aboutsummaryrefslogtreecommitdiffstats
path: root/lib/Target/CellSPU/SPUISelLowering.cpp
blob: e0049fd7973c47a2eab8eb5695462d820440d889 (plain)
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
//
//===-- SPUISelLowering.cpp - Cell SPU 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 SPUTargetLowering class.
//
//===----------------------------------------------------------------------===//

#include "SPURegisterNames.h"
#include "SPUISelLowering.h"
#include "SPUTargetMachine.h"
#include "SPUFrameInfo.h"
#include "SPUMachineFunction.h"
#include "llvm/Constants.h"
#include "llvm/Function.h"
#include "llvm/Intrinsics.h"
#include "llvm/CallingConv.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/SelectionDAG.h"
#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/ADT/VectorExtras.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include <map>

using namespace llvm;

// Used in getTargetNodeName() below
namespace {
  std::map<unsigned, const char *> node_names;

  //! EVT mapping to useful data for Cell SPU
  struct valtype_map_s {
    EVT   valtype;
    int   prefslot_byte;
  };

  const valtype_map_s valtype_map[] = {
    { MVT::i1,   3 },
    { MVT::i8,   3 },
    { MVT::i16,  2 },
    { MVT::i32,  0 },
    { MVT::f32,  0 },
    { MVT::i64,  0 },
    { MVT::f64,  0 },
    { MVT::i128, 0 }
  };

  const size_t n_valtype_map = sizeof(valtype_map) / sizeof(valtype_map[0]);

  const valtype_map_s *getValueTypeMapEntry(EVT VT) {
    const valtype_map_s *retval = 0;

    for (size_t i = 0; i < n_valtype_map; ++i) {
      if (valtype_map[i].valtype == VT) {
        retval = valtype_map + i;
        break;
      }
    }

#ifndef NDEBUG
    if (retval == 0) {
      report_fatal_error("getValueTypeMapEntry returns NULL for " +
                         Twine(VT.getEVTString()));
    }
#endif

    return retval;
  }

  //! Expand a library call into an actual call DAG node
  /*!
   \note
   This code is taken from SelectionDAGLegalize, since it is not exposed as
   part of the LLVM SelectionDAG API.
   */

  SDValue
  ExpandLibCall(RTLIB::Libcall LC, SDValue Op, SelectionDAG &DAG,
                bool isSigned, SDValue &Hi, const SPUTargetLowering &TLI) {
    // The input chain to this libcall is the entry node of the function.
    // Legalizing the call will automatically add the previous call to the
    // dependence.
    SDValue InChain = DAG.getEntryNode();

    TargetLowering::ArgListTy Args;
    TargetLowering::ArgListEntry Entry;
    for (unsigned i = 0, e = Op.getNumOperands(); i != e; ++i) {
      EVT ArgVT = Op.getOperand(i).getValueType();
      const Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
      Entry.Node = Op.getOperand(i);
      Entry.Ty = ArgTy;
      Entry.isSExt = isSigned;
      Entry.isZExt = !isSigned;
      Args.push_back(Entry);
    }
    SDValue Callee = DAG.getExternalSymbol(TLI.getLibcallName(LC),
                                           TLI.getPointerTy());

    // Splice the libcall in wherever FindInputOutputChains tells us to.
    const Type *RetTy =
                Op.getNode()->getValueType(0).getTypeForEVT(*DAG.getContext());
    std::pair<SDValue, SDValue> CallInfo =
            TLI.LowerCallTo(InChain, RetTy, isSigned, !isSigned, false, false,
                            0, TLI.getLibcallCallingConv(LC), false,
                            /*isReturnValueUsed=*/true,
                            Callee, Args, DAG, Op.getDebugLoc());

    return CallInfo.first;
  }
}

SPUTargetLowering::SPUTargetLowering(SPUTargetMachine &TM)
  : TargetLowering(TM, new TargetLoweringObjectFileELF()),
    SPUTM(TM) {
  // Fold away setcc operations if possible.
  setPow2DivIsCheap();

  // Use _setjmp/_longjmp instead of setjmp/longjmp.
  setUseUnderscoreSetJmp(true);
  setUseUnderscoreLongJmp(true);

  // Set RTLIB libcall names as used by SPU:
  setLibcallName(RTLIB::DIV_F64, "__fast_divdf3");

  // Set up the SPU's register classes:
  addRegisterClass(MVT::i8,   SPU::R8CRegisterClass);
  addRegisterClass(MVT::i16,  SPU::R16CRegisterClass);
  addRegisterClass(MVT::i32,  SPU::R32CRegisterClass);
  addRegisterClass(MVT::i64,  SPU::R64CRegisterClass);
  addRegisterClass(MVT::f32,  SPU::R32FPRegisterClass);
  addRegisterClass(MVT::f64,  SPU::R64FPRegisterClass);
  addRegisterClass(MVT::i128, SPU::GPRCRegisterClass);

  // SPU has no sign or zero extended loads for i1, i8, i16:
  setLoadExtAction(ISD::EXTLOAD,  MVT::i1, Promote);
  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
  setLoadExtAction(ISD::ZEXTLOAD, MVT::i1, Promote);

  setLoadExtAction(ISD::EXTLOAD,  MVT::f32, Expand);
  setLoadExtAction(ISD::EXTLOAD,  MVT::f64, Expand);

  setTruncStoreAction(MVT::i128, MVT::i64, Expand);
  setTruncStoreAction(MVT::i128, MVT::i32, Expand);
  setTruncStoreAction(MVT::i128, MVT::i16, Expand);
  setTruncStoreAction(MVT::i128, MVT::i8, Expand);

  setTruncStoreAction(MVT::f64, MVT::f32, Expand);

  // SPU constant load actions are custom lowered:
  setOperationAction(ISD::ConstantFP, MVT::f32, Legal);
  setOperationAction(ISD::ConstantFP, MVT::f64, Custom);

  // SPU's loads and stores have to be custom lowered:
  for (unsigned sctype = (unsigned) MVT::i8; sctype < (unsigned) MVT::i128;
       ++sctype) {
    MVT::SimpleValueType VT = (MVT::SimpleValueType)sctype;

    setOperationAction(ISD::LOAD,   VT, Custom);
    setOperationAction(ISD::STORE,  VT, Custom);
    setLoadExtAction(ISD::EXTLOAD,  VT, Custom);
    setLoadExtAction(ISD::ZEXTLOAD, VT, Custom);
    setLoadExtAction(ISD::SEXTLOAD, VT, Custom);

    for (unsigned stype = sctype - 1; stype >= (unsigned) MVT::i8; --stype) {
      MVT::SimpleValueType StoreVT = (MVT::SimpleValueType) stype;
      setTruncStoreAction(VT, StoreVT, Expand);
    }
  }

  for (unsigned sctype = (unsigned) MVT::f32; sctype < (unsigned) MVT::f64;
       ++sctype) {
    MVT::SimpleValueType VT = (MVT::SimpleValueType) sctype;

    setOperationAction(ISD::LOAD,   VT, Custom);
    setOperationAction(ISD::STORE,  VT, Custom);

    for (unsigned stype = sctype - 1; stype >= (unsigned) MVT::f32; --stype) {
      MVT::SimpleValueType StoreVT = (MVT::SimpleValueType) stype;
      setTruncStoreAction(VT, StoreVT, Expand);
    }
  }

  // Expand the jumptable branches
  setOperationAction(ISD::BR_JT,        MVT::Other, Expand);
  setOperationAction(ISD::BR_CC,        MVT::Other, Expand);

  // Custom lower SELECT_CC for most cases, but expand by default
  setOperationAction(ISD::SELECT_CC,    MVT::Other, Expand);
  setOperationAction(ISD::SELECT_CC,    MVT::i8,    Custom);
  setOperationAction(ISD::SELECT_CC,    MVT::i16,   Custom);
  setOperationAction(ISD::SELECT_CC,    MVT::i32,   Custom);
  setOperationAction(ISD::SELECT_CC,    MVT::i64,   Custom);

  // SPU has no intrinsics for these particular operations:
  setOperationAction(ISD::MEMBARRIER, MVT::Other, Expand);

  // SPU has no division/remainder instructions
  setOperationAction(ISD::SREM,    MVT::i8,   Expand);
  setOperationAction(ISD::UREM,    MVT::i8,   Expand);
  setOperationAction(ISD::SDIV,    MVT::i8,   Expand);
  setOperationAction(ISD::UDIV,    MVT::i8,   Expand);
  setOperationAction(ISD::SDIVREM, MVT::i8,   Expand);
  setOperationAction(ISD::UDIVREM, MVT::i8,   Expand);
  setOperationAction(ISD::SREM,    MVT::i16,  Expand);
  setOperationAction(ISD::UREM,    MVT::i16,  Expand);
  setOperationAction(ISD::SDIV,    MVT::i16,  Expand);
  setOperationAction(ISD::UDIV,    MVT::i16,  Expand);
  setOperationAction(ISD::SDIVREM, MVT::i16,  Expand);
  setOperationAction(ISD::UDIVREM, MVT::i16,  Expand);
  setOperationAction(ISD::SREM,    MVT::i32,  Expand);
  setOperationAction(ISD::UREM,    MVT::i32,  Expand);
  setOperationAction(ISD::SDIV,    MVT::i32,  Expand);
  setOperationAction(ISD::UDIV,    MVT::i32,  Expand);
  setOperationAction(ISD::SDIVREM, MVT::i32,  Expand);
  setOperationAction(ISD::UDIVREM, MVT::i32,  Expand);
  setOperationAction(ISD::SREM,    MVT::i64,  Expand);
  setOperationAction(ISD::UREM,    MVT::i64,  Expand);
  setOperationAction(ISD::SDIV,    MVT::i64,  Expand);
  setOperationAction(ISD::UDIV,    MVT::i64,  Expand);
  setOperationAction(ISD::SDIVREM, MVT::i64,  Expand);
  setOperationAction(ISD::UDIVREM, MVT::i64,  Expand);
  setOperationAction(ISD::SREM,    MVT::i128, Expand);
  setOperationAction(ISD::UREM,    MVT::i128, Expand);
  setOperationAction(ISD::SDIV,    MVT::i128, Expand);
  setOperationAction(ISD::UDIV,    MVT::i128, Expand);
  setOperationAction(ISD::SDIVREM, MVT::i128, Expand);
  setOperationAction(ISD::UDIVREM, MVT::i128, Expand);

  // We don't support sin/cos/sqrt/fmod
  setOperationAction(ISD::FSIN , MVT::f64, Expand);
  setOperationAction(ISD::FCOS , MVT::f64, Expand);
  setOperationAction(ISD::FREM , MVT::f64, Expand);
  setOperationAction(ISD::FSIN , MVT::f32, Expand);
  setOperationAction(ISD::FCOS , MVT::f32, Expand);
  setOperationAction(ISD::FREM , MVT::f32, Expand);

  // Expand fsqrt to the appropriate libcall (NOTE: should use h/w fsqrt
  // for f32!)
  setOperationAction(ISD::FSQRT, MVT::f64, Expand);
  setOperationAction(ISD::FSQRT, MVT::f32, Expand);

  setOperationAction(ISD::FCOPYSIGN, MVT::f64, Expand);
  setOperationAction(ISD::FCOPYSIGN, MVT::f32, Expand);

  // SPU can do rotate right and left, so legalize it... but customize for i8
  // because instructions don't exist.

  // FIXME: Change from "expand" to appropriate type once ROTR is supported in
  //        .td files.
  setOperationAction(ISD::ROTR, MVT::i32,    Expand /*Legal*/);
  setOperationAction(ISD::ROTR, MVT::i16,    Expand /*Legal*/);
  setOperationAction(ISD::ROTR, MVT::i8,     Expand /*Custom*/);

  setOperationAction(ISD::ROTL, MVT::i32,    Legal);
  setOperationAction(ISD::ROTL, MVT::i16,    Legal);
  setOperationAction(ISD::ROTL, MVT::i8,     Custom);

  // SPU has no native version of shift left/right for i8
  setOperationAction(ISD::SHL,  MVT::i8,     Custom);
  setOperationAction(ISD::SRL,  MVT::i8,     Custom);
  setOperationAction(ISD::SRA,  MVT::i8,     Custom);

  // Make these operations legal and handle them during instruction selection:
  setOperationAction(ISD::SHL,  MVT::i64,    Legal);
  setOperationAction(ISD::SRL,  MVT::i64,    Legal);
  setOperationAction(ISD::SRA,  MVT::i64,    Legal);

  // Custom lower i8, i32 and i64 multiplications
  setOperationAction(ISD::MUL,  MVT::i8,     Custom);
  setOperationAction(ISD::MUL,  MVT::i32,    Legal);
  setOperationAction(ISD::MUL,  MVT::i64,    Legal);

  // Expand double-width multiplication
  // FIXME: It would probably be reasonable to support some of these operations
  setOperationAction(ISD::UMUL_LOHI, MVT::i8,  Expand);
  setOperationAction(ISD::SMUL_LOHI, MVT::i8,  Expand);
  setOperationAction(ISD::MULHU,     MVT::i8,  Expand);
  setOperationAction(ISD::MULHS,     MVT::i8,  Expand);
  setOperationAction(ISD::UMUL_LOHI, MVT::i16, Expand);
  setOperationAction(ISD::SMUL_LOHI, MVT::i16, Expand);
  setOperationAction(ISD::MULHU,     MVT::i16, Expand);
  setOperationAction(ISD::MULHS,     MVT::i16, Expand);
  setOperationAction(ISD::UMUL_LOHI, MVT::i32, Expand);
  setOperationAction(ISD::SMUL_LOHI, MVT::i32, Expand);
  setOperationAction(ISD::MULHU,     MVT::i32, Expand);
  setOperationAction(ISD::MULHS,     MVT::i32, Expand);
  setOperationAction(ISD::UMUL_LOHI, MVT::i64, Expand);
  setOperationAction(ISD::SMUL_LOHI, MVT::i64, Expand);
  setOperationAction(ISD::MULHU,     MVT::i64, Expand);
  setOperationAction(ISD::MULHS,     MVT::i64, Expand);

  // Need to custom handle (some) common i8, i64 math ops
  setOperationAction(ISD::ADD,  MVT::i8,     Custom);
  setOperationAction(ISD::ADD,  MVT::i64,    Legal);
  setOperationAction(ISD::SUB,  MVT::i8,     Custom);
  setOperationAction(ISD::SUB,  MVT::i64,    Legal);

  // SPU does not have BSWAP. It does have i32 support CTLZ.
  // CTPOP has to be custom lowered.
  setOperationAction(ISD::BSWAP, MVT::i32,   Expand);
  setOperationAction(ISD::BSWAP, MVT::i64,   Expand);

  setOperationAction(ISD::CTPOP, MVT::i8,    Custom);
  setOperationAction(ISD::CTPOP, MVT::i16,   Custom);
  setOperationAction(ISD::CTPOP, MVT::i32,   Custom);
  setOperationAction(ISD::CTPOP, MVT::i64,   Custom);
  setOperationAction(ISD::CTPOP, MVT::i128,  Expand);

  setOperationAction(ISD::CTTZ , MVT::i8,    Expand);
  setOperationAction(ISD::CTTZ , MVT::i16,   Expand);
  setOperationAction(ISD::CTTZ , MVT::i32,   Expand);
  setOperationAction(ISD::CTTZ , MVT::i64,   Expand);
  setOperationAction(ISD::CTTZ , MVT::i128,  Expand);

  setOperationAction(ISD::CTLZ , MVT::i8,    Promote);
  setOperationAction(ISD::CTLZ , MVT::i16,   Promote);
  setOperationAction(ISD::CTLZ , MVT::i32,   Legal);
  setOperationAction(ISD::CTLZ , MVT::i64,   Expand);
  setOperationAction(ISD::CTLZ , MVT::i128,  Expand);

