summaryrefslogtreecommitdiff
path: root/src/jit/lowerxarch.cpp
blob: 585f97b3c61fb931bfee20b678d5cb4ebabed27c (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
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
// Licensed to the .NET Foundation under one or more agreements.
// The .NET Foundation licenses this file to you under the MIT license.
// See the LICENSE file in the project root for more information.

/*XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XX                                                                           XX
XX                           Lowering for AMD64                              XX
XX                                                                           XX
XX  This encapsulates all the logic for lowering trees for the AMD64         XX
XX  architecture.  For a more detailed view of what is lowering, please      XX
XX  take a look at Lower.cpp                                                 XX
XX                                                                           XX
XX                                                                           XX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
*/

#include "jitpch.h"
#ifdef _MSC_VER
#pragma hdrstop
#endif

#ifndef LEGACY_BACKEND // This file is ONLY used for the RyuJIT backend that uses the linear scan register allocator

#ifdef _TARGET_XARCH_

#include "jit.h"
#include "lower.h"

// xarch supports both ROL and ROR instructions so no lowering is required.
void Lowering::LowerRotate(GenTreePtr tree)
{
}

// there is not much lowering to do with storing a local but 
// we do some handling of contained immediates and widening operations of unsigneds
void Lowering::LowerStoreLoc(GenTreeLclVarCommon* storeLoc)
{
    TreeNodeInfo* info = &(storeLoc->gtLsraInfo);

    // Is this the case of var = call where call is returning
    // a value in multiple return registers?
    GenTree* op1 = storeLoc->gtGetOp1();
    if (op1->IsMultiRegCall())
    {
        // backend expects to see this case only for store lclvar.
        assert(storeLoc->OperGet() == GT_STORE_LCL_VAR);

        // srcCount = number of registers in which the value is returned by call
        GenTreeCall* call = op1->AsCall();
        ReturnTypeDesc* retTypeDesc = call->GetReturnTypeDesc();
        info->srcCount = retTypeDesc->GetReturnRegCount();

        // Call node srcCandidates = Bitwise-OR(allregs(GetReturnRegType(i))) for all i=0..RetRegCount-1
        regMaskTP srcCandidates = m_lsra->allMultiRegCallNodeRegs(call);
        op1->gtLsraInfo.setSrcCandidates(m_lsra, srcCandidates);
        return;
    }

#ifdef FEATURE_SIMD    
    if (storeLoc->TypeGet() == TYP_SIMD12)
    {
        // Need an additional register to extract upper 4 bytes of Vector3.
        info->internalFloatCount = 1;
        info->setInternalCandidates(m_lsra, m_lsra->allSIMDRegs());

        // In this case don't mark the operand as contained as we want it to
        // be evaluated into an xmm register
        return;
    }
#endif // FEATURE_SIMD

    // If the source is a containable immediate, make it contained, unless it is
    // an int-size or larger store of zero to memory, because we can generate smaller code
    // by zeroing a register and then storing it.
    if (IsContainableImmed(storeLoc, op1) && (!op1->IsIntegralConst(0) || varTypeIsSmall(storeLoc)))
    {
        MakeSrcContained(storeLoc, op1);
    }

    // Try to widen the ops if they are going into a local var.
    if ((storeLoc->gtOper == GT_STORE_LCL_VAR) &&
        (storeLoc->gtOp1->gtOper == GT_CNS_INT))
    {
        GenTreeIntCon* con = storeLoc->gtOp1->AsIntCon();
        ssize_t       ival = con->gtIconVal;

        unsigned        varNum = storeLoc->gtLclNum;
        LclVarDsc*      varDsc = comp->lvaTable + varNum;

        if (varDsc->lvIsSIMDType())
        {
            noway_assert(storeLoc->gtType != TYP_STRUCT);
        }
        unsigned size = genTypeSize(storeLoc);
        // If we are storing a constant into a local variable
        // we extend the size of the store here 
        if ((size < 4) && !varTypeIsStruct(varDsc))
        {
            if (!varTypeIsUnsigned(varDsc))
            {
                if (genTypeSize(storeLoc) == 1)
                {
                    if ((ival & 0x7f) != ival)
                    {
                        ival = ival | 0xffffff00;
                    }
                }
                else
                {
                    assert(genTypeSize(storeLoc) == 2);
                    if ((ival & 0x7fff) != ival)
                    {
                        ival = ival | 0xffff0000;
                    }
                }
            }

            // A local stack slot is at least 4 bytes in size, regardless of
            // what the local var is typed as, so auto-promote it here
            // unless it is a field of a promoted struct
            // TODO-XArch-CQ: if the field is promoted shouldn't we also be able to do this?
            if (!varDsc->lvIsStructField)
            {
                storeLoc->gtType = TYP_INT;
                con->SetIconValue(ival);
            }
        }
    }
}


/**
 * Takes care of annotating the register requirements 
 * for every TreeNodeInfo struct that maps to each tree node.
 * Preconditions:
 *    LSRA Has been initialized and there is a TreeNodeInfo node
 *    already allocated and initialized for every tree in the IR.
 * Postconditions:
 *    Every TreeNodeInfo instance has the right annotations on register
 *    requirements needed by LSRA to build the Interval Table (source, 
 *    destination and internal [temp] register counts).
 *    This code is refactored originally from LSRA.
 */
void Lowering::TreeNodeInfoInit(GenTree* stmt)
{
    LinearScan* l = m_lsra;
    Compiler* compiler = comp;

    assert(stmt->gtStmt.gtStmtIsTopLevel());
    GenTree* tree = stmt->gtStmt.gtStmtList;
    
    while (tree)
    {
        TreeNodeInfo* info = &(tree->gtLsraInfo);
        GenTree* next = tree->gtNext;

        switch (tree->OperGet())
        {
            GenTree* op1;
            GenTree* op2;

        default:
            TreeNodeInfoInitSimple(tree);
            break;

        case GT_LCL_FLD:
            info->srcCount = 0;
            info->dstCount = 1;

#ifdef FEATURE_SIMD
            // Need an additional register to read upper 4 bytes of Vector3.
            if (tree->TypeGet() == TYP_SIMD12)
            {
                // We need an internal register different from targetReg in which 'tree' produces its result
                // because both targetReg and internal reg will be in use at the same time. This is achieved
                // by asking for two internal registers.
                info->internalFloatCount = 2;
                info->setInternalCandidates(m_lsra, m_lsra->allSIMDRegs());
            }
#endif
            break;
            
        case GT_STORE_LCL_FLD:
        case GT_STORE_LCL_VAR:
            info->srcCount = 1;
            info->dstCount = 0;
            LowerStoreLoc(tree->AsLclVarCommon());
            break;

        case GT_BOX:
            noway_assert(!"box should not exist here");
            // The result of 'op1' is also the final result
            info->srcCount = 0;
            info->dstCount = 0;
            break;

        case GT_PHYSREGDST:
            info->srcCount = 1;
            info->dstCount = 0;
            break;

        case GT_COMMA:
            {
                GenTreePtr firstOperand;
                GenTreePtr secondOperand;
                if (tree->gtFlags & GTF_REVERSE_OPS)
                {
                    firstOperand  = tree->gtOp.gtOp2;
                    secondOperand = tree->gtOp.gtOp1;
                }
                else
                {
                    firstOperand  = tree->gtOp.gtOp1;
                    secondOperand = tree->gtOp.gtOp2;
                }
                if (firstOperand->TypeGet() != TYP_VOID)
                {
                    firstOperand->gtLsraInfo.isLocalDefUse = true;
                    firstOperand->gtLsraInfo.dstCount = 0;
                }
                if (tree->TypeGet() == TYP_VOID && secondOperand->TypeGet() != TYP_VOID)
                {
                    secondOperand->gtLsraInfo.isLocalDefUse = true;
                    secondOperand->gtLsraInfo.dstCount = 0;
                }
            }
            info->srcCount = 0;
            info->dstCount = 0;
            break;

        case GT_LIST:
        case GT_ARGPLACE:
        case GT_NO_OP:
        case GT_START_NONGC:
        case GT_PROF_HOOK:
            info->srcCount = 0;
            info->dstCount = 0;
            break;

        case GT_CNS_DBL:
            info->srcCount = 0;
            info->dstCount = 1;
            break;

#if !defined(_TARGET_64BIT_)

        case GT_LONG:
            if (tree->gtNext == nullptr)
            {
                // An uncontained GT_LONG node needs to consume its source operands
                info->srcCount = 2;
            }
            else
            {
                // Passthrough
                info->srcCount = 0;
            }
            info->dstCount = 0;
            break;

#endif // !defined(_TARGET_64BIT_)

        case GT_QMARK:
        case GT_COLON:
            info->srcCount = 0;
            info->dstCount = 0;
            unreached();
            break;

        case GT_RETURN:
            TreeNodeInfoInitReturn(tree);
            break;

        case GT_RETFILT:
            if (tree->TypeGet() == TYP_VOID)
            {
                info->srcCount = 0;
                info->dstCount = 0;
            }
            else
            {
                assert(tree->TypeGet() == TYP_INT);

                info->srcCount = 1;
                info->dstCount = 1;

                info->setSrcCandidates(l, RBM_INTRET);
                tree->gtOp.gtOp1->gtLsraInfo.setSrcCandidates(l, RBM_INTRET);
            }
            break;

            // A GT_NOP is either a passthrough (if it is void, or if it has
            // a child), but must be considered to produce a dummy value if it
            // has a type but no child
        case GT_NOP:
            info->srcCount = 0;
            if (tree->TypeGet() != TYP_VOID && tree->gtOp.gtOp1 == nullptr)
            {
                info->dstCount = 1;
            }
            else
            {
                info->dstCount = 0;
            }
            break;

        case GT_JTRUE:
            info->srcCount = 0;
            info->dstCount = 0;
            l->clearDstCount(tree->gtOp.gtOp1);
            break;

        case GT_JMP:
            info->srcCount = 0;
            info->dstCount = 0;
            break;

        case GT_SWITCH:
            // This should never occur since switch nodes must not be visible at this
            // point in the JIT.
            info->srcCount = 0;
            info->dstCount = 0;  // To avoid getting uninit errors.
            noway_assert(!"Switch must be lowered at this point");
            break;

        case GT_JMPTABLE:
            info->srcCount = 0;
            info->dstCount = 1;
            break;

        case GT_SWITCH_TABLE:
            info->srcCount = 2;
            info->internalIntCount = 1;
            info->dstCount = 0;
            break;

        case GT_ASG:
        case GT_ASG_ADD:
        case GT_ASG_SUB:
            noway_assert(!"We should never hit any assignment operator in lowering");
            info->srcCount = 0;
            info->dstCount = 0;
            break;

#if !defined(_TARGET_64BIT_)
        case GT_ADD_LO:
        case GT_ADD_HI:
        case GT_SUB_LO:
        case GT_SUB_HI:
#endif
        case GT_ADD:
        case GT_SUB:
            // SSE2 arithmetic instructions doesn't support the form "op mem, xmm".  
            // Rather they only support "op xmm, mem/xmm" form.
            if (varTypeIsFloating(tree->TypeGet()))
            {
                // overflow operations aren't supported on float/double types.
                assert(!tree->gtOverflow());

                op1 = tree->gtGetOp1();
                op2 = tree->gtGetOp2();

                // No implicit conversions at this stage as the expectation is that
                // everything is made explicit by adding casts.
                assert(op1->TypeGet() == op2->TypeGet());

                info->srcCount = 2;
                info->dstCount = 1;              
 
                if (op2->isMemoryOp() || op2->IsCnsNonZeroFltOrDbl())
                {
                    MakeSrcContained(tree, op2);
                }
                else if (tree->OperIsCommutative() && 
                         (op1->IsCnsNonZeroFltOrDbl() || (op1->isMemoryOp() && IsSafeToContainMem(tree, op1))))
                {
                    // Though we have GT_ADD(op1=memOp, op2=non-memOp, we try to reorder the operands
                    // as long as it is safe so that the following efficient code sequence is generated:
                    //      addss/sd targetReg, memOp    (if op1Reg == targetReg) OR
                    //      movaps targetReg, op2Reg; addss/sd targetReg, [memOp]
                    //
                    // Instead of
                    //      movss op1Reg, [memOp]; addss/sd targetReg, Op2Reg  (if op1Reg == targetReg) OR
                    //      movss op1Reg, [memOp]; movaps targetReg, op1Reg, addss/sd targetReg, Op2Reg  
                    MakeSrcContained(tree, op1);
                }
                else
                {
                    // If there are no containable operands, we can make an operand reg optional.
                    SetRegOptionalForBinOp(tree);
                }
                break;
            }

            __fallthrough;

        case GT_AND:
        case GT_OR:
        case GT_XOR:
            TreeNodeInfoInitLogicalOp(tree);
            break;
      
        case GT_RETURNTRAP:
            // this just turns into a compare of its child with an int
            // + a conditional call
            info->srcCount = 1;
            info->dstCount = 0;
            if (tree->gtOp.gtOp1->isIndir())
            {
                MakeSrcContained(tree, tree->gtOp.gtOp1);
            }
            info->internalIntCount = 1;
            info->setInternalCandidates(l, l->allRegs(TYP_INT));
            break;

        case GT_MOD:
        case GT_DIV:
        case GT_UMOD:
        case GT_UDIV:
            TreeNodeInfoInitModDiv(tree);
            break;

        case GT_MUL:
        case GT_MULHI:
            SetMulOpCounts(tree);
            break;
        
        case GT_INTRINSIC:
            TreeNodeInfoInitIntrinsic(tree);
            break;

#ifdef FEATURE_SIMD
        case GT_SIMD:
            TreeNodeInfoInitSIMD(tree);
            break;
#endif // FEATURE_SIMD

        case GT_CAST:
            TreeNodeInfoInitCast(tree);
            break;

        case GT_NEG:
            info->srcCount = 1;
            info->dstCount = 1;

            // TODO-XArch-CQ:
            // SSE instruction set doesn't have an instruction to negate a number.
            // The recommended way is to xor the float/double number with a bitmask.
            // The only way to xor is using xorps or xorpd both of which operate on 
            // 128-bit operands.  To hold the bit-mask we would need another xmm
            // register or a 16-byte aligned 128-bit data constant. Right now emitter
            // lacks the support for emitting such constants or instruction with mem
            // addressing mode referring to a 128-bit operand. For now we use an 
            // internal xmm register to load 32/64-bit bitmask from data section.
            // Note that by trading additional data section memory (128-bit) we can
            // save on the need for an internal register and also a memory-to-reg
            // move.
            //
            // Note: another option to avoid internal register requirement is by
            // lowering as GT_SUB(0, src).  This will generate code different from
            // Jit64 and could possibly result in compat issues (?).
            if (varTypeIsFloating(tree))
            {
                info->internalFloatCount = 1;
                info->setInternalCandidates(l, l->internalFloatRegCandidates());
            }
            break;
        
        case GT_NOT:
            info->srcCount = 1;
            info->dstCount = 1;
            break;

        case GT_LSH:
        case GT_RSH:
        case GT_RSZ:
        case GT_ROL:
        case GT_ROR:
            TreeNodeInfoInitShiftRotate(tree);
            break;

        case GT_EQ:
        case GT_NE:
        case GT_LT:
        case GT_LE:
        case GT_GE:
        case GT_GT:
            LowerCmp(tree);
            break;

        case GT_CKFINITE:
            info->srcCount = 1;
            info->dstCount = 1;
            info->internalIntCount = 1;
            break;

        case GT_CMPXCHG:
            info->srcCount = 3;
            info->dstCount = 1;

            // comparand is preferenced to RAX.
            // Remaining two operands can be in any reg other than RAX.
            tree->gtCmpXchg.gtOpComparand->gtLsraInfo.setSrcCandidates(l, RBM_RAX);
            tree->gtCmpXchg.gtOpLocation->gtLsraInfo.setSrcCandidates(l, l->allRegs(TYP_INT) & ~RBM_RAX);
            tree->gtCmpXchg.gtOpValue->gtLsraInfo.setSrcCandidates(l, l->allRegs(TYP_INT) & ~RBM_RAX);
            tree->gtLsraInfo.setDstCandidates(l, RBM_RAX);
            break;

        case GT_LOCKADD:
            info->srcCount = 2;
            info->dstCount = 0;

            CheckImmedAndMakeContained(tree, tree->gtOp.gtOp2);
            break;

        case GT_CALL:
            TreeNodeInfoInitCall(tree->AsCall());
            break;

        case GT_ADDR:
            {
                // For a GT_ADDR, the child node should not be evaluated into a register
                GenTreePtr child = tree->gtOp.gtOp1;
                assert(!l->isCandidateLocalRef(child));
                l->clearDstCount(child);
                info->srcCount = 0;
                info->dstCount = 1;
            }
            break;

#ifdef _TARGET_X86_
        case GT_OBJ:
            NYI_X86("GT_OBJ");
#endif //_TARGET_X86_

        case GT_INITBLK:
        case GT_COPYBLK:
        case GT_COPYOBJ:
            TreeNodeInfoInitBlockStore(tree->AsBlkOp());
            break;

#ifdef FEATURE_UNIX_AMD64_STRUCT_PASSING
        case GT_PUTARG_STK:
            TreeNodeInfoInitPutArgStk(tree);
            break;
#endif // FEATURE_UNIX_AMD64_STRUCT_PASSING

        case GT_LCLHEAP:
            TreeNodeInfoInitLclHeap(tree);
            break;

        case GT_ARR_BOUNDS_CHECK:
#ifdef FEATURE_SIMD
        case GT_SIMD_CHK:
#endif // FEATURE_SIMD
            {
                GenTreeBoundsChk* node = tree->AsBoundsChk();
                // Consumes arrLen & index - has no result
                info->srcCount = 2;
                info->dstCount = 0;
            
