summaryrefslogtreecommitdiff
path: root/src/intel/compiler/brw_fs_nir.cpp
blob: ce350fc00b5c2962f65323b46d1d764de8fc42c8 (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
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118
5119
5120
5121
5122
5123
5124
5125
5126
5127
5128
5129
5130
5131
5132
5133
5134
5135
5136
5137
5138
5139
5140
5141
5142
5143
5144
5145
5146
5147
5148
5149
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160
5161
5162
5163
5164
5165
5166
5167
5168
5169
5170
5171
5172
5173
5174
5175
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220
5221
5222
5223
5224
5225
5226
5227
5228
5229
5230
5231
5232
5233
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
5271
5272
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289
5290
5291
5292
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308
5309
5310
5311
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334
5335
5336
5337
5338
5339
5340
5341
5342
5343
5344
5345
5346
5347
5348
5349
5350
5351
5352
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364
5365
5366
5367
5368
5369
5370
5371
5372
5373
5374
5375
5376
5377
5378
5379
5380
5381
5382
5383
5384
5385
5386
5387
5388
5389
5390
5391
5392
5393
5394
5395
5396
5397
5398
5399
5400
5401
5402
5403
5404
5405
5406
/*
 * Copyright © 2010 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

#include "compiler/glsl/ir.h"
#include "brw_fs.h"
#include "brw_fs_surface_builder.h"
#include "brw_nir.h"

using namespace brw;
using namespace brw::surface_access;

void
fs_visitor::emit_nir_code()
{
   /* emit the arrays used for inputs and outputs - load/store intrinsics will
    * be converted to reads/writes of these arrays
    */
   nir_setup_outputs();
   nir_setup_uniforms();
   nir_emit_system_values();

   /* get the main function and emit it */
   nir_foreach_function(function, nir) {
      assert(strcmp(function->name, "main") == 0);
      assert(function->impl);
      nir_emit_impl(function->impl);
   }
}

void
fs_visitor::nir_setup_outputs()
{
   if (stage == MESA_SHADER_TESS_CTRL || stage == MESA_SHADER_FRAGMENT)
      return;

   unsigned vec4s[VARYING_SLOT_TESS_MAX] = { 0, };

   /* Calculate the size of output registers in a separate pass, before
    * allocating them.  With ARB_enhanced_layouts, multiple output variables
    * may occupy the same slot, but have different type sizes.
    */
   nir_foreach_variable(var, &nir->outputs) {
      const int loc = var->data.driver_location;
      const unsigned var_vec4s =
         var->data.compact ? DIV_ROUND_UP(glsl_get_length(var->type), 4)
                           : type_size_vec4(var->type);
      vec4s[loc] = MAX2(vec4s[loc], var_vec4s);
   }

   for (unsigned loc = 0; loc < ARRAY_SIZE(vec4s);) {
      if (vec4s[loc] == 0) {
         loc++;
         continue;
      }

      unsigned reg_size = vec4s[loc];

      /* Check if there are any ranges that start within this range and extend
       * past it. If so, include them in this allocation.
       */
      for (unsigned i = 1; i < reg_size; i++)
         reg_size = MAX2(vec4s[i + loc] + i, reg_size);

      fs_reg reg = bld.vgrf(BRW_REGISTER_TYPE_F, 4 * reg_size);
      for (unsigned i = 0; i < reg_size; i++)
         outputs[loc + i] = offset(reg, bld, 4 * i);

      loc += reg_size;
   }
}

void
fs_visitor::nir_setup_uniforms()
{
   /* Only the first compile gets to set up uniforms. */
   if (push_constant_loc) {
      assert(pull_constant_loc);
      return;
   }

   uniforms = nir->num_uniforms / 4;

   if (stage == MESA_SHADER_COMPUTE) {
      /* Add a uniform for the thread local id.  It must be the last uniform
       * on the list.
       */
      assert(uniforms == prog_data->nr_params);
      uint32_t *param = brw_stage_prog_data_add_params(prog_data, 1);
      *param = BRW_PARAM_BUILTIN_SUBGROUP_ID;
      subgroup_id = fs_reg(UNIFORM, uniforms++, BRW_REGISTER_TYPE_UD);
   }
}

static bool
emit_system_values_block(nir_block *block, fs_visitor *v)
{
   fs_reg *reg;

   nir_foreach_instr(instr, block) {
      if (instr->type != nir_instr_type_intrinsic)
         continue;

      nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
      switch (intrin->intrinsic) {
      case nir_intrinsic_load_vertex_id:
      case nir_intrinsic_load_base_vertex:
         unreachable("should be lowered by nir_lower_system_values().");

      case nir_intrinsic_load_vertex_id_zero_base:
      case nir_intrinsic_load_is_indexed_draw:
      case nir_intrinsic_load_first_vertex:
      case nir_intrinsic_load_instance_id:
      case nir_intrinsic_load_base_instance:
      case nir_intrinsic_load_draw_id:
         unreachable("should be lowered by brw_nir_lower_vs_inputs().");

      case nir_intrinsic_load_invocation_id:
         if (v->stage == MESA_SHADER_TESS_CTRL)
            break;
         assert(v->stage == MESA_SHADER_GEOMETRY);
         reg = &v->nir_system_values[SYSTEM_VALUE_INVOCATION_ID];
         if (reg->file == BAD_FILE) {
            const fs_builder abld = v->bld.annotate("gl_InvocationID", NULL);
            fs_reg g1(retype(brw_vec8_grf(1, 0), BRW_REGISTER_TYPE_UD));
            fs_reg iid = abld.vgrf(BRW_REGISTER_TYPE_UD, 1);
            abld.SHR(iid, g1, brw_imm_ud(27u));
            *reg = iid;
         }
         break;

      case nir_intrinsic_load_sample_pos:
         assert(v->stage == MESA_SHADER_FRAGMENT);
         reg = &v->nir_system_values[SYSTEM_VALUE_SAMPLE_POS];
         if (reg->file == BAD_FILE)
            *reg = *v->emit_samplepos_setup();
         break;

      case nir_intrinsic_load_sample_id:
         assert(v->stage == MESA_SHADER_FRAGMENT);
         reg = &v->nir_system_values[SYSTEM_VALUE_SAMPLE_ID];
         if (reg->file == BAD_FILE)
            *reg = *v->emit_sampleid_setup();
         break;

      case nir_intrinsic_load_sample_mask_in:
         assert(v->stage == MESA_SHADER_FRAGMENT);
         assert(v->devinfo->gen >= 7);
         reg = &v->nir_system_values[SYSTEM_VALUE_SAMPLE_MASK_IN];
         if (reg->file == BAD_FILE)
            *reg = *v->emit_samplemaskin_setup();
         break;

      case nir_intrinsic_load_work_group_id:
         assert(v->stage == MESA_SHADER_COMPUTE);
         reg = &v->nir_system_values[SYSTEM_VALUE_WORK_GROUP_ID];
         if (reg->file == BAD_FILE)
            *reg = *v->emit_cs_work_group_id_setup();
         break;

      case nir_intrinsic_load_helper_invocation:
         assert(v->stage == MESA_SHADER_FRAGMENT);
         reg = &v->nir_system_values[SYSTEM_VALUE_HELPER_INVOCATION];
         if (reg->file == BAD_FILE) {
            const fs_builder abld =
               v->bld.annotate("gl_HelperInvocation", NULL);

            /* On Gen6+ (gl_HelperInvocation is only exposed on Gen7+) the
             * pixel mask is in g1.7 of the thread payload.
             *
             * We move the per-channel pixel enable bit to the low bit of each
             * channel by shifting the byte containing the pixel mask by the
             * vector immediate 0x76543210UV.
             *
             * The region of <1,8,0> reads only 1 byte (the pixel masks for
             * subspans 0 and 1) in SIMD8 and an additional byte (the pixel
             * masks for 2 and 3) in SIMD16.
             */
            fs_reg shifted = abld.vgrf(BRW_REGISTER_TYPE_UW, 1);

            for (unsigned i = 0; i < DIV_ROUND_UP(v->dispatch_width, 16); i++) {
               const fs_builder hbld = abld.group(MIN2(16, v->dispatch_width), i);
               hbld.SHR(offset(shifted, hbld, i),
                        stride(retype(brw_vec1_grf(1 + i, 7),
                                      BRW_REGISTER_TYPE_UB),
                               1, 8, 0),
                        brw_imm_v(0x76543210));
            }

            /* A set bit in the pixel mask means the channel is enabled, but
             * that is the opposite of gl_HelperInvocation so we need to invert
             * the mask.
             *
             * The negate source-modifier bit of logical instructions on Gen8+
             * performs 1's complement negation, so we can use that instead of
             * a NOT instruction.
             */
            fs_reg inverted = negate(shifted);
            if (v->devinfo->gen < 8) {
               inverted = abld.vgrf(BRW_REGISTER_TYPE_UW);
               abld.NOT(inverted, shifted);
            }

            /* We then resolve the 0/1 result to 0/~0 boolean values by ANDing
             * with 1 and negating.
             */
            fs_reg anded = abld.vgrf(BRW_REGISTER_TYPE_UD, 1);
            abld.AND(anded, inverted, brw_imm_uw(1));

            fs_reg dst = abld.vgrf(BRW_REGISTER_TYPE_D, 1);
            abld.MOV(dst, negate(retype(anded, BRW_REGISTER_TYPE_D)));
            *reg = dst;
         }
         break;

      default:
         break;
      }
   }

   return true;
}

void
fs_visitor::nir_emit_system_values()
{
   nir_system_values = ralloc_array(mem_ctx, fs_reg, SYSTEM_VALUE_MAX);
   for (unsigned i = 0; i < SYSTEM_VALUE_MAX; i++) {
      nir_system_values[i] = fs_reg();
   }

   /* Always emit SUBGROUP_INVOCATION.  Dead code will clean it up if we
    * never end up using it.
    */
   {
      const fs_builder abld = bld.annotate("gl_SubgroupInvocation", NULL);
      fs_reg &reg = nir_system_values[SYSTEM_VALUE_SUBGROUP_INVOCATION];
      reg = abld.vgrf(BRW_REGISTER_TYPE_UW);

      const fs_builder allbld8 = abld.group(8, 0).exec_all();
      allbld8.MOV(reg, brw_imm_v(0x76543210));
      if (dispatch_width > 8)
         allbld8.ADD(byte_offset(reg, 16), reg, brw_imm_uw(8u));
      if (dispatch_width > 16) {
         const fs_builder allbld16 = abld.group(16, 0).exec_all();
         allbld16.ADD(byte_offset(reg, 32), reg, brw_imm_uw(16u));
      }
   }

   nir_foreach_function(function, nir) {
      assert(strcmp(function->name, "main") == 0);
      assert(function->impl);
      nir_foreach_block(block, function->impl) {
         emit_system_values_block(block, this);
      }
   }
}

/*
 * Returns a type based on a reference_type (word, float, half-float) and a
 * given bit_size.
 *
 * Reference BRW_REGISTER_TYPE are HF,F,DF,W,D,UW,UD.
 *
 * @FIXME: 64-bit return types are always DF on integer types to maintain
 * compability with uses of DF previously to the introduction of int64
 * support.
 */
static brw_reg_type
brw_reg_type_from_bit_size(const unsigned bit_size,
                           const brw_reg_type reference_type)
{
   switch(reference_type) {
   case BRW_REGISTER_TYPE_HF:
   case BRW_REGISTER_TYPE_F:
   case BRW_REGISTER_TYPE_DF:
      switch(bit_size) {
      case 16:
         return BRW_REGISTER_TYPE_HF;
      case 32:
         return BRW_REGISTER_TYPE_F;
      case 64:
         return BRW_REGISTER_TYPE_DF;
      default:
         unreachable("Invalid bit size");
      }
   case BRW_REGISTER_TYPE_W:
   case BRW_REGISTER_TYPE_D:
   case BRW_REGISTER_TYPE_Q:
      switch(bit_size) {
      case 16:
         return BRW_REGISTER_TYPE_W;
      case 32:
         return BRW_REGISTER_TYPE_D;
      case 64:
         return BRW_REGISTER_TYPE_Q;
      default:
         unreachable("Invalid bit size");
      }
   case BRW_REGISTER_TYPE_UW:
   case BRW_REGISTER_TYPE_UD:
   case BRW_REGISTER_TYPE_UQ:
      switch(bit_size) {
      case 16:
         return BRW_REGISTER_TYPE_UW;
      case 32:
         return BRW_REGISTER_TYPE_UD;
      case 64:
         return BRW_REGISTER_TYPE_UQ;
      default:
         unreachable("Invalid bit size");
      }
   default:
      unreachable("Unknown type");
   }
}

void
fs_visitor::nir_emit_impl(nir_function_impl *impl)
{
   nir_locals = ralloc_array(mem_ctx, fs_reg, impl->reg_alloc);
   for (unsigned i = 0; i < impl->reg_alloc; i++) {
      nir_locals[i] = fs_reg();
   }

   foreach_list_typed(nir_register, reg, node, &impl->registers) {
      unsigned array_elems =
         reg->num_array_elems == 0 ? 1 : reg->num_array_elems;
      unsigned size = array_elems * reg->num_components;
      const brw_reg_type reg_type =
         brw_reg_type_from_bit_size(reg->bit_size, BRW_REGISTER_TYPE_F);
      nir_locals[reg->index] = bld.vgrf(reg_type, size);
   }

   nir_ssa_values = reralloc(mem_ctx, nir_ssa_values, fs_reg,
                             impl->ssa_alloc);

   nir_emit_cf_list(&impl->body);
}

void
fs_visitor::nir_emit_cf_list(exec_list *list)
{
   exec_list_validate(list);
   foreach_list_typed(nir_cf_node, node, node, list) {
      switch (node->type) {
      case nir_cf_node_if:
         nir_emit_if(nir_cf_node_as_if(node));
         break;

      case nir_cf_node_loop:
         nir_emit_loop(nir_cf_node_as_loop(node));
         break;

      case nir_cf_node_block:
         nir_emit_block(nir_cf_node_as_block(node));
         break;

      default:
         unreachable("Invalid CFG node block");
      }
   }
}

void
fs_visitor::nir_emit_if(nir_if *if_stmt)
{
   /* first, put the condition into f0 */
   fs_inst *inst = bld.MOV(bld.null_reg_d(),
                            retype(get_nir_src(if_stmt->condition),
                                   BRW_REGISTER_TYPE_D));
   inst->conditional_mod = BRW_CONDITIONAL_NZ;

   bld.IF(BRW_PREDICATE_NORMAL);

   nir_emit_cf_list(&if_stmt->then_list);

   /* note: if the else is empty, dead CF elimination will remove it */
   bld.emit(BRW_OPCODE_ELSE);

   nir_emit_cf_list(&if_stmt->else_list);

   bld.emit(BRW_OPCODE_ENDIF);

   if (devinfo->gen < 7)
      limit_dispatch_width(16, "Non-uniform control flow unsupported "
                           "in SIMD32 mode.");
}

void
fs_visitor::nir_emit_loop(nir_loop *loop)
{
   bld.emit(BRW_OPCODE_DO);

   nir_emit_cf_list(&loop->body);

   bld.emit(BRW_OPCODE_WHILE);

   if (devinfo->gen < 7)
      limit_dispatch_width(16, "Non-uniform control flow unsupported "
                           "in SIMD32 mode.");
}

void
fs_visitor::nir_emit_block(nir_block *block)
{
   nir_foreach_instr(instr, block) {
      nir_emit_instr(instr);
   }
}

void
fs_visitor::nir_emit_instr(nir_instr *instr)
{
   const fs_builder abld = bld.annotate(NULL, instr);

   switch (instr->type) {
   case nir_instr_type_alu:
      nir_emit_alu(abld, nir_instr_as_alu(instr));
      break;

   case nir_instr_type_deref:
      /* Derefs can exist for images but they do nothing */
      break;

   case nir_instr_type_intrinsic:
      switch (stage) {
      case MESA_SHADER_VERTEX:
         nir_emit_vs_intrinsic(abld, nir_instr_as_intrinsic(instr));
         break;
      case MESA_SHADER_TESS_CTRL:
         nir_emit_tcs_intrinsic(abld, nir_instr_as_intrinsic(instr));
         break;
      case MESA_SHADER_TESS_EVAL:
         nir_emit_tes_intrinsic(abld, nir_instr_as_intrinsic(instr));
         break;
      case MESA_SHADER_GEOMETRY:
         nir_emit_gs_intrinsic(abld, nir_instr_as_intrinsic(instr));
         break;
      case MESA_SHADER_FRAGMENT:
         nir_emit_fs_intrinsic(abld, nir_instr_as_intrinsic(instr));
         break;
      case MESA_SHADER_COMPUTE:
         nir_emit_cs_intrinsic(abld, nir_instr_as_intrinsic(instr));
         break;
      default:
         unreachable("unsupported shader stage");
      }
      break;

   case nir_instr_type_tex:
      nir_emit_texture(abld, nir_instr_as_tex(instr));
      break;

   case nir_instr_type_load_const:
      nir_emit_load_const(abld, nir_instr_as_load_const(instr));
      break;

   case nir_instr_type_ssa_undef:
      /* We create a new VGRF for undefs on every use (by handling
       * them in get_nir_src()), rather than for each definition.
       * This helps register coalescing eliminate MOVs from undef.
       */
      break;

   case nir_instr_type_jump:
      nir_emit_jump(abld, nir_instr_as_jump(instr));
      break;

   default:
      unreachable("unknown instruction type");
   }
}

/**
 * Recognizes a parent instruction of nir_op_extract_* and changes the type to
 * match instr.
 */
bool
fs_visitor::optimize_extract_to_float(nir_alu_instr *instr,
                                      const fs_reg &result)
{
   if (!instr->src[0].src.is_ssa ||
       !instr->src[0].src.ssa->parent_instr)
      return false;

   if (instr->src[0].src.ssa->parent_instr->type != nir_instr_type_alu)
      return false;

   nir_alu_instr *src0 =
      nir_instr_as_alu(instr->src[0].src.ssa->parent_instr);

   if (src0->op != nir_op_extract_u8 && src0->op != nir_op_extract_u16 &&
       src0->op != nir_op_extract_i8 && src0->op != nir_op_extract_i16)
      return false;

   nir_const_value *element = nir_src_as_const_value(src0->src[1].src);
   assert(element != NULL);

   /* Element type to extract.*/
   const brw_reg_type type = brw_int_type(
      src0->op == nir_op_extract_u16 || src0->op == nir_op_extract_i16 ? 2 : 1,
      src0->op == nir_op_extract_i16 || src0->op == nir_op_extract_i8);

   fs_reg op0 = get_nir_src(src0->src[0].src);
   op0.type = brw_type_for_nir_type(devinfo,
      (nir_alu_type)(nir_op_infos[src0->op].input_types[0] |
                     nir_src_bit_size(src0->src[0].src)));
   op0 = offset(op0, bld, src0->src[0].swizzle[0]);

   set_saturate(instr->dest.saturate,
                bld.MOV(result, subscript(op0, type, element->u32[0])));
   return true;
}

bool
fs_visitor::optimize_frontfacing_ternary(nir_alu_instr *instr,
                                         const fs_reg &result)
{
   if (!instr->src[0].src.is_ssa ||
       instr->src[0].src.ssa->parent_instr->type != nir_instr_type_intrinsic)
      return false;

   nir_intrinsic_instr *src0 =
      nir_instr_as_intrinsic(instr->src[0].src.ssa->parent_instr);

   if (src0->intrinsic != nir_intrinsic_load_front_face)
      return false;

   nir_const_value *value1 = nir_src_as_const_value(instr->src[1].src);
   if (!value1 || fabsf(value1->f32[0]) != 1.0f)
      return false;

   nir_const_value *value2 = nir_src_as_const_value(instr->src[2].src);
   if (!value2 || fabsf(value2->f32[0]) != 1.0f)
      return false;

   fs_reg tmp = vgrf(glsl_type::int_type);

   if (devinfo->gen >= 6) {
      /* Bit 15 of g0.0 is 0 if the polygon is front facing. */
      fs_reg g0 = fs_reg(retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_W));

      /* For (gl_FrontFacing ? 1.0 : -1.0), emit:
       *
       *    or(8)  tmp.1<2>W  g0.0<0,1,0>W  0x00003f80W
       *    and(8) dst<1>D    tmp<8,8,1>D   0xbf800000D
       *
       * and negate g0.0<0,1,0>W for (gl_FrontFacing ? -1.0 : 1.0).
       *
       * This negation looks like it's safe in practice, because bits 0:4 will
       * surely be TRIANGLES
       */

      if (value1->f32[0] == -1.0f) {
         g0.negate = true;
      }

      bld.OR(subscript(tmp, BRW_REGISTER_TYPE_W, 1),
             g0, brw_imm_uw(0x3f80));
   } else {
      /* Bit 31 of g1.6 is 0 if the polygon is front facing. */
      fs_reg g1_6 = fs_reg(retype(brw_vec1_grf(1, 6), BRW_REGISTER_TYPE_D));

      /* For (gl_FrontFacing ? 1.0 : -1.0), emit:
       *
       *    or(8)  tmp<1>D  g1.6<0,1,0>D  0x3f800000D
       *    and(8) dst<1>D  tmp<8,8,1>D   0xbf800000D
       *
       * and negate g1.6<0,1,0>D for (gl_FrontFacing ? -1.0 : 1.0).
       *
       * This negation looks like it's safe in practice, because bits 0:4 will
       * surely be TRIANGLES
       */

      if (value1->f32[0] == -1.0f) {
         g1_6.negate = true;
      }

      bld.OR(tmp, g1_6, brw_imm_d(0x3f800000));
   }
   bld.AND(retype(result, BRW_REGISTER_TYPE_D), tmp, brw_imm_d(0xbf800000));

   return true;
}

static void
emit_find_msb_using_lzd(const fs_builder &bld,
                        const fs_reg &result,
                        const fs_reg &src,
                        bool is_signed)
{
   fs_inst *inst;
   fs_reg temp = src;

   if (is_signed) {
      /* LZD of an absolute value source almost always does the right
       * thing.  There are two problem values:
       *
       * * 0x80000000.  Since abs(0x80000000) == 0x80000000, LZD returns
       *   0.  However, findMSB(int(0x80000000)) == 30.
       *
       * * 0xffffffff.  Since abs(0xffffffff) == 1, LZD returns
       *   31.  Section 8.8 (Integer Functions) of the GLSL 4.50 spec says:
       *
       *    For a value of zero or negative one, -1 will be returned.
       *
       * * Negative powers of two.  LZD(abs(-(1<<x))) returns x, but
       *   findMSB(-(1<<x)) should return x-1.
       *
       * For all negative number cases, including 0x80000000 and
       * 0xffffffff, the correct value is obtained from LZD if instead of
       * negating the (already negative) value the logical-not is used.  A
       * conditonal logical-not can be achieved in two instructions.
       */
      temp = bld.vgrf(BRW_REGISTER_TYPE_D);

      bld.ASR(temp, src, brw_imm_d(31));
      bld.XOR(temp, temp, src);
   }

   bld.LZD(retype(result, BRW_REGISTER_TYPE_UD),
           retype(temp, BRW_REGISTER_TYPE_UD));

   /* LZD counts from the MSB side, while GLSL's findMSB() wants the count
    * from the LSB side. Subtract the result from 31 to convert the MSB
    * count into an LSB count.  If no bits are set, LZD will return 32.
    * 31-32 = -1, which is exactly what findMSB() is supposed to return.
    */
   inst = bld.ADD(result, retype(result, BRW_REGISTER_TYPE_D), brw_imm_d(31));
   inst->src[0].negate = true;
}

static brw_rnd_mode
brw_rnd_mode_from_nir_op (const nir_op op) {
   switch (op) {
   case nir_op_f2f16_rtz:
      return BRW_RND_MODE_RTZ;
   case nir_op_f2f16_rtne:
      return BRW_RND_MODE_RTNE;
   default:
      unreachable("Operation doesn't support rounding mode");
   }
}

void
fs_visitor::nir_emit_alu(const fs_builder &bld, nir_alu_instr *instr)
{
   struct brw_wm_prog_key *fs_key = (struct brw_wm_prog_key *) this->key;
   fs_inst *inst;

   fs_reg result = get_nir_dest(instr->dest.dest);
   result.type = brw_type_for_nir_type(devinfo,
      (nir_alu_type)(nir_op_infos[instr->op].output_type |
                     nir_dest_bit_size(instr->dest.dest)));

   fs_reg op[4];
   for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++) {
      op[i] = get_nir_src(instr->src[i].src);
      op[i].type = brw_type_for_nir_type(devinfo,
         (nir_alu_type)(nir_op_infos[instr->op].input_types[i] |
                        nir_src_bit_size(instr->src[i].src)));
      op[i].abs = instr->src[i].abs;
      op[i].negate = instr->src[i].negate;
   }

   /* We get a bunch of mov's out of the from_ssa pass and they may still
    * be vectorized.  We'll handle them as a special-case.  We'll also
    * handle vecN here because it's basically the same thing.
    */
   switch (instr->op) {
   case nir_op_imov:
   case nir_op_fmov:
   case nir_op_vec2:
   case nir_op_vec3:
   case nir_op_vec4: {
      fs_reg temp = result;
      bool need_extra_copy = false;
      for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++) {
         if (!instr->src[i].src.is_ssa &&
             instr->dest.dest.reg.reg == instr->src[i].src.reg.reg) {
            need_extra_copy = true;
            temp = bld.vgrf(result.type, 4);
            break;
         }
      }

      for (unsigned i = 0; i < 4; i++) {
         if (!(instr->dest.write_mask & (1 << i)))
            continue;

         if (instr->op == nir_op_imov || instr->op == nir_op_fmov) {
            inst = bld.MOV(offset(temp, bld, i),
                           offset(op[0], bld, instr->src[0].swizzle[i]));
         } else {
            inst = bld.MOV(offset(temp, bld, i),
                           offset(op[i], bld, instr->src[i].swizzle[0]));
         }
         inst->saturate = instr->dest.saturate;
      }

