , Expand);
setOperationAction(ISD::BSWAP, MVT::i64 , Expand);
setOperationAction(ISD::CTPOP, MVT::i64 , Expand);
setOperationAction(ISD::CTTZ , MVT::i64 , Expand);
// PowerPC does not have ROTR
setOperationAction(ISD::ROTR, MVT::i32 , Expand);
// PowerPC does not have Select
setOperationAction(ISD::SELECT, MVT::i32, Expand);
setOperationAction(ISD::SELECT, MVT::i64, Expand);
setOperationAction(ISD::SELECT, MVT::f32, Expand);
setOperationAction(ISD::SELECT, MVT::f64, Expand);
// PowerPC wants to turn select_cc of FP into fsel when possible.
setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
setOperationAction(ISD::SELECT_CC, MVT::f64, Custom);
// PowerPC wants to optimize integer setcc a bit
setOperationAction(ISD::SETCC, MVT::i32, Custom);
// PowerPC does not have BRCOND which requires SetCC
setOperationAction(ISD::BRCOND, MVT::Other, Expand);
setOperationAction(ISD::BR_JT, MVT::Other, Expand);
// PowerPC turns FP_TO_SINT into FCTIWZ and some load/stores.
setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
// PowerPC does not have [U|S]INT_TO_FP
setOperationAction(ISD::SINT_TO_FP, MVT::i32, Expand);
setOperationAction(ISD::UINT_TO_FP, MVT::i32, Expand);
setOperationAction(ISD::BIT_CONVERT, MVT::f32, Expand);
setOperationAction(ISD::BIT_CONVERT, MVT::i32, Expand);
setOperationAction(ISD::BIT_CONVERT, MVT::i64, Expand);
setOperationAction(ISD::BIT_CONVERT, MVT::f64, Expand);
// We cannot sextinreg(i1). Expand to shifts.
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
// Support label based line numbers.
setOperationAction(ISD::LOCATION, MVT::Other, Expand);
setOperationAction(ISD::DEBUG_LOC, MVT::Other, Expand);
setOperationAction(ISD::EXCEPTIONADDR, MVT::i64, Expand);
setOperationAction(ISD::EHSELECTION, MVT::i64, Expand);
setOperationAction(ISD::EXCEPTIONADDR, MVT::i32, Expand);
setOperationAction(ISD::EHSELECTION, MVT::i32, Expand);
// We want to legalize GlobalAddress and ConstantPool nodes into the
// appropriate instructions to materialize the address.
setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
setOperationAction(ISD::GlobalTLSAddress, MVT::i32, Custom);
setOperationAction(ISD::ConstantPool, MVT::i32, Custom);
setOperationAction(ISD::JumpTable, MVT::i32, Custom);
setOperationAction(ISD::GlobalAddress, MVT::i64, Custom);
setOperationAction(ISD::GlobalTLSAddress, MVT::i64, Custom);
setOperationAction(ISD::ConstantPool, MVT::i64, Custom);
setOperationAction(ISD::JumpTable, MVT::i64, Custom);
// RET must be custom lowered, to meet ABI requirements
setOperationAction(ISD::RET , MVT::Other, Custom);
// VASTART needs to be custom lowered to use the VarArgsFrameIndex
setOperationAction(ISD::VASTART , MVT::Other, Custom);
// VAARG is custom lowered with ELF 32 ABI
if (TM.getSubtarget<PPCSubtarget>().isELF32_ABI())
setOperationAction(ISD::VAARG, MVT::Other, Custom);
else
setOperationAction(ISD::VAARG, MVT::Other, Expand);
// Use the default implementation.
setOperationAction(ISD::VACOPY , MVT::Other, Expand);
setOperationAction(ISD::VAEND , MVT::Other, Expand);
setOperationAction(ISD::STACKSAVE , MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE , MVT::Other, Custom);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32 , Custom);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64 , Custom);
// We want to custom lower some of our intrinsics.
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
if (TM.getSubtarget<PPCSubtarget>().has64BitSupport()) {
// They also have instructions for converting between i64 and fp.
setOperationAction(ISD::FP_TO_SINT, MVT::i64, Custom);
setOperationAction(ISD::FP_TO_UINT, MVT::i64, Expand);
setOperationAction(ISD::SINT_TO_FP, MVT::i64, Custom);
setOperationAction(ISD::UINT_TO_FP, MVT::i64, Expand);
setOperationAction(ISD::FP_TO_UINT, MVT::i32, Expand);
// FIXME: disable this lowered code. This generates 64-bit register values,
// and we don't model the fact that the top part is clobbered by calls. We
// need to flag these together so that the value isn't live across a call.
//setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
// To take advantage of the above i64 FP_TO_SINT, promote i32 FP_TO_UINT
setOperationAction(ISD::FP_TO_UINT, MVT::i32, Promote);
} else {
// PowerPC does not have FP_TO_UINT on 32-bit implementations.
setOperationAction(ISD::FP_TO_UINT, MVT::i32, Expand);
}
if (TM.getSubtarget<PPCSubtarget>().use64BitRegs()) {
// 64-bit PowerPC implementations can support i64 types directly
addRegisterClass(MVT::i64, PPC::G8RCRegisterClass);
// BUILD_PAIR can't be handled natively, and should be expanded to shl/or
setOperationAction(ISD::BUILD_PAIR, MVT::i64, Expand);
} else {
// 32-bit PowerPC wants to expand i64 shifts itself.
setOperationAction(ISD::SHL_PARTS, MVT::i32, Custom);
setOperationAction(ISD::SRA_PARTS, MVT::i32, Custom);
setOperationAction(ISD::SRL_PARTS, MVT::i32, Custom);
}
if (TM.getSubtarget<PPCSubtarget>().hasAltivec()) {
// First set operation action for all vector types to expand. Then we
// will selectively turn on ones that can be effectively codegen'd.
for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
VT <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++VT) {
// add/sub are legal for all supported vector VT's.
setOperationAction(ISD::ADD , (MVT::ValueType)VT, Legal);
setOperationAction(ISD::SUB , (MVT::ValueType)VT, Legal);
// We promote all shuffles to v16i8.
setOperationAction(ISD::VECTOR_SHUFFLE, (MVT::ValueType)VT, Promote);
AddPromotedToType (ISD::VECTOR_SHUFFLE, (MVT::ValueType)VT, MVT::v16i8);
// We promote all non-typed operations to v4i32.
setOperationAction(ISD::AND , (MVT::ValueType)VT, Promote);
AddPromotedToType (ISD::AND , (MVT::ValueType)VT, MVT::v4i32);
setOperationAction(ISD::OR , (MVT::ValueType)VT, Promote);
AddPromotedToType (ISD::OR , (MVT::ValueType)VT, MVT::v4i32);
setOperationAction(ISD::XOR , (MVT::ValueType)VT, Promote);
AddPromotedToType (ISD::XOR , (MVT::ValueType)VT, MVT::v4i32);
setOperationAction(ISD::LOAD , (MVT::ValueType)VT, Promote);
AddPromotedToType (ISD::LOAD , (MVT::ValueType)VT, MVT::v4i32);
setOperationAction(ISD::SELECT, (MVT::ValueType)VT, Promote);
AddPromotedToType (ISD::SELECT, (MVT::ValueType)VT, MVT::v4i32);
setOperationAction(ISD::STORE, (MVT::ValueType)VT, Promote);
AddPromotedToType (ISD::STORE, (MVT::ValueType)VT, MVT::v4i32);
// No other operations are legal.
setOperationAction(ISD::MUL , (MVT::ValueType)VT, Expand);
setOperationAction(ISD::SDIV, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::SREM, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::UDIV, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::UREM, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::FDIV, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::FNEG, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::INSERT_VECTOR_ELT, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::BUILD_VECTOR, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::UMUL_LOHI, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::SMUL_LOHI, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::UDIVREM, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::SDIVREM, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::SCALAR_TO_VECTOR, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::FPOW, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::CTPOP, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::CTLZ, (MVT::ValueType)VT, Expand);
setOperationAction(ISD::CTTZ, (MVT::ValueType)VT, Expand);
}
// We can custom expand all VECTOR_SHUFFLEs to VPERM, others we can handle
// with merges, splats, etc.
setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v16i8, Custom);
setOperationAction(ISD::AND , MVT::v4i32, Legal);
setOperationAction(ISD::OR , MVT::v4i32, Legal);
setOperationAction(ISD::XOR , MVT::v4i32, Legal);
setOperationAction(ISD::LOAD , MVT::v4i32, Legal);
setOperationAction(ISD::SELECT, MVT::v4i32, Expand);
setOperationAction(ISD::STORE , MVT::v4i32, Legal);
addRegisterClass(MVT::v4f32, PPC::VRRCRegisterClass);
addRegisterClass(MVT::v4i32, PPC::VRRCRegisterClass);
addRegisterClass(MVT::v8i16, PPC::VRRCRegisterClass);
addRegisterClass(MVT::v16i8, PPC::VRRCRegisterClass);
setOperationAction(ISD::MUL, MVT::v4f32, Legal);
setOperationAction(ISD::MUL, MVT::v4i32, Custom);
setOperationAction(ISD::MUL, MVT::v8i16, Custom);
setOperationAction(ISD::MUL, MVT::v16i8, Custom);
setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v4f32, Custom);
setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v4i32, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v16i8, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v8i16, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v4i32, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v4f32, Custom);
}
setSetCCResultType(MVT::i32);
setShiftAmountType(MVT::i32);
setSetCCResultContents(ZeroOrOneSetCCResult);
if (TM.getSubtarget<PPCSubtarget>().isPPC64()) {
setStackPointerRegisterToSaveRestore(PPC::X1);
setExceptionPointerRegister(PPC::X3);
setExceptionSelectorRegister(PPC::X4);
} else {
setStackPointerRegisterToSaveRestore(PPC::R1);
setExceptionPointerRegister(PPC::R3);
setExceptionSelectorRegister(PPC::R4);
}
// We have target-specific dag combine patterns for the following nodes:
setTargetDAGCombine(ISD::SINT_TO_FP);
setTargetDAGCombine(ISD::STORE);
setTargetDAGCombine(ISD::BR_CC);
setTargetDAGCombine(ISD::BSWAP);
// Darwin long double math library functions have $LDBL128 appended.
if (TM.getSubtarget<PPCSubtarget>().isDarwin()) {
setLibcallName(RTLIB::COS_PPCF128, "cosl$LDBL128");
setLibcallName(RTLIB::POW_PPCF128, "powl$LDBL128");
setLibcallName(RTLIB::REM_PPCF128, "fmodl$LDBL128");
setLibcallName(RTLIB::SIN_PPCF128, "sinl$LDBL128");
setLibcallName(RTLIB::SQRT_PPCF128, "sqrtl$LDBL128");
}
computeRegisterProperties();
}
const char *PPCTargetLowering::getTargetNodeName(unsigned Opcode) const {
switch (Opcode) {
default: return 0;
case PPCISD::FSEL: return "PPCISD::FSEL";
case PPCISD::FCFID: return "PPCISD::FCFID";
case PPCISD::FCTIDZ: return "PPCISD::FCTIDZ";
case PPCISD::FCTIWZ: return "PPCISD::FCTIWZ";
case PPCISD::STFIWX: return "PPCISD::STFIWX";
case PPCISD::VMADDFP: return "PPCISD::VMADDFP";
case PPCISD::VNMSUBFP: return "PPCISD::VNMSUBFP";
case PPCISD::VPERM: return "PPCISD::VPERM";
case PPCISD::Hi: return "PPCISD::Hi";
case PPCISD::Lo: return "PPCISD::Lo";
case PPCISD::DYNALLOC: return "PPCISD::DYNALLOC";
case PPCISD::GlobalBaseReg: return "PPCISD::GlobalBaseReg";
case PPCISD::SRL: return "PPCISD::SRL";
case PPCISD::SRA: return "PPCISD::SRA";
case PPCISD::SHL: return "PPCISD::SHL";
case PPCISD::EXTSW_32: return "PPCISD::EXTSW_32";
case PPCISD::STD_32: return "PPCISD::STD_32";
case PPCISD::CALL_ELF: return "PPCISD::CALL_ELF";
case PPCISD::CALL_Macho: return "PPCISD::CALL_Macho";
case PPCISD::MTCTR: return "PPCISD::MTCTR";
case PPCISD::BCTRL_Macho: return "PPCISD::BCTRL_Macho";
case PPCISD::BCTRL_ELF: return "PPCISD::BCTRL_ELF";
case PPCISD::RET_FLAG: return "PPCISD::RET_FLAG";
case PPCISD::MFCR: return "PPCISD::MFCR";
case PPCISD::VCMP: return "PPCISD::VCMP";
case PPCISD::VCMPo: return "PPCISD::VCMPo";
case PPCISD::LBRX: return "PPCISD::LBRX";
case PPCISD::STBRX: return "PPCISD::STBRX";
case PPCISD::COND_BRANCH: return "PPCISD::COND_BRANCH";
case PPCISD::MFFS: return "PPCISD::MFFS";
case PPCISD::MTFSB0: return "PPCISD::MTFSB0";
case PPCISD::MTFSB1: return "PPCISD::MTFSB1";
case PPCISD::FADDRTZ: return "PPCISD::FADDRTZ";
case PPCISD::MTFSF: return "PPCISD::MTFSF";
}
}
//===----------------------------------------------------------------------===//
// Node matching predicates, for use by the tblgen matching code.
//===----------------------------------------------------------------------===//
/// isFloatingPointZero - Return true if this is 0.0 or -0.0.
static bool isFloatingPointZero(SDOperand Op) {
if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(Op))
return CFP->getValueAPF().isZero();
else if (ISD::isEXTLoad(Op.Val) || ISD::isNON_EXTLoad(Op.Val)) {
// Maybe this has already been legalized into the constant pool?
if (ConstantPoolSDNode *CP = dyn_cast<ConstantPoolSDNode>(Op.getOperand(1)))
if (ConstantFP *CFP = dyn_cast<ConstantFP>(CP->getConstVal()))
return CFP->getValueAPF().isZero();
}
return false;
}
/// isConstantOrUndef - Op is either an undef node or a ConstantSDNode. Return
/// true if Op is undef or if it matches the specified value.
static bool isConstantOrUndef(SDOperand Op, unsigned Val) {
return Op.getOpcode() == ISD::UNDEF ||
cast<ConstantSDNode>(Op)->getValue() == Val;
}
/// isVPKUHUMShuffleMask - Return true if this is the shuffle mask for a
/// VPKUHUM instruction.
bool PPC::isVPKUHUMShuffleMask(SDNode *N, bool isUnary) {
if (!isUnary) {
for (unsigned i = 0; i != 16; ++i)
if (!isConstantOrUndef(N->getOperand(i), i*2+1))
return false;
} else {
for (unsigned i = 0; i != 8; ++i)
if (!isConstantOrUndef(N->getOperand(i), i*2+1) ||
!isConstantOrUndef(N->getOperand(i+8), i*2+1))
return false;
}
return true;
}
/// isVPKUWUMShuffleMask - Return true if this is the shuffle mask for a
/// VPKUWUM instruction.
bool PPC::isVPKUWUMShuffleMask(SDNode *N, bool isUnary) {
if (!isUnary) {
for (unsigned i = 0; i != 16; i += 2)
if (!isConstantOrUndef(N->getOperand(i ), i*2+2) ||
!isConstantOrUndef(N->getOperand(i+1), i*2+3))
return false;
} else {
for (unsigned i = 0; i != 8; i += 2)
if (!isConstantOrUndef(N->getOperand(i ), i*2+2) ||
!isConstantOrUndef(N->getOperand(i+1), i*2+3) ||
!isConstantOrUndef(N->getOperand(i+8), i*2+2) ||
!isConstantOrUndef(N->getOperand(i+9), i*2+3))
return false;
}
return true;
}
/// isVMerge - Common function, used to match vmrg* shuffles.
///
static bool isVMerge(SDNode *N, unsigned UnitSize,
unsigned LHSStart, unsigned RHSStart) {
assert(N->getOpcode() == ISD::BUILD_VECTOR &&
N->getNumOperands() == 16 && "PPC only supports shuffles by bytes!");
assert((UnitSize == 1 || UnitSize == 2 || UnitSize == 4) &&
"Unsupported merge size!");
for (unsigned i = 0; i != 8/UnitSize; ++i) // Step over units
for (unsigned j = 0; j != UnitSize; ++j) { // Step over bytes within unit
if (!isConstantOrUndef(N->getOperand(i*UnitSize*2+j),
LHSStart+j+i*UnitSize) ||
!isConstantOrUndef(N->getOperand(i*UnitSize*2+UnitSize+j),
RHSStart+j+i*UnitSize))
return false;
}
return true;
}
/// isVMRGLShuffleMask - Return true if this is a shuffle mask suitable for
/// a VRGL* instruction with the specified unit size (1,2 or 4 bytes).
bool PPC::isVMRGLShuffleMask(SDNode *N, unsigned UnitSize, bool isUnary) {
if (!isUnary)
return isVMerge(N, UnitSize, 8, 24);
return isVMerge(N, UnitSize, 8, 8);
}
/// isVMRGHShuffleMask - Return true if this is a shuffle mask suitable for
/// a VRGH* instruction with the specified unit size (1,2 or 4 bytes).
bool PPC::isVMRGHShuffleMask(SDNode *N, unsigned UnitSize, bool isUnary) {
if (!isUnary)
return isVMerge(N, UnitSize, 0, 16);
return isVMerge(N, UnitSize, 0, 0);
}
/// isVSLDOIShuffleMask - If this is a vsldoi shuffle mask, return the shift
/// amount, otherwise return -1.
int PPC::isVSLDOIShuffleMask(SDNode *N, bool isUnary) {
assert(N->getOpcode() == ISD::BUILD_VECTOR &&
N->getNumOperands() == 16 && "PPC only supports shuffles by bytes!");
// Find the first non-undef value in the shuffle mask.
unsigned i;
for (i = 0; i != 16 && N->getOperand(i).getOpcode() == ISD::UNDEF; ++i)
/*search*/;
if (i == 16) return -1; // all undef.