                GenTreePtr other = nullptr;
                if (CheckImmedAndMakeContained(tree, node->gtIndex))
                {
                    other = node->gtArrLen;
                }
                else if (CheckImmedAndMakeContained(tree, node->gtArrLen))
                {
                    other = node->gtIndex;
                }
                else if (node->gtIndex->isMemoryOp())
                {
                    other = node->gtIndex;
                }
                else
                {
                    other = node->gtArrLen;
                }

                if (other->isMemoryOp())
                {
                    if (node->gtIndex->TypeGet() == node->gtArrLen->TypeGet())
                    {
                        MakeSrcContained(tree, other);
                    }
                }
                else
                {
                    // since 'other' operand is not contained, we can mark it as reg optional
                    TryToSetRegOptional(other);
                }
            }
            break;

        case GT_ARR_ELEM:
            // These must have been lowered to GT_ARR_INDEX
            noway_assert(!"We should never see a GT_ARR_ELEM in lowering");
            info->srcCount = 0;
            info->dstCount = 0;
            break;

        case GT_ARR_INDEX:
            info->srcCount = 2;
            info->dstCount = 1;
            // For GT_ARR_INDEX, the lifetime of the arrObj must be extended because it is actually used multiple
            // times while the result is being computed.
            tree->AsArrIndex()->ArrObj()->gtLsraInfo.isDelayFree = true;
            info->hasDelayFreeSrc = true;
            break;

        case GT_ARR_OFFSET:
            // This consumes the offset, if any, the arrObj and the effective index,
            // and produces the flattened offset for this dimension.
            info->srcCount = 3;
            info->dstCount = 1;
            info->internalIntCount = 1;
            // we don't want to generate code for this
            if (tree->gtArrOffs.gtOffset->IsIntegralConst(0))
            {
                MakeSrcContained(tree, tree->gtArrOffs.gtOffset);
            }
            break;

        case GT_LEA:
            // The LEA usually passes its operands through to the GT_IND, in which case we'll
            // clear the info->srcCount and info->dstCount later, but we may be instantiating an address,
            // so we set them here.
            info->srcCount = 0;
            if (tree->AsAddrMode()->HasBase())
            {
                info->srcCount++;
            }
            if (tree->AsAddrMode()->HasIndex())
            {
                info->srcCount++;
            }
            info->dstCount = 1;
            break;

        case GT_STOREIND:
            {
                info->srcCount = 2;
                info->dstCount = 0;
                GenTree* src = tree->gtOp.gtOp2;

                if (compiler->codeGen->gcInfo.gcIsWriteBarrierAsgNode(tree))
                {
                    LowerGCWriteBarrier(tree);
                    break;
                }

                // If the source is a containable immediate, make it contained, unless it is
                // an int-size or larger store of zero to memory, because we can generate smaller code
                // by zeroing a register and then storing it.
                if (IsContainableImmed(tree, src) &&
                    (!src->IsIntegralConst(0) || varTypeIsSmall(tree) || tree->gtGetOp1()->OperGet() == GT_CLS_VAR_ADDR))
                {
                    MakeSrcContained(tree, src);
                }
                else if (!varTypeIsFloating(tree))
                {
                    // Perform recognition of trees with the following structure:
                    //        StoreInd(addr, BinOp(expr, GT_IND(addr)))
                    // to be able to fold this into an instruction of the form
                    //        BINOP [addr], register
                    // where register is the actual place where 'expr' is computed.
                    //
                    // SSE2 doesn't support RMW form of instructions.
                    if (SetStoreIndOpCountsIfRMWMemOp(tree))
                    {
                        break;
                    }
                }

                SetIndirAddrOpCounts(tree);
            }
            break;
        
        case GT_NULLCHECK:
            info->dstCount = 0;
            info->srcCount = 1;
            info->isLocalDefUse = true;
            break;

        case GT_IND:
            info->dstCount = 1;
            info->srcCount = 1;            
            SetIndirAddrOpCounts(tree);
            break;

        case GT_CATCH_ARG:
            info->srcCount = 0;
            info->dstCount = 1;
            info->setDstCandidates(l, RBM_EXCEPTION_OBJECT);
            break;

#if !FEATURE_EH_FUNCLETS
        case GT_END_LFIN:
            info->srcCount = 0;
            info->dstCount = 0;
            break;
#endif

        case GT_CLS_VAR:
            info->srcCount = 0;
            // GT_CLS_VAR, by the time we reach the backend, must always
            // be a pure use.
            // It will produce a result of the type of the
            // node, and use an internal register for the address.

            info->dstCount = 1;
            assert((tree->gtFlags & (GTF_VAR_DEF|GTF_VAR_USEASG|GTF_VAR_USEDEF)) == 0);
            info->internalIntCount = 1;
            break;
        } // end switch (tree->OperGet())

        // If op2 of a binary-op gets marked as contained, then binary-op srcCount will be 1.
        // Even then we would like to set isTgtPref on Op1.
        if (tree->OperIsBinary() && info->srcCount >= 1)
        {
            if (isRMWRegOper(tree))
            {
                GenTree* op1 = tree->gtOp.gtOp1;
                GenTree* op2 = tree->gtOp.gtOp2;
                
                // Commutative opers like add/mul/and/or/xor could reverse the order of
                // operands if it is safe to do so.  In such a case we would like op2 to be
                // target preferenced instead of op1.
                if (tree->OperIsCommutative() && op1->gtLsraInfo.dstCount == 0 && op2 != nullptr)
                {
                    op1 = op2;
                    op2 = tree->gtOp.gtOp1;
                }

                // If we have a read-modify-write operation, we want to preference op1 to the target.
                // If op1 is contained, we don't want to preference it, but it won't
                // show up as a source in that case, so it will be ignored.
                op1->gtLsraInfo.isTgtPref = true;

                // Is this a non-commutative operator, or is op2 a contained memory op?
                // (Note that we can't call IsContained() at this point because it uses exactly the
                // same information we're currently computing.)
                // In either case, we need to make op2 remain live until the op is complete, by marking
                // the source(s) associated with op2 as "delayFree".
                // Note that if op2 of a binary RMW operator is a memory op, even if the operator
                // is commutative, codegen cannot reverse them.
                // TODO-XArch-CQ: This is not actually the case for all RMW binary operators, but there's
                // more work to be done to correctly reverse the operands if they involve memory
                // operands.  Also, we may need to handle more cases than GT_IND, especially once
                // we've modified the register allocator to not require all nodes to be assigned
                // a register (e.g. a spilled lclVar can often be referenced directly from memory).
                // Note that we may have a null op2, even with 2 sources, if op1 is a base/index memory op.

                GenTree* delayUseSrc = nullptr;
                // TODO-XArch-Cleanup: We should make the indirection explicit on these nodes so that we don't have
                // to special case them.
                if (tree->OperGet() == GT_XADD || tree->OperGet() == GT_XCHG || tree->OperGet() == GT_LOCKADD)
                {
                    delayUseSrc = op1;
                }
                else if ((op2 != nullptr) &&
                         (!tree->OperIsCommutative() || (op2->isMemoryOp() && (op2->gtLsraInfo.srcCount == 0))))
                {
                    delayUseSrc = op2;
                }
                if (delayUseSrc != nullptr)
                {
                    // If delayUseSrc is an indirection and it doesn't produce a result, then we need to set "delayFree'
                    // on the base & index, if any.
                    // Otherwise, we set it on delayUseSrc itself.
                    if (delayUseSrc->isIndir() && (delayUseSrc->gtLsraInfo.dstCount == 0))
                    {
                        GenTree* base = delayUseSrc->AsIndir()->Base();
                        GenTree* index = delayUseSrc->AsIndir()->Index();
                        if (base != nullptr)
                        {
                            base->gtLsraInfo.isDelayFree = true;
                        }
                        if (index != nullptr)
                        {
                            index->gtLsraInfo.isDelayFree = true;
                        }
                    }
                    else
                    {
                        delayUseSrc->gtLsraInfo.isDelayFree = true;
                    }
                    info->hasDelayFreeSrc = true;
                }
            }
        }

#ifdef _TARGET_X86_
        // Exclude RBM_NON_BYTE_REGS from dst candidates of tree node and src candidates of operands
        // if the tree node is a byte type.
        // 
        // Example1: GT_STOREIND(byte, addr, op2) - storeind of byte sized value from op2 into mem 'addr'
        // Storeind itself will not produce any value and hence dstCount=0. But op2 could be TYP_INT
        // value. In this case we need to exclude esi/edi from the src candidates of op2.
        // 
        // Example2: GT_CAST(int <- bool <- int) - here type of GT_CAST node is int and castToType is bool.
        //
        // Example3: GT_EQ(int, op1 of type ubyte, op2 of type ubyte) - in this case codegen uses
        // ubyte as the result of comparison and if the result needs to be materialized into a reg
        // simply zero extend it to TYP_INT size.  Here is an example of generated code:
        //         cmp dl, byte ptr[addr mode]
        //         movzx edx, dl
        //
        // Though this looks conservative in theory, in practice we could not think of a case where
        // the below logic leads to conservative register specification.  In future when or if we find
        // one such case, this logic needs to be fine tuned for that case(s).
        if (varTypeIsByte(tree) || 
            ((tree->OperGet() == GT_CAST) && varTypeIsByte(tree->CastToType())) ||
            (tree->OperIsCompare() && varTypeIsByte(tree->gtGetOp1()) && varTypeIsByte(tree->gtGetOp2()))
           )
        {
            regMaskTP regMask;
            if (info->dstCount > 0)
            {
                regMask = info->getDstCandidates(l);
                assert(regMask != RBM_NONE);
                info->setDstCandidates(l, regMask & ~RBM_NON_BYTE_REGS);
            }

            if (tree->OperIsSimple() && (info->srcCount > 0))
            {
                // No need to set src candidates on a contained child operand.
                GenTree *op = tree->gtOp.gtOp1;
                assert(op != nullptr);
                bool containedNode = (op->gtLsraInfo.srcCount == 0) && (op->gtLsraInfo.dstCount == 0);
                if (!containedNode)
                {
                    regMask = op->gtLsraInfo.getSrcCandidates(l);
                    assert(regMask != RBM_NONE);
                    op->gtLsraInfo.setSrcCandidates(l, regMask & ~RBM_NON_BYTE_REGS);
                }

                if (tree->OperIsBinary() && (tree->gtOp.gtOp2 != nullptr))
                {
                    op = tree->gtOp.gtOp2;
                    containedNode = (op->gtLsraInfo.srcCount == 0) && (op->gtLsraInfo.dstCount == 0);                
                    if (!containedNode)
                    {
                        regMask = op->gtLsraInfo.getSrcCandidates(l);
                        assert(regMask != RBM_NONE);
                        op->gtLsraInfo.setSrcCandidates(l, regMask & ~RBM_NON_BYTE_REGS);
                    }
                }
            }
        }
#endif //_TARGET_X86_

        // We need to be sure that we've set info->srcCount and info->dstCount appropriately
        assert((info->dstCount < 2) ||
               (tree->IsMultiRegCall() && info->dstCount == MAX_RET_REG_COUNT));

        tree = next;
    }
}


//------------------------------------------------------------------------
// TreeNodeInfoInitSimple: Sets the srcCount and dstCount for all the trees
// without special handling based on the tree node type.
//
// Arguments:
//    tree      - The node of interest
//
// Return Value:
//    None.
//
void Lowering::TreeNodeInfoInitSimple(GenTree* tree)
{
    TreeNodeInfo* info = &(tree->gtLsraInfo);
    unsigned kind = tree->OperKind();
    info->dstCount = (tree->TypeGet() == TYP_VOID) ? 0 : 1;
    if (kind & (GTK_CONST | GTK_LEAF))
    {
        info->srcCount = 0;
    }
    else if (kind & (GTK_SMPOP))
    {
        if (tree->gtGetOp2() != nullptr)
        {
            info->srcCount = 2;
        }
        else
        {
            info->srcCount = 1;
        }
    }
    else
    {
        unreached();
    }
}


//------------------------------------------------------------------------
// TreeNodeInfoInitReturn: Set the NodeInfo for a GT_RETURN.
//
// Arguments:
//    tree      - The node of interest
//
// Return Value:
//    None.
//
void
Lowering::TreeNodeInfoInitReturn(GenTree* tree)
{
    TreeNodeInfo* info = &(tree->gtLsraInfo);
    LinearScan* l = m_lsra;
    Compiler* compiler = comp;

#if !defined(_TARGET_64BIT_)
    if (tree->TypeGet() == TYP_LONG)
    {
        GenTree* op1 = tree->gtGetOp1();
        noway_assert(op1->OperGet() == GT_LONG);
        GenTree* loVal = op1->gtGetOp1();
        GenTree* hiVal = op1->gtGetOp2();
        info->srcCount = 2;
        loVal->gtLsraInfo.setSrcCandidates(l, RBM_LNGRET_LO);
        hiVal->gtLsraInfo.setSrcCandidates(l, RBM_LNGRET_HI);
        info->dstCount = 0;
    }
    else
#endif // !defined(_TARGET_64BIT_)
    {
        GenTree* op1 = tree->gtGetOp1();
        regMaskTP useCandidates = RBM_NONE;

        info->srcCount = (tree->TypeGet() == TYP_VOID) ? 0 : 1;
        info->dstCount = 0;

#ifdef FEATURE_UNIX_AMD64_STRUCT_PASSING
        if (varTypeIsStruct(tree))
        {                    
            // op1 has to be either an lclvar or a multi-reg returning call
            if (op1->OperGet() == GT_LCL_VAR)
            {
                GenTreeLclVarCommon* lclVarCommon = op1->AsLclVarCommon();
                LclVarDsc* varDsc = &(compiler->lvaTable[lclVarCommon->gtLclNum]);
                assert(varDsc->lvIsMultiRegRet);

                // Mark var as contained if not enregistrable.
                if (!varTypeIsEnregisterableStruct(op1))
                {
                    MakeSrcContained(tree, op1);
                }
            }
            else
            {
                noway_assert(op1->IsMultiRegCall());

                ReturnTypeDesc* retTypeDesc = op1->AsCall()->GetReturnTypeDesc();
                info->srcCount = retTypeDesc->GetReturnRegCount();
                useCandidates = retTypeDesc->GetABIReturnRegs();
            }
        }
        else
#endif // FEATURE_UNIX_AMD64_STRUCT_PASSING
        {
            // Non-struct type return - determine useCandidates                   
            switch (tree->TypeGet())
            {
            case TYP_VOID:   useCandidates = RBM_NONE; break;
            case TYP_FLOAT:  useCandidates = RBM_FLOATRET; break;
            case TYP_DOUBLE: useCandidates = RBM_DOUBLERET; break;
#if defined(_TARGET_64BIT_)
            case TYP_LONG:   useCandidates = RBM_LNGRET; break;
#endif // defined(_TARGET_64BIT_)
            default:         useCandidates = RBM_INTRET; break;
            }
        }

        if (useCandidates != RBM_NONE)
        {
            op1->gtLsraInfo.setSrcCandidates(l, useCandidates);
        }
    }
}


//------------------------------------------------------------------------
// TreeNodeInfoInitShiftRotate: Set the NodeInfo for a shift or rotate.
//
// Arguments:
//    tree      - The node of interest
//
// Return Value:
//    None.
//
void
Lowering::TreeNodeInfoInitShiftRotate(GenTree* tree)
{
    TreeNodeInfo* info = &(tree->gtLsraInfo);
    LinearScan* l = m_lsra;

    info->srcCount = 2;
    info->dstCount = 1;

    // For shift operations, we need that the number
    // of bits moved gets stored in CL in case 
    // the number of bits to shift is not a constant.
    GenTreePtr shiftBy = tree->gtOp.gtOp2;
    GenTreePtr source = tree->gtOp.gtOp1;

    // x64 can encode 8 bits of shift and it will use 5 or 6. (the others are masked off)
    // We will allow whatever can be encoded - hope you know what you are doing.
    if (!IsContainableImmed(tree, shiftBy) ||
        (shiftBy->gtIntConCommon.IconValue() > 255) ||
        (shiftBy->gtIntConCommon.IconValue() < 0))
    {
        source->gtLsraInfo.setSrcCandidates(l, l->allRegs(TYP_INT) & ~RBM_RCX);
        shiftBy->gtLsraInfo.setSrcCandidates(l, RBM_RCX);
        info->setDstCandidates(l, l->allRegs(TYP_INT) & ~RBM_RCX);
    }
    else
    {
        MakeSrcContained(tree, shiftBy);
    }
}


//------------------------------------------------------------------------
// TreeNodeInfoInitCall: Set the NodeInfo for a call.
//
// Arguments:
//    call      - The call node of interest
//
// Return Value:
//    None.
//
void
Lowering::TreeNodeInfoInitCall(GenTreeCall* call)
{
    TreeNodeInfo* info = &(call->gtLsraInfo);
    LinearScan* l = m_lsra;
    Compiler* compiler = comp;
    bool hasMultiRegRetVal = false;
    ReturnTypeDesc* retTypeDesc = nullptr;

    info->srcCount = 0;
    if (call->TypeGet() != TYP_VOID)
    {
        hasMultiRegRetVal = call->HasMultiRegRetVal();
        if (hasMultiRegRetVal)
        {
            // dst count = number of registers in which the value is returned by call
            retTypeDesc = call->GetReturnTypeDesc();
            info->dstCount = retTypeDesc->GetReturnRegCount();
        }
        else
        {
            info->dstCount = 1;
        }
    }
    else
    {
        info->dstCount = 0;
    }
    