      /* In this case the source and destination registers were the same,
       * so we need to insert an extra set of moves in order to deal with
       * any swizzling.
       */
      if (need_extra_copy) {
         for (unsigned i = 0; i < 4; i++) {
            if (!(instr->dest.write_mask & (1 << i)))
               continue;

            bld.MOV(offset(result, bld, i), offset(temp, bld, i));
         }
      }
      return;
   }
   default:
      break;
   }

   /* At this point, we have dealt with any instruction that operates on
    * more than a single channel.  Therefore, we can just adjust the source
    * and destination registers for that channel and emit the instruction.
    */
   unsigned channel = 0;
   if (nir_op_infos[instr->op].output_size == 0) {
      /* Since NIR is doing the scalarizing for us, we should only ever see
       * vectorized operations with a single channel.
       */
      assert(_mesa_bitcount(instr->dest.write_mask) == 1);
      channel = ffs(instr->dest.write_mask) - 1;

      result = offset(result, bld, channel);
   }

   for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++) {
      assert(nir_op_infos[instr->op].input_sizes[i] < 2);
      op[i] = offset(op[i], bld, instr->src[i].swizzle[channel]);
   }

   switch (instr->op) {
   case nir_op_i2f32:
   case nir_op_u2f32:
      if (optimize_extract_to_float(instr, result))
         return;
      inst = bld.MOV(result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_f2f16_rtne:
   case nir_op_f2f16_rtz:
      bld.emit(SHADER_OPCODE_RND_MODE, bld.null_reg_ud(),
               brw_imm_d(brw_rnd_mode_from_nir_op(instr->op)));
      /* fallthrough */

      /* In theory, it would be better to use BRW_OPCODE_F32TO16. Depending
       * on the HW gen, it is a special hw opcode or just a MOV, and
       * brw_F32TO16 (at brw_eu_emit) would do the work to chose.
       *
       * But if we want to use that opcode, we need to provide support on
       * different optimizations and lowerings. As right now HF support is
       * only for gen8+, it will be better to use directly the MOV, and use
       * BRW_OPCODE_F32TO16 when/if we work for HF support on gen7.
       */

   case nir_op_f2f16_undef:
      inst = bld.MOV(result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_f2f64:
   case nir_op_f2i64:
   case nir_op_f2u64:
   case nir_op_i2f64:
   case nir_op_i2i64:
   case nir_op_u2f64:
   case nir_op_u2u64:
      /* CHV PRM, vol07, 3D Media GPGPU Engine, Register Region Restrictions:
       *
       *    "When source or destination is 64b (...), regioning in Align1
       *     must follow these rules:
       *
       *     1. Source and destination horizontal stride must be aligned to
       *        the same qword.
       *     (...)"
       *
       * This means that conversions from bit-sizes smaller than 64-bit to
       * 64-bit need to have the source data elements aligned to 64-bit.
       * This restriction does not apply to BDW and later.
       */
      if (nir_dest_bit_size(instr->dest.dest) == 64 &&
          nir_src_bit_size(instr->src[0].src) < 64 &&
          (devinfo->is_cherryview || gen_device_info_is_9lp(devinfo))) {
         fs_reg tmp = bld.vgrf(result.type, 1);
         tmp = subscript(tmp, op[0].type, 0);
         inst = bld.MOV(tmp, op[0]);
         inst = bld.MOV(result, tmp);
         inst->saturate = instr->dest.saturate;
         break;
      }
      /* fallthrough */
   case nir_op_f2f32:
   case nir_op_f2i32:
   case nir_op_f2u32:
   case nir_op_f2i16:
   case nir_op_f2u16:
   case nir_op_i2i32:
   case nir_op_u2u32:
   case nir_op_i2i16:
   case nir_op_u2u16:
   case nir_op_i2f16:
   case nir_op_u2f16:
      inst = bld.MOV(result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_fsign: {
      if (op[0].abs) {
         /* Straightforward since the source can be assumed to be
          * non-negative.
          */
         set_condmod(BRW_CONDITIONAL_NZ, bld.MOV(result, op[0]));
         set_predicate(BRW_PREDICATE_NORMAL, bld.MOV(result, brw_imm_f(1.0f)));

      } else if (type_sz(op[0].type) < 8) {
         /* AND(val, 0x80000000) gives the sign bit.
          *
          * Predicated OR ORs 1.0 (0x3f800000) with the sign bit if val is not
          * zero.
          */
         bld.CMP(bld.null_reg_f(), op[0], brw_imm_f(0.0f), BRW_CONDITIONAL_NZ);

         fs_reg result_int = retype(result, BRW_REGISTER_TYPE_UD);
         op[0].type = BRW_REGISTER_TYPE_UD;
         result.type = BRW_REGISTER_TYPE_UD;
         bld.AND(result_int, op[0], brw_imm_ud(0x80000000u));

         inst = bld.OR(result_int, result_int, brw_imm_ud(0x3f800000u));
         inst->predicate = BRW_PREDICATE_NORMAL;
         if (instr->dest.saturate) {
            inst = bld.MOV(result, result);
            inst->saturate = true;
         }
      } else {
         /* For doubles we do the same but we need to consider:
          *
          * - 2-src instructions can't operate with 64-bit immediates
          * - The sign is encoded in the high 32-bit of each DF
          * - We need to produce a DF result.
          */

         fs_reg zero = vgrf(glsl_type::double_type);
         bld.MOV(zero, setup_imm_df(bld, 0.0));
         bld.CMP(bld.null_reg_df(), op[0], zero, BRW_CONDITIONAL_NZ);

         bld.MOV(result, zero);

         fs_reg r = subscript(result, BRW_REGISTER_TYPE_UD, 1);
         bld.AND(r, subscript(op[0], BRW_REGISTER_TYPE_UD, 1),
                 brw_imm_ud(0x80000000u));

         set_predicate(BRW_PREDICATE_NORMAL,
                       bld.OR(r, r, brw_imm_ud(0x3ff00000u)));

         if (instr->dest.saturate) {
            inst = bld.MOV(result, result);
            inst->saturate = true;
         }
      }
      break;
   }

   case nir_op_isign: {
      /*  ASR(val, 31) -> negative val generates 0xffffffff (signed -1).
       *               -> non-negative val generates 0x00000000.
       *  Predicated OR sets 1 if val is positive.
       */
      uint32_t bit_size = nir_dest_bit_size(instr->dest.dest);
      assert(bit_size == 32 || bit_size == 16);

      fs_reg zero = bit_size == 32 ? brw_imm_d(0) : brw_imm_w(0);
      fs_reg one = bit_size == 32 ? brw_imm_d(1) : brw_imm_w(1);
      fs_reg shift = bit_size == 32 ? brw_imm_d(31) : brw_imm_w(15);

      bld.CMP(bld.null_reg_d(), op[0], zero, BRW_CONDITIONAL_G);
      bld.ASR(result, op[0], shift);
      inst = bld.OR(result, result, one);
      inst->predicate = BRW_PREDICATE_NORMAL;
      break;
   }

   case nir_op_frcp:
      inst = bld.emit(SHADER_OPCODE_RCP, result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_fexp2:
      inst = bld.emit(SHADER_OPCODE_EXP2, result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_flog2:
      inst = bld.emit(SHADER_OPCODE_LOG2, result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_fsin:
      inst = bld.emit(SHADER_OPCODE_SIN, result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_fcos:
      inst = bld.emit(SHADER_OPCODE_COS, result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_fddx:
      if (fs_key->high_quality_derivatives) {
         inst = bld.emit(FS_OPCODE_DDX_FINE, result, op[0]);
      } else {
         inst = bld.emit(FS_OPCODE_DDX_COARSE, result, op[0]);
      }
      inst->saturate = instr->dest.saturate;
      break;
   case nir_op_fddx_fine:
      inst = bld.emit(FS_OPCODE_DDX_FINE, result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;
   case nir_op_fddx_coarse:
      inst = bld.emit(FS_OPCODE_DDX_COARSE, result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;
   case nir_op_fddy:
      if (fs_key->high_quality_derivatives) {
         inst = bld.emit(FS_OPCODE_DDY_FINE, result, op[0]);
      } else {
         inst = bld.emit(FS_OPCODE_DDY_COARSE, result, op[0]);
      }
      inst->saturate = instr->dest.saturate;
      break;
   case nir_op_fddy_fine:
      inst = bld.emit(FS_OPCODE_DDY_FINE, result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;
   case nir_op_fddy_coarse:
      inst = bld.emit(FS_OPCODE_DDY_COARSE, result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_iadd:
   case nir_op_fadd:
      inst = bld.ADD(result, op[0], op[1]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_fmul:
      inst = bld.MUL(result, op[0], op[1]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_imul:
      assert(nir_dest_bit_size(instr->dest.dest) < 64);
      bld.MUL(result, op[0], op[1]);
      break;

   case nir_op_imul_high:
   case nir_op_umul_high:
      assert(nir_dest_bit_size(instr->dest.dest) < 64);
      bld.emit(SHADER_OPCODE_MULH, result, op[0], op[1]);
      break;

   case nir_op_idiv:
   case nir_op_udiv:
      assert(nir_dest_bit_size(instr->dest.dest) < 64);
      bld.emit(SHADER_OPCODE_INT_QUOTIENT, result, op[0], op[1]);
      break;

   case nir_op_uadd_carry:
      unreachable("Should have been lowered by carry_to_arith().");

   case nir_op_usub_borrow:
      unreachable("Should have been lowered by borrow_to_arith().");

   case nir_op_umod:
   case nir_op_irem:
      /* According to the sign table for INT DIV in the Ivy Bridge PRM, it
       * appears that our hardware just does the right thing for signed
       * remainder.
       */
      assert(nir_dest_bit_size(instr->dest.dest) < 64);
      bld.emit(SHADER_OPCODE_INT_REMAINDER, result, op[0], op[1]);
      break;

   case nir_op_imod: {
      /* Get a regular C-style remainder.  If a % b == 0, set the predicate. */
      bld.emit(SHADER_OPCODE_INT_REMAINDER, result, op[0], op[1]);

      /* Math instructions don't support conditional mod */
      inst = bld.MOV(bld.null_reg_d(), result);
      inst->conditional_mod = BRW_CONDITIONAL_NZ;

      /* Now, we need to determine if signs of the sources are different.
       * When we XOR the sources, the top bit is 0 if they are the same and 1
       * if they are different.  We can then use a conditional modifier to
       * turn that into a predicate.  This leads us to an XOR.l instruction.
       *
       * Technically, according to the PRM, you're not allowed to use .l on a
       * XOR instruction.  However, emperical experiments and Curro's reading
       * of the simulator source both indicate that it's safe.
       */
      fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_D);
      inst = bld.XOR(tmp, op[0], op[1]);
      inst->predicate = BRW_PREDICATE_NORMAL;
      inst->conditional_mod = BRW_CONDITIONAL_L;

      /* If the result of the initial remainder operation is non-zero and the
       * two sources have different signs, add in a copy of op[1] to get the
       * final integer modulus value.
       */
      inst = bld.ADD(result, result, op[1]);
      inst->predicate = BRW_PREDICATE_NORMAL;
      break;
   }

   case nir_op_flt:
   case nir_op_fge:
   case nir_op_feq:
   case nir_op_fne: {
      fs_reg dest = result;

      const uint32_t bit_size =  nir_src_bit_size(instr->src[0].src);
      if (bit_size != 32)
         dest = bld.vgrf(op[0].type, 1);

      brw_conditional_mod cond;
      switch (instr->op) {
      case nir_op_flt:
         cond = BRW_CONDITIONAL_L;
         break;
      case nir_op_fge:
         cond = BRW_CONDITIONAL_GE;
         break;
      case nir_op_feq:
         cond = BRW_CONDITIONAL_Z;
         break;
      case nir_op_fne:
         cond = BRW_CONDITIONAL_NZ;
         break;
      default:
         unreachable("bad opcode");
      }

      bld.CMP(dest, op[0], op[1], cond);

      if (bit_size > 32) {
         bld.MOV(result, subscript(dest, BRW_REGISTER_TYPE_UD, 0));
      } else if(bit_size < 32) {
         /* When we convert the result to 32-bit we need to be careful and do
          * it as a signed conversion to get sign extension (for 32-bit true)
          */
         const brw_reg_type src_type =
            brw_reg_type_from_bit_size(bit_size, BRW_REGISTER_TYPE_D);

         bld.MOV(retype(result, BRW_REGISTER_TYPE_D), retype(dest, src_type));
      }
      break;
   }

   case nir_op_ilt:
   case nir_op_ult:
   case nir_op_ige:
   case nir_op_uge:
   case nir_op_ieq:
   case nir_op_ine: {
      fs_reg dest = result;

      const uint32_t bit_size = nir_src_bit_size(instr->src[0].src);
      if (bit_size != 32)
         dest = bld.vgrf(op[0].type, 1);

      brw_conditional_mod cond;
      switch (instr->op) {
      case nir_op_ilt:
      case nir_op_ult:
         cond = BRW_CONDITIONAL_L;
         break;
      case nir_op_ige:
      case nir_op_uge:
         cond = BRW_CONDITIONAL_GE;
         break;
      case nir_op_ieq:
         cond = BRW_CONDITIONAL_Z;
         break;
      case nir_op_ine:
         cond = BRW_CONDITIONAL_NZ;
         break;
      default:
         unreachable("bad opcode");
      }
      bld.CMP(dest, op[0], op[1], cond);

      if (bit_size > 32) {
         bld.MOV(result, subscript(dest, BRW_REGISTER_TYPE_UD, 0));
      } else if (bit_size < 32) {
         /* When we convert the result to 32-bit we need to be careful and do
          * it as a signed conversion to get sign extension (for 32-bit true)
          */
         const brw_reg_type src_type =
            brw_reg_type_from_bit_size(bit_size, BRW_REGISTER_TYPE_D);

         bld.MOV(retype(result, BRW_REGISTER_TYPE_D), retype(dest, src_type));
      }
      break;
   }

   case nir_op_inot:
      if (devinfo->gen >= 8) {
         op[0] = resolve_source_modifiers(op[0]);
      }
      bld.NOT(result, op[0]);
      break;
   case nir_op_ixor:
      if (devinfo->gen >= 8) {
         op[0] = resolve_source_modifiers(op[0]);
         op[1] = resolve_source_modifiers(op[1]);
      }
      bld.XOR(result, op[0], op[1]);
      break;
   case nir_op_ior:
      if (devinfo->gen >= 8) {
         op[0] = resolve_source_modifiers(op[0]);
         op[1] = resolve_source_modifiers(op[1]);
      }
      bld.OR(result, op[0], op[1]);
      break;
   case nir_op_iand:
      if (devinfo->gen >= 8) {
         op[0] = resolve_source_modifiers(op[0]);
         op[1] = resolve_source_modifiers(op[1]);
      }
      bld.AND(result, op[0], op[1]);
      break;

   case nir_op_fdot2:
   case nir_op_fdot3:
   case nir_op_fdot4:
   case nir_op_ball_fequal2:
   case nir_op_ball_iequal2:
   case nir_op_ball_fequal3:
   case nir_op_ball_iequal3:
   case nir_op_ball_fequal4:
   case nir_op_ball_iequal4:
   case nir_op_bany_fnequal2:
   case nir_op_bany_inequal2:
   case nir_op_bany_fnequal3:
   case nir_op_bany_inequal3:
   case nir_op_bany_fnequal4:
   case nir_op_bany_inequal4:
      unreachable("Lowered by nir_lower_alu_reductions");

   case nir_op_fnoise1_1:
   case nir_op_fnoise1_2:
   case nir_op_fnoise1_3:
   case nir_op_fnoise1_4:
   case nir_op_fnoise2_1:
   case nir_op_fnoise2_2:
   case nir_op_fnoise2_3:
   case nir_op_fnoise2_4:
   case nir_op_fnoise3_1:
   case nir_op_fnoise3_2:
   case nir_op_fnoise3_3:
   case nir_op_fnoise3_4:
   case nir_op_fnoise4_1:
   case nir_op_fnoise4_2:
   case nir_op_fnoise4_3:
   case nir_op_fnoise4_4:
      unreachable("not reached: should be handled by lower_noise");

   case nir_op_ldexp:
      unreachable("not reached: should be handled by ldexp_to_arith()");

   case nir_op_fsqrt:
      inst = bld.emit(SHADER_OPCODE_SQRT, result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_frsq:
      inst = bld.emit(SHADER_OPCODE_RSQ, result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_b2i:
   case nir_op_b2f:
      bld.MOV(result, negate(op[0]));
      break;

   case nir_op_i2b:
   case nir_op_f2b: {
      uint32_t bit_size = nir_src_bit_size(instr->src[0].src);
      if (bit_size == 64) {
         /* two-argument instructions can't take 64-bit immediates */
         fs_reg zero;
         fs_reg tmp;

         if (instr->op == nir_op_f2b) {
            zero = vgrf(glsl_type::double_type);
            tmp = vgrf(glsl_type::double_type);
            bld.MOV(zero, setup_imm_df(bld, 0.0));
         } else {
            zero = vgrf(glsl_type::int64_t_type);
            tmp = vgrf(glsl_type::int64_t_type);
            bld.MOV(zero, brw_imm_q(0));
         }

         /* A SIMD16 execution needs to be split in two instructions, so use
          * a vgrf instead of the flag register as dst so instruction splitting
          * works
          */
         bld.CMP(tmp, op[0], zero, BRW_CONDITIONAL_NZ);
         bld.MOV(result, subscript(tmp, BRW_REGISTER_TYPE_UD, 0));
      } else {
         fs_reg zero;
         if (bit_size == 32) {
            zero = instr->op == nir_op_f2b ? brw_imm_f(0.0f) : brw_imm_d(0);
         } else {
            assert(bit_size == 16);
            zero = instr->op == nir_op_f2b ?
               retype(brw_imm_w(0), BRW_REGISTER_TYPE_HF) : brw_imm_w(0);
         }
         bld.CMP(result, op[0], zero, BRW_CONDITIONAL_NZ);
      }
      break;
   }

   case nir_op_ftrunc:
      inst = bld.RNDZ(result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_fceil: {
      op[0].negate = !op[0].negate;
      fs_reg temp = vgrf(glsl_type::float_type);
      bld.RNDD(temp, op[0]);
      temp.negate = true;
      inst = bld.MOV(result, temp);
      inst->saturate = instr->dest.saturate;
      break;
   }
   case nir_op_ffloor:
      inst = bld.RNDD(result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;
   case nir_op_ffract:
      inst = bld.FRC(result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;
   case nir_op_fround_even:
      inst = bld.RNDE(result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_fquantize2f16: {
      fs_reg tmp16 = bld.vgrf(BRW_REGISTER_TYPE_D);
      fs_reg tmp32 = bld.vgrf(BRW_REGISTER_TYPE_F);
      fs_reg zero = bld.vgrf(BRW_REGISTER_TYPE_F);

      /* The destination stride must be at least as big as the source stride. */
      tmp16.type = BRW_REGISTER_TYPE_W;
      tmp16.stride = 2;

      /* Check for denormal */
      fs_reg abs_src0 = op[0];
      abs_src0.abs = true;
      bld.CMP(bld.null_reg_f(), abs_src0, brw_imm_f(ldexpf(1.0, -14)),
              BRW_CONDITIONAL_L);
      /* Get the appropriately signed zero */
      bld.AND(retype(zero, BRW_REGISTER_TYPE_UD),
              retype(op[0], BRW_REGISTER_TYPE_UD),
              brw_imm_ud(0x80000000));
      /* Do the actual F32 -> F16 -> F32 conversion */
      bld.emit(BRW_OPCODE_F32TO16, tmp16, op[0]);
      bld.emit(BRW_OPCODE_F16TO32, tmp32, tmp16);
      /* Select that or zero based on normal status */
      inst = bld.SEL(result, zero, tmp32);
      inst->predicate = BRW_PREDICATE_NORMAL;
      inst->saturate = instr->dest.saturate;
      break;
   }

   case nir_op_imin:
   case nir_op_umin:
   case nir_op_fmin:
      inst = bld.emit_minmax(result, op[0], op[1], BRW_CONDITIONAL_L);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_imax:
   case nir_op_umax:
   case nir_op_fmax:
      inst = bld.emit_minmax(result, op[0], op[1], BRW_CONDITIONAL_GE);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_pack_snorm_2x16:
   case nir_op_pack_snorm_4x8:
   case nir_op_pack_unorm_2x16:
   case nir_op_pack_unorm_4x8:
   case nir_op_unpack_snorm_2x16:
   case nir_op_unpack_snorm_4x8:
   case nir_op_unpack_unorm_2x16:
   case nir_op_unpack_unorm_4x8:
   case nir_op_unpack_half_2x16:
   case nir_op_pack_half_2x16:
      unreachable("not reached: should be handled by lower_packing_builtins");

   case nir_op_unpack_half_2x16_split_x:
      inst = bld.emit(FS_OPCODE_UNPACK_HALF_2x16_SPLIT_X, result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;
   case nir_op_unpack_half_2x16_split_y:
      inst = bld.emit(FS_OPCODE_UNPACK_HALF_2x16_SPLIT_Y, result, op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_pack_64_2x32_split:
   case nir_op_pack_32_2x16_split:
      bld.emit(FS_OPCODE_PACK, result, op[0], op[1]);
      break;

   case nir_op_unpack_64_2x32_split_x:
   case nir_op_unpack_64_2x32_split_y: {
      if (instr->op == nir_op_unpack_64_2x32_split_x)
         bld.MOV(result, subscript(op[0], BRW_REGISTER_TYPE_UD, 0));
      else
         bld.MOV(result, subscript(op[0], BRW_REGISTER_TYPE_UD, 1));
      break;
   }

   case nir_op_unpack_32_2x16_split_x:
   case nir_op_unpack_32_2x16_split_y: {
      if (instr->op == nir_op_unpack_32_2x16_split_x)
         bld.MOV(result, subscript(op[0], BRW_REGISTER_TYPE_UW, 0));
      else
         bld.MOV(result, subscript(op[0], BRW_REGISTER_TYPE_UW, 1));
      break;
   }

   case nir_op_fpow:
      inst = bld.emit(SHADER_OPCODE_POW, result, op[0], op[1]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_bitfield_reverse:
      assert(nir_dest_bit_size(instr->dest.dest) < 64);
      bld.BFREV(result, op[0]);
      break;

   case nir_op_bit_count:
      assert(nir_dest_bit_size(instr->dest.dest) < 64);
      bld.CBIT(result, op[0]);
      break;

   case nir_op_ufind_msb: {
      assert(nir_dest_bit_size(instr->dest.dest) < 64);
      emit_find_msb_using_lzd(bld, result, op[0], false);
      break;
   }

   case nir_op_ifind_msb: {
      assert(nir_dest_bit_size(instr->dest.dest) < 64);

      if (devinfo->gen < 7) {
         emit_find_msb_using_lzd(bld, result, op[0], true);
      } else {
         bld.FBH(retype(result, BRW_REGISTER_TYPE_UD), op[0]);

         /* FBH counts from the MSB side, while GLSL's findMSB() wants the
          * count from the LSB side. If FBH didn't return an error
          * (0xFFFFFFFF), then subtract the result from 31 to convert the MSB
          * count into an LSB count.
          */
         bld.CMP(bld.null_reg_d(), result, brw_imm_d(-1), BRW_CONDITIONAL_NZ);

         inst = bld.ADD(result, result, brw_imm_d(31));
         inst->predicate = BRW_PREDICATE_NORMAL;
         inst->src[0].negate = true;
      }
      break;
   }

   case nir_op_find_lsb:
      assert(nir_dest_bit_size(instr->dest.dest) < 64);

      if (devinfo->gen < 7) {
         fs_reg temp = vgrf(glsl_type::int_type);

         /* (x & -x) generates a value that consists of only the LSB of x.
          * For all powers of 2, findMSB(y) == findLSB(y).
          */
         fs_reg src = retype(op[0], BRW_REGISTER_TYPE_D);
         fs_reg negated_src = src;

         /* One must be negated, and the other must be non-negated.  It
          * doesn't matter which is which.
          */
         negated_src.negate = true;
         src.negate = false;

         bld.AND(temp, src, negated_src);
         emit_find_msb_using_lzd(bld, result, temp, false);
      } else {
         bld.FBL(result, op[0]);
      }
      break;

   case nir_op_ubitfield_extract:
   case nir_op_ibitfield_extract:
      unreachable("should have been lowered");
   case nir_op_ubfe:
   case nir_op_ibfe:
      assert(nir_dest_bit_size(instr->dest.dest) < 64);
      bld.BFE(result, op[2], op[1], op[0]);
      break;
   case nir_op_bfm:
      assert(nir_dest_bit_size(instr->dest.dest) < 64);
      bld.BFI1(result, op[0], op[1]);
      break;
   case nir_op_bfi:
      assert(nir_dest_bit_size(instr->dest.dest) < 64);
      bld.BFI2(result, op[0], op[1], op[2]);
      break;

   case nir_op_bitfield_insert:
      unreachable("not reached: should have been lowered");

   case nir_op_ishl:
   case nir_op_ishr:
   case nir_op_ushr: {
      fs_reg shift_count = op[1];

      if (devinfo->is_cherryview || gen_device_info_is_9lp(devinfo)) {
         if (op[1].file == VGRF &&
             (result.type == BRW_REGISTER_TYPE_Q ||
              result.type == BRW_REGISTER_TYPE_UQ)) {
            shift_count = fs_reg(VGRF, alloc.allocate(dispatch_width / 4),
                                 BRW_REGISTER_TYPE_UD);
            shift_count.stride = 2;
            bld.MOV(shift_count, op[1]);
         }
      }

      switch (instr->op) {
      case nir_op_ishl:
         bld.SHL(result, op[0], shift_count);
         break;
      case nir_op_ishr:
         bld.ASR(result, op[0], shift_count);
         break;
      case nir_op_ushr:
         bld.SHR(result, op[0], shift_count);
         break;
      default:
         unreachable("not reached");
      }
      break;
   }

   case nir_op_pack_half_2x16_split:
      bld.emit(FS_OPCODE_PACK_HALF_2x16_SPLIT, result, op[0], op[1]);
      break;

   case nir_op_ffma:
      inst = bld.MAD(result, op[2], op[1], op[0]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_flrp:
      inst = bld.LRP(result, op[0], op[1], op[2]);
      inst->saturate = instr->dest.saturate;
      break;

   case nir_op_bcsel:
      if (optimize_frontfacing_ternary(instr, result))
         return;

      bld.CMP(bld.null_reg_d(), op[0], brw_imm_d(0), BRW_CONDITIONAL_NZ);
      inst = bld.SEL(result, op[1], op[2]);
      inst->predicate = BRW_PREDICATE_NORMAL;
      break;

   case nir_op_extract_u8:
   case nir_op_extract_i8: {
      nir_const_value *byte = nir_src_as_const_value(instr->src[1].src);
      assert(byte != NULL);