// Otherwise, check to see if the rest of the elements are consequtively
// numbered from this value.
unsigned ShiftAmt = cast<ConstantSDNode>(N->getOperand(i))->getValue();
if (ShiftAmt < i) return -1;
ShiftAmt -= i;
if (!isUnary) {
// Check the rest of the elements to see if they are consequtive.
for (++i; i != 16; ++i)
if (!isConstantOrUndef(N->getOperand(i), ShiftAmt+i))
return -1;
} else {
// Check the rest of the elements to see if they are consequtive.
for (++i; i != 16; ++i)
if (!isConstantOrUndef(N->getOperand(i), (ShiftAmt+i) & 15))
return -1;
}
return ShiftAmt;
}
/// isSplatShuffleMask - Return true if the specified VECTOR_SHUFFLE operand
/// specifies a splat of a single element that is suitable for input to
/// VSPLTB/VSPLTH/VSPLTW.
bool PPC::isSplatShuffleMask(SDNode *N, unsigned EltSize) {
assert(N->getOpcode() == ISD::BUILD_VECTOR &&
N->getNumOperands() == 16 &&
(EltSize == 1 || EltSize == 2 || EltSize == 4));
// This is a splat operation if each element of the permute is the same, and
// if the value doesn't reference the second vector.
unsigned ElementBase = 0;
SDOperand Elt = N->getOperand(0);
if (ConstantSDNode *EltV = dyn_cast<ConstantSDNode>(Elt))
ElementBase = EltV->getValue();
else
return false; // FIXME: Handle UNDEF elements too!
if (cast<ConstantSDNode>(Elt)->getValue() >= 16)
return false;
// Check that they are consequtive.
for (unsigned i = 1; i != EltSize; ++i) {
if (!isa<ConstantSDNode>(N->getOperand(i)) ||
cast<ConstantSDNode>(N->getOperand(i))->getValue() != i+ElementBase)
return false;
}
assert(isa<ConstantSDNode>(Elt) && "Invalid VECTOR_SHUFFLE mask!");
for (unsigned i = EltSize, e = 16; i != e; i += EltSize) {
if (N->getOperand(i).getOpcode() == ISD::UNDEF) continue;
assert(isa<ConstantSDNode>(N->getOperand(i)) &&
"Invalid VECTOR_SHUFFLE mask!");
for (unsigned j = 0; j != EltSize; ++j)
if (N->getOperand(i+j) != N->getOperand(j))
return false;
}
return true;
}
/// isAllNegativeZeroVector - Returns true if all elements of build_vector
/// are -0.0.
bool PPC::isAllNegativeZeroVector(SDNode *N) {
assert(N->getOpcode() == ISD::BUILD_VECTOR);
if (PPC::isSplatShuffleMask(N, N->getNumOperands()))
if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(N))
return CFP->getValueAPF().isNegZero();
return false;
}
/// getVSPLTImmediate - Return the appropriate VSPLT* immediate to splat the
/// specified isSplatShuffleMask VECTOR_SHUFFLE mask.
unsigned PPC::getVSPLTImmediate(SDNode *N, unsigned EltSize) {
assert(isSplatShuffleMask(N, EltSize));
return cast<ConstantSDNode>(N->getOperand(0))->getValue() / EltSize;
}
/// get_VSPLTI_elt - If this is a build_vector of constants which can be formed
/// by using a vspltis[bhw] instruction of the specified element size, return
/// the constant being splatted. The ByteSize field indicates the number of
/// bytes of each element [124] -> [bhw].
SDOperand PPC::get_VSPLTI_elt(SDNode *N, unsigned ByteSize, SelectionDAG &DAG) {
SDOperand OpVal(0, 0);
// If ByteSize of the splat is bigger than the element size of the
// build_vector, then we have a case where we are checking for a splat where
// multiple elements of the buildvector are folded together into a single
// logical element of the splat (e.g. "vsplish 1" to splat {0,1}*8).
unsigned EltSize = 16/N->getNumOperands();
if (EltSize < ByteSize) {
unsigned Multiple = ByteSize/EltSize; // Number of BV entries per spltval.
SDOperand UniquedVals[4];
assert(Multiple > 1 && Multiple <= 4 && "How can this happen?");
// See if all of the elements in the buildvector agree across.
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
if (N->getOperand(i).getOpcode() == ISD::UNDEF) continue;
// If the element isn't a constant, bail fully out.
if (!isa<ConstantSDNode>(N->getOperand(i))) return SDOperand();
if (UniquedVals[i&(Multiple-1)].Val == 0)
UniquedVals[i&(Multiple-1)] = N->getOperand(i);
else if (UniquedVals[i&(Multiple-1)] != N->getOperand(i))
return SDOperand(); // no match.
}
// Okay, if we reached this point, UniquedVals[0..Multiple-1] contains
// either constant or undef values that are identical for each chunk. See
// if these chunks can form into a larger vspltis*.
// Check to see if all of the leading entries are either 0 or -1. If
// neither, then this won't fit into the immediate field.
bool LeadingZero = true;
bool LeadingOnes = true;
for (unsigned i = 0; i != Multiple-1; ++i) {
if (UniquedVals[i].Val == 0) continue; // Must have been undefs.
LeadingZero &= cast<ConstantSDNode>(UniquedVals[i])->isNullValue();
LeadingOnes &= cast<ConstantSDNode>(UniquedVals[i])->isAllOnesValue();
}
// Finally, check the least significant entry.
if (LeadingZero) {
if (UniquedVals[Multiple-1].Val == 0)
return DAG.getTargetConstant(0, MVT::i32); // 0,0,0,undef
int Val = cast<ConstantSDNode>(UniquedVals[Multiple-1])->getValue();
if (Val < 16)
return DAG.getTargetConstant(Val, MVT::i32); // 0,0,0,4 -> vspltisw(4)
}
if (LeadingOnes) {
if (UniquedVals[Multiple-1].Val == 0)
return DAG.getTargetConstant(~0U, MVT::i32); // -1,-1,-1,undef
int Val =cast<ConstantSDNode>(UniquedVals[Multiple-1])->getSignExtended();
if (Val >= -16) // -1,-1,-1,-2 -> vspltisw(-2)
return DAG.getTargetConstant(Val, MVT::i32);
}
return SDOperand();
}
// Check to see if this buildvec has a single non-undef value in its elements.
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
if (N->getOperand(i).getOpcode() == ISD::UNDEF) continue;
if (OpVal.Val == 0)
OpVal = N->getOperand(i);
else if (OpVal != N->getOperand(i))
return SDOperand();
}
if (OpVal.Val == 0) return SDOperand(); // All UNDEF: use implicit def.
unsigned ValSizeInBytes = 0;
uint64_t Value = 0;
if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(OpVal)) {
Value = CN->getValue();
ValSizeInBytes = MVT::getSizeInBits(CN->getValueType(0))/8;
} else if (ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(OpVal)) {
assert(CN->getValueType(0) == MVT::f32 && "Only one legal FP vector type!");
Value = FloatToBits(CN->getValueAPF().convertToFloat());
ValSizeInBytes = 4;
}
// If the splat value is larger than the element value, then we can never do
// this splat. The only case that we could fit the replicated bits into our
// immediate field for would be zero, and we prefer to use vxor for it.
if (ValSizeInBytes < ByteSize) return SDOperand();
// If the element value is larger than the splat value, cut it in half and
// check to see if the two halves are equal. Continue doing this until we
// get to ByteSize. This allows us to handle 0x01010101 as 0x01.
while (ValSizeInBytes > ByteSize) {
ValSizeInBytes >>= 1;
// If the top half equals the bottom half, we're still ok.
if (((Value >> (ValSizeInBytes*8)) & ((1 << (8*ValSizeInBytes))-1)) !=
(Value & ((1 << (8*ValSizeInBytes))-1)))
return SDOperand();
}
// Properly sign extend the value.
int ShAmt = (4-ByteSize)*8;
int MaskVal = ((int)Value << ShAmt) >> ShAmt;
// If this is zero, don't match, zero matches ISD::isBuildVectorAllZeros.
if (MaskVal == 0) return SDOperand();
// Finally, if this value fits in a 5 bit sext field, return it
if (((MaskVal << (32-5)) >> (32-5)) == MaskVal)
return DAG.getTargetConstant(MaskVal, MVT::i32);
return SDOperand();
}
//===----------------------------------------------------------------------===//
// Addressing Mode Selection
//===----------------------------------------------------------------------===//
/// isIntS16Immediate - This method tests to see if the node is either a 32-bit
/// or 64-bit immediate, and if the value can be accurately represented as a
/// sign extension from a 16-bit value. If so, this returns true and the
/// immediate.
static bool isIntS16Immediate(SDNode *N, short &Imm) {
if (N->getOpcode() != ISD::Constant)
return false;
Imm = (short)cast<ConstantSDNode>(N)->getValue();
if (N->getValueType(0) == MVT::i32)
return Imm == (int32_t)cast<ConstantSDNode>(N)->getValue();
else
return Imm == (int64_t)cast<ConstantSDNode>(N)->getValue();
}
static bool isIntS16Immediate(SDOperand Op, short &Imm) {
return isIntS16Immediate(Op.Val, Imm);
}
/// SelectAddressRegReg - Given the specified addressed, check to see if it
/// can be represented as an indexed [r+r] operation. Returns false if it
/// can be more efficiently represented with [r+imm].
bool PPCTargetLowering::SelectAddressRegReg(SDOperand N, SDOperand &Base,
SDOperand &Index,
SelectionDAG &DAG) {
short imm = 0;
if (N.getOpcode() == ISD::ADD) {
if (isIntS16Immediate(N.getOperand(1), imm))
return false; // r+i
if (N.getOperand(1).getOpcode() == PPCISD::Lo)
return false; // r+i
Base = N.getOperand(0);
Index = N.getOperand(1);
return true;
} else if (N.getOpcode() == ISD::OR) {
if (isIntS16Immediate(N.getOperand(1), imm))
return false; // r+i can fold it if we can.
// If this is an or of disjoint bitfields, we can codegen this as an add
// (for better address arithmetic) if the LHS and RHS of the OR are provably
// disjoint.
uint64_t LHSKnownZero, LHSKnownOne;
uint64_t RHSKnownZero, RHSKnownOne;
DAG.ComputeMaskedBits(N.getOperand(0), ~0U, LHSKnownZero, LHSKnownOne);
if (LHSKnownZero) {
DAG.ComputeMaskedBits(N.getOperand(1), ~0U, RHSKnownZero, RHSKnownOne);
// If all of the bits are known zero on the LHS or RHS, the add won't
// carry.
if ((LHSKnownZero | RHSKnownZero) == ~0U) {
Base = N.getOperand(0);
Index = N.getOperand(1);
return true;
}
}
}
return false;
}
/// Returns true if the address N can be represented by a base register plus
/// a signed 16-bit displacement [r+imm], and if it is not better
/// represented as reg+reg.
bool PPCTargetLowering::SelectAddressRegImm(SDOperand N, SDOperand &Disp,
SDOperand &Base, SelectionDAG &DAG){
// If this can be more profitably realized as r+r, fail.
if (SelectAddressRegReg(N, Disp, Base, DAG))
return false;
if (N.getOpcode() == ISD::ADD) {
short imm = 0;
if (isIntS16Immediate(N.getOperand(1), imm)) {
Disp = DAG.getTargetConstant((int)imm & 0xFFFF, MVT::i32);
if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(N.getOperand(0))) {
Base = DAG.getTargetFrameIndex(FI->getIndex(), N.getValueType());
} else {
Base = N.getOperand(0);
}
return true; // [r+i]
} else if (N.getOperand(1).getOpcode() == PPCISD::Lo) {
// Match LOAD (ADD (X, Lo(G))).
assert(!cast<ConstantSDNode>(N.getOperand(1).getOperand(1))->getValue()
&& "Cannot handle constant offsets yet!");
Disp = N.getOperand(1).getOperand(0); // The global address.
assert(Disp.getOpcode() == ISD::TargetGlobalAddress ||
Disp.getOpcode() == ISD::TargetConstantPool ||
Disp.getOpcode() == ISD::TargetJumpTable);
Base = N.getOperand(0);
return true; // [&g+r]
}
} else if (N.getOpcode() == ISD::OR) {
short imm = 0;
if (isIntS16Immediate(N.getOperand(1), imm)) {
// If this is an or of disjoint bitfields, we can codegen this as an add
// (for better address arithmetic) if the LHS and RHS of the OR are
// provably disjoint.
uint64_t LHSKnownZero, LHSKnownOne;
DAG.ComputeMaskedBits(N.getOperand(0), ~0U, LHSKnownZero, LHSKnownOne);
if ((LHSKnownZero|~(unsigned)imm) == ~0U) {
// If all of the bits are known zero on the LHS or RHS, the add won't
// carry.
Base = N.getOperand(0);
Disp = DAG.getTargetConstant((int)imm & 0xFFFF, MVT::i32);
return true;
}
}
} else if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N)) {
// Loading from a constant address.
// If this address fits entirely in a 16-bit sext immediate field, codegen
// this as "d, 0"
short Imm;
if (isIntS16Immediate(CN, Imm)) {
Disp = DAG.getTargetConstant(Imm, CN->getValueType(0));
Base = DAG.getRegister(PPC::R0, CN->getValueType(0));
return true;
}
// Handle 32-bit sext immediates with LIS + addr mode.
if (CN->getValueType(0) == MVT::i32 ||
(int64_t)CN->getValue() == (int)CN->getValue()) {
int Addr = (int)CN->getValue();
// Otherwise, break this down into an LIS + disp.
Disp = DAG.getTargetConstant((short)Addr, MVT::i32);
Base = DAG.getTargetConstant((Addr - (signed short)Addr) >> 16, MVT::i32);
unsigned Opc = CN->getValueType(0) == MVT::i32 ? PPC::LIS : PPC::LIS8;
Base = SDOperand(DAG.getTargetNode(Opc, CN->getValueType(0), Base), 0);
return true;
}
}
Disp = DAG.getTargetConstant(0, getPointerTy());
if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(N))
Base = DAG.getTargetFrameIndex(FI->getIndex(), N.getValueType());
else
Base = N;
return true; // [r+0]
}
/// SelectAddressRegRegOnly - Given the specified addressed, force it to be
/// represented as an indexed [r+r] operation.
bool PPCTargetLowering::SelectAddressRegRegOnly(SDOperand N, SDOperand &Base,
SDOperand &Index,
SelectionDAG &DAG) {
// Check to see if we can easily represent this as an [r+r] address. This
// will fail if it thinks that the address is more profitably represented as
// reg+imm, e.g. where imm = 0.
if (SelectAddressRegReg(N, Base, Index, DAG))
return true;
// If the operand is an addition, always emit this as [r+r], since this is
// better (for code size, and execution, as the memop does the add for free)
// than emitting an explicit add.
if (N.getOpcode() == ISD::ADD) {
Base = N.getOperand(0);
Index = N.getOperand(1);
return true;
}
// Otherwise, do it the hard way, using R0 as the base register.
Base = DAG.getRegister(PPC::R0, N.getValueType());
Index = N;
return true;
}
/// SelectAddressRegImmShift - Returns true if the address N can be
/// represented by a base register plus a signed 14-bit displacement
/// [r+imm*4]. Suitable for use by STD and friends.
bool PPCTargetLowering::SelectAddressRegImmShift(SDOperand N, SDOperand &Disp,
SDOperand &Base,
SelectionDAG &DAG) {
// If this can be more profitably realized as r+r, fail.
if (SelectAddressRegReg(N, Disp, Base, DAG))
return false;
if (N.getOpcode() == ISD::ADD) {
short imm = 0;
if (isIntS16Immediate(N.getOperand(1), imm) && (imm & 3) == 0) {
Disp = DAG.getTargetConstant(((int)imm & 0xFFFF) >> 2, MVT::i32);
if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(N.getOperand(0))) {
Base = DAG.getTargetFrameIndex(FI->getIndex(), N.getValueType());
} else {
Base = N.getOperand(0);
}
return true; // [r+i]
} else if (N.getOperand(1).getOpcode() == PPCISD::Lo) {
// Match LOAD (ADD (X, Lo(G))).
assert(!cast<ConstantSDNode>(N.getOperand(1).getOperand(1))->getValue()
&& "Cannot handle constant offsets yet!");
Disp = N.getOperand(1).getOperand(0); // The global address.
assert(Disp.getOpcode() == ISD::TargetGlobalAddress ||
Disp.getOpcode() == ISD::TargetConstantPool ||
Disp.getOpcode() == ISD::TargetJumpTable);
Base = N.getOperand(0);
return true; // [&g+r]
}
} else if (N.getOpcode() == ISD::OR) {
short imm = 0;
if (isIntS16Immediate(N.getOperand(1), imm) && (imm & 3) == 0) {
// If this is an or of disjoint bitfields, we can codegen this as an add
// (for better address arithmetic) if the LHS and RHS of the OR are
// provably disjoint.
uint64_t LHSKnownZero, LHSKnownOne;
DAG.ComputeMaskedBits(N.getOperand(0), ~0U, LHSKnownZero, LHSKnownOne);
if ((LHSKnownZero|~(unsigned)imm) == ~0U) {
// If all of the bits are known zero on the LHS or RHS, the add won't
// carry.
Base = N.getOperand(0);
Disp = DAG.getTargetConstant(((int)imm & 0xFFFF) >> 2, MVT::i32);
return true;
}
}
} else if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N)) {
// Loading from a constant address. Verify low two bits are clear.
if ((CN->getValue() & 3) == 0) {
// If this address fits entirely in a 14-bit sext immediate field, codegen
// this as "d, 0"
short Imm;
if (isIntS16Immediate(CN, Imm)) {
Disp = DAG.getTargetConstant((unsigned short)Imm >> 2, getPointerTy());
Base = DAG.getRegister(PPC::R0, CN->getValueType(0));
return true;
}
// Fold the low-part of 32-bit absolute addresses into addr mode.
if (CN->getValueType(0) == MVT::i32 ||
(int64_t)CN->getValue() == (int)CN->getValue()) {
int Addr = (int)CN->getValue();
// Otherwise, break this down into an LIS + disp.
Disp = DAG.getTargetConstant((short)Addr >> 2, MVT::i32);
Base = DAG.getTargetConstant((Addr-(signed short)Addr) >> 16, MVT::i32);
unsigned Opc = CN->getValueType(0) == MVT::i32 ? PPC::LIS : PPC::LIS8;
Base = SDOperand(DAG.getTargetNode(Opc, CN->getValueType(0), Base), 0);
return true;
}
}
}
Disp = DAG.getTargetConstant(0, getPointerTy());
if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(N))
Base = DAG.getTargetFrameIndex(FI->getIndex(), N.getValueType());
else
Base = N;
return true; // [r+0]
}
/// getPreIndexedAddressParts - returns true by value, base pointer and
/// offset pointer and addressing mode by reference if the node's address
/// can be legally represented as pre-indexed load / store address.
bool PPCTargetLowering::getPreIndexedAddressParts(SDNode *N, SDOperand &Base,
SDOperand &Offset,
ISD::MemIndexedMode &AM,
SelectionDAG &DAG) {
// Disabled by default for now.
if (!EnablePPCPreinc) return false;
SDOperand Ptr;
MVT::ValueType VT;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
Ptr = LD->getBasePtr();
VT = LD->getMemoryVT();
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
ST = ST;
Ptr = ST->getBasePtr();
VT = ST->getMemoryVT();
} else
return false;
// PowerPC doesn't have preinc load/store instructions for vectors.
if (MVT::isVector(VT))
return false;
// TODO: Check reg+reg first.
// LDU/STU use reg+imm*4, others use reg+imm.
if (VT != MVT::i64) {
// reg + imm
if (!SelectAddressRegImm(Ptr, Offset, Base, DAG))
return false;
} else {
// reg + imm * 4.
if (!SelectAddressRegImmShift(Ptr, Offset, Base, DAG))
return false;
}
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
// PPC64 doesn't have lwau, but it does have lwaux. Reject preinc load of
// sext i32 to i64 when addr mode is r+i.
if (LD->getValueType(0) == MVT::i64 && LD->getMemoryVT() == MVT::i32 &&
LD->getExtensionType() == ISD::SEXTLOAD &&
isa<ConstantSDNode>(Offset))
return false;
}
AM = ISD::PRE_INC;
return true;
}
//===----------------------------------------------------------------------===//
// LowerOperation implementation
//===----------------------------------------------------------------------===//
static SDOperand LowerConstantPool(SDOperand Op, SelectionDAG &DAG) {
MVT::ValueType PtrVT = Op.getValueType();
ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
Constant *C = CP->getConstVal();
SDOperand CPI = DAG.getTargetConstantPool(C, PtrVT, CP->getAlignment());
SDOperand Zero = DAG.getConstant(0, PtrVT);
const TargetMachine &TM = DAG.getTarget();
SDOperand Hi = DAG.getNode(PPCISD::Hi, PtrVT, CPI, Zero);
SDOperand Lo = DAG.getNode(PPCISD::Lo, PtrVT, CPI, Zero);
// If this is a non-darwin platform, we don't support non-static relo models
// yet.
if (TM.getRelocationModel() == Reloc::Static ||
!TM.getSubtarget<PPCSubtarget>().isDarwin()) {
// Generate non-pic code that has direct accesses to the constant pool.