    GenTree* ctrlExpr = call->gtControlExpr;
    if (call->gtCallType == CT_INDIRECT)
    {
        // either gtControlExpr != null or gtCallAddr != null.
        // Both cannot be non-null at the same time.
        assert(ctrlExpr == nullptr);
        assert(call->gtCallAddr != nullptr);
        ctrlExpr = call->gtCallAddr;
    }

    // set reg requirements on call target represented as control sequence.
    if (ctrlExpr != nullptr)
    {
        // we should never see a gtControlExpr whose type is void.
        assert(ctrlExpr->TypeGet() != TYP_VOID);                

        // call can take a Rm op on x64
        info->srcCount++;               

        // In case of fast tail implemented as jmp, make sure that gtControlExpr is
        // computed into a register.
        if (!call->IsFastTailCall())
        {
            if (ctrlExpr->isIndir())
            {
                MakeSrcContained(call, ctrlExpr);
            }
        }
        else
        {
            // Fast tail call - make sure that call target is always computed in RAX
            // so that epilog sequence can generate "jmp rax" to achieve fast tail call.
            ctrlExpr->gtLsraInfo.setSrcCandidates(l, RBM_RAX);
        }
    }

    // If this is a varargs call, we will clear the internal candidates in case we need
    // to reserve some integer registers for copying float args.
    // We have to do this because otherwise the default candidates are allRegs, and adding
    // the individual specific registers will have no effect.
    if (call->IsVarargs())
    {
        info->setInternalCandidates(l, RBM_NONE);
    }

    RegisterType registerType = call->TypeGet();

    // Set destination candidates for return value of the call.
#ifdef _TARGET_X86_
    if (call->IsHelperCall(compiler, CORINFO_HELP_INIT_PINVOKE_FRAME))
    {
        // The x86 CORINFO_HELP_INIT_PINVOKE_FRAME helper uses a custom calling convention that returns with
        // TCB in REG_PINVOKE_TCB. AMD64/ARM64 use the standard calling convention. fgMorphCall() sets the
        // correct argument registers.
        info->setDstCandidates(l, RBM_PINVOKE_TCB);
    }
    else
#endif // _TARGET_X86_
    if (hasMultiRegRetVal)
    {
        assert(retTypeDesc != nullptr);
        info->setDstCandidates(l, retTypeDesc->GetABIReturnRegs());
    }
    else if (varTypeIsFloating(registerType))
    {
#ifdef _TARGET_X86_
        // The return value will be on the X87 stack, and we will need to move it.
        info->setDstCandidates(l, l->allRegs(registerType));
#else // !_TARGET_X86_
        info->setDstCandidates(l, RBM_FLOATRET);
#endif // !_TARGET_X86_
    }
    else if (registerType == TYP_LONG)
    {
        info->setDstCandidates(l, RBM_LNGRET);
    }
    else
    {
        info->setDstCandidates(l, RBM_INTRET);
    }

    // number of args to a call = 
    // callRegArgs + (callargs - placeholders, setup, etc)
    // there is an explicit thisPtr but it is redundant 

    // If there is an explicit this pointer, we don't want that node to produce anything
    // as it is redundant
    if (call->gtCallObjp != nullptr)
    {
        GenTreePtr thisPtrNode = call->gtCallObjp;

        if (thisPtrNode->gtOper == GT_PUTARG_REG)
        {
            l->clearOperandCounts(thisPtrNode);
            l->clearDstCount(thisPtrNode->gtOp.gtOp1);
        }
        else
        {
            l->clearDstCount(thisPtrNode);
        }
    }

    // First, count reg args
#if FEATURE_VARARG
    bool callHasFloatRegArgs = false;
#endif // !FEATURE_VARARG
    
    for (GenTreePtr list = call->gtCallLateArgs; list; list = list->MoveNext())
    {
        assert(list->IsList());

        GenTreePtr argNode = list->Current();

        fgArgTabEntryPtr curArgTabEntry = compiler->gtArgEntryByNode(call, argNode);
        assert(curArgTabEntry);

        if (curArgTabEntry->regNum == REG_STK)
        {
            // late arg that is not passed in a register
            DISPNODE(argNode);
            assert(argNode->gtOper == GT_PUTARG_STK);
            argNode->gtLsraInfo.srcCount = 1;
            argNode->gtLsraInfo.dstCount = 0;

#ifdef FEATURE_UNIX_AMD64_STRUCT_PASSING
            // If the node is TYP_STRUCT and it is put on stack with
            // putarg_stk operation, we consume and produce no registers.
            // In this case the embedded Obj node should not produce 
            // registers too since it is contained.
            // Note that if it is a SIMD type the argument will be in a register.
            if (argNode->TypeGet() == TYP_STRUCT)
            {
                assert(argNode->gtOp.gtOp1 != nullptr && argNode->gtOp.gtOp1->OperGet() == GT_OBJ);
                argNode->gtOp.gtOp1->gtLsraInfo.dstCount = 0;
                argNode->gtLsraInfo.srcCount = 0;
            }
#endif // FEATURE_UNIX_AMD64_STRUCT_PASSING
            continue;
        }

        regNumber argReg = REG_NA;
        regMaskTP argMask = RBM_NONE;
        short regCount = 0;
        bool isOnStack = true;
        if (curArgTabEntry->regNum != REG_STK)
        {
            isOnStack = false;
            var_types argType = argNode->TypeGet();

#if FEATURE_VARARG
            callHasFloatRegArgs |= varTypeIsFloating(argType);
#endif // !FEATURE_VARARG

            argReg = curArgTabEntry->regNum;
            regCount = 1;

            // Default case is that we consume one source; modify this later (e.g. for
            // promoted structs)
            info->srcCount++;

            argMask = genRegMask(argReg);
            argNode = argNode->gtEffectiveVal();
        }

        // If the struct arg is wrapped in CPYBLK the type of the param will be TYP_VOID.
        // Use the curArgTabEntry's isStruct to get whether the param is a struct.
        if (varTypeIsStruct(argNode)
            FEATURE_UNIX_AMD64_STRUCT_PASSING_ONLY(|| curArgTabEntry->isStruct))
        {
            unsigned originalSize = 0;
            LclVarDsc* varDsc = nullptr;
            if (argNode->gtOper == GT_LCL_VAR)
            {
                varDsc = compiler->lvaTable + argNode->gtLclVarCommon.gtLclNum;
                originalSize = varDsc->lvSize();
            }
            else if (argNode->gtOper == GT_MKREFANY)
            {
                originalSize = 2 * TARGET_POINTER_SIZE;
            }
            else if (argNode->gtOper == GT_OBJ)
            {
                noway_assert(!"GT_OBJ not supported for amd64");
            }
#ifdef FEATURE_UNIX_AMD64_STRUCT_PASSING
            else if (argNode->gtOper == GT_PUTARG_REG)
            {
                originalSize = genTypeSize(argNode->gtType);
            }
            else if (argNode->gtOper == GT_LIST)
            {
                originalSize = 0;

                // There could be up to 2 PUTARG_REGs in the list
                GenTreeArgList* argListPtr = argNode->AsArgList();
                unsigned iterationNum = 0;
                for (; argListPtr; argListPtr = argListPtr->Rest())
                {
                    GenTreePtr putArgRegNode = argListPtr->gtOp.gtOp1;
                    assert(putArgRegNode->gtOper == GT_PUTARG_REG);

                    if (iterationNum == 0)
                    {
                        varDsc = compiler->lvaTable + putArgRegNode->gtOp.gtOp1->gtLclVarCommon.gtLclNum;
                        originalSize = varDsc->lvSize();
                        assert(originalSize != 0);
                    }
                    else
                    {
                        // Need an extra source for every node, but the first in the list.
                        info->srcCount++;

                        // Get the mask for the second putarg_reg
                        argMask = genRegMask(curArgTabEntry->otherRegNum);
                    }

                    putArgRegNode->gtLsraInfo.setDstCandidates(l, argMask);
                    putArgRegNode->gtLsraInfo.setSrcCandidates(l, argMask);

                    // To avoid redundant moves, have the argument child tree computed in the
                    // register in which the argument is passed to the call.
                    putArgRegNode->gtOp.gtOp1->gtLsraInfo.setSrcCandidates(l, l->getUseCandidates(putArgRegNode));
                    iterationNum++;
                }

                assert(iterationNum <= CLR_SYSTEMV_MAX_EIGHTBYTES_COUNT_TO_PASS_IN_REGISTERS);
            }
#endif // FEATURE_UNIX_AMD64_STRUCT_PASSING
            else
            {
                noway_assert(!"Can't predict unsupported TYP_STRUCT arg kind");
            }

            unsigned slots = ((unsigned)(roundUp(originalSize, TARGET_POINTER_SIZE))) / REGSIZE_BYTES; 
            unsigned remainingSlots = slots;

            if (!isOnStack)
            {
                remainingSlots = slots - 1;

                regNumber reg = (regNumber)(argReg + 1);
                while (remainingSlots > 0 && reg <= REG_ARG_LAST)
                {
                    argMask |= genRegMask(reg);
                    reg = (regNumber)(reg + 1);
                    remainingSlots--;
                    regCount++;
                }
            }

            short internalIntCount = 0;
            if (remainingSlots > 0)
            {
                // This TYP_STRUCT argument is also passed in the outgoing argument area
                // We need a register to address the TYP_STRUCT
                // And we may need 2
#ifdef FEATURE_UNIX_AMD64_STRUCT_PASSING
                internalIntCount = 1;
#else // FEATURE_UNIX_AMD64_STRUCT_PASSING
                internalIntCount = 2;
#endif // FEATURE_UNIX_AMD64_STRUCT_PASSING
            }
            argNode->gtLsraInfo.internalIntCount = internalIntCount;

#ifdef FEATURE_UNIX_AMD64_STRUCT_PASSING
            if (argNode->gtOper == GT_PUTARG_REG)
            {
                argNode->gtLsraInfo.setDstCandidates(l, argMask);
                argNode->gtLsraInfo.setSrcCandidates(l, argMask);
            }
#endif // FEATURE_UNIX_AMD64_STRUCT_PASSING
        }
        else
        {
            argNode->gtLsraInfo.setDstCandidates(l, argMask);
            argNode->gtLsraInfo.setSrcCandidates(l, argMask);
        }

        // To avoid redundant moves, have the argument child tree computed in the
        // register in which the argument is passed to the call.
        if (argNode->gtOper == GT_PUTARG_REG)
        {
            argNode->gtOp.gtOp1->gtLsraInfo.setSrcCandidates(l, l->getUseCandidates(argNode));
        }

#if FEATURE_VARARG
        // In the case of a varargs call, the ABI dictates that if we have floating point args,
        // we must pass the enregistered arguments in both the integer and floating point registers.
        // Since the integer register is not associated with this arg node, we will reserve it as
        // an internal register so that it is not used during the evaluation of the call node
        // (e.g. for the target).
        if (call->IsVarargs() && varTypeIsFloating(argNode))
        {
            regNumber targetReg = compiler->getCallArgIntRegister(argReg);
            info->setInternalIntCount(info->internalIntCount + 1);
            info->addInternalCandidates(l, genRegMask(targetReg));
        }
#endif // FEATURE_VARARG
    }

    // Now, count stack args
    // Note that these need to be computed into a register, but then
    // they're just stored to the stack - so the reg doesn't
    // need to remain live until the call.  In fact, it must not
    // because the code generator doesn't actually consider it live,
    // so it can't be spilled.

    GenTreePtr args = call->gtCallArgs;
    while (args)
    {
        GenTreePtr arg = args->gtOp.gtOp1;
        if (!(args->gtFlags & GTF_LATE_ARG))
        {                    
            TreeNodeInfo* argInfo = &(arg->gtLsraInfo);
#if !defined(_TARGET_64BIT_)
            if (arg->TypeGet() == TYP_LONG)
            {
                assert(arg->OperGet() == GT_LONG);
                GenTreePtr loArg = arg->gtGetOp1();
                GenTreePtr hiArg = arg->gtGetOp2();
                assert((loArg->OperGet() == GT_PUTARG_STK) && (hiArg->OperGet() == GT_PUTARG_STK));
                assert((loArg->gtLsraInfo.dstCount == 1) && (hiArg->gtLsraInfo.dstCount == 1));
                loArg->gtLsraInfo.isLocalDefUse = true;
                hiArg->gtLsraInfo.isLocalDefUse = true;
            }
            else
#endif // !defined(_TARGET_64BIT_)
            {
                if (argInfo->dstCount != 0)
                {
                    argInfo->isLocalDefUse = true;
                }

                // If the child of GT_PUTARG_STK is a constant, we don't need a register to 
                // move it to memory (stack location).
                // We don't want to make 0 contained, because we can generate smaller code
                // by zeroing a register and then storing it.
                argInfo->dstCount = 0;
                if (arg->gtOper == GT_PUTARG_STK) 
                {
                    GenTree* op1 = arg->gtOp.gtOp1;
                    if (IsContainableImmed(arg, op1) && !op1->IsIntegralConst(0))
                    {
                        MakeSrcContained(arg, op1);
                    }
                }
            }
        }
        args = args->gtOp.gtOp2;
    }

#if FEATURE_VARARG
    // If it is a fast tail call, it is already preferenced to use RAX.
    // Therefore, no need set src candidates on call tgt again.
    if (call->IsVarargs() && 
        callHasFloatRegArgs &&                 
        !call->IsFastTailCall() &&
        (ctrlExpr != nullptr))
    {
        // Don't assign the call target to any of the argument registers because
        // we will use them to also pass floating point arguments as required
        // by Amd64 ABI.
        ctrlExpr->gtLsraInfo.setSrcCandidates(l, l->allRegs(TYP_INT) & ~(RBM_ARG_REGS));
    }
#endif // !FEATURE_VARARG
}


//------------------------------------------------------------------------
// TreeNodeInfoInitBlockStore: Set the NodeInfo for a block store.
//
// Arguments:
//    blkNode       - The block store node of interest
//
// Return Value:
//    None.
//
void
Lowering::TreeNodeInfoInitBlockStore(GenTreeBlkOp* blkNode)
{
    GenTree*  dstAddr = blkNode->Dest();
    unsigned  size;
    LinearScan* l = m_lsra;
    Compiler* compiler = comp;

    // Sources are dest address, initVal or source, and size
    blkNode->gtLsraInfo.srcCount = 3;
    blkNode->gtLsraInfo.dstCount = 0;

    if (blkNode->OperGet() == GT_INITBLK)
    {
        GenTreeInitBlk* initBlkNode = blkNode->AsInitBlk();

        GenTreePtr blockSize = initBlkNode->Size();
        GenTreePtr   initVal = initBlkNode->InitVal();

        // If we have an InitBlk with constant block size we can optimize several ways:
        // a) If the size is smaller than a small memory page but larger than INITBLK_UNROLL_LIMIT bytes 
        //    we use rep stosb since this reduces the register pressure in LSRA and we have
        //    roughly the same performance as calling the helper.
        // b) If the size is <= INITBLK_UNROLL_LIMIT bytes and the fill byte is a constant, 
        //    we can speed this up by unrolling the loop using SSE2 stores.  The reason for
        //    this threshold is because our last investigation (Fall 2013), more than 95% of initblks 
        //    in our framework assemblies are actually <= INITBLK_UNROLL_LIMIT bytes size, so this is the
        //    preferred code sequence for the vast majority of cases.