  // SPU has a version of select that implements (a&~c)|(b&c), just like
  // select ought to work:
  setOperationAction(ISD::SELECT, MVT::i8,   Legal);
  setOperationAction(ISD::SELECT, MVT::i16,  Legal);
  setOperationAction(ISD::SELECT, MVT::i32,  Legal);
  setOperationAction(ISD::SELECT, MVT::i64,  Legal);

  setOperationAction(ISD::SETCC, MVT::i8,    Legal);
  setOperationAction(ISD::SETCC, MVT::i16,   Legal);
  setOperationAction(ISD::SETCC, MVT::i32,   Legal);
  setOperationAction(ISD::SETCC, MVT::i64,   Legal);
  setOperationAction(ISD::SETCC, MVT::f64,   Custom);

  // Custom lower i128 -> i64 truncates
  setOperationAction(ISD::TRUNCATE, MVT::i64, Custom);

  // Custom lower i32/i64 -> i128 sign extend
  setOperationAction(ISD::SIGN_EXTEND, MVT::i128, Custom);

  setOperationAction(ISD::FP_TO_SINT, MVT::i8, Promote);
  setOperationAction(ISD::FP_TO_UINT, MVT::i8, Promote);
  setOperationAction(ISD::FP_TO_SINT, MVT::i16, Promote);
  setOperationAction(ISD::FP_TO_UINT, MVT::i16, Promote);
  // SPU has a legal FP -> signed INT instruction for f32, but for f64, need
  // to expand to a libcall, hence the custom lowering:
  setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
  setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
  setOperationAction(ISD::FP_TO_SINT, MVT::i64, Expand);
  setOperationAction(ISD::FP_TO_UINT, MVT::i64, Expand);
  setOperationAction(ISD::FP_TO_SINT, MVT::i128, Expand);
  setOperationAction(ISD::FP_TO_UINT, MVT::i128, Expand);

  // FDIV on SPU requires custom lowering
  setOperationAction(ISD::FDIV, MVT::f64, Expand);      // to libcall

  // SPU has [U|S]INT_TO_FP for f32->i32, but not for f64->i32, f64->i64:
  setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
  setOperationAction(ISD::SINT_TO_FP, MVT::i16, Promote);
  setOperationAction(ISD::SINT_TO_FP, MVT::i8,  Promote);
  setOperationAction(ISD::UINT_TO_FP, MVT::i32, Custom);
  setOperationAction(ISD::UINT_TO_FP, MVT::i16, Promote);
  setOperationAction(ISD::UINT_TO_FP, MVT::i8,  Promote);
  setOperationAction(ISD::SINT_TO_FP, MVT::i64, Custom);
  setOperationAction(ISD::UINT_TO_FP, MVT::i64, Custom);

  setOperationAction(ISD::BIT_CONVERT, MVT::i32, Legal);
  setOperationAction(ISD::BIT_CONVERT, MVT::f32, Legal);
  setOperationAction(ISD::BIT_CONVERT, MVT::i64, Legal);
  setOperationAction(ISD::BIT_CONVERT, MVT::f64, Legal);

  // We cannot sextinreg(i1).  Expand to shifts.
  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);

  // We want to legalize GlobalAddress and ConstantPool nodes into the
  // appropriate instructions to materialize the address.
  for (unsigned sctype = (unsigned) MVT::i8; sctype < (unsigned) MVT::f128;
       ++sctype) {
    MVT::SimpleValueType VT = (MVT::SimpleValueType)sctype;

    setOperationAction(ISD::GlobalAddress,  VT, Custom);
    setOperationAction(ISD::ConstantPool,   VT, Custom);
    setOperationAction(ISD::JumpTable,      VT, Custom);
  }

  // VASTART needs to be custom lowered to use the VarArgsFrameIndex
  setOperationAction(ISD::VASTART           , MVT::Other, Custom);

  // Use the default implementation.
  setOperationAction(ISD::VAARG             , MVT::Other, Expand);
  setOperationAction(ISD::VACOPY            , MVT::Other, Expand);
  setOperationAction(ISD::VAEND             , MVT::Other, Expand);
  setOperationAction(ISD::STACKSAVE         , MVT::Other, Expand);
  setOperationAction(ISD::STACKRESTORE      , MVT::Other, Expand);
  setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32  , Expand);
  setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64  , Expand);

  // Cell SPU has instructions for converting between i64 and fp.
  setOperationAction(ISD::FP_TO_SINT, MVT::i64, Custom);
  setOperationAction(ISD::SINT_TO_FP, MVT::i64, Custom);

  // To take advantage of the above i64 FP_TO_SINT, promote i32 FP_TO_UINT
  setOperationAction(ISD::FP_TO_UINT, MVT::i32, Promote);

  // BUILD_PAIR can't be handled natively, and should be expanded to shl/or
  setOperationAction(ISD::BUILD_PAIR, MVT::i64, Expand);

  // First set operation action for all vector types to expand. Then we
  // will selectively turn on ones that can be effectively codegen'd.
  addRegisterClass(MVT::v16i8, SPU::VECREGRegisterClass);
  addRegisterClass(MVT::v8i16, SPU::VECREGRegisterClass);
  addRegisterClass(MVT::v4i32, SPU::VECREGRegisterClass);
  addRegisterClass(MVT::v2i64, SPU::VECREGRegisterClass);
  addRegisterClass(MVT::v4f32, SPU::VECREGRegisterClass);
  addRegisterClass(MVT::v2f64, SPU::VECREGRegisterClass);

  for (unsigned i = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
       i <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++i) {
    MVT::SimpleValueType VT = (MVT::SimpleValueType)i;

    // add/sub are legal for all supported vector VT's.
    setOperationAction(ISD::ADD,     VT, Legal);
    setOperationAction(ISD::SUB,     VT, Legal);
    // mul has to be custom lowered.
    setOperationAction(ISD::MUL,     VT, Legal);

    setOperationAction(ISD::AND,     VT, Legal);
    setOperationAction(ISD::OR,      VT, Legal);
    setOperationAction(ISD::XOR,     VT, Legal);
    setOperationAction(ISD::LOAD,    VT, Legal);
    setOperationAction(ISD::SELECT,  VT, Legal);
    setOperationAction(ISD::STORE,   VT, Legal);

    // These operations need to be expanded:
    setOperationAction(ISD::SDIV,    VT, Expand);
    setOperationAction(ISD::SREM,    VT, Expand);
    setOperationAction(ISD::UDIV,    VT, Expand);
    setOperationAction(ISD::UREM,    VT, Expand);

    // Custom lower build_vector, constant pool spills, insert and
    // extract vector elements:
    setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
    setOperationAction(ISD::ConstantPool, VT, Custom);
    setOperationAction(ISD::SCALAR_TO_VECTOR, VT, Custom);
    setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
    setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
    setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
  }

  setOperationAction(ISD::AND, MVT::v16i8, Custom);
  setOperationAction(ISD::OR,  MVT::v16i8, Custom);
  setOperationAction(ISD::XOR, MVT::v16i8, Custom);
  setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v4f32, Custom);

  setOperationAction(ISD::FDIV, MVT::v4f32, Legal);

  setShiftAmountType(MVT::i32);
  setBooleanContents(ZeroOrNegativeOneBooleanContent);

  setStackPointerRegisterToSaveRestore(SPU::R1);

  // We have target-specific dag combine patterns for the following nodes:
  setTargetDAGCombine(ISD::ADD);
  setTargetDAGCombine(ISD::ZERO_EXTEND);
  setTargetDAGCombine(ISD::SIGN_EXTEND);
  setTargetDAGCombine(ISD::ANY_EXTEND);

  computeRegisterProperties();

  // Set pre-RA register scheduler default to BURR, which produces slightly
  // better code than the default (could also be TDRR, but TargetLowering.h
  // needs a mod to support that model):
  setSchedulingPreference(Sched::RegPressure);
}

const char *
SPUTargetLowering::getTargetNodeName(unsigned Opcode) const
{
  if (node_names.empty()) {
    node_names[(unsigned) SPUISD::RET_FLAG] = "SPUISD::RET_FLAG";
    node_names[(unsigned) SPUISD::Hi] = "SPUISD::Hi";
    node_names[(unsigned) SPUISD::Lo] = "SPUISD::Lo";
    node_names[(unsigned) SPUISD::PCRelAddr] = "SPUISD::PCRelAddr";
    node_names[(unsigned) SPUISD::AFormAddr] = "SPUISD::AFormAddr";
    node_names[(unsigned) SPUISD::IndirectAddr] = "SPUISD::IndirectAddr";
    node_names[(unsigned) SPUISD::LDRESULT] = "SPUISD::LDRESULT";
    node_names[(unsigned) SPUISD::CALL] = "SPUISD::CALL";
    node_names[(unsigned) SPUISD::SHUFB] = "SPUISD::SHUFB";
    node_names[(unsigned) SPUISD::SHUFFLE_MASK] = "SPUISD::SHUFFLE_MASK";
    node_names[(unsigned) SPUISD::CNTB] = "SPUISD::CNTB";
    node_names[(unsigned) SPUISD::PREFSLOT2VEC] = "SPUISD::PREFSLOT2VEC";
    node_names[(unsigned) SPUISD::VEC2PREFSLOT] = "SPUISD::VEC2PREFSLOT";
    node_names[(unsigned) SPUISD::SHLQUAD_L_BITS] = "SPUISD::SHLQUAD_L_BITS";
    node_names[(unsigned) SPUISD::SHLQUAD_L_BYTES] = "SPUISD::SHLQUAD_L_BYTES";
    node_names[(unsigned) SPUISD::VEC_ROTL] = "SPUISD::VEC_ROTL";
    node_names[(unsigned) SPUISD::VEC_ROTR] = "SPUISD::VEC_ROTR";
    node_names[(unsigned) SPUISD::ROTBYTES_LEFT] = "SPUISD::ROTBYTES_LEFT";
    node_names[(unsigned) SPUISD::ROTBYTES_LEFT_BITS] =
            "SPUISD::ROTBYTES_LEFT_BITS";
    node_names[(unsigned) SPUISD::SELECT_MASK] = "SPUISD::SELECT_MASK";
    node_names[(unsigned) SPUISD::SELB] = "SPUISD::SELB";
    node_names[(unsigned) SPUISD::ADD64_MARKER] = "SPUISD::ADD64_MARKER";
    node_names[(unsigned) SPUISD::SUB64_MARKER] = "SPUISD::SUB64_MARKER";
    node_names[(unsigned) SPUISD::MUL64_MARKER] = "SPUISD::MUL64_MARKER";
  }

  std::map<unsigned, const char *>::iterator i = node_names.find(Opcode);

  return ((i != node_names.end()) ? i->second : 0);
}

/// getFunctionAlignment - Return the Log2 alignment of this function.
unsigned SPUTargetLowering::getFunctionAlignment(const Function *) const {
  return 3;
}

//===----------------------------------------------------------------------===//
// Return the Cell SPU's SETCC result type
//===----------------------------------------------------------------------===//

MVT::SimpleValueType SPUTargetLowering::getSetCCResultType(EVT VT) const {
  // i16 and i32 are valid SETCC result types
  return ((VT == MVT::i8 || VT == MVT::i16 || VT == MVT::i32) ?
    VT.getSimpleVT().SimpleTy :
    MVT::i32);
}

//===----------------------------------------------------------------------===//
// Calling convention code:
//===----------------------------------------------------------------------===//

#include "SPUGenCallingConv.inc"

//===----------------------------------------------------------------------===//
//  LowerOperation implementation
//===----------------------------------------------------------------------===//

/// Custom lower loads for CellSPU
/*!
 All CellSPU loads and stores are aligned to 16-byte boundaries, so for elements
 within a 16-byte block, we have to rotate to extract the requested element.

 For extending loads, we also want to ensure that the following sequence is
 emitted, e.g. for MVT::f32 extending load to MVT::f64:

\verbatim
%1  v16i8,ch = load
%2  v16i8,ch = rotate %1
%3  v4f8, ch = bitconvert %2
%4  f32      = vec2perfslot %3
%5  f64      = fp_extend %4
\endverbatim
*/
static SDValue
LowerLOAD(SDValue Op, SelectionDAG &DAG, const SPUSubtarget *ST) {
  LoadSDNode *LN = cast<LoadSDNode>(Op);
  SDValue the_chain = LN->getChain();
  EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
  EVT InVT = LN->getMemoryVT();
  EVT OutVT = Op.getValueType();
  ISD::LoadExtType ExtType = LN->getExtensionType();
  unsigned alignment = LN->getAlignment();
  const valtype_map_s *vtm = getValueTypeMapEntry(InVT);
  DebugLoc dl = Op.getDebugLoc();

  switch (LN->getAddressingMode()) {
  case ISD::UNINDEXED: {
    SDValue result;
    SDValue basePtr = LN->getBasePtr();
    SDValue rotate;

    if (alignment == 16) {
      ConstantSDNode *CN;

      // Special cases for a known aligned load to simplify the base pointer
      // and the rotation amount:
      if (basePtr.getOpcode() == ISD::ADD
          && (CN = dyn_cast<ConstantSDNode > (basePtr.getOperand(1))) != 0) {
        // Known offset into basePtr
        int64_t offset = CN->getSExtValue();
        int64_t rotamt = int64_t((offset & 0xf) - vtm->prefslot_byte);

        if (rotamt < 0)
          rotamt += 16;

        rotate = DAG.getConstant(rotamt, MVT::i16);

        // Simplify the base pointer for this case:
        basePtr = basePtr.getOperand(0);
        if ((offset & ~0xf) > 0) {
          basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
                                basePtr,
                                DAG.getConstant((offset & ~0xf), PtrVT));
        }
      } else if ((basePtr.getOpcode() == SPUISD::AFormAddr)
                 || (basePtr.getOpcode() == SPUISD::IndirectAddr
                     && basePtr.getOperand(0).getOpcode() == SPUISD::Hi
                     && basePtr.getOperand(1).getOpcode() == SPUISD::Lo)) {
        // Plain aligned a-form address: rotate into preferred slot
        // Same for (SPUindirect (SPUhi ...), (SPUlo ...))
        int64_t rotamt = -vtm->prefslot_byte;
        if (rotamt < 0)
          rotamt += 16;
        rotate = DAG.getConstant(rotamt, MVT::i16);
      } else {
        // Offset the rotate amount by the basePtr and the preferred slot
        // byte offset
        int64_t rotamt = -vtm->prefslot_byte;
        if (rotamt < 0)
          rotamt += 16;
        rotate = DAG.getNode(ISD::ADD, dl, PtrVT,
                             basePtr,
                             DAG.getConstant(rotamt, PtrVT));
      }
    } else {
      // Unaligned load: must be more pessimistic about addressing modes:
      if (basePtr.getOpcode() == ISD::ADD) {
        MachineFunction &MF = DAG.getMachineFunction();
        MachineRegisterInfo &RegInfo = MF.getRegInfo();
        unsigned VReg = RegInfo.createVirtualRegister(&SPU::R32CRegClass);
        SDValue Flag;

        SDValue Op0 = basePtr.getOperand(0);
        SDValue Op1 = basePtr.getOperand(1);

        if (isa<ConstantSDNode>(Op1)) {
          // Convert the (add <ptr>, <const>) to an indirect address contained
          // in a register. Note that this is done because we need to avoid
          // creating a 0(reg) d-form address due to the SPU's block loads.
          basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Op0, Op1);
          the_chain = DAG.getCopyToReg(the_chain, dl, VReg, basePtr, Flag);
          basePtr = DAG.getCopyFromReg(the_chain, dl, VReg, PtrVT);
        } else {
          // Convert the (add <arg1>, <arg2>) to an indirect address, which
          // will likely be lowered as a reg(reg) x-form address.
          basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Op0, Op1);
        }
      } else {
        basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
                              basePtr,
                              DAG.getConstant(0, PtrVT));
      }