// The address of the global is just (hi(&g)+lo(&g)).
return DAG.getNode(ISD::ADD, PtrVT, Hi, Lo);
}
if (TM.getRelocationModel() == Reloc::PIC_) {
// With PIC, the first instruction is actually "GR+hi(&G)".
Hi = DAG.getNode(ISD::ADD, PtrVT,
DAG.getNode(PPCISD::GlobalBaseReg, PtrVT), Hi);
}
Lo = DAG.getNode(ISD::ADD, PtrVT, Hi, Lo);
return Lo;
}
static SDOperand LowerJumpTable(SDOperand Op, SelectionDAG &DAG) {
MVT::ValueType PtrVT = Op.getValueType();
JumpTableSDNode *JT = cast<JumpTableSDNode>(Op);
SDOperand JTI = DAG.getTargetJumpTable(JT->getIndex(), PtrVT);
SDOperand Zero = DAG.getConstant(0, PtrVT);
const TargetMachine &TM = DAG.getTarget();
SDOperand Hi = DAG.getNode(PPCISD::Hi, PtrVT, JTI, Zero);
SDOperand Lo = DAG.getNode(PPCISD::Lo, PtrVT, JTI, Zero);
// If this is a non-darwin platform, we don't support non-static relo models
// yet.
if (TM.getRelocationModel() == Reloc::Static ||
!TM.getSubtarget<PPCSubtarget>().isDarwin()) {
// Generate non-pic code that has direct accesses to the constant pool.
// The address of the global is just (hi(&g)+lo(&g)).
return DAG.getNode(ISD::ADD, PtrVT, Hi, Lo);
}
if (TM.getRelocationModel() == Reloc::PIC_) {
// With PIC, the first instruction is actually "GR+hi(&G)".
Hi = DAG.getNode(ISD::ADD, PtrVT,
DAG.getNode(PPCISD::GlobalBaseReg, PtrVT), Hi);
}
Lo = DAG.getNode(ISD::ADD, PtrVT, Hi, Lo);
return Lo;
}
static SDOperand LowerGlobalTLSAddress(SDOperand Op, SelectionDAG &DAG) {
assert(0 && "TLS not implemented for PPC.");
}
static SDOperand LowerGlobalAddress(SDOperand Op, SelectionDAG &DAG) {
MVT::ValueType PtrVT = Op.getValueType();
GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op);
GlobalValue *GV = GSDN->getGlobal();
SDOperand GA = DAG.getTargetGlobalAddress(GV, PtrVT, GSDN->getOffset());
SDOperand Zero = DAG.getConstant(0, PtrVT);
const TargetMachine &TM = DAG.getTarget();
SDOperand Hi = DAG.getNode(PPCISD::Hi, PtrVT, GA, Zero);
SDOperand Lo = DAG.getNode(PPCISD::Lo, PtrVT, GA, Zero);
// If this is a non-darwin platform, we don't support non-static relo models
// yet.
if (TM.getRelocationModel() == Reloc::Static ||
!TM.getSubtarget<PPCSubtarget>().isDarwin()) {
// Generate non-pic code that has direct accesses to globals.
// The address of the global is just (hi(&g)+lo(&g)).
return DAG.getNode(ISD::ADD, PtrVT, Hi, Lo);
}
if (TM.getRelocationModel() == Reloc::PIC_) {
// With PIC, the first instruction is actually "GR+hi(&G)".
Hi = DAG.getNode(ISD::ADD, PtrVT,
DAG.getNode(PPCISD::GlobalBaseReg, PtrVT), Hi);
}
Lo = DAG.getNode(ISD::ADD, PtrVT, Hi, Lo);
if (!TM.getSubtarget<PPCSubtarget>().hasLazyResolverStub(GV))
return Lo;
// If the global is weak or external, we have to go through the lazy
// resolution stub.
return DAG.getLoad(PtrVT, DAG.getEntryNode(), Lo, NULL, 0);
}
static SDOperand LowerSETCC(SDOperand Op, SelectionDAG &DAG) {
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
// If we're comparing for equality to zero, expose the fact that this is
// implented as a ctlz/srl pair on ppc, so that the dag combiner can
// fold the new nodes.
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
if (C->isNullValue() && CC == ISD::SETEQ) {
MVT::ValueType VT = Op.getOperand(0).getValueType();
SDOperand Zext = Op.getOperand(0);
if (VT < MVT::i32) {
VT = MVT::i32;
Zext = DAG.getNode(ISD::ZERO_EXTEND, VT, Op.getOperand(0));
}
unsigned Log2b = Log2_32(MVT::getSizeInBits(VT));
SDOperand Clz = DAG.getNode(ISD::CTLZ, VT, Zext);
SDOperand Scc = DAG.getNode(ISD::SRL, VT, Clz,
DAG.getConstant(Log2b, MVT::i32));
return DAG.getNode(ISD::TRUNCATE, MVT::i32, Scc);
}
// Leave comparisons against 0 and -1 alone for now, since they're usually
// optimized. FIXME: revisit this when we can custom lower all setcc
// optimizations.
if (C->isAllOnesValue() || C->isNullValue())
return SDOperand();
}
// If we have an integer seteq/setne, turn it into a compare against zero
// by xor'ing the rhs with the lhs, which is faster than setting a
// condition register, reading it back out, and masking the correct bit. The
// normal approach here uses sub to do this instead of xor. Using xor exposes
// the result to other bit-twiddling opportunities.
MVT::ValueType LHSVT = Op.getOperand(0).getValueType();
if (MVT::isInteger(LHSVT) && (CC == ISD::SETEQ || CC == ISD::SETNE)) {
MVT::ValueType VT = Op.getValueType();
SDOperand Sub = DAG.getNode(ISD::XOR, LHSVT, Op.getOperand(0),
Op.getOperand(1));
return DAG.getSetCC(VT, Sub, DAG.getConstant(0, LHSVT), CC);
}
return SDOperand();
}
static SDOperand LowerVAARG(SDOperand Op, SelectionDAG &DAG,
int VarArgsFrameIndex,
int VarArgsStackOffset,
unsigned VarArgsNumGPR,
unsigned VarArgsNumFPR,
const PPCSubtarget &Subtarget) {
assert(0 && "VAARG in ELF32 ABI not implemented yet!");
}
static SDOperand LowerVASTART(SDOperand Op, SelectionDAG &DAG,
int VarArgsFrameIndex,
int VarArgsStackOffset,
unsigned VarArgsNumGPR,
unsigned VarArgsNumFPR,
const PPCSubtarget &Subtarget) {
if (Subtarget.isMachoABI()) {
// vastart just stores the address of the VarArgsFrameIndex slot into the
// memory location argument.
MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDOperand FR = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
return DAG.getStore(Op.getOperand(0), FR, Op.getOperand(1), SV, 0);
}
// For ELF 32 ABI we follow the layout of the va_list struct.
// We suppose the given va_list is already allocated.
//
// typedef struct {
// char gpr; /* index into the array of 8 GPRs
// * stored in the register save area
// * gpr=0 corresponds to r3,
// * gpr=1 to r4, etc.
// */
// char fpr; /* index into the array of 8 FPRs
// * stored in the register save area
// * fpr=0 corresponds to f1,
// * fpr=1 to f2, etc.
// */
// char *overflow_arg_area;
// /* location on stack that holds
// * the next overflow argument
// */
// char *reg_save_area;
// /* where r3:r10 and f1:f8 (if saved)
// * are stored
// */
// } va_list[1];
SDOperand ArgGPR = DAG.getConstant(VarArgsNumGPR, MVT::i8);
SDOperand ArgFPR = DAG.getConstant(VarArgsNumFPR, MVT::i8);
MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDOperand StackOffsetFI = DAG.getFrameIndex(VarArgsStackOffset, PtrVT);
SDOperand FR = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
uint64_t FrameOffset = MVT::getSizeInBits(PtrVT)/8;
SDOperand ConstFrameOffset = DAG.getConstant(FrameOffset, PtrVT);
uint64_t StackOffset = MVT::getSizeInBits(PtrVT)/8 - 1;
SDOperand ConstStackOffset = DAG.getConstant(StackOffset, PtrVT);
uint64_t FPROffset = 1;
SDOperand ConstFPROffset = DAG.getConstant(FPROffset, PtrVT);
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
// Store first byte : number of int regs
SDOperand firstStore = DAG.getStore(Op.getOperand(0), ArgGPR,
Op.getOperand(1), SV, 0);
uint64_t nextOffset = FPROffset;
SDOperand nextPtr = DAG.getNode(ISD::ADD, PtrVT, Op.getOperand(1),
ConstFPROffset);
// Store second byte : number of float regs
SDOperand secondStore =
DAG.getStore(firstStore, ArgFPR, nextPtr, SV, nextOffset);
nextOffset += StackOffset;
nextPtr = DAG.getNode(ISD::ADD, PtrVT, nextPtr, ConstStackOffset);
// Store second word : arguments given on stack
SDOperand thirdStore =
DAG.getStore(secondStore, StackOffsetFI, nextPtr, SV, nextOffset);
nextOffset += FrameOffset;
nextPtr = DAG.getNode(ISD::ADD, PtrVT, nextPtr, ConstFrameOffset);
// Store third word : arguments given in registers
return DAG.getStore(thirdStore, FR, nextPtr, SV, nextOffset);
}
#include "PPCGenCallingConv.inc"
/// GetFPR - Get the set of FP registers that should be allocated for arguments,
/// depending on which subtarget is selected.
static const unsigned *GetFPR(const PPCSubtarget &Subtarget) {
if (Subtarget.isMachoABI()) {
static const unsigned FPR[] = {
PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
PPC::F8, PPC::F9, PPC::F10, PPC::F11, PPC::F12, PPC::F13
};
return FPR;
}
static const unsigned FPR[] = {
PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
PPC::F8
};
return FPR;
}
static SDOperand LowerFORMAL_ARGUMENTS(SDOperand Op, SelectionDAG &DAG,
int &VarArgsFrameIndex,
int &VarArgsStackOffset,
unsigned &VarArgsNumGPR,
unsigned &VarArgsNumFPR,
const PPCSubtarget &Subtarget) {
// TODO: add description of PPC stack frame format, or at least some docs.
//
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
SmallVector<SDOperand, 8> ArgValues;
SDOperand Root = Op.getOperand(0);
MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = PtrVT == MVT::i64;
bool isMachoABI = Subtarget.isMachoABI();
bool isELF32_ABI = Subtarget.isELF32_ABI();
unsigned PtrByteSize = isPPC64 ? 8 : 4;
unsigned ArgOffset = PPCFrameInfo::getLinkageSize(isPPC64, isMachoABI);
static const unsigned GPR_32[] = { // 32-bit registers.
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10,
};
static const unsigned GPR_64[] = { // 64-bit registers.
PPC::X3, PPC::X4, PPC::X5, PPC::X6,
PPC::X7, PPC::X8, PPC::X9, PPC::X10,
};
static const unsigned *FPR = GetFPR(Subtarget);
static const unsigned VR[] = {
PPC::V2, PPC::V3, PPC::V4, PPC::V5, PPC::V6, PPC::V7, PPC::V8,
PPC::V9, PPC::V10, PPC::V11, PPC::V12, PPC::V13
};
const unsigned Num_GPR_Regs = array_lengthof(GPR_32);
const unsigned Num_FPR_Regs = isMachoABI ? 13 : 8;
const unsigned Num_VR_Regs = array_lengthof( VR);
unsigned GPR_idx = 0, FPR_idx = 0, VR_idx = 0;
const unsigned *GPR = isPPC64 ? GPR_64 : GPR_32;
// Add DAG nodes to load the arguments or copy them out of registers. On
// entry to a function on PPC, the arguments start after the linkage area,
// although the first ones are often in registers.
//
// In the ELF 32 ABI, GPRs and stack are double word align: an argument
// represented with two words (long long or double) must be copied to an
// even GPR_idx value or to an even ArgOffset value.
for (unsigned ArgNo = 0, e = Op.Val->getNumValues()-1; ArgNo != e; ++ArgNo) {
SDOperand ArgVal;
bool needsLoad = false;
MVT::ValueType ObjectVT = Op.getValue(ArgNo).getValueType();
unsigned ObjSize = MVT::getSizeInBits(ObjectVT)/8;
unsigned ArgSize = ObjSize;
unsigned Flags = cast<ConstantSDNode>(Op.getOperand(ArgNo+3))->getValue();
unsigned AlignFlag = 1 << ISD::ParamFlags::OrigAlignmentOffs;
// See if next argument requires stack alignment in ELF
bool Expand = (ObjectVT == MVT::f64) || ((ArgNo + 1 < e) &&
(cast<ConstantSDNode>(Op.getOperand(ArgNo+4))->getValue() & AlignFlag) &&
(!(Flags & AlignFlag)));
unsigned CurArgOffset = ArgOffset;
switch (ObjectVT) {
default: assert(0 && "Unhandled argument type!");
case MVT::i32:
// Double word align in ELF
if (Expand && isELF32_ABI) GPR_idx += (GPR_idx % 2);
if (GPR_idx != Num_GPR_Regs) {
unsigned VReg = RegInfo.createVirtualRegister(&PPC::GPRCRegClass);
RegInfo.addLiveIn(GPR[GPR_idx], VReg);
ArgVal = DAG.getCopyFromReg(Root, VReg, MVT::i32);
++GPR_idx;
} else {
needsLoad = true;
ArgSize = PtrByteSize;
}
// Stack align in ELF
if (needsLoad && Expand && isELF32_ABI)
ArgOffset += ((ArgOffset/4) % 2) * PtrByteSize;
// All int arguments reserve stack space in Macho ABI.
if (isMachoABI || needsLoad) ArgOffset += PtrByteSize;
break;
case MVT::i64: // PPC64
if (GPR_idx != Num_GPR_Regs) {
unsigned VReg = RegInfo.createVirtualRegister(&PPC::G8RCRegClass);
RegInfo.addLiveIn(GPR[GPR_idx], VReg);
ArgVal = DAG.getCopyFromReg(Root, VReg, MVT::i64);
++GPR_idx;
} else {
needsLoad = true;
}
// All int arguments reserve stack space in Macho ABI.
if (isMachoABI || needsLoad) ArgOffset += 8;
break;
case MVT::f32:
case MVT::f64:
// Every 4 bytes of argument space consumes one of the GPRs available for
// argument passing.
if (GPR_idx != Num_GPR_Regs && isMachoABI) {
++GPR_idx;
if (ObjSize == 8 && GPR_idx != Num_GPR_Regs && !isPPC64)
++GPR_idx;
}
if (FPR_idx != Num_FPR_Regs) {
unsigned VReg;
if (ObjectVT == MVT::f32)
VReg = RegInfo.createVirtualRegister(&PPC::F4RCRegClass);
else
VReg = RegInfo.createVirtualRegister(&PPC::F8RCRegClass);
RegInfo.addLiveIn(FPR[FPR_idx], VReg);
ArgVal = DAG.getCopyFromReg(Root, VReg, ObjectVT);
++FPR_idx;
} else {
needsLoad = true;
}
// Stack align in ELF
if (needsLoad && Expand && isELF32_ABI)
ArgOffset += ((ArgOffset/4) % 2) * PtrByteSize;
// All FP arguments reserve stack space in Macho ABI.
if (isMachoABI || needsLoad) ArgOffset += isPPC64 ? 8 : ObjSize;
break;
case MVT::v4f32:
case MVT::v4i32:
case MVT::v8i16:
case MVT::v16i8:
// Note that vector arguments in registers don't reserve stack space.
if (VR_idx != Num_VR_Regs) {
unsigned VReg = RegInfo.createVirtualRegister(&PPC::VRRCRegClass);
RegInfo.addLiveIn(VR[VR_idx], VReg);
ArgVal = DAG.getCopyFromReg(Root, VReg, ObjectVT);
++VR_idx;
} else {
// This should be simple, but requires getting 16-byte aligned stack
// values.
assert(0 && "Loading VR argument not implemented yet!");
needsLoad = true;
}
break;
}
// We need to load the argument to a virtual register if we determined above
// that we ran out of physical registers of the appropriate type
if (needsLoad) {
// If the argument is actually used, emit a load from the right stack
// slot.
if (!Op.Val->hasNUsesOfValue(0, ArgNo)) {
int FI = MFI->CreateFixedObject(ObjSize,
CurArgOffset + (ArgSize - ObjSize));
SDOperand FIN = DAG.getFrameIndex(FI, PtrVT);
ArgVal = DAG.getLoad(ObjectVT, Root, FIN, NULL, 0);
} else {
// Don't emit a dead load.
ArgVal = DAG.getNode(ISD::UNDEF, ObjectVT);
}
}
ArgValues.push_back(ArgVal);
}
// If the function takes variable number of arguments, make a frame index for
// the start of the first vararg value... for expansion of llvm.va_start.
bool isVarArg = cast<ConstantSDNode>(Op.getOperand(2))->getValue() != 0;
if (isVarArg) {
int depth;
if (isELF32_ABI) {
VarArgsNumGPR = GPR_idx;
VarArgsNumFPR = FPR_idx;
// Make room for Num_GPR_Regs, Num_FPR_Regs and for a possible frame
// pointer.
depth = -(Num_GPR_Regs * MVT::getSizeInBits(PtrVT)/8 +
Num_FPR_Regs * MVT::getSizeInBits(MVT::f64)/8 +
MVT::getSizeInBits(PtrVT)/8);
VarArgsStackOffset = MFI->CreateFixedObject(MVT::getSizeInBits(PtrVT)/8,
ArgOffset);
}
else
depth = ArgOffset;
VarArgsFrameIndex = MFI->CreateFixedObject(MVT::getSizeInBits(PtrVT)/8,
depth);
SDOperand FIN = DAG.getFrameIndex(VarArgsFrameIndex, PtrVT);
SmallVector<SDOperand, 8> MemOps;
// In ELF 32 ABI, the fixed integer arguments of a variadic function are
// stored to the VarArgsFrameIndex on the stack.
if (isELF32_ABI) {
for (GPR_idx = 0; GPR_idx != VarArgsNumGPR; ++GPR_idx) {
SDOperand Val = DAG.getRegister(GPR[GPR_idx], PtrVT);
SDOperand Store = DAG.getStore(Root, Val, FIN, NULL, 0);
MemOps.push_back(Store);
// Increment the address by four for the next argument to store
SDOperand PtrOff = DAG.getConstant(MVT::getSizeInBits(PtrVT)/8, PtrVT);
FIN = DAG.getNode(ISD::ADD, PtrOff.getValueType(), FIN, PtrOff);
}
}
// If this function is vararg, store any remaining integer argument regs
// to their spots on the stack so that they may be loaded by deferencing the
// result of va_next.
for (; GPR_idx != Num_GPR_Regs; ++GPR_idx) {
unsigned VReg;
if (isPPC64)
VReg = RegInfo.createVirtualRegister(&PPC::G8RCRegClass);
else
VReg = RegInfo.createVirtualRegister(&PPC::GPRCRegClass);
RegInfo.addLiveIn(GPR[GPR_idx], VReg);
SDOperand Val = DAG.getCopyFromReg(Root, VReg, PtrVT);
SDOperand Store = DAG.getStore(Val.getValue(1), Val, FIN, NULL, 0);
MemOps.push_back(Store);
// Increment the address by four for the next argument to store
SDOperand PtrOff = DAG.getConstant(MVT::getSizeInBits(PtrVT)/8, PtrVT);
FIN = DAG.getNode(ISD::ADD, PtrOff.getValueType(), FIN, PtrOff);
}
// In ELF 32 ABI, the double arguments are stored to the VarArgsFrameIndex
// on the stack.
if (isELF32_ABI) {
for (FPR_idx = 0; FPR_idx != VarArgsNumFPR; ++FPR_idx) {
SDOperand Val = DAG.getRegister(FPR[FPR_idx], MVT::f64);
SDOperand Store = DAG.getStore(Root, Val, FIN, NULL, 0);
MemOps.push_back(Store);
// Increment the address by eight for the next argument to store
SDOperand PtrOff = DAG.getConstant(MVT::getSizeInBits(MVT::f64)/8,
PtrVT);
FIN = DAG.getNode(ISD::ADD, PtrOff.getValueType(), FIN, PtrOff);
}
for (; FPR_idx != Num_FPR_Regs; ++FPR_idx) {
unsigned VReg;
VReg = RegInfo.createVirtualRegister(&PPC::F8RCRegClass);
RegInfo.addLiveIn(FPR[FPR_idx], VReg);
SDOperand Val = DAG.getCopyFromReg(Root, VReg, MVT::f64);
SDOperand Store = DAG.getStore(Val.getValue(1), Val, FIN, NULL, 0);
MemOps.push_back(Store);
// Increment the address by eight for the next argument to store
SDOperand PtrOff = DAG.getConstant(MVT::getSizeInBits(MVT::f64)/8,
PtrVT);
FIN = DAG.getNode(ISD::ADD, PtrOff.getValueType(), FIN, PtrOff);
}
}
if (!MemOps.empty())
Root = DAG.getNode(ISD::TokenFactor, MVT::Other,&MemOps[0],MemOps.size());
}
ArgValues.push_back(Root);
// Return the new list of results.
std::vector<MVT::ValueType> RetVT(Op.Val->value_begin(),
Op.Val->value_end());
return DAG.getNode(ISD::MERGE_VALUES, RetVT, &ArgValues[0], ArgValues.size());
}
/// isCallCompatibleAddress - Return the immediate to use if the specified
/// 32-bit value is representable in the immediate field of a BxA instruction.
static SDNode *isBLACompatibleAddress(SDOperand Op, SelectionDAG &DAG) {
ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op);
if (!C) return 0;
int Addr = C->getValue();
if ((Addr & 3) != 0 || // Low 2 bits are implicitly zero.