        // This threshold will decide from using the helper or let the JIT decide to inline
        // a code sequence of its choice.
        ssize_t helperThreshold = max(INITBLK_STOS_LIMIT, INITBLK_UNROLL_LIMIT);

        // TODO-X86-CQ: Investigate whether a helper call would be beneficial on x86
        if (blockSize->IsCnsIntOrI() && blockSize->gtIntCon.gtIconVal <= helperThreshold)
        {
            ssize_t size = blockSize->gtIntCon.gtIconVal;

            // Always favor unrolling vs rep stos.
            if (size <= INITBLK_UNROLL_LIMIT && initVal->IsCnsIntOrI())
            {
                // The fill value of an initblk is interpreted to hold a
                // value of (unsigned int8) however a constant of any size
                // may practically reside on the evaluation stack. So extract
                // the lower byte out of the initVal constant and replicate
                // it to a larger constant whose size is sufficient to support
                // the largest width store of the desired inline expansion.

                ssize_t fill = initVal->gtIntCon.gtIconVal & 0xFF;
#ifdef _TARGET_AMD64_
                if (size < REGSIZE_BYTES)
                {
                    initVal->gtIntCon.gtIconVal = 0x01010101 * fill;
                }
                else
                {
                    initVal->gtIntCon.gtIconVal = 0x0101010101010101LL * fill;
                    initVal->gtType = TYP_LONG;
                }
#else // !_TARGET_AMD64_
                initVal->gtIntCon.gtIconVal = 0x01010101 * fill;
#endif // !_TARGET_AMD64_

                MakeSrcContained(blkNode, blockSize);

                // In case we have a buffer >= 16 bytes
                // we can use SSE2 to do a 128-bit store in a single
                // instruction.
                if (size >= XMM_REGSIZE_BYTES)
                {
                    // Reserve an XMM register to fill it with 
                    // a pack of 16 init value constants.
                    blkNode->gtLsraInfo.internalFloatCount = 1;
                    blkNode->gtLsraInfo.setInternalCandidates(l, l->internalFloatRegCandidates());
                }
                initBlkNode->gtBlkOpKind = GenTreeBlkOp::BlkOpKindUnroll;
            }
            else
            {
                // rep stos has the following register requirements:
                // a) The memory address to be in RDI.
                // b) The fill value has to be in RAX.
                // c) The buffer size must be in RCX.
                dstAddr->gtLsraInfo.setSrcCandidates(l, RBM_RDI);
                initVal->gtLsraInfo.setSrcCandidates(l, RBM_RAX);
                blockSize->gtLsraInfo.setSrcCandidates(l, RBM_RCX);
                initBlkNode->gtBlkOpKind = GenTreeBlkOp::BlkOpKindRepInstr;
            }
        }
        else
        {
#ifdef _TARGET_AMD64_
            // The helper follows the regular AMD64 ABI.
            dstAddr->gtLsraInfo.setSrcCandidates(l, RBM_ARG_0);
            initVal->gtLsraInfo.setSrcCandidates(l, RBM_ARG_1);
            blockSize->gtLsraInfo.setSrcCandidates(l, RBM_ARG_2);
            initBlkNode->gtBlkOpKind = GenTreeBlkOp::BlkOpKindHelper;
#else // !_TARGET_AMD64_
            dstAddr->gtLsraInfo.setSrcCandidates(l, RBM_RDI);
            initVal->gtLsraInfo.setSrcCandidates(l, RBM_RAX);
            blockSize->gtLsraInfo.setSrcCandidates(l, RBM_RCX);
            initBlkNode->gtBlkOpKind = GenTreeBlkOp::BlkOpKindRepInstr;
#endif // !_TARGET_AMD64_
        }
    }
    else if (blkNode->OperGet() == GT_COPYOBJ)
    {
        GenTreeCpObj* cpObjNode = blkNode->AsCpObj();
            
        GenTreePtr  clsTok = cpObjNode->ClsTok();
        GenTreePtr srcAddr = cpObjNode->Source();
            
        unsigned slots = cpObjNode->gtSlots;

#ifdef DEBUG
        // CpObj must always have at least one GC-Pointer as a member.
        assert(cpObjNode->gtGcPtrCount > 0);
            
        assert(dstAddr->gtType == TYP_BYREF || dstAddr->gtType == TYP_I_IMPL);
        assert(clsTok->IsIconHandle());
            
        CORINFO_CLASS_HANDLE clsHnd = (CORINFO_CLASS_HANDLE)clsTok->gtIntCon.gtIconVal;
        size_t classSize = compiler->info.compCompHnd->getClassSize(clsHnd);
        size_t blkSize = roundUp(classSize, TARGET_POINTER_SIZE);
            
        // Currently, the EE always round up a class data structure so 
        // we are not handling the case where we have a non multiple of pointer sized 
        // struct. This behavior may change in the future so in order to keeps things correct
        // let's assert it just to be safe. Going forward we should simply
        // handle this case.
        assert(classSize == blkSize);
        assert((blkSize / TARGET_POINTER_SIZE) == slots);
        assert(cpObjNode->HasGCPtr());
#endif

        bool IsRepMovsProfitable = false;
            
        // If the destination is not on the stack, let's find out if we
        // can improve code size by using rep movsq instead of generating
        // sequences of movsq instructions.
        if (!dstAddr->OperIsLocalAddr())
        {
            // Let's inspect the struct/class layout and determine if it's profitable
            // to use rep movsq for copying non-gc memory instead of using single movsq 
            // instructions for each memory slot.
            unsigned i = 0;
            BYTE* gcPtrs = cpObjNode->gtGcPtrs;
                
            do {
                unsigned nonGCSlots = 0;
                // Measure a contiguous non-gc area inside the struct and note the maximum.
                while (i < slots && gcPtrs[i] == TYPE_GC_NONE)
                {
                    nonGCSlots++;
                    i++;
                }
                    
                while (i < slots && gcPtrs[i] != TYPE_GC_NONE)
                {
                    i++;
                }

                if (nonGCSlots >= CPOBJ_NONGC_SLOTS_LIMIT)
                {
                    IsRepMovsProfitable = true;
                    break;
                }
            } while (i < slots);
        }
        else if (slots >= CPOBJ_NONGC_SLOTS_LIMIT)
        {
            IsRepMovsProfitable = true;
        }

        // There are two cases in which we need to materialize the 
        // struct size:
        // a) When the destination is on the stack we don't need to use the 
        //    write barrier, we can just simply call rep movsq and get a win in codesize.
        // b) If we determine we have contiguous non-gc regions in the struct where it's profitable
        //    to use rep movsq instead of a sequence of single movsq instructions.  According to the
        //    Intel Manual, the sweet spot for small structs is between 4 to 12 slots of size where
        //    the entire operation takes 20 cycles and encodes in 5 bytes (moving RCX, and calling rep movsq).
        if (IsRepMovsProfitable)
        {
            // We need the size of the contiguous Non-GC-region to be in RCX to call rep movsq.
            MakeSrcContained(blkNode, clsTok);
            blkNode->gtLsraInfo.internalIntCount = 1;
            blkNode->gtLsraInfo.setInternalCandidates(l, RBM_RCX);
        }
        else
        {
            // We don't need to materialize the struct size because we will unroll
            // the loop using movsq that automatically increments the pointers.
            MakeSrcContained(blkNode, clsTok);
        }

        dstAddr->gtLsraInfo.setSrcCandidates(l, RBM_RDI);
        srcAddr->gtLsraInfo.setSrcCandidates(l, RBM_RSI);
    }
    else
    {
        assert(blkNode->OperGet() == GT_COPYBLK);
        GenTreeCpBlk* cpBlkNode = blkNode->AsCpBlk();

        GenTreePtr blockSize = cpBlkNode->Size();
        GenTreePtr   srcAddr = cpBlkNode->Source();

        // In case of a CpBlk with a constant size and less than CPBLK_MOVS_LIMIT size
        // we can use rep movs to generate code instead of the helper call.

        // This threshold will decide from using the helper or let the JIT decide to inline
        // a code sequence of its choice.
        ssize_t helperThreshold = max(CPBLK_MOVS_LIMIT, CPBLK_UNROLL_LIMIT);

        // TODO-X86-CQ: Investigate whether a helper call would be beneficial on x86
        if (blockSize->IsCnsIntOrI() && blockSize->gtIntCon.gtIconVal <= helperThreshold)
        {
            assert(!blockSize->IsIconHandle());
            ssize_t size = blockSize->gtIntCon.gtIconVal;

            // If we have a buffer between XMM_REGSIZE_BYTES and CPBLK_UNROLL_LIMIT bytes, we'll use SSE2. 
            // Structs and buffer with sizes <= CPBLK_UNROLL_LIMIT bytes are occurring in more than 95% of
            // our framework assemblies, so this is the main code generation scheme we'll use.
            if (size <= CPBLK_UNROLL_LIMIT)
            {
                MakeSrcContained(blkNode, blockSize);
                    
                // If we have a remainder smaller than XMM_REGSIZE_BYTES, we need an integer temp reg.
                // 
                // x86 specific note: if the size is odd, the last copy operation would be of size 1 byte.
                // But on x86 only RBM_BYTE_REGS could be used as byte registers.  Therefore, exclude
                // RBM_NON_BYTE_REGS from internal candidates.
                if ((size & (XMM_REGSIZE_BYTES - 1)) != 0)
                {
                    blkNode->gtLsraInfo.internalIntCount++;
                    regMaskTP regMask = l->allRegs(TYP_INT);

#ifdef _TARGET_X86_
                    if ((size % 2) != 0)
                    {
                        regMask &= ~RBM_NON_BYTE_REGS;
                    }
#endif
                    blkNode->gtLsraInfo.setInternalCandidates(l, regMask);
                }

                if (size >= XMM_REGSIZE_BYTES)
                {
                    // If we have a buffer larger than XMM_REGSIZE_BYTES, 
                    // reserve an XMM register to use it for a 
                    // series of 16-byte loads and stores.
                    blkNode->gtLsraInfo.internalFloatCount = 1;
                    blkNode->gtLsraInfo.addInternalCandidates(l, l->internalFloatRegCandidates());
                }

                // If src or dst are on stack, we don't have to generate the address into a register
                // because it's just some constant+SP
                if (srcAddr->OperIsLocalAddr())
                {
                    MakeSrcContained(blkNode, srcAddr);
                }

                if (dstAddr->OperIsLocalAddr())
                {
                    MakeSrcContained(blkNode, dstAddr);
                }
                            
                cpBlkNode->gtBlkOpKind = GenTreeBlkOp::BlkOpKindUnroll;
            }
            else
            {
                dstAddr->gtLsraInfo.setSrcCandidates(l, RBM_RDI);
                srcAddr->gtLsraInfo.setSrcCandidates(l, RBM_RSI);
                blockSize->gtLsraInfo.setSrcCandidates(l, RBM_RCX);
                cpBlkNode->gtBlkOpKind = GenTreeBlkOp::BlkOpKindRepInstr;
            }
        }
#ifdef _TARGET_AMD64_
        else
        {
            // In case we have a constant integer this means we went beyond
            // CPBLK_MOVS_LIMIT bytes of size, still we should never have the case of
            // any GC-Pointers in the src struct.
            if (blockSize->IsCnsIntOrI())
            {
                assert(!blockSize->IsIconHandle());
            }

            dstAddr->gtLsraInfo.setSrcCandidates(l, RBM_ARG_0);
            srcAddr->gtLsraInfo.setSrcCandidates(l, RBM_ARG_1);
            blockSize->gtLsraInfo.setSrcCandidates(l, RBM_ARG_2);
            cpBlkNode->gtBlkOpKind = GenTreeBlkOp::BlkOpKindHelper;
        }
#elif defined(_TARGET_X86_)
        else
        {
            dstAddr->gtLsraInfo.setSrcCandidates(l, RBM_RDI);
            srcAddr->gtLsraInfo.setSrcCandidates(l, RBM_RSI);
            blockSize->gtLsraInfo.setSrcCandidates(l, RBM_RCX);
            cpBlkNode->gtBlkOpKind = GenTreeBlkOp::BlkOpKindRepInstr;
        }
#endif // _TARGET_X86_
    }
}


#ifdef FEATURE_UNIX_AMD64_STRUCT_PASSING
//------------------------------------------------------------------------
// TreeNodeInfoInitPutArgStk: Set the NodeInfo for a GT_PUTARG_STK.
//
// Arguments:
//    tree      - The node of interest
//
// Return Value:
//    None.
//
void
Lowering::TreeNodeInfoInitPutArgStk(GenTree* tree)
{
    TreeNodeInfo* info = &(tree->gtLsraInfo);
    LinearScan* l = m_lsra;

    if (tree->TypeGet() != TYP_STRUCT)
    {
        TreeNodeInfoInitSimple(tree);
        return;
    }

    GenTreePutArgStk* putArgStkTree = tree->AsPutArgStk();
    
    GenTreePtr   dst = tree;
    GenTreePtr   src = tree->gtOp.gtOp1;
    GenTreePtr   srcAddr = nullptr;

    if ((src->OperGet() == GT_OBJ) || (src->OperGet() == GT_IND))
    {
        srcAddr = src->gtOp.gtOp1;
    }
    else
    {
        assert(varTypeIsSIMD(tree));
    }
    info->srcCount = src->gtLsraInfo.dstCount;

    // If this is a stack variable address,
    // make the op1 contained, so this way 
    // there is no unnecessary copying between registers.
    // To avoid assertion, increment the parent's source.
    // It is recovered below.
    bool haveLocalAddr = ((srcAddr != nullptr) && (srcAddr->OperIsLocalAddr()));
    if (haveLocalAddr)
    {
        info->srcCount += 1;
    }
    
    info->dstCount = 0;
    
    // In case of a CpBlk we could use a helper call. In case of putarg_stk we 
    // can't do that since the helper call could kill some already set up outgoing args.
    // TODO-Amd64-Unix: converge the code for putarg_stk with cpyblk/cpyobj.
    // The cpyXXXX code is rather complex and this could cause it to be more complex, but
    // it might be the right thing to do.
    
    // This threshold will decide from using the helper or let the JIT decide to inline
    // a code sequence of its choice.
    ssize_t helperThreshold = max(CPBLK_MOVS_LIMIT, CPBLK_UNROLL_LIMIT);
    ssize_t size = putArgStkTree->gtNumSlots * TARGET_POINTER_SIZE;
    
    // TODO-X86-CQ: The helper call either is not supported on x86 or required more work
    // (I don't know which).
    
    // If we have a buffer between XMM_REGSIZE_BYTES and CPBLK_UNROLL_LIMIT bytes, we'll use SSE2. 
    // Structs and buffer with sizes <= CPBLK_UNROLL_LIMIT bytes are occurring in more than 95% of
    // our framework assemblies, so this is the main code generation scheme we'll use.
    if (size <= CPBLK_UNROLL_LIMIT && putArgStkTree->gtNumberReferenceSlots == 0)
    {
        // If we have a remainder smaller than XMM_REGSIZE_BYTES, we need an integer temp reg.
        // 
        // x86 specific note: if the size is odd, the last copy operation would be of size 1 byte.
        // But on x86 only RBM_BYTE_REGS could be used as byte registers.  Therefore, exclude
        // RBM_NON_BYTE_REGS from internal candidates.
        if ((size & (XMM_REGSIZE_BYTES - 1)) != 0)
        {
            info->internalIntCount++;
            regMaskTP regMask = l->allRegs(TYP_INT);
            
#ifdef _TARGET_X86_
            if ((size % 2) != 0)
            {
                regMask &= ~RBM_NON_BYTE_REGS;
            }
#endif
            info->setInternalCandidates(l, regMask);
        }
        
        if (size >= XMM_REGSIZE_BYTES)
        {
            // If we have a buffer larger than XMM_REGSIZE_BYTES, 
            // reserve an XMM register to use it for a 
            // series of 16-byte loads and stores.
            info->internalFloatCount = 1;
            info->addInternalCandidates(l, l->internalFloatRegCandidates());
        }
        
        if (haveLocalAddr)
        {
            MakeSrcContained(putArgStkTree, srcAddr);
        }
        
        // If src or dst are on stack, we don't have to generate the address into a register
        // because it's just some constant+SP
        putArgStkTree->gtPutArgStkKind = GenTreePutArgStk::PutArgStkKindUnroll;
    }
    else
    {
        info->internalIntCount += 3;
        info->setInternalCandidates(l, (RBM_RDI | RBM_RCX | RBM_RSI));
        if (haveLocalAddr)
        {
            MakeSrcContained(putArgStkTree, srcAddr);
        }
        
        putArgStkTree->gtPutArgStkKind = GenTreePutArgStk::PutArgStkKindRepInstr;
    }
    
    // Always mark the OBJ and ADDR as contained trees by the putarg_stk. The codegen will deal with this tree.
    MakeSrcContained(putArgStkTree, src);
    
    // Balance up the inc above.
    if (haveLocalAddr)
    {
        info->srcCount -= 1;
    }
}
#endif // FEATURE_UNIX_AMD64_STRUCT_PASSING


//------------------------------------------------------------------------
// TreeNodeInfoInitLclHeap: Set the NodeInfo for a GT_LCLHEAP.
//
// Arguments:
//    tree      - The node of interest
//
// Return Value:
//    None.
//
void
Lowering::TreeNodeInfoInitLclHeap(GenTree* tree)
{
    TreeNodeInfo* info = &(tree->gtLsraInfo);
    LinearScan* l = m_lsra;
    Compiler* compiler = comp;

    info->srcCount = 1;
    info->dstCount = 1;

    // Need a variable number of temp regs (see genLclHeap() in codegenamd64.cpp):
    // Here '-' means don't care.
    //
    //     Size?                    Init Memory?         # temp regs
    //      0                            -                  0
    //      const and <=6 ptr words      -                  0
    //      const and >6 ptr words       Yes                0
    //      const and <PageSize          No                 0 (amd64) 1 (x86)
    //      const and >=PageSize         No                 2
    //      Non-const                    Yes                0
    //      Non-const                    No                 2            

    GenTreePtr size = tree->gtOp.gtOp1;
    if (size->IsCnsIntOrI())
    {
        MakeSrcContained(tree, size);

        size_t sizeVal = size->gtIntCon.gtIconVal;

        if (sizeVal == 0)
        {
            info->internalIntCount = 0;
        }
        else 
        {
            // Compute the amount of memory to properly STACK_ALIGN.
            // Note: The Gentree node is not updated here as it is cheap to recompute stack aligned size.
            // This should also help in debugging as we can examine the original size specified with localloc.
            sizeVal = AlignUp(sizeVal, STACK_ALIGN);
            size_t cntStackAlignedWidthItems = (sizeVal >> STACK_ALIGN_SHIFT);

            // For small allocations up to 6 pointer sized words (i.e. 48 bytes of localloc)
            // we will generate 'push 0'.
            if (cntStackAlignedWidthItems <= 6)
            {
                info->internalIntCount = 0;
            }
            else if (!compiler->info.compInitMem)
            {
                // No need to initialize allocated stack space.
                if (sizeVal < compiler->eeGetPageSize())
                {
#ifdef _TARGET_X86_
                    info->internalIntCount = 1;     // x86 needs a register here to avoid generating "sub" on ESP.
#else // !_TARGET_X86_
                    info->internalIntCount = 0;
#endif // !_TARGET_X86_
                }
                else
                {
                    // We need two registers: regCnt and RegTmp
                    info->internalIntCount = 2;
                }
            }
            else
            {
                // >6 and need to zero initialize allocated stack space.
                info->internalIntCount = 0;
            }
        }
    }
    else
    {
        if (!compiler->info.compInitMem)
        {
            info->internalIntCount = 2;
        }
        else
        {
            info->internalIntCount = 0;
        }
    }
}


//------------------------------------------------------------------------
// TreeNodeInfoInitLogicalOp: Set the NodeInfo for GT_AND/GT_OR/GT_XOR,
// as well as GT_ADD/GT_SUB.
//
// Arguments:
//    tree      - The node of interest
//
// Return Value:
//    None.
//
void
Lowering::TreeNodeInfoInitLogicalOp(GenTree* tree)
{
    TreeNodeInfo* info = &(tree->gtLsraInfo);
    LinearScan* l = m_lsra;