      /* The PRMs say:
       *
       *    BDW+
       *    There is no direct conversion from B/UB to Q/UQ or Q/UQ to B/UB.
       *    Use two instructions and a word or DWord intermediate integer type.
       */
      if (nir_dest_bit_size(instr->dest.dest) == 64) {
         const brw_reg_type type = brw_int_type(2, instr->op == nir_op_extract_i8);

         if (instr->op == nir_op_extract_i8) {
            /* If we need to sign extend, extract to a word first */
            fs_reg w_temp = bld.vgrf(BRW_REGISTER_TYPE_W);
            bld.MOV(w_temp, subscript(op[0], type, byte->u32[0]));
            bld.MOV(result, w_temp);
         } else {
            /* Otherwise use an AND with 0xff and a word type */
            bld.AND(result, subscript(op[0], type, byte->u32[0] / 2), brw_imm_uw(0xff));
         }
      } else {
         const brw_reg_type type = brw_int_type(1, instr->op == nir_op_extract_i8);
         bld.MOV(result, subscript(op[0], type, byte->u32[0]));
      }
      break;
   }

   case nir_op_extract_u16:
   case nir_op_extract_i16: {
      const brw_reg_type type = brw_int_type(2, instr->op == nir_op_extract_i16);
      nir_const_value *word = nir_src_as_const_value(instr->src[1].src);
      assert(word != NULL);
      bld.MOV(result, subscript(op[0], type, word->u32[0]));
      break;
   }

   default:
      unreachable("unhandled instruction");
   }

   /* If we need to do a boolean resolve, replace the result with -(x & 1)
    * to sign extend the low bit to 0/~0
    */
   if (devinfo->gen <= 5 &&
       (instr->instr.pass_flags & BRW_NIR_BOOLEAN_MASK) == BRW_NIR_BOOLEAN_NEEDS_RESOLVE) {
      fs_reg masked = vgrf(glsl_type::int_type);
      bld.AND(masked, result, brw_imm_d(1));
      masked.negate = true;
      bld.MOV(retype(result, BRW_REGISTER_TYPE_D), masked);
   }
}

void
fs_visitor::nir_emit_load_const(const fs_builder &bld,
                                nir_load_const_instr *instr)
{
   const brw_reg_type reg_type =
      brw_reg_type_from_bit_size(instr->def.bit_size, BRW_REGISTER_TYPE_D);
   fs_reg reg = bld.vgrf(reg_type, instr->def.num_components);

   switch (instr->def.bit_size) {
   case 16:
      for (unsigned i = 0; i < instr->def.num_components; i++)
         bld.MOV(offset(reg, bld, i), brw_imm_w(instr->value.i16[i]));
      break;

   case 32:
      for (unsigned i = 0; i < instr->def.num_components; i++)
         bld.MOV(offset(reg, bld, i), brw_imm_d(instr->value.i32[i]));
      break;

   case 64:
      assert(devinfo->gen >= 7);
      if (devinfo->gen == 7) {
         /* We don't get 64-bit integer types until gen8 */
         for (unsigned i = 0; i < instr->def.num_components; i++) {
            bld.MOV(retype(offset(reg, bld, i), BRW_REGISTER_TYPE_DF),
                    setup_imm_df(bld, instr->value.f64[i]));
         }
      } else {
         for (unsigned i = 0; i < instr->def.num_components; i++)
            bld.MOV(offset(reg, bld, i), brw_imm_q(instr->value.i64[i]));
      }
      break;

   default:
      unreachable("Invalid bit size");
   }

   nir_ssa_values[instr->def.index] = reg;
}

fs_reg
fs_visitor::get_nir_src(const nir_src &src)
{
   fs_reg reg;
   if (src.is_ssa) {
      if (src.ssa->parent_instr->type == nir_instr_type_ssa_undef) {
         const brw_reg_type reg_type =
            brw_reg_type_from_bit_size(src.ssa->bit_size, BRW_REGISTER_TYPE_D);
         reg = bld.vgrf(reg_type, src.ssa->num_components);
      } else {
         reg = nir_ssa_values[src.ssa->index];
      }
   } else {
      /* We don't handle indirects on locals */
      assert(src.reg.indirect == NULL);
      reg = offset(nir_locals[src.reg.reg->index], bld,
                   src.reg.base_offset * src.reg.reg->num_components);
   }

   if (nir_src_bit_size(src) == 64 && devinfo->gen == 7) {
      /* The only 64-bit type available on gen7 is DF, so use that. */
      reg.type = BRW_REGISTER_TYPE_DF;
   } else {
      /* To avoid floating-point denorm flushing problems, set the type by
       * default to an integer type - instructions that need floating point
       * semantics will set this to F if they need to
       */
      reg.type = brw_reg_type_from_bit_size(nir_src_bit_size(src),
                                            BRW_REGISTER_TYPE_D);
   }

   return reg;
}

/**
 * Return an IMM for constants; otherwise call get_nir_src() as normal.
 *
 * This function should not be called on any value which may be 64 bits.
 * We could theoretically support 64-bit on gen8+ but we choose not to
 * because it wouldn't work in general (no gen7 support) and there are
 * enough restrictions in 64-bit immediates that you can't take the return
 * value and treat it the same as the result of get_nir_src().
 */
fs_reg
fs_visitor::get_nir_src_imm(const nir_src &src)
{
   nir_const_value *val = nir_src_as_const_value(src);
   assert(nir_src_bit_size(src) == 32);
   return val ? fs_reg(brw_imm_d(val->i32[0])) : get_nir_src(src);
}

fs_reg
fs_visitor::get_nir_dest(const nir_dest &dest)
{
   if (dest.is_ssa) {
      const brw_reg_type reg_type =
         brw_reg_type_from_bit_size(dest.ssa.bit_size, BRW_REGISTER_TYPE_F);
      nir_ssa_values[dest.ssa.index] =
         bld.vgrf(reg_type, dest.ssa.num_components);
      return nir_ssa_values[dest.ssa.index];
   } else {
      /* We don't handle indirects on locals */
      assert(dest.reg.indirect == NULL);
      return offset(nir_locals[dest.reg.reg->index], bld,
                    dest.reg.base_offset * dest.reg.reg->num_components);
   }
}

fs_reg
fs_visitor::get_nir_image_deref(nir_deref_instr *deref)
{
   fs_reg arr_offset = brw_imm_ud(0);
   unsigned array_size = BRW_IMAGE_PARAM_SIZE * 4;
   nir_deref_instr *head = deref;
   while (head->deref_type != nir_deref_type_var) {
      assert(head->deref_type == nir_deref_type_array);

      /* This level's element size is the previous level's array size */
      const unsigned elem_size = array_size;

      fs_reg index = retype(get_nir_src_imm(head->arr.index),
                            BRW_REGISTER_TYPE_UD);
      if (arr_offset.file == BRW_IMMEDIATE_VALUE &&
          index.file == BRW_IMMEDIATE_VALUE) {
         arr_offset.ud += index.ud * elem_size;
      } else if (index.file == BRW_IMMEDIATE_VALUE) {
         bld.ADD(arr_offset, arr_offset, brw_imm_ud(index.ud * elem_size));
      } else {
         fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_UD);
         bld.MUL(tmp, index, brw_imm_ud(elem_size));
         bld.ADD(tmp, tmp, arr_offset);
         arr_offset = tmp;
      }

      head = nir_deref_instr_parent(head);
      assert(glsl_type_is_array(head->type));
      array_size = elem_size * glsl_get_length(head->type);
   }

   assert(head->deref_type == nir_deref_type_var);
   const unsigned max_arr_offset = array_size - (BRW_IMAGE_PARAM_SIZE * 4);
   fs_reg image(UNIFORM, head->var->data.driver_location / 4,
                BRW_REGISTER_TYPE_UD);

   if (arr_offset.file == BRW_IMMEDIATE_VALUE) {
      /* The offset is in bytes but we want it in dwords */
      return offset(image, bld, MIN2(arr_offset.ud, max_arr_offset) / 4);
   } else {
      /* Accessing an invalid surface index with the dataport can result
       * in a hang.  According to the spec "if the index used to
       * select an individual element is negative or greater than or
       * equal to the size of the array, the results of the operation
       * are undefined but may not lead to termination" -- which is one
       * of the possible outcomes of the hang.  Clamp the index to
       * prevent access outside of the array bounds.
       */
      bld.emit_minmax(arr_offset, arr_offset, brw_imm_ud(max_arr_offset),
                      BRW_CONDITIONAL_L);

      /* Emit a pile of MOVs to load the uniform into a temporary.  The
       * dead-code elimination pass will get rid of what we don't use.
       */
      fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_UD, BRW_IMAGE_PARAM_SIZE);
      for (unsigned j = 0; j < BRW_IMAGE_PARAM_SIZE; j++) {
         bld.emit(SHADER_OPCODE_MOV_INDIRECT,
                  offset(tmp, bld, j), offset(image, bld, j),
                  arr_offset, brw_imm_ud(max_arr_offset + 4));
      }
      return tmp;
   }
}

void
fs_visitor::emit_percomp(const fs_builder &bld, const fs_inst &inst,
                         unsigned wr_mask)
{
   for (unsigned i = 0; i < 4; i++) {
      if (!((wr_mask >> i) & 1))
         continue;

      fs_inst *new_inst = new(mem_ctx) fs_inst(inst);
      new_inst->dst = offset(new_inst->dst, bld, i);
      for (unsigned j = 0; j < new_inst->sources; j++)
         if (new_inst->src[j].file == VGRF)
            new_inst->src[j] = offset(new_inst->src[j], bld, i);

      bld.emit(new_inst);
   }
}

/**
 * Get the matching channel register datatype for an image intrinsic of the
 * specified GLSL image type.
 */
static brw_reg_type
get_image_base_type(const glsl_type *type)
{
   switch ((glsl_base_type)type->sampled_type) {
   case GLSL_TYPE_UINT:
      return BRW_REGISTER_TYPE_UD;
   case GLSL_TYPE_INT:
      return BRW_REGISTER_TYPE_D;
   case GLSL_TYPE_FLOAT:
      return BRW_REGISTER_TYPE_F;
   default:
      unreachable("Not reached.");
   }
}

/**
 * Get the appropriate atomic op for an image atomic intrinsic.
 */
static unsigned
get_image_atomic_op(nir_intrinsic_op op, const glsl_type *type)
{
   switch (op) {
   case nir_intrinsic_image_deref_atomic_add:
      return BRW_AOP_ADD;
   case nir_intrinsic_image_deref_atomic_min:
      return (get_image_base_type(type) == BRW_REGISTER_TYPE_D ?
              BRW_AOP_IMIN : BRW_AOP_UMIN);
   case nir_intrinsic_image_deref_atomic_max:
      return (get_image_base_type(type) == BRW_REGISTER_TYPE_D ?
              BRW_AOP_IMAX : BRW_AOP_UMAX);
   case nir_intrinsic_image_deref_atomic_and:
      return BRW_AOP_AND;
   case nir_intrinsic_image_deref_atomic_or:
      return BRW_AOP_OR;
   case nir_intrinsic_image_deref_atomic_xor:
      return BRW_AOP_XOR;
   case nir_intrinsic_image_deref_atomic_exchange:
      return BRW_AOP_MOV;
   case nir_intrinsic_image_deref_atomic_comp_swap:
      return BRW_AOP_CMPWR;
   default:
      unreachable("Not reachable.");
   }
}

static fs_inst *
emit_pixel_interpolater_send(const fs_builder &bld,
                             enum opcode opcode,
                             const fs_reg &dst,
                             const fs_reg &src,
                             const fs_reg &desc,
                             glsl_interp_mode interpolation)
{
   struct brw_wm_prog_data *wm_prog_data =
      brw_wm_prog_data(bld.shader->stage_prog_data);

   fs_inst *inst = bld.emit(opcode, dst, src, desc);
   /* 2 floats per slot returned */
   inst->size_written = 2 * dst.component_size(inst->exec_size);
   inst->pi_noperspective = interpolation == INTERP_MODE_NOPERSPECTIVE;

   wm_prog_data->pulls_bary = true;

   return inst;
}

/**
 * Computes 1 << x, given a D/UD register containing some value x.
 */
static fs_reg
intexp2(const fs_builder &bld, const fs_reg &x)
{
   assert(x.type == BRW_REGISTER_TYPE_UD || x.type == BRW_REGISTER_TYPE_D);

   fs_reg result = bld.vgrf(x.type, 1);
   fs_reg one = bld.vgrf(x.type, 1);

   bld.MOV(one, retype(brw_imm_d(1), one.type));
   bld.SHL(result, one, x);
   return result;
}

void
fs_visitor::emit_gs_end_primitive(const nir_src &vertex_count_nir_src)
{
   assert(stage == MESA_SHADER_GEOMETRY);

   struct brw_gs_prog_data *gs_prog_data = brw_gs_prog_data(prog_data);

   if (gs_compile->control_data_header_size_bits == 0)
      return;

   /* We can only do EndPrimitive() functionality when the control data
    * consists of cut bits.  Fortunately, the only time it isn't is when the
    * output type is points, in which case EndPrimitive() is a no-op.
    */
   if (gs_prog_data->control_data_format !=
       GEN7_GS_CONTROL_DATA_FORMAT_GSCTL_CUT) {
      return;
   }

   /* Cut bits use one bit per vertex. */
   assert(gs_compile->control_data_bits_per_vertex == 1);

   fs_reg vertex_count = get_nir_src(vertex_count_nir_src);
   vertex_count.type = BRW_REGISTER_TYPE_UD;

   /* Cut bit n should be set to 1 if EndPrimitive() was called after emitting
    * vertex n, 0 otherwise.  So all we need to do here is mark bit
    * (vertex_count - 1) % 32 in the cut_bits register to indicate that
    * EndPrimitive() was called after emitting vertex (vertex_count - 1);
    * vec4_gs_visitor::emit_control_data_bits() will take care of the rest.
    *
    * Note that if EndPrimitive() is called before emitting any vertices, this
    * will cause us to set bit 31 of the control_data_bits register to 1.
    * That's fine because:
    *
    * - If max_vertices < 32, then vertex number 31 (zero-based) will never be
    *   output, so the hardware will ignore cut bit 31.
    *
    * - If max_vertices == 32, then vertex number 31 is guaranteed to be the
    *   last vertex, so setting cut bit 31 has no effect (since the primitive
    *   is automatically ended when the GS terminates).
    *
    * - If max_vertices > 32, then the ir_emit_vertex visitor will reset the
    *   control_data_bits register to 0 when the first vertex is emitted.
    */

   const fs_builder abld = bld.annotate("end primitive");

   /* control_data_bits |= 1 << ((vertex_count - 1) % 32) */
   fs_reg prev_count = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
   abld.ADD(prev_count, vertex_count, brw_imm_ud(0xffffffffu));
   fs_reg mask = intexp2(abld, prev_count);
   /* Note: we're relying on the fact that the GEN SHL instruction only pays
    * attention to the lower 5 bits of its second source argument, so on this
    * architecture, 1 << (vertex_count - 1) is equivalent to 1 <<
    * ((vertex_count - 1) % 32).
    */
   abld.OR(this->control_data_bits, this->control_data_bits, mask);
}

void
fs_visitor::emit_gs_control_data_bits(const fs_reg &vertex_count)
{
   assert(stage == MESA_SHADER_GEOMETRY);
   assert(gs_compile->control_data_bits_per_vertex != 0);

   struct brw_gs_prog_data *gs_prog_data = brw_gs_prog_data(prog_data);

   const fs_builder abld = bld.annotate("emit control data bits");
   const fs_builder fwa_bld = bld.exec_all();

   /* We use a single UD register to accumulate control data bits (32 bits
    * for each of the SIMD8 channels).  So we need to write a DWord (32 bits)
    * at a time.
    *
    * Unfortunately, the URB_WRITE_SIMD8 message uses 128-bit (OWord) offsets.
    * We have select a 128-bit group via the Global and Per-Slot Offsets, then
    * use the Channel Mask phase to enable/disable which DWord within that
    * group to write.  (Remember, different SIMD8 channels may have emitted
    * different numbers of vertices, so we may need per-slot offsets.)
    *
    * Channel masking presents an annoying problem: we may have to replicate
    * the data up to 4 times:
    *
    * Msg = Handles, Per-Slot Offsets, Channel Masks, Data, Data, Data, Data.
    *
    * To avoid penalizing shaders that emit a small number of vertices, we
    * can avoid these sometimes: if the size of the control data header is
    * <= 128 bits, then there is only 1 OWord.  All SIMD8 channels will land
    * land in the same 128-bit group, so we can skip per-slot offsets.
    *
    * Similarly, if the control data header is <= 32 bits, there is only one
    * DWord, so we can skip channel masks.
    */
   enum opcode opcode = SHADER_OPCODE_URB_WRITE_SIMD8;

   fs_reg channel_mask, per_slot_offset;

   if (gs_compile->control_data_header_size_bits > 32) {
      opcode = SHADER_OPCODE_URB_WRITE_SIMD8_MASKED;
      channel_mask = vgrf(glsl_type::uint_type);
   }

   if (gs_compile->control_data_header_size_bits > 128) {
      opcode = SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT;
      per_slot_offset = vgrf(glsl_type::uint_type);
   }

   /* Figure out which DWord we're trying to write to using the formula:
    *
    *    dword_index = (vertex_count - 1) * bits_per_vertex / 32
    *
    * Since bits_per_vertex is a power of two, and is known at compile
    * time, this can be optimized to:
    *
    *    dword_index = (vertex_count - 1) >> (6 - log2(bits_per_vertex))
    */
   if (opcode != SHADER_OPCODE_URB_WRITE_SIMD8) {
      fs_reg dword_index = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
      fs_reg prev_count = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
      abld.ADD(prev_count, vertex_count, brw_imm_ud(0xffffffffu));
      unsigned log2_bits_per_vertex =
         util_last_bit(gs_compile->control_data_bits_per_vertex);
      abld.SHR(dword_index, prev_count, brw_imm_ud(6u - log2_bits_per_vertex));

      if (per_slot_offset.file != BAD_FILE) {
         /* Set the per-slot offset to dword_index / 4, so that we'll write to
          * the appropriate OWord within the control data header.
          */
         abld.SHR(per_slot_offset, dword_index, brw_imm_ud(2u));
      }

      /* Set the channel masks to 1 << (dword_index % 4), so that we'll
       * write to the appropriate DWORD within the OWORD.
       */
      fs_reg channel = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
      fwa_bld.AND(channel, dword_index, brw_imm_ud(3u));
      channel_mask = intexp2(fwa_bld, channel);
      /* Then the channel masks need to be in bits 23:16. */
      fwa_bld.SHL(channel_mask, channel_mask, brw_imm_ud(16u));
   }

   /* Store the control data bits in the message payload and send it. */
   int mlen = 2;
   if (channel_mask.file != BAD_FILE)
      mlen += 4; /* channel masks, plus 3 extra copies of the data */
   if (per_slot_offset.file != BAD_FILE)
      mlen++;

   fs_reg payload = bld.vgrf(BRW_REGISTER_TYPE_UD, mlen);
   fs_reg *sources = ralloc_array(mem_ctx, fs_reg, mlen);
   int i = 0;
   sources[i++] = fs_reg(retype(brw_vec8_grf(1, 0), BRW_REGISTER_TYPE_UD));
   if (per_slot_offset.file != BAD_FILE)
      sources[i++] = per_slot_offset;
   if (channel_mask.file != BAD_FILE)
      sources[i++] = channel_mask;
   while (i < mlen) {
      sources[i++] = this->control_data_bits;
   }

   abld.LOAD_PAYLOAD(payload, sources, mlen, mlen);
   fs_inst *inst = abld.emit(opcode, reg_undef, payload);
   inst->mlen = mlen;
   /* We need to increment Global Offset by 256-bits to make room for
    * Broadwell's extra "Vertex Count" payload at the beginning of the
    * URB entry.  Since this is an OWord message, Global Offset is counted
    * in 128-bit units, so we must set it to 2.
    */
   if (gs_prog_data->static_vertex_count == -1)
      inst->offset = 2;
}

void
fs_visitor::set_gs_stream_control_data_bits(const fs_reg &vertex_count,
                                            unsigned stream_id)
{
   /* control_data_bits |= stream_id << ((2 * (vertex_count - 1)) % 32) */

   /* Note: we are calling this *before* increasing vertex_count, so
    * this->vertex_count == vertex_count - 1 in the formula above.
    */

   /* Stream mode uses 2 bits per vertex */
   assert(gs_compile->control_data_bits_per_vertex == 2);

   /* Must be a valid stream */
   assert(stream_id < MAX_VERTEX_STREAMS);

   /* Control data bits are initialized to 0 so we don't have to set any
    * bits when sending vertices to stream 0.
    */
   if (stream_id == 0)
      return;

   const fs_builder abld = bld.annotate("set stream control data bits", NULL);

   /* reg::sid = stream_id */
   fs_reg sid = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
   abld.MOV(sid, brw_imm_ud(stream_id));

   /* reg:shift_count = 2 * (vertex_count - 1) */
   fs_reg shift_count = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
   abld.SHL(shift_count, vertex_count, brw_imm_ud(1u));

   /* Note: we're relying on the fact that the GEN SHL instruction only pays
    * attention to the lower 5 bits of its second source argument, so on this
    * architecture, stream_id << 2 * (vertex_count - 1) is equivalent to
    * stream_id << ((2 * (vertex_count - 1)) % 32).
    */
   fs_reg mask = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
   abld.SHL(mask, sid, shift_count);
   abld.OR(this->control_data_bits, this->control_data_bits, mask);
}

void
fs_visitor::emit_gs_vertex(const nir_src &vertex_count_nir_src,
                           unsigned stream_id)
{
   assert(stage == MESA_SHADER_GEOMETRY);

   struct brw_gs_prog_data *gs_prog_data = brw_gs_prog_data(prog_data);

   fs_reg vertex_count = get_nir_src(vertex_count_nir_src);
   vertex_count.type = BRW_REGISTER_TYPE_UD;

   /* Haswell and later hardware ignores the "Render Stream Select" bits
    * from the 3DSTATE_STREAMOUT packet when the SOL stage is disabled,
    * and instead sends all primitives down the pipeline for rasterization.
    * If the SOL stage is enabled, "Render Stream Select" is honored and
    * primitives bound to non-zero streams are discarded after stream output.
    *
    * Since the only purpose of primives sent to non-zero streams is to
    * be recorded by transform feedback, we can simply discard all geometry
    * bound to these streams when transform feedback is disabled.
    */
   if (stream_id > 0 && !nir->info.has_transform_feedback_varyings)
      return;

   /* If we're outputting 32 control data bits or less, then we can wait
    * until the shader is over to output them all.  Otherwise we need to
    * output them as we go.  Now is the time to do it, since we're about to
    * output the vertex_count'th vertex, so it's guaranteed that the
    * control data bits associated with the (vertex_count - 1)th vertex are
    * correct.
    */
   if (gs_compile->control_data_header_size_bits > 32) {
      const fs_builder abld =
         bld.annotate("emit vertex: emit control data bits");

      /* Only emit control data bits if we've finished accumulating a batch
       * of 32 bits.  This is the case when:
       *
       *     (vertex_count * bits_per_vertex) % 32 == 0
       *
       * (in other words, when the last 5 bits of vertex_count *
       * bits_per_vertex are 0).  Assuming bits_per_vertex == 2^n for some
       * integer n (which is always the case, since bits_per_vertex is
       * always 1 or 2), this is equivalent to requiring that the last 5-n
       * bits of vertex_count are 0:
       *
       *     vertex_count & (2^(5-n) - 1) == 0
       *
       * 2^(5-n) == 2^5 / 2^n == 32 / bits_per_vertex, so this is
       * equivalent to:
       *
       *     vertex_count & (32 / bits_per_vertex - 1) == 0
       *
       * TODO: If vertex_count is an immediate, we could do some of this math
       *       at compile time...
       */
      fs_inst *inst =
         abld.AND(bld.null_reg_d(), vertex_count,
                  brw_imm_ud(32u / gs_compile->control_data_bits_per_vertex - 1u));
      inst->conditional_mod = BRW_CONDITIONAL_Z;

      abld.IF(BRW_PREDICATE_NORMAL);
      /* If vertex_count is 0, then no control data bits have been
       * accumulated yet, so we can skip emitting them.
       */
      abld.CMP(bld.null_reg_d(), vertex_count, brw_imm_ud(0u),
               BRW_CONDITIONAL_NEQ);
      abld.IF(BRW_PREDICATE_NORMAL);
      emit_gs_control_data_bits(vertex_count);
      abld.emit(BRW_OPCODE_ENDIF);

      /* Reset control_data_bits to 0 so we can start accumulating a new
       * batch.
       *
       * Note: in the case where vertex_count == 0, this neutralizes the
       * effect of any call to EndPrimitive() that the shader may have
       * made before outputting its first vertex.
       */
      inst = abld.MOV(this->control_data_bits, brw_imm_ud(0u));
      inst->force_writemask_all = true;
      abld.emit(BRW_OPCODE_ENDIF);
   }

   emit_urb_writes(vertex_count);