      // Offset the rotate amount by the basePtr and the preferred slot
      // byte offset
      rotate = DAG.getNode(ISD::ADD, dl, PtrVT,
                           basePtr,
                           DAG.getConstant(-vtm->prefslot_byte, PtrVT));
    }

    // Re-emit as a v16i8 vector load
    result = DAG.getLoad(MVT::v16i8, dl, the_chain, basePtr,
                         LN->getPointerInfo(),
                         LN->isVolatile(), LN->isNonTemporal(), 16);

    // Update the chain
    the_chain = result.getValue(1);

    // Rotate into the preferred slot:
    result = DAG.getNode(SPUISD::ROTBYTES_LEFT, dl, MVT::v16i8,
                         result.getValue(0), rotate);

    // Convert the loaded v16i8 vector to the appropriate vector type
    // specified by the operand:
    EVT vecVT = EVT::getVectorVT(*DAG.getContext(), 
                                 InVT, (128 / InVT.getSizeInBits()));
    result = DAG.getNode(SPUISD::VEC2PREFSLOT, dl, InVT,
                         DAG.getNode(ISD::BIT_CONVERT, dl, vecVT, result));

    // Handle extending loads by extending the scalar result:
    if (ExtType == ISD::SEXTLOAD) {
      result = DAG.getNode(ISD::SIGN_EXTEND, dl, OutVT, result);
    } else if (ExtType == ISD::ZEXTLOAD) {
      result = DAG.getNode(ISD::ZERO_EXTEND, dl, OutVT, result);
    } else if (ExtType == ISD::EXTLOAD) {
      unsigned NewOpc = ISD::ANY_EXTEND;

      if (OutVT.isFloatingPoint())
        NewOpc = ISD::FP_EXTEND;

      result = DAG.getNode(NewOpc, dl, OutVT, result);
    }

    SDVTList retvts = DAG.getVTList(OutVT, MVT::Other);
    SDValue retops[2] = {
      result,
      the_chain
    };

    result = DAG.getNode(SPUISD::LDRESULT, dl, retvts,
                         retops, sizeof(retops) / sizeof(retops[0]));
    return result;
  }
  case ISD::PRE_INC:
  case ISD::PRE_DEC:
  case ISD::POST_INC:
  case ISD::POST_DEC:
  case ISD::LAST_INDEXED_MODE:
    {
      report_fatal_error("LowerLOAD: Got a LoadSDNode with an addr mode other "
                         "than UNINDEXED\n" +
                         Twine((unsigned)LN->getAddressingMode()));
      /*NOTREACHED*/
    }
  }

  return SDValue();
}

/// Custom lower stores for CellSPU
/*!
 All CellSPU stores are aligned to 16-byte boundaries, so for elements
 within a 16-byte block, we have to generate a shuffle to insert the
 requested element into its place, then store the resulting block.
 */
static SDValue
LowerSTORE(SDValue Op, SelectionDAG &DAG, const SPUSubtarget *ST) {
  StoreSDNode *SN = cast<StoreSDNode>(Op);
  SDValue Value = SN->getValue();
  EVT VT = Value.getValueType();
  EVT StVT = (!SN->isTruncatingStore() ? VT : SN->getMemoryVT());
  EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
  DebugLoc dl = Op.getDebugLoc();
  unsigned alignment = SN->getAlignment();

  switch (SN->getAddressingMode()) {
  case ISD::UNINDEXED: {
    // The vector type we really want to load from the 16-byte chunk.
    EVT vecVT = EVT::getVectorVT(*DAG.getContext(),
                                 VT, (128 / VT.getSizeInBits()));

    SDValue alignLoadVec;
    SDValue basePtr = SN->getBasePtr();
    SDValue the_chain = SN->getChain();
    SDValue insertEltOffs;

    if (alignment == 16) {
      ConstantSDNode *CN;
      // Special cases for a known aligned load to simplify the base pointer
      // and insertion byte:
      if (basePtr.getOpcode() == ISD::ADD
          && (CN = dyn_cast<ConstantSDNode>(basePtr.getOperand(1))) != 0) {
        // Known offset into basePtr
        int64_t offset = CN->getSExtValue();

        // Simplify the base pointer for this case:
        basePtr = basePtr.getOperand(0);
        insertEltOffs = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
                                    basePtr,
                                    DAG.getConstant((offset & 0xf), PtrVT));

        if ((offset & ~0xf) > 0) {
          basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
                                basePtr,
                                DAG.getConstant((offset & ~0xf), PtrVT));
        }
      } else {
        // Otherwise, assume it's at byte 0 of basePtr
        insertEltOffs = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
                                    basePtr,
                                    DAG.getConstant(0, PtrVT));
        basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
                                    basePtr,
                                    DAG.getConstant(0, PtrVT));
      }
    } else {
      // Unaligned load: must be more pessimistic about addressing modes:
      if (basePtr.getOpcode() == ISD::ADD) {
        MachineFunction &MF = DAG.getMachineFunction();
        MachineRegisterInfo &RegInfo = MF.getRegInfo();
        unsigned VReg = RegInfo.createVirtualRegister(&SPU::R32CRegClass);
        SDValue Flag;

        SDValue Op0 = basePtr.getOperand(0);
        SDValue Op1 = basePtr.getOperand(1);

        if (isa<ConstantSDNode>(Op1)) {
          // Convert the (add <ptr>, <const>) to an indirect address contained
          // in a register. Note that this is done because we need to avoid
          // creating a 0(reg) d-form address due to the SPU's block loads.
          basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Op0, Op1);
          the_chain = DAG.getCopyToReg(the_chain, dl, VReg, basePtr, Flag);
          basePtr = DAG.getCopyFromReg(the_chain, dl, VReg, PtrVT);
        } else {
          // Convert the (add <arg1>, <arg2>) to an indirect address, which
          // will likely be lowered as a reg(reg) x-form address.
          basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Op0, Op1);
        }
      } else {
        basePtr = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
                              basePtr,
                              DAG.getConstant(0, PtrVT));
      }

      // Insertion point is solely determined by basePtr's contents
      insertEltOffs = DAG.getNode(ISD::ADD, dl, PtrVT,
                                  basePtr,
                                  DAG.getConstant(0, PtrVT));
    }

    // Load the memory to which to store.
    alignLoadVec = DAG.getLoad(vecVT, dl, the_chain, basePtr,
                               SN->getPointerInfo(),
                               SN->isVolatile(), SN->isNonTemporal(), 16);

    // Update the chain
    the_chain = alignLoadVec.getValue(1);

    LoadSDNode *LN = cast<LoadSDNode>(alignLoadVec);
    SDValue theValue = SN->getValue();
    SDValue result;

    if (StVT != VT
        && (theValue.getOpcode() == ISD::AssertZext
            || theValue.getOpcode() == ISD::AssertSext)) {
      // Drill down and get the value for zero- and sign-extended
      // quantities
      theValue = theValue.getOperand(0);
    }

    // If the base pointer is already a D-form address, then just create
    // a new D-form address with a slot offset and the orignal base pointer.
    // Otherwise generate a D-form address with the slot offset relative
    // to the stack pointer, which is always aligned.
#if !defined(NDEBUG)
      if (DebugFlag && isCurrentDebugType(DEBUG_TYPE)) {
        errs() << "CellSPU LowerSTORE: basePtr = ";
        basePtr.getNode()->dump(&DAG);
        errs() << "\n";
      }
#endif

    SDValue insertEltOp = DAG.getNode(SPUISD::SHUFFLE_MASK, dl, vecVT,
                                      insertEltOffs);
    SDValue vectorizeOp = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, vecVT, 
                                      theValue);

    result = DAG.getNode(SPUISD::SHUFB, dl, vecVT,
                         vectorizeOp, alignLoadVec,
                         DAG.getNode(ISD::BIT_CONVERT, dl,
                                     MVT::v4i32, insertEltOp));

    result = DAG.getStore(the_chain, dl, result, basePtr,
                          LN->getSrcValue(), LN->getSrcValueOffset(),
                          LN->isVolatile(), LN->isNonTemporal(),
                          LN->getAlignment());

#if 0 && !defined(NDEBUG)
    if (DebugFlag && isCurrentDebugType(DEBUG_TYPE)) {
      const SDValue &currentRoot = DAG.getRoot();

      DAG.setRoot(result);
      errs() << "------- CellSPU:LowerStore result:\n";
      DAG.dump();
      errs() << "-------\n";
      DAG.setRoot(currentRoot);
    }
#endif

    return result;
    /*UNREACHED*/
  }
  case ISD::PRE_INC:
  case ISD::PRE_DEC:
  case ISD::POST_INC:
  case ISD::POST_DEC:
  case ISD::LAST_INDEXED_MODE:
    {
      report_fatal_error("LowerLOAD: Got a LoadSDNode with an addr mode other "
                         "than UNINDEXED\n" +
                         Twine((unsigned)SN->getAddressingMode()));
      /*NOTREACHED*/
    }
  }

  return SDValue();
}

//! Generate the address of a constant pool entry.
static SDValue
LowerConstantPool(SDValue Op, SelectionDAG &DAG, const SPUSubtarget *ST) {
  EVT PtrVT = Op.getValueType();
  ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
  const Constant *C = CP->getConstVal();
  SDValue CPI = DAG.getTargetConstantPool(C, PtrVT, CP->getAlignment());
  SDValue Zero = DAG.getConstant(0, PtrVT);
  const TargetMachine &TM = DAG.getTarget();
  // FIXME there is no actual debug info here
  DebugLoc dl = Op.getDebugLoc();

  if (TM.getRelocationModel() == Reloc::Static) {
    if (!ST->usingLargeMem()) {
      // Just return the SDValue with the constant pool address in it.
      return DAG.getNode(SPUISD::AFormAddr, dl, PtrVT, CPI, Zero);
    } else {
      SDValue Hi = DAG.getNode(SPUISD::Hi, dl, PtrVT, CPI, Zero);
      SDValue Lo = DAG.getNode(SPUISD::Lo, dl, PtrVT, CPI, Zero);
      return DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Hi, Lo);
    }
  }

  llvm_unreachable("LowerConstantPool: Relocation model other than static"
                   " not supported.");
  return SDValue();
}

//! Alternate entry point for generating the address of a constant pool entry
SDValue
SPU::LowerConstantPool(SDValue Op, SelectionDAG &DAG, const SPUTargetMachine &TM) {
  return ::LowerConstantPool(Op, DAG, TM.getSubtargetImpl());
}

static SDValue
LowerJumpTable(SDValue Op, SelectionDAG &DAG, const SPUSubtarget *ST) {
  EVT PtrVT = Op.getValueType();
  JumpTableSDNode *JT = cast<JumpTableSDNode>(Op);
  SDValue JTI = DAG.getTargetJumpTable(JT->getIndex(), PtrVT);
  SDValue Zero = DAG.getConstant(0, PtrVT);
  const TargetMachine &TM = DAG.getTarget();
  // FIXME there is no actual debug info here
  DebugLoc dl = Op.getDebugLoc();

  if (TM.getRelocationModel() == Reloc::Static) {
    if (!ST->usingLargeMem()) {
      return DAG.getNode(SPUISD::AFormAddr, dl, PtrVT, JTI, Zero);
    } else {
      SDValue Hi = DAG.getNode(SPUISD::Hi, dl, PtrVT, JTI, Zero);
      SDValue Lo = DAG.getNode(SPUISD::Lo, dl, PtrVT, JTI, Zero);
      return DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Hi, Lo);
    }
  }

  llvm_unreachable("LowerJumpTable: Relocation model other than static"
                   " not supported.");
  return SDValue();
}

static SDValue
LowerGlobalAddress(SDValue Op, SelectionDAG &DAG, const SPUSubtarget *ST) {
  EVT PtrVT = Op.getValueType();
  GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op);
  const GlobalValue *GV = GSDN->getGlobal();
  SDValue GA = DAG.getTargetGlobalAddress(GV, Op.getDebugLoc(),
                                          PtrVT, GSDN->getOffset());
  const TargetMachine &TM = DAG.getTarget();
  SDValue Zero = DAG.getConstant(0, PtrVT);
  // FIXME there is no actual debug info here
  DebugLoc dl = Op.getDebugLoc();

  if (TM.getRelocationModel() == Reloc::Static) {
    if (!ST->usingLargeMem()) {
      return DAG.getNode(SPUISD::AFormAddr, dl, PtrVT, GA, Zero);
    } else {
      SDValue Hi = DAG.getNode(SPUISD::Hi, dl, PtrVT, GA, Zero);
      SDValue Lo = DAG.getNode(SPUISD::Lo, dl, PtrVT, GA, Zero);
      return DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, Hi, Lo);
    }
  } else {
    report_fatal_error("LowerGlobalAddress: Relocation model other than static"
                      "not supported.");
    /*NOTREACHED*/
  }

  return SDValue();
}

//! Custom lower double precision floating point constants
static SDValue
LowerConstantFP(SDValue Op, SelectionDAG &DAG) {
  EVT VT = Op.getValueType();
  // FIXME there is no actual debug info here
  DebugLoc dl = Op.getDebugLoc();

  if (VT == MVT::f64) {
    ConstantFPSDNode *FP = cast<ConstantFPSDNode>(Op.getNode());

    assert((FP != 0) &&
           "LowerConstantFP: Node is not ConstantFPSDNode");

    uint64_t dbits = DoubleToBits(FP->getValueAPF().convertToDouble());
    SDValue T = DAG.getConstant(dbits, MVT::i64);
    SDValue Tvec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v2i64, T, T);
    return DAG.getNode(SPUISD::VEC2PREFSLOT, dl, VT,
                       DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2f64, Tvec));
  }

  return SDValue();
}

SDValue
SPUTargetLowering::LowerFormalArguments(SDValue Chain,
                                        CallingConv::ID CallConv, bool isVarArg,
                                        const SmallVectorImpl<ISD::InputArg>
                                          &Ins,
                                        DebugLoc dl, SelectionDAG &DAG,
                                        SmallVectorImpl<SDValue> &InVals)
                                          const {

  MachineFunction &MF = DAG.getMachineFunction();
  MachineFrameInfo *MFI = MF.getFrameInfo();
  MachineRegisterInfo &RegInfo = MF.getRegInfo();
  SPUFunctionInfo *FuncInfo = MF.getInfo<SPUFunctionInfo>();

  unsigned ArgOffset = SPUFrameInfo::minStackSize();
  unsigned ArgRegIdx = 0;
  unsigned StackSlotSize = SPUFrameInfo::stackSlotSize();

  EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();

  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(CallConv, isVarArg, getTargetMachine(), ArgLocs,
                 *DAG.getContext());
  // FIXME: allow for other calling conventions
  CCInfo.AnalyzeFormalArguments(Ins, CCC_SPU);