(Addr << 6 >> 6) != Addr)
return 0; // Top 6 bits have to be sext of immediate.
return DAG.getConstant((int)C->getValue() >> 2,
DAG.getTargetLoweringInfo().getPointerTy()).Val;
}
static SDOperand LowerCALL(SDOperand Op, SelectionDAG &DAG,
const PPCSubtarget &Subtarget) {
SDOperand Chain = Op.getOperand(0);
bool isVarArg = cast<ConstantSDNode>(Op.getOperand(2))->getValue() != 0;
SDOperand Callee = Op.getOperand(4);
unsigned NumOps = (Op.getNumOperands() - 5) / 2;
bool isMachoABI = Subtarget.isMachoABI();
bool isELF32_ABI = Subtarget.isELF32_ABI();
MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = PtrVT == MVT::i64;
unsigned PtrByteSize = isPPC64 ? 8 : 4;
// args_to_use will accumulate outgoing args for the PPCISD::CALL case in
// SelectExpr to use to put the arguments in the appropriate registers.
std::vector<SDOperand> args_to_use;
// Count how many bytes are to be pushed on the stack, including the linkage
// area, and parameter passing area. We start with 24/48 bytes, which is
// prereserved space for [SP][CR][LR][3 x unused].
unsigned NumBytes = PPCFrameInfo::getLinkageSize(isPPC64, isMachoABI);
// Add up all the space actually used.
for (unsigned i = 0; i != NumOps; ++i) {
unsigned ArgSize =MVT::getSizeInBits(Op.getOperand(5+2*i).getValueType())/8;
ArgSize = std::max(ArgSize, PtrByteSize);
NumBytes += ArgSize;
}
// The prolog code of the callee may store up to 8 GPR argument registers to
// the stack, allowing va_start to index over them in memory if its varargs.
// Because we cannot tell if this is needed on the caller side, we have to
// conservatively assume that it is needed. As such, make sure we have at
// least enough stack space for the caller to store the 8 GPRs.
NumBytes = std::max(NumBytes,
PPCFrameInfo::getMinCallFrameSize(isPPC64, isMachoABI));
// Adjust the stack pointer for the new arguments...
// These operations are automatically eliminated by the prolog/epilog pass
Chain = DAG.getCALLSEQ_START(Chain,
DAG.getConstant(NumBytes, PtrVT));
// Set up a copy of the stack pointer for use loading and storing any
// arguments that may not fit in the registers available for argument
// passing.
SDOperand StackPtr;
if (isPPC64)
StackPtr = DAG.getRegister(PPC::X1, MVT::i64);
else
StackPtr = DAG.getRegister(PPC::R1, MVT::i32);
// Figure out which arguments are going to go in registers, and which in
// memory. Also, if this is a vararg function, floating point operations
// must be stored to our stack, and loaded into integer regs as well, if
// any integer regs are available for argument passing.
unsigned ArgOffset = PPCFrameInfo::getLinkageSize(isPPC64, isMachoABI);
unsigned GPR_idx = 0, FPR_idx = 0, VR_idx = 0;
static const unsigned GPR_32[] = { // 32-bit registers.
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10,
};
static const unsigned GPR_64[] = { // 64-bit registers.
PPC::X3, PPC::X4, PPC::X5, PPC::X6,
PPC::X7, PPC::X8, PPC::X9, PPC::X10,
};
static const unsigned *FPR = GetFPR(Subtarget);
static const unsigned VR[] = {
PPC::V2, PPC::V3, PPC::V4, PPC::V5, PPC::V6, PPC::V7, PPC::V8,
PPC::V9, PPC::V10, PPC::V11, PPC::V12, PPC::V13
};
const unsigned NumGPRs = array_lengthof(GPR_32);
const unsigned NumFPRs = isMachoABI ? 13 : 8;
const unsigned NumVRs = array_lengthof( VR);
const unsigned *GPR = isPPC64 ? GPR_64 : GPR_32;
std::vector<std::pair<unsigned, SDOperand> > RegsToPass;
SmallVector<SDOperand, 8> MemOpChains;
for (unsigned i = 0; i != NumOps; ++i) {
bool inMem = false;
SDOperand Arg = Op.getOperand(5+2*i);
unsigned Flags = cast<ConstantSDNode>(Op.getOperand(5+2*i+1))->getValue();
unsigned AlignFlag = 1 << ISD::ParamFlags::OrigAlignmentOffs;
// See if next argument requires stack alignment in ELF
unsigned next = 5+2*(i+1)+1;
bool Expand = (Arg.getValueType() == MVT::f64) || ((i + 1 < NumOps) &&
(cast<ConstantSDNode>(Op.getOperand(next))->getValue() & AlignFlag) &&
(!(Flags & AlignFlag)));
// PtrOff will be used to store the current argument to the stack if a
// register cannot be found for it.
SDOperand PtrOff;
// Stack align in ELF 32
if (isELF32_ABI && Expand)
PtrOff = DAG.getConstant(ArgOffset + ((ArgOffset/4) % 2) * PtrByteSize,
StackPtr.getValueType());
else
PtrOff = DAG.getConstant(ArgOffset, StackPtr.getValueType());
PtrOff = DAG.getNode(ISD::ADD, PtrVT, StackPtr, PtrOff);
// On PPC64, promote integers to 64-bit values.
if (isPPC64 && Arg.getValueType() == MVT::i32) {
unsigned ExtOp = (Flags & 1) ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND;
Arg = DAG.getNode(ExtOp, MVT::i64, Arg);
}
switch (Arg.getValueType()) {
default: assert(0 && "Unexpected ValueType for argument!");
case MVT::i32:
case MVT::i64:
// Double word align in ELF
if (isELF32_ABI && Expand) GPR_idx += (GPR_idx % 2);
if (GPR_idx != NumGPRs) {
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Arg));
} else {
MemOpChains.push_back(DAG.getStore(Chain, Arg, PtrOff, NULL, 0));
inMem = true;
}
if (inMem || isMachoABI) {
// Stack align in ELF
if (isELF32_ABI && Expand)
ArgOffset += ((ArgOffset/4) % 2) * PtrByteSize;
ArgOffset += PtrByteSize;
}
break;
case MVT::f32:
case MVT::f64:
if (isVarArg) {
// Float varargs need to be promoted to double.
if (Arg.getValueType() == MVT::f32)
Arg = DAG.getNode(ISD::FP_EXTEND, MVT::f64, Arg);
}
if (FPR_idx != NumFPRs) {
RegsToPass.push_back(std::make_pair(FPR[FPR_idx++], Arg));
if (isVarArg) {
SDOperand Store = DAG.getStore(Chain, Arg, PtrOff, NULL, 0);
MemOpChains.push_back(Store);
// Float varargs are always shadowed in available integer registers
if (GPR_idx != NumGPRs) {
SDOperand Load = DAG.getLoad(PtrVT, Store, PtrOff, NULL, 0);
MemOpChains.push_back(Load.getValue(1));
if (isMachoABI) RegsToPass.push_back(std::make_pair(GPR[GPR_idx++],
Load));
}
if (GPR_idx != NumGPRs && Arg.getValueType() == MVT::f64 && !isPPC64){
SDOperand ConstFour = DAG.getConstant(4, PtrOff.getValueType());
PtrOff = DAG.getNode(ISD::ADD, PtrVT, PtrOff, ConstFour);
SDOperand Load = DAG.getLoad(PtrVT, Store, PtrOff, NULL, 0);
MemOpChains.push_back(Load.getValue(1));
if (isMachoABI) RegsToPass.push_back(std::make_pair(GPR[GPR_idx++],
Load));
}
} else {
// If we have any FPRs remaining, we may also have GPRs remaining.
// Args passed in FPRs consume either 1 (f32) or 2 (f64) available
// GPRs.
if (isMachoABI) {
if (GPR_idx != NumGPRs)
++GPR_idx;
if (GPR_idx != NumGPRs && Arg.getValueType() == MVT::f64 &&
!isPPC64) // PPC64 has 64-bit GPR's obviously :)
++GPR_idx;
}
}
} else {
MemOpChains.push_back(DAG.getStore(Chain, Arg, PtrOff, NULL, 0));
inMem = true;
}
if (inMem || isMachoABI) {
// Stack align in ELF
if (isELF32_ABI && Expand)
ArgOffset += ((ArgOffset/4) % 2) * PtrByteSize;
if (isPPC64)
ArgOffset += 8;
else
ArgOffset += Arg.getValueType() == MVT::f32 ? 4 : 8;
}
break;
case MVT::v4f32:
case MVT::v4i32:
case MVT::v8i16:
case MVT::v16i8:
assert(!isVarArg && "Don't support passing vectors to varargs yet!");
assert(VR_idx != NumVRs &&
"Don't support passing more than 12 vector args yet!");
RegsToPass.push_back(std::make_pair(VR[VR_idx++], Arg));
break;
}
}
if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, MVT::Other,
&MemOpChains[0], MemOpChains.size());
// Build a sequence of copy-to-reg nodes chained together with token chain
// and flag operands which copy the outgoing args into the appropriate regs.
SDOperand InFlag;
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
Chain = DAG.getCopyToReg(Chain, RegsToPass[i].first, RegsToPass[i].second,
InFlag);
InFlag = Chain.getValue(1);
}
// With the ELF 32 ABI, set CR6 to true if this is a vararg call.
if (isVarArg && isELF32_ABI) {
SDOperand SetCR(DAG.getTargetNode(PPC::SETCR, MVT::i32), 0);
Chain = DAG.getCopyToReg(Chain, PPC::CR6, SetCR, InFlag);
InFlag = Chain.getValue(1);
}
std::vector<MVT::ValueType> NodeTys;
NodeTys.push_back(MVT::Other); // Returns a chain
NodeTys.push_back(MVT::Flag); // Returns a flag for retval copy to use.
SmallVector<SDOperand, 8> Ops;
unsigned CallOpc = isMachoABI? PPCISD::CALL_Macho : PPCISD::CALL_ELF;
// If the callee is a GlobalAddress/ExternalSymbol node (quite common, every
// direct call is) turn it into a TargetGlobalAddress/TargetExternalSymbol
// node so that legalize doesn't hack it.
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
Callee = DAG.getTargetGlobalAddress(G->getGlobal(), Callee.getValueType());
else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee))
Callee = DAG.getTargetExternalSymbol(S->getSymbol(), Callee.getValueType());
else if (SDNode *Dest = isBLACompatibleAddress(Callee, DAG))
// If this is an absolute destination address, use the munged value.
Callee = SDOperand(Dest, 0);
else {
// Otherwise, this is an indirect call. We have to use a MTCTR/BCTRL pair
// to do the call, we can't use PPCISD::CALL.
SDOperand MTCTROps[] = {Chain, Callee, InFlag};
Chain = DAG.getNode(PPCISD::MTCTR, NodeTys, MTCTROps, 2+(InFlag.Val!=0));
InFlag = Chain.getValue(1);
// Copy the callee address into R12 on darwin.
if (isMachoABI) {
Chain = DAG.getCopyToReg(Chain, PPC::R12, Callee, InFlag);
InFlag = Chain.getValue(1);
}
NodeTys.clear();
NodeTys.push_back(MVT::Other);
NodeTys.push_back(MVT::Flag);
Ops.push_back(Chain);
CallOpc = isMachoABI ? PPCISD::BCTRL_Macho : PPCISD::BCTRL_ELF;
Callee.Val = 0;
}
// If this is a direct call, pass the chain and the callee.
if (Callee.Val) {
Ops.push_back(Chain);
Ops.push_back(Callee);
}
// Add argument registers to the end of the list so that they are known live
// into the call.
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i)
Ops.push_back(DAG.getRegister(RegsToPass[i].first,
RegsToPass[i].second.getValueType()));
if (InFlag.Val)
Ops.push_back(InFlag);
Chain = DAG.getNode(CallOpc, NodeTys, &Ops[0], Ops.size());
InFlag = Chain.getValue(1);
Chain = DAG.getCALLSEQ_END(Chain,
DAG.getConstant(NumBytes, PtrVT),
DAG.getConstant(0, PtrVT),
InFlag);
if (Op.Val->getValueType(0) != MVT::Other)
InFlag = Chain.getValue(1);
SDOperand ResultVals[3];
unsigned NumResults = 0;
NodeTys.clear();
// If the call has results, copy the values out of the ret val registers.
switch (Op.Val->getValueType(0)) {
default: assert(0 && "Unexpected ret value!");
case MVT::Other: break;
case MVT::i32:
if (Op.Val->getValueType(1) == MVT::i32) {
Chain = DAG.getCopyFromReg(Chain, PPC::R3, MVT::i32, InFlag).getValue(1);
ResultVals[0] = Chain.getValue(0);
Chain = DAG.getCopyFromReg(Chain, PPC::R4, MVT::i32,
Chain.getValue(2)).getValue(1);
ResultVals[1] = Chain.getValue(0);
NumResults = 2;
NodeTys.push_back(MVT::i32);
} else {
Chain = DAG.getCopyFromReg(Chain, PPC::R3, MVT::i32, InFlag).getValue(1);
ResultVals[0] = Chain.getValue(0);
NumResults = 1;
}
NodeTys.push_back(MVT::i32);
break;
case MVT::i64:
Chain = DAG.getCopyFromReg(Chain, PPC::X3, MVT::i64, InFlag).getValue(1);
ResultVals[0] = Chain.getValue(0);
NumResults = 1;
NodeTys.push_back(MVT::i64);
break;
case MVT::f64:
if (Op.Val->getValueType(1) == MVT::f64) {
Chain = DAG.getCopyFromReg(Chain, PPC::F1, MVT::f64, InFlag).getValue(1);
ResultVals[0] = Chain.getValue(0);
Chain = DAG.getCopyFromReg(Chain, PPC::F2, MVT::f64,
Chain.getValue(2)).getValue(1);
ResultVals[1] = Chain.getValue(0);
NumResults = 2;
NodeTys.push_back(MVT::f64);
NodeTys.push_back(MVT::f64);
break;
}
// else fall through
case MVT::f32:
Chain = DAG.getCopyFromReg(Chain, PPC::F1, Op.Val->getValueType(0),
InFlag).getValue(1);
ResultVals[0] = Chain.getValue(0);
NumResults = 1;
NodeTys.push_back(Op.Val->getValueType(0));
break;
case MVT::v4f32:
case MVT::v4i32:
case MVT::v8i16:
case MVT::v16i8:
Chain = DAG.getCopyFromReg(Chain, PPC::V2, Op.Val->getValueType(0),
InFlag).getValue(1);
ResultVals[0] = Chain.getValue(0);
NumResults = 1;
NodeTys.push_back(Op.Val->getValueType(0));
break;
}
NodeTys.push_back(MVT::Other);
// If the function returns void, just return the chain.
if (NumResults == 0)
return Chain;
// Otherwise, merge everything together with a MERGE_VALUES node.
ResultVals[NumResults++] = Chain;
SDOperand Res = DAG.getNode(ISD::MERGE_VALUES, NodeTys,
ResultVals, NumResults);
return Res.getValue(Op.ResNo);
}
static SDOperand LowerRET(SDOperand Op, SelectionDAG &DAG, TargetMachine &TM) {
SmallVector<CCValAssign, 16> RVLocs;
unsigned CC = DAG.getMachineFunction().getFunction()->getCallingConv();
bool isVarArg = DAG.getMachineFunction().getFunction()->isVarArg();
CCState CCInfo(CC, isVarArg, TM, RVLocs);
CCInfo.AnalyzeReturn(Op.Val, RetCC_PPC);
// If this is the first return lowered for this function, add the regs to the
// liveout set for the function.
if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
for (unsigned i = 0; i != RVLocs.size(); ++i)
DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
}
SDOperand Chain = Op.getOperand(0);
SDOperand Flag;
// Copy the result values into the output registers.
for (unsigned i = 0; i != RVLocs.size(); ++i) {
CCValAssign &VA = RVLocs[i];
assert(VA.isRegLoc() && "Can only return in registers!");
Chain = DAG.getCopyToReg(Chain, VA.getLocReg(), Op.getOperand(i*2+1), Flag);
Flag = Chain.getValue(1);
}
if (Flag.Val)
return DAG.getNode(PPCISD::RET_FLAG, MVT::Other, Chain, Flag);
else
return DAG.getNode(PPCISD::RET_FLAG, MVT::Other, Chain);
}
static SDOperand LowerSTACKRESTORE(SDOperand Op, SelectionDAG &DAG,
const PPCSubtarget &Subtarget) {
// When we pop the dynamic allocation we need to restore the SP link.
// Get the corect type for pointers.
MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Construct the stack pointer operand.
bool IsPPC64 = Subtarget.isPPC64();
unsigned SP = IsPPC64 ? PPC::X1 : PPC::R1;
SDOperand StackPtr = DAG.getRegister(SP, PtrVT);
// Get the operands for the STACKRESTORE.
SDOperand Chain = Op.getOperand(0);
SDOperand SaveSP = Op.getOperand(1);
// Load the old link SP.
SDOperand LoadLinkSP = DAG.getLoad(PtrVT, Chain, StackPtr, NULL, 0);
// Restore the stack pointer.
Chain = DAG.getCopyToReg(LoadLinkSP.getValue(1), SP, SaveSP);
// Store the old link SP.
return DAG.getStore(Chain, LoadLinkSP, StackPtr, NULL, 0);
}
static SDOperand LowerDYNAMIC_STACKALLOC(SDOperand Op, SelectionDAG &DAG,
const PPCSubtarget &Subtarget) {
MachineFunction &MF = DAG.getMachineFunction();
bool IsPPC64 = Subtarget.isPPC64();
bool isMachoABI = Subtarget.isMachoABI();
// Get current frame pointer save index. The users of this index will be
// primarily DYNALLOC instructions.
PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
int FPSI = FI->getFramePointerSaveIndex();
// If the frame pointer save index hasn't been defined yet.
if (!FPSI) {
// Find out what the fix offset of the frame pointer save area.
int FPOffset = PPCFrameInfo::getFramePointerSaveOffset(IsPPC64, isMachoABI);
// Allocate the frame index for frame pointer save area.
FPSI = MF.getFrameInfo()->CreateFixedObject(IsPPC64? 8 : 4, FPOffset);
// Save the result.
FI->setFramePointerSaveIndex(FPSI);
}
// Get the inputs.
SDOperand Chain = Op.getOperand(0);
SDOperand Size = Op.getOperand(1);
// Get the corect type for pointers.
MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
// Negate the size.
SDOperand NegSize = DAG.getNode(ISD::SUB, PtrVT,
DAG.getConstant(0, PtrVT), Size);
// Construct a node for the frame pointer save index.
SDOperand FPSIdx = DAG.getFrameIndex(FPSI, PtrVT);
// Build a DYNALLOC node.
SDOperand Ops[3] = { Chain, NegSize, FPSIdx };
SDVTList VTs = DAG.getVTList(PtrVT, MVT::Other);
return DAG.getNode(PPCISD::DYNALLOC, VTs, Ops, 3);
}
/// LowerSELECT_CC - Lower floating point select_cc's into fsel instruction when
/// possible.
static SDOperand LowerSELECT_CC(SDOperand Op, SelectionDAG &DAG) {
// Not FP? Not a fsel.
if (!MVT::isFloatingPoint(Op.getOperand(0).getValueType()) ||
!MVT::isFloatingPoint(Op.getOperand(2).getValueType()))
return SDOperand();
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(4))->get();
// Cannot handle SETEQ/SETNE.
if (CC == ISD::SETEQ || CC == ISD::SETNE) return SDOperand();
MVT::ValueType ResVT = Op.getValueType();
MVT::ValueType CmpVT = Op.getOperand(0).getValueType();
SDOperand LHS = Op.getOperand(0), RHS = Op.getOperand(1);
SDOperand TV = Op.getOperand(2), FV = Op.getOperand(3);
// If the RHS of the comparison is a 0.0, we don't need to do the
// subtraction at all.
if (isFloatingPointZero(RHS))
switch (CC) {
default: break; // SETUO etc aren't handled by fsel.
case ISD::SETULT:
case ISD::SETOLT:
case ISD::SETLT:
std::swap(TV, FV); // fsel is natively setge, swap operands for setlt
case ISD::SETUGE:
case ISD::SETOGE:
case ISD::SETGE:
if (LHS.getValueType() == MVT::f32) // Comparison is always 64-bits
LHS = DAG.getNode(ISD::FP_EXTEND, MVT::f64, LHS);
return DAG.getNode(PPCISD::FSEL, ResVT, LHS, TV, FV);
case ISD::SETUGT:
case ISD::SETOGT:
case ISD::SETGT:
std::swap(TV, FV); // fsel is natively setge, swap operands for setlt
case ISD::SETULE:
case ISD::SETOLE:
case ISD::SETLE:
if (LHS.getValueType() == MVT::f32) // Comparison is always 64-bits
LHS = DAG.getNode(ISD::FP_EXTEND, MVT::f64, LHS);
return DAG.getNode(PPCISD::FSEL, ResVT,
DAG.getNode(ISD::FNEG, MVT::f64, LHS), TV, FV);
}
SDOperand Cmp;
switch (CC) {
default: break; // SETUO etc aren't handled by fsel.
case ISD::SETULT:
case ISD::SETOLT:
case ISD::SETLT:
Cmp = DAG.getNode(ISD::FSUB, CmpVT, LHS, RHS);
if (Cmp.getValueType() == MVT::f32) // Comparison is always 64-bits
Cmp = DAG.getNode(ISD::FP_EXTEND, MVT::f64, Cmp);
return DAG.getNode(PPCISD::FSEL, ResVT, Cmp, FV, TV);
case ISD::SETUGE:
case ISD::SETOGE:
case ISD::SETGE:
Cmp = DAG.getNode(ISD::FSUB, CmpVT, LHS, RHS);
if (Cmp.getValueType() == MVT::f32) // Comparison is always 64-bits
Cmp = DAG.getNode(ISD::FP_EXTEND, MVT::f64, Cmp);
return DAG.getNode(PPCISD::FSEL, ResVT, Cmp, TV, FV);
case ISD::SETUGT:
case ISD::SETOGT:
case ISD::SETGT:
Cmp = DAG.getNode(ISD::FSUB, CmpVT, RHS, LHS);
if (Cmp.getValueType() == MVT::f32) // Comparison is always 64-bits
Cmp = DAG.getNode(ISD::FP_EXTEND, MVT::f64, Cmp);
return DAG.getNode(PPCISD::FSEL, ResVT, Cmp, FV, TV);
case ISD::SETULE:
case ISD::SETOLE:
case ISD::SETLE:
Cmp = DAG.getNode(ISD::FSUB, CmpVT, RHS, LHS);
if (Cmp.getValueType() == MVT::f32) // Comparison is always 64-bits
Cmp = DAG.getNode(ISD::FP_EXTEND, MVT::f64, Cmp);
return DAG.getNode(PPCISD::FSEL, ResVT, Cmp, TV, FV);
}
return SDOperand();
}
// FIXME: Split this code up when LegalizeDAGTypes lands.
static SDOperand LowerFP_TO_SINT(SDOperand Op, SelectionDAG &DAG) {
assert(MVT::isFloatingPoint(Op.getOperand(0).getValueType()));
SDOperand Src = Op.getOperand(0);
if (Src.getValueType() == MVT::f32)
Src = DAG.getNode(ISD::FP_EXTEND, MVT::f64, Src);
SDOperand Tmp;
switch (Op.getValueType()) {
default: assert(0 && "Unhandled FP_TO_SINT type in custom expander!");
case MVT::i32:
Tmp = DAG.getNode(PPCISD::FCTIWZ, MVT::f64, Src);
break;
case MVT::i64:
Tmp = DAG.getNode(PPCISD::FCTIDZ, MVT::f64, Src);
break;
}
// Convert the FP value to an int value through memory.
SDOperand FIPtr = DAG.CreateStackTemporary(MVT::f64);
// Emit a store to the stack slot.
SDOperand Chain = DAG.getStore(DAG.getEntryNode(), Tmp, FIPtr, NULL, 0);
// Result is a load from the stack slot. If loading 4 bytes, make sure to
// add in a bias.
if (Op.getValueType() == MVT::i32)
FIPtr = DAG.getNode(ISD::ADD, FIPtr.getValueType(), FIPtr,
DAG.getConstant(4, FIPtr.getValueType()));
return DAG.getLoad(Op.getValueType(), Chain, FIPtr, NULL, 0);
}
static SDOperand LowerFP_ROUND_INREG(SDOperand Op, SelectionDAG &DAG) {
assert(Op.getValueType() == MVT::ppcf128);
SDNode *Node = Op.Val;
assert(Node->getOperand(0).getValueType() == MVT::ppcf128);
assert(Node->getOperand(0).Val->getOpcode() == ISD::BUILD_PAIR);
SDOperand Lo = Node->getOperand(0).Val->getOperand(0);
SDOperand Hi = Node->getOperand(0).Val->getOperand(1);
// This sequence changes FPSCR to do round-to-zero, adds the two halves
// of the long double, and puts FPSCR back the way it was. We do not
// actually model FPSCR.
std::vector<MVT::ValueType> NodeTys;
SDOperand Ops[4], Result, MFFSreg, InFlag, FPreg;
NodeTys.push_back(MVT::f64); // Return register
NodeTys.push_back(MVT::Flag); // Returns a flag for later insns
Result = DAG.getNode(PPCISD::MFFS, NodeTys, &InFlag, 0);
MFFSreg = Result.getValue(0);
InFlag = Result.getValue(1);
NodeTys.clear();
NodeTys.push_back(MVT::Flag); // Returns a flag
Ops[0] = DAG.getConstant(31, MVT::i32);
Ops[1] = InFlag;
Result = DAG.getNode(PPCISD::MTFSB1, NodeTys, Ops, 2);
InFlag = Result.getValue(0);
NodeTys.clear();
NodeTys.push_back(MVT::Flag); // Returns a flag
Ops[0] = DAG.getConstant(30, MVT::i32);
Ops[1] = InFlag;
Result = DAG.getNode(PPCISD::MTFSB0, NodeTys, Ops, 2);
InFlag = Result.getValue(0);
NodeTys.clear();
NodeTys.push_back(MVT::f64); // result of add
NodeTys.push_back(MVT::Flag); // Returns a flag
Ops[0] = Lo;
Ops[1] = Hi;
Ops[2] = InFlag;
Result = DAG.getNode(PPCISD::FADDRTZ, NodeTys, Ops, 3);
FPreg = Result.getValue(0);
InFlag = Result.getValue(1);
NodeTys.clear();
NodeTys.push_back(MVT::f64);
Ops[0] = DAG.getConstant(1, MVT::i32);
Ops[1] = MFFSreg;
Ops[2] = FPreg;
Ops[3] = InFlag;
Result = DAG.getNode(PPCISD::MTFSF, NodeTys, Ops, 4);
FPreg = Result.getValue(0);
// We know the low half is about to be thrown away, so just use something
// convenient.
return DAG.getNode(ISD::BUILD_PAIR, Lo.getValueType(), FPreg, FPreg);
}
static SDOperand LowerSINT_TO_FP(SDOperand Op, SelectionDAG &DAG) {
if (Op.getOperand(0).getValueType() == MVT::i64) {
SDOperand Bits = DAG.getNode(ISD::BIT_CONVERT, MVT::f64, Op.getOperand(0));
SDOperand FP = DAG.getNode(PPCISD::FCFID, MVT::f64, Bits);
if (Op.getValueType() == MVT::f32)
FP = DAG.getNode(ISD::FP_ROUND, MVT::f32, FP, DAG.getIntPtrConstant(0));
return FP;
}
assert(Op.getOperand(0).getValueType() == MVT::i32 &&
"Unhandled SINT_TO_FP type in custom expander!");
// Since we only generate this in 64-bit mode, we can take advantage of
// 64-bit registers. In particular, sign extend the input value into the
// 64-bit register with extsw, store the WHOLE 64-bit value into the stack
// then lfd it and fcfid it.
MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
int FrameIdx = FrameInfo->CreateStackObject(8, 8);
MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDOperand FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
SDOperand Ext64 = DAG.getNode(PPCISD::EXTSW_32, MVT::i32,
Op.getOperand(0));
// STD the extended value into the stack slot.
MemOperand MO(&PseudoSourceValue::FPRel,
MemOperand::MOStore, FrameIdx, 8, 8);
SDOperand Store = DAG.getNode(PPCISD::STD_32, MVT::Other,
DAG.getEntryNode(), Ext64, FIdx,
DAG.getMemOperand(MO));
// Load the value as a double.
SDOperand Ld = DAG.getLoad(MVT::f64, Store, FIdx, NULL, 0);
// FCFID it and return it.
SDOperand FP = DAG.getNode(PPCISD::FCFID, MVT::f64, Ld);
if (Op.getValueType() == MVT::f32)
FP = DAG.getNode(ISD::FP_ROUND, MVT::f32, FP, DAG.getIntPtrConstant(0));
return FP;
}
static SDOperand LowerFLT_ROUNDS(SDOperand Op, SelectionDAG &DAG) {
/*
The rounding mode is in bits 30:31 of FPSR, and has the following
settings:
00 Round to nearest
01 Round to 0
10 Round to +inf
11 Round to -inf
FLT_ROUNDS, on the other hand, expects the following:
-1 Undefined
0 Round to 0
1 Round to nearest
2 Round to +inf
3 Round to -inf
To perform the conversion, we do:
((FPSCR & 0x3) ^ ((~FPSCR & 0x3) >> 1))
*/
MachineFunction &MF = DAG.getMachineFunction();
MVT::ValueType VT = Op.getValueType();
MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
std::vector<MVT::ValueType> NodeTys;
SDOperand MFFSreg, InFlag;
// Save FP Control Word to register
NodeTys.push_back(MVT::f64); // return register
NodeTys.push_back(MVT::Flag); // unused in this context
SDOperand Chain = DAG.getNode(PPCISD::MFFS, NodeTys, &InFlag, 0);
// Save FP register to stack slot
int SSFI = MF.getFrameInfo()->CreateStackObject(8, 8);
SDOperand StackSlot = DAG.getFrameIndex(SSFI, PtrVT);
SDOperand Store = DAG.getStore(DAG.getEntryNode(), Chain,
StackSlot, NULL, 0);
// Load FP Control Word from low 32 bits of stack slot.
SDOperand Four = DAG.getConstant(4, PtrVT);
SDOperand Addr = DAG.getNode(ISD::ADD, PtrVT, StackSlot, Four);
SDOperand CWD = DAG.getLoad(MVT::i32, Store, Addr, NULL, 0);
// Transform as necessary
SDOperand CWD1 =
DAG.getNode(ISD::AND, MVT::i32,
CWD, DAG.getConstant(3, MVT::i32));
SDOperand CWD2 =
DAG.getNode(ISD::SRL, MVT::i32,
DAG.getNode(ISD::AND, MVT::i32,
DAG.getNode(ISD::XOR, MVT::i32,
CWD, DAG.getConstant(3, MVT::i32)),
DAG.getConstant(3, MVT::i32)),
DAG.getConstant(1, MVT::i8));
SDOperand RetVal =
DAG.getNode(ISD::XOR, MVT::i32, CWD1, CWD2);
return DAG.getNode((MVT::getSizeInBits(VT) < 16 ?
ISD::TRUNCATE : ISD::ZERO_EXTEND), VT, RetVal);
}
static SDOperand LowerSHL_PARTS(SDOperand Op, SelectionDAG &DAG) {
assert(Op.getNumOperands() == 3 && Op.getValueType() == MVT::i32 &&
Op.getOperand(1).getValueType() == MVT::i32 && "Unexpected SHL!");
// Expand into a bunch of logical ops. Note that these ops
// depend on the PPC behavior for oversized shift amounts.
SDOperand Lo = Op.getOperand(0);
SDOperand Hi = Op.getOperand(1);
SDOperand Amt = Op.getOperand(2);
SDOperand Tmp1 = DAG.getNode(ISD::SUB, MVT::i32,
DAG.getConstant(32, MVT::i32), Amt);
SDOperand Tmp2 = DAG.getNode(PPCISD::SHL, MVT::i32, Hi, Amt);
SDOperand Tmp3 = DAG.getNode(PPCISD::SRL, MVT::i32, Lo, Tmp1);
SDOperand Tmp4 = DAG.getNode(ISD::OR , MVT::i32, Tmp2, Tmp3);
SDOperand Tmp5 = DAG.getNode(ISD::ADD, MVT::i32, Amt,
DAG.getConstant(-32U, MVT::i32));
SDOperand Tmp6 = DAG.getNode(PPCISD::SHL, MVT::i32, Lo, Tmp5);
SDOperand OutHi = DAG.getNode(ISD::OR, MVT::i32, Tmp4, Tmp6);
SDOperand OutLo = DAG.getNode(PPCISD::SHL, MVT::i32, Lo, Amt);
SDOperand OutOps[] = { OutLo, OutHi };
return DAG.getNode(ISD::MERGE_VALUES, DAG.getVTList(MVT::i32, MVT::i32),
OutOps, 2);
}
static SDOperand LowerSRL_PARTS(SDOperand Op, SelectionDAG &DAG) {
assert(Op.getNumOperands() == 3 && Op.getValueType() == MVT::i32 &&
Op.getOperand(1).getValueType() == MVT::i32 && "Unexpected SRL!");
// Otherwise, expand into a bunch of logical ops. Note that these ops
// depend on the PPC behavior for oversized shift amounts.
SDOperand Lo = Op.getOperand(0);
SDOperand Hi = Op.getOperand(1);
SDOperand Amt = Op.getOperand(2);
SDOperand Tmp1 = DAG.getNode(ISD::SUB, MVT::i32,
DAG.getConstant(32, MVT::i32), Amt);
SDOperand Tmp2 = DAG.getNode(PPCISD::SRL, MVT::i32, Lo, Amt);
SDOperand Tmp3 = DAG.getNode(PPCISD::SHL, MVT::i32, Hi, Tmp1);
SDOperand Tmp4 = DAG.getNode(ISD::OR , MVT::i32, Tmp2, Tmp3);
SDOperand Tmp5 = DAG.getNode(ISD::ADD, MVT::i32, Amt,
DAG.getConstant(-32U, MVT::i32));
SDOperand Tmp6 = DAG.getNode(PPCISD::SRL, MVT::i32, Hi, Tmp5);
SDOperand OutLo = DAG.getNode(ISD::OR, MVT::i32, Tmp4, Tmp6);
SDOperand OutHi = DAG.getNode(PPCISD::SRL, MVT::i32, Hi, Amt);
SDOperand OutOps[] = { OutLo, OutHi };
return DAG.getNode(ISD::MERGE_VALUES, DAG.getVTList(MVT::i32, MVT::i32),
OutOps, 2);
}
static SDOperand LowerSRA_PARTS(SDOperand Op, SelectionDAG &DAG) {
assert(Op.getNumOperands() == 3 && Op.getValueType() == MVT::i32 &&
Op.getOperand(1).getValueType() == MVT::i32 && "Unexpected SRA!");
// Otherwise, expand into a bunch of logical ops, followed by a select_cc.
SDOperand Lo = Op.getOperand(0);
SDOperand Hi = Op.getOperand(1);
SDOperand Amt = Op.getOperand(2);
SDOperand Tmp1 = DAG.getNode(ISD::SUB, MVT::i32,
DAG.getConstant(32, MVT::i32), Amt);
SDOperand Tmp2 = DAG.getNode(PPCISD::SRL, MVT::i32, Lo, Amt);
SDOperand Tmp3 = DAG.getNode(PPCISD::SHL, MVT::i32, Hi, Tmp1);
SDOperand Tmp4 = DAG.getNode(ISD::OR , MVT::i32, Tmp2, Tmp3);
SDOperand Tmp5 = DAG.getNode(ISD::ADD, MVT::i32, Amt,
DAG.getConstant(-32U, MVT::i32));
SDOperand Tmp6 = DAG.getNode(PPCISD::SRA, MVT::i32, Hi, Tmp5);
SDOperand OutHi = DAG.getNode(PPCISD::SRA, MVT::i32, Hi, Amt);
SDOperand OutLo = DAG.getSelectCC(Tmp5, DAG.getConstant(0, MVT::i32),
Tmp4, Tmp6, ISD::SETLE);
SDOperand OutOps[] = { OutLo, OutHi };
return DAG.getNode(ISD::MERGE_VALUES, DAG.getVTList(MVT::i32, MVT::i32),
OutOps, 2);
}
//===----------------------------------------------------------------------===//
// Vector related lowering.
//
// If this is a vector of constants or undefs, get the bits. A bit in
// UndefBits is set if the corresponding element of the vector is an
// ISD::UNDEF value. For undefs, the corresponding VectorBits values are
// zero. Return true if this is not an array of constants, false if it is.
//
static bool GetConstantBuildVectorBits(SDNode *BV, uint64_t VectorBits[2],
uint64_t UndefBits[2]) {
// Start with zero'd results.