   /* In stream mode we have to set control data bits for all vertices
    * unless we have disabled control data bits completely (which we do
    * do for GL_POINTS outputs that don't use streams).
    */
   if (gs_compile->control_data_header_size_bits > 0 &&
       gs_prog_data->control_data_format ==
          GEN7_GS_CONTROL_DATA_FORMAT_GSCTL_SID) {
      set_gs_stream_control_data_bits(vertex_count, stream_id);
   }
}

void
fs_visitor::emit_gs_input_load(const fs_reg &dst,
                               const nir_src &vertex_src,
                               unsigned base_offset,
                               const nir_src &offset_src,
                               unsigned num_components,
                               unsigned first_component)
{
   struct brw_gs_prog_data *gs_prog_data = brw_gs_prog_data(prog_data);

   nir_const_value *vertex_const = nir_src_as_const_value(vertex_src);
   nir_const_value *offset_const = nir_src_as_const_value(offset_src);
   const unsigned push_reg_count = gs_prog_data->base.urb_read_length * 8;

   /* TODO: figure out push input layout for invocations == 1 */
   /* TODO: make this work with 64-bit inputs */
   if (gs_prog_data->invocations == 1 &&
       type_sz(dst.type) <= 4 &&
       offset_const != NULL && vertex_const != NULL &&
       4 * (base_offset + offset_const->u32[0]) < push_reg_count) {
      int imm_offset = (base_offset + offset_const->u32[0]) * 4 +
                       vertex_const->u32[0] * push_reg_count;
      for (unsigned i = 0; i < num_components; i++) {
         bld.MOV(offset(dst, bld, i),
                 fs_reg(ATTR, imm_offset + i + first_component, dst.type));
      }
      return;
   }

   /* Resort to the pull model.  Ensure the VUE handles are provided. */
   assert(gs_prog_data->base.include_vue_handles);

   unsigned first_icp_handle = gs_prog_data->include_primitive_id ? 3 : 2;
   fs_reg icp_handle = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);

   if (gs_prog_data->invocations == 1) {
      if (vertex_const) {
         /* The vertex index is constant; just select the proper URB handle. */
         icp_handle =
            retype(brw_vec8_grf(first_icp_handle + vertex_const->i32[0], 0),
                   BRW_REGISTER_TYPE_UD);
      } else {
         /* The vertex index is non-constant.  We need to use indirect
          * addressing to fetch the proper URB handle.
          *
          * First, we start with the sequence <7, 6, 5, 4, 3, 2, 1, 0>
          * indicating that channel <n> should read the handle from
          * DWord <n>.  We convert that to bytes by multiplying by 4.
          *
          * Next, we convert the vertex index to bytes by multiplying
          * by 32 (shifting by 5), and add the two together.  This is
          * the final indirect byte offset.
          */
         fs_reg sequence = bld.vgrf(BRW_REGISTER_TYPE_UW, 1);
         fs_reg channel_offsets = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
         fs_reg vertex_offset_bytes = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
         fs_reg icp_offset_bytes = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);

         /* sequence = <7, 6, 5, 4, 3, 2, 1, 0> */
         bld.MOV(sequence, fs_reg(brw_imm_v(0x76543210)));
         /* channel_offsets = 4 * sequence = <28, 24, 20, 16, 12, 8, 4, 0> */
         bld.SHL(channel_offsets, sequence, brw_imm_ud(2u));
         /* Convert vertex_index to bytes (multiply by 32) */
         bld.SHL(vertex_offset_bytes,
                 retype(get_nir_src(vertex_src), BRW_REGISTER_TYPE_UD),
                 brw_imm_ud(5u));
         bld.ADD(icp_offset_bytes, vertex_offset_bytes, channel_offsets);

         /* Use first_icp_handle as the base offset.  There is one register
          * of URB handles per vertex, so inform the register allocator that
          * we might read up to nir->info.gs.vertices_in registers.
          */
         bld.emit(SHADER_OPCODE_MOV_INDIRECT, icp_handle,
                  retype(brw_vec8_grf(first_icp_handle, 0), icp_handle.type),
                  fs_reg(icp_offset_bytes),
                  brw_imm_ud(nir->info.gs.vertices_in * REG_SIZE));
      }
   } else {
      assert(gs_prog_data->invocations > 1);

      if (vertex_const) {
         assert(devinfo->gen >= 9 || vertex_const->i32[0] <= 5);
         bld.MOV(icp_handle,
                 retype(brw_vec1_grf(first_icp_handle +
                                     vertex_const->i32[0] / 8,
                                     vertex_const->i32[0] % 8),
                        BRW_REGISTER_TYPE_UD));
      } else {
         /* The vertex index is non-constant.  We need to use indirect
          * addressing to fetch the proper URB handle.
          *
          */
         fs_reg icp_offset_bytes = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);

         /* Convert vertex_index to bytes (multiply by 4) */
         bld.SHL(icp_offset_bytes,
                 retype(get_nir_src(vertex_src), BRW_REGISTER_TYPE_UD),
                 brw_imm_ud(2u));

         /* Use first_icp_handle as the base offset.  There is one DWord
          * of URB handles per vertex, so inform the register allocator that
          * we might read up to ceil(nir->info.gs.vertices_in / 8) registers.
          */
         bld.emit(SHADER_OPCODE_MOV_INDIRECT, icp_handle,
                  retype(brw_vec8_grf(first_icp_handle, 0), icp_handle.type),
                  fs_reg(icp_offset_bytes),
                  brw_imm_ud(DIV_ROUND_UP(nir->info.gs.vertices_in, 8) *
                             REG_SIZE));
      }
   }

   fs_inst *inst;

   fs_reg tmp_dst = dst;
   fs_reg indirect_offset = get_nir_src(offset_src);
   unsigned num_iterations = 1;
   unsigned orig_num_components = num_components;

   if (type_sz(dst.type) == 8) {
      if (num_components > 2) {
         num_iterations = 2;
         num_components = 2;
      }
      fs_reg tmp = fs_reg(VGRF, alloc.allocate(4), dst.type);
      tmp_dst = tmp;
      first_component = first_component / 2;
   }

   for (unsigned iter = 0; iter < num_iterations; iter++) {
      if (offset_const) {
         /* Constant indexing - use global offset. */
         if (first_component != 0) {
            unsigned read_components = num_components + first_component;
            fs_reg tmp = bld.vgrf(dst.type, read_components);
            inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, tmp, icp_handle);
            inst->size_written = read_components *
                                 tmp.component_size(inst->exec_size);
            for (unsigned i = 0; i < num_components; i++) {
               bld.MOV(offset(tmp_dst, bld, i),
                       offset(tmp, bld, i + first_component));
            }
         } else {
            inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, tmp_dst,
                            icp_handle);
            inst->size_written = num_components *
                                 tmp_dst.component_size(inst->exec_size);
         }
         inst->offset = base_offset + offset_const->u32[0];
         inst->mlen = 1;
      } else {
         /* Indirect indexing - use per-slot offsets as well. */
         const fs_reg srcs[] = { icp_handle, indirect_offset };
         unsigned read_components = num_components + first_component;
         fs_reg tmp = bld.vgrf(dst.type, read_components);
         fs_reg payload = bld.vgrf(BRW_REGISTER_TYPE_UD, 2);
         bld.LOAD_PAYLOAD(payload, srcs, ARRAY_SIZE(srcs), 0);
         if (first_component != 0) {
            inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, tmp,
                            payload);
            inst->size_written = read_components *
                                 tmp.component_size(inst->exec_size);
            for (unsigned i = 0; i < num_components; i++) {
               bld.MOV(offset(tmp_dst, bld, i),
                       offset(tmp, bld, i + first_component));
            }
         } else {
            inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, tmp_dst,
                         payload);
            inst->size_written = num_components *
                                 tmp_dst.component_size(inst->exec_size);
         }
         inst->offset = base_offset;
         inst->mlen = 2;
      }

      if (type_sz(dst.type) == 8) {
         shuffle_from_32bit_read(bld,
                                 offset(dst, bld, iter * 2),
                                 retype(tmp_dst, BRW_REGISTER_TYPE_D),
                                 0,
                                 num_components);
      }

      if (num_iterations > 1) {
         num_components = orig_num_components - 2;
         if(offset_const) {
            base_offset++;
         } else {
            fs_reg new_indirect = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
            bld.ADD(new_indirect, indirect_offset, brw_imm_ud(1u));
            indirect_offset = new_indirect;
         }
      }
   }
}

fs_reg
fs_visitor::get_indirect_offset(nir_intrinsic_instr *instr)
{
   nir_src *offset_src = nir_get_io_offset_src(instr);
   nir_const_value *const_value = nir_src_as_const_value(*offset_src);

   if (const_value) {
      /* The only constant offset we should find is 0.  brw_nir.c's
       * add_const_offset_to_base() will fold other constant offsets
       * into instr->const_index[0].
       */
      assert(const_value->u32[0] == 0);
      return fs_reg();
   }

   return get_nir_src(*offset_src);
}

static void
do_untyped_vector_read(const fs_builder &bld,
                       const fs_reg dest,
                       const fs_reg surf_index,
                       const fs_reg offset_reg,
                       unsigned num_components)
{
   if (type_sz(dest.type) <= 2) {
      assert(dest.stride == 1);
      boolean is_const_offset = offset_reg.file == BRW_IMMEDIATE_VALUE;

      if (is_const_offset) {
         uint32_t start = offset_reg.ud & ~3;
         uint32_t end = offset_reg.ud + num_components * type_sz(dest.type);
         end = ALIGN(end, 4);
         assert (end - start <= 16);

         /* At this point we have 16-bit component/s that have constant
          * offset aligned to 4-bytes that can be read with untyped_reads.
          * untyped_read message requires 32-bit aligned offsets.
          */
         unsigned first_component = (offset_reg.ud & 3) / type_sz(dest.type);
         unsigned num_components_32bit = (end - start) / 4;

         fs_reg read_result =
            emit_untyped_read(bld, surf_index, brw_imm_ud(start),
                              1 /* dims */,
                              num_components_32bit,
                              BRW_PREDICATE_NONE);
         shuffle_from_32bit_read(bld, dest, read_result, first_component,
                                 num_components);
      } else {
         fs_reg read_offset = bld.vgrf(BRW_REGISTER_TYPE_UD);
         for (unsigned i = 0; i < num_components; i++) {
            if (i == 0) {
               bld.MOV(read_offset, offset_reg);
            } else {
               bld.ADD(read_offset, offset_reg,
                       brw_imm_ud(i * type_sz(dest.type)));
            }
            /* Non constant offsets are not guaranteed to be aligned 32-bits
             * so they are read using one byte_scattered_read message
             * for each component.
             */
            fs_reg read_result =
               emit_byte_scattered_read(bld, surf_index, read_offset,
                                        1 /* dims */, 1,
                                        type_sz(dest.type) * 8 /* bit_size */,
                                        BRW_PREDICATE_NONE);
            bld.MOV(offset(dest, bld, i),
                    subscript (read_result, dest.type, 0));
         }
      }
   } else if (type_sz(dest.type) == 4) {
      fs_reg read_result = emit_untyped_read(bld, surf_index, offset_reg,
                                             1 /* dims */,
                                             num_components,
                                             BRW_PREDICATE_NONE);
      read_result.type = dest.type;
      for (unsigned i = 0; i < num_components; i++)
         bld.MOV(offset(dest, bld, i), offset(read_result, bld, i));
   } else if (type_sz(dest.type) == 8) {
      /* Reading a dvec, so we need to:
       *
       * 1. Multiply num_components by 2, to account for the fact that we
       *    need to read 64-bit components.
       * 2. Shuffle the result of the load to form valid 64-bit elements
       * 3. Emit a second load (for components z/w) if needed.
       */
      fs_reg read_offset = bld.vgrf(BRW_REGISTER_TYPE_UD);
      bld.MOV(read_offset, offset_reg);

      int iters = num_components <= 2 ? 1 : 2;

      /* Load the dvec, the first iteration loads components x/y, the second
       * iteration, if needed, loads components z/w
       */
      for (int it = 0; it < iters; it++) {
         /* Compute number of components to read in this iteration */
         int iter_components = MIN2(2, num_components);
         num_components -= iter_components;

         /* Read. Since this message reads 32-bit components, we need to
          * read twice as many components.
          */
         fs_reg read_result = emit_untyped_read(bld, surf_index, read_offset,
                                                1 /* dims */,
                                                iter_components * 2,
                                                BRW_PREDICATE_NONE);

         /* Shuffle the 32-bit load result into valid 64-bit data */
         shuffle_from_32bit_read(bld, offset(dest, bld, it * 2),
                                 read_result, 0, iter_components);

         bld.ADD(read_offset, read_offset, brw_imm_ud(16));
      }
   } else {
      unreachable("Unsupported type");
   }
}

void
fs_visitor::nir_emit_vs_intrinsic(const fs_builder &bld,
                                  nir_intrinsic_instr *instr)
{
   assert(stage == MESA_SHADER_VERTEX);

   fs_reg dest;
   if (nir_intrinsic_infos[instr->intrinsic].has_dest)
      dest = get_nir_dest(instr->dest);

   switch (instr->intrinsic) {
   case nir_intrinsic_load_vertex_id:
   case nir_intrinsic_load_base_vertex:
      unreachable("should be lowered by nir_lower_system_values()");

   case nir_intrinsic_load_vertex_id_zero_base:
   case nir_intrinsic_load_instance_id:
   case nir_intrinsic_load_base_instance:
   case nir_intrinsic_load_draw_id: {
      gl_system_value sv = nir_system_value_from_intrinsic(instr->intrinsic);
      fs_reg val = nir_system_values[sv];
      assert(val.file != BAD_FILE);
      dest.type = val.type;
      bld.MOV(dest, val);
      break;
   }

   case nir_intrinsic_load_input: {
      fs_reg src = fs_reg(ATTR, nir_intrinsic_base(instr) * 4, dest.type);
      unsigned first_component = nir_intrinsic_component(instr);
      unsigned num_components = instr->num_components;

      nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);
      assert(const_offset && "Indirect input loads not allowed");
      src = offset(src, bld, const_offset->u32[0]);

      if (type_sz(dest.type) == 8)
         first_component /= 2;

      /* For 16-bit support maybe a temporary will be needed to copy from
       * the ATTR file.
       */
      shuffle_from_32bit_read(bld, dest, retype(src, BRW_REGISTER_TYPE_D),
                              first_component, num_components);
      break;
   }

   case nir_intrinsic_load_first_vertex:
   case nir_intrinsic_load_is_indexed_draw:
      unreachable("lowered by brw_nir_lower_vs_inputs");

   default:
      nir_emit_intrinsic(bld, instr);
      break;
   }
}

void
fs_visitor::nir_emit_tcs_intrinsic(const fs_builder &bld,
                                   nir_intrinsic_instr *instr)
{
   assert(stage == MESA_SHADER_TESS_CTRL);
   struct brw_tcs_prog_key *tcs_key = (struct brw_tcs_prog_key *) key;
   struct brw_tcs_prog_data *tcs_prog_data = brw_tcs_prog_data(prog_data);

   fs_reg dst;
   if (nir_intrinsic_infos[instr->intrinsic].has_dest)
      dst = get_nir_dest(instr->dest);

   switch (instr->intrinsic) {
   case nir_intrinsic_load_primitive_id:
      bld.MOV(dst, fs_reg(brw_vec1_grf(0, 1)));
      break;
   case nir_intrinsic_load_invocation_id:
      bld.MOV(retype(dst, invocation_id.type), invocation_id);
      break;
   case nir_intrinsic_load_patch_vertices_in:
      bld.MOV(retype(dst, BRW_REGISTER_TYPE_D),
              brw_imm_d(tcs_key->input_vertices));
      break;

   case nir_intrinsic_barrier: {
      if (tcs_prog_data->instances == 1)
         break;

      fs_reg m0 = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
      fs_reg m0_2 = component(m0, 2);

      const fs_builder chanbld = bld.exec_all().group(1, 0);

      /* Zero the message header */
      bld.exec_all().MOV(m0, brw_imm_ud(0u));

      /* Copy "Barrier ID" from r0.2, bits 16:13 */
      chanbld.AND(m0_2, retype(brw_vec1_grf(0, 2), BRW_REGISTER_TYPE_UD),
                  brw_imm_ud(INTEL_MASK(16, 13)));

      /* Shift it up to bits 27:24. */
      chanbld.SHL(m0_2, m0_2, brw_imm_ud(11));

      /* Set the Barrier Count and the enable bit */
      chanbld.OR(m0_2, m0_2,
                 brw_imm_ud(tcs_prog_data->instances << 9 | (1 << 15)));

      bld.emit(SHADER_OPCODE_BARRIER, bld.null_reg_ud(), m0);
      break;
   }

   case nir_intrinsic_load_input:
      unreachable("nir_lower_io should never give us these.");
      break;

   case nir_intrinsic_load_per_vertex_input: {
      fs_reg indirect_offset = get_indirect_offset(instr);
      unsigned imm_offset = instr->const_index[0];

      const nir_src &vertex_src = instr->src[0];
      nir_const_value *vertex_const = nir_src_as_const_value(vertex_src);

      fs_inst *inst;

      fs_reg icp_handle;

      if (vertex_const) {
         /* Emit a MOV to resolve <0,1,0> regioning. */
         icp_handle = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
         bld.MOV(icp_handle,
                 retype(brw_vec1_grf(1 + (vertex_const->i32[0] >> 3),
                                     vertex_const->i32[0] & 7),
                        BRW_REGISTER_TYPE_UD));
      } else if (tcs_prog_data->instances == 1 &&
                 vertex_src.is_ssa &&
                 vertex_src.ssa->parent_instr->type == nir_instr_type_intrinsic &&
                 nir_instr_as_intrinsic(vertex_src.ssa->parent_instr)->intrinsic == nir_intrinsic_load_invocation_id) {
         /* For the common case of only 1 instance, an array index of
          * gl_InvocationID means reading g1.  Skip all the indirect work.
          */
         icp_handle = retype(brw_vec8_grf(1, 0), BRW_REGISTER_TYPE_UD);
      } else {
         /* The vertex index is non-constant.  We need to use indirect
          * addressing to fetch the proper URB handle.
          */
         icp_handle = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);

         /* Each ICP handle is a single DWord (4 bytes) */
         fs_reg vertex_offset_bytes = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
         bld.SHL(vertex_offset_bytes,
                 retype(get_nir_src(vertex_src), BRW_REGISTER_TYPE_UD),
                 brw_imm_ud(2u));

         /* Start at g1.  We might read up to 4 registers. */
         bld.emit(SHADER_OPCODE_MOV_INDIRECT, icp_handle,
                  retype(brw_vec8_grf(1, 0), icp_handle.type), vertex_offset_bytes,
                  brw_imm_ud(4 * REG_SIZE));
      }

      /* We can only read two double components with each URB read, so
       * we send two read messages in that case, each one loading up to
       * two double components.
       */
      unsigned num_iterations = 1;
      unsigned num_components = instr->num_components;
      unsigned first_component = nir_intrinsic_component(instr);
      fs_reg orig_dst = dst;
      if (type_sz(dst.type) == 8) {
         first_component = first_component / 2;
         if (instr->num_components > 2) {
            num_iterations = 2;
            num_components = 2;
         }

         fs_reg tmp = fs_reg(VGRF, alloc.allocate(4), dst.type);
         dst = tmp;
      }

      for (unsigned iter = 0; iter < num_iterations; iter++) {
         if (indirect_offset.file == BAD_FILE) {
            /* Constant indexing - use global offset. */
            if (first_component != 0) {
               unsigned read_components = num_components + first_component;
               fs_reg tmp = bld.vgrf(dst.type, read_components);
               inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, tmp, icp_handle);
               for (unsigned i = 0; i < num_components; i++) {
                  bld.MOV(offset(dst, bld, i),
                          offset(tmp, bld, i + first_component));
               }
            } else {
               inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, dst, icp_handle);
            }
            inst->offset = imm_offset;
            inst->mlen = 1;
         } else {
            /* Indirect indexing - use per-slot offsets as well. */
            const fs_reg srcs[] = { icp_handle, indirect_offset };
            fs_reg payload = bld.vgrf(BRW_REGISTER_TYPE_UD, 2);
            bld.LOAD_PAYLOAD(payload, srcs, ARRAY_SIZE(srcs), 0);
            if (first_component != 0) {
               unsigned read_components = num_components + first_component;
               fs_reg tmp = bld.vgrf(dst.type, read_components);
               inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, tmp,
                               payload);
               for (unsigned i = 0; i < num_components; i++) {
                  bld.MOV(offset(dst, bld, i),
                          offset(tmp, bld, i + first_component));
               }
            } else {
               inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, dst,
                               payload);
            }
            inst->offset = imm_offset;
            inst->mlen = 2;
         }
         inst->size_written = (num_components + first_component) *
                              inst->dst.component_size(inst->exec_size);

         /* If we are reading 64-bit data using 32-bit read messages we need
          * build proper 64-bit data elements by shuffling the low and high
          * 32-bit components around like we do for other things like UBOs
          * or SSBOs.
          */
         if (type_sz(dst.type) == 8) {
            shuffle_from_32bit_read(bld,
                                    offset(orig_dst, bld, iter * 2),
                                    retype(dst, BRW_REGISTER_TYPE_D),
                                    0, num_components);
         }

         /* Copy the temporary to the destination to deal with writemasking.
          *
          * Also attempt to deal with gl_PointSize being in the .w component.
          */
         if (inst->offset == 0 && indirect_offset.file == BAD_FILE) {
            assert(type_sz(dst.type) < 8);
            inst->dst = bld.vgrf(dst.type, 4);
            inst->size_written = 4 * REG_SIZE;
            bld.MOV(dst, offset(inst->dst, bld, 3));
         }

         /* If we are loading double data and we need a second read message
          * adjust the write offset
          */
         if (num_iterations > 1) {
            num_components = instr->num_components - 2;
            imm_offset++;
         }
      }
      break;
   }

   case nir_intrinsic_load_output:
   case nir_intrinsic_load_per_vertex_output: {
      fs_reg indirect_offset = get_indirect_offset(instr);
      unsigned imm_offset = instr->const_index[0];
      unsigned first_component = nir_intrinsic_component(instr);

      fs_inst *inst;
      if (indirect_offset.file == BAD_FILE) {
         /* Replicate the patch handle to all enabled channels */
         fs_reg patch_handle = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
         bld.MOV(patch_handle,
                 retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UD));

         {
            if (first_component != 0) {
               unsigned read_components =
                  instr->num_components + first_component;
               fs_reg tmp = bld.vgrf(dst.type, read_components);
               inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, tmp,
                               patch_handle);
               inst->size_written = read_components * REG_SIZE;
               for (unsigned i = 0; i < instr->num_components; i++) {
                  bld.MOV(offset(dst, bld, i),
                          offset(tmp, bld, i + first_component));
               }
            } else {
               inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, dst,
                               patch_handle);
               inst->size_written = instr->num_components * REG_SIZE;
            }
            inst->offset = imm_offset;
            inst->mlen = 1;
         }
      } else {
         /* Indirect indexing - use per-slot offsets as well. */
         const fs_reg srcs[] = {
            retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UD),
            indirect_offset
         };
         fs_reg payload = bld.vgrf(BRW_REGISTER_TYPE_UD, 2);
         bld.LOAD_PAYLOAD(payload, srcs, ARRAY_SIZE(srcs), 0);
         if (first_component != 0) {
            unsigned read_components =
               instr->num_components + first_component;
            fs_reg tmp = bld.vgrf(dst.type, read_components);
            inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, tmp,
                            payload);
            inst->size_written = read_components * REG_SIZE;
            for (unsigned i = 0; i < instr->num_components; i++) {
               bld.MOV(offset(dst, bld, i),
                       offset(tmp, bld, i + first_component));
            }
         } else {
            inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, dst,
                            payload);
            inst->size_written = instr->num_components * REG_SIZE;
         }
         inst->offset = imm_offset;
         inst->mlen = 2;
      }
      break;
   }

   case nir_intrinsic_store_output:
   case nir_intrinsic_store_per_vertex_output: {
      fs_reg value = get_nir_src(instr->src[0]);
      bool is_64bit = (instr->src[0].is_ssa ?
         instr->src[0].ssa->bit_size : instr->src[0].reg.reg->bit_size) == 64;
      fs_reg indirect_offset = get_indirect_offset(instr);
      unsigned imm_offset = instr->const_index[0];
      unsigned mask = instr->const_index[1];
      unsigned header_regs = 0;
      fs_reg srcs[7];
      srcs[header_regs++] = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UD);

      if (indirect_offset.file != BAD_FILE) {
         srcs[header_regs++] = indirect_offset;
      }

      if (mask == 0)
         break;

      unsigned num_components = util_last_bit(mask);
      enum opcode opcode;

      /* We can only pack two 64-bit components in a single message, so send
       * 2 messages if we have more components
       */
      unsigned num_iterations = 1;
      unsigned iter_components = num_components;
      unsigned first_component = nir_intrinsic_component(instr);
      if (is_64bit) {
         first_component = first_component / 2;
         if (instr->num_components > 2) {
            num_iterations = 2;
            iter_components = 2;
         }
      }

      mask = mask << first_component;

      for (unsigned iter = 0; iter < num_iterations; iter++) {
         if (!is_64bit && mask != WRITEMASK_XYZW) {
            srcs[header_regs++] = brw_imm_ud(mask << 16);
            opcode = indirect_offset.file != BAD_FILE ?
               SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT :
               SHADER_OPCODE_URB_WRITE_SIMD8_MASKED;
         } else if (is_64bit && ((mask & WRITEMASK_XY) != WRITEMASK_XY)) {
            /* Expand the 64-bit mask to 32-bit channels. We only handle
             * two channels in each iteration, so we only care about X/Y.
             */
            unsigned mask32 = 0;
            if (mask & WRITEMASK_X)
               mask32 |= WRITEMASK_XY;
            if (mask & WRITEMASK_Y)
               mask32 |= WRITEMASK_ZW;