  // Add DAG nodes to load the arguments or copy them out of registers.
  for (unsigned ArgNo = 0, e = Ins.size(); ArgNo != e; ++ArgNo) {
    EVT ObjectVT = Ins[ArgNo].VT;
    unsigned ObjSize = ObjectVT.getSizeInBits()/8;
    SDValue ArgVal;
    CCValAssign &VA = ArgLocs[ArgNo];

    if (VA.isRegLoc()) {
      const TargetRegisterClass *ArgRegClass;

      switch (ObjectVT.getSimpleVT().SimpleTy) {
      default:
        report_fatal_error("LowerFormalArguments Unhandled argument type: " +
                           Twine(ObjectVT.getEVTString()));
      case MVT::i8:
        ArgRegClass = &SPU::R8CRegClass;
        break;
      case MVT::i16:
        ArgRegClass = &SPU::R16CRegClass;
        break;
      case MVT::i32:
        ArgRegClass = &SPU::R32CRegClass;
        break;
      case MVT::i64:
        ArgRegClass = &SPU::R64CRegClass;
        break;
      case MVT::i128:
        ArgRegClass = &SPU::GPRCRegClass;
        break;
      case MVT::f32:
        ArgRegClass = &SPU::R32FPRegClass;
        break;
      case MVT::f64:
        ArgRegClass = &SPU::R64FPRegClass;
        break;
      case MVT::v2f64:
      case MVT::v4f32:
      case MVT::v2i64:
      case MVT::v4i32:
      case MVT::v8i16:
      case MVT::v16i8:
        ArgRegClass = &SPU::VECREGRegClass;
        break;
      }

      unsigned VReg = RegInfo.createVirtualRegister(ArgRegClass);
      RegInfo.addLiveIn(VA.getLocReg(), VReg);
      ArgVal = DAG.getCopyFromReg(Chain, dl, VReg, ObjectVT);
      ++ArgRegIdx;
    } else {
      // 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
      // or we're forced to do vararg
      int FI = MFI->CreateFixedObject(ObjSize, ArgOffset, true);
      SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
      ArgVal = DAG.getLoad(ObjectVT, dl, Chain, FIN, MachinePointerInfo(),
                           false, false, 0);
      ArgOffset += StackSlotSize;
    }

    InVals.push_back(ArgVal);
    // Update the chain
    Chain = ArgVal.getOperand(0);
  }

  // vararg handling:
  if (isVarArg) {
    // FIXME: we should be able to query the argument registers from 
    //        tablegen generated code. 
    static const unsigned ArgRegs[] = {
      SPU::R3,  SPU::R4,  SPU::R5,  SPU::R6,  SPU::R7,  SPU::R8,  SPU::R9,
      SPU::R10, SPU::R11, SPU::R12, SPU::R13, SPU::R14, SPU::R15, SPU::R16,
      SPU::R17, SPU::R18, SPU::R19, SPU::R20, SPU::R21, SPU::R22, SPU::R23,
      SPU::R24, SPU::R25, SPU::R26, SPU::R27, SPU::R28, SPU::R29, SPU::R30,
      SPU::R31, SPU::R32, SPU::R33, SPU::R34, SPU::R35, SPU::R36, SPU::R37,
      SPU::R38, SPU::R39, SPU::R40, SPU::R41, SPU::R42, SPU::R43, SPU::R44,
      SPU::R45, SPU::R46, SPU::R47, SPU::R48, SPU::R49, SPU::R50, SPU::R51,
      SPU::R52, SPU::R53, SPU::R54, SPU::R55, SPU::R56, SPU::R57, SPU::R58,
      SPU::R59, SPU::R60, SPU::R61, SPU::R62, SPU::R63, SPU::R64, SPU::R65,
      SPU::R66, SPU::R67, SPU::R68, SPU::R69, SPU::R70, SPU::R71, SPU::R72,
      SPU::R73, SPU::R74, SPU::R75, SPU::R76, SPU::R77, SPU::R78, SPU::R79
    };
    // size of ArgRegs array
    unsigned NumArgRegs = 77;

    // We will spill (79-3)+1 registers to the stack
    SmallVector<SDValue, 79-3+1> MemOps;

    // Create the frame slot
    for (; ArgRegIdx != NumArgRegs; ++ArgRegIdx) {
      FuncInfo->setVarArgsFrameIndex(
        MFI->CreateFixedObject(StackSlotSize, ArgOffset, true));
      SDValue FIN = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(), PtrVT);
      unsigned VReg = MF.addLiveIn(ArgRegs[ArgRegIdx], &SPU::R32CRegClass);
      SDValue ArgVal = DAG.getRegister(VReg, MVT::v16i8);
      SDValue Store = DAG.getStore(Chain, dl, ArgVal, FIN, NULL, 0,
                                   false, false, 0);
      Chain = Store.getOperand(0);
      MemOps.push_back(Store);

      // Increment address by stack slot size for the next stored argument
      ArgOffset += StackSlotSize;
    }
    if (!MemOps.empty())
      Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
                          &MemOps[0], MemOps.size());
  }

  return Chain;
}

/// isLSAAddress - Return the immediate to use if the specified
/// value is representable as a LSA address.
static SDNode *isLSAAddress(SDValue Op, SelectionDAG &DAG) {
  ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op);
  if (!C) return 0;

  int Addr = C->getZExtValue();
  if ((Addr & 3) != 0 ||  // Low 2 bits are implicitly zero.
      (Addr << 14 >> 14) != Addr)
    return 0;  // Top 14 bits have to be sext of immediate.

  return DAG.getConstant((int)C->getZExtValue() >> 2, MVT::i32).getNode();
}

SDValue
SPUTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
                             CallingConv::ID CallConv, bool isVarArg,
                             bool &isTailCall,
                             const SmallVectorImpl<ISD::OutputArg> &Outs,
                             const SmallVectorImpl<SDValue> &OutVals,
                             const SmallVectorImpl<ISD::InputArg> &Ins,
                             DebugLoc dl, SelectionDAG &DAG,
                             SmallVectorImpl<SDValue> &InVals) const {
  // CellSPU target does not yet support tail call optimization.
  isTailCall = false;

  const SPUSubtarget *ST = SPUTM.getSubtargetImpl();
  unsigned NumOps     = Outs.size();
  unsigned StackSlotSize = SPUFrameInfo::stackSlotSize();

  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(CallConv, isVarArg, getTargetMachine(), ArgLocs,
                 *DAG.getContext()); 
  // FIXME: allow for other calling conventions
  CCInfo.AnalyzeCallOperands(Outs, CCC_SPU);
  
  const unsigned NumArgRegs = ArgLocs.size();


  // Handy pointer type
  EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();

  // 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.
  SDValue StackPtr = DAG.getRegister(SPU::R1, MVT::i32);

  // Figure out which arguments are going to go in registers, and which in
  // memory.
  unsigned ArgOffset = SPUFrameInfo::minStackSize(); // Just below [LR]
  unsigned ArgRegIdx = 0;

  // Keep track of registers passing arguments
  std::vector<std::pair<unsigned, SDValue> > RegsToPass;
  // And the arguments passed on the stack
  SmallVector<SDValue, 8> MemOpChains;

  for (; ArgRegIdx != NumOps; ++ArgRegIdx) {
    SDValue Arg = OutVals[ArgRegIdx];
    CCValAssign &VA = ArgLocs[ArgRegIdx];

    // PtrOff will be used to store the current argument to the stack if a
    // register cannot be found for it.
    SDValue PtrOff = DAG.getConstant(ArgOffset, StackPtr.getValueType());
    PtrOff = DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr, PtrOff);

    switch (Arg.getValueType().getSimpleVT().SimpleTy) {
    default: llvm_unreachable("Unexpected ValueType for argument!");
    case MVT::i8:
    case MVT::i16:
    case MVT::i32:
    case MVT::i64:
    case MVT::i128:
    case MVT::f32:
    case MVT::f64:
    case MVT::v2i64:
    case MVT::v2f64:
    case MVT::v4f32:
    case MVT::v4i32:
    case MVT::v8i16:
    case MVT::v16i8:
      if (ArgRegIdx != NumArgRegs) {
        RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
      } else {
        MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff, NULL, 0,
                                           false, false, 0));
        ArgOffset += StackSlotSize;
      }
      break;
    }
  }

  // Accumulate how many bytes are to be pushed on the stack, including the
  // linkage area, and parameter passing area.  According to the SPU ABI,
  // we minimally need space for [LR] and [SP].
  unsigned NumStackBytes = ArgOffset - SPUFrameInfo::minStackSize();

  // Insert a call sequence start
  Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumStackBytes,
                                                            true));

  if (!MemOpChains.empty()) {
    // Adjust the stack pointer for the stack arguments.
    Chain = DAG.getNode(ISD::TokenFactor, dl, 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.
  SDValue InFlag;
  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
    Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
                             RegsToPass[i].second, InFlag);
    InFlag = Chain.getValue(1);
  }

  SmallVector<SDValue, 8> Ops;
  unsigned CallOpc = SPUISD::CALL;

  // 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)) {
    const GlobalValue *GV = G->getGlobal();
    EVT CalleeVT = Callee.getValueType();
    SDValue Zero = DAG.getConstant(0, PtrVT);
    SDValue GA = DAG.getTargetGlobalAddress(GV, dl, CalleeVT);

    if (!ST->usingLargeMem()) {
      // Turn calls to targets that are defined (i.e., have bodies) into BRSL
      // style calls, otherwise, external symbols are BRASL calls. This assumes
      // that declared/defined symbols are in the same compilation unit and can
      // be reached through PC-relative jumps.
      //
      // NOTE:
      // This may be an unsafe assumption for JIT and really large compilation
      // units.
      if (GV->isDeclaration()) {
        Callee = DAG.getNode(SPUISD::AFormAddr, dl, CalleeVT, GA, Zero);
      } else {
        Callee = DAG.getNode(SPUISD::PCRelAddr, dl, CalleeVT, GA, Zero);
      }
    } else {
      // "Large memory" mode: Turn all calls into indirect calls with a X-form
      // address pairs:
      Callee = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, GA, Zero);
    }
  } else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
    EVT CalleeVT = Callee.getValueType();
    SDValue Zero = DAG.getConstant(0, PtrVT);
    SDValue ExtSym = DAG.getTargetExternalSymbol(S->getSymbol(),
        Callee.getValueType());

    if (!ST->usingLargeMem()) {
      Callee = DAG.getNode(SPUISD::AFormAddr, dl, CalleeVT, ExtSym, Zero);
    } else {
      Callee = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT, ExtSym, Zero);
    }
  } else if (SDNode *Dest = isLSAAddress(Callee, DAG)) {
    // If this is an absolute destination address that appears to be a legal
    // local store address, use the munged value.
    Callee = SDValue(Dest, 0);
  }

  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.getNode())
    Ops.push_back(InFlag);
  // Returns a chain and a flag for retval copy to use.
  Chain = DAG.getNode(CallOpc, dl, DAG.getVTList(MVT::Other, MVT::Flag),
                      &Ops[0], Ops.size());
  InFlag = Chain.getValue(1);

  Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumStackBytes, true),
                             DAG.getIntPtrConstant(0, true), InFlag);
  if (!Ins.empty())
    InFlag = Chain.getValue(1);

  // If the function returns void, just return the chain.
  if (Ins.empty())
    return Chain;

  // Now handle the return value(s)
  SmallVector<CCValAssign, 16> RVLocs;
  CCState CCRetInfo(CallConv, isVarArg, getTargetMachine(),
                    RVLocs, *DAG.getContext());
  CCRetInfo.AnalyzeCallResult(Ins, CCC_SPU);


  // If the call has results, copy the values out of the ret val registers.
  for (unsigned i = 0; i != RVLocs.size(); ++i) {
    CCValAssign VA = RVLocs[i];
    
    SDValue Val = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), VA.getLocVT(),
                                     InFlag);
    Chain = Val.getValue(1);
    InFlag = Val.getValue(2);
    InVals.push_back(Val);
   }

  return Chain;
}

SDValue
SPUTargetLowering::LowerReturn(SDValue Chain,
                               CallingConv::ID CallConv, bool isVarArg,
                               const SmallVectorImpl<ISD::OutputArg> &Outs,
                               const SmallVectorImpl<SDValue> &OutVals,
                               DebugLoc dl, SelectionDAG &DAG) const {

  SmallVector<CCValAssign, 16> RVLocs;
  CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
                 RVLocs, *DAG.getContext());
  CCInfo.AnalyzeReturn(Outs, RetCC_SPU);

  // 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());
  }

  SDValue 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, dl, VA.getLocReg(),
                             OutVals[i], Flag);
    Flag = Chain.getValue(1);
  }

  if (Flag.getNode())
    return DAG.getNode(SPUISD::RET_FLAG, dl, MVT::Other, Chain, Flag);
  else
    return DAG.getNode(SPUISD::RET_FLAG, dl, MVT::Other, Chain);
}


//===----------------------------------------------------------------------===//
// Vector related lowering:
//===----------------------------------------------------------------------===//

static ConstantSDNode *
getVecImm(SDNode *N) {
  SDValue OpVal(0, 0);

  // 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.getNode() == 0)
      OpVal = N->getOperand(i);
    else if (OpVal != N->getOperand(i))
      return 0;
  }

  if (OpVal.getNode() != 0) {
    if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(OpVal)) {
      return CN;
    }
  }

  return 0;
}

/// get_vec_i18imm - Test if this vector is a vector filled with the same value
/// and the value fits into an unsigned 18-bit constant, and if so, return the
/// constant
SDValue SPU::get_vec_u18imm(SDNode *N, SelectionDAG &DAG,
                              EVT ValueType) {
  if (ConstantSDNode *CN = getVecImm(N)) {
    uint64_t Value = CN->getZExtValue();
    if (ValueType == MVT::i64) {
      uint64_t UValue = CN->getZExtValue();
      uint32_t upper = uint32_t(UValue >> 32);
      uint32_t lower = uint32_t(UValue);
      if (upper != lower)
        return SDValue();
      Value = Value >> 32;
    }
    if (Value <= 0x3ffff)
      return DAG.getTargetConstant(Value, ValueType);
  }

  return SDValue();
}

/// get_vec_i16imm - Test if this vector is a vector filled with the same value
/// and the value fits into a signed 16-bit constant, and if so, return the
/// constant
SDValue SPU::get_vec_i16imm(SDNode *N, SelectionDAG &DAG,
                              EVT ValueType) {
  if (ConstantSDNode *CN = getVecImm(N)) {
    int64_t Value = CN->getSExtValue();
    if (ValueType == MVT::i64) {
      uint64_t UValue = CN->getZExtValue();
      uint32_t upper = uint32_t(UValue >> 32);
      uint32_t lower = uint32_t(UValue);
      if (upper != lower)
        return SDValue();
      Value = Value >> 32;
    }
    if (Value >= -(1 << 15) && Value <= ((1 << 15) - 1)) {
      return DAG.getTargetConstant(Value, ValueType);
    }
  }

  return SDValue();
}

/// get_vec_i10imm - Test if this vector is a vector filled with the same value
/// and the value fits into a signed 10-bit constant, and if so, return the
/// constant
SDValue SPU::get_vec_i10imm(SDNode *N, SelectionDAG &DAG,
                              EVT ValueType) {
  if (ConstantSDNode *CN = getVecImm(N)) {
    int64_t Value = CN->getSExtValue();
    if (ValueType == MVT::i64) {
      uint64_t UValue = CN->getZExtValue();
      uint32_t upper = uint32_t(UValue >> 32);
      uint32_t lower = uint32_t(UValue);
      if (upper != lower)
        return SDValue();
      Value = Value >> 32;
    }
    if (isInt<10>(Value))
      return DAG.getTargetConstant(Value, ValueType);
  }

  return SDValue();
}

/// get_vec_i8imm - Test if this vector is a vector filled with the same value
/// and the value fits into a signed 8-bit constant, and if so, return the
/// constant.
///
/// @note: The incoming vector is v16i8 because that's the only way we can load
/// constant vectors. Thus, we test to see if the upper and lower bytes are the
/// same value.
SDValue SPU::get_vec_i8imm(SDNode *N, SelectionDAG &DAG,
                             EVT ValueType) {
  if (ConstantSDNode *CN = getVecImm(N)) {
    int Value = (int) CN->getZExtValue();
    if (ValueType == MVT::i16
        && Value <= 0xffff                 /* truncated from uint64_t */
        && ((short) Value >> 8) == ((short) Value & 0xff))
      return DAG.getTargetConstant(Value & 0xff, ValueType);
    else if (ValueType == MVT::i8
             && (Value & 0xff) == Value)
      return DAG.getTargetConstant(Value, ValueType);
  }

  return SDValue();
}

/// get_ILHUvec_imm - Test if this vector is a vector filled with the same value
/// and the value fits into a signed 16-bit constant, and if so, return the
/// constant
SDValue SPU::get_ILHUvec_imm(SDNode *N, SelectionDAG &DAG,
                               EVT ValueType) {
  if (ConstantSDNode *CN = getVecImm(N)) {
    uint64_t Value = CN->getZExtValue();
    if ((ValueType == MVT::i32
          && ((unsigned) Value & 0xffff0000) == (unsigned) Value)
        || (ValueType == MVT::i64 && (Value & 0xffff0000) == Value))
      return DAG.getTargetConstant(Value >> 16, ValueType);
  }

  return SDValue();
}

/// get_v4i32_imm - Catch-all for general 32-bit constant vectors
SDValue SPU::get_v4i32_imm(SDNode *N, SelectionDAG &DAG) {
  if (ConstantSDNode *CN = getVecImm(N)) {
    return DAG.getTargetConstant((unsigned) CN->getZExtValue(), MVT::i32);
  }

  return SDValue();
}

/// get_v4i32_imm - Catch-all for general 64-bit constant vectors
SDValue SPU::get_v2i64_imm(SDNode *N, SelectionDAG &DAG) {
  if (ConstantSDNode *CN = getVecImm(N)) {
    return DAG.getTargetConstant((unsigned) CN->getZExtValue(), MVT::i64);
  }

  return SDValue();
}

//! Lower a BUILD_VECTOR instruction creatively:
static SDValue
LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) {
  EVT VT = Op.getValueType();
  EVT EltVT = VT.getVectorElementType();
  DebugLoc dl = Op.getDebugLoc();
  BuildVectorSDNode *BCN = dyn_cast<BuildVectorSDNode>(Op.getNode());
  assert(BCN != 0 && "Expected BuildVectorSDNode in SPU LowerBUILD_VECTOR");
  unsigned minSplatBits = EltVT.getSizeInBits();

  if (minSplatBits < 16)
    minSplatBits = 16;