    // We're not marking a constant hanging on the left of the add
    // as containable so we assign it to a register having CQ impact.
    // TODO-XArch-CQ: Detect this case and support both generating a single instruction 
    // for GT_ADD(Constant, SomeTree)
    info->srcCount = 2;
    info->dstCount = 1;

    GenTree* op1 = tree->gtGetOp1();
    GenTree* op2 = tree->gtGetOp2();            

    // We can directly encode the second operand if it is either a containable constant or a memory-op.
    // In case of memory-op, we can encode it directly provided its type matches with 'tree' type.
    // This is because during codegen, type of 'tree' is used to determine emit Type size. If the types
    // do not match, they get normalized (i.e. sign/zero extended) on load into a register.
    bool directlyEncodable = false;
    bool binOpInRMW = false;
    GenTreePtr operand = nullptr;

    if (IsContainableImmed(tree, op2))
    {
        directlyEncodable = true;
        operand = op2;
    }
    else
    {
        binOpInRMW = IsBinOpInRMWStoreInd(tree);
        if (!binOpInRMW)
        {
            if (op2->isMemoryOp() && tree->TypeGet() == op2->TypeGet())
            {
                directlyEncodable = true;
                operand = op2;
            }
            else if (tree->OperIsCommutative())
            {
                if (IsContainableImmed(tree, op1) ||
                    (op1->isMemoryOp() && tree->TypeGet() == op1->TypeGet() && IsSafeToContainMem(tree, op1)))
                {
                    // If it is safe, we can reverse the order of operands of commutative operations for efficient codegen
                    directlyEncodable = true;
                    operand = op1;
                }
            }
        }
    }

    if (directlyEncodable)
    {
        assert(operand != nullptr);
        MakeSrcContained(tree, operand);
    }
    else if (!binOpInRMW)
    {
        // If this binary op neither has contained operands, nor is a 
        // Read-Modify-Write (RMW) operation, we can mark its operands
        // as reg optional.
        SetRegOptionalForBinOp(tree);
    }
}


//------------------------------------------------------------------------
// TreeNodeInfoInitModDiv: Set the NodeInfo for GT_MOD/GT_DIV/GT_UMOD/GT_UDIV.
//
// Arguments:
//    tree      - The node of interest
//
// Return Value:
//    None.
//
void
Lowering::TreeNodeInfoInitModDiv(GenTree* tree)
{
    TreeNodeInfo* info = &(tree->gtLsraInfo);
    LinearScan* l = m_lsra;

    GenTree* op1 = tree->gtGetOp1();
    GenTree* op2 = tree->gtGetOp2();

    info->srcCount = 2;
    info->dstCount = 1;

    switch (tree->OperGet())
    {
    case GT_MOD:
    case GT_DIV:
        if (varTypeIsFloating(tree->TypeGet()))
        {   
            // No implicit conversions at this stage as the expectation is that
            // everything is made explicit by adding casts.
            assert(op1->TypeGet() == op2->TypeGet());

            if (op2->isMemoryOp() || op2->IsCnsNonZeroFltOrDbl())
            {
                MakeSrcContained(tree, op2);
            }
            else
            {
                // If there are no containable operands, we can make an operand reg optional.
                SetRegOptionalForBinOp(tree);
            }

            return;
        }
        break;

    default:
        break;
    }

    // Amd64 Div/Idiv instruction: 
    //    Dividend in RAX:RDX  and computes
    //    Quotient in RAX, Remainder in RDX

    if (tree->OperGet() == GT_MOD || tree->OperGet() == GT_UMOD)
    {
        // We are interested in just the remainder.
        // RAX is used as a trashable register during computation of remainder.
        info->setDstCandidates(l, RBM_RDX);
    }
    else 
    {
        // We are interested in just the quotient.
        // RDX gets used as trashable register during computation of quotient
        info->setDstCandidates(l, RBM_RAX);            
    }

    // If possible would like to have op1 in RAX to avoid a register move
    op1->gtLsraInfo.setSrcCandidates(l, RBM_RAX);

    // divisor can be an r/m, but the memory indirection must be of the same size as the divide
    if (op2->isMemoryOp() && (op2->TypeGet() == tree->TypeGet()))
    {
        MakeSrcContained(tree, op2);
    }
    else
    {
        op2->gtLsraInfo.setSrcCandidates(l, l->allRegs(TYP_INT) & ~(RBM_RAX | RBM_RDX));

        // If there are no containable operands, we can make an operand reg optional.
        SetRegOptionalForBinOp(tree);
    }
}


//------------------------------------------------------------------------
// TreeNodeInfoInitIntrinsic: Set the NodeInfo for a GT_INTRINSIC.
//
// Arguments:
//    tree      - The node of interest
//
// Return Value:
//    None.
//
void
Lowering::TreeNodeInfoInitIntrinsic(GenTree* tree)
{
    TreeNodeInfo* info = &(tree->gtLsraInfo);
    LinearScan* l = m_lsra;

    // Both operand and its result must be of floating point type.
    GenTree* op1 = tree->gtGetOp1();
    assert(varTypeIsFloating(op1));
    assert(op1->TypeGet() == tree->TypeGet());

    info->srcCount = 1;
    info->dstCount = 1;

    switch(tree->gtIntrinsic.gtIntrinsicId)
    {
    case CORINFO_INTRINSIC_Sqrt:
        if (op1->isMemoryOp() || op1->IsCnsNonZeroFltOrDbl())
        {
            MakeSrcContained(tree, op1);
        }
        else
        {
            // Mark the operand as reg optional since codegen can still 
            // generate code if op1 is on stack.
            TryToSetRegOptional(op1);
        }
        break;

     case CORINFO_INTRINSIC_Abs:
        // Abs(float x) = x & 0x7fffffff
        // Abs(double x) = x & 0x7ffffff ffffffff

        // In case of Abs we need an internal register to hold mask.
         
        // TODO-XArch-CQ: avoid using an internal register for the mask.
        // Andps or andpd both will operate on 128-bit operands.  
        // The data section constant to hold the mask is a 64-bit size.
        // Therefore, we need both the operand and mask to be in 
        // xmm register. When we add support in emitter to emit 128-bit
        // data constants and instructions that operate on 128-bit
        // memory operands we can avoid the need for an internal register.
        if (tree->gtIntrinsic.gtIntrinsicId == CORINFO_INTRINSIC_Abs)
        {
            info->internalFloatCount = 1;
            info->setInternalCandidates(l, l->internalFloatRegCandidates());
        }
        break;

#ifdef _TARGET_X86_
    case CORINFO_INTRINSIC_Cos:                
    case CORINFO_INTRINSIC_Sin:                
    case CORINFO_INTRINSIC_Round:
        NYI_X86("Math intrinsics Cos, Sin and Round");
        break;
#endif // _TARGET_X86_

    default:
        // Right now only Sqrt/Abs are treated as math intrinsics
        noway_assert(!"Unsupported math intrinsic");
        unreached();
        break;
    }
}

#ifdef FEATURE_SIMD
//------------------------------------------------------------------------
// TreeNodeInfoInitSIMD: Set the NodeInfo for a GT_SIMD tree.
//
// Arguments:
//    tree       - The GT_SIMD node of interest
//
// Return Value:
//    None.

void
Lowering::TreeNodeInfoInitSIMD(GenTree* tree)
{
    GenTreeSIMD* simdTree = tree->AsSIMD();
    TreeNodeInfo* info = &(tree->gtLsraInfo);
    LinearScan* lsra = m_lsra;
    info->dstCount = 1;
    switch(simdTree->gtSIMDIntrinsicID)
    {
        GenTree* op2;

    case SIMDIntrinsicInit:
        {
            info->srcCount = 1;
            GenTree* op1 = tree->gtOp.gtOp1;

            // This sets all fields of a SIMD struct to the given value.
            // Mark op1 as contained if it is either zero or int constant of all 1's,
            // or a float constant with 16 or 32 byte simdType (AVX case)
            //
            // Should never see small int base type vectors except for zero initialization.
            assert(!varTypeIsSmallInt(simdTree->gtSIMDBaseType) || op1->IsIntegralConst(0));

            if (op1->IsFPZero() ||
                op1->IsIntegralConst(0) || 
                (varTypeIsIntegral(simdTree->gtSIMDBaseType) && op1->IsIntegralConst(-1)))
            {
                MakeSrcContained(tree, tree->gtOp.gtOp1);
                info->srcCount = 0;
            }            
            else if ((comp->getSIMDInstructionSet() == InstructionSet_AVX) &&
                     ((simdTree->gtSIMDSize == 16) || (simdTree->gtSIMDSize == 32)))
            {
                // Either op1 is a float or dbl constant or an addr
                if (op1->IsCnsFltOrDbl() || op1->OperIsLocalAddr())
                {
                    MakeSrcContained(tree, tree->gtOp.gtOp1);
                    info->srcCount = 0;
                }
            }
        }
        break;

    case SIMDIntrinsicInitN:
        {
            info->srcCount = (short)(simdTree->gtSIMDSize / genTypeSize(simdTree->gtSIMDBaseType));

            // Need an internal register to stitch together all the values into a single vector in a SIMD reg
            info->internalFloatCount = 1;
            info->setInternalCandidates(lsra, lsra->allSIMDRegs());
        }
        break;

    case SIMDIntrinsicInitArray:
        // We have an array and an index, which may be contained.
        info->srcCount = 2;
        CheckImmedAndMakeContained(tree,  tree->gtGetOp2());
        break;

    case SIMDIntrinsicDiv:
        // SSE2 has no instruction support for division on integer vectors
        noway_assert(varTypeIsFloating(simdTree->gtSIMDBaseType));
        info->srcCount = 2;
        break;

    case SIMDIntrinsicAbs:
        // This gets implemented as bitwise-And operation with a mask
        // and hence should never see it here.
        unreached();
        break;

    case SIMDIntrinsicSqrt:
        // SSE2 has no instruction support for sqrt on integer vectors.
        noway_assert(varTypeIsFloating(simdTree->gtSIMDBaseType));
        info->srcCount = 1;
        break;

    case SIMDIntrinsicAdd:
    case SIMDIntrinsicSub:
    case SIMDIntrinsicMul:    
    case SIMDIntrinsicBitwiseAnd:
    case SIMDIntrinsicBitwiseAndNot:
    case SIMDIntrinsicBitwiseOr:
    case SIMDIntrinsicBitwiseXor:
    case SIMDIntrinsicMin:
    case SIMDIntrinsicMax:
        info->srcCount = 2;

        // SSE2 32-bit integer multiplication requires two temp regs
        if (simdTree->gtSIMDIntrinsicID == SIMDIntrinsicMul && 
            simdTree->gtSIMDBaseType == TYP_INT)
        {
            info->internalFloatCount = 2;
            info->setInternalCandidates(lsra, lsra->allSIMDRegs());
        }
        break;

    case SIMDIntrinsicEqual:
        info->srcCount = 2;
        break;

    // SSE2 doesn't support < and <= directly on int vectors.
    // Instead we need to use > and >= with swapped operands.
    case SIMDIntrinsicLessThan:
    case SIMDIntrinsicLessThanOrEqual:
        info->srcCount = 2;
        noway_assert(!varTypeIsIntegral(simdTree->gtSIMDBaseType));
        break;

    // SIMDIntrinsicEqual is supported only on non-floating point base type vectors.
    // SSE2 cmpps/pd doesn't support > and >=  directly on float/double vectors.
    // Instead we need to use <  and <= with swapped operands.
    case SIMDIntrinsicGreaterThan:
        noway_assert(!varTypeIsFloating(simdTree->gtSIMDBaseType));
        info->srcCount = 2;
        break;

    case SIMDIntrinsicOpEquality:
    case SIMDIntrinsicOpInEquality:
        // Need two SIMD registers as scratch.
        // See genSIMDIntrinsicRelOp() for details on code sequence generate and
        // the need for two scratch registers.
        info->srcCount = 2;
        info->internalFloatCount = 2;
        info->setInternalCandidates(lsra, lsra->allSIMDRegs());
        break;

    case SIMDIntrinsicDotProduct:
        if ((comp->getSIMDInstructionSet() == InstructionSet_SSE2) || (simdTree->gtOp.gtOp1->TypeGet() == TYP_SIMD32))
        {
            // For SSE, or AVX with 32-byte vectors, we also need an internal register as scratch.
            // Further we need the targetReg and internal reg to be distinct registers.
            // This is achieved by requesting two internal registers; thus one of them
            // will be different from targetReg.
            // Note that if this is a TYP_SIMD16 or smaller on AVX, then we don't need a tmpReg.
            //
            // See genSIMDIntrinsicDotProduct() for details on code sequence generated and
            // the need for scratch registers.
            info->internalFloatCount = 2;
            info->setInternalCandidates(lsra, lsra->allSIMDRegs());
        }
        info->srcCount = 2;
        break;

    case SIMDIntrinsicGetItem:
        // This implements get_Item method. The sources are:
        //  - the source SIMD struct
        //  - index (which element to get)
        // The result is baseType of SIMD struct.
        info->srcCount = 2;
        op2 = tree->gtOp.gtOp2;

        // If the index is a constant, mark it as contained.
        if (CheckImmedAndMakeContained(tree, op2))
        {
            info->srcCount = 1;
        }

        // If the index is not a constant, we will use the SIMD temp location to store the vector.
        // Otherwise, if the baseType is floating point, the targetReg will be a xmm reg and we
        // can use that in the process of extracting the element.
        // 
        // If the index is a constant and base type is a small int we can use pextrw, but on AVX
        // we will need a temp if are indexing into the upper half of the AVX register.
        // In all other cases with constant index, we need a temp xmm register to extract the 
        // element if index is other than zero.

        if (!op2->IsCnsIntOrI())
        {
            (void) comp->getSIMDInitTempVarNum();
        }
        else if (!varTypeIsFloating(simdTree->gtSIMDBaseType))
        {
            bool needFloatTemp;
            if (varTypeIsSmallInt(simdTree->gtSIMDBaseType) && (comp->getSIMDInstructionSet() == InstructionSet_AVX))
            {
                int byteShiftCnt = (int) op2->AsIntCon()->gtIconVal * genTypeSize(simdTree->gtSIMDBaseType);
                needFloatTemp = (byteShiftCnt >= 16);
            }
            else
            {
                needFloatTemp = !op2->IsIntegralConst(0);
            }
            if (needFloatTemp)
            {
                info->internalFloatCount = 1;
                info->setInternalCandidates(lsra, lsra->allSIMDRegs());
            }
        }
        break;

    case SIMDIntrinsicSetX:
    case SIMDIntrinsicSetY:
    case SIMDIntrinsicSetZ:
    case SIMDIntrinsicSetW:
        // We need an internal integer register
        info->srcCount = 2;
        info->internalIntCount = 1;
        info->setInternalCandidates(lsra, lsra->allRegs(TYP_INT));
        break;    

    case SIMDIntrinsicCast:
        info->srcCount = 1;
        break;

    case SIMDIntrinsicShuffleSSE2:
        info->srcCount = 2;
        // Second operand is an integer constant and marked as contained.
        op2 = tree->gtOp.gtOp2;
        noway_assert(op2->IsCnsIntOrI());
        MakeSrcContained(tree, op2);
        break;

    case SIMDIntrinsicGetX:
    case SIMDIntrinsicGetY:
    case SIMDIntrinsicGetZ:
    case SIMDIntrinsicGetW:
    case SIMDIntrinsicGetOne:
    case SIMDIntrinsicGetZero:
    case SIMDIntrinsicGetCount:
    case SIMDIntrinsicGetAllOnes:
        assert(!"Get intrinsics should not be seen during Lowering.");
        unreached();

    default:
        noway_assert(!"Unimplemented SIMD node type.");
        unreached();
    }
}
#endif // FEATURE_SIMD


//------------------------------------------------------------------------
// TreeNodeInfoInitCast: Set the NodeInfo for a GT_CAST.
//
// Arguments:
//    tree      - The node of interest
//
// Return Value:
//    None.
//
void
Lowering::TreeNodeInfoInitCast(GenTree* tree)
{
    TreeNodeInfo* info = &(tree->gtLsraInfo);

    // TODO-XArch-CQ: Int-To-Int conversions - castOp cannot be a memory op and must have an assigned register.
    //         see CodeGen::genIntToIntCast() 

    info->srcCount = 1;
    info->dstCount = 1;

    // Non-overflow casts to/from float/double are done using SSE2 instructions
    // and that allow the source operand to be either a reg or memop. Given the
    // fact that casts from small int to float/double are done as two-level casts, 
    // the source operand is always guaranteed to be of size 4 or 8 bytes.
    var_types castToType = tree->CastToType();
    GenTreePtr castOp    = tree->gtCast.CastOp();
    var_types castOpType = castOp->TypeGet();
    if (tree->gtFlags & GTF_UNSIGNED)
    {
        castOpType = genUnsignedType(castOpType);
    }

    if (!tree->gtOverflow() && (varTypeIsFloating(castToType) || varTypeIsFloating(castOpType)))
    {
#ifdef DEBUG
        // If converting to float/double, the operand must be 4 or 8 byte in size.
        if (varTypeIsFloating(castToType))
        {
            unsigned opSize = genTypeSize(castOpType);
            assert(opSize == 4 || opSize == 8);
        }
#endif //DEBUG

        // U8 -> R8 conversion requires that the operand be in a register.
        if (castOpType != TYP_ULONG)
        {
            if (castOp->isMemoryOp() || castOp->IsCnsNonZeroFltOrDbl())
            {
                MakeSrcContained(tree, castOp);
            }
            else
            {
                // Mark castOp as reg optional to indicate codegen
                // can still generate code if it is on stack.
                TryToSetRegOptional(castOp);
            }
        }
    }

#if !defined(_TARGET_64BIT_)
    if (varTypeIsLong(castOpType))
    {
        noway_assert(castOp->OperGet() == GT_LONG);
        info->srcCount = 2;
    }
#endif // !defined(_TARGET_64BIT_)

    // some overflow checks need a temp reg:
    //  - GT_CAST from INT64/UINT64 to UINT32
    if (tree->gtOverflow() && (castToType == TYP_UINT))
    {
        if (genTypeSize(castOpType) == 8)
        {
            info->internalIntCount = 1;
        }
    }
}


void Lowering::LowerGCWriteBarrier(GenTree* tree)
{
    assert(tree->OperGet() == GT_STOREIND);

    GenTreeStoreInd* dst = tree->AsStoreInd();
    GenTreePtr addr = dst->Addr();
    GenTreePtr src  = dst->Data();

    if (addr->OperGet() == GT_LEA)
    {
        // In the case where we are doing a helper assignment, if the dst
        // is an indir through an lea, we need to actually instantiate the
        // lea in a register
        GenTreeAddrMode* lea = addr->AsAddrMode();

        int leaSrcCount = 0;
        if (lea->HasBase())
        {
            leaSrcCount++;
        }
        if (lea->HasIndex())
        {
            leaSrcCount++;
        }
        lea->gtLsraInfo.srcCount = leaSrcCount;
        lea->gtLsraInfo.dstCount = 1;
    }

    bool useOptimizedWriteBarrierHelper = false; // By default, assume no optimized write barriers.