            /* If the mask does not include any of the channels X or Y there
             * is nothing to do in this iteration. Move on to the next couple
             * of 64-bit channels.
             */
            if (!mask32) {
               mask >>= 2;
               imm_offset++;
               continue;
            }

            srcs[header_regs++] = brw_imm_ud(mask32 << 16);
            opcode = indirect_offset.file != BAD_FILE ?
               SHADER_OPCODE_URB_WRITE_SIMD8_MASKED_PER_SLOT :
               SHADER_OPCODE_URB_WRITE_SIMD8_MASKED;
         } else {
            opcode = indirect_offset.file != BAD_FILE ?
               SHADER_OPCODE_URB_WRITE_SIMD8_PER_SLOT :
               SHADER_OPCODE_URB_WRITE_SIMD8;
         }

         for (unsigned i = 0; i < iter_components; i++) {
            if (!(mask & (1 << (i + first_component))))
               continue;

            if (!is_64bit) {
               srcs[header_regs + i + first_component] = offset(value, bld, i);
            } else {
               /* We need to shuffle the 64-bit data to match the layout
                * expected by our 32-bit URB write messages. We use a temporary
                * for that.
                */
               unsigned channel = iter * 2 + i;
               fs_reg dest = shuffle_for_32bit_write(bld, value, channel, 1);

               srcs[header_regs + (i + first_component) * 2] = dest;
               srcs[header_regs + (i + first_component) * 2 + 1] =
                  offset(dest, bld, 1);
            }
         }

         unsigned mlen =
            header_regs + (is_64bit ? 2 * iter_components : iter_components) +
            (is_64bit ? 2 * first_component : first_component);
         fs_reg payload =
            bld.vgrf(BRW_REGISTER_TYPE_UD, mlen);
         bld.LOAD_PAYLOAD(payload, srcs, mlen, header_regs);

         fs_inst *inst = bld.emit(opcode, bld.null_reg_ud(), payload);
         inst->offset = imm_offset;
         inst->mlen = mlen;

         /* If this is a 64-bit attribute, select the next two 64-bit channels
          * to be handled in the next iteration.
          */
         if (is_64bit) {
            mask >>= 2;
            imm_offset++;
         }
      }
      break;
   }

   default:
      nir_emit_intrinsic(bld, instr);
      break;
   }
}

void
fs_visitor::nir_emit_tes_intrinsic(const fs_builder &bld,
                                   nir_intrinsic_instr *instr)
{
   assert(stage == MESA_SHADER_TESS_EVAL);
   struct brw_tes_prog_data *tes_prog_data = brw_tes_prog_data(prog_data);

   fs_reg dest;
   if (nir_intrinsic_infos[instr->intrinsic].has_dest)
      dest = get_nir_dest(instr->dest);

   switch (instr->intrinsic) {
   case nir_intrinsic_load_primitive_id:
      bld.MOV(dest, fs_reg(brw_vec1_grf(0, 1)));
      break;
   case nir_intrinsic_load_tess_coord:
      /* gl_TessCoord is part of the payload in g1-3 */
      for (unsigned i = 0; i < 3; i++) {
         bld.MOV(offset(dest, bld, i), fs_reg(brw_vec8_grf(1 + i, 0)));
      }
      break;

   case nir_intrinsic_load_input:
   case nir_intrinsic_load_per_vertex_input: {
      fs_reg indirect_offset = get_indirect_offset(instr);
      unsigned imm_offset = instr->const_index[0];
      unsigned first_component = nir_intrinsic_component(instr);

      if (type_sz(dest.type) == 8) {
         first_component = first_component / 2;
      }

      fs_inst *inst;
      if (indirect_offset.file == BAD_FILE) {
         /* Arbitrarily only push up to 32 vec4 slots worth of data,
          * which is 16 registers (since each holds 2 vec4 slots).
          */
         unsigned slot_count = 1;
         if (type_sz(dest.type) == 8 && instr->num_components > 2)
            slot_count++;

         const unsigned max_push_slots = 32;
         if (imm_offset + slot_count <= max_push_slots) {
            fs_reg src = fs_reg(ATTR, imm_offset / 2, dest.type);
            for (int i = 0; i < instr->num_components; i++) {
               unsigned comp = 16 / type_sz(dest.type) * (imm_offset % 2) +
                  i + first_component;
               bld.MOV(offset(dest, bld, i), component(src, comp));
            }

            tes_prog_data->base.urb_read_length =
               MAX2(tes_prog_data->base.urb_read_length,
                    DIV_ROUND_UP(imm_offset + slot_count, 2));
         } else {
            /* Replicate the patch handle to all enabled channels */
            const fs_reg srcs[] = {
               retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UD)
            };
            fs_reg patch_handle = bld.vgrf(BRW_REGISTER_TYPE_UD, 1);
            bld.LOAD_PAYLOAD(patch_handle, srcs, ARRAY_SIZE(srcs), 0);

            if (first_component != 0) {
               unsigned read_components =
                  instr->num_components + first_component;
               fs_reg tmp = bld.vgrf(dest.type, read_components);
               inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, tmp,
                               patch_handle);
               inst->size_written = read_components * REG_SIZE;
               for (unsigned i = 0; i < instr->num_components; i++) {
                  bld.MOV(offset(dest, bld, i),
                          offset(tmp, bld, i + first_component));
               }
            } else {
               inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8, dest,
                               patch_handle);
               inst->size_written = instr->num_components * REG_SIZE;
            }
            inst->mlen = 1;
            inst->offset = imm_offset;
         }
      } else {
         /* Indirect indexing - use per-slot offsets as well. */

         /* We can only read two double components with each URB read, so
          * we send two read messages in that case, each one loading up to
          * two double components.
          */
         unsigned num_iterations = 1;
         unsigned num_components = instr->num_components;
         fs_reg orig_dest = dest;
         if (type_sz(dest.type) == 8) {
            if (instr->num_components > 2) {
               num_iterations = 2;
               num_components = 2;
            }
            fs_reg tmp = fs_reg(VGRF, alloc.allocate(4), dest.type);
            dest = tmp;
         }

         for (unsigned iter = 0; iter < num_iterations; iter++) {
            const fs_reg srcs[] = {
               retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UD),
               indirect_offset
            };
            fs_reg payload = bld.vgrf(BRW_REGISTER_TYPE_UD, 2);
            bld.LOAD_PAYLOAD(payload, srcs, ARRAY_SIZE(srcs), 0);

            if (first_component != 0) {
               unsigned read_components =
                   num_components + first_component;
               fs_reg tmp = bld.vgrf(dest.type, read_components);
               inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, tmp,
                               payload);
               for (unsigned i = 0; i < num_components; i++) {
                  bld.MOV(offset(dest, bld, i),
                          offset(tmp, bld, i + first_component));
               }
            } else {
               inst = bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, dest,
                               payload);
            }
            inst->mlen = 2;
            inst->offset = imm_offset;
            inst->size_written = (num_components + first_component) *
                                 inst->dst.component_size(inst->exec_size);

            /* If we are reading 64-bit data using 32-bit read messages we need
             * build proper 64-bit data elements by shuffling the low and high
             * 32-bit components around like we do for other things like UBOs
             * or SSBOs.
             */
            if (type_sz(dest.type) == 8) {
               shuffle_from_32bit_read(bld,
                                       offset(orig_dest, bld, iter * 2),
                                       retype(dest, BRW_REGISTER_TYPE_D),
                                       0, num_components);
            }

            /* If we are loading double data and we need a second read message
             * adjust the offset
             */
            if (num_iterations > 1) {
               num_components = instr->num_components - 2;
               imm_offset++;
            }
         }
      }
      break;
   }
   default:
      nir_emit_intrinsic(bld, instr);
      break;
   }
}

void
fs_visitor::nir_emit_gs_intrinsic(const fs_builder &bld,
                                  nir_intrinsic_instr *instr)
{
   assert(stage == MESA_SHADER_GEOMETRY);
   fs_reg indirect_offset;

   fs_reg dest;
   if (nir_intrinsic_infos[instr->intrinsic].has_dest)
      dest = get_nir_dest(instr->dest);

   switch (instr->intrinsic) {
   case nir_intrinsic_load_primitive_id:
      assert(stage == MESA_SHADER_GEOMETRY);
      assert(brw_gs_prog_data(prog_data)->include_primitive_id);
      bld.MOV(retype(dest, BRW_REGISTER_TYPE_UD),
              retype(fs_reg(brw_vec8_grf(2, 0)), BRW_REGISTER_TYPE_UD));
      break;

   case nir_intrinsic_load_input:
      unreachable("load_input intrinsics are invalid for the GS stage");

   case nir_intrinsic_load_per_vertex_input:
      emit_gs_input_load(dest, instr->src[0], instr->const_index[0],
                         instr->src[1], instr->num_components,
                         nir_intrinsic_component(instr));
      break;

   case nir_intrinsic_emit_vertex_with_counter:
      emit_gs_vertex(instr->src[0], instr->const_index[0]);
      break;

   case nir_intrinsic_end_primitive_with_counter:
      emit_gs_end_primitive(instr->src[0]);
      break;

   case nir_intrinsic_set_vertex_count:
      bld.MOV(this->final_gs_vertex_count, get_nir_src(instr->src[0]));
      break;

   case nir_intrinsic_load_invocation_id: {
      fs_reg val = nir_system_values[SYSTEM_VALUE_INVOCATION_ID];
      assert(val.file != BAD_FILE);
      dest.type = val.type;
      bld.MOV(dest, val);
      break;
   }

   default:
      nir_emit_intrinsic(bld, instr);
      break;
   }
}

/**
 * Fetch the current render target layer index.
 */
static fs_reg
fetch_render_target_array_index(const fs_builder &bld)
{
   if (bld.shader->devinfo->gen >= 6) {
      /* The render target array index is provided in the thread payload as
       * bits 26:16 of r0.0.
       */
      const fs_reg idx = bld.vgrf(BRW_REGISTER_TYPE_UD);
      bld.AND(idx, brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, 0, 1),
              brw_imm_uw(0x7ff));
      return idx;
   } else {
      /* Pre-SNB we only ever render into the first layer of the framebuffer
       * since layered rendering is not implemented.
       */
      return brw_imm_ud(0);
   }
}

/**
 * Fake non-coherent framebuffer read implemented using TXF to fetch from the
 * framebuffer at the current fragment coordinates and sample index.
 */
fs_inst *
fs_visitor::emit_non_coherent_fb_read(const fs_builder &bld, const fs_reg &dst,
                                      unsigned target)
{
   const struct gen_device_info *devinfo = bld.shader->devinfo;

   assert(bld.shader->stage == MESA_SHADER_FRAGMENT);
   const brw_wm_prog_key *wm_key =
      reinterpret_cast<const brw_wm_prog_key *>(key);
   assert(!wm_key->coherent_fb_fetch);
   const struct brw_wm_prog_data *wm_prog_data =
      brw_wm_prog_data(stage_prog_data);

   /* Calculate the surface index relative to the start of the texture binding
    * table block, since that's what the texturing messages expect.
    */
   const unsigned surface = target +
      wm_prog_data->binding_table.render_target_read_start -
      wm_prog_data->base.binding_table.texture_start;

   brw_mark_surface_used(
      bld.shader->stage_prog_data,
      wm_prog_data->binding_table.render_target_read_start + target);

   /* Calculate the fragment coordinates. */
   const fs_reg coords = bld.vgrf(BRW_REGISTER_TYPE_UD, 3);
   bld.MOV(offset(coords, bld, 0), pixel_x);
   bld.MOV(offset(coords, bld, 1), pixel_y);
   bld.MOV(offset(coords, bld, 2), fetch_render_target_array_index(bld));

   /* Calculate the sample index and MCS payload when multisampling.  Luckily
    * the MCS fetch message behaves deterministically for UMS surfaces, so it
    * shouldn't be necessary to recompile based on whether the framebuffer is
    * CMS or UMS.
    */
   if (wm_key->multisample_fbo &&
       nir_system_values[SYSTEM_VALUE_SAMPLE_ID].file == BAD_FILE)
      nir_system_values[SYSTEM_VALUE_SAMPLE_ID] = *emit_sampleid_setup();

   const fs_reg sample = nir_system_values[SYSTEM_VALUE_SAMPLE_ID];
   const fs_reg mcs = wm_key->multisample_fbo ?
      emit_mcs_fetch(coords, 3, brw_imm_ud(surface)) : fs_reg();

   /* Use either a normal or a CMS texel fetch message depending on whether
    * the framebuffer is single or multisample.  On SKL+ use the wide CMS
    * message just in case the framebuffer uses 16x multisampling, it should
    * be equivalent to the normal CMS fetch for lower multisampling modes.
    */
   const opcode op = !wm_key->multisample_fbo ? SHADER_OPCODE_TXF_LOGICAL :
                     devinfo->gen >= 9 ? SHADER_OPCODE_TXF_CMS_W_LOGICAL :
                     SHADER_OPCODE_TXF_CMS_LOGICAL;

   /* Emit the instruction. */
   const fs_reg srcs[] = { coords, fs_reg(), brw_imm_ud(0), fs_reg(),
                           sample, mcs,
                           brw_imm_ud(surface), brw_imm_ud(0),
                           fs_reg(), brw_imm_ud(3), brw_imm_ud(0) };
   STATIC_ASSERT(ARRAY_SIZE(srcs) == TEX_LOGICAL_NUM_SRCS);

   fs_inst *inst = bld.emit(op, dst, srcs, ARRAY_SIZE(srcs));
   inst->size_written = 4 * inst->dst.component_size(inst->exec_size);

   return inst;
}

/**
 * Actual coherent framebuffer read implemented using the native render target
 * read message.  Requires SKL+.
 */
static fs_inst *
emit_coherent_fb_read(const fs_builder &bld, const fs_reg &dst, unsigned target)
{
   assert(bld.shader->devinfo->gen >= 9);
   fs_inst *inst = bld.emit(FS_OPCODE_FB_READ_LOGICAL, dst);
   inst->target = target;
   inst->size_written = 4 * inst->dst.component_size(inst->exec_size);

   return inst;
}

static fs_reg
alloc_temporary(const fs_builder &bld, unsigned size, fs_reg *regs, unsigned n)
{
   if (n && regs[0].file != BAD_FILE) {
      return regs[0];

   } else {
      const fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_F, size);

      for (unsigned i = 0; i < n; i++)
         regs[i] = tmp;

      return tmp;
   }
}

static fs_reg
alloc_frag_output(fs_visitor *v, unsigned location)
{
   assert(v->stage == MESA_SHADER_FRAGMENT);
   const brw_wm_prog_key *const key =
      reinterpret_cast<const brw_wm_prog_key *>(v->key);
   const unsigned l = GET_FIELD(location, BRW_NIR_FRAG_OUTPUT_LOCATION);
   const unsigned i = GET_FIELD(location, BRW_NIR_FRAG_OUTPUT_INDEX);

   if (i > 0 || (key->force_dual_color_blend && l == FRAG_RESULT_DATA1))
      return alloc_temporary(v->bld, 4, &v->dual_src_output, 1);

   else if (l == FRAG_RESULT_COLOR)
      return alloc_temporary(v->bld, 4, v->outputs,
                             MAX2(key->nr_color_regions, 1));

   else if (l == FRAG_RESULT_DEPTH)
      return alloc_temporary(v->bld, 1, &v->frag_depth, 1);

   else if (l == FRAG_RESULT_STENCIL)
      return alloc_temporary(v->bld, 1, &v->frag_stencil, 1);

   else if (l == FRAG_RESULT_SAMPLE_MASK)
      return alloc_temporary(v->bld, 1, &v->sample_mask, 1);

   else if (l >= FRAG_RESULT_DATA0 &&
            l < FRAG_RESULT_DATA0 + BRW_MAX_DRAW_BUFFERS)
      return alloc_temporary(v->bld, 4,
                             &v->outputs[l - FRAG_RESULT_DATA0], 1);

   else
      unreachable("Invalid location");
}

void
fs_visitor::nir_emit_fs_intrinsic(const fs_builder &bld,
                                  nir_intrinsic_instr *instr)
{
   assert(stage == MESA_SHADER_FRAGMENT);

   fs_reg dest;
   if (nir_intrinsic_infos[instr->intrinsic].has_dest)
      dest = get_nir_dest(instr->dest);

   switch (instr->intrinsic) {
   case nir_intrinsic_load_front_face:
      bld.MOV(retype(dest, BRW_REGISTER_TYPE_D),
              *emit_frontfacing_interpolation());
      break;

   case nir_intrinsic_load_sample_pos: {
      fs_reg sample_pos = nir_system_values[SYSTEM_VALUE_SAMPLE_POS];
      assert(sample_pos.file != BAD_FILE);
      dest.type = sample_pos.type;
      bld.MOV(dest, sample_pos);
      bld.MOV(offset(dest, bld, 1), offset(sample_pos, bld, 1));
      break;
   }

   case nir_intrinsic_load_layer_id:
      dest.type = BRW_REGISTER_TYPE_UD;
      bld.MOV(dest, fetch_render_target_array_index(bld));
      break;

   case nir_intrinsic_load_helper_invocation:
   case nir_intrinsic_load_sample_mask_in:
   case nir_intrinsic_load_sample_id: {
      gl_system_value sv = nir_system_value_from_intrinsic(instr->intrinsic);
      fs_reg val = nir_system_values[sv];
      assert(val.file != BAD_FILE);
      dest.type = val.type;
      bld.MOV(dest, val);
      break;
   }

   case nir_intrinsic_store_output: {
      const fs_reg src = get_nir_src(instr->src[0]);
      const nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]);
      assert(const_offset && "Indirect output stores not allowed");
      const unsigned location = nir_intrinsic_base(instr) +
         SET_FIELD(const_offset->u32[0], BRW_NIR_FRAG_OUTPUT_LOCATION);
      const fs_reg new_dest = retype(alloc_frag_output(this, location),
                                     src.type);

      for (unsigned j = 0; j < instr->num_components; j++)
         bld.MOV(offset(new_dest, bld, nir_intrinsic_component(instr) + j),
                 offset(src, bld, j));

      break;
   }

   case nir_intrinsic_load_output: {
      const unsigned l = GET_FIELD(nir_intrinsic_base(instr),
                                   BRW_NIR_FRAG_OUTPUT_LOCATION);
      assert(l >= FRAG_RESULT_DATA0);
      nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);
      assert(const_offset && "Indirect output loads not allowed");
      const unsigned target = l - FRAG_RESULT_DATA0 + const_offset->u32[0];
      const fs_reg tmp = bld.vgrf(dest.type, 4);

      if (reinterpret_cast<const brw_wm_prog_key *>(key)->coherent_fb_fetch)
         emit_coherent_fb_read(bld, tmp, target);
      else
         emit_non_coherent_fb_read(bld, tmp, target);

      for (unsigned j = 0; j < instr->num_components; j++) {
         bld.MOV(offset(dest, bld, j),
                 offset(tmp, bld, nir_intrinsic_component(instr) + j));
      }

      break;
   }

   case nir_intrinsic_discard:
   case nir_intrinsic_discard_if: {
      /* We track our discarded pixels in f0.1.  By predicating on it, we can
       * update just the flag bits that aren't yet discarded.  If there's no
       * condition, we emit a CMP of g0 != g0, so all currently executing
       * channels will get turned off.
       */
      fs_inst *cmp;
      if (instr->intrinsic == nir_intrinsic_discard_if) {
         cmp = bld.CMP(bld.null_reg_f(), get_nir_src(instr->src[0]),
                       brw_imm_d(0), BRW_CONDITIONAL_Z);
      } else {
         fs_reg some_reg = fs_reg(retype(brw_vec8_grf(0, 0),
                                       BRW_REGISTER_TYPE_UW));
         cmp = bld.CMP(bld.null_reg_f(), some_reg, some_reg, BRW_CONDITIONAL_NZ);
      }
      cmp->predicate = BRW_PREDICATE_NORMAL;
      cmp->flag_subreg = 1;

      if (devinfo->gen >= 6) {
         emit_discard_jump();
      }

      limit_dispatch_width(16, "Fragment discard not implemented in SIMD32 mode.");
      break;
   }

   case nir_intrinsic_load_input: {
      /* load_input is only used for flat inputs */
      unsigned base = nir_intrinsic_base(instr);
      unsigned comp = nir_intrinsic_component(instr);
      unsigned num_components = instr->num_components;
      fs_reg orig_dest = dest;
      enum brw_reg_type type = dest.type;

      /* Special case fields in the VUE header */
      if (base == VARYING_SLOT_LAYER)
         comp = 1;
      else if (base == VARYING_SLOT_VIEWPORT)
         comp = 2;

      if (nir_dest_bit_size(instr->dest) == 64) {
         /* const_index is in 32-bit type size units that could not be aligned
          * with DF. We need to read the double vector as if it was a float
          * vector of twice the number of components to fetch the right data.
          */
         type = BRW_REGISTER_TYPE_F;
         num_components *= 2;
         dest = bld.vgrf(type, num_components);
      }

      for (unsigned int i = 0; i < num_components; i++) {
         bld.MOV(offset(retype(dest, type), bld, i),
                 retype(component(interp_reg(base, comp + i), 3), type));
      }

      if (nir_dest_bit_size(instr->dest) == 64) {
         shuffle_from_32bit_read(bld, orig_dest, dest, 0,
                                 instr->num_components);
      }
      break;
   }

   case nir_intrinsic_load_barycentric_pixel:
   case nir_intrinsic_load_barycentric_centroid:
   case nir_intrinsic_load_barycentric_sample:
      /* Do nothing - load_interpolated_input handling will handle it later. */
      break;

   case nir_intrinsic_load_barycentric_at_sample: {
      const glsl_interp_mode interpolation =
         (enum glsl_interp_mode) nir_intrinsic_interp_mode(instr);

      nir_const_value *const_sample = nir_src_as_const_value(instr->src[0]);

      if (const_sample) {
         unsigned msg_data = const_sample->i32[0] << 4;

         emit_pixel_interpolater_send(bld,
                                      FS_OPCODE_INTERPOLATE_AT_SAMPLE,
                                      dest,
                                      fs_reg(), /* src */
                                      brw_imm_ud(msg_data),
                                      interpolation);
      } else {
         const fs_reg sample_src = retype(get_nir_src(instr->src[0]),
                                          BRW_REGISTER_TYPE_UD);

         if (nir_src_is_dynamically_uniform(instr->src[0])) {
            const fs_reg sample_id = bld.emit_uniformize(sample_src);
            const fs_reg msg_data = vgrf(glsl_type::uint_type);
            bld.exec_all().group(1, 0)
               .SHL(msg_data, sample_id, brw_imm_ud(4u));
            emit_pixel_interpolater_send(bld,
                                         FS_OPCODE_INTERPOLATE_AT_SAMPLE,
                                         dest,
                                         fs_reg(), /* src */
                                         msg_data,
                                         interpolation);
         } else {
            /* Make a loop that sends a message to the pixel interpolater
             * for the sample number in each live channel. If there are
             * multiple channels with the same sample number then these
             * will be handled simultaneously with a single interation of
             * the loop.
             */
            bld.emit(BRW_OPCODE_DO);

            /* Get the next live sample number into sample_id_reg */
            const fs_reg sample_id = bld.emit_uniformize(sample_src);

            /* Set the flag register so that we can perform the send
             * message on all channels that have the same sample number
             */
            bld.CMP(bld.null_reg_ud(),
                    sample_src, sample_id,
                    BRW_CONDITIONAL_EQ);
            const fs_reg msg_data = vgrf(glsl_type::uint_type);
            bld.exec_all().group(1, 0)
               .SHL(msg_data, sample_id, brw_imm_ud(4u));
            fs_inst *inst =
               emit_pixel_interpolater_send(bld,
                                            FS_OPCODE_INTERPOLATE_AT_SAMPLE,
                                            dest,
                                            fs_reg(), /* src */
                                            component(msg_data, 0),
                                            interpolation);
            set_predicate(BRW_PREDICATE_NORMAL, inst);

            /* Continue the loop if there are any live channels left */
            set_predicate_inv(BRW_PREDICATE_NORMAL,
                              true, /* inverse */
                              bld.emit(BRW_OPCODE_WHILE));
         }
      }
      break;
   }

   case nir_intrinsic_load_barycentric_at_offset: {
      const glsl_interp_mode interpolation =
         (enum glsl_interp_mode) nir_intrinsic_interp_mode(instr);

      nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);

      if (const_offset) {
         unsigned off_x = MIN2((int)(const_offset->f32[0] * 16), 7) & 0xf;
         unsigned off_y = MIN2((int)(const_offset->f32[1] * 16), 7) & 0xf;

         emit_pixel_interpolater_send(bld,
                                      FS_OPCODE_INTERPOLATE_AT_SHARED_OFFSET,
                                      dest,
                                      fs_reg(), /* src */
                                      brw_imm_ud(off_x | (off_y << 4)),
                                      interpolation);
      } else {
         fs_reg src = vgrf(glsl_type::ivec2_type);
         fs_reg offset_src = retype(get_nir_src(instr->src[0]),
                                    BRW_REGISTER_TYPE_F);
         for (int i = 0; i < 2; i++) {
            fs_reg temp = vgrf(glsl_type::float_type);
            bld.MUL(temp, offset(offset_src, bld, i), brw_imm_f(16.0f));
            fs_reg itemp = vgrf(glsl_type::int_type);
            /* float to int */
            bld.MOV(itemp, temp);