VectorBits[0] = VectorBits[1] = UndefBits[0] = UndefBits[1] = 0;
unsigned EltBitSize = MVT::getSizeInBits(BV->getOperand(0).getValueType());
for (unsigned i = 0, e = BV->getNumOperands(); i != e; ++i) {
SDOperand OpVal = BV->getOperand(i);
unsigned PartNo = i >= e/2; // In the upper 128 bits?
unsigned SlotNo = e/2 - (i & (e/2-1))-1; // Which subpiece of the uint64_t.
uint64_t EltBits = 0;
if (OpVal.getOpcode() == ISD::UNDEF) {
uint64_t EltUndefBits = ~0U >> (32-EltBitSize);
UndefBits[PartNo] |= EltUndefBits << (SlotNo*EltBitSize);
continue;
} else if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(OpVal)) {
EltBits = CN->getValue() & (~0U >> (32-EltBitSize));
} else if (ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(OpVal)) {
assert(CN->getValueType(0) == MVT::f32 &&
"Only one legal FP vector type!");
EltBits = FloatToBits(CN->getValueAPF().convertToFloat());
} else {
// Nonconstant element.
return true;
}
VectorBits[PartNo] |= EltBits << (SlotNo*EltBitSize);
}
//printf("%llx %llx %llx %llx\n",
// VectorBits[0], VectorBits[1], UndefBits[0], UndefBits[1]);
return false;
}
// If this is a splat (repetition) of a value across the whole vector, return
// the smallest size that splats it. For example, "0x01010101010101..." is a
// splat of 0x01, 0x0101, and 0x01010101. We return SplatBits = 0x01 and
// SplatSize = 1 byte.
static bool isConstantSplat(const uint64_t Bits128[2],
const uint64_t Undef128[2],
unsigned &SplatBits, unsigned &SplatUndef,
unsigned &SplatSize) {
// Don't let undefs prevent splats from matching. See if the top 64-bits are
// the same as the lower 64-bits, ignoring undefs.
if ((Bits128[0] & ~Undef128[1]) != (Bits128[1] & ~Undef128[0]))
return false; // Can't be a splat if two pieces don't match.
uint64_t Bits64 = Bits128[0] | Bits128[1];
uint64_t Undef64 = Undef128[0] & Undef128[1];
// Check that the top 32-bits are the same as the lower 32-bits, ignoring
// undefs.
if ((Bits64 & (~Undef64 >> 32)) != ((Bits64 >> 32) & ~Undef64))
return false; // Can't be a splat if two pieces don't match.
uint32_t Bits32 = uint32_t(Bits64) | uint32_t(Bits64 >> 32);
uint32_t Undef32 = uint32_t(Undef64) & uint32_t(Undef64 >> 32);
// If the top 16-bits are different than the lower 16-bits, ignoring
// undefs, we have an i32 splat.
if ((Bits32 & (~Undef32 >> 16)) != ((Bits32 >> 16) & ~Undef32)) {
SplatBits = Bits32;
SplatUndef = Undef32;
SplatSize = 4;
return true;
}
uint16_t Bits16 = uint16_t(Bits32) | uint16_t(Bits32 >> 16);
uint16_t Undef16 = uint16_t(Undef32) & uint16_t(Undef32 >> 16);
// If the top 8-bits are different than the lower 8-bits, ignoring
// undefs, we have an i16 splat.
if ((Bits16 & (uint16_t(~Undef16) >> 8)) != ((Bits16 >> 8) & ~Undef16)) {
SplatBits = Bits16;
SplatUndef = Undef16;
SplatSize = 2;
return true;
}
// Otherwise, we have an 8-bit splat.
SplatBits = uint8_t(Bits16) | uint8_t(Bits16 >> 8);
SplatUndef = uint8_t(Undef16) & uint8_t(Undef16 >> 8);
SplatSize = 1;
return true;
}
/// BuildSplatI - Build a canonical splati of Val with an element size of
/// SplatSize. Cast the result to VT.
static SDOperand BuildSplatI(int Val, unsigned SplatSize, MVT::ValueType VT,
SelectionDAG &DAG) {
assert(Val >= -16 && Val <= 15 && "vsplti is out of range!");
static const MVT::ValueType VTys[] = { // canonical VT to use for each size.
MVT::v16i8, MVT::v8i16, MVT::Other, MVT::v4i32
};
MVT::ValueType ReqVT = VT != MVT::Other ? VT : VTys[SplatSize-1];
// Force vspltis[hw] -1 to vspltisb -1 to canonicalize.
if (Val == -1)
SplatSize = 1;
MVT::ValueType CanonicalVT = VTys[SplatSize-1];
// Build a canonical splat for this value.
SDOperand Elt = DAG.getConstant(Val, MVT::getVectorElementType(CanonicalVT));
SmallVector<SDOperand, 8> Ops;
Ops.assign(MVT::getVectorNumElements(CanonicalVT), Elt);
SDOperand Res = DAG.getNode(ISD::BUILD_VECTOR, CanonicalVT,
&Ops[0], Ops.size());
return DAG.getNode(ISD::BIT_CONVERT, ReqVT, Res);
}
/// BuildIntrinsicOp - Return a binary operator intrinsic node with the
/// specified intrinsic ID.
static SDOperand BuildIntrinsicOp(unsigned IID, SDOperand LHS, SDOperand RHS,
SelectionDAG &DAG,
MVT::ValueType DestVT = MVT::Other) {
if (DestVT == MVT::Other) DestVT = LHS.getValueType();
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DestVT,
DAG.getConstant(IID, MVT::i32), LHS, RHS);
}
/// BuildIntrinsicOp - Return a ternary operator intrinsic node with the
/// specified intrinsic ID.
static SDOperand BuildIntrinsicOp(unsigned IID, SDOperand Op0, SDOperand Op1,
SDOperand Op2, SelectionDAG &DAG,
MVT::ValueType DestVT = MVT::Other) {
if (DestVT == MVT::Other) DestVT = Op0.getValueType();
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DestVT,
DAG.getConstant(IID, MVT::i32), Op0, Op1, Op2);
}
/// BuildVSLDOI - Return a VECTOR_SHUFFLE that is a vsldoi of the specified
/// amount. The result has the specified value type.
static SDOperand BuildVSLDOI(SDOperand LHS, SDOperand RHS, unsigned Amt,
MVT::ValueType VT, SelectionDAG &DAG) {
// Force LHS/RHS to be the right type.
LHS = DAG.getNode(ISD::BIT_CONVERT, MVT::v16i8, LHS);
RHS = DAG.getNode(ISD::BIT_CONVERT, MVT::v16i8, RHS);
SDOperand Ops[16];
for (unsigned i = 0; i != 16; ++i)
Ops[i] = DAG.getConstant(i+Amt, MVT::i32);
SDOperand T = DAG.getNode(ISD::VECTOR_SHUFFLE, MVT::v16i8, LHS, RHS,
DAG.getNode(ISD::BUILD_VECTOR, MVT::v16i8, Ops,16));
return DAG.getNode(ISD::BIT_CONVERT, VT, T);
}
// If this is a case we can't handle, return null and let the default
// expansion code take care of it. If we CAN select this case, and if it
// selects to a single instruction, return Op. Otherwise, if we can codegen
// this case more efficiently than a constant pool load, lower it to the
// sequence of ops that should be used.
static SDOperand LowerBUILD_VECTOR(SDOperand Op, SelectionDAG &DAG) {
// If this is a vector of constants or undefs, get the bits. A bit in
// UndefBits is set if the corresponding element of the vector is an
// ISD::UNDEF value. For undefs, the corresponding VectorBits values are
// zero.
uint64_t VectorBits[2];
uint64_t UndefBits[2];
if (GetConstantBuildVectorBits(Op.Val, VectorBits, UndefBits))
return SDOperand(); // Not a constant vector.
// If this is a splat (repetition) of a value across the whole vector, return
// the smallest size that splats it. For example, "0x01010101010101..." is a
// splat of 0x01, 0x0101, and 0x01010101. We return SplatBits = 0x01 and
// SplatSize = 1 byte.
unsigned SplatBits, SplatUndef, SplatSize;
if (isConstantSplat(VectorBits, UndefBits, SplatBits, SplatUndef, SplatSize)){
bool HasAnyUndefs = (UndefBits[0] | UndefBits[1]) != 0;
// First, handle single instruction cases.
// All zeros?
if (SplatBits == 0) {
// Canonicalize all zero vectors to be v4i32.
if (Op.getValueType() != MVT::v4i32 || HasAnyUndefs) {
SDOperand Z = DAG.getConstant(0, MVT::i32);
Z = DAG.getNode(ISD::BUILD_VECTOR, MVT::v4i32, Z, Z, Z, Z);
Op = DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), Z);
}
return Op;
}
// If the sign extended value is in the range [-16,15], use VSPLTI[bhw].
int32_t SextVal= int32_t(SplatBits << (32-8*SplatSize)) >> (32-8*SplatSize);
if (SextVal >= -16 && SextVal <= 15)
return BuildSplatI(SextVal, SplatSize, Op.getValueType(), DAG);
// Two instruction sequences.
// If this value is in the range [-32,30] and is even, use:
// tmp = VSPLTI[bhw], result = add tmp, tmp
if (SextVal >= -32 && SextVal <= 30 && (SextVal & 1) == 0) {
Op = BuildSplatI(SextVal >> 1, SplatSize, Op.getValueType(), DAG);
return DAG.getNode(ISD::ADD, Op.getValueType(), Op, Op);
}
// If this is 0x8000_0000 x 4, turn into vspltisw + vslw. If it is
// 0x7FFF_FFFF x 4, turn it into not(0x8000_0000). This is important
// for fneg/fabs.
if (SplatSize == 4 && SplatBits == (0x7FFFFFFF&~SplatUndef)) {
// Make -1 and vspltisw -1:
SDOperand OnesV = BuildSplatI(-1, 4, MVT::v4i32, DAG);
// Make the VSLW intrinsic, computing 0x8000_0000.
SDOperand Res = BuildIntrinsicOp(Intrinsic::ppc_altivec_vslw, OnesV,
OnesV, DAG);
// xor by OnesV to invert it.
Res = DAG.getNode(ISD::XOR, MVT::v4i32, Res, OnesV);
return DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), Res);
}
// Check to see if this is a wide variety of vsplti*, binop self cases.
unsigned SplatBitSize = SplatSize*8;
static const signed char SplatCsts[] = {
-1, 1, -2, 2, -3, 3, -4, 4, -5, 5, -6, 6, -7, 7,
-8, 8, -9, 9, -10, 10, -11, 11, -12, 12, -13, 13, 14, -14, 15, -15, -16
};
for (unsigned idx = 0; idx < array_lengthof(SplatCsts); ++idx) {
// Indirect through the SplatCsts array so that we favor 'vsplti -1' for
// cases which are ambiguous (e.g. formation of 0x8000_0000). 'vsplti -1'
int i = SplatCsts[idx];
// Figure out what shift amount will be used by altivec if shifted by i in
// this splat size.
unsigned TypeShiftAmt = i & (SplatBitSize-1);
// vsplti + shl self.
if (SextVal == (i << (int)TypeShiftAmt)) {
SDOperand Res = BuildSplatI(i, SplatSize, MVT::Other, DAG);
static const unsigned IIDs[] = { // Intrinsic to use for each size.
Intrinsic::ppc_altivec_vslb, Intrinsic::ppc_altivec_vslh, 0,
Intrinsic::ppc_altivec_vslw
};
Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG);
return DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), Res);
}
// vsplti + srl self.
if (SextVal == (int)((unsigned)i >> TypeShiftAmt)) {
SDOperand Res = BuildSplatI(i, SplatSize, MVT::Other, DAG);
static const unsigned IIDs[] = { // Intrinsic to use for each size.
Intrinsic::ppc_altivec_vsrb, Intrinsic::ppc_altivec_vsrh, 0,
Intrinsic::ppc_altivec_vsrw
};
Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG);
return DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), Res);
}
// vsplti + sra self.
if (SextVal == (int)((unsigned)i >> TypeShiftAmt)) {
SDOperand Res = BuildSplatI(i, SplatSize, MVT::Other, DAG);
static const unsigned IIDs[] = { // Intrinsic to use for each size.
Intrinsic::ppc_altivec_vsrab, Intrinsic::ppc_altivec_vsrah, 0,
Intrinsic::ppc_altivec_vsraw
};
Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG);
return DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), Res);
}
// vsplti + rol self.
if (SextVal == (int)(((unsigned)i << TypeShiftAmt) |
((unsigned)i >> (SplatBitSize-TypeShiftAmt)))) {
SDOperand Res = BuildSplatI(i, SplatSize, MVT::Other, DAG);
static const unsigned IIDs[] = { // Intrinsic to use for each size.
Intrinsic::ppc_altivec_vrlb, Intrinsic::ppc_altivec_vrlh, 0,
Intrinsic::ppc_altivec_vrlw
};
Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG);
return DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), Res);
}
// t = vsplti c, result = vsldoi t, t, 1
if (SextVal == ((i << 8) | (i >> (TypeShiftAmt-8)))) {
SDOperand T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG);
return BuildVSLDOI(T, T, 1, Op.getValueType(), DAG);
}
// t = vsplti c, result = vsldoi t, t, 2
if (SextVal == ((i << 16) | (i >> (TypeShiftAmt-16)))) {
SDOperand T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG);
return BuildVSLDOI(T, T, 2, Op.getValueType(), DAG);
}
// t = vsplti c, result = vsldoi t, t, 3
if (SextVal == ((i << 24) | (i >> (TypeShiftAmt-24)))) {
SDOperand T = BuildSplatI(i, SplatSize, MVT::v16i8, DAG);
return BuildVSLDOI(T, T, 3, Op.getValueType(), DAG);
}
}
// Three instruction sequences.
// Odd, in range [17,31]: (vsplti C)-(vsplti -16).
if (SextVal >= 0 && SextVal <= 31) {
SDOperand LHS = BuildSplatI(SextVal-16, SplatSize, MVT::Other, DAG);
SDOperand RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG);
LHS = DAG.getNode(ISD::SUB, LHS.getValueType(), LHS, RHS);
return DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), LHS);
}
// Odd, in range [-31,-17]: (vsplti C)+(vsplti -16).
if (SextVal >= -31 && SextVal <= 0) {
SDOperand LHS = BuildSplatI(SextVal+16, SplatSize, MVT::Other, DAG);
SDOperand RHS = BuildSplatI(-16, SplatSize, MVT::Other, DAG);
LHS = DAG.getNode(ISD::ADD, LHS.getValueType(), LHS, RHS);
return DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), LHS);
}
}
return SDOperand();
}
/// GeneratePerfectShuffle - Given an entry in the perfect-shuffle table, emit
/// the specified operations to build the shuffle.
static SDOperand GeneratePerfectShuffle(unsigned PFEntry, SDOperand LHS,
SDOperand RHS, SelectionDAG &DAG) {
unsigned OpNum = (PFEntry >> 26) & 0x0F;
unsigned LHSID = (PFEntry >> 13) & ((1 << 13)-1);
unsigned RHSID = (PFEntry >> 0) & ((1 << 13)-1);
enum {
OP_COPY = 0, // Copy, used for things like <u,u,u,3> to say it is <0,1,2,3>
OP_VMRGHW,
OP_VMRGLW,
OP_VSPLTISW0,
OP_VSPLTISW1,
OP_VSPLTISW2,
OP_VSPLTISW3,
OP_VSLDOI4,
OP_VSLDOI8,
OP_VSLDOI12
};
if (OpNum == OP_COPY) {
if (LHSID == (1*9+2)*9+3) return LHS;
assert(LHSID == ((4*9+5)*9+6)*9+7 && "Illegal OP_COPY!");
return RHS;
}
SDOperand OpLHS, OpRHS;
OpLHS = GeneratePerfectShuffle(PerfectShuffleTable[LHSID], LHS, RHS, DAG);
OpRHS = GeneratePerfectShuffle(PerfectShuffleTable[RHSID], LHS, RHS, DAG);
unsigned ShufIdxs[16];
switch (OpNum) {
default: assert(0 && "Unknown i32 permute!");
case OP_VMRGHW:
ShufIdxs[ 0] = 0; ShufIdxs[ 1] = 1; ShufIdxs[ 2] = 2; ShufIdxs[ 3] = 3;
ShufIdxs[ 4] = 16; ShufIdxs[ 5] = 17; ShufIdxs[ 6] = 18; ShufIdxs[ 7] = 19;
ShufIdxs[ 8] = 4; ShufIdxs[ 9] = 5; ShufIdxs[10] = 6; ShufIdxs[11] = 7;
ShufIdxs[12] = 20; ShufIdxs[13] = 21; ShufIdxs[14] = 22; ShufIdxs[15] = 23;
break;
case OP_VMRGLW:
ShufIdxs[ 0] = 8; ShufIdxs[ 1] = 9; ShufIdxs[ 2] = 10; ShufIdxs[ 3] = 11;
ShufIdxs[ 4] = 24; ShufIdxs[ 5] = 25; ShufIdxs[ 6] = 26; ShufIdxs[ 7] = 27;
ShufIdxs[ 8] = 12; ShufIdxs[ 9] = 13; ShufIdxs[10] = 14; ShufIdxs[11] = 15;
ShufIdxs[12] = 28; ShufIdxs[13] = 29; ShufIdxs[14] = 30; ShufIdxs[15] = 31;
break;
case OP_VSPLTISW0:
for (unsigned i = 0; i != 16; ++i)
ShufIdxs[i] = (i&3)+0;
break;
case OP_VSPLTISW1:
for (unsigned i = 0; i != 16; ++i)
ShufIdxs[i] = (i&3)+4;
break;
case OP_VSPLTISW2:
for (unsigned i = 0; i != 16; ++i)
ShufIdxs[i] = (i&3)+8;
break;
case OP_VSPLTISW3:
for (unsigned i = 0; i != 16; ++i)
ShufIdxs[i] = (i&3)+12;
break;
case OP_VSLDOI4:
return BuildVSLDOI(OpLHS, OpRHS, 4, OpLHS.getValueType(), DAG);
case OP_VSLDOI8:
return BuildVSLDOI(OpLHS, OpRHS, 8, OpLHS.getValueType(), DAG);
case OP_VSLDOI12:
return BuildVSLDOI(OpLHS, OpRHS, 12, OpLHS.getValueType(), DAG);
}
SDOperand Ops[16];
for (unsigned i = 0; i != 16; ++i)
Ops[i] = DAG.getConstant(ShufIdxs[i], MVT::i32);
return DAG.getNode(ISD::VECTOR_SHUFFLE, OpLHS.getValueType(), OpLHS, OpRHS,
DAG.getNode(ISD::BUILD_VECTOR, MVT::v16i8, Ops, 16));
}
/// LowerVECTOR_SHUFFLE - Return the code we lower for VECTOR_SHUFFLE. If this
/// is a shuffle we can handle in a single instruction, return it. Otherwise,
/// return the code it can be lowered into. Worst case, it can always be
/// lowered into a vperm.
static SDOperand LowerVECTOR_SHUFFLE(SDOperand Op, SelectionDAG &DAG) {
SDOperand V1 = Op.getOperand(0);
SDOperand V2 = Op.getOperand(1);
SDOperand PermMask = Op.getOperand(2);
// Cases that are handled by instructions that take permute immediates
// (such as vsplt*) should be left as VECTOR_SHUFFLE nodes so they can be
// selected by the instruction selector.
if (V2.getOpcode() == ISD::UNDEF) {
if (PPC::isSplatShuffleMask(PermMask.Val, 1) ||
PPC::isSplatShuffleMask(PermMask.Val, 2) ||
PPC::isSplatShuffleMask(PermMask.Val, 4) ||
PPC::isVPKUWUMShuffleMask(PermMask.Val, true) ||
PPC::isVPKUHUMShuffleMask(PermMask.Val, true) ||
PPC::isVSLDOIShuffleMask(PermMask.Val, true) != -1 ||
PPC::isVMRGLShuffleMask(PermMask.Val, 1, true) ||
PPC::isVMRGLShuffleMask(PermMask.Val, 2, true) ||
PPC::isVMRGLShuffleMask(PermMask.Val, 4, true) ||
PPC::isVMRGHShuffleMask(PermMask.Val, 1, true) ||
PPC::isVMRGHShuffleMask(PermMask.Val, 2, true) ||
PPC::isVMRGHShuffleMask(PermMask.Val, 4, true)) {
return Op;
}
}
// Altivec has a variety of "shuffle immediates" that take two vector inputs
// and produce a fixed permutation. If any of these match, do not lower to
// VPERM.
if (PPC::isVPKUWUMShuffleMask(PermMask.Val, false) ||
PPC::isVPKUHUMShuffleMask(PermMask.Val, false) ||
PPC::isVSLDOIShuffleMask(PermMask.Val, false) != -1 ||
PPC::isVMRGLShuffleMask(PermMask.Val, 1, false) ||
PPC::isVMRGLShuffleMask(PermMask.Val, 2, false) ||
PPC::isVMRGLShuffleMask(PermMask.Val, 4, false) ||
PPC::isVMRGHShuffleMask(PermMask.Val, 1, false) ||
PPC::isVMRGHShuffleMask(PermMask.Val, 2, false) ||
PPC::isVMRGHShuffleMask(PermMask.Val, 4, false))
return Op;
// Check to see if this is a shuffle of 4-byte values. If so, we can use our
// perfect shuffle table to emit an optimal matching sequence.
unsigned PFIndexes[4];
bool isFourElementShuffle = true;
for (unsigned i = 0; i != 4 && isFourElementShuffle; ++i) { // Element number
unsigned EltNo = 8; // Start out undef.
for (unsigned j = 0; j != 4; ++j) { // Intra-element byte.
if (PermMask.getOperand(i*4+j).getOpcode() == ISD::UNDEF)
continue; // Undef, ignore it.
unsigned ByteSource =
cast<ConstantSDNode>(PermMask.getOperand(i*4+j))->getValue();
if ((ByteSource & 3) != j) {
isFourElementShuffle = false;
break;
}
if (EltNo == 8) {
EltNo = ByteSource/4;
} else if (EltNo != ByteSource/4) {
isFourElementShuffle = false;
break;
}
}
PFIndexes[i] = EltNo;
}
// If this shuffle can be expressed as a shuffle of 4-byte elements, use the
// perfect shuffle vector to determine if it is cost effective to do this as
// discrete instructions, or whether we should use a vperm.
if (isFourElementShuffle) {
// Compute the index in the perfect shuffle table.
unsigned PFTableIndex =
PFIndexes[0]*9*9*9+PFIndexes[1]*9*9+PFIndexes[2]*9+PFIndexes[3];
unsigned PFEntry = PerfectShuffleTable[PFTableIndex];
unsigned Cost = (PFEntry >> 30);
// Determining when to avoid vperm is tricky. Many things affect the cost
// of vperm, particularly how many times the perm mask needs to be computed.