#if NOGC_WRITE_BARRIERS

#if defined(_TARGET_X86_)

    useOptimizedWriteBarrierHelper = true;  // On x86, use the optimized write barriers by default.
#ifdef DEBUG
    GCInfo::WriteBarrierForm wbf = comp->codeGen->gcInfo.gcIsWriteBarrierCandidate(tree, src);
    if (wbf == GCInfo::WBF_NoBarrier_CheckNotHeapInDebug) // This one is always a call to a C++ method.
    {
        useOptimizedWriteBarrierHelper = false;
    }
#endif

    if (useOptimizedWriteBarrierHelper)
    {
        // Special write barrier:
        // op1 (addr) goes into REG_WRITE_BARRIER (rdx) and
        // op2 (src) goes into any int register.
        addr->gtLsraInfo.setSrcCandidates(m_lsra, RBM_WRITE_BARRIER);
        src->gtLsraInfo.setSrcCandidates(m_lsra, RBM_WRITE_BARRIER_SRC);
    }

#else // !defined(_TARGET_X86_)
#error "NOGC_WRITE_BARRIERS is not supported"
#endif // !defined(_TARGET_X86_)

#endif // NOGC_WRITE_BARRIERS

    if (!useOptimizedWriteBarrierHelper)
    {
        // For the standard JIT Helper calls:
        // op1 (addr) goes into REG_ARG_0 and
        // op2 (src) goes into REG_ARG_1
        addr->gtLsraInfo.setSrcCandidates(m_lsra, RBM_ARG_0);
        src->gtLsraInfo.setSrcCandidates(m_lsra, RBM_ARG_1);
    }

    // Both src and dst must reside in a register, which they should since we haven't set
    // either of them as contained.
    assert(addr->gtLsraInfo.dstCount == 1);
    assert(src->gtLsraInfo.dstCount == 1);
}

//-----------------------------------------------------------------------------------------
// Specify register requirements for address expression of an indirection operation.
//
// Arguments:
//    indirTree    -   GT_IND or GT_STOREIND gentree node
//
void Lowering::SetIndirAddrOpCounts(GenTreePtr indirTree)
{
    assert(indirTree->isIndir());

    GenTreePtr addr = indirTree->gtGetOp1();
    TreeNodeInfo* info = &(indirTree->gtLsraInfo);

    GenTreePtr base = nullptr;
    GenTreePtr index = nullptr;
    unsigned mul, cns;
    bool rev;
    bool modifiedSources = false;    

    // If indirTree is of TYP_SIMD12, don't mark addr as contained
    // so that it always get computed to a register.  This would
    // mean codegen side logic doesn't need to handle all possible
    // addr expressions that could be contained.
    // 
    // TODO-XArch-CQ: handle other addr mode expressions that could be marked
    // as contained.
#ifdef FEATURE_SIMD
    if (indirTree->TypeGet() == TYP_SIMD12)
    {
        // Vector3 is read/written as two reads/writes: 8 byte and 4 byte.
        // To assemble the vector properly we would need an additional
        // XMM register.
        info->internalFloatCount = 1;

        // In case of GT_IND we need an internal register different from targetReg and
        // both of the registers are used at the same time. This achieved by reserving
        // two internal registers
        if (indirTree->OperGet() == GT_IND)
        {
            (info->internalFloatCount)++;
        }

        info->setInternalCandidates(m_lsra, m_lsra->allSIMDRegs());

        return ;
    }
#endif //FEATURE_SIMD

    // These nodes go into an addr mode:
    // - GT_CLS_VAR_ADDR turns into a constant.
    // - GT_LCL_VAR_ADDR is a stack addr mode.
    if ((addr->OperGet() == GT_CLS_VAR_ADDR) || (addr->OperGet() == GT_LCL_VAR_ADDR))
    {
        // make this contained, it turns into a constant that goes into an addr mode
        MakeSrcContained(indirTree, addr);
    }
    else if (addr->IsCnsIntOrI() &&
             addr->AsIntConCommon()->FitsInAddrBase(comp) &&
             addr->gtLsraInfo.getDstCandidates(m_lsra) != RBM_VIRTUAL_STUB_PARAM)
    {
        // Amd64:
        // We can mark any pc-relative 32-bit addr as containable, except for a direct VSD call address.
        // (i.e. those VSD calls for which stub addr is known during JIT compilation time).  In this case,
        // VM requires us to pass stub addr in REG_VIRTUAL_STUB_PARAM - see LowerVirtualStubCall().  For
        // that reason we cannot mark such an addr as contained.  Note that this is not an issue for
        // indirect VSD calls since morphArgs() is explicitly materializing hidden param as a non-standard
        // argument.
        //
        // Workaround:
        // Note that LowerVirtualStubCall() sets addr->gtRegNum to REG_VIRTUAL_STUB_PARAM and Lowering::doPhase()
        // sets destination candidates on such nodes and resets addr->gtRegNum to REG_NA before calling
        // TreeNodeInfoInit(). Ideally we should set a flag on addr nodes that shouldn't be marked as contained
        // (in LowerVirtualStubCall()), but we don't have any GTF_* flags left for that purpose.  As a workaround
        // an explicit check is made here.
        //
        // On x86, direct VSD is done via a relative branch, and in fact it MUST be contained.
        MakeSrcContained(indirTree, addr);
    }
    else if (addr->OperGet() == GT_LEA)
    {
        GenTreeAddrMode* lea = addr->AsAddrMode();
        base  = lea->Base();
        index = lea->Index();

        m_lsra->clearOperandCounts(addr);
        // The srcCount is decremented because addr is now "contained", 
        // then we account for the base and index below, if they are non-null.    
        info->srcCount--;
    }
    else if (comp->codeGen->genCreateAddrMode(addr, -1, true, 0, &rev, &base, &index, &mul, &cns, true /*nogen*/)
             && !(modifiedSources = AreSourcesPossiblyModified(indirTree, base, index)))
    {
        // An addressing mode will be constructed that may cause some
        // nodes to not need a register, and cause others' lifetimes to be extended
        // to the GT_IND or even its parent if it's an assignment

        assert(base != addr);
        m_lsra->clearOperandCounts(addr);

        GenTreePtr arrLength = nullptr;

        // Traverse the computation below GT_IND to find the operands
        // for the addressing mode, marking the various constants and
        // intermediate results as not consuming/producing.
        // If the traversal were more complex, we might consider using
        // a traversal function, but the addressing mode is only made
        // up of simple arithmetic operators, and the code generator
        // only traverses one leg of each node.

        bool foundBase = (base == nullptr);
        bool foundIndex = (index == nullptr);
        GenTreePtr nextChild = nullptr;
        for (GenTreePtr child = addr;
             child != nullptr && !child->OperIsLeaf();
             child = nextChild)
        {
            nextChild = nullptr;
            GenTreePtr op1 = child->gtOp.gtOp1;
            GenTreePtr op2 = (child->OperIsBinary()) ? child->gtOp.gtOp2 : nullptr;

            if (op1 == base)
            {
                foundBase = true;
            }
            else if (op1 == index)
            {
                foundIndex = true;
            }
            else
            {
                m_lsra->clearOperandCounts(op1);
                if (!op1->OperIsLeaf())
                {
                    nextChild = op1;
                }
            }

            if (op2 != nullptr)
            {
                if (op2 == base)
                {
                    foundBase = true;
                }
                else if (op2 == index)
                {
                    foundIndex = true;
                }
                else
                {
                    m_lsra->clearOperandCounts(op2);
                    if (!op2->OperIsLeaf())
                    {
                        assert(nextChild == nullptr);
                        nextChild = op2;
                    }
                }
            }
        }
        assert(foundBase && foundIndex);
        info->srcCount--; // it gets incremented below.
    }
    else if (addr->gtOper == GT_ARR_ELEM)
    {
        // The GT_ARR_ELEM consumes all the indices and produces the offset.
        // The array object lives until the mem access.
        // We also consume the target register to which the address is
        // computed

        info->srcCount++;
        assert(addr->gtLsraInfo.srcCount >= 2);
        addr->gtLsraInfo.srcCount -= 1;
    }
    else
    {
        // it is nothing but a plain indir
        info->srcCount--; //base gets added in below
        base = addr;
    }

    if (base != nullptr)
    {
        info->srcCount++;
    }

    if (index != nullptr && !modifiedSources)
    {
        info->srcCount++;
    }
}
            

void Lowering::LowerCmp(GenTreePtr tree)
{
    TreeNodeInfo* info = &(tree->gtLsraInfo);
    
    info->srcCount = 2;
    info->dstCount = 1;

#ifdef _TARGET_X86_
    info->setDstCandidates(m_lsra, RBM_BYTE_REGS);
#endif // _TARGET_X86_

    GenTreePtr op1 = tree->gtOp.gtOp1;
    GenTreePtr op2 = tree->gtOp.gtOp2;
    var_types op1Type = op1->TypeGet();
    var_types op2Type = op2->TypeGet();

#if !defined(_TARGET_64BIT_)
    // Long compares will consume GT_LONG nodes, each of which produces two results.
    // Thus for each long operand there will be an additional source.
    // TODO-X86-CQ: Mark hiOp2 and loOp2 as contained if it is a constant or a memory op.
    if (varTypeIsLong(op1Type))
    {
        info->srcCount++;
    }
    if (varTypeIsLong(op2Type))
    {
        info->srcCount++;
    }
#endif // !defined(_TARGET_64BIT_)

    // If either of op1 or op2 is floating point values, then we need to use
    // ucomiss or ucomisd to compare, both of which support the following form
    // ucomis[s|d] xmm, xmm/mem.  That is only the second operand can be a memory
    // op.  
    //
    // Second operand is a memory Op:  Note that depending on comparison operator,
    // the operands of ucomis[s|d] need to be reversed.  Therefore, either op1 or 
    // op2 can be a memory op depending on the comparison operator.
    if (varTypeIsFloating(op1Type))
    {        
        // The type of the operands has to be the same and no implicit conversions at this stage.
        assert(op1Type == op2Type);

        bool reverseOps;
        if ((tree->gtFlags & GTF_RELOP_NAN_UN) != 0)
        {
            // Unordered comparison case
            reverseOps = (tree->gtOper == GT_GT || tree->gtOper == GT_GE);
        }
        else
        {
            reverseOps = (tree->gtOper == GT_LT || tree->gtOper == GT_LE);
        }

        GenTreePtr otherOp;
        if (reverseOps)
        {
            otherOp = op1;
        }
        else
        {
            otherOp = op2;
        }

        assert(otherOp != nullptr);
        if (otherOp->IsCnsNonZeroFltOrDbl())
        {
            MakeSrcContained(tree, otherOp);
        }
        else if (otherOp->isMemoryOp())
        {
            if ((otherOp == op2) || IsSafeToContainMem(tree, otherOp)) 
            {
                MakeSrcContained(tree, otherOp);
            }
        }
        else
        {
            // Mark otherOp as reg optional to indicate codgen can still generate
            // code even if otherOp is on stack.
            TryToSetRegOptional(otherOp);
        }

        return;
    }

    // TODO-XArch-CQ: factor out cmp optimization in 'genCondSetFlags' to be used here
    // or in other backend.
    
    bool hasShortCast = false;
    if (CheckImmedAndMakeContained(tree, op2))
    {
        bool op1CanBeContained  = (op1Type == op2Type);
        if (!op1CanBeContained)
        {
            if (genTypeSize(op1Type) == genTypeSize(op2Type))          
            {
                // The constant is of the correct size, but we don't have an exact type match
                // We can treat the isMemoryOp as "contained"
                op1CanBeContained = true;
            }
        }

        // Do we have a short compare against a constant in op2
        //
        if (varTypeIsSmall(op1Type))
        {
            GenTreeIntCon* con  = op2->AsIntCon();
            ssize_t        ival = con->gtIconVal;

            bool    isEqualityCompare = (tree->gtOper == GT_EQ || tree->gtOper == GT_NE);
            bool    useTest           = isEqualityCompare && (ival == 0);

            if (!useTest)
            {
                ssize_t lo = 0;  // minimum imm value allowed for cmp reg,imm
                ssize_t hi = 0;  // maximum imm value allowed for cmp reg,imm
                bool    isUnsigned = false;

                switch (op1Type) {
                case TYP_BOOL:
                    op1Type = TYP_UBYTE;
                    __fallthrough;
                case TYP_UBYTE:
                    lo = 0;
                    hi = 0x7f;
                    isUnsigned = true;
                    break;
                case TYP_BYTE:
                    lo = -0x80;
                    hi =  0x7f;
                    break;
                case TYP_CHAR:
                    lo = 0;
                    hi = 0x7fff;
                    isUnsigned = true;
                    break;
                case TYP_SHORT:
                    lo = -0x8000;
                    hi =  0x7fff;
                    break;
                default:
                    unreached();
                }

                if ((ival >= lo) && (ival <= hi))
                {
                    // We can perform a small compare with the immediate 'ival'
                    tree->gtFlags |= GTF_RELOP_SMALL;
                    if (isUnsigned && !isEqualityCompare)
                    {
                        tree->gtFlags |= GTF_UNSIGNED;
                    }
                    // We can treat the isMemoryOp as "contained"
                    op1CanBeContained = true;
                }
            }
        }
        if (op1CanBeContained)
        {
            if (op1->isMemoryOp())
            {
                MakeSrcContained(tree, op1);
            }
            else 
            {
                // When op1 is a GT_AND we can often generate a single "test" instruction
                // instead of two instructions (an "and" instruction followed by a "cmp"/"test")
                //
                // This instruction can only be used for equality or inequality comparions.
                // and we must have a compare against zero.
                //
                // If we have a postive test for a single bit we can reverse the condition and 
                // make the compare be against zero
                //
                // Example:
                //                  GT_EQ                              GT_NE
                //                  /   \                              /   \
                //             GT_AND   GT_CNS (0x100)  ==>>      GT_AND   GT_CNS (0)
                //             /    \                             /    \
                //          andOp1  GT_CNS (0x100)             andOp1  GT_CNS (0x100)
                //
                // We will mark the GT_AND node as contained if the tree is a equality compare with zero
                // Additionally when we do this we also allow for a contained memory operand for "andOp1".
                //
                bool isEqualityCompare = (tree->gtOper == GT_EQ || tree->gtOper == GT_NE);

                if (isEqualityCompare && (op1->OperGet() == GT_AND))
                {
                    GenTreePtr andOp2 = op1->gtOp.gtOp2;
                    if (IsContainableImmed(op1, andOp2))
                    {
                        ssize_t andOp2CnsVal = andOp2->AsIntConCommon()->IconValue();
                        ssize_t relOp2CnsVal = op2->AsIntConCommon()->IconValue();

                        if ((relOp2CnsVal == andOp2CnsVal) && isPow2(andOp2CnsVal))
                        {
                            // We have a single bit test, so now we can change the
                            // tree into the alternative form, 
                            // so that we can generate a test instruction.