  APInt APSplatBits, APSplatUndef;
  unsigned SplatBitSize;
  bool HasAnyUndefs;

  if (!BCN->isConstantSplat(APSplatBits, APSplatUndef, SplatBitSize,
                            HasAnyUndefs, minSplatBits)
      || minSplatBits < SplatBitSize)
    return SDValue();   // Wasn't a constant vector or splat exceeded min

  uint64_t SplatBits = APSplatBits.getZExtValue();

  switch (VT.getSimpleVT().SimpleTy) {
  default:
    report_fatal_error("CellSPU: Unhandled VT in LowerBUILD_VECTOR, VT = " +
                       Twine(VT.getEVTString()));
    /*NOTREACHED*/
  case MVT::v4f32: {
    uint32_t Value32 = uint32_t(SplatBits);
    assert(SplatBitSize == 32
           && "LowerBUILD_VECTOR: Unexpected floating point vector element.");
    // NOTE: pretend the constant is an integer. LLVM won't load FP constants
    SDValue T = DAG.getConstant(Value32, MVT::i32);
    return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v4f32,
                       DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, T,T,T,T));
    break;
  }
  case MVT::v2f64: {
    uint64_t f64val = uint64_t(SplatBits);
    assert(SplatBitSize == 64
           && "LowerBUILD_VECTOR: 64-bit float vector size > 8 bytes.");
    // NOTE: pretend the constant is an integer. LLVM won't load FP constants
    SDValue T = DAG.getConstant(f64val, MVT::i64);
    return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2f64,
                       DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v2i64, T, T));
    break;
  }
  case MVT::v16i8: {
   // 8-bit constants have to be expanded to 16-bits
   unsigned short Value16 = SplatBits /* | (SplatBits << 8) */;
   SmallVector<SDValue, 8> Ops;

   Ops.assign(8, DAG.getConstant(Value16, MVT::i16));
   return DAG.getNode(ISD::BIT_CONVERT, dl, VT,
                      DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i16, &Ops[0], Ops.size()));
  }
  case MVT::v8i16: {
    unsigned short Value16 = SplatBits;
    SDValue T = DAG.getConstant(Value16, EltVT);
    SmallVector<SDValue, 8> Ops;

    Ops.assign(8, T);
    return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Ops[0], Ops.size());
  }
  case MVT::v4i32: {
    SDValue T = DAG.getConstant(unsigned(SplatBits), VT.getVectorElementType());
    return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, T, T, T, T);
  }
  case MVT::v2i64: {
    return SPU::LowerV2I64Splat(VT, DAG, SplatBits, dl);
  }
  }

  return SDValue();
}

/*!
 */
SDValue
SPU::LowerV2I64Splat(EVT OpVT, SelectionDAG& DAG, uint64_t SplatVal,
                     DebugLoc dl) {
  uint32_t upper = uint32_t(SplatVal >> 32);
  uint32_t lower = uint32_t(SplatVal);

  if (upper == lower) {
    // Magic constant that can be matched by IL, ILA, et. al.
    SDValue Val = DAG.getTargetConstant(upper, MVT::i32);
    return DAG.getNode(ISD::BIT_CONVERT, dl, OpVT,
                       DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
                                   Val, Val, Val, Val));
  } else {
    bool upper_special, lower_special;

    // NOTE: This code creates common-case shuffle masks that can be easily
    // detected as common expressions. It is not attempting to create highly
    // specialized masks to replace any and all 0's, 0xff's and 0x80's.

    // Detect if the upper or lower half is a special shuffle mask pattern:
    upper_special = (upper == 0 || upper == 0xffffffff || upper == 0x80000000);
    lower_special = (lower == 0 || lower == 0xffffffff || lower == 0x80000000);

    // Both upper and lower are special, lower to a constant pool load:
    if (lower_special && upper_special) {
      SDValue SplatValCN = DAG.getConstant(SplatVal, MVT::i64);
      return DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v2i64,
                         SplatValCN, SplatValCN);
    }

    SDValue LO32;
    SDValue HI32;
    SmallVector<SDValue, 16> ShufBytes;
    SDValue Result;

    // Create lower vector if not a special pattern
    if (!lower_special) {
      SDValue LO32C = DAG.getConstant(lower, MVT::i32);
      LO32 = DAG.getNode(ISD::BIT_CONVERT, dl, OpVT,
                         DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
                                     LO32C, LO32C, LO32C, LO32C));
    }

    // Create upper vector if not a special pattern
    if (!upper_special) {
      SDValue HI32C = DAG.getConstant(upper, MVT::i32);
      HI32 = DAG.getNode(ISD::BIT_CONVERT, dl, OpVT,
                         DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
                                     HI32C, HI32C, HI32C, HI32C));
    }

    // If either upper or lower are special, then the two input operands are
    // the same (basically, one of them is a "don't care")
    if (lower_special)
      LO32 = HI32;
    if (upper_special)
      HI32 = LO32;

    for (int i = 0; i < 4; ++i) {
      uint64_t val = 0;
      for (int j = 0; j < 4; ++j) {
        SDValue V;
        bool process_upper, process_lower;
        val <<= 8;
        process_upper = (upper_special && (i & 1) == 0);
        process_lower = (lower_special && (i & 1) == 1);

        if (process_upper || process_lower) {
          if ((process_upper && upper == 0)
                  || (process_lower && lower == 0))
            val |= 0x80;
          else if ((process_upper && upper == 0xffffffff)
                  || (process_lower && lower == 0xffffffff))
            val |= 0xc0;
          else if ((process_upper && upper == 0x80000000)
                  || (process_lower && lower == 0x80000000))
            val |= (j == 0 ? 0xe0 : 0x80);
        } else
          val |= i * 4 + j + ((i & 1) * 16);
      }

      ShufBytes.push_back(DAG.getConstant(val, MVT::i32));
    }

    return DAG.getNode(SPUISD::SHUFB, dl, OpVT, HI32, LO32,
                       DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
                                   &ShufBytes[0], ShufBytes.size()));
  }
}

/// LowerVECTOR_SHUFFLE - Lower a vector shuffle (V1, V2, V3) to something on
/// which the Cell can operate. The code inspects V3 to ascertain whether the
/// permutation vector, V3, is monotonically increasing with one "exception"
/// element, e.g., (0, 1, _, 3). If this is the case, then generate a
/// SHUFFLE_MASK synthetic instruction. Otherwise, spill V3 to the constant pool.
/// In either case, the net result is going to eventually invoke SHUFB to
/// permute/shuffle the bytes from V1 and V2.
/// \note
/// SHUFFLE_MASK is eventually selected as one of the C*D instructions, generate
/// control word for byte/halfword/word insertion. This takes care of a single
/// element move from V2 into V1.
/// \note
/// SPUISD::SHUFB is eventually selected as Cell's <i>shufb</i> instructions.
static SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) {
  const ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(Op);
  SDValue V1 = Op.getOperand(0);
  SDValue V2 = Op.getOperand(1);
  DebugLoc dl = Op.getDebugLoc();

  if (V2.getOpcode() == ISD::UNDEF) V2 = V1;

  // If we have a single element being moved from V1 to V2, this can be handled
  // using the C*[DX] compute mask instructions, but the vector elements have
  // to be monotonically increasing with one exception element, and the source
  // slot of the element to move must be the same as the destination.
  EVT VecVT = V1.getValueType();
  EVT EltVT = VecVT.getVectorElementType();
  unsigned EltsFromV2 = 0;
  unsigned V2EltOffset = 0;
  unsigned V2EltIdx0 = 0;
  unsigned CurrElt = 0;
  unsigned MaxElts = VecVT.getVectorNumElements();
  unsigned PrevElt = 0;
  bool monotonic = true;
  bool rotate = true;
  int rotamt=0;
  EVT maskVT;             // which of the c?d instructions to use

  if (EltVT == MVT::i8) {
    V2EltIdx0 = 16;
    maskVT = MVT::v16i8; 
  } else if (EltVT == MVT::i16) {
    V2EltIdx0 = 8;
    maskVT = MVT::v8i16;
  } else if (EltVT == MVT::i32 || EltVT == MVT::f32) {
    V2EltIdx0 = 4;
    maskVT = MVT::v4i32;
  } else if (EltVT == MVT::i64 || EltVT == MVT::f64) {
    V2EltIdx0 = 2;
    maskVT = MVT::v2i64;
  } else
    llvm_unreachable("Unhandled vector type in LowerVECTOR_SHUFFLE");

  for (unsigned i = 0; i != MaxElts; ++i) {
    if (SVN->getMaskElt(i) < 0)
      continue;
    
    unsigned SrcElt = SVN->getMaskElt(i);

    if (monotonic) {
      if (SrcElt >= V2EltIdx0) {
        // TODO: optimize for the monotonic case when several consecutive
        // elements are taken form V2. Do we ever get such a case?
        if (EltsFromV2 == 0 && CurrElt == (SrcElt - V2EltIdx0))
          V2EltOffset = (SrcElt - V2EltIdx0) * (EltVT.getSizeInBits()/8);
        else
          monotonic = false;
        ++EltsFromV2;
      } else if (CurrElt != SrcElt) {
        monotonic = false;
      }

      ++CurrElt;
    }

    if (rotate) {
      if (PrevElt > 0 && SrcElt < MaxElts) {
        if ((PrevElt == SrcElt - 1)
            || (PrevElt == MaxElts - 1 && SrcElt == 0)) {
          rotamt = SrcElt-i;
          PrevElt = SrcElt;
        } else {
          rotate = false;
        }
      } else if (i == 0 || (PrevElt==0 && SrcElt==1)) {
        // First time or after a "wrap around"
        PrevElt = SrcElt;
      } else {
        // This isn't a rotation, takes elements from vector 2
        rotate = false;
      }
    }
  }

  if (EltsFromV2 == 1 && monotonic) {
    // Compute mask and shuffle
    EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();

    // As SHUFFLE_MASK becomes a c?d instruction, feed it an address
    // R1 ($sp) is used here only as it is guaranteed to have last bits zero
    SDValue Pointer = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
                                DAG.getRegister(SPU::R1, PtrVT),
                                DAG.getConstant(V2EltOffset, MVT::i32));
    SDValue ShufMaskOp = DAG.getNode(SPUISD::SHUFFLE_MASK, dl, 
                                     maskVT, Pointer);

    // Use shuffle mask in SHUFB synthetic instruction:
    return DAG.getNode(SPUISD::SHUFB, dl, V1.getValueType(), V2, V1,
                       ShufMaskOp);
  } else if (rotate) {
    if (rotamt < 0)
      rotamt +=MaxElts;
    rotamt *= EltVT.getSizeInBits()/8;
    return DAG.getNode(SPUISD::ROTBYTES_LEFT, dl, V1.getValueType(),
                       V1, DAG.getConstant(rotamt, MVT::i16));
  } else {
   // Convert the SHUFFLE_VECTOR mask's input element units to the
   // actual bytes.
    unsigned BytesPerElement = EltVT.getSizeInBits()/8;

    SmallVector<SDValue, 16> ResultMask;
    for (unsigned i = 0, e = MaxElts; i != e; ++i) {
      unsigned SrcElt = SVN->getMaskElt(i) < 0 ? 0 : SVN->getMaskElt(i);

      for (unsigned j = 0; j < BytesPerElement; ++j)
        ResultMask.push_back(DAG.getConstant(SrcElt*BytesPerElement+j,MVT::i8));
    }
    SDValue VPermMask = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v16i8,
                                    &ResultMask[0], ResultMask.size());
    return DAG.getNode(SPUISD::SHUFB, dl, V1.getValueType(), V1, V2, VPermMask);
  }
}

static SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) {
  SDValue Op0 = Op.getOperand(0);                     // Op0 = the scalar
  DebugLoc dl = Op.getDebugLoc();

  if (Op0.getNode()->getOpcode() == ISD::Constant) {
    // For a constant, build the appropriate constant vector, which will
    // eventually simplify to a vector register load.