// For example, if the perm mask can be hoisted out of a loop or is already
// used (perhaps because there are multiple permutes with the same shuffle
// mask?) the vperm has a cost of 1. OTOH, hoisting the permute mask out of
// the loop requires an extra register.
//
// As a compromise, we only emit discrete instructions if the shuffle can be
// generated in 3 or fewer operations. When we have loop information
// available, if this block is within a loop, we should avoid using vperm
// for 3-operation perms and use a constant pool load instead.
if (Cost < 3)
return GeneratePerfectShuffle(PFEntry, V1, V2, DAG);
}
// Lower this to a VPERM(V1, V2, V3) expression, where V3 is a constant
// vector that will get spilled to the constant pool.
if (V2.getOpcode() == ISD::UNDEF) V2 = V1;
// The SHUFFLE_VECTOR mask is almost exactly what we want for vperm, except
// that it is in input element units, not in bytes. Convert now.
MVT::ValueType EltVT = MVT::getVectorElementType(V1.getValueType());
unsigned BytesPerElement = MVT::getSizeInBits(EltVT)/8;
SmallVector<SDOperand, 16> ResultMask;
for (unsigned i = 0, e = PermMask.getNumOperands(); i != e; ++i) {
unsigned SrcElt;
if (PermMask.getOperand(i).getOpcode() == ISD::UNDEF)
SrcElt = 0;
else
SrcElt = cast<ConstantSDNode>(PermMask.getOperand(i))->getValue();
for (unsigned j = 0; j != BytesPerElement; ++j)
ResultMask.push_back(DAG.getConstant(SrcElt*BytesPerElement+j,
MVT::i8));
}
SDOperand VPermMask = DAG.getNode(ISD::BUILD_VECTOR, MVT::v16i8,
&ResultMask[0], ResultMask.size());
return DAG.getNode(PPCISD::VPERM, V1.getValueType(), V1, V2, VPermMask);
}
/// getAltivecCompareInfo - Given an intrinsic, return false if it is not an
/// altivec comparison. If it is, return true and fill in Opc/isDot with
/// information about the intrinsic.
static bool getAltivecCompareInfo(SDOperand Intrin, int &CompareOpc,
bool &isDot) {
unsigned IntrinsicID = cast<ConstantSDNode>(Intrin.getOperand(0))->getValue();
CompareOpc = -1;
isDot = false;
switch (IntrinsicID) {
default: return false;
// Comparison predicates.
case Intrinsic::ppc_altivec_vcmpbfp_p: CompareOpc = 966; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpeqfp_p: CompareOpc = 198; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpequb_p: CompareOpc = 6; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpequh_p: CompareOpc = 70; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpequw_p: CompareOpc = 134; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpgefp_p: CompareOpc = 454; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpgtfp_p: CompareOpc = 710; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpgtsb_p: CompareOpc = 774; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpgtsh_p: CompareOpc = 838; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpgtsw_p: CompareOpc = 902; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpgtub_p: CompareOpc = 518; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpgtuh_p: CompareOpc = 582; isDot = 1; break;
case Intrinsic::ppc_altivec_vcmpgtuw_p: CompareOpc = 646; isDot = 1; break;
// Normal Comparisons.
case Intrinsic::ppc_altivec_vcmpbfp: CompareOpc = 966; isDot = 0; break;
case Intrinsic::ppc_altivec_vcmpeqfp: CompareOpc = 198; isDot = 0; break;
case Intrinsic::ppc_altivec_vcmpequb: CompareOpc = 6; isDot = 0; break;
case Intrinsic::ppc_altivec_vcmpequh: CompareOpc = 70; isDot = 0; break;
case Intrinsic::ppc_altivec_vcmpequw: CompareOpc = 134; isDot = 0; break;
case Intrinsic::ppc_altivec_vcmpgefp: CompareOpc = 454; isDot = 0; break;
case Intrinsic::ppc_altivec_vcmpgtfp: CompareOpc = 710; isDot = 0; break;
case Intrinsic::ppc_altivec_vcmpgtsb: CompareOpc = 774; isDot = 0; break;
case Intrinsic::ppc_altivec_vcmpgtsh: CompareOpc = 838; isDot = 0; break;
case Intrinsic::ppc_altivec_vcmpgtsw: CompareOpc = 902; isDot = 0; break;
case Intrinsic::ppc_altivec_vcmpgtub: CompareOpc = 518; isDot = 0; break;
case Intrinsic::ppc_altivec_vcmpgtuh: CompareOpc = 582; isDot = 0; break;
case Intrinsic::ppc_altivec_vcmpgtuw: CompareOpc = 646; isDot = 0; break;
}
return true;
}
/// LowerINTRINSIC_WO_CHAIN - If this is an intrinsic that we want to custom
/// lower, do it, otherwise return null.
static SDOperand LowerINTRINSIC_WO_CHAIN(SDOperand Op, SelectionDAG &DAG) {
// If this is a lowered altivec predicate compare, CompareOpc is set to the
// opcode number of the comparison.
int CompareOpc;
bool isDot;
if (!getAltivecCompareInfo(Op, CompareOpc, isDot))
return SDOperand(); // Don't custom lower most intrinsics.
// If this is a non-dot comparison, make the VCMP node and we are done.
if (!isDot) {
SDOperand Tmp = DAG.getNode(PPCISD::VCMP, Op.getOperand(2).getValueType(),
Op.getOperand(1), Op.getOperand(2),
DAG.getConstant(CompareOpc, MVT::i32));
return DAG.getNode(ISD::BIT_CONVERT, Op.getValueType(), Tmp);
}
// Create the PPCISD altivec 'dot' comparison node.
SDOperand Ops[] = {
Op.getOperand(2), // LHS
Op.getOperand(3), // RHS
DAG.getConstant(CompareOpc, MVT::i32)
};
std::vector<MVT::ValueType> VTs;
VTs.push_back(Op.getOperand(2).getValueType());
VTs.push_back(MVT::Flag);
SDOperand CompNode = DAG.getNode(PPCISD::VCMPo, VTs, Ops, 3);
// Now that we have the comparison, emit a copy from the CR to a GPR.
// This is flagged to the above dot comparison.
SDOperand Flags = DAG.getNode(PPCISD::MFCR, MVT::i32,
DAG.getRegister(PPC::CR6, MVT::i32),
CompNode.getValue(1));
// Unpack the result based on how the target uses it.
unsigned BitNo; // Bit # of CR6.
bool InvertBit; // Invert result?
switch (cast<ConstantSDNode>(Op.getOperand(1))->getValue()) {
default: // Can't happen, don't crash on invalid number though.
case 0: // Return the value of the EQ bit of CR6.
BitNo = 0; InvertBit = false;
break;
case 1: // Return the inverted value of the EQ bit of CR6.
BitNo = 0; InvertBit = true;
break;
case 2: // Return the value of the LT bit of CR6.
BitNo = 2; InvertBit = false;
break;
case 3: // Return the inverted value of the LT bit of CR6.
BitNo = 2; InvertBit = true;
break;
}
// Shift the bit into the low position.
Flags = DAG.getNode(ISD::SRL, MVT::i32, Flags,
DAG.getConstant(8-(3-BitNo), MVT::i32));
// Isolate the bit.
Flags = DAG.getNode(ISD::AND, MVT::i32, Flags,
DAG.getConstant(1, MVT::i32));
// If we are supposed to, toggle the bit.
if (InvertBit)
Flags = DAG.getNode(ISD::XOR, MVT::i32, Flags,
DAG.getConstant(1, MVT::i32));
return Flags;
}
static SDOperand LowerSCALAR_TO_VECTOR(SDOperand Op, SelectionDAG &DAG) {
// Create a stack slot that is 16-byte aligned.
MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
int FrameIdx = FrameInfo->CreateStackObject(16, 16);
MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
SDOperand FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
// Store the input value into Value#0 of the stack slot.
SDOperand Store = DAG.getStore(DAG.getEntryNode(),
Op.getOperand(0), FIdx, NULL, 0);
// Load it out.
return DAG.getLoad(Op.getValueType(), Store, FIdx, NULL, 0);
}
static SDOperand LowerMUL(SDOperand Op, SelectionDAG &DAG) {
if (Op.getValueType() == MVT::v4i32) {
SDOperand LHS = Op.getOperand(0), RHS = Op.getOperand(1);
SDOperand Zero = BuildSplatI( 0, 1, MVT::v4i32, DAG);
SDOperand Neg16 = BuildSplatI(-16, 4, MVT::v4i32, DAG); // +16 as shift amt.
SDOperand RHSSwap = // = vrlw RHS, 16
BuildIntrinsicOp(Intrinsic::ppc_altivec_vrlw, RHS, Neg16, DAG);
// Shrinkify inputs to v8i16.
LHS = DAG.getNode(ISD::BIT_CONVERT, MVT::v8i16, LHS);
RHS = DAG.getNode(ISD::BIT_CONVERT, MVT::v8i16, RHS);
RHSSwap = DAG.getNode(ISD::BIT_CONVERT, MVT::v8i16, RHSSwap);
// Low parts multiplied together, generating 32-bit results (we ignore the
// top parts).
SDOperand LoProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmulouh,
LHS, RHS, DAG, MVT::v4i32);
SDOperand HiProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmsumuhm,
LHS, RHSSwap, Zero, DAG, MVT::v4i32);
// Shift the high parts up 16 bits.
HiProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vslw, HiProd, Neg16, DAG);
return DAG.getNode(ISD::ADD, MVT::v4i32, LoProd, HiProd);
} else if (Op.getValueType() == MVT::v8i16) {
SDOperand LHS = Op.getOperand(0), RHS = Op.getOperand(1);
SDOperand Zero = BuildSplatI(0, 1, MVT::v8i16, DAG);
return BuildIntrinsicOp(Intrinsic::ppc_altivec_vmladduhm,
LHS, RHS, Zero, DAG);
} else if (Op.getValueType() == MVT::v16i8) {
SDOperand LHS = Op.getOperand(0), RHS = Op.getOperand(1);
// Multiply the even 8-bit parts, producing 16-bit sums.
SDOperand EvenParts = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmuleub,
LHS, RHS, DAG, MVT::v8i16);
EvenParts = DAG.getNode(ISD::BIT_CONVERT, MVT::v16i8, EvenParts);
// Multiply the odd 8-bit parts, producing 16-bit sums.
SDOperand OddParts = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmuloub,
LHS, RHS, DAG, MVT::v8i16);
OddParts = DAG.getNode(ISD::BIT_CONVERT, MVT::v16i8, OddParts);
// Merge the results together.
SDOperand Ops[16];
for (unsigned i = 0; i != 8; ++i) {
Ops[i*2 ] = DAG.getConstant(2*i+1, MVT::i8);
Ops[i*2+1] = DAG.getConstant(2*i+1+16, MVT::i8);
}
return DAG.getNode(ISD::VECTOR_SHUFFLE, MVT::v16i8, EvenParts, OddParts,
DAG.getNode(ISD::BUILD_VECTOR, MVT::v16i8, Ops, 16));
} else {
assert(0 && "Unknown mul to lower!");
abort();
}
}
/// LowerOperation - Provide custom lowering hooks for some operations.
///
SDOperand PPCTargetLowering::LowerOperation(SDOperand Op, SelectionDAG &DAG) {
switch (Op.getOpcode()) {
default: assert(0 && "Wasn't expecting to be able to lower this!");
case ISD::ConstantPool: return LowerConstantPool(Op, DAG);
case ISD::GlobalAddress: return LowerGlobalAddress(Op, DAG);
case ISD::GlobalTLSAddress: return LowerGlobalTLSAddress(Op, DAG);
case ISD::JumpTable: return LowerJumpTable(Op, DAG);
case ISD::SETCC: return LowerSETCC(Op, DAG);
case ISD::VASTART:
return LowerVASTART(Op, DAG, VarArgsFrameIndex, VarArgsStackOffset,
VarArgsNumGPR, VarArgsNumFPR, PPCSubTarget);
case ISD::VAARG:
return LowerVAARG(Op, DAG, VarArgsFrameIndex, VarArgsStackOffset,
VarArgsNumGPR, VarArgsNumFPR, PPCSubTarget);
case ISD::FORMAL_ARGUMENTS:
return LowerFORMAL_ARGUMENTS(Op, DAG, VarArgsFrameIndex,
VarArgsStackOffset, VarArgsNumGPR,
VarArgsNumFPR, PPCSubTarget);
case ISD::CALL: return LowerCALL(Op, DAG, PPCSubTarget);
case ISD::RET: return LowerRET(Op, DAG, getTargetMachine());
case ISD::STACKRESTORE: return LowerSTACKRESTORE(Op, DAG, PPCSubTarget);
case ISD::DYNAMIC_STACKALLOC:
return LowerDYNAMIC_STACKALLOC(Op, DAG, PPCSubTarget);
case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
case ISD::FP_TO_SINT: return LowerFP_TO_SINT(Op, DAG);
case ISD::SINT_TO_FP: return LowerSINT_TO_FP(Op, DAG);
case ISD::FP_ROUND_INREG: return LowerFP_ROUND_INREG(Op, DAG);
case ISD::FLT_ROUNDS: return LowerFLT_ROUNDS(Op, DAG);
// Lower 64-bit shifts.
case ISD::SHL_PARTS: return LowerSHL_PARTS(Op, DAG);
case ISD::SRL_PARTS: return LowerSRL_PARTS(Op, DAG);
case ISD::SRA_PARTS: return LowerSRA_PARTS(Op, DAG);
// Vector-related lowering.
case ISD::BUILD_VECTOR: return LowerBUILD_VECTOR(Op, DAG);
case ISD::VECTOR_SHUFFLE: return LowerVECTOR_SHUFFLE(Op, DAG);
case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
case ISD::SCALAR_TO_VECTOR: return LowerSCALAR_TO_VECTOR(Op, DAG);
case ISD::MUL: return LowerMUL(Op, DAG);
// Frame & Return address.
case ISD::RETURNADDR: return LowerRETURNADDR(Op, DAG);
case ISD::FRAMEADDR: return LowerFRAMEADDR(Op, DAG);
}
return SDOperand();
}
SDNode *PPCTargetLowering::ExpandOperationResult(SDNode *N, SelectionDAG &DAG) {
switch (N->getOpcode()) {
default: assert(0 && "Wasn't expecting to be able to lower this!");
case ISD::FP_TO_SINT: return LowerFP_TO_SINT(SDOperand(N, 0), DAG).Val;
}
}
//===----------------------------------------------------------------------===//
// Other Lowering Code
//===----------------------------------------------------------------------===//
MachineBasicBlock *
PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr *MI,
MachineBasicBlock *BB) {
const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
assert((MI->getOpcode() == PPC::SELECT_CC_I4 ||
MI->getOpcode() == PPC::SELECT_CC_I8 ||
MI->getOpcode() == PPC::SELECT_CC_F4 ||
MI->getOpcode() == PPC::SELECT_CC_F8 ||
MI->getOpcode() == PPC::SELECT_CC_VRRC) &&
"Unexpected instr type to insert");
// To "insert" a SELECT_CC instruction, we actually have to insert the diamond
// control-flow pattern. The incoming instruction knows the destination vreg
// to set, the condition code register to branch on, the true/false values to
// select between, and a branch opcode to use.
const BasicBlock *LLVM_BB = BB->getBasicBlock();
ilist<MachineBasicBlock>::iterator It = BB;
++It;
// thisMBB:
// ...
// TrueVal = ...
// cmpTY ccX, r1, r2
// bCC copy1MBB
// fallthrough --> copy0MBB
MachineBasicBlock *thisMBB = BB;
MachineBasicBlock *copy0MBB = new MachineBasicBlock(LLVM_BB);
MachineBasicBlock *sinkMBB = new MachineBasicBlock(LLVM_BB);
unsigned SelectPred = MI->getOperand(4).getImm();
BuildMI(BB, TII->get(PPC::BCC))
.addImm(SelectPred).addReg(MI->getOperand(1).getReg()).addMBB(sinkMBB);
MachineFunction *F = BB->getParent();
F->getBasicBlockList().insert(It, copy0MBB);
F->getBasicBlockList().insert(It, sinkMBB);
// Update machine-CFG edges by first adding all successors of the current
// block to the new block which will contain the Phi node for the select.
for(MachineBasicBlock::succ_iterator i = BB->succ_begin(),
e = BB->succ_end(); i != e; ++i)
sinkMBB->addSuccessor(*i);
// Next, remove all successors of the current block, and add the true
// and fallthrough blocks as its successors.
while(!BB->succ_empty())
BB->removeSuccessor(BB->succ_begin());
BB->addSuccessor(copy0MBB);
BB->addSuccessor(sinkMBB);
// copy0MBB:
// %FalseValue = ...
// # fallthrough to sinkMBB
BB = copy0MBB;
// Update machine-CFG edges
BB->addSuccessor(sinkMBB);
// sinkMBB:
// %Result = phi [ %FalseValue, copy0MBB ], [ %TrueValue, thisMBB ]
// ...