                            // Reverse the equality comparison
                            tree->gtOper = (tree->gtOper == GT_EQ) ? GT_NE : GT_EQ;

                            // Change the relOp2CnsVal to zero
                            relOp2CnsVal = 0;
                            op2->AsIntConCommon()->SetIconValue(0);
                        }

                        // Now do we have a equality compare with zero?
                        //
                        if (relOp2CnsVal == 0)
                        {
                            // Note that child nodes must be made contained before parent nodes

                            // Check for a memory operand for op1 with the test instruction
                            //
                            GenTreePtr andOp1 = op1->gtOp.gtOp1;
                            if (andOp1->isMemoryOp())
                            {
                                // If the type of value memoryOp (andOp1) is not the same as the type of constant (andOp2)
                                // check to see whether it is safe to mark AndOp1 as contained.  For e.g. in the following
                                // case it is not safe to mark andOp1 as contained
                                //    AndOp1 = signed byte and andOp2 is an int constant of value 512.
                                //
                                // If it is safe, we update the type and value of andOp2 to match with andOp1.
                                bool containable = (andOp1->TypeGet() == op1->TypeGet());
                                if (!containable)
                                {
                                    ssize_t newIconVal = 0;

                                    switch (andOp1->TypeGet())
                                    {
                                    default:
                                        break;
                                    case TYP_BYTE:
                                        newIconVal = (signed char)andOp2CnsVal; 
                                        containable = FitsIn<signed char>(andOp2CnsVal);
                                        break;
                                    case TYP_BOOL:
                                    case TYP_UBYTE:
                                        newIconVal = andOp2CnsVal & 0xFF; 
                                        containable = true;
                                        break;
                                    case TYP_SHORT:
                                        newIconVal = (signed short)andOp2CnsVal; 
                                        containable = FitsIn<signed short>(andOp2CnsVal);
                                        break;
                                    case TYP_CHAR:
                                        newIconVal = andOp2CnsVal & 0xFFFF;
                                        containable = true;
                                        break;
                                    case TYP_INT:
                                        newIconVal = (INT32)andOp2CnsVal;
                                        containable = FitsIn<INT32>(andOp2CnsVal);
                                        break;
                                    case TYP_UINT:
                                        newIconVal = andOp2CnsVal & 0xFFFFFFFF;
                                        containable = true;
                                        break;

#ifdef _TARGET_64BIT_
                                    case TYP_LONG:
                                        newIconVal = (INT64)andOp2CnsVal; 
                                        containable = true;
                                        break;
                                    case TYP_ULONG:
                                        newIconVal = (UINT64)andOp2CnsVal; 
                                        containable = true;
                                        break;
#endif //_TARGET_64BIT_
                                    }


                                    if (containable)
                                    {
                                        andOp2->gtType = andOp1->TypeGet();
                                        andOp2->AsIntConCommon()->SetIconValue(newIconVal);
                                    }
                                }

                                // Mark the 'andOp1' memory operand as contained
                                // Note that for equality comparisons we don't need
                                // to deal with any signed or unsigned issues.
                                if (containable)
                                {
                                    MakeSrcContained(op1, andOp1);
                                }
                            }
                            // Mark the 'op1' (the GT_AND) operand as contained
                            MakeSrcContained(tree, op1);

                            // During Codegen we will now generate "test andOp1, andOp2CnsVal"
                        }
                    }
                }
                else if (op1->OperGet() == GT_CAST)
                {
                    //If the op1 is a cast operation, and cast type is one byte sized unsigned type, 
                    //we can directly use the number in register, instead of doing an extra cast step.
                    var_types   dstType       = op1->CastToType();
                    bool        isUnsignedDst = varTypeIsUnsigned(dstType);
                    emitAttr    castSize      = EA_ATTR(genTypeSize(dstType));
                    GenTreePtr  castOp1       = op1->gtOp.gtOp1;
                    genTreeOps  castOp1Oper   = castOp1->OperGet();
                    bool        safeOper      = false;

                    // It is not always safe to change the gtType of 'castOp1' to TYP_UBYTE
                    // For example when 'castOp1Oper' is a GT_RSZ or GT_RSH then we are shifting
                    // bits from the left into the lower bits.  If we change the type to a TYP_UBYTE
                    // we will instead generate a byte sized shift operation:  shr  al, 24
                    // For the following ALU operations is it safe to change the gtType to the
                    // smaller type:   
                    //
                    if ((castOp1Oper == GT_CNS_INT) || 
                        (castOp1Oper == GT_CALL)    ||    // the return value from a Call
                        (castOp1Oper == GT_LCL_VAR) ||
                        castOp1->OperIsLogical()    ||    // GT_AND, GT_OR, GT_XOR
                        castOp1->isMemoryOp()         )   // isIndir() || isLclField();
                    {
                        safeOper = true;
                    }

                    if ((castSize == EA_1BYTE) && isUnsignedDst &&    // Unsigned cast to TYP_UBYTE
                        safeOper &&                                   // Must be a safe operation
                        !op1->gtOverflow()                         )  // Must not be an overflow checking cast
                    {
                        // Currently all of the Oper accepted as 'safeOper' are 
                        // non-overflow checking operations.  If we were to add 
                        // an overflow checking operation then this assert needs 
                        // to be moved above to guard entry to this block.
                        // 
                        assert(!castOp1->gtOverflowEx());             // Must not be an overflow checking operation
                        
                        GenTreePtr removeTreeNode = op1;
                        GenTreePtr removeTreeNodeChild = castOp1;
                        tree->gtOp.gtOp1 = castOp1;
                        castOp1->gtType = TYP_UBYTE;

                        // trim down the value if castOp1 is an int constant since its type changed to UBYTE.
                        if (castOp1Oper == GT_CNS_INT)
                        {                            
                            castOp1->gtIntCon.gtIconVal = (UINT8)castOp1->gtIntCon.gtIconVal;
                        }

                        if (op2->isContainedIntOrIImmed())
                        {
                            ssize_t val = (ssize_t)op2->AsIntConCommon()->IconValue();
                            if (val >= 0 && val <= 255)
                            {
                                op2->gtType    = TYP_UBYTE;
                                tree->gtFlags |= GTF_UNSIGNED;
                                
                                //right now the op1's type is the same as op2's type.
                                //if op1 is MemoryOp, we should make the op1 as contained node.
                                if (castOp1->isMemoryOp())
                                {
                                    MakeSrcContained(tree, op1);
                                }
                            }
                        }
                        comp->fgSnipNode(comp->compCurStmt->AsStmt(), removeTreeNode);
#ifdef DEBUG
                        if (comp->verbose)
                        {
                            printf("LowerCmp: Removing a GT_CAST to TYP_UBYTE and changing castOp1->gtType to TYP_UBYTE\n");
                            comp->gtDispTree(tree);
                        }
#endif
                    }
                }
            }
        }
    }
	else if (op1Type == op2Type)
	{
		if (op2->isMemoryOp())
		{
			MakeSrcContained(tree, op2);
		}
		else if (op1->isMemoryOp() && IsSafeToContainMem(tree, op1))
		{
			MakeSrcContained(tree, op1);
		}
        else
        {
            // One of op1 or op2 could be marked as reg optional
            // to indicate that codgen can still generate code 
            // if one of them is on stack.
            TryToSetRegOptional(op2);

            if (!op2->IsRegOptional())
            {
                TryToSetRegOptional(op1);
            }
        }

		if (varTypeIsSmall(op1Type) && varTypeIsUnsigned(op1Type))
		{
			// Mark the tree as doing unsigned comparison if
			// both the operands are small and unsigned types.
			// Otherwise we will end up performing a signed comparison
			// of two small unsigned values without zero extending them to
			// TYP_INT size and which is incorrect.
			tree->gtFlags |= GTF_UNSIGNED;
		}
	}
}

/* Lower GT_CAST(srcType, DstType) nodes. 
 *
 * Casts from small int type to float/double are transformed as follows:
 * GT_CAST(byte, float/double)     =   GT_CAST(GT_CAST(byte, int32), float/double)
 * GT_CAST(sbyte, float/double)    =   GT_CAST(GT_CAST(sbyte, int32), float/double)
 * GT_CAST(int16, float/double)    =   GT_CAST(GT_CAST(int16, int32), float/double)
 * GT_CAST(uint16, float/double)   =   GT_CAST(GT_CAST(uint16, int32), float/double)
 *
 * SSE2 conversion instructions operate on signed integers. casts from Uint32/Uint64 
 * are morphed as follows by front-end and hence should not be seen here.
 * GT_CAST(uint32, float/double)   =   GT_CAST(GT_CAST(uint32, long), float/double)
 * GT_CAST(uint64, float)          =   GT_CAST(GT_CAST(uint64, double), float)
 *
 *
 * Similarly casts from float/double to a smaller int type are transformed as follows:
 * GT_CAST(float/double, byte)     =   GT_CAST(GT_CAST(float/double, int32), byte)
 * GT_CAST(float/double, sbyte)    =   GT_CAST(GT_CAST(float/double, int32), sbyte)
 * GT_CAST(float/double, int16)    =   GT_CAST(GT_CAST(double/double, int32), int16)
 * GT_CAST(float/double, uint16)   =   GT_CAST(GT_CAST(double/double, int32), uint16)
 *
 * SSE2 has instructions to convert a float/double vlaue into a signed 32/64-bit
 * integer.  The above transformations help us to leverage those instructions.
 * 
 * Note that for the following conversions we still depend on helper calls and
 * don't expect to see them here. 
 *  i) GT_CAST(float/double, uint64)
 * ii) GT_CAST(float/double, int type with overflow detection) 
 *
 * TODO-XArch-CQ: (Low-pri): Jit64 generates in-line code of 8 instructions for (i) above.
 * There are hardly any occurrences of this conversion operation in platform
 * assemblies or in CQ perf benchmarks (1 occurrence in mscorlib, microsoft.jscript,
 * 1 occurence in Roslyn and no occurrences in system, system.core, system.numerics
 * system.windows.forms, scimark, fractals, bio mums). If we ever find evidence that
 * doing this optimization is a win, should consider generating in-lined code.
 */
void Lowering::LowerCast( GenTreePtr* ppTree) 
{
    GenTreePtr  tree    = *ppTree;
    assert(tree->OperGet() == GT_CAST);

    GenTreePtr  op1     = tree->gtOp.gtOp1;
    var_types   dstType = tree->CastToType();
    var_types   srcType = op1->TypeGet();
    var_types   tmpType = TYP_UNDEF;
    bool        srcUns  = false;

    // force the srcType to unsigned if GT_UNSIGNED flag is set
    if (tree->gtFlags & GTF_UNSIGNED)
    {
        srcType = genUnsignedType(srcType);
    }

    // We should never see the following casts as they are expected to be lowered 
    // apropriately or converted into helper calls by front-end.
    //   srcType = float/double                    dstType = * and overflow detecting cast
    //       Reason: must be converted to a helper call
    //   srcType = float/double,                   dstType = ulong
    //       Reason: must be converted to a helper call
    //   srcType = uint                            dstType = float/double
    //       Reason: uint -> float/double = uint -> long -> float/double
    //   srcType = ulong                           dstType = float
    //       Reason: ulong -> float = ulong -> double -> float
    if (varTypeIsFloating(srcType))
    {
        noway_assert(!tree->gtOverflow());
        noway_assert(dstType != TYP_ULONG);
    }
    else if (srcType == TYP_UINT)
    {
        noway_assert(!varTypeIsFloating(dstType));
    }
    else if (srcType == TYP_ULONG)
    {
        noway_assert(dstType != TYP_FLOAT);
    }

    // Case of src is a small type and dst is a floating point type.
    if (varTypeIsSmall(srcType) && varTypeIsFloating(dstType))
    {
        // These conversions can never be overflow detecting ones.
        noway_assert(!tree->gtOverflow());
        tmpType = TYP_INT;
    }
    // case of src is a floating point type and dst is a small type.
    else if (varTypeIsFloating(srcType) && varTypeIsSmall(dstType))
    {
        tmpType = TYP_INT;
    }

    if (tmpType != TYP_UNDEF)
    {
        GenTreePtr tmp = comp->gtNewCastNode(tmpType, op1, tmpType);
        tmp->gtFlags |= (tree->gtFlags & (GTF_UNSIGNED|GTF_OVERFLOW|GTF_EXCEPT));

        tree->gtFlags &= ~GTF_UNSIGNED;
        tree->gtOp.gtOp1 = tmp;
        op1->InsertAfterSelf(tmp);
    }
}

 //----------------------------------------------------------------------------------------------
 // Returns true if this tree is bin-op of a GT_STOREIND of the following form
 //      storeInd(subTreeA, binOp(gtInd(subTreeA), subtreeB)) or
 //      storeInd(subTreeA, binOp(subtreeB, gtInd(subTreeA)) in case of commutative bin-ops
 // 
 // The above form for storeInd represents a read-modify-write memory binary operation.
 //
 // Parameters
 //     tree   -   GentreePtr of binOp
 //
 // Return Value 
 //     True if 'tree' is part of a RMW memory operation pattern
 //
bool Lowering::IsBinOpInRMWStoreInd(GenTreePtr tree)
{
    // Must be a non floating-point type binary operator since SSE2 doesn't support RMW memory ops
    assert(!varTypeIsFloating(tree));
    assert(GenTree::OperIsBinary(tree->OperGet()));

    // Cheap bail out check before more expensive checks are performed.
    // RMW memory op pattern requires that one of the operands of binOp to be GT_IND.
    if (tree->gtGetOp1()->OperGet() != GT_IND && tree->gtGetOp2()->OperGet() != GT_IND)
    {
        return false;
    }

    GenTreePtr parent = tree->gtGetParent(nullptr);
    if (parent == nullptr || parent->OperGet() != GT_STOREIND || parent->gtGetOp2() != tree)
    {
        return false;
    }

    // Since it is not relatively cheap to recognize RMW memory op pattern, we
    // cache the result in GT_STOREIND node so that while lowering GT_STOREIND
    // we can use the result.
    GenTreePtr indirCandidate = nullptr;
    GenTreePtr indirOpSource = nullptr;
    return IsRMWMemOpRootedAtStoreInd(parent, &indirCandidate, &indirOpSource);
}
 
 //----------------------------------------------------------------------------------------------
 // This method recognizes the case where we have a treeNode with the following structure:
 //         storeInd(IndirDst, binOp(gtInd(IndirDst), indirOpSource)) OR
 //         storeInd(IndirDst, binOp(indirOpSource, gtInd(IndirDst)) in case of commutative operations OR
 //         storeInd(IndirDst, unaryOp(gtInd(IndirDst)) in case of unary operations
 //
 // Terminology:
 //         indirDst = memory write of an addr mode  (i.e. storeind destination)
 //         indirSrc = value being written to memory (i.e. storeind source which could either be a binary or unary op)
 //         indirCandidate = memory read i.e. a gtInd of an addr mode
 //         indirOpSource = source operand used in binary/unary op (i.e. source operand of indirSrc node)
 //
 // In x86/x64 this storeInd pattern can be effectively encoded in a single instruction of the
 // following form in case of integer operations:
 //         binOp [addressing mode], RegIndirOpSource
 //         binOp [addressing mode], immediateVal
 // where RegIndirOpSource is the register where indirOpSource was computed.
 //
 // Right now, we recognize few cases:
 //     a) The gtInd child is a lea/lclVar/lclVarAddr/clsVarAddr/constant
 //     b) BinOp is either add, sub, xor, or, and, shl, rsh, rsz.
 //     c) unaryOp is either not/neg
 //
 // Implementation Note: The following routines need to be in sync for RMW memory op optimization
 // to be correct and functional.
 //     IndirsAreEquivalent()
 //     NodesAreEquivalentLeaves()
 //     Codegen of GT_STOREIND and genCodeForShiftRMW()
 //     emitInsRMW()
 //     
 //  TODO-CQ: Enable support for more complex indirections (if needed) or use the value numbering
 //  package to perform more complex tree recognition.
 //
 //  TODO-XArch-CQ: Add support for RMW of lcl fields (e.g. lclfield binop= source)
 //
 //  Parameters:
 //     tree               -  GT_STOREIND node
 //     outIndirCandidate  -  out param set to indirCandidate as described above
 //     ouutIndirOpSource  -  out param set to indirOpSource as described above
 //
 //  Return value
 //     True if there is a RMW memory operation rooted at a GT_STOREIND tree
 //     and out params indirCandidate and indirOpSource are set to non-null values.
 //     Otherwise, returns false with indirCandidate and indirOpSource set to null.
 //     Also updates flags of GT_STOREIND tree with its RMW status.
 //
bool Lowering::IsRMWMemOpRootedAtStoreInd(GenTreePtr tree, GenTreePtr *outIndirCandidate, GenTreePtr *outIndirOpSource)
{
    assert(!varTypeIsFloating(tree));
    assert(outIndirCandidate != nullptr);
    assert(outIndirOpSource != nullptr);
     
    *outIndirCandidate = nullptr;
    *outIndirOpSource = nullptr;

    // Early out if storeInd is already known to be a non-RMW memory op
    GenTreeStoreInd* storeInd = tree->AsStoreInd();
    if (storeInd->IsNonRMWMemoryOp())
    {
        return false;
    }

    GenTreePtr indirDst = storeInd->gtGetOp1();
    GenTreePtr indirSrc = storeInd->gtGetOp2();
    genTreeOps oper = indirSrc->OperGet();

    // Early out if it is already known to be a RMW memory op
    if (storeInd->IsRMWMemoryOp())
    {
        if (GenTree::OperIsBinary(oper))
        {
            if (storeInd->IsRMWDstOp1())
            {
                *outIndirCandidate = indirSrc->gtGetOp1();
                *outIndirOpSource = indirSrc->gtGetOp2();
            }
            else
            {
                assert(storeInd->IsRMWDstOp2());
                *outIndirCandidate = indirSrc->gtGetOp2();
                *outIndirOpSource = indirSrc->gtGetOp1();
            }
            assert(IndirsAreEquivalent(*outIndirCandidate, storeInd));
        }
        else
        {
            assert(GenTree::OperIsUnary(oper));
            assert(IndirsAreEquivalent(indirSrc->gtGetOp1(), storeInd));
            *outIndirCandidate = indirSrc->gtGetOp1();
            *outIndirOpSource = indirSrc->gtGetOp1();
        }

        return true;
    }

    // If reached here means that we do not know RMW status of tree rooted at storeInd
    assert(storeInd->IsRMWStatusUnknown());

    // Early out if indirDst is not one of the supported memory operands.
    if (indirDst->OperGet() != GT_LEA &&
        indirDst->OperGet() != GT_LCL_VAR &&
        indirDst->OperGet() != GT_LCL_VAR_ADDR &&
        indirDst->OperGet() != GT_CLS_VAR_ADDR &&
        indirDst->OperGet() != GT_CNS_INT)
    {
        storeInd->SetRMWStatus(STOREIND_RMW_UNSUPPORTED_ADDR);
        return false;
    }
    