            /* Clamp the upper end of the range to +7/16.
             * ARB_gpu_shader5 requires that we support a maximum offset
             * of +0.5, which isn't representable in a S0.4 value -- if
             * we didn't clamp it, we'd end up with -8/16, which is the
             * opposite of what the shader author wanted.
             *
             * This is legal due to ARB_gpu_shader5's quantization
             * rules:
             *
             * "Not all values of <offset> may be supported; x and y
             * offsets may be rounded to fixed-point values with the
             * number of fraction bits given by the
             * implementation-dependent constant
             * FRAGMENT_INTERPOLATION_OFFSET_BITS"
             */
            set_condmod(BRW_CONDITIONAL_L,
                        bld.SEL(offset(src, bld, i), itemp, brw_imm_d(7)));
         }

         const enum opcode opcode = FS_OPCODE_INTERPOLATE_AT_PER_SLOT_OFFSET;
         emit_pixel_interpolater_send(bld,
                                      opcode,
                                      dest,
                                      src,
                                      brw_imm_ud(0u),
                                      interpolation);
      }
      break;
   }

   case nir_intrinsic_load_interpolated_input: {
      if (nir_intrinsic_base(instr) == VARYING_SLOT_POS) {
         emit_fragcoord_interpolation(dest);
         break;
      }

      assert(instr->src[0].ssa &&
             instr->src[0].ssa->parent_instr->type == nir_instr_type_intrinsic);
      nir_intrinsic_instr *bary_intrinsic =
         nir_instr_as_intrinsic(instr->src[0].ssa->parent_instr);
      nir_intrinsic_op bary_intrin = bary_intrinsic->intrinsic;
      enum glsl_interp_mode interp_mode =
         (enum glsl_interp_mode) nir_intrinsic_interp_mode(bary_intrinsic);
      fs_reg dst_xy;

      if (bary_intrin == nir_intrinsic_load_barycentric_at_offset ||
          bary_intrin == nir_intrinsic_load_barycentric_at_sample) {
         /* Use the result of the PI message */
         dst_xy = retype(get_nir_src(instr->src[0]), BRW_REGISTER_TYPE_F);
      } else {
         /* Use the delta_xy values computed from the payload */
         enum brw_barycentric_mode bary =
            brw_barycentric_mode(interp_mode, bary_intrin);

         dst_xy = this->delta_xy[bary];
      }

      for (unsigned int i = 0; i < instr->num_components; i++) {
         fs_reg interp =
            component(interp_reg(nir_intrinsic_base(instr),
                                 nir_intrinsic_component(instr) + i), 0);
         interp.type = BRW_REGISTER_TYPE_F;
         dest.type = BRW_REGISTER_TYPE_F;

         if (devinfo->gen < 6 && interp_mode == INTERP_MODE_SMOOTH) {
            fs_reg tmp = vgrf(glsl_type::float_type);
            bld.emit(FS_OPCODE_LINTERP, tmp, dst_xy, interp);
            bld.MUL(offset(dest, bld, i), tmp, this->pixel_w);
         } else {
            bld.emit(FS_OPCODE_LINTERP, offset(dest, bld, i), dst_xy, interp);
         }
      }
      break;
   }

   default:
      nir_emit_intrinsic(bld, instr);
      break;
   }
}

void
fs_visitor::nir_emit_cs_intrinsic(const fs_builder &bld,
                                  nir_intrinsic_instr *instr)
{
   assert(stage == MESA_SHADER_COMPUTE);
   struct brw_cs_prog_data *cs_prog_data = brw_cs_prog_data(prog_data);

   fs_reg dest;
   if (nir_intrinsic_infos[instr->intrinsic].has_dest)
      dest = get_nir_dest(instr->dest);

   switch (instr->intrinsic) {
   case nir_intrinsic_barrier:
      emit_barrier();
      cs_prog_data->uses_barrier = true;
      break;

   case nir_intrinsic_load_subgroup_id:
      bld.MOV(retype(dest, BRW_REGISTER_TYPE_UD), subgroup_id);
      break;

   case nir_intrinsic_load_local_invocation_id:
   case nir_intrinsic_load_work_group_id: {
      gl_system_value sv = nir_system_value_from_intrinsic(instr->intrinsic);
      fs_reg val = nir_system_values[sv];
      assert(val.file != BAD_FILE);
      dest.type = val.type;
      for (unsigned i = 0; i < 3; i++)
         bld.MOV(offset(dest, bld, i), offset(val, bld, i));
      break;
   }

   case nir_intrinsic_load_num_work_groups: {
      const unsigned surface =
         cs_prog_data->binding_table.work_groups_start;

      cs_prog_data->uses_num_work_groups = true;

      fs_reg surf_index = brw_imm_ud(surface);
      brw_mark_surface_used(prog_data, surface);

      /* Read the 3 GLuint components of gl_NumWorkGroups */
      for (unsigned i = 0; i < 3; i++) {
         fs_reg read_result =
            emit_untyped_read(bld, surf_index,
                              brw_imm_ud(i << 2),
                              1 /* dims */, 1 /* size */,
                              BRW_PREDICATE_NONE);
         read_result.type = dest.type;
         bld.MOV(dest, read_result);
         dest = offset(dest, bld, 1);
      }
      break;
   }

   case nir_intrinsic_shared_atomic_add:
      nir_emit_shared_atomic(bld, BRW_AOP_ADD, instr);
      break;
   case nir_intrinsic_shared_atomic_imin:
      nir_emit_shared_atomic(bld, BRW_AOP_IMIN, instr);
      break;
   case nir_intrinsic_shared_atomic_umin:
      nir_emit_shared_atomic(bld, BRW_AOP_UMIN, instr);
      break;
   case nir_intrinsic_shared_atomic_imax:
      nir_emit_shared_atomic(bld, BRW_AOP_IMAX, instr);
      break;
   case nir_intrinsic_shared_atomic_umax:
      nir_emit_shared_atomic(bld, BRW_AOP_UMAX, instr);
      break;
   case nir_intrinsic_shared_atomic_and:
      nir_emit_shared_atomic(bld, BRW_AOP_AND, instr);
      break;
   case nir_intrinsic_shared_atomic_or:
      nir_emit_shared_atomic(bld, BRW_AOP_OR, instr);
      break;
   case nir_intrinsic_shared_atomic_xor:
      nir_emit_shared_atomic(bld, BRW_AOP_XOR, instr);
      break;
   case nir_intrinsic_shared_atomic_exchange:
      nir_emit_shared_atomic(bld, BRW_AOP_MOV, instr);
      break;
   case nir_intrinsic_shared_atomic_comp_swap:
      nir_emit_shared_atomic(bld, BRW_AOP_CMPWR, instr);
      break;

   case nir_intrinsic_load_shared: {
      assert(devinfo->gen >= 7);

      fs_reg surf_index = brw_imm_ud(GEN7_BTI_SLM);

      /* Get the offset to read from */
      fs_reg offset_reg;
      nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);
      if (const_offset) {
         offset_reg = brw_imm_ud(instr->const_index[0] + const_offset->u32[0]);
      } else {
         offset_reg = vgrf(glsl_type::uint_type);
         bld.ADD(offset_reg,
                 retype(get_nir_src(instr->src[0]), BRW_REGISTER_TYPE_UD),
                 brw_imm_ud(instr->const_index[0]));
      }

      /* Read the vector */
      do_untyped_vector_read(bld, dest, surf_index, offset_reg,
                             instr->num_components);
      break;
   }

   case nir_intrinsic_store_shared: {
      assert(devinfo->gen >= 7);

      /* Block index */
      fs_reg surf_index = brw_imm_ud(GEN7_BTI_SLM);

      /* Value */
      fs_reg val_reg = get_nir_src(instr->src[0]);

      /* Writemask */
      unsigned writemask = instr->const_index[1];

      /* get_nir_src() retypes to integer. Be wary of 64-bit types though
       * since the untyped writes below operate in units of 32-bits, which
       * means that we need to write twice as many components each time.
       * Also, we have to suffle 64-bit data to be in the appropriate layout
       * expected by our 32-bit write messages.
       */
      unsigned type_size = 4;
      if (nir_src_bit_size(instr->src[0]) == 64) {
         type_size = 8;
         val_reg = shuffle_for_32bit_write(bld, val_reg, 0,
                                           instr->num_components);
      }

      unsigned type_slots = type_size / 4;

      /* Combine groups of consecutive enabled channels in one write
       * message. We use ffs to find the first enabled channel and then ffs on
       * the bit-inverse, down-shifted writemask to determine the length of
       * the block of enabled bits.
       */
      while (writemask) {
         unsigned first_component = ffs(writemask) - 1;
         unsigned length = ffs(~(writemask >> first_component)) - 1;

         /* We can't write more than 2 64-bit components at once. Limit the
          * length of the write to what we can do and let the next iteration
          * handle the rest
          */
         if (type_size > 4)
            length = MIN2(2, length);

         fs_reg offset_reg;
         nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]);
         if (const_offset) {
            offset_reg = brw_imm_ud(instr->const_index[0] + const_offset->u32[0] +
                                    type_size * first_component);
         } else {
            offset_reg = vgrf(glsl_type::uint_type);
            bld.ADD(offset_reg,
                    retype(get_nir_src(instr->src[1]), BRW_REGISTER_TYPE_UD),
                    brw_imm_ud(instr->const_index[0] + type_size * first_component));
         }

         emit_untyped_write(bld, surf_index, offset_reg,
                            offset(val_reg, bld, first_component * type_slots),
                            1 /* dims */, length * type_slots,
                            BRW_PREDICATE_NONE);

         /* Clear the bits in the writemask that we just wrote, then try
          * again to see if more channels are left.
          */
         writemask &= (15 << (first_component + length));
      }

      break;
   }

   default:
      nir_emit_intrinsic(bld, instr);
      break;
   }
}

static fs_reg
brw_nir_reduction_op_identity(const fs_builder &bld,
                              nir_op op, brw_reg_type type)
{
   nir_const_value value = nir_alu_binop_identity(op, type_sz(type) * 8);
   switch (type_sz(type)) {
   case 2:
      assert(type != BRW_REGISTER_TYPE_HF);
      return retype(brw_imm_uw(value.u16[0]), type);
   case 4:
      return retype(brw_imm_ud(value.u32[0]), type);
   case 8:
      if (type == BRW_REGISTER_TYPE_DF)
         return setup_imm_df(bld, value.f64[0]);
      else
         return retype(brw_imm_u64(value.u64[0]), type);
   default:
      unreachable("Invalid type size");
   }
}

static opcode
brw_op_for_nir_reduction_op(nir_op op)
{
   switch (op) {
   case nir_op_iadd: return BRW_OPCODE_ADD;
   case nir_op_fadd: return BRW_OPCODE_ADD;
   case nir_op_imul: return BRW_OPCODE_MUL;
   case nir_op_fmul: return BRW_OPCODE_MUL;
   case nir_op_imin: return BRW_OPCODE_SEL;
   case nir_op_umin: return BRW_OPCODE_SEL;
   case nir_op_fmin: return BRW_OPCODE_SEL;
   case nir_op_imax: return BRW_OPCODE_SEL;
   case nir_op_umax: return BRW_OPCODE_SEL;
   case nir_op_fmax: return BRW_OPCODE_SEL;
   case nir_op_iand: return BRW_OPCODE_AND;
   case nir_op_ior:  return BRW_OPCODE_OR;
   case nir_op_ixor: return BRW_OPCODE_XOR;
   default:
      unreachable("Invalid reduction operation");
   }
}

static brw_conditional_mod
brw_cond_mod_for_nir_reduction_op(nir_op op)
{
   switch (op) {
   case nir_op_iadd: return BRW_CONDITIONAL_NONE;
   case nir_op_fadd: return BRW_CONDITIONAL_NONE;
   case nir_op_imul: return BRW_CONDITIONAL_NONE;
   case nir_op_fmul: return BRW_CONDITIONAL_NONE;
   case nir_op_imin: return BRW_CONDITIONAL_L;
   case nir_op_umin: return BRW_CONDITIONAL_L;
   case nir_op_fmin: return BRW_CONDITIONAL_L;
   case nir_op_imax: return BRW_CONDITIONAL_GE;
   case nir_op_umax: return BRW_CONDITIONAL_GE;
   case nir_op_fmax: return BRW_CONDITIONAL_GE;
   case nir_op_iand: return BRW_CONDITIONAL_NONE;
   case nir_op_ior:  return BRW_CONDITIONAL_NONE;
   case nir_op_ixor: return BRW_CONDITIONAL_NONE;
   default:
      unreachable("Invalid reduction operation");
   }
}

void
fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr)
{
   fs_reg dest;
   if (nir_intrinsic_infos[instr->intrinsic].has_dest)
      dest = get_nir_dest(instr->dest);

   switch (instr->intrinsic) {
   case nir_intrinsic_image_deref_load:
   case nir_intrinsic_image_deref_store:
   case nir_intrinsic_image_deref_atomic_add:
   case nir_intrinsic_image_deref_atomic_min:
   case nir_intrinsic_image_deref_atomic_max:
   case nir_intrinsic_image_deref_atomic_and:
   case nir_intrinsic_image_deref_atomic_or:
   case nir_intrinsic_image_deref_atomic_xor:
   case nir_intrinsic_image_deref_atomic_exchange:
   case nir_intrinsic_image_deref_atomic_comp_swap: {
      using namespace image_access;

      if (stage == MESA_SHADER_FRAGMENT &&
          instr->intrinsic != nir_intrinsic_image_deref_load)
         brw_wm_prog_data(prog_data)->has_side_effects = true;

      /* Get the referenced image variable and type. */
      nir_deref_instr *deref = nir_src_as_deref(instr->src[0]);
      const nir_variable *var = nir_deref_instr_get_variable(deref);
      const glsl_type *type = var->type->without_array();
      const brw_reg_type base_type = get_image_base_type(type);

      /* Get some metadata from the image intrinsic. */
      const nir_intrinsic_info *info = &nir_intrinsic_infos[instr->intrinsic];
      const unsigned arr_dims = type->sampler_array ? 1 : 0;
      const unsigned surf_dims = type->coordinate_components() - arr_dims;
      const unsigned format = var->data.image.format;
      const unsigned dest_components = nir_intrinsic_dest_components(instr);

      /* Get the arguments of the image intrinsic. */
      const fs_reg image = get_nir_image_deref(deref);
      const fs_reg addr = retype(get_nir_src(instr->src[1]),
                                 BRW_REGISTER_TYPE_UD);
      const fs_reg src0 = (info->num_srcs >= 4 ?
                           retype(get_nir_src(instr->src[3]), base_type) :
                           fs_reg());
      const fs_reg src1 = (info->num_srcs >= 5 ?
                           retype(get_nir_src(instr->src[4]), base_type) :
                           fs_reg());
      fs_reg tmp;

      /* Emit an image load, store or atomic op. */
      if (instr->intrinsic == nir_intrinsic_image_deref_load)
         tmp = emit_image_load(bld, image, addr, surf_dims, arr_dims, format);

      else if (instr->intrinsic == nir_intrinsic_image_deref_store)
         emit_image_store(bld, image, addr, src0, surf_dims, arr_dims,
                          var->data.image.write_only ? GL_NONE : format);

      else
         tmp = emit_image_atomic(bld, image, addr, src0, src1,
                                 surf_dims, arr_dims, dest_components,
                                 get_image_atomic_op(instr->intrinsic, type));

      /* Assign the result. */
      for (unsigned c = 0; c < dest_components; ++c) {
         bld.MOV(offset(retype(dest, base_type), bld, c),
               offset(tmp, bld, c));
      }
      break;
   }

   case nir_intrinsic_memory_barrier_atomic_counter:
   case nir_intrinsic_memory_barrier_buffer:
   case nir_intrinsic_memory_barrier_image:
   case nir_intrinsic_memory_barrier: {
      const fs_builder ubld = bld.group(8, 0);
      const fs_reg tmp = ubld.vgrf(BRW_REGISTER_TYPE_UD, 2);
      ubld.emit(SHADER_OPCODE_MEMORY_FENCE, tmp)
         ->size_written = 2 * REG_SIZE;
      break;
   }

   case nir_intrinsic_group_memory_barrier:
   case nir_intrinsic_memory_barrier_shared:
      /* We treat these workgroup-level barriers as no-ops.  This should be
       * safe at present and as long as:
       *
       *  - Memory access instructions are not subsequently reordered by the
       *    compiler back-end.
       *
       *  - All threads from a given compute shader workgroup fit within a
       *    single subslice and therefore talk to the same HDC shared unit
       *    what supposedly guarantees ordering and coherency between threads
       *    from the same workgroup.  This may change in the future when we
       *    start splitting workgroups across multiple subslices.
       *
       *  - The context is not in fault-and-stream mode, which could cause
       *    memory transactions (including to SLM) prior to the barrier to be
       *    replayed after the barrier if a pagefault occurs.  This shouldn't
       *    be a problem up to and including SKL because fault-and-stream is
       *    not usable due to hardware issues, but that's likely to change in
       *    the future.
       */
      break;

   case nir_intrinsic_shader_clock: {
      /* We cannot do anything if there is an event, so ignore it for now */
      const fs_reg shader_clock = get_timestamp(bld);
      const fs_reg srcs[] = { component(shader_clock, 0),
                              component(shader_clock, 1) };
      bld.LOAD_PAYLOAD(dest, srcs, ARRAY_SIZE(srcs), 0);
      break;
   }

   case nir_intrinsic_image_deref_size: {
      /* Get the referenced image variable and type. */
      nir_deref_instr *deref = nir_src_as_deref(instr->src[0]);
      const nir_variable *var = nir_deref_instr_get_variable(deref);
      const glsl_type *type = var->type->without_array();

      /* Get the size of the image. */
      const fs_reg image = get_nir_image_deref(deref);
      const fs_reg size = offset(image, bld, BRW_IMAGE_PARAM_SIZE_OFFSET);

      /* For 1DArray image types, the array index is stored in the Z component.
       * Fix this by swizzling the Z component to the Y component.
       */
      const bool is_1d_array_image =
                  type->sampler_dimensionality == GLSL_SAMPLER_DIM_1D &&
                  type->sampler_array;

      /* For CubeArray images, we should count the number of cubes instead
       * of the number of faces. Fix it by dividing the (Z component) by 6.
       */
      const bool is_cube_array_image =
                  type->sampler_dimensionality == GLSL_SAMPLER_DIM_CUBE &&
                  type->sampler_array;

      /* Copy all the components. */
      for (unsigned c = 0; c < instr->dest.ssa.num_components; ++c) {
         if ((int)c >= type->coordinate_components()) {
             bld.MOV(offset(retype(dest, BRW_REGISTER_TYPE_D), bld, c),
                     brw_imm_d(1));
         } else if (c == 1 && is_1d_array_image) {
            bld.MOV(offset(retype(dest, BRW_REGISTER_TYPE_D), bld, c),
                    offset(size, bld, 2));
         } else if (c == 2 && is_cube_array_image) {
            bld.emit(SHADER_OPCODE_INT_QUOTIENT,
                     offset(retype(dest, BRW_REGISTER_TYPE_D), bld, c),
                     offset(size, bld, c), brw_imm_d(6));
         } else {
            bld.MOV(offset(retype(dest, BRW_REGISTER_TYPE_D), bld, c),
                    offset(size, bld, c));
         }
       }

      break;
   }

   case nir_intrinsic_image_deref_samples:
      /* The driver does not support multi-sampled images. */
      bld.MOV(retype(dest, BRW_REGISTER_TYPE_D), brw_imm_d(1));
      break;

   case nir_intrinsic_load_uniform: {
      /* Offsets are in bytes but they should always aligned to
       * the type size
       */
      assert(instr->const_index[0] % 4 == 0 ||
             instr->const_index[0] % type_sz(dest.type) == 0);

      fs_reg src(UNIFORM, instr->const_index[0] / 4, dest.type);

      nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);
      if (const_offset) {
         assert(const_offset->u32[0] % type_sz(dest.type) == 0);
         /* For 16-bit types we add the module of the const_index[0]
          * offset to access to not 32-bit aligned element
          */
         src.offset = const_offset->u32[0] + instr->const_index[0] % 4;

         for (unsigned j = 0; j < instr->num_components; j++) {
            bld.MOV(offset(dest, bld, j), offset(src, bld, j));
         }
      } else {
         fs_reg indirect = retype(get_nir_src(instr->src[0]),
                                  BRW_REGISTER_TYPE_UD);

         /* We need to pass a size to the MOV_INDIRECT but we don't want it to
          * go past the end of the uniform.  In order to keep the n'th
          * component from running past, we subtract off the size of all but
          * one component of the vector.
          */
         assert(instr->const_index[1] >=
                instr->num_components * (int) type_sz(dest.type));
         unsigned read_size = instr->const_index[1] -
            (instr->num_components - 1) * type_sz(dest.type);

         bool supports_64bit_indirects =
            !devinfo->is_cherryview && !gen_device_info_is_9lp(devinfo);

         if (type_sz(dest.type) != 8 || supports_64bit_indirects) {
            for (unsigned j = 0; j < instr->num_components; j++) {
               bld.emit(SHADER_OPCODE_MOV_INDIRECT,
                        offset(dest, bld, j), offset(src, bld, j),
                        indirect, brw_imm_ud(read_size));
            }
         } else {
            const unsigned num_mov_indirects =
               type_sz(dest.type) / type_sz(BRW_REGISTER_TYPE_UD);
            /* We read a little bit less per MOV INDIRECT, as they are now
             * 32-bits ones instead of 64-bit. Fix read_size then.
             */
            const unsigned read_size_32bit = read_size -
                (num_mov_indirects - 1) * type_sz(BRW_REGISTER_TYPE_UD);
            for (unsigned j = 0; j < instr->num_components; j++) {
               for (unsigned i = 0; i < num_mov_indirects; i++) {
                  bld.emit(SHADER_OPCODE_MOV_INDIRECT,
                           subscript(offset(dest, bld, j), BRW_REGISTER_TYPE_UD, i),
                           subscript(offset(src, bld, j), BRW_REGISTER_TYPE_UD, i),
                           indirect, brw_imm_ud(read_size_32bit));
               }
            }
         }
      }
      break;
   }

   case nir_intrinsic_load_ubo: {
      nir_const_value *const_index = nir_src_as_const_value(instr->src[0]);
      fs_reg surf_index;

      if (const_index) {
         const unsigned index = stage_prog_data->binding_table.ubo_start +
                                const_index->u32[0];
         surf_index = brw_imm_ud(index);
         brw_mark_surface_used(prog_data, index);
      } else {
         /* The block index is not a constant. Evaluate the index expression
          * per-channel and add the base UBO index; we have to select a value
          * from any live channel.
          */
         surf_index = vgrf(glsl_type::uint_type);
         bld.ADD(surf_index, get_nir_src(instr->src[0]),
                 brw_imm_ud(stage_prog_data->binding_table.ubo_start));
         surf_index = bld.emit_uniformize(surf_index);

         /* Assume this may touch any UBO. It would be nice to provide
          * a tighter bound, but the array information is already lowered away.
          */
         brw_mark_surface_used(prog_data,
                               stage_prog_data->binding_table.ubo_start +
                               nir->info.num_ubos - 1);
      }

      nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]);
      if (const_offset == NULL) {
         fs_reg base_offset = retype(get_nir_src(instr->src[1]),
                                     BRW_REGISTER_TYPE_UD);

         for (int i = 0; i < instr->num_components; i++)
            VARYING_PULL_CONSTANT_LOAD(bld, offset(dest, bld, i), surf_index,
                                       base_offset, i * type_sz(dest.type));
      } else {
         /* Even if we are loading doubles, a pull constant load will load
          * a 32-bit vec4, so should only reserve vgrf space for that. If we
          * need to load a full dvec4 we will have to emit 2 loads. This is
          * similar to demote_pull_constants(), except that in that case we
          * see individual accesses to each component of the vector and then
          * we let CSE deal with duplicate loads. Here we see a vector access
          * and we have to split it if necessary.
          */
         const unsigned type_size = type_sz(dest.type);

         /* See if we've selected this as a push constant candidate */
         if (const_index) {
            const unsigned ubo_block = const_index->u32[0];
            const unsigned offset_256b = const_offset->u32[0] / 32;

            fs_reg push_reg;
            for (int i = 0; i < 4; i++) {
               const struct brw_ubo_range *range = &prog_data->ubo_ranges[i];
               if (range->block == ubo_block &&
                   offset_256b >= range->start &&
                   offset_256b < range->start + range->length) {

                  push_reg = fs_reg(UNIFORM, UBO_START + i, dest.type);
                  push_reg.offset = const_offset->u32[0] - 32 * range->start;
                  break;
               }
            }

            if (push_reg.file != BAD_FILE) {
               for (unsigned i = 0; i < instr->num_components; i++) {
                  bld.MOV(offset(dest, bld, i),
                          byte_offset(push_reg, i * type_size));
               }
               break;
            }
         }

         const unsigned block_sz = 64; /* Fetch one cacheline at a time. */
         const fs_builder ubld = bld.exec_all().group(block_sz / 4, 0);
         const fs_reg packed_consts = ubld.vgrf(BRW_REGISTER_TYPE_UD);

         for (unsigned c = 0; c < instr->num_components;) {
            const unsigned base = const_offset->u32[0] + c * type_size;
            /* Number of usable components in the next block-aligned load. */
            const unsigned count = MIN2(instr->num_components - c,
                                        (block_sz - base % block_sz) / type_size);

            ubld.emit(FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD,
                      packed_consts, surf_index,
                      brw_imm_ud(base & ~(block_sz - 1)));

            const fs_reg consts =
               retype(byte_offset(packed_consts, base & (block_sz - 1)),
                      dest.type);

            for (unsigned d = 0; d < count; d++)
               bld.MOV(offset(dest, bld, c + d), component(consts, d));

            c += count;
         }
      }
      break;
   }

   case nir_intrinsic_load_ssbo: {
      assert(devinfo->gen >= 7);

      nir_const_value *const_uniform_block =
         nir_src_as_const_value(instr->src[0]);

      fs_reg surf_index;
      if (const_uniform_block) {
         unsigned index = stage_prog_data->binding_table.ssbo_start +
                          const_uniform_block->u32[0];
         surf_index = brw_imm_ud(index);
         brw_mark_surface_used(prog_data, index);
      } else {
         surf_index = vgrf(glsl_type::uint_type);
         bld.ADD(surf_index, get_nir_src(instr->src[0]),
                 brw_imm_ud(stage_prog_data->binding_table.ssbo_start));