BB = sinkMBB;
BuildMI(BB, TII->get(PPC::PHI), MI->getOperand(0).getReg())
.addReg(MI->getOperand(3).getReg()).addMBB(copy0MBB)
.addReg(MI->getOperand(2).getReg()).addMBB(thisMBB);
delete MI; // The pseudo instruction is gone now.
return BB;
}
//===----------------------------------------------------------------------===//
// Target Optimization Hooks
//===----------------------------------------------------------------------===//
SDOperand PPCTargetLowering::PerformDAGCombine(SDNode *N,
DAGCombinerInfo &DCI) const {
TargetMachine &TM = getTargetMachine();
SelectionDAG &DAG = DCI.DAG;
switch (N->getOpcode()) {
default: break;
case PPCISD::SHL:
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(0))) {
if (C->getValue() == 0) // 0 << V -> 0.
return N->getOperand(0);
}
break;
case PPCISD::SRL:
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(0))) {
if (C->getValue() == 0) // 0 >>u V -> 0.
return N->getOperand(0);
}
break;
case PPCISD::SRA:
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(0))) {
if (C->getValue() == 0 || // 0 >>s V -> 0.
C->isAllOnesValue()) // -1 >>s V -> -1.
return N->getOperand(0);
}
break;
case ISD::SINT_TO_FP:
if (TM.getSubtarget<PPCSubtarget>().has64BitSupport()) {
if (N->getOperand(0).getOpcode() == ISD::FP_TO_SINT) {
// Turn (sint_to_fp (fp_to_sint X)) -> fctidz/fcfid without load/stores.
// We allow the src/dst to be either f32/f64, but the intermediate
// type must be i64.
if (N->getOperand(0).getValueType() == MVT::i64 &&
N->getOperand(0).getOperand(0).getValueType() != MVT::ppcf128) {
SDOperand Val = N->getOperand(0).getOperand(0);
if (Val.getValueType() == MVT::f32) {
Val = DAG.getNode(ISD::FP_EXTEND, MVT::f64, Val);
DCI.AddToWorklist(Val.Val);
}
Val = DAG.getNode(PPCISD::FCTIDZ, MVT::f64, Val);
DCI.AddToWorklist(Val.Val);
Val = DAG.getNode(PPCISD::FCFID, MVT::f64, Val);
DCI.AddToWorklist(Val.Val);
if (N->getValueType(0) == MVT::f32) {
Val = DAG.getNode(ISD::FP_ROUND, MVT::f32, Val,
DAG.getIntPtrConstant(0));
DCI.AddToWorklist(Val.Val);
}
return Val;
} else if (N->getOperand(0).getValueType() == MVT::i32) {
// If the intermediate type is i32, we can avoid the load/store here
// too.
}
}
}
break;
case ISD::STORE:
// Turn STORE (FP_TO_SINT F) -> STFIWX(FCTIWZ(F)).
if (TM.getSubtarget<PPCSubtarget>().hasSTFIWX() &&
!cast<StoreSDNode>(N)->isTruncatingStore() &&
N->getOperand(1).getOpcode() == ISD::FP_TO_SINT &&
N->getOperand(1).getValueType() == MVT::i32 &&
N->getOperand(1).getOperand(0).getValueType() != MVT::ppcf128) {
SDOperand Val = N->getOperand(1).getOperand(0);
if (Val.getValueType() == MVT::f32) {
Val = DAG.getNode(ISD::FP_EXTEND, MVT::f64, Val);
DCI.AddToWorklist(Val.Val);
}
Val = DAG.getNode(PPCISD::FCTIWZ, MVT::f64, Val);
DCI.AddToWorklist(Val.Val);
Val = DAG.getNode(PPCISD::STFIWX, MVT::Other, N->getOperand(0), Val,
N->getOperand(2), N->getOperand(3));
DCI.AddToWorklist(Val.Val);
return Val;
}
// Turn STORE (BSWAP) -> sthbrx/stwbrx.
if (N->getOperand(1).getOpcode() == ISD::BSWAP &&
N->getOperand(1).Val->hasOneUse() &&
(N->getOperand(1).getValueType() == MVT::i32 ||
N->getOperand(1).getValueType() == MVT::i16)) {
SDOperand BSwapOp = N->getOperand(1).getOperand(0);
// Do an any-extend to 32-bits if this is a half-word input.
if (BSwapOp.getValueType() == MVT::i16)
BSwapOp = DAG.getNode(ISD::ANY_EXTEND, MVT::i32, BSwapOp);
return DAG.getNode(PPCISD::STBRX, MVT::Other, N->getOperand(0), BSwapOp,
N->getOperand(2), N->getOperand(3),
DAG.getValueType(N->getOperand(1).getValueType()));
}
break;
case ISD::BSWAP:
// Turn BSWAP (LOAD) -> lhbrx/lwbrx.
if (ISD::isNON_EXTLoad(N->getOperand(0).Val) &&
N->getOperand(0).hasOneUse() &&
(N->getValueType(0) == MVT::i32 || N->getValueType(0) == MVT::i16)) {
SDOperand Load = N->getOperand(0);
LoadSDNode *LD = cast<LoadSDNode>(Load);
// Create the byte-swapping load.
std::vector<MVT::ValueType> VTs;
VTs.push_back(MVT::i32);
VTs.push_back(MVT::Other);
SDOperand MO = DAG.getMemOperand(LD->getMemOperand());
SDOperand Ops[] = {
LD->getChain(), // Chain
LD->getBasePtr(), // Ptr
MO, // MemOperand
DAG.getValueType(N->getValueType(0)) // VT
};
SDOperand BSLoad = DAG.getNode(PPCISD::LBRX, VTs, Ops, 4);
// If this is an i16 load, insert the truncate.
SDOperand ResVal = BSLoad;
if (N->getValueType(0) == MVT::i16)
ResVal = DAG.getNode(ISD::TRUNCATE, MVT::i16, BSLoad);
// First, combine the bswap away. This makes the value produced by the
// load dead.
DCI.CombineTo(N, ResVal);
// Next, combine the load away, we give it a bogus result value but a real
// chain result. The result value is dead because the bswap is dead.
DCI.CombineTo(Load.Val, ResVal, BSLoad.getValue(1));
// Return N so it doesn't get rechecked!
return SDOperand(N, 0);
}
break;
case PPCISD::VCMP: {
// If a VCMPo node already exists with exactly the same operands as this
// node, use its result instead of this node (VCMPo computes both a CR6 and
// a normal output).
//
if (!N->getOperand(0).hasOneUse() &&
!N->getOperand(1).hasOneUse() &&
!N->getOperand(2).hasOneUse()) {
// Scan all of the users of the LHS, looking for VCMPo's that match.
SDNode *VCMPoNode = 0;
SDNode *LHSN = N->getOperand(0).Val;
for (SDNode::use_iterator UI = LHSN->use_begin(), E = LHSN->use_end();
UI != E; ++UI)
if ((*UI)->getOpcode() == PPCISD::VCMPo &&
(*UI)->getOperand(1) == N->getOperand(1) &&
(*UI)->getOperand(2) == N->getOperand(2) &&
(*UI)->getOperand(0) == N->getOperand(0)) {
VCMPoNode = *UI;
break;
}
// If there is no VCMPo node, or if the flag value has a single use, don't
// transform this.
if (!VCMPoNode || VCMPoNode->hasNUsesOfValue(0, 1))
break;
// Look at the (necessarily single) use of the flag value. If it has a
// chain, this transformation is more complex. Note that multiple things
// could use the value result, which we should ignore.
SDNode *FlagUser = 0;
for (SDNode::use_iterator UI = VCMPoNode->use_begin();
FlagUser == 0; ++UI) {
assert(UI != VCMPoNode->use_end() && "Didn't find user!");
SDNode *User = *UI;
for (unsigned i = 0, e = User->getNumOperands(); i != e; ++i) {
if (User->getOperand(i) == SDOperand(VCMPoNode, 1)) {
FlagUser = User;
break;
}
}
}
// If the user is a MFCR instruction, we know this is safe. Otherwise we
// give up for right now.
if (FlagUser->getOpcode() == PPCISD::MFCR)
return SDOperand(VCMPoNode, 0);
}
break;
}
case ISD::BR_CC: {
// If this is a branch on an altivec predicate comparison, lower this so
// that we don't have to do a MFCR: instead, branch directly on CR6. This
// lowering is done pre-legalize, because the legalizer lowers the predicate
// compare down to code that is difficult to reassemble.
ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(1))->get();
SDOperand LHS = N->getOperand(2), RHS = N->getOperand(3);
int CompareOpc;
bool isDot;
if (LHS.getOpcode() == ISD::INTRINSIC_WO_CHAIN &&
isa<ConstantSDNode>(RHS) && (CC == ISD::SETEQ || CC == ISD::SETNE) &&
getAltivecCompareInfo(LHS, CompareOpc, isDot)) {
assert(isDot && "Can't compare against a vector result!");
// If this is a comparison against something other than 0/1, then we know
// that the condition is never/always true.
unsigned Val = cast<ConstantSDNode>(RHS)->getValue();
if (Val != 0 && Val != 1) {
if (CC == ISD::SETEQ) // Cond never true, remove branch.
return N->getOperand(0);
// Always !=, turn it into an unconditional branch.
return DAG.getNode(ISD::BR, MVT::Other,
N->getOperand(0), N->getOperand(4));
}
bool BranchOnWhenPredTrue = (CC == ISD::SETEQ) ^ (Val == 0);
// Create the PPCISD altivec 'dot' comparison node.
std::vector<MVT::ValueType> VTs;
SDOperand Ops[] = {
LHS.getOperand(2), // LHS of compare
LHS.getOperand(3), // RHS of compare
DAG.getConstant(CompareOpc, MVT::i32)
};
VTs.push_back(LHS.getOperand(2).getValueType());
VTs.push_back(MVT::Flag);
SDOperand CompNode = DAG.getNode(PPCISD::VCMPo, VTs, Ops, 3);
// Unpack the result based on how the target uses it.
PPC::Predicate CompOpc;
switch (cast<ConstantSDNode>(LHS.getOperand(1))->getValue()) {
default: // Can't happen, don't crash on invalid number though.
case 0: // Branch on the value of the EQ bit of CR6.
CompOpc = BranchOnWhenPredTrue ? PPC::PRED_EQ : PPC::PRED_NE;
break;
case 1: // Branch on the inverted value of the EQ bit of CR6.
CompOpc = BranchOnWhenPredTrue ? PPC::PRED_NE : PPC::PRED_EQ;
break;
case 2: // Branch on the value of the LT bit of CR6.
CompOpc = BranchOnWhenPredTrue ? PPC::PRED_LT : PPC::PRED_GE;
break;
case 3: // Branch on the inverted value of the LT bit of CR6.
CompOpc = BranchOnWhenPredTrue ? PPC::PRED_GE : PPC::PRED_LT;
break;
}
return DAG.getNode(PPCISD::COND_BRANCH, MVT::Other, N->getOperand(0),
DAG.getConstant(CompOpc, MVT::i32),
DAG.getRegister(PPC::CR6, MVT::i32),
N->getOperand(4), CompNode.getValue(1));
}
break;
}
}
return SDOperand();
}
//===----------------------------------------------------------------------===//
// Inline Assembly Support
//===----------------------------------------------------------------------===//
void PPCTargetLowering::computeMaskedBitsForTargetNode(const SDOperand Op,
uint64_t Mask,
uint64_t &KnownZero,
uint64_t &KnownOne,
const SelectionDAG &DAG,
unsigned Depth) const {
KnownZero = 0;
KnownOne = 0;
switch (Op.getOpcode()) {
default: break;
case PPCISD::LBRX: {
// lhbrx is known to have the top bits cleared out.
if (cast<VTSDNode>(Op.getOperand(3))->getVT() == MVT::i16)
KnownZero = 0xFFFF0000;
break;
}
case ISD::INTRINSIC_WO_CHAIN: {
switch (cast<ConstantSDNode>(Op.getOperand(0))->getValue()) {
default: break;
case Intrinsic::ppc_altivec_vcmpbfp_p:
case Intrinsic::ppc_altivec_vcmpeqfp_p:
case Intrinsic::ppc_altivec_vcmpequb_p:
case Intrinsic::ppc_altivec_vcmpequh_p:
case Intrinsic::ppc_altivec_vcmpequw_p:
case Intrinsic::ppc_altivec_vcmpgefp_p:
case Intrinsic::ppc_altivec_vcmpgtfp_p:
case Intrinsic::ppc_altivec_vcmpgtsb_p:
case Intrinsic::ppc_altivec_vcmpgtsh_p:
case Intrinsic::ppc_altivec_vcmpgtsw_p:
case Intrinsic::ppc_altivec_vcmpgtub_p:
case Intrinsic::ppc_altivec_vcmpgtuh_p:
case Intrinsic::ppc_altivec_vcmpgtuw_p:
KnownZero = ~1U; // All bits but the low one are known to be zero.
break;
}
}
}
}
/// getConstraintType - Given a constraint, return the type of
/// constraint it is for this target.
PPCTargetLowering::ConstraintType
PPCTargetLowering::getConstraintType(const std::string &Constraint) const {
if (Constraint.size() == 1) {
switch (Constraint[0]) {
default: break;
case 'b':
case 'r':
case 'f':
case 'v':
case 'y':
return C_RegisterClass;
}
}
return TargetLowering::getConstraintType(Constraint);
}
std::pair<unsigned, const TargetRegisterClass*>
PPCTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
MVT::ValueType VT) const {
if (Constraint.size() == 1) {
// GCC RS6000 Constraint Letters
switch (Constraint[0]) {
case 'b': // R1-R31
case 'r': // R0-R31
if (VT == MVT::i64 && PPCSubTarget.isPPC64())
return std::make_pair(0U, PPC::G8RCRegisterClass);
return std::make_pair(0U, PPC::GPRCRegisterClass);
case 'f':
if (VT == MVT::f32)
return std::make_pair(0U, PPC::F4RCRegisterClass);
else if (VT == MVT::f64)
return std::make_pair(0U, PPC::F8RCRegisterClass);
break;
case 'v':
return std::make_pair(0U, PPC::VRRCRegisterClass);
case 'y': // crrc
return std::make_pair(0U, PPC::CRRCRegisterClass);
}
}
return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
}
/// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
/// vector. If it is invalid, don't add anything to Ops.
void PPCTargetLowering::LowerAsmOperandForConstraint(SDOperand Op, char Letter,
std::vector<SDOperand>&Ops,
SelectionDAG &DAG) {
SDOperand Result(0,0);
switch (Letter) {
default: break;
case 'I':
case 'J':
case 'K':
case 'L':
case 'M':
case 'N':
case 'O':
case 'P': {
ConstantSDNode *CST = dyn_cast<ConstantSDNode>(Op);
if (!CST) return; // Must be an immediate to match.
unsigned Value = CST->getValue();
switch (Letter) {
default: assert(0 && "Unknown constraint letter!");
case 'I': // "I" is a signed 16-bit constant.
if ((short)Value == (int)Value)
Result = DAG.getTargetConstant(Value, Op.getValueType());
break;
case 'J': // "J" is a constant with only the high-order 16 bits nonzero.
case 'L': // "L" is a signed 16-bit constant shifted left 16 bits.
if ((short)Value == 0)
Result = DAG.getTargetConstant(Value, Op.getValueType());
break;
case 'K': // "K" is a constant with only the low-order 16 bits nonzero.
if ((Value >> 16) == 0)
Result = DAG.getTargetConstant(Value, Op.getValueType());
break;
case 'M': // "M" is a constant that is greater than 31.
if (Value > 31)
Result = DAG.getTargetConstant(Value, Op.getValueType());
break;
case 'N': // "N" is a positive constant that is an exact power of two.
if ((int)Value > 0 && isPowerOf2_32(Value))
Result = DAG.getTargetConstant(Value, Op.getValueType());
break;
case 'O': // "O" is the constant zero.
if (Value == 0)
Result = DAG.getTargetConstant(Value, Op.getValueType());
break;
case 'P': // "P" is a constant whose negation is a signed 16-bit constant.
if ((short)-Value == (int)-Value)
Result = DAG.getTargetConstant(Value, Op.getValueType());
break;
}
break;
}
}
if (Result.Val) {
Ops.push_back(Result);
return;
}
// Handle standard constraint letters.
TargetLowering::LowerAsmOperandForConstraint(Op, Letter, Ops, DAG);
}
// isLegalAddressingMode - Return true if the addressing mode represented
// by AM is legal for this target, for a load/store of the specified type.
bool PPCTargetLowering::isLegalAddressingMode(const AddrMode &AM,
const Type *Ty) const {
// FIXME: PPC does not allow r+i addressing modes for vectors!
// PPC allows a sign-extended 16-bit immediate field.
if (AM.BaseOffs <= -(1LL << 16) || AM.BaseOffs >= (1LL << 16)-1)
return false;
// No global is ever allowed as a base.
if (AM.BaseGV)
return false;
// PPC only support r+r,
switch (AM.Scale) {
case 0: // "r+i" or just "i", depending on HasBaseReg.
break;
case 1:
if (AM.HasBaseReg && AM.BaseOffs) // "r+r+i" is not allowed.
return false;
// Otherwise we have r+r or r+i.
break;
case 2:
if (AM.HasBaseReg || AM.BaseOffs) // 2*r+r or 2*r+i is not allowed.
return false;
// Allow 2*r as r+r.
break;
default:
// No other scales are supported.
return false;
}
return true;
}
/// isLegalAddressImmediate - Return true if the integer value can be used
/// as the offset of the target addressing mode for load / store of the
/// given type.
bool PPCTargetLowering::isLegalAddressImmediate(int64_t V,const Type *Ty) const{
// PPC allows a sign-extended 16-bit immediate field.
return (V > -(1 << 16) && V < (1 << 16)-1);
}
bool PPCTargetLowering::isLegalAddressImmediate(llvm::GlobalValue* GV) const {
return false;
}
SDOperand PPCTargetLowering::LowerRETURNADDR(SDOperand Op, SelectionDAG &DAG) {
// Depths > 0 not supported yet!
if (cast<ConstantSDNode>(Op.getOperand(0))->getValue() > 0)
return SDOperand();
MachineFunction &MF = DAG.getMachineFunction();
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
int RAIdx = FuncInfo->getReturnAddrSaveIndex();
if (RAIdx == 0) {
bool isPPC64 = PPCSubTarget.isPPC64();
int Offset =
PPCFrameInfo::getReturnSaveOffset(isPPC64, PPCSubTarget.isMachoABI());
// Set up a frame object for the return address.
RAIdx = MF.getFrameInfo()->CreateFixedObject(isPPC64 ? 8 : 4, Offset);
// Remember it for next time.
FuncInfo->setReturnAddrSaveIndex(RAIdx);
// Make sure the function really does not optimize away the store of the RA
// to the stack.
FuncInfo->setLRStoreRequired();
}
// Just load the return address off the stack.
SDOperand RetAddrFI = DAG.getFrameIndex(RAIdx, getPointerTy());
return DAG.getLoad(getPointerTy(), DAG.getEntryNode(), RetAddrFI, NULL, 0);
}
SDOperand PPCTargetLowering::LowerFRAMEADDR(SDOperand Op, SelectionDAG &DAG) {
// Depths > 0 not supported yet!
if (cast<ConstantSDNode>(Op.getOperand(0))->getValue() > 0)
return SDOperand();
MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = PtrVT == MVT::i64;
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo *MFI = MF.getFrameInfo();
bool is31 = (NoFramePointerElim || MFI->hasVarSizedObjects())
&& MFI->getStackSize();
if (isPPC64)
return DAG.getCopyFromReg(DAG.getEntryNode(), is31 ? PPC::X31 : PPC::X1,
MVT::i64);
else
return DAG.getCopyFromReg(DAG.getEntryNode(), is31 ? PPC::R31 : PPC::R1,
MVT::i32);
}
|