    ConstantSDNode *CN = cast<ConstantSDNode>(Op0.getNode());
    SmallVector<SDValue, 16> ConstVecValues;
    EVT VT;
    size_t n_copies;

    // Create a constant vector:
    switch (Op.getValueType().getSimpleVT().SimpleTy) {
    default: llvm_unreachable("Unexpected constant value type in "
                              "LowerSCALAR_TO_VECTOR");
    case MVT::v16i8: n_copies = 16; VT = MVT::i8; break;
    case MVT::v8i16: n_copies = 8; VT = MVT::i16; break;
    case MVT::v4i32: n_copies = 4; VT = MVT::i32; break;
    case MVT::v4f32: n_copies = 4; VT = MVT::f32; break;
    case MVT::v2i64: n_copies = 2; VT = MVT::i64; break;
    case MVT::v2f64: n_copies = 2; VT = MVT::f64; break;
    }

    SDValue CValue = DAG.getConstant(CN->getZExtValue(), VT);
    for (size_t j = 0; j < n_copies; ++j)
      ConstVecValues.push_back(CValue);

    return DAG.getNode(ISD::BUILD_VECTOR, dl, Op.getValueType(),
                       &ConstVecValues[0], ConstVecValues.size());
  } else {
    // Otherwise, copy the value from one register to another:
    switch (Op0.getValueType().getSimpleVT().SimpleTy) {
    default: llvm_unreachable("Unexpected value type in LowerSCALAR_TO_VECTOR");
    case MVT::i8:
    case MVT::i16:
    case MVT::i32:
    case MVT::i64:
    case MVT::f32:
    case MVT::f64:
      return DAG.getNode(SPUISD::PREFSLOT2VEC, dl, Op.getValueType(), Op0, Op0);
    }
  }

  return SDValue();
}

static SDValue LowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) {
  EVT VT = Op.getValueType();
  SDValue N = Op.getOperand(0);
  SDValue Elt = Op.getOperand(1);
  DebugLoc dl = Op.getDebugLoc();
  SDValue retval;

  if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Elt)) {
    // Constant argument:
    int EltNo = (int) C->getZExtValue();

    // sanity checks:
    if (VT == MVT::i8 && EltNo >= 16)
      llvm_unreachable("SPU LowerEXTRACT_VECTOR_ELT: i8 extraction slot > 15");
    else if (VT == MVT::i16 && EltNo >= 8)
      llvm_unreachable("SPU LowerEXTRACT_VECTOR_ELT: i16 extraction slot > 7");
    else if (VT == MVT::i32 && EltNo >= 4)
      llvm_unreachable("SPU LowerEXTRACT_VECTOR_ELT: i32 extraction slot > 4");
    else if (VT == MVT::i64 && EltNo >= 2)
      llvm_unreachable("SPU LowerEXTRACT_VECTOR_ELT: i64 extraction slot > 2");

    if (EltNo == 0 && (VT == MVT::i32 || VT == MVT::i64)) {
      // i32 and i64: Element 0 is the preferred slot
      return DAG.getNode(SPUISD::VEC2PREFSLOT, dl, VT, N);
    }

    // Need to generate shuffle mask and extract:
    int prefslot_begin = -1, prefslot_end = -1;
    int elt_byte = EltNo * VT.getSizeInBits() / 8;

    switch (VT.getSimpleVT().SimpleTy) {
    default:
      assert(false && "Invalid value type!");
    case MVT::i8: {
      prefslot_begin = prefslot_end = 3;
      break;
    }
    case MVT::i16: {
      prefslot_begin = 2; prefslot_end = 3;
      break;
    }
    case MVT::i32:
    case MVT::f32: {
      prefslot_begin = 0; prefslot_end = 3;
      break;
    }
    case MVT::i64:
    case MVT::f64: {
      prefslot_begin = 0; prefslot_end = 7;
      break;
    }
    }

    assert(prefslot_begin != -1 && prefslot_end != -1 &&
           "LowerEXTRACT_VECTOR_ELT: preferred slots uninitialized");

    unsigned int ShufBytes[16] = {
      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
    };
    for (int i = 0; i < 16; ++i) {
      // zero fill uppper part of preferred slot, don't care about the
      // other slots:
      unsigned int mask_val;
      if (i <= prefslot_end) {
        mask_val =
          ((i < prefslot_begin)
           ? 0x80
           : elt_byte + (i - prefslot_begin));

        ShufBytes[i] = mask_val;
      } else
        ShufBytes[i] = ShufBytes[i % (prefslot_end + 1)];
    }

    SDValue ShufMask[4];
    for (unsigned i = 0; i < sizeof(ShufMask)/sizeof(ShufMask[0]); ++i) {
      unsigned bidx = i * 4;
      unsigned int bits = ((ShufBytes[bidx] << 24) |
                           (ShufBytes[bidx+1] << 16) |
                           (ShufBytes[bidx+2] << 8) |
                           ShufBytes[bidx+3]);
      ShufMask[i] = DAG.getConstant(bits, MVT::i32);
    }

    SDValue ShufMaskVec =
      DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
                  &ShufMask[0], sizeof(ShufMask)/sizeof(ShufMask[0]));

    retval = DAG.getNode(SPUISD::VEC2PREFSLOT, dl, VT,
                         DAG.getNode(SPUISD::SHUFB, dl, N.getValueType(),
                                     N, N, ShufMaskVec));
  } else {
    // Variable index: Rotate the requested element into slot 0, then replicate
    // slot 0 across the vector
    EVT VecVT = N.getValueType();
    if (!VecVT.isSimple() || !VecVT.isVector()) {
      report_fatal_error("LowerEXTRACT_VECTOR_ELT: Must have a simple, 128-bit"
                        "vector type!");
    }

    // Make life easier by making sure the index is zero-extended to i32
    if (Elt.getValueType() != MVT::i32)
      Elt = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, Elt);

    // Scale the index to a bit/byte shift quantity
    APInt scaleFactor =
            APInt(32, uint64_t(16 / N.getValueType().getVectorNumElements()), false);
    unsigned scaleShift = scaleFactor.logBase2();
    SDValue vecShift;

    if (scaleShift > 0) {
      // Scale the shift factor:
      Elt = DAG.getNode(ISD::SHL, dl, MVT::i32, Elt,
                        DAG.getConstant(scaleShift, MVT::i32));
    }

    vecShift = DAG.getNode(SPUISD::SHLQUAD_L_BYTES, dl, VecVT, N, Elt);

    // Replicate the bytes starting at byte 0 across the entire vector (for
    // consistency with the notion of a unified register set)
    SDValue replicate;

    switch (VT.getSimpleVT().SimpleTy) {
    default:
      report_fatal_error("LowerEXTRACT_VECTOR_ELT(varable): Unhandled vector"
                        "type");
      /*NOTREACHED*/
    case MVT::i8: {
      SDValue factor = DAG.getConstant(0x00000000, MVT::i32);
      replicate = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
                              factor, factor, factor, factor);
      break;
    }
    case MVT::i16: {
      SDValue factor = DAG.getConstant(0x00010001, MVT::i32);
      replicate = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
                              factor, factor, factor, factor);
      break;
    }
    case MVT::i32:
    case MVT::f32: {
      SDValue factor = DAG.getConstant(0x00010203, MVT::i32);
      replicate = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
                              factor, factor, factor, factor);
      break;
    }
    case MVT::i64:
    case MVT::f64: {
      SDValue loFactor = DAG.getConstant(0x00010203, MVT::i32);
      SDValue hiFactor = DAG.getConstant(0x04050607, MVT::i32);
      replicate = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
                              loFactor, hiFactor, loFactor, hiFactor);
      break;
    }
    }

    retval = DAG.getNode(SPUISD::VEC2PREFSLOT, dl, VT,
                         DAG.getNode(SPUISD::SHUFB, dl, VecVT,
                                     vecShift, vecShift, replicate));
  }

  return retval;
}

static SDValue LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) {
  SDValue VecOp = Op.getOperand(0);
  SDValue ValOp = Op.getOperand(1);
  SDValue IdxOp = Op.getOperand(2);
  DebugLoc dl = Op.getDebugLoc();
  EVT VT = Op.getValueType();
  EVT eltVT = ValOp.getValueType();

  // use 0 when the lane to insert to is 'undef'
  int64_t Offset=0;
  if (IdxOp.getOpcode() != ISD::UNDEF) {
    ConstantSDNode *CN = cast<ConstantSDNode>(IdxOp);
    assert(CN != 0 && "LowerINSERT_VECTOR_ELT: Index is not constant!");
    Offset = (CN->getSExtValue()) * eltVT.getSizeInBits()/8;
  }

  EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
  // Use $sp ($1) because it's always 16-byte aligned and it's available:
  SDValue Pointer = DAG.getNode(SPUISD::IndirectAddr, dl, PtrVT,
                                DAG.getRegister(SPU::R1, PtrVT),
                                DAG.getConstant(Offset, PtrVT));
  // widen the mask when dealing with half vectors
  EVT maskVT = EVT::getVectorVT(*(DAG.getContext()), VT.getVectorElementType(), 
                                128/ VT.getVectorElementType().getSizeInBits());
  SDValue ShufMask = DAG.getNode(SPUISD::SHUFFLE_MASK, dl, maskVT, Pointer);

  SDValue result =
    DAG.getNode(SPUISD::SHUFB, dl, VT,
                DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, ValOp),
                VecOp,
                DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v4i32, ShufMask));

  return result;
}

static SDValue LowerI8Math(SDValue Op, SelectionDAG &DAG, unsigned Opc,
                           const TargetLowering &TLI)
{
  SDValue N0 = Op.getOperand(0);      // Everything has at least one operand
  DebugLoc dl = Op.getDebugLoc();
  EVT ShiftVT = TLI.getShiftAmountTy();

  assert(Op.getValueType() == MVT::i8);
  switch (Opc) {
  default:
    llvm_unreachable("Unhandled i8 math operator");
    /*NOTREACHED*/
    break;
  case ISD::ADD: {
    // 8-bit addition: Promote the arguments up to 16-bits and truncate
    // the result:
    SDValue N1 = Op.getOperand(1);
    N0 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N0);
    N1 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N1);
    return DAG.getNode(ISD::TRUNCATE, dl, MVT::i8,
                       DAG.getNode(Opc, dl, MVT::i16, N0, N1));

  }

  case ISD::SUB: {
    // 8-bit subtraction: Promote the arguments up to 16-bits and truncate
    // the result:
    SDValue N1 = Op.getOperand(1);
    N0 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N0);
    N1 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N1);
    return DAG.getNode(ISD::TRUNCATE, dl, MVT::i8,
                       DAG.getNode(Opc, dl, MVT::i16, N0, N1));
  }
  case ISD::ROTR:
  case ISD::ROTL: {
    SDValue N1 = Op.getOperand(1);
    EVT N1VT = N1.getValueType();

    N0 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i16, N0);
    if (!N1VT.bitsEq(ShiftVT)) {
      unsigned N1Opc = N1.getValueType().bitsLT(ShiftVT)
                       ? ISD::ZERO_EXTEND
                       : ISD::TRUNCATE;
      N1 = DAG.getNode(N1Opc, dl, ShiftVT, N1);
    }

    // Replicate lower 8-bits into upper 8:
    SDValue ExpandArg =
      DAG.getNode(ISD::OR, dl, MVT::i16, N0,
                  DAG.getNode(ISD::SHL, dl, MVT::i16,
                              N0, DAG.getConstant(8, MVT::i32)));

    // Truncate back down to i8
    return DAG.getNode(ISD::TRUNCATE, dl, MVT::i8,
                       DAG.getNode(Opc, dl, MVT::i16, ExpandArg, N1));
  }
  case ISD::SRL:
  case ISD::SHL: {
    SDValue N1 = Op.getOperand(1);
    EVT N1VT = N1.getValueType();

    N0 = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i16, N0);
    if (!N1VT.bitsEq(ShiftVT)) {
      unsigned N1Opc = ISD::ZERO_EXTEND;

      if (N1.getValueType().bitsGT(ShiftVT))
        N1Opc = ISD::TRUNCATE;

      N1 = DAG.getNode(N1Opc, dl, ShiftVT, N1);
    }

    return DAG.getNode(ISD::TRUNCATE, dl, MVT::i8,
                       DAG.getNode(Opc, dl, MVT::i16, N0, N1));
  }
  case ISD::SRA: {
    SDValue N1 = Op.getOperand(1);
    EVT N1VT = N1.getValueType();

    N0 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N0);
    if (!N1VT.bitsEq(ShiftVT)) {
      unsigned N1Opc = ISD::SIGN_EXTEND;

      if (N1VT.bitsGT(ShiftVT))
        N1Opc = ISD::TRUNCATE;
      N1 = DAG.getNode(N1Opc, dl, ShiftVT, N1);
    }

    return DAG.getNode(ISD::TRUNCATE, dl, MVT::i8,
                       DAG.getNode(Opc, dl, MVT::i16, N0, N1));
  }
  case ISD::MUL: {
    SDValue N1 = Op.getOperand(1);

    N0 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N0);
    N1 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i16, N1);
    return DAG.getNode(ISD::TRUNCATE, dl, MVT::i8,
                       DAG.getNode(Opc, dl, MVT::i16, N0, N1));
    break;
  }
  }

  return SDValue();
}

//! Lower byte immediate operations for v16i8 vectors:
static SDValue
LowerByteImmed(SDValue Op, SelectionDAG &DAG) {
  SDValue ConstVec;
  SDValue Arg;
  EVT VT = Op.getValueType();
  DebugLoc dl = Op.getDebugLoc();

  ConstVec = Op.getOperand(0);
  Arg = Op.getOperand(1);
  if (ConstVec.getNode()->getOpcode() != ISD::BUILD_VECTOR) {
    if (ConstVec.getNode()->getOpcode() == ISD::BIT_CONVERT) {
      ConstVec = ConstVec.getOperand(0);
    } else {
      ConstVec = Op.getOperand(1);
      Arg = Op.getOperand(0);
      if (ConstVec.getNode()->getOpcode() == ISD::BIT_CONVERT) {
        ConstVec = ConstVec.getOperand(0);
      }
    }
  }

  if (ConstVec.getNode()->getOpcode() == ISD::BUILD_VECTOR) {
    BuildVectorSDNode *BCN = dyn_cast<BuildVectorSDNode>(ConstVec.getNode());
    assert(BCN != 0 && "Expected BuildVectorSDNode in SPU LowerByteImmed");

    APInt APSplatBits, APSplatUndef;
    unsigned SplatBitSize;
    bool HasAnyUndefs;
    unsigned minSplatBits = VT.getVectorElementType().getSizeInBits();

    if (BCN->isConstantSplat(APSplatBits, APSplatUndef, SplatBitSize,
                              HasAnyUndefs, minSplatBits)
        && minSplatBits <= SplatBitSize) {
      uint64_t SplatBits = APSplatBits.getZExtValue();
      SDValue tc = DAG.getTargetConstant(SplatBits & 0xff, MVT::i8);

      SmallVector<SDValue, 16> tcVec;
      tcVec.assign(16, tc);
      return DAG.getNode(Op.getNode()->getOpcode(), dl, VT, Arg,
                         DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &tcVec[0], tcVec.size()));
    }
  }

  // These operations (AND, OR, XOR) are legal, they just couldn't be custom
  // lowered.  Return the operation, rather than a null SDValue.
  return Op;
}

//! Custom lowering for CTPOP (count population)
/*!
  Custom lowering code that counts the number ones in the input
  operand. SPU has such an instruction, but it counts the number of
  ones per byte, which then have to be accumulated.
*/
static SDValue LowerCTPOP(SDValue Op, SelectionDAG &DAG) {
  EVT VT = Op.getValueType();
  EVT vecVT = EVT::getVectorVT(*DAG.getContext(), 
                               VT, (128 / VT.getSizeInBits()));
  DebugLoc dl = Op.getDebugLoc();

  switch (VT.getSimpleVT().SimpleTy) {
  default:
    assert(false && "Invalid value type!");
  case MVT::i8: {
    SDValue N = Op.getOperand(0);
    SDValue Elt0 = DAG.getConstant(0, MVT::i32);

    SDValue Promote = DAG.getNode(SPUISD::PREFSLOT2VEC, dl, vecVT, N, N);
    SDValue CNTB = DAG.getNode(SPUISD::CNTB, dl, vecVT, Promote);

    return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i8, CNTB, Elt0);
  }

  case MVT::i16: {
    MachineFunction &MF = DAG.getMachineFunction();
    MachineRegisterInfo &RegInfo = MF.getRegInfo();

    unsigned CNTB_reg = RegInfo.createVirtualRegister(&SPU::R16CRegClass);

    SDValue N = Op.getOperand(0);
    SDValue Elt0 = DAG.getConstant(0, MVT::i16);
    SDValue Mask0 = DAG.getConstant(0x0f, MVT::i16);
    SDValue Shift1 = DAG.getConstant(8, MVT::i32);

    SDValue Promote = DAG.getNode(SPUISD::PREFSLOT2VEC, dl, vecVT, N, N);
    SDValue CNTB = DAG.getNode(SPUISD::CNTB, dl, vecVT, Promote);

    // CNTB_result becomes the chain to which all of the virtual registers
    // CNTB_reg, SUM1_reg become associated:
    SDValue CNTB_result =
      DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i16, CNTB, Elt0);