         /* Assume this may touch any UBO. It would be nice to provide
          * a tighter bound, but the array information is already lowered away.
          */
         brw_mark_surface_used(prog_data,
                               stage_prog_data->binding_table.ssbo_start +
                               nir->info.num_ssbos - 1);
      }

      fs_reg offset_reg;
      nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]);
      if (const_offset) {
         offset_reg = brw_imm_ud(const_offset->u32[0]);
      } else {
         offset_reg = retype(get_nir_src(instr->src[1]), BRW_REGISTER_TYPE_UD);
      }

      /* Read the vector */
      do_untyped_vector_read(bld, dest, surf_index, offset_reg,
                             instr->num_components);

      break;
   }

   case nir_intrinsic_store_ssbo: {
      assert(devinfo->gen >= 7);

      if (stage == MESA_SHADER_FRAGMENT)
         brw_wm_prog_data(prog_data)->has_side_effects = true;

      /* Block index */
      fs_reg surf_index;
      nir_const_value *const_uniform_block =
         nir_src_as_const_value(instr->src[1]);
      if (const_uniform_block) {
         unsigned index = stage_prog_data->binding_table.ssbo_start +
                          const_uniform_block->u32[0];
         surf_index = brw_imm_ud(index);
         brw_mark_surface_used(prog_data, index);
      } else {
         surf_index = vgrf(glsl_type::uint_type);
         bld.ADD(surf_index, get_nir_src(instr->src[1]),
                  brw_imm_ud(stage_prog_data->binding_table.ssbo_start));

         brw_mark_surface_used(prog_data,
                               stage_prog_data->binding_table.ssbo_start +
                               nir->info.num_ssbos - 1);
      }

      /* Value */
      fs_reg val_reg = get_nir_src(instr->src[0]);

      /* Writemask */
      unsigned writemask = instr->const_index[0];

      /* get_nir_src() retypes to integer. Be wary of 64-bit types though
       * since the untyped writes below operate in units of 32-bits, which
       * means that we need to write twice as many components each time.
       * Also, we have to suffle 64-bit data to be in the appropriate layout
       * expected by our 32-bit write messages.
       */
      unsigned bit_size = nir_src_bit_size(instr->src[0]);
      unsigned type_size = bit_size / 8;

      /* Combine groups of consecutive enabled channels in one write
       * message. We use ffs to find the first enabled channel and then ffs on
       * the bit-inverse, down-shifted writemask to determine the num_components
       * of the block of enabled bits.
       */
      while (writemask) {
         unsigned first_component = ffs(writemask) - 1;
         unsigned num_components = ffs(~(writemask >> first_component)) - 1;
         fs_reg write_src = offset(val_reg, bld, first_component);

         nir_const_value *const_offset = nir_src_as_const_value(instr->src[2]);

         if (type_size > 4) {
            /* We can't write more than 2 64-bit components at once. Limit
             * the num_components of the write to what we can do and let the next
             * iteration handle the rest.
             */
            num_components = MIN2(2, num_components);
            write_src = shuffle_for_32bit_write(bld, write_src, 0,
                                                num_components);
         } else if (type_size < 4) {
            assert(type_size == 2);
            /* For 16-bit types we pack two consecutive values into a 32-bit
             * word and use an untyped write message. For single values or not
             * 32-bit-aligned we need to use byte-scattered writes because
             * untyped writes works with 32-bit components with 32-bit
             * alignment. byte_scattered_write messages only support one
             * 16-bit component at a time. As VK_KHR_relaxed_block_layout
             * could be enabled we can not guarantee that not constant offsets
             * to be 32-bit aligned for 16-bit types. For example an array, of
             * 16-bit vec3 with array element stride of 6.
             *
             * In the case of 32-bit aligned constant offsets if there is
             * a 3-components vector we submit one untyped-write message
             * of 32-bit (first two components), and one byte-scattered
             * write message (the last component).
             */

            if ( !const_offset || ((const_offset->u32[0] +
                                   type_size * first_component) % 4)) {
               /* If we use a .yz writemask we also need to emit 2
                * byte-scattered write messages because of y-component not
                * being aligned to 32-bit.
                */
               num_components = 1;
            } else if (num_components > 2 && (num_components % 2)) {
               /* If there is an odd number of consecutive components we left
                * the not paired component for a following emit of length == 1
                * with byte_scattered_write.
                */
               num_components --;
            }
            /* For num_components == 1 we are also shuffling the component
             * because byte scattered writes of 16-bit need values to be dword
             * aligned. Shuffling only one component would be the same as
             * striding it.
             */
            write_src = shuffle_for_32bit_write(bld, write_src, 0,
                                                num_components);
         }

         fs_reg offset_reg;

         if (const_offset) {
            offset_reg = brw_imm_ud(const_offset->u32[0] +
                                    type_size * first_component);
         } else {
            offset_reg = vgrf(glsl_type::uint_type);
            bld.ADD(offset_reg,
                    retype(get_nir_src(instr->src[2]), BRW_REGISTER_TYPE_UD),
                    brw_imm_ud(type_size * first_component));
         }

         if (type_size < 4 && num_components == 1) {
            assert(type_size == 2);
            /* Untyped Surface messages have a fixed 32-bit size, so we need
             * to rely on byte scattered in order to write 16-bit elements.
             * The byte_scattered_write message needs that every written 16-bit
             * type to be aligned 32-bits (stride=2).
             */
            emit_byte_scattered_write(bld, surf_index, offset_reg,
                                      write_src,
                                      1 /* dims */, 1,
                                      bit_size,
                                      BRW_PREDICATE_NONE);
         } else {
            assert(num_components * type_size <= 16);
            assert((num_components * type_size) % 4 == 0);
            assert(offset_reg.file != BRW_IMMEDIATE_VALUE ||
                   offset_reg.ud % 4 == 0);
            unsigned num_slots = (num_components * type_size) / 4;

            emit_untyped_write(bld, surf_index, offset_reg,
                               write_src,
                               1 /* dims */, num_slots,
                               BRW_PREDICATE_NONE);
         }

         /* Clear the bits in the writemask that we just wrote, then try
          * again to see if more channels are left.
          */
         writemask &= (15 << (first_component + num_components));
      }
      break;
   }

   case nir_intrinsic_store_output: {
      fs_reg src = get_nir_src(instr->src[0]);

      nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]);
      assert(const_offset && "Indirect output stores not allowed");

      unsigned num_components = instr->num_components;
      unsigned first_component = nir_intrinsic_component(instr);
      if (nir_src_bit_size(instr->src[0]) == 64) {
         src = shuffle_for_32bit_write(bld, src, 0, num_components);
         num_components *= 2;
      }

      fs_reg new_dest = retype(offset(outputs[instr->const_index[0]], bld,
                                      4 * const_offset->u32[0]), src.type);
      for (unsigned j = 0; j < num_components; j++) {
         bld.MOV(offset(new_dest, bld, j + first_component),
                 offset(src, bld, j));
      }
      break;
   }

   case nir_intrinsic_ssbo_atomic_add:
      nir_emit_ssbo_atomic(bld, BRW_AOP_ADD, instr);
      break;
   case nir_intrinsic_ssbo_atomic_imin:
      nir_emit_ssbo_atomic(bld, BRW_AOP_IMIN, instr);
      break;
   case nir_intrinsic_ssbo_atomic_umin:
      nir_emit_ssbo_atomic(bld, BRW_AOP_UMIN, instr);
      break;
   case nir_intrinsic_ssbo_atomic_imax:
      nir_emit_ssbo_atomic(bld, BRW_AOP_IMAX, instr);
      break;
   case nir_intrinsic_ssbo_atomic_umax:
      nir_emit_ssbo_atomic(bld, BRW_AOP_UMAX, instr);
      break;
   case nir_intrinsic_ssbo_atomic_and:
      nir_emit_ssbo_atomic(bld, BRW_AOP_AND, instr);
      break;
   case nir_intrinsic_ssbo_atomic_or:
      nir_emit_ssbo_atomic(bld, BRW_AOP_OR, instr);
      break;
   case nir_intrinsic_ssbo_atomic_xor:
      nir_emit_ssbo_atomic(bld, BRW_AOP_XOR, instr);
      break;
   case nir_intrinsic_ssbo_atomic_exchange:
      nir_emit_ssbo_atomic(bld, BRW_AOP_MOV, instr);
      break;
   case nir_intrinsic_ssbo_atomic_comp_swap:
      nir_emit_ssbo_atomic(bld, BRW_AOP_CMPWR, instr);
      break;

   case nir_intrinsic_get_buffer_size: {
      nir_const_value *const_uniform_block = nir_src_as_const_value(instr->src[0]);
      unsigned ssbo_index = const_uniform_block ? const_uniform_block->u32[0] : 0;

      /* A resinfo's sampler message is used to get the buffer size.  The
       * SIMD8's writeback message consists of four registers and SIMD16's
       * writeback message consists of 8 destination registers (two per each
       * component).  Because we are only interested on the first channel of
       * the first returned component, where resinfo returns the buffer size
       * for SURFTYPE_BUFFER, we can just use the SIMD8 variant regardless of
       * the dispatch width.
       */
      const fs_builder ubld = bld.exec_all().group(8, 0);
      fs_reg src_payload = ubld.vgrf(BRW_REGISTER_TYPE_UD);
      fs_reg ret_payload = ubld.vgrf(BRW_REGISTER_TYPE_UD, 4);

      /* Set LOD = 0 */
      ubld.MOV(src_payload, brw_imm_d(0));

      const unsigned index = prog_data->binding_table.ssbo_start + ssbo_index;
      fs_inst *inst = ubld.emit(SHADER_OPCODE_GET_BUFFER_SIZE, ret_payload,
                                src_payload, brw_imm_ud(index));
      inst->header_size = 0;
      inst->mlen = 1;
      inst->size_written = 4 * REG_SIZE;

      /* SKL PRM, vol07, 3D Media GPGPU Engine, Bounds Checking and Faulting:
       *
       * "Out-of-bounds checking is always performed at a DWord granularity. If
       * any part of the DWord is out-of-bounds then the whole DWord is
       * considered out-of-bounds."
       *
       * This implies that types with size smaller than 4-bytes need to be
       * padded if they don't complete the last dword of the buffer. But as we
       * need to maintain the original size we need to reverse the padding
       * calculation to return the correct size to know the number of elements
       * of an unsized array. As we stored in the last two bits of the surface
       * size the needed padding for the buffer, we calculate here the
       * original buffer_size reversing the surface_size calculation:
       *
       * surface_size = isl_align(buffer_size, 4) +
       *                (isl_align(buffer_size) - buffer_size)
       *
       * buffer_size = surface_size & ~3 - surface_size & 3
       */

      fs_reg size_aligned4 = ubld.vgrf(BRW_REGISTER_TYPE_UD);
      fs_reg size_padding = ubld.vgrf(BRW_REGISTER_TYPE_UD);
      fs_reg buffer_size = ubld.vgrf(BRW_REGISTER_TYPE_UD);

      ubld.AND(size_padding, ret_payload, brw_imm_ud(3));
      ubld.AND(size_aligned4, ret_payload, brw_imm_ud(~3));
      ubld.ADD(buffer_size, size_aligned4, negate(size_padding));

      bld.MOV(retype(dest, ret_payload.type), component(buffer_size, 0));

      brw_mark_surface_used(prog_data, index);
      break;
   }

   case nir_intrinsic_load_subgroup_invocation:
      bld.MOV(retype(dest, BRW_REGISTER_TYPE_D),
              nir_system_values[SYSTEM_VALUE_SUBGROUP_INVOCATION]);
      break;

   case nir_intrinsic_load_subgroup_eq_mask:
   case nir_intrinsic_load_subgroup_ge_mask:
   case nir_intrinsic_load_subgroup_gt_mask:
   case nir_intrinsic_load_subgroup_le_mask:
   case nir_intrinsic_load_subgroup_lt_mask:
      unreachable("not reached");

   case nir_intrinsic_vote_any: {
      const fs_builder ubld = bld.exec_all().group(1, 0);

      /* The any/all predicates do not consider channel enables. To prevent
       * dead channels from affecting the result, we initialize the flag with
       * with the identity value for the logical operation.
       */
      if (dispatch_width == 32) {
         /* For SIMD32, we use a UD type so we fill both f0.0 and f0.1. */
         ubld.MOV(retype(brw_flag_reg(0, 0), BRW_REGISTER_TYPE_UD),
                         brw_imm_ud(0));
      } else {
         ubld.MOV(brw_flag_reg(0, 0), brw_imm_uw(0));
      }
      bld.CMP(bld.null_reg_d(), get_nir_src(instr->src[0]), brw_imm_d(0), BRW_CONDITIONAL_NZ);

      /* For some reason, the any/all predicates don't work properly with
       * SIMD32.  In particular, it appears that a SEL with a QtrCtrl of 2H
       * doesn't read the correct subset of the flag register and you end up
       * getting garbage in the second half.  Work around this by using a pair
       * of 1-wide MOVs and scattering the result.
       */
      fs_reg res1 = ubld.vgrf(BRW_REGISTER_TYPE_D);
      ubld.MOV(res1, brw_imm_d(0));
      set_predicate(dispatch_width == 8  ? BRW_PREDICATE_ALIGN1_ANY8H :
                    dispatch_width == 16 ? BRW_PREDICATE_ALIGN1_ANY16H :
                                           BRW_PREDICATE_ALIGN1_ANY32H,
                    ubld.MOV(res1, brw_imm_d(-1)));

      bld.MOV(retype(dest, BRW_REGISTER_TYPE_D), component(res1, 0));
      break;
   }
   case nir_intrinsic_vote_all: {
      const fs_builder ubld = bld.exec_all().group(1, 0);

      /* The any/all predicates do not consider channel enables. To prevent
       * dead channels from affecting the result, we initialize the flag with
       * with the identity value for the logical operation.
       */
      if (dispatch_width == 32) {
         /* For SIMD32, we use a UD type so we fill both f0.0 and f0.1. */
         ubld.MOV(retype(brw_flag_reg(0, 0), BRW_REGISTER_TYPE_UD),
                         brw_imm_ud(0xffffffff));
      } else {
         ubld.MOV(brw_flag_reg(0, 0), brw_imm_uw(0xffff));
      }
      bld.CMP(bld.null_reg_d(), get_nir_src(instr->src[0]), brw_imm_d(0), BRW_CONDITIONAL_NZ);

      /* For some reason, the any/all predicates don't work properly with
       * SIMD32.  In particular, it appears that a SEL with a QtrCtrl of 2H
       * doesn't read the correct subset of the flag register and you end up
       * getting garbage in the second half.  Work around this by using a pair
       * of 1-wide MOVs and scattering the result.
       */
      fs_reg res1 = ubld.vgrf(BRW_REGISTER_TYPE_D);
      ubld.MOV(res1, brw_imm_d(0));
      set_predicate(dispatch_width == 8  ? BRW_PREDICATE_ALIGN1_ALL8H :
                    dispatch_width == 16 ? BRW_PREDICATE_ALIGN1_ALL16H :
                                           BRW_PREDICATE_ALIGN1_ALL32H,
                    ubld.MOV(res1, brw_imm_d(-1)));

      bld.MOV(retype(dest, BRW_REGISTER_TYPE_D), component(res1, 0));
      break;
   }
   case nir_intrinsic_vote_feq:
   case nir_intrinsic_vote_ieq: {
      fs_reg value = get_nir_src(instr->src[0]);
      if (instr->intrinsic == nir_intrinsic_vote_feq) {
         const unsigned bit_size = nir_src_bit_size(instr->src[0]);
         value.type = brw_reg_type_from_bit_size(bit_size, BRW_REGISTER_TYPE_F);
      }

      fs_reg uniformized = bld.emit_uniformize(value);
      const fs_builder ubld = bld.exec_all().group(1, 0);

      /* The any/all predicates do not consider channel enables. To prevent
       * dead channels from affecting the result, we initialize the flag with
       * with the identity value for the logical operation.
       */
      if (dispatch_width == 32) {
         /* For SIMD32, we use a UD type so we fill both f0.0 and f0.1. */
         ubld.MOV(retype(brw_flag_reg(0, 0), BRW_REGISTER_TYPE_UD),
                         brw_imm_ud(0xffffffff));
      } else {
         ubld.MOV(brw_flag_reg(0, 0), brw_imm_uw(0xffff));
      }
      bld.CMP(bld.null_reg_d(), value, uniformized, BRW_CONDITIONAL_Z);

      /* For some reason, the any/all predicates don't work properly with
       * SIMD32.  In particular, it appears that a SEL with a QtrCtrl of 2H
       * doesn't read the correct subset of the flag register and you end up
       * getting garbage in the second half.  Work around this by using a pair
       * of 1-wide MOVs and scattering the result.
       */
      fs_reg res1 = ubld.vgrf(BRW_REGISTER_TYPE_D);
      ubld.MOV(res1, brw_imm_d(0));
      set_predicate(dispatch_width == 8  ? BRW_PREDICATE_ALIGN1_ALL8H :
                    dispatch_width == 16 ? BRW_PREDICATE_ALIGN1_ALL16H :
                                           BRW_PREDICATE_ALIGN1_ALL32H,
                    ubld.MOV(res1, brw_imm_d(-1)));

      bld.MOV(retype(dest, BRW_REGISTER_TYPE_D), component(res1, 0));
      break;
   }

   case nir_intrinsic_ballot: {
      const fs_reg value = retype(get_nir_src(instr->src[0]),
                                  BRW_REGISTER_TYPE_UD);
      struct brw_reg flag = brw_flag_reg(0, 0);
      /* FIXME: For SIMD32 programs, this causes us to stomp on f0.1 as well
       * as f0.0.  This is a problem for fragment programs as we currently use
       * f0.1 for discards.  Fortunately, we don't support SIMD32 fragment
       * programs yet so this isn't a problem.  When we do, something will
       * have to change.
       */
      if (dispatch_width == 32)
         flag.type = BRW_REGISTER_TYPE_UD;

      bld.exec_all().group(1, 0).MOV(flag, brw_imm_ud(0u));
      bld.CMP(bld.null_reg_ud(), value, brw_imm_ud(0u), BRW_CONDITIONAL_NZ);

      if (instr->dest.ssa.bit_size > 32) {
         dest.type = BRW_REGISTER_TYPE_UQ;
      } else {
         dest.type = BRW_REGISTER_TYPE_UD;
      }
      bld.MOV(dest, flag);
      break;
   }

   case nir_intrinsic_read_invocation: {
      const fs_reg value = get_nir_src(instr->src[0]);
      const fs_reg invocation = get_nir_src(instr->src[1]);
      fs_reg tmp = bld.vgrf(value.type);

      bld.exec_all().emit(SHADER_OPCODE_BROADCAST, tmp, value,
                          bld.emit_uniformize(invocation));

      bld.MOV(retype(dest, value.type), fs_reg(component(tmp, 0)));
      break;
   }

   case nir_intrinsic_read_first_invocation: {
      const fs_reg value = get_nir_src(instr->src[0]);
      bld.MOV(retype(dest, value.type), bld.emit_uniformize(value));
      break;
   }

   case nir_intrinsic_shuffle: {
      const fs_reg value = get_nir_src(instr->src[0]);
      const fs_reg index = get_nir_src(instr->src[1]);

      bld.emit(SHADER_OPCODE_SHUFFLE, retype(dest, value.type), value, index);
      break;
   }

   case nir_intrinsic_first_invocation: {
      fs_reg tmp = bld.vgrf(BRW_REGISTER_TYPE_UD);
      bld.exec_all().emit(SHADER_OPCODE_FIND_LIVE_CHANNEL, tmp);
      bld.MOV(retype(dest, BRW_REGISTER_TYPE_UD),
              fs_reg(component(tmp, 0)));
      break;
   }

   case nir_intrinsic_quad_broadcast: {
      const fs_reg value = get_nir_src(instr->src[0]);
      nir_const_value *index = nir_src_as_const_value(instr->src[1]);
      assert(nir_src_bit_size(instr->src[1]) == 32);

      bld.emit(SHADER_OPCODE_CLUSTER_BROADCAST, retype(dest, value.type),
               value, brw_imm_ud(index->u32[0]), brw_imm_ud(4));
      break;
   }

   case nir_intrinsic_quad_swap_horizontal: {
      const fs_reg value = get_nir_src(instr->src[0]);
      const fs_reg tmp = bld.vgrf(value.type);
      const fs_builder ubld = bld.exec_all().group(dispatch_width / 2, 0);

      const fs_reg src_left = horiz_stride(value, 2);
      const fs_reg src_right = horiz_stride(horiz_offset(value, 1), 2);
      const fs_reg tmp_left = horiz_stride(tmp, 2);
      const fs_reg tmp_right = horiz_stride(horiz_offset(tmp, 1), 2);

      /* From the Cherryview PRM Vol. 7, "Register Region Restrictiosn":
       *
       *    "When source or destination datatype is 64b or operation is
       *    integer DWord multiply, regioning in Align1 must follow
       *    these rules:
       *
       *    [...]
       *
       *    3. Source and Destination offset must be the same, except
       *       the case of scalar source."
       *
       * In order to work around this, we have to emit two 32-bit MOVs instead
       * of a single 64-bit MOV to do the shuffle.
       */
      if (type_sz(value.type) > 4 &&
          (devinfo->is_cherryview || gen_device_info_is_9lp(devinfo))) {
         ubld.MOV(subscript(tmp_left, BRW_REGISTER_TYPE_D, 0),
                  subscript(src_right, BRW_REGISTER_TYPE_D, 0));
         ubld.MOV(subscript(tmp_left, BRW_REGISTER_TYPE_D, 1),
                  subscript(src_right, BRW_REGISTER_TYPE_D, 1));
         ubld.MOV(subscript(tmp_right, BRW_REGISTER_TYPE_D, 0),
                  subscript(src_left, BRW_REGISTER_TYPE_D, 0));
         ubld.MOV(subscript(tmp_right, BRW_REGISTER_TYPE_D, 1),
                  subscript(src_left, BRW_REGISTER_TYPE_D, 1));
      } else {
         ubld.MOV(tmp_left, src_right);
         ubld.MOV(tmp_right, src_left);
      }
      bld.MOV(retype(dest, value.type), tmp);
      break;
   }

   case nir_intrinsic_quad_swap_vertical: {
      const fs_reg value = get_nir_src(instr->src[0]);
      if (nir_src_bit_size(instr->src[0]) == 32) {
         /* For 32-bit, we can use a SIMD4x2 instruction to do this easily */
         const fs_reg tmp = bld.vgrf(value.type);
         const fs_builder ubld = bld.exec_all();
         ubld.emit(SHADER_OPCODE_QUAD_SWIZZLE, tmp, value,
                   brw_imm_ud(BRW_SWIZZLE4(2,3,0,1)));
         bld.MOV(retype(dest, value.type), tmp);
      } else {
         /* For larger data types, we have to either emit dispatch_width many
          * MOVs or else fall back to doing indirects.
          */
         fs_reg idx = bld.vgrf(BRW_REGISTER_TYPE_W);
         bld.XOR(idx, nir_system_values[SYSTEM_VALUE_SUBGROUP_INVOCATION],
                      brw_imm_w(0x2));
         bld.emit(SHADER_OPCODE_SHUFFLE, retype(dest, value.type), value, idx);
      }
      break;
   }

   case nir_intrinsic_quad_swap_diagonal: {
      const fs_reg value = get_nir_src(instr->src[0]);
      if (nir_src_bit_size(instr->src[0]) == 32) {
         /* For 32-bit, we can use a SIMD4x2 instruction to do this easily */
         const fs_reg tmp = bld.vgrf(value.type);
         const fs_builder ubld = bld.exec_all();
         ubld.emit(SHADER_OPCODE_QUAD_SWIZZLE, tmp, value,
                   brw_imm_ud(BRW_SWIZZLE4(3,2,1,0)));
         bld.MOV(retype(dest, value.type), tmp);
      } else {
         /* For larger data types, we have to either emit dispatch_width many
          * MOVs or else fall back to doing indirects.
          */
         fs_reg idx = bld.vgrf(BRW_REGISTER_TYPE_W);
         bld.XOR(idx, nir_system_values[SYSTEM_VALUE_SUBGROUP_INVOCATION],
                      brw_imm_w(0x3));
         bld.emit(SHADER_OPCODE_SHUFFLE, retype(dest, value.type), value, idx);
      }
      break;
   }

   case nir_intrinsic_reduce: {
      fs_reg src = get_nir_src(instr->src[0]);
      nir_op redop = (nir_op)nir_intrinsic_reduction_op(instr);
      unsigned cluster_size = nir_intrinsic_cluster_size(instr);
      if (cluster_size == 0 || cluster_size > dispatch_width)
         cluster_size = dispatch_width;

      /* Figure out the source type */
      src.type = brw_type_for_nir_type(devinfo,
         (nir_alu_type)(nir_op_infos[redop].input_types[0] |
                        nir_src_bit_size(instr->src[0])));

      fs_reg identity = brw_nir_reduction_op_identity(bld, redop, src.type);
      opcode brw_op = brw_op_for_nir_reduction_op(redop);
      brw_conditional_mod cond_mod = brw_cond_mod_for_nir_reduction_op(redop);