    // We can not use Read-Modify-Write instruction forms with overflow checking instructions
    // because we are not allowed to modify the target until after the overflow check. 
    if (indirSrc->gtOverflowEx())
    {
        storeInd->SetRMWStatus(STOREIND_RMW_UNSUPPORTED_OPER);
        return false;
    }

    if (GenTree::OperIsBinary(oper))
    {
        // Return if binary op is not one of the supported operations for RMW of memory.
        if (oper != GT_ADD &&
            oper != GT_SUB &&
            oper != GT_AND &&
            oper != GT_OR  &&
            oper != GT_XOR &&
            !GenTree::OperIsShiftOrRotate(oper))
        {
            storeInd->SetRMWStatus(STOREIND_RMW_UNSUPPORTED_OPER);
            return false;
        }

        if (GenTree::OperIsShiftOrRotate(oper) &&
            varTypeIsSmall(storeInd))
        {
            // In ldind, Integer values smaller than 4 bytes, a boolean, or a character converted to 4 bytes
            // by sign or zero-extension as appropriate. If we directly shift the short type data using sar, we
            // will lose the sign or zero-extension bits.
            storeInd->SetRMWStatus(STOREIND_RMW_UNSUPPORTED_TYPE);
            return false;
        }

        GenTreePtr rhsLeft = indirSrc->gtGetOp1();
        GenTreePtr rhsRight = indirSrc->gtGetOp2();

        // The most common case is rhsRight is GT_IND
        if (GenTree::OperIsCommutative(oper) &&
            rhsRight->OperGet() == GT_IND &&
            rhsRight->gtGetOp1()->OperGet() == indirDst->OperGet() &&
            IndirsAreEquivalent(rhsRight, storeInd))
        {
            *outIndirCandidate = rhsRight;
            *outIndirOpSource = rhsLeft;
            storeInd->SetRMWStatus(STOREIND_RMW_DST_IS_OP2);
            return true;
        }
        else if (rhsLeft->OperGet() == GT_IND &&
                 rhsLeft->gtGetOp1()->OperGet() == indirDst->OperGet() &&
                 IsSafeToContainMem(indirSrc, rhsLeft) &&
                 IndirsAreEquivalent(rhsLeft, storeInd))
        {
            *outIndirCandidate = rhsLeft;
            *outIndirOpSource = rhsRight;
            storeInd->SetRMWStatus(STOREIND_RMW_DST_IS_OP1);
            return true;
        }

        storeInd->SetRMWStatus(STOREIND_RMW_UNSUPPORTED_ADDR);
        return false;
    }
    else if (GenTree::OperIsUnary(oper))
    {
        // Nodes other than GT_NOT and GT_NEG are not yet supported.    
        if (oper != GT_NOT && oper != GT_NEG)
        {
            storeInd->SetRMWStatus(STOREIND_RMW_UNSUPPORTED_OPER);
            return false;
        }

        if (indirSrc->gtGetOp1()->OperGet() != GT_IND)
        {
            storeInd->SetRMWStatus(STOREIND_RMW_UNSUPPORTED_ADDR);
            return false;
        }

        GenTreePtr indirCandidate = indirSrc->gtGetOp1();
        if (indirCandidate->gtGetOp1()->OperGet() == indirDst->OperGet() &&
            IndirsAreEquivalent(indirCandidate, storeInd))
        {
            // src and dest are the same in case of unary ops
            *outIndirCandidate = indirCandidate;
            *outIndirOpSource = indirCandidate; 
            storeInd->SetRMWStatus(STOREIND_RMW_DST_IS_OP1);
            return true;
        }
    }

    assert(*outIndirCandidate == nullptr);
    assert(*outIndirOpSource == nullptr);
    storeInd->SetRMWStatus(STOREIND_RMW_UNSUPPORTED_OPER);
    return false;
}

 //--------------------------------------------------------------------------------------------
 // SetStoreIndOpCountsIfRMWMemOp checks to see if there is a RMW memory operation rooted at
 // GT_STOREIND node and if so will mark register requirements for nodes under storeInd so
 // that CodeGen will generate a single instruction of the form:
 //
 //         binOp [addressing mode], reg
 //
 // Parameters
 //         storeInd   - GT_STOREIND node
 //
 // Return value
 //         True, if RMW memory op tree pattern is recognized and op counts are set.
 //         False otherwise.
 //
bool Lowering::SetStoreIndOpCountsIfRMWMemOp(GenTreePtr storeInd)
{
    assert(storeInd->OperGet() == GT_STOREIND);

    // SSE2 doesn't support RMW on float values
    assert(!varTypeIsFloating(storeInd));

    // Terminology:
    // indirDst = memory write of an addr mode  (i.e. storeind destination)
    // indirSrc = value being written to memory (i.e. storeind source which could a binary/unary op)
    // indirCandidate = memory read i.e. a gtInd of an addr mode
    // indirOpSource = source operand used in binary/unary op (i.e. source operand of indirSrc node)

    GenTreePtr indirCandidate = nullptr;
    GenTreePtr indirOpSource = nullptr;

    if (!IsRMWMemOpRootedAtStoreInd(storeInd, &indirCandidate, &indirOpSource))
    {
        JITDUMP("Lower of StoreInd didn't mark the node as self contained for reason: %d\n", storeInd->AsStoreInd()->GetRMWStatus());
        DISPTREE(storeInd);
        return false;
    }

    GenTreePtr indirDst = storeInd->gtGetOp1();
    GenTreePtr indirSrc = storeInd->gtGetOp2();
    genTreeOps oper = indirSrc->OperGet();

    // At this point we have successfully detected a RMW memory op of one of the following forms
    //         storeInd(indirDst, indirSrc(indirCandidate, indirOpSource)) OR
    //         storeInd(indirDst, indirSrc(indirOpSource, indirCandidate) in case of commutative operations OR
    //         storeInd(indirDst, indirSrc(indirCandidate) in case of unary operations
    //
    // Here indirSrc = one of the supported binary or unary operation for RMW of memory
    //      indirCandidate = a GT_IND node
    //      indirCandidateChild = operand of GT_IND indirCandidate
    //
    // The logic below essentially does the following
    //      set storeInd src count to that of the dst count of indirOpSource
    //      clear operand counts on indirSrc  (i.e. marked as contained and storeInd will generate code for it)
    //      clear operand counts on indirCandidate
    //      clear operand counts on indirDst except when it is a GT_LCL_VAR or GT_CNS_INT that doesn't fit within addr base
    //      Increment src count of storeInd to account for the registers required to form indirDst addr mode
    //      clear operand counts on indirCandidateChild

    TreeNodeInfo* info = &(storeInd->gtLsraInfo);
    info->dstCount = 0;

    if (GenTree::OperIsBinary(oper))
    {
        // On Xarch RMW operations require that the source memory-op be in a register.
        assert(!indirOpSource->isMemoryOp() || indirOpSource->gtLsraInfo.dstCount == 1);
        JITDUMP("Lower succesfully detected an assignment of the form: *addrMode BinOp= source\n");
        info->srcCount = indirOpSource->gtLsraInfo.dstCount;
    } 
    else 
    {
        assert(GenTree::OperIsUnary(oper));
        JITDUMP("Lower succesfully detected an assignment of the form: *addrMode = UnaryOp(*addrMode)\n");
        info->srcCount = 0;
    }
    DISPTREE(storeInd);
    
    m_lsra->clearOperandCounts(indirSrc);
    m_lsra->clearOperandCounts(indirCandidate);

    GenTreePtr indirCandidateChild = indirCandidate->gtGetOp1();
    if (indirCandidateChild->OperGet() == GT_LEA)
    {
        GenTreeAddrMode* addrMode = indirCandidateChild->AsAddrMode();

        if (addrMode->HasBase())
        {
            assert(addrMode->Base()->OperIsLeaf());
            m_lsra->clearOperandCounts(addrMode->Base());
            info->srcCount++;
        }

        if (addrMode->HasIndex())
        {
            assert(addrMode->Index()->OperIsLeaf());
            m_lsra->clearOperandCounts(addrMode->Index());
            info->srcCount++;
        }

        m_lsra->clearOperandCounts(indirDst);
    }
    else
    {
        assert(indirCandidateChild->OperGet() == GT_LCL_VAR || 
               indirCandidateChild->OperGet() == GT_LCL_VAR_ADDR ||
               indirCandidateChild->OperGet() == GT_CLS_VAR_ADDR ||
               indirCandidateChild->OperGet() == GT_CNS_INT);                         

        // If it is a GT_LCL_VAR, it still needs the reg to hold the address. 
        // We would still need a reg for GT_CNS_INT if it doesn't fit within addressing mode base.
        // For GT_CLS_VAR_ADDR, we don't need a reg to hold the address, because field address value is known at jit time.
        // Also, we don't need a reg for GT_CLS_VAR_ADDR.
        if (indirCandidateChild->OperGet() == GT_LCL_VAR_ADDR || indirCandidateChild->OperGet() == GT_CLS_VAR_ADDR)
        {
            m_lsra->clearOperandCounts(indirDst);
        }
        else if (indirCandidateChild->IsCnsIntOrI() &&
                 indirCandidateChild->AsIntConCommon()->FitsInAddrBase(comp))
        {
            m_lsra->clearOperandCounts(indirDst);
        }
        else
        {
            // Need a reg and hence increment src count of storeind
            info->srcCount += indirCandidateChild->gtLsraInfo.dstCount;
        }
    }
    m_lsra->clearOperandCounts(indirCandidateChild);

    return true;
}

/**
 * Takes care of annotating the src and dst register 
 * requirements for a GT_MUL treenode.
 */
void Lowering::SetMulOpCounts(GenTreePtr tree)
{
    assert(tree->OperGet() == GT_MUL || tree->OperGet() == GT_MULHI);

    TreeNodeInfo* info = &(tree->gtLsraInfo);

    info->srcCount = 2;
    info->dstCount = 1;

    GenTreePtr op1 = tree->gtOp.gtOp1;
    GenTreePtr op2 = tree->gtOp.gtOp2;

    // Case of float/double mul.
    if (varTypeIsFloating(tree->TypeGet()))
    {
        assert(tree->OperGet() == GT_MUL);

        if (op2->isMemoryOp() || op2->IsCnsNonZeroFltOrDbl())
        {
            MakeSrcContained(tree, op2);
        }
        else if (op1->IsCnsNonZeroFltOrDbl() || (op1->isMemoryOp() && IsSafeToContainMem(tree, op1)))
        {
            // Since  GT_MUL is commutative, we will try to re-order operands if it is safe to
            // generate more efficient code sequence for the case of GT_MUL(op1=memOp, op2=non-memOp)
            MakeSrcContained(tree, op1);
        }
        else
        {
            // If there are no containable operands, we can make an operand reg optional.
            SetRegOptionalForBinOp(tree);
        }
        return;
    }
    
    bool isUnsignedMultiply    = ((tree->gtFlags & GTF_UNSIGNED) != 0);
    bool requiresOverflowCheck = tree->gtOverflowEx();
    bool useLeaEncoding = false;
    GenTreePtr memOp = nullptr;

    // There are three forms of x86 multiply:
    // one-op form:     RDX:RAX = RAX * r/m
    // two-op form:     reg *= r/m
    // three-op form:   reg = r/m * imm

    // This special widening 32x32->64 MUL is not used on x64
    assert((tree->gtFlags & GTF_MUL_64RSLT) == 0);

    // Multiply should never be using small types
    assert(!varTypeIsSmall(tree->TypeGet()));

    // We do use the widening multiply to implement 
    // the overflow checking for unsigned multiply
    // 
    if (isUnsignedMultiply && requiresOverflowCheck)
    {
        // The only encoding provided is RDX:RAX = RAX * rm
        // 
        // Here we set RAX as the only destination candidate 
        // In LSRA we set the kill set for this operation to RBM_RAX|RBM_RDX
        //
        info->setDstCandidates(m_lsra,RBM_RAX);
    }
    else if (tree->gtOper == GT_MULHI)
    {
        // have to use the encoding:RDX:RAX = RAX * rm
        info->setDstCandidates(m_lsra, RBM_RAX);
    }
    else if (IsContainableImmed(tree, op2) || IsContainableImmed(tree, op1))
    {
        GenTreeIntConCommon* imm;
        GenTreePtr other;

        if (IsContainableImmed(tree, op2))
        { 
            imm = op2->AsIntConCommon();
            other = op1; 
        }
        else
        { 
            imm = op1->AsIntConCommon();
            other = op2; 
        }

        // CQ: We want to rewrite this into a LEA
        ssize_t immVal = imm->AsIntConCommon()->IconValue();
        if (!requiresOverflowCheck && (immVal == 3 || immVal == 5 || immVal == 9))
        {
            useLeaEncoding = true;
        }

        MakeSrcContained(tree, imm);   // The imm is always contained
        if (other->isIndir())
        {
            memOp = other;             // memOp may be contained below
        }
    }
    // We allow one operand to be a contained memory operand.
    // The memory op type must match with the 'tree' type.
    // This is because during codegen we use 'tree' type to derive EmitTypeSize.
    // E.g op1 type = byte, op2 type = byte but GT_MUL tree type is int.
    //
    if (memOp == nullptr && op2->isMemoryOp())
    {
        memOp = op2;
    }

    // To generate an LEA we need to force memOp into a register
    // so don't allow memOp to be 'contained'
    //
    if (!useLeaEncoding)
    {
        if (memOp != nullptr)
        {
            if ((memOp->TypeGet() == tree->TypeGet()) &&
                IsSafeToContainMem(tree, memOp))
            {
                MakeSrcContained(tree, memOp);
            }
        }
        else
        {
            // If there are no containable operands, we can make an operand reg optional.
            SetRegOptionalForBinOp(tree);
        }
    }
}

//------------------------------------------------------------------------------
// isRMWRegOper: Can this binary tree node be used in a Read-Modify-Write format
//
// Arguments:
//    tree      - a binary tree node
//
// Return Value:
//    Returns true if we can use the read-modify-write instruction form
//
// Notes:
//    This is used to determine whether to preference the source to the destination register.
//
bool Lowering::isRMWRegOper(GenTreePtr tree)
{
    // TODO-XArch-CQ: Make this more accurate.
    // For now, We assume that most binary operators are of the RMW form.
    assert(tree->OperIsBinary());

    if (tree->OperIsCompare())
    {
        return false;
    }

    // These Opers either support a three op form (i.e. GT_LEA), or do not read/write their first operand 
    if ((tree->OperGet() == GT_LEA) || (tree->OperGet() == GT_STOREIND) || (tree->OperGet() == GT_ARR_INDEX))
        return false;

    // x86/x64 does support a three op multiply when op2|op1 is a contained immediate
    if ((tree->OperGet() == GT_MUL) &&
        (Lowering::IsContainableImmed(tree, tree->gtOp.gtOp2) ||
        Lowering::IsContainableImmed(tree, tree->gtOp.gtOp1)))
    {
        return false;
    }

    // otherwise we return true.
    return true;
}

// anything is in range for AMD64
bool Lowering::IsCallTargetInRange(void* addr)
{
    return true;
}

// return true if the immediate can be folded into an instruction, for example small enough and non-relocatable
bool Lowering:: IsContainableImmed(GenTree* parentNode, GenTree* childNode)
{
    if (!childNode->IsIntCnsFitsInI32())
    {
        return false;
    }

    // At this point we know that it is an int const fits within 4-bytes and hence can safely cast to IntConCommon.
    // Icons that need relocation should never be marked as contained immed
    if (childNode->AsIntConCommon()->ImmedValNeedsReloc(comp))
    {
        return false;
    }

    return true;
}

//----------------------------------------------------------------------
// TryToSetRegOptional - sets a bit to indicate to LSRA that register
// for a given tree node is optional for codegen purpose.  If no
// register is allocated to such a tree node, its parent node treats
// it as a contained memory operand during codegen.
//
// Arguments:
//    tree    -   GenTree node
//
// Returns
//    None
//
// Note: Right now a tree node is marked as reg optional only
// if is it a GT_LCL_VAR.  This routine needs to be modified if
// in future if lower/codegen needs to support other tree node
// types.
void Lowering::TryToSetRegOptional(GenTree* tree)
{
    if (tree->OperGet() == GT_LCL_VAR)
    {
        tree->gtLsraInfo.regOptional = true;
    }
}

// ------------------------------------------------------------------
// SetRegOptionalBinOp - Indicates which of the operands of a bin-op
// register requirement is optional. Xarch instruction set allows
// either of op1 or op2 of binary operation (e.g. add, mul etc) to be
// a memory operand.  This routine provides info to register allocator
// which of its operands optionally require a register.  Lsra might not
// allocate a register to RefTypeUse positions of such operands if it
// is beneficial. In such a case codegen will treat them as memory
// operands.
//
// Arguments:
//     tree  -  Gentree of a bininary operation.
//
// Returns 
//     None.
// 
// Note: On xarch at most only one of the operands will be marked as
// reg optional, even when both operands could be considered register
// optional.
void Lowering::SetRegOptionalForBinOp(GenTree* tree)
{
    assert(GenTree::OperIsBinary(tree->OperGet()));

    GenTree* op1 = tree->gtGetOp1();
    GenTree* op2 = tree->gtGetOp2();

    if (tree->TypeGet() == op2->TypeGet())
    {
        TryToSetRegOptional(op2);
    }

    if (!op2->IsRegOptional() &&
        tree->OperIsCommutative() &&
        tree->TypeGet() == op1->TypeGet())
    {
        TryToSetRegOptional(op1);
    }
}

#endif // _TARGET_XARCH_

#endif // !LEGACY_BACKEND