    SDValue CNTB_rescopy =
      DAG.getCopyToReg(CNTB_result, dl, CNTB_reg, CNTB_result);

    SDValue Tmp1 = DAG.getCopyFromReg(CNTB_rescopy, dl, CNTB_reg, MVT::i16);

    return DAG.getNode(ISD::AND, dl, MVT::i16,
                       DAG.getNode(ISD::ADD, dl, MVT::i16,
                                   DAG.getNode(ISD::SRL, dl, MVT::i16,
                                               Tmp1, Shift1),
                                   Tmp1),
                       Mask0);
  }

  case MVT::i32: {
    MachineFunction &MF = DAG.getMachineFunction();
    MachineRegisterInfo &RegInfo = MF.getRegInfo();

    unsigned CNTB_reg = RegInfo.createVirtualRegister(&SPU::R32CRegClass);
    unsigned SUM1_reg = RegInfo.createVirtualRegister(&SPU::R32CRegClass);

    SDValue N = Op.getOperand(0);
    SDValue Elt0 = DAG.getConstant(0, MVT::i32);
    SDValue Mask0 = DAG.getConstant(0xff, MVT::i32);
    SDValue Shift1 = DAG.getConstant(16, MVT::i32);
    SDValue Shift2 = DAG.getConstant(8, MVT::i32);

    SDValue Promote = DAG.getNode(SPUISD::PREFSLOT2VEC, dl, vecVT, N, N);
    SDValue CNTB = DAG.getNode(SPUISD::CNTB, dl, vecVT, Promote);

    // CNTB_result becomes the chain to which all of the virtual registers
    // CNTB_reg, SUM1_reg become associated:
    SDValue CNTB_result =
      DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i32, CNTB, Elt0);

    SDValue CNTB_rescopy =
      DAG.getCopyToReg(CNTB_result, dl, CNTB_reg, CNTB_result);

    SDValue Comp1 =
      DAG.getNode(ISD::SRL, dl, MVT::i32,
                  DAG.getCopyFromReg(CNTB_rescopy, dl, CNTB_reg, MVT::i32),
                  Shift1);

    SDValue Sum1 =
      DAG.getNode(ISD::ADD, dl, MVT::i32, Comp1,
                  DAG.getCopyFromReg(CNTB_rescopy, dl, CNTB_reg, MVT::i32));

    SDValue Sum1_rescopy =
      DAG.getCopyToReg(CNTB_result, dl, SUM1_reg, Sum1);

    SDValue Comp2 =
      DAG.getNode(ISD::SRL, dl, MVT::i32,
                  DAG.getCopyFromReg(Sum1_rescopy, dl, SUM1_reg, MVT::i32),
                  Shift2);
    SDValue Sum2 =
      DAG.getNode(ISD::ADD, dl, MVT::i32, Comp2,
                  DAG.getCopyFromReg(Sum1_rescopy, dl, SUM1_reg, MVT::i32));

    return DAG.getNode(ISD::AND, dl, MVT::i32, Sum2, Mask0);
  }

  case MVT::i64:
    break;
  }

  return SDValue();
}

//! Lower ISD::FP_TO_SINT, ISD::FP_TO_UINT for i32
/*!
 f32->i32 passes through unchanged, whereas f64->i32 expands to a libcall.
 All conversions to i64 are expanded to a libcall.
 */
static SDValue LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG,
                              const SPUTargetLowering &TLI) {
  EVT OpVT = Op.getValueType();
  SDValue Op0 = Op.getOperand(0);
  EVT Op0VT = Op0.getValueType();

  if ((OpVT == MVT::i32 && Op0VT == MVT::f64)
      || OpVT == MVT::i64) {
    // Convert f32 / f64 to i32 / i64 via libcall.
    RTLIB::Libcall LC =
            (Op.getOpcode() == ISD::FP_TO_SINT)
             ? RTLIB::getFPTOSINT(Op0VT, OpVT)
             : RTLIB::getFPTOUINT(Op0VT, OpVT);
    assert(LC != RTLIB::UNKNOWN_LIBCALL && "Unexpectd fp-to-int conversion!");
    SDValue Dummy;
    return ExpandLibCall(LC, Op, DAG, false, Dummy, TLI);
  }

  return Op;
}

//! Lower ISD::SINT_TO_FP, ISD::UINT_TO_FP for i32
/*!
 i32->f32 passes through unchanged, whereas i32->f64 is expanded to a libcall.
 All conversions from i64 are expanded to a libcall.
 */
static SDValue LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG,
                              const SPUTargetLowering &TLI) {
  EVT OpVT = Op.getValueType();
  SDValue Op0 = Op.getOperand(0);
  EVT Op0VT = Op0.getValueType();

  if ((OpVT == MVT::f64 && Op0VT == MVT::i32)
      || Op0VT == MVT::i64) {
    // Convert i32, i64 to f64 via libcall:
    RTLIB::Libcall LC =
            (Op.getOpcode() == ISD::SINT_TO_FP)
             ? RTLIB::getSINTTOFP(Op0VT, OpVT)
             : RTLIB::getUINTTOFP(Op0VT, OpVT);
    assert(LC != RTLIB::UNKNOWN_LIBCALL && "Unexpectd int-to-fp conversion!");
    SDValue Dummy;
    return ExpandLibCall(LC, Op, DAG, false, Dummy, TLI);
  }

  return Op;
}

//! Lower ISD::SETCC
/*!
 This handles MVT::f64 (double floating point) condition lowering
 */
static SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG,
                          const TargetLowering &TLI) {
  CondCodeSDNode *CC = dyn_cast<CondCodeSDNode>(Op.getOperand(2));
  DebugLoc dl = Op.getDebugLoc();
  assert(CC != 0 && "LowerSETCC: CondCodeSDNode should not be null here!\n");

  SDValue lhs = Op.getOperand(0);
  SDValue rhs = Op.getOperand(1);
  EVT lhsVT = lhs.getValueType();
  assert(lhsVT == MVT::f64 && "LowerSETCC: type other than MVT::64\n");

  EVT ccResultVT = TLI.getSetCCResultType(lhs.getValueType());
  APInt ccResultOnes = APInt::getAllOnesValue(ccResultVT.getSizeInBits());
  EVT IntVT(MVT::i64);

  // Take advantage of the fact that (truncate (sra arg, 32)) is efficiently
  // selected to a NOP:
  SDValue i64lhs = DAG.getNode(ISD::BIT_CONVERT, dl, IntVT, lhs);
  SDValue lhsHi32 =
          DAG.getNode(ISD::TRUNCATE, dl, MVT::i32,
                      DAG.getNode(ISD::SRL, dl, IntVT,
                                  i64lhs, DAG.getConstant(32, MVT::i32)));
  SDValue lhsHi32abs =
          DAG.getNode(ISD::AND, dl, MVT::i32,
                      lhsHi32, DAG.getConstant(0x7fffffff, MVT::i32));
  SDValue lhsLo32 =
          DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, i64lhs);

  // SETO and SETUO only use the lhs operand:
  if (CC->get() == ISD::SETO) {
    // Evaluates to true if Op0 is not [SQ]NaN - lowers to the inverse of
    // SETUO
    APInt ccResultAllOnes = APInt::getAllOnesValue(ccResultVT.getSizeInBits());
    return DAG.getNode(ISD::XOR, dl, ccResultVT,
                       DAG.getSetCC(dl, ccResultVT,
                                    lhs, DAG.getConstantFP(0.0, lhsVT),
                                    ISD::SETUO),
                       DAG.getConstant(ccResultAllOnes, ccResultVT));
  } else if (CC->get() == ISD::SETUO) {
    // Evaluates to true if Op0 is [SQ]NaN
    return DAG.getNode(ISD::AND, dl, ccResultVT,
                       DAG.getSetCC(dl, ccResultVT,
                                    lhsHi32abs,
                                    DAG.getConstant(0x7ff00000, MVT::i32),
                                    ISD::SETGE),
                       DAG.getSetCC(dl, ccResultVT,
                                    lhsLo32,
                                    DAG.getConstant(0, MVT::i32),
                                    ISD::SETGT));
  }

  SDValue i64rhs = DAG.getNode(ISD::BIT_CONVERT, dl, IntVT, rhs);
  SDValue rhsHi32 =
          DAG.getNode(ISD::TRUNCATE, dl, MVT::i32,
                      DAG.getNode(ISD::SRL, dl, IntVT,
                                  i64rhs, DAG.getConstant(32, MVT::i32)));

  // If a value is negative, subtract from the sign magnitude constant:
  SDValue signMag2TC = DAG.getConstant(0x8000000000000000ULL, IntVT);

  // Convert the sign-magnitude representation into 2's complement:
  SDValue lhsSelectMask = DAG.getNode(ISD::SRA, dl, ccResultVT,
                                      lhsHi32, DAG.getConstant(31, MVT::i32));
  SDValue lhsSignMag2TC = DAG.getNode(ISD::SUB, dl, IntVT, signMag2TC, i64lhs);
  SDValue lhsSelect =
          DAG.getNode(ISD::SELECT, dl, IntVT,
                      lhsSelectMask, lhsSignMag2TC, i64lhs);

  SDValue rhsSelectMask = DAG.getNode(ISD::SRA, dl, ccResultVT,
                                      rhsHi32, DAG.getConstant(31, MVT::i32));
  SDValue rhsSignMag2TC = DAG.getNode(ISD::SUB, dl, IntVT, signMag2TC, i64rhs);
  SDValue rhsSelect =
          DAG.getNode(ISD::SELECT, dl, IntVT,
                      rhsSelectMask, rhsSignMag2TC, i64rhs);

  unsigned compareOp;

  switch (CC->get()) {
  case ISD::SETOEQ:
  case ISD::SETUEQ:
    compareOp = ISD::SETEQ; break;
  case ISD::SETOGT:
  case ISD::SETUGT:
    compareOp = ISD::SETGT; break;
  case ISD::SETOGE:
  case ISD::SETUGE:
    compareOp = ISD::SETGE; break;
  case ISD::SETOLT:
  case ISD::SETULT:
    compareOp = ISD::SETLT; break;
  case ISD::SETOLE:
  case ISD::SETULE:
    compareOp = ISD::SETLE; break;
  case ISD::SETUNE:
  case ISD::SETONE:
    compareOp = ISD::SETNE; break;
  default:
    report_fatal_error("CellSPU ISel Select: unimplemented f64 condition");
  }

  SDValue result =
          DAG.getSetCC(dl, ccResultVT, lhsSelect, rhsSelect,
                       (ISD::CondCode) compareOp);

  if ((CC->get() & 0x8) == 0) {
    // Ordered comparison:
    SDValue lhsNaN = DAG.getSetCC(dl, ccResultVT,
                                  lhs, DAG.getConstantFP(0.0, MVT::f64),
                                  ISD::SETO);
    SDValue rhsNaN = DAG.getSetCC(dl, ccResultVT,
                                  rhs, DAG.getConstantFP(0.0, MVT::f64),
                                  ISD::SETO);
    SDValue ordered = DAG.getNode(ISD::AND, dl, ccResultVT, lhsNaN, rhsNaN);

    result = DAG.getNode(ISD::AND, dl, ccResultVT, ordered, result);
  }

  return result;
}

//! Lower ISD::SELECT_CC
/*!
  ISD::SELECT_CC can (generally) be implemented directly on the SPU using the
  SELB instruction.

  \note Need to revisit this in the future: if the code path through the true
  and false value computations is longer than the latency of a branch (6
  cycles), then it would be more advantageous to branch and insert a new basic
  block and branch on the condition. However, this code does not make that
  assumption, given the simplisitc uses so far.
 */

static SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG,
                              const TargetLowering &TLI) {
  EVT VT = Op.getValueType();
  SDValue lhs = Op.getOperand(0);
  SDValue rhs = Op.getOperand(1);
  SDValue trueval = Op.getOperand(2);
  SDValue falseval = Op.getOperand(3);
  SDValue condition = Op.getOperand(4);
  DebugLoc dl = Op.getDebugLoc();

  // NOTE: SELB's arguments: $rA, $rB, $mask
  //
  // SELB selects bits from $rA where bits in $mask are 0, bits from $rB
  // where bits in $mask are 1. CCond will be inverted, having 1s where the
  // condition was true and 0s where the condition was false. Hence, the
  // arguments to SELB get reversed.

  // Note: Really should be ISD::SELECT instead of SPUISD::SELB, but LLVM's
  // legalizer insists on combining SETCC/SELECT into SELECT_CC, so we end up
  // with another "cannot select select_cc" assert:

  SDValue compare = DAG.getNode(ISD::SETCC, dl,
                                TLI.getSetCCResultType(Op.getValueType()),
                                lhs, rhs, condition);
  return DAG.getNode(SPUISD::SELB, dl, VT, falseval, trueval, compare);
}

//! Custom lower ISD::TRUNCATE
static SDValue LowerTRUNCATE(SDValue Op, SelectionDAG &DAG)
{
  // Type to truncate to
  EVT VT = Op.getValueType();
  MVT simpleVT = VT.getSimpleVT();
  EVT VecVT = EVT::getVectorVT(*DAG.getContext(), 
                               VT, (128 / VT.getSizeInBits()));
  DebugLoc dl = Op.getDebugLoc();

  // Type to truncate from
  SDValue Op0 = Op.getOperand(0);
  EVT Op0VT = Op0.getValueType();

  if (Op0VT.getSimpleVT() == MVT::i128 && simpleVT == MVT::i64) {
    // Create shuffle mask, least significant doubleword of quadword
    unsigned maskHigh = 0x08090a0b;
    unsigned maskLow = 0x0c0d0e0f;
    // Use a shuffle to perform the truncation
    SDValue shufMask = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
                                   DAG.getConstant(maskHigh, MVT::i32),
                                   DAG.getConstant(maskLow, MVT::i32),
                                   DAG.getConstant(maskHigh, MVT::i32),
                                   DAG.getConstant(maskLow, MVT::i32));

    SDValue truncShuffle = DAG.getNode(SPUISD::SHUFB, dl, VecVT,
                                       Op0, Op0, shufMask);

    return DAG.getNode(SPUISD::VEC2PREFSLOT, dl, VT, truncShuffle);
  }

  return SDValue();             // Leave the truncate unmolested
}

/*!
 * Emit the instruction sequence for i64/i32 -> i128 sign extend. The basic
 * algorithm is to duplicate the sign bit using rotmai to generate at
 * least one byte full of sign bits. Then propagate the "sign-byte" into
 * the leftmost words and the i64/i32 into the rightmost words using shufb.
 *
 * @param Op The sext operand
 * @param DAG The current DAG
 * @return The SDValue with the entire instruction sequence
 */
static SDValue LowerSIGN_EXTEND(SDValue Op, SelectionDAG &DAG)
{
  DebugLoc dl = Op.getDebugLoc();

  // Type to extend to
  MVT OpVT = Op.getValueType().getSimpleVT();

  // Type to extend from
  SDValue Op0 = Op.getOperand(0);
  MVT Op0VT = Op0.getValueType().getSimpleVT();

  // The type to extend to needs to be a i128 and
  // the type to extend from needs to be i64 or i32.
  assert((OpVT == MVT::i128 && (Op0VT == MVT::i64 || Op0VT == MVT::i32)) &&
          "LowerSIGN_EXTEND: input and/or output operand have wrong size");

  // Create shuffle mask
  unsigned mask1 = 0x10101010; // byte 0 - 3 and 4 - 7
  unsigned mask2 = Op0VT == MVT::i64 ? 0x00010203 : 0x10101010; // byte  8 - 11
  unsigned mask3 = Op0VT == MVT::i64 ? 0x04050607 : 0x00010203; // byte 12 - 15
  SDValue shufMask = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32,
                                 DAG.getConstant(mask1, MVT::i32),
                                 DAG.getConstant(mask1, MVT::i32),
                                 DAG.getConstant(mask2, MVT::i32),
                                 DAG.getConstant(mask3, MVT::i32));