      /* Set up a register for all of our scratching around and initialize it
       * to reduction operation's identity value.
       */
      fs_reg scan = bld.vgrf(src.type);
      bld.exec_all().emit(SHADER_OPCODE_SEL_EXEC, scan, src, identity);

      bld.emit_scan(brw_op, scan, cluster_size, cond_mod);

      dest.type = src.type;
      if (cluster_size * type_sz(src.type) >= REG_SIZE * 2) {
         /* In this case, CLUSTER_BROADCAST instruction isn't needed because
          * the distance between clusters is at least 2 GRFs.  In this case,
          * we don't need the weird striding of the CLUSTER_BROADCAST
          * instruction and can just do regular MOVs.
          */
         assert((cluster_size * type_sz(src.type)) % (REG_SIZE * 2) == 0);
         const unsigned groups =
            (dispatch_width * type_sz(src.type)) / (REG_SIZE * 2);
         const unsigned group_size = dispatch_width / groups;
         for (unsigned i = 0; i < groups; i++) {
            const unsigned cluster = (i * group_size) / cluster_size;
            const unsigned comp = cluster * cluster_size + (cluster_size - 1);
            bld.group(group_size, i).MOV(horiz_offset(dest, i * group_size),
                                         component(scan, comp));
         }
      } else {
         bld.emit(SHADER_OPCODE_CLUSTER_BROADCAST, dest, scan,
                  brw_imm_ud(cluster_size - 1), brw_imm_ud(cluster_size));
      }
      break;
   }

   case nir_intrinsic_inclusive_scan:
   case nir_intrinsic_exclusive_scan: {
      fs_reg src = get_nir_src(instr->src[0]);
      nir_op redop = (nir_op)nir_intrinsic_reduction_op(instr);

      /* Figure out the source type */
      src.type = brw_type_for_nir_type(devinfo,
         (nir_alu_type)(nir_op_infos[redop].input_types[0] |
                        nir_src_bit_size(instr->src[0])));

      fs_reg identity = brw_nir_reduction_op_identity(bld, redop, src.type);
      opcode brw_op = brw_op_for_nir_reduction_op(redop);
      brw_conditional_mod cond_mod = brw_cond_mod_for_nir_reduction_op(redop);

      /* Set up a register for all of our scratching around and initialize it
       * to reduction operation's identity value.
       */
      fs_reg scan = bld.vgrf(src.type);
      const fs_builder allbld = bld.exec_all();
      allbld.emit(SHADER_OPCODE_SEL_EXEC, scan, src, identity);

      if (instr->intrinsic == nir_intrinsic_exclusive_scan) {
         /* Exclusive scan is a bit harder because we have to do an annoying
          * shift of the contents before we can begin.  To make things worse,
          * we can't do this with a normal stride; we have to use indirects.
          */
         fs_reg shifted = bld.vgrf(src.type);
         fs_reg idx = bld.vgrf(BRW_REGISTER_TYPE_W);
         allbld.ADD(idx, nir_system_values[SYSTEM_VALUE_SUBGROUP_INVOCATION],
                         brw_imm_w(-1));
         allbld.emit(SHADER_OPCODE_SHUFFLE, shifted, scan, idx);
         allbld.group(1, 0).MOV(component(shifted, 0), identity);
         scan = shifted;
      }

      bld.emit_scan(brw_op, scan, dispatch_width, cond_mod);

      bld.MOV(retype(dest, src.type), scan);
      break;
   }

   case nir_intrinsic_begin_invocation_interlock: {
      const fs_builder ubld = bld.group(8, 0);
      const fs_reg tmp = ubld.vgrf(BRW_REGISTER_TYPE_UD, 2);

      ubld.emit(SHADER_OPCODE_INTERLOCK, tmp)->size_written = 2 *
         REG_SIZE;

      break;
   }

   case nir_intrinsic_end_invocation_interlock: {
      /* We don't need to do anything here */
      break;
   }

   default:
      unreachable("unknown intrinsic");
   }
}

void
fs_visitor::nir_emit_ssbo_atomic(const fs_builder &bld,
                                 int op, nir_intrinsic_instr *instr)
{
   if (stage == MESA_SHADER_FRAGMENT)
      brw_wm_prog_data(prog_data)->has_side_effects = true;

   fs_reg dest;
   if (nir_intrinsic_infos[instr->intrinsic].has_dest)
      dest = get_nir_dest(instr->dest);

   fs_reg surface;
   nir_const_value *const_surface = nir_src_as_const_value(instr->src[0]);
   if (const_surface) {
      unsigned surf_index = stage_prog_data->binding_table.ssbo_start +
                            const_surface->u32[0];
      surface = brw_imm_ud(surf_index);
      brw_mark_surface_used(prog_data, surf_index);
   } else {
      surface = vgrf(glsl_type::uint_type);
      bld.ADD(surface, get_nir_src(instr->src[0]),
              brw_imm_ud(stage_prog_data->binding_table.ssbo_start));

      /* Assume this may touch any SSBO. This is the same we do for other
       * UBO/SSBO accesses with non-constant surface.
       */
      brw_mark_surface_used(prog_data,
                            stage_prog_data->binding_table.ssbo_start +
                            nir->info.num_ssbos - 1);
   }

   fs_reg offset = get_nir_src(instr->src[1]);
   fs_reg data1 = get_nir_src(instr->src[2]);
   fs_reg data2;
   if (op == BRW_AOP_CMPWR)
      data2 = get_nir_src(instr->src[3]);

   /* Emit the actual atomic operation */

   fs_reg atomic_result = emit_untyped_atomic(bld, surface, offset,
                                              data1, data2,
                                              1 /* dims */, 1 /* rsize */,
                                              op,
                                              BRW_PREDICATE_NONE);
   dest.type = atomic_result.type;
   bld.MOV(dest, atomic_result);
}

void
fs_visitor::nir_emit_shared_atomic(const fs_builder &bld,
                                   int op, nir_intrinsic_instr *instr)
{
   fs_reg dest;
   if (nir_intrinsic_infos[instr->intrinsic].has_dest)
      dest = get_nir_dest(instr->dest);

   fs_reg surface = brw_imm_ud(GEN7_BTI_SLM);
   fs_reg offset;
   fs_reg data1 = get_nir_src(instr->src[1]);
   fs_reg data2;
   if (op == BRW_AOP_CMPWR)
      data2 = get_nir_src(instr->src[2]);

   /* Get the offset */
   nir_const_value *const_offset = nir_src_as_const_value(instr->src[0]);
   if (const_offset) {
      offset = brw_imm_ud(instr->const_index[0] + const_offset->u32[0]);
   } else {
      offset = vgrf(glsl_type::uint_type);
      bld.ADD(offset,
	      retype(get_nir_src(instr->src[0]), BRW_REGISTER_TYPE_UD),
	      brw_imm_ud(instr->const_index[0]));
   }

   /* Emit the actual atomic operation operation */

   fs_reg atomic_result = emit_untyped_atomic(bld, surface, offset,
                                              data1, data2,
                                              1 /* dims */, 1 /* rsize */,
                                              op,
                                              BRW_PREDICATE_NONE);
   dest.type = atomic_result.type;
   bld.MOV(dest, atomic_result);
}

void
fs_visitor::nir_emit_texture(const fs_builder &bld, nir_tex_instr *instr)
{
   unsigned texture = instr->texture_index;
   unsigned sampler = instr->sampler_index;

   fs_reg srcs[TEX_LOGICAL_NUM_SRCS];

   srcs[TEX_LOGICAL_SRC_SURFACE] = brw_imm_ud(texture);
   srcs[TEX_LOGICAL_SRC_SAMPLER] = brw_imm_ud(sampler);

   int lod_components = 0;

   /* The hardware requires a LOD for buffer textures */
   if (instr->sampler_dim == GLSL_SAMPLER_DIM_BUF)
      srcs[TEX_LOGICAL_SRC_LOD] = brw_imm_d(0);

   uint32_t header_bits = 0;
   for (unsigned i = 0; i < instr->num_srcs; i++) {
      fs_reg src = get_nir_src(instr->src[i].src);
      switch (instr->src[i].src_type) {
      case nir_tex_src_bias:
         srcs[TEX_LOGICAL_SRC_LOD] =
            retype(get_nir_src_imm(instr->src[i].src), BRW_REGISTER_TYPE_F);
         break;
      case nir_tex_src_comparator:
         srcs[TEX_LOGICAL_SRC_SHADOW_C] = retype(src, BRW_REGISTER_TYPE_F);
         break;
      case nir_tex_src_coord:
         switch (instr->op) {
         case nir_texop_txf:
         case nir_texop_txf_ms:
         case nir_texop_txf_ms_mcs:
         case nir_texop_samples_identical:
            srcs[TEX_LOGICAL_SRC_COORDINATE] = retype(src, BRW_REGISTER_TYPE_D);
            break;
         default:
            srcs[TEX_LOGICAL_SRC_COORDINATE] = retype(src, BRW_REGISTER_TYPE_F);
            break;
         }
         break;
      case nir_tex_src_ddx:
         srcs[TEX_LOGICAL_SRC_LOD] = retype(src, BRW_REGISTER_TYPE_F);
         lod_components = nir_tex_instr_src_size(instr, i);
         break;
      case nir_tex_src_ddy:
         srcs[TEX_LOGICAL_SRC_LOD2] = retype(src, BRW_REGISTER_TYPE_F);
         break;
      case nir_tex_src_lod:
         switch (instr->op) {
         case nir_texop_txs:
            srcs[TEX_LOGICAL_SRC_LOD] =
               retype(get_nir_src_imm(instr->src[i].src), BRW_REGISTER_TYPE_UD);
            break;
         case nir_texop_txf:
            srcs[TEX_LOGICAL_SRC_LOD] =
               retype(get_nir_src_imm(instr->src[i].src), BRW_REGISTER_TYPE_D);
            break;
         default:
            srcs[TEX_LOGICAL_SRC_LOD] =
               retype(get_nir_src_imm(instr->src[i].src), BRW_REGISTER_TYPE_F);
            break;
         }
         break;
      case nir_tex_src_ms_index:
         srcs[TEX_LOGICAL_SRC_SAMPLE_INDEX] = retype(src, BRW_REGISTER_TYPE_UD);
         break;

      case nir_tex_src_offset: {
         nir_const_value *const_offset =
            nir_src_as_const_value(instr->src[i].src);
         unsigned offset_bits = 0;
         if (const_offset &&
             brw_texture_offset(const_offset->i32,
                                nir_tex_instr_src_size(instr, i),
                                &offset_bits)) {
            header_bits |= offset_bits;
         } else {
            srcs[TEX_LOGICAL_SRC_TG4_OFFSET] =
               retype(src, BRW_REGISTER_TYPE_D);
         }
         break;
      }

      case nir_tex_src_projector:
         unreachable("should be lowered");

      case nir_tex_src_texture_offset: {
         /* Figure out the highest possible texture index and mark it as used */
         uint32_t max_used = texture + instr->texture_array_size - 1;
         if (instr->op == nir_texop_tg4 && devinfo->gen < 8) {
            max_used += stage_prog_data->binding_table.gather_texture_start;
         } else {
            max_used += stage_prog_data->binding_table.texture_start;
         }
         brw_mark_surface_used(prog_data, max_used);

         /* Emit code to evaluate the actual indexing expression */
         fs_reg tmp = vgrf(glsl_type::uint_type);
         bld.ADD(tmp, src, brw_imm_ud(texture));
         srcs[TEX_LOGICAL_SRC_SURFACE] = bld.emit_uniformize(tmp);
         break;
      }

      case nir_tex_src_sampler_offset: {
         /* Emit code to evaluate the actual indexing expression */
         fs_reg tmp = vgrf(glsl_type::uint_type);
         bld.ADD(tmp, src, brw_imm_ud(sampler));
         srcs[TEX_LOGICAL_SRC_SAMPLER] = bld.emit_uniformize(tmp);
         break;
      }

      case nir_tex_src_ms_mcs:
         assert(instr->op == nir_texop_txf_ms);
         srcs[TEX_LOGICAL_SRC_MCS] = retype(src, BRW_REGISTER_TYPE_D);
         break;

      case nir_tex_src_plane: {
         nir_const_value *const_plane =
            nir_src_as_const_value(instr->src[i].src);
         const uint32_t plane = const_plane->u32[0];
         const uint32_t texture_index =
            instr->texture_index +
            stage_prog_data->binding_table.plane_start[plane] -
            stage_prog_data->binding_table.texture_start;

         srcs[TEX_LOGICAL_SRC_SURFACE] = brw_imm_ud(texture_index);
         break;
      }

      default:
         unreachable("unknown texture source");
      }
   }

   if (srcs[TEX_LOGICAL_SRC_MCS].file == BAD_FILE &&
       (instr->op == nir_texop_txf_ms ||
        instr->op == nir_texop_samples_identical)) {
      if (devinfo->gen >= 7 &&
          key_tex->compressed_multisample_layout_mask & (1 << texture)) {
         srcs[TEX_LOGICAL_SRC_MCS] =
            emit_mcs_fetch(srcs[TEX_LOGICAL_SRC_COORDINATE],
                           instr->coord_components,
                           srcs[TEX_LOGICAL_SRC_SURFACE]);
      } else {
         srcs[TEX_LOGICAL_SRC_MCS] = brw_imm_ud(0u);
      }
   }

   srcs[TEX_LOGICAL_SRC_COORD_COMPONENTS] = brw_imm_d(instr->coord_components);
   srcs[TEX_LOGICAL_SRC_GRAD_COMPONENTS] = brw_imm_d(lod_components);

   enum opcode opcode;
   switch (instr->op) {
   case nir_texop_tex:
      opcode = (stage == MESA_SHADER_FRAGMENT ? SHADER_OPCODE_TEX_LOGICAL :
                SHADER_OPCODE_TXL_LOGICAL);
      break;
   case nir_texop_txb:
      opcode = FS_OPCODE_TXB_LOGICAL;
      break;
   case nir_texop_txl:
      opcode = SHADER_OPCODE_TXL_LOGICAL;
      break;
   case nir_texop_txd:
      opcode = SHADER_OPCODE_TXD_LOGICAL;
      break;
   case nir_texop_txf:
      opcode = SHADER_OPCODE_TXF_LOGICAL;
      break;
   case nir_texop_txf_ms:
      if ((key_tex->msaa_16 & (1 << sampler)))
         opcode = SHADER_OPCODE_TXF_CMS_W_LOGICAL;
      else
         opcode = SHADER_OPCODE_TXF_CMS_LOGICAL;
      break;
   case nir_texop_txf_ms_mcs:
      opcode = SHADER_OPCODE_TXF_MCS_LOGICAL;
      break;
   case nir_texop_query_levels:
   case nir_texop_txs:
      opcode = SHADER_OPCODE_TXS_LOGICAL;
      break;
   case nir_texop_lod:
      opcode = SHADER_OPCODE_LOD_LOGICAL;
      break;
   case nir_texop_tg4:
      if (srcs[TEX_LOGICAL_SRC_TG4_OFFSET].file != BAD_FILE)
         opcode = SHADER_OPCODE_TG4_OFFSET_LOGICAL;
      else
         opcode = SHADER_OPCODE_TG4_LOGICAL;
      break;
   case nir_texop_texture_samples:
      opcode = SHADER_OPCODE_SAMPLEINFO_LOGICAL;
      break;
   case nir_texop_samples_identical: {
      fs_reg dst = retype(get_nir_dest(instr->dest), BRW_REGISTER_TYPE_D);

      /* If mcs is an immediate value, it means there is no MCS.  In that case
       * just return false.
       */
      if (srcs[TEX_LOGICAL_SRC_MCS].file == BRW_IMMEDIATE_VALUE) {
         bld.MOV(dst, brw_imm_ud(0u));
      } else if ((key_tex->msaa_16 & (1 << sampler))) {
         fs_reg tmp = vgrf(glsl_type::uint_type);
         bld.OR(tmp, srcs[TEX_LOGICAL_SRC_MCS],
                offset(srcs[TEX_LOGICAL_SRC_MCS], bld, 1));
         bld.CMP(dst, tmp, brw_imm_ud(0u), BRW_CONDITIONAL_EQ);
      } else {
         bld.CMP(dst, srcs[TEX_LOGICAL_SRC_MCS], brw_imm_ud(0u),
                 BRW_CONDITIONAL_EQ);
      }
      return;
   }
   default:
      unreachable("unknown texture opcode");
   }

   if (instr->op == nir_texop_tg4) {
      if (instr->component == 1 &&
          key_tex->gather_channel_quirk_mask & (1 << texture)) {
         /* gather4 sampler is broken for green channel on RG32F --
          * we must ask for blue instead.
          */
         header_bits |= 2 << 16;
      } else {
         header_bits |= instr->component << 16;
      }
   }

   fs_reg dst = bld.vgrf(brw_type_for_nir_type(devinfo, instr->dest_type), 4);
   fs_inst *inst = bld.emit(opcode, dst, srcs, ARRAY_SIZE(srcs));
   inst->offset = header_bits;

   const unsigned dest_size = nir_tex_instr_dest_size(instr);
   if (devinfo->gen >= 9 &&
       instr->op != nir_texop_tg4 && instr->op != nir_texop_query_levels) {
      unsigned write_mask = instr->dest.is_ssa ?
                            nir_ssa_def_components_read(&instr->dest.ssa):
                            (1 << dest_size) - 1;
      assert(write_mask != 0); /* dead code should have been eliminated */
      inst->size_written = util_last_bit(write_mask) *
                           inst->dst.component_size(inst->exec_size);
   } else {
      inst->size_written = 4 * inst->dst.component_size(inst->exec_size);
   }

   if (srcs[TEX_LOGICAL_SRC_SHADOW_C].file != BAD_FILE)
      inst->shadow_compare = true;

   if (instr->op == nir_texop_tg4 && devinfo->gen == 6)
      emit_gen6_gather_wa(key_tex->gen6_gather_wa[texture], dst);

   fs_reg nir_dest[4];
   for (unsigned i = 0; i < dest_size; i++)
      nir_dest[i] = offset(dst, bld, i);

   if (instr->op == nir_texop_query_levels) {
      /* # levels is in .w */
      nir_dest[0] = offset(dst, bld, 3);
   } else if (instr->op == nir_texop_txs &&
              dest_size >= 3 && devinfo->gen < 7) {
      /* Gen4-6 return 0 instead of 1 for single layer surfaces. */
      fs_reg depth = offset(dst, bld, 2);
      nir_dest[2] = vgrf(glsl_type::int_type);
      bld.emit_minmax(nir_dest[2], depth, brw_imm_d(1), BRW_CONDITIONAL_GE);
   }

   bld.LOAD_PAYLOAD(get_nir_dest(instr->dest), nir_dest, dest_size, 0);
}

void
fs_visitor::nir_emit_jump(const fs_builder &bld, nir_jump_instr *instr)
{
   switch (instr->type) {
   case nir_jump_break:
      bld.emit(BRW_OPCODE_BREAK);
      break;
   case nir_jump_continue:
      bld.emit(BRW_OPCODE_CONTINUE);
      break;
   case nir_jump_return:
   default:
      unreachable("unknown jump");
   }
}

/*
 * This helper takes a source register and un/shuffles it into the destination
 * register.
 *
 * If source type size is smaller than destination type size the operation
 * needed is a component shuffle. The opposite case would be an unshuffle. If
 * source/destination type size is equal a shuffle is done that would be
 * equivalent to a simple MOV.
 *
 * For example, if source is a 16-bit type and destination is 32-bit. A 3
 * components .xyz 16-bit vector on SIMD8 would be.
 *
 *    |x1|x2|x3|x4|x5|x6|x7|x8|y1|y2|y3|y4|y5|y6|y7|y8|
 *    |z1|z2|z3|z4|z5|z6|z7|z8|  |  |  |  |  |  |  |  |
 *
 * This helper will return the following 2 32-bit components with the 16-bit
 * values shuffled:
 *
 *    |x1 y1|x2 y2|x3 y3|x4 y4|x5 y5|x6 y6|x7 y7|x8 y8|
 *    |z1   |z2   |z3   |z4   |z5   |z6   |z7   |z8   |
 *
 * For unshuffle, the example would be the opposite, a 64-bit type source
 * and a 32-bit destination. A 2 component .xy 64-bit vector on SIMD8
 * would be:
 *
 *    | x1l   x1h | x2l   x2h | x3l   x3h | x4l   x4h |
 *    | x5l   x5h | x6l   x6h | x7l   x7h | x8l   x8h |
 *    | y1l   y1h | y2l   y2h | y3l   y3h | y4l   y4h |
 *    | y5l   y5h | y6l   y6h | y7l   y7h | y8l   y8h |
 *
 * The returned result would be the following 4 32-bit components unshuffled:
 *
 *    | x1l | x2l | x3l | x4l | x5l | x6l | x7l | x8l |
 *    | x1h | x2h | x3h | x4h | x5h | x6h | x7h | x8h |
 *    | y1l | y2l | y3l | y4l | y5l | y6l | y7l | y8l |
 *    | y1h | y2h | y3h | y4h | y5h | y6h | y7h | y8h |
 *
 * - Source and destination register must not be overlapped.
 * - components units are measured in terms of the smaller type between
 *   source and destination because we are un/shuffling the smaller
 *   components from/into the bigger ones.
 * - first_component parameter allows skipping source components.
 */
void
shuffle_src_to_dst(const fs_builder &bld,
                   const fs_reg &dst,
                   const fs_reg &src,
                   uint32_t first_component,
                   uint32_t components)
{
   if (type_sz(src.type) == type_sz(dst.type)) {
      assert(!regions_overlap(dst,
         type_sz(dst.type) * bld.dispatch_width() * components,
         offset(src, bld, first_component),
         type_sz(src.type) * bld.dispatch_width() * components));
      for (unsigned i = 0; i < components; i++) {
         bld.MOV(retype(offset(dst, bld, i), src.type),
                 offset(src, bld, i + first_component));
      }
   } else if (type_sz(src.type) < type_sz(dst.type)) {
      /* Source is shuffled into destination */
      unsigned size_ratio = type_sz(dst.type) / type_sz(src.type);
      assert(!regions_overlap(dst,
         type_sz(dst.type) * bld.dispatch_width() *
         DIV_ROUND_UP(components, size_ratio),
         offset(src, bld, first_component),
         type_sz(src.type) * bld.dispatch_width() * components));

      brw_reg_type shuffle_type =
         brw_reg_type_from_bit_size(8 * type_sz(src.type),
                                    BRW_REGISTER_TYPE_D);
      for (unsigned i = 0; i < components; i++) {
         fs_reg shuffle_component_i =
            subscript(offset(dst, bld, i / size_ratio),
                      shuffle_type, i % size_ratio);
         bld.MOV(shuffle_component_i,
                 retype(offset(src, bld, i + first_component), shuffle_type));
      }
   } else {
      /* Source is unshuffled into destination */
      unsigned size_ratio = type_sz(src.type) / type_sz(dst.type);
      assert(!regions_overlap(dst,
         type_sz(dst.type) * bld.dispatch_width() * components,
         offset(src, bld, first_component / size_ratio),
         type_sz(src.type) * bld.dispatch_width() *
         DIV_ROUND_UP(components + (first_component % size_ratio),
                      size_ratio)));

      brw_reg_type shuffle_type =
         brw_reg_type_from_bit_size(8 * type_sz(dst.type),
                                    BRW_REGISTER_TYPE_D);
      for (unsigned i = 0; i < components; i++) {
         fs_reg shuffle_component_i =
            subscript(offset(src, bld, (first_component + i) / size_ratio),
                      shuffle_type, (first_component + i) % size_ratio);
         bld.MOV(retype(offset(dst, bld, i), shuffle_type),
                 shuffle_component_i);
      }
   }
}

void
shuffle_from_32bit_read(const fs_builder &bld,
                        const fs_reg &dst,
                        const fs_reg &src,
                        uint32_t first_component,
                        uint32_t components)
{
   assert(type_sz(src.type) == 4);

   /* This function takes components in units of the destination type while
    * shuffle_src_to_dst takes components in units of the smallest type
    */
   if (type_sz(dst.type) > 4) {
      assert(type_sz(dst.type) == 8);
      first_component *= 2;
      components *= 2;
   }

   shuffle_src_to_dst(bld, dst, src, first_component, components);
}

fs_reg
shuffle_for_32bit_write(const fs_builder &bld,
                        const fs_reg &src,
                        uint32_t first_component,
                        uint32_t components)
{
   fs_reg dst = bld.vgrf(BRW_REGISTER_TYPE_D,
                         DIV_ROUND_UP (components * type_sz(src.type), 4));
   /* This function takes components in units of the source type while
    * shuffle_src_to_dst takes components in units of the smallest type
    */
   if (type_sz(src.type) > 4) {
      assert(type_sz(src.type) == 8);
      first_component *= 2;
      components *= 2;
   }

   shuffle_src_to_dst(bld, dst, src, first_component, components);

   return dst;
}

fs_reg
setup_imm_df(const fs_builder &bld, double v)
{
   const struct gen_device_info *devinfo = bld.shader->devinfo;
   assert(devinfo->gen >= 7);

   if (devinfo->gen >= 8)
      return brw_imm_df(v);

   /* gen7.5 does not support DF immediates straighforward but the DIM
    * instruction allows to set the 64-bit immediate value.
    */
   if (devinfo->is_haswell) {
      const fs_builder ubld = bld.exec_all().group(1, 0);
      fs_reg dst = ubld.vgrf(BRW_REGISTER_TYPE_DF, 1);
      ubld.DIM(dst, brw_imm_df(v));
      return component(dst, 0);
   }

   /* gen7 does not support DF immediates, so we generate a 64-bit constant by
    * writing the low 32-bit of the constant to suboffset 0 of a VGRF and
    * the high 32-bit to suboffset 4 and then applying a stride of 0.
    *
    * Alternatively, we could also produce a normal VGRF (without stride 0)
    * by writing to all the channels in the VGRF, however, that would hit the
    * gen7 bug where we have to split writes that span more than 1 register
    * into instructions with a width of 4 (otherwise the write to the second
    * register written runs into an execmask hardware bug) which isn't very
    * nice.
    */
   union {
      double d;
      struct {
         uint32_t i1;
         uint32_t i2;
      };
   } di;

   di.d = v;

   const fs_builder ubld = bld.exec_all().group(1, 0);
   const fs_reg tmp = ubld.vgrf(BRW_REGISTER_TYPE_UD, 2);
   ubld.MOV(tmp, brw_imm_ud(di.i1));
   ubld.MOV(horiz_offset(tmp, 1), brw_imm_ud(di.i2));

   return component(retype(tmp, BRW_REGISTER_TYPE_DF), 0);
}