  // Word wise arithmetic right shift to generate at least one byte
  // that contains sign bits.
  MVT mvt = Op0VT == MVT::i64 ? MVT::v2i64 : MVT::v4i32;
  SDValue sraVal = DAG.getNode(ISD::SRA,
                 dl,
                 mvt,
                 DAG.getNode(SPUISD::PREFSLOT2VEC, dl, mvt, Op0, Op0),
                 DAG.getConstant(31, MVT::i32));

  // Shuffle bytes - Copy the sign bits into the upper 64 bits
  // and the input value into the lower 64 bits.
  SDValue extShuffle = DAG.getNode(SPUISD::SHUFB, dl, mvt,
      DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i128, Op0), sraVal, shufMask);

  return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i128, extShuffle);
}

//! Custom (target-specific) lowering entry point
/*!
  This is where LLVM's DAG selection process calls to do target-specific
  lowering of nodes.
 */
SDValue
SPUTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const
{
  unsigned Opc = (unsigned) Op.getOpcode();
  EVT VT = Op.getValueType();

  switch (Opc) {
  default: {
#ifndef NDEBUG
    errs() << "SPUTargetLowering::LowerOperation(): need to lower this!\n";
    errs() << "Op.getOpcode() = " << Opc << "\n";
    errs() << "*Op.getNode():\n";
    Op.getNode()->dump();
#endif
    llvm_unreachable(0);
  }
  case ISD::LOAD:
  case ISD::EXTLOAD:
  case ISD::SEXTLOAD:
  case ISD::ZEXTLOAD:
    return LowerLOAD(Op, DAG, SPUTM.getSubtargetImpl());
  case ISD::STORE:
    return LowerSTORE(Op, DAG, SPUTM.getSubtargetImpl());
  case ISD::ConstantPool:
    return LowerConstantPool(Op, DAG, SPUTM.getSubtargetImpl());
  case ISD::GlobalAddress:
    return LowerGlobalAddress(Op, DAG, SPUTM.getSubtargetImpl());
  case ISD::JumpTable:
    return LowerJumpTable(Op, DAG, SPUTM.getSubtargetImpl());
  case ISD::ConstantFP:
    return LowerConstantFP(Op, DAG);

  // i8, i64 math ops:
  case ISD::ADD:
  case ISD::SUB:
  case ISD::ROTR:
  case ISD::ROTL:
  case ISD::SRL:
  case ISD::SHL:
  case ISD::SRA: {
    if (VT == MVT::i8)
      return LowerI8Math(Op, DAG, Opc, *this);
    break;
  }

  case ISD::FP_TO_SINT:
  case ISD::FP_TO_UINT:
    return LowerFP_TO_INT(Op, DAG, *this);

  case ISD::SINT_TO_FP:
  case ISD::UINT_TO_FP:
    return LowerINT_TO_FP(Op, DAG, *this);

  // Vector-related lowering.
  case ISD::BUILD_VECTOR:
    return LowerBUILD_VECTOR(Op, DAG);
  case ISD::SCALAR_TO_VECTOR:
    return LowerSCALAR_TO_VECTOR(Op, DAG);
  case ISD::VECTOR_SHUFFLE:
    return LowerVECTOR_SHUFFLE(Op, DAG);
  case ISD::EXTRACT_VECTOR_ELT:
    return LowerEXTRACT_VECTOR_ELT(Op, DAG);
  case ISD::INSERT_VECTOR_ELT:
    return LowerINSERT_VECTOR_ELT(Op, DAG);

  // Look for ANDBI, ORBI and XORBI opportunities and lower appropriately:
  case ISD::AND:
  case ISD::OR:
  case ISD::XOR:
    return LowerByteImmed(Op, DAG);

  // Vector and i8 multiply:
  case ISD::MUL:
    if (VT == MVT::i8)
      return LowerI8Math(Op, DAG, Opc, *this);

  case ISD::CTPOP:
    return LowerCTPOP(Op, DAG);

  case ISD::SELECT_CC:
    return LowerSELECT_CC(Op, DAG, *this);

  case ISD::SETCC:
    return LowerSETCC(Op, DAG, *this);

  case ISD::TRUNCATE:
    return LowerTRUNCATE(Op, DAG);

  case ISD::SIGN_EXTEND:
    return LowerSIGN_EXTEND(Op, DAG);
  }

  return SDValue();
}

void SPUTargetLowering::ReplaceNodeResults(SDNode *N,
                                           SmallVectorImpl<SDValue>&Results,
                                           SelectionDAG &DAG) const
{
#if 0
  unsigned Opc = (unsigned) N->getOpcode();
  EVT OpVT = N->getValueType(0);

  switch (Opc) {
  default: {
    errs() << "SPUTargetLowering::ReplaceNodeResults(): need to fix this!\n";
    errs() << "Op.getOpcode() = " << Opc << "\n";
    errs() << "*Op.getNode():\n";
    N->dump();
    abort();
    /*NOTREACHED*/
  }
  }
#endif

  /* Otherwise, return unchanged */
}

//===----------------------------------------------------------------------===//
// Target Optimization Hooks
//===----------------------------------------------------------------------===//

SDValue
SPUTargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const
{
#if 0
  TargetMachine &TM = getTargetMachine();
#endif
  const SPUSubtarget *ST = SPUTM.getSubtargetImpl();
  SelectionDAG &DAG = DCI.DAG;
  SDValue Op0 = N->getOperand(0);       // everything has at least one operand
  EVT NodeVT = N->getValueType(0);      // The node's value type
  EVT Op0VT = Op0.getValueType();       // The first operand's result
  SDValue Result;                       // Initially, empty result
  DebugLoc dl = N->getDebugLoc();

  switch (N->getOpcode()) {
  default: break;
  case ISD::ADD: {
    SDValue Op1 = N->getOperand(1);

    if (Op0.getOpcode() == SPUISD::IndirectAddr
        || Op1.getOpcode() == SPUISD::IndirectAddr) {
      // Normalize the operands to reduce repeated code
      SDValue IndirectArg = Op0, AddArg = Op1;

      if (Op1.getOpcode() == SPUISD::IndirectAddr) {
        IndirectArg = Op1;
        AddArg = Op0;
      }

      if (isa<ConstantSDNode>(AddArg)) {
        ConstantSDNode *CN0 = cast<ConstantSDNode > (AddArg);
        SDValue IndOp1 = IndirectArg.getOperand(1);

        if (CN0->isNullValue()) {
          // (add (SPUindirect <arg>, <arg>), 0) ->
          // (SPUindirect <arg>, <arg>)

#if !defined(NDEBUG)
          if (DebugFlag && isCurrentDebugType(DEBUG_TYPE)) {
            errs() << "\n"
                 << "Replace: (add (SPUindirect <arg>, <arg>), 0)\n"
                 << "With:    (SPUindirect <arg>, <arg>)\n";
          }
#endif

          return IndirectArg;
        } else if (isa<ConstantSDNode>(IndOp1)) {
          // (add (SPUindirect <arg>, <const>), <const>) ->
          // (SPUindirect <arg>, <const + const>)
          ConstantSDNode *CN1 = cast<ConstantSDNode > (IndOp1);
          int64_t combinedConst = CN0->getSExtValue() + CN1->getSExtValue();
          SDValue combinedValue = DAG.getConstant(combinedConst, Op0VT);

#if !defined(NDEBUG)
          if (DebugFlag && isCurrentDebugType(DEBUG_TYPE)) {
            errs() << "\n"
                 << "Replace: (add (SPUindirect <arg>, " << CN1->getSExtValue()
                 << "), " << CN0->getSExtValue() << ")\n"
                 << "With:    (SPUindirect <arg>, "
                 << combinedConst << ")\n";
          }
#endif

          return DAG.getNode(SPUISD::IndirectAddr, dl, Op0VT,
                             IndirectArg, combinedValue);
        }
      }
    }
    break;
  }
  case ISD::SIGN_EXTEND:
  case ISD::ZERO_EXTEND:
  case ISD::ANY_EXTEND: {
    if (Op0.getOpcode() == SPUISD::VEC2PREFSLOT && NodeVT == Op0VT) {
      // (any_extend (SPUextract_elt0 <arg>)) ->
      // (SPUextract_elt0 <arg>)
      // Types must match, however...
#if !defined(NDEBUG)
      if (DebugFlag && isCurrentDebugType(DEBUG_TYPE)) {
        errs() << "\nReplace: ";
        N->dump(&DAG);
        errs() << "\nWith:    ";
        Op0.getNode()->dump(&DAG);
        errs() << "\n";
      }
#endif

      return Op0;
    }
    break;
  }
  case SPUISD::IndirectAddr: {
    if (!ST->usingLargeMem() && Op0.getOpcode() == SPUISD::AFormAddr) {
      ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N->getOperand(1));
      if (CN != 0 && CN->isNullValue()) {
        // (SPUindirect (SPUaform <addr>, 0), 0) ->
        // (SPUaform <addr>, 0)

        DEBUG(errs() << "Replace: ");
        DEBUG(N->dump(&DAG));
        DEBUG(errs() << "\nWith:    ");
        DEBUG(Op0.getNode()->dump(&DAG));
        DEBUG(errs() << "\n");

        return Op0;
      }
    } else if (Op0.getOpcode() == ISD::ADD) {
      SDValue Op1 = N->getOperand(1);
      if (ConstantSDNode *CN1 = dyn_cast<ConstantSDNode>(Op1)) {
        // (SPUindirect (add <arg>, <arg>), 0) ->
        // (SPUindirect <arg>, <arg>)
        if (CN1->isNullValue()) {

#if !defined(NDEBUG)
          if (DebugFlag && isCurrentDebugType(DEBUG_TYPE)) {
            errs() << "\n"
                 << "Replace: (SPUindirect (add <arg>, <arg>), 0)\n"
                 << "With:    (SPUindirect <arg>, <arg>)\n";
          }
#endif

          return DAG.getNode(SPUISD::IndirectAddr, dl, Op0VT,
                             Op0.getOperand(0), Op0.getOperand(1));
        }
      }
    }
    break;
  }
  case SPUISD::SHLQUAD_L_BITS:
  case SPUISD::SHLQUAD_L_BYTES:
  case SPUISD::ROTBYTES_LEFT: {
    SDValue Op1 = N->getOperand(1);

    // Kill degenerate vector shifts:
    if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(Op1)) {
      if (CN->isNullValue()) {
        Result = Op0;
      }
    }
    break;
  }
  case SPUISD::PREFSLOT2VEC: {
    switch (Op0.getOpcode()) {
    default:
      break;
    case ISD::ANY_EXTEND:
    case ISD::ZERO_EXTEND:
    case ISD::SIGN_EXTEND: {
      // (SPUprefslot2vec (any|zero|sign_extend (SPUvec2prefslot <arg>))) ->
      // <arg>
      // but only if the SPUprefslot2vec and <arg> types match.
      SDValue Op00 = Op0.getOperand(0);
      if (Op00.getOpcode() == SPUISD::VEC2PREFSLOT) {
        SDValue Op000 = Op00.getOperand(0);
        if (Op000.getValueType() == NodeVT) {
          Result = Op000;
        }
      }
      break;
    }
    case SPUISD::VEC2PREFSLOT: {
      // (SPUprefslot2vec (SPUvec2prefslot <arg>)) ->
      // <arg>
      Result = Op0.getOperand(0);
      break;
    }
    }
    break;
  }
  }

  // Otherwise, return unchanged.
#ifndef NDEBUG
  if (Result.getNode()) {
    DEBUG(errs() << "\nReplace.SPU: ");
    DEBUG(N->dump(&DAG));
    DEBUG(errs() << "\nWith:        ");
    DEBUG(Result.getNode()->dump(&DAG));
    DEBUG(errs() << "\n");
  }
#endif

  return Result;
}

//===----------------------------------------------------------------------===//
// Inline Assembly Support
//===----------------------------------------------------------------------===//

/// getConstraintType - Given a constraint letter, return the type of
/// constraint it is for this target.
SPUTargetLowering::ConstraintType
SPUTargetLowering::getConstraintType(const std::string &ConstraintLetter) const {
  if (ConstraintLetter.size() == 1) {
    switch (ConstraintLetter[0]) {
    default: break;
    case 'b':
    case 'r':
    case 'f':
    case 'v':
    case 'y':
      return C_RegisterClass;
    }
  }
  return TargetLowering::getConstraintType(ConstraintLetter);
}

std::pair<unsigned, const TargetRegisterClass*>
SPUTargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
                                                EVT 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)
        return std::make_pair(0U, SPU::R64CRegisterClass);
      return std::make_pair(0U, SPU::R32CRegisterClass);
    case 'f':
      if (VT == MVT::f32)
        return std::make_pair(0U, SPU::R32FPRegisterClass);
      else if (VT == MVT::f64)
        return std::make_pair(0U, SPU::R64FPRegisterClass);
      break;
    case 'v':
      return std::make_pair(0U, SPU::GPRCRegisterClass);
    }
  }

  return TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
}

//! Compute used/known bits for a SPU operand
void
SPUTargetLowering::computeMaskedBitsForTargetNode(const SDValue Op,
                                                  const APInt &Mask,
                                                  APInt &KnownZero,
                                                  APInt &KnownOne,
                                                  const SelectionDAG &DAG,
                                                  unsigned Depth ) const {
#if 0
  const uint64_t uint64_sizebits = sizeof(uint64_t) * CHAR_BIT;

  switch (Op.getOpcode()) {
  default:
    // KnownZero = KnownOne = APInt(Mask.getBitWidth(), 0);
    break;
  case CALL:
  case SHUFB:
  case SHUFFLE_MASK:
  case CNTB:
  case SPUISD::PREFSLOT2VEC:
  case SPUISD::LDRESULT:
  case SPUISD::VEC2PREFSLOT:
  case SPUISD::SHLQUAD_L_BITS:
  case SPUISD::SHLQUAD_L_BYTES:
  case SPUISD::VEC_ROTL:
  case SPUISD::VEC_ROTR:
  case SPUISD::ROTBYTES_LEFT:
  case SPUISD::SELECT_MASK:
  case SPUISD::SELB:
  }
#endif
}

unsigned
SPUTargetLowering::ComputeNumSignBitsForTargetNode(SDValue Op,
                                                   unsigned Depth) const {
  switch (Op.getOpcode()) {
  default:
    return 1;

  case ISD::SETCC: {
    EVT VT = Op.getValueType();

    if (VT != MVT::i8 && VT != MVT::i16 && VT != MVT::i32) {
      VT = MVT::i32;
    }
    return VT.getSizeInBits();
  }
  }
}

// LowerAsmOperandForConstraint
void
SPUTargetLowering::LowerAsmOperandForConstraint(SDValue Op,
                                                char ConstraintLetter,
                                                std::vector<SDValue> &Ops,
                                                SelectionDAG &DAG) const {
  // Default, for the time being, to the base class handler
  TargetLowering::LowerAsmOperandForConstraint(Op, ConstraintLetter, Ops, DAG);
}

/// isLegalAddressImmediate - Return true if the integer value can be used
/// as the offset of the target addressing mode.
bool SPUTargetLowering::isLegalAddressImmediate(int64_t V,
                                                const Type *Ty) const {
  // SPU's addresses are 256K:
  return (V > -(1 << 18) && V < (1 << 18) - 1);
}

bool SPUTargetLowering::isLegalAddressImmediate(llvm::GlobalValue* GV) const {
  return false;
}

bool
SPUTargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const {
  // The SPU target isn't yet aware of offsets.
  return false;
}