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
|
/*
* 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.
*/
/** @file brw_fs.cpp
*
* This file drives the GLSL IR -> LIR translation, contains the
* optimizations on the LIR, and drives the generation of native code
* from the LIR.
*/
extern "C" {
#include <sys/types.h>
#include "main/hash_table.h"
#include "main/macros.h"
#include "main/shaderobj.h"
#include "main/fbobject.h"
#include "program/prog_parameter.h"
#include "program/prog_print.h"
#include "program/register_allocate.h"
#include "program/sampler.h"
#include "program/hash_table.h"
#include "brw_context.h"
#include "brw_eu.h"
#include "brw_wm.h"
}
#include "brw_fs.h"
#include "main/uniforms.h"
#include "glsl/glsl_types.h"
void
fs_inst::init()
{
memset(this, 0, sizeof(*this));
this->opcode = BRW_OPCODE_NOP;
this->conditional_mod = BRW_CONDITIONAL_NONE;
this->dst = reg_undef;
this->src[0] = reg_undef;
this->src[1] = reg_undef;
this->src[2] = reg_undef;
/* This will be the case for almost all instructions. */
this->regs_written = 1;
}
fs_inst::fs_inst()
{
init();
}
fs_inst::fs_inst(enum opcode opcode)
{
init();
this->opcode = opcode;
}
fs_inst::fs_inst(enum opcode opcode, fs_reg dst)
{
init();
this->opcode = opcode;
this->dst = dst;
if (dst.file == GRF)
assert(dst.reg_offset >= 0);
}
fs_inst::fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0)
{
init();
this->opcode = opcode;
this->dst = dst;
this->src[0] = src0;
if (dst.file == GRF)
assert(dst.reg_offset >= 0);
if (src[0].file == GRF)
assert(src[0].reg_offset >= 0);
}
fs_inst::fs_inst(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1)
{
init();
this->opcode = opcode;
this->dst = dst;
this->src[0] = src0;
this->src[1] = src1;
if (dst.file == GRF)
assert(dst.reg_offset >= 0);
if (src[0].file == GRF)
assert(src[0].reg_offset >= 0);
if (src[1].file == GRF)
assert(src[1].reg_offset >= 0);
}
fs_inst::fs_inst(enum opcode opcode, fs_reg dst,
fs_reg src0, fs_reg src1, fs_reg src2)
{
init();
this->opcode = opcode;
this->dst = dst;
this->src[0] = src0;
this->src[1] = src1;
this->src[2] = src2;
if (dst.file == GRF)
assert(dst.reg_offset >= 0);
if (src[0].file == GRF)
assert(src[0].reg_offset >= 0);
if (src[1].file == GRF)
assert(src[1].reg_offset >= 0);
if (src[2].file == GRF)
assert(src[2].reg_offset >= 0);
}
#define ALU1(op) \
fs_inst * \
fs_visitor::op(fs_reg dst, fs_reg src0) \
{ \
return new(mem_ctx) fs_inst(BRW_OPCODE_##op, dst, src0); \
}
#define ALU2(op) \
fs_inst * \
fs_visitor::op(fs_reg dst, fs_reg src0, fs_reg src1) \
{ \
return new(mem_ctx) fs_inst(BRW_OPCODE_##op, dst, src0, src1); \
}
#define ALU3(op) \
fs_inst * \
fs_visitor::op(fs_reg dst, fs_reg src0, fs_reg src1, fs_reg src2) \
{ \
return new(mem_ctx) fs_inst(BRW_OPCODE_##op, dst, src0, src1, src2);\
}
ALU1(NOT)
ALU1(MOV)
ALU1(FRC)
ALU1(RNDD)
ALU1(RNDE)
ALU1(RNDZ)
ALU2(ADD)
ALU2(MUL)
ALU2(MACH)
ALU2(AND)
ALU2(OR)
ALU2(XOR)
ALU2(SHL)
ALU2(SHR)
ALU2(ASR)
ALU3(LRP)
ALU1(BFREV)
ALU3(BFE)
ALU2(BFI1)
ALU3(BFI2)
ALU1(FBH)
ALU1(FBL)
ALU1(CBIT)
ALU3(MAD)
ALU2(ADDC)
ALU2(SUBB)
/** Gen4 predicated IF. */
fs_inst *
fs_visitor::IF(uint32_t predicate)
{
fs_inst *inst = new(mem_ctx) fs_inst(BRW_OPCODE_IF);
inst->predicate = predicate;
return inst;
}
/** Gen6+ IF with embedded comparison. */
fs_inst *
fs_visitor::IF(fs_reg src0, fs_reg src1, uint32_t condition)
{
assert(brw->gen >= 6);
fs_inst *inst = new(mem_ctx) fs_inst(BRW_OPCODE_IF,
reg_null_d, src0, src1);
inst->conditional_mod = condition;
return inst;
}
/**
* CMP: Sets the low bit of the destination channels with the result
* of the comparison, while the upper bits are undefined, and updates
* the flag register with the packed 16 bits of the result.
*/
fs_inst *
fs_visitor::CMP(fs_reg dst, fs_reg src0, fs_reg src1, uint32_t condition)
{
fs_inst *inst;
/* Take the instruction:
*
* CMP null<d> src0<f> src1<f>
*
* Original gen4 does type conversion to the destination type before
* comparison, producing garbage results for floating point comparisons.
* gen5 does the comparison on the execution type (resolved source types),
* so dst type doesn't matter. gen6 does comparison and then uses the
* result as if it was the dst type with no conversion, which happens to
* mostly work out for float-interpreted-as-int since our comparisons are
* for >0, =0, <0.
*/
if (brw->gen == 4) {
dst.type = src0.type;
if (dst.file == HW_REG)
dst.fixed_hw_reg.type = dst.type;
}
resolve_ud_negate(&src0);
resolve_ud_negate(&src1);
inst = new(mem_ctx) fs_inst(BRW_OPCODE_CMP, dst, src0, src1);
inst->conditional_mod = condition;
return inst;
}
exec_list
fs_visitor::VARYING_PULL_CONSTANT_LOAD(fs_reg dst, fs_reg surf_index,
fs_reg varying_offset,
uint32_t const_offset)
{
exec_list instructions;
fs_inst *inst;
/* We have our constant surface use a pitch of 4 bytes, so our index can
* be any component of a vector, and then we load 4 contiguous
* components starting from that.
*
* We break down the const_offset to a portion added to the variable
* offset and a portion done using reg_offset, which means that if you
* have GLSL using something like "uniform vec4 a[20]; gl_FragColor =
* a[i]", we'll temporarily generate 4 vec4 loads from offset i * 4, and
* CSE can later notice that those loads are all the same and eliminate
* the redundant ones.
*/
fs_reg vec4_offset = fs_reg(this, glsl_type::int_type);
instructions.push_tail(ADD(vec4_offset,
varying_offset, const_offset & ~3));
int scale = 1;
if (brw->gen == 4 && dispatch_width == 8) {
/* Pre-gen5, we can either use a SIMD8 message that requires (header,
* u, v, r) as parameters, or we can just use the SIMD16 message
* consisting of (header, u). We choose the second, at the cost of a
* longer return length.
*/
scale = 2;
}
enum opcode op;
if (brw->gen >= 7)
op = FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7;
else
op = FS_OPCODE_VARYING_PULL_CONSTANT_LOAD;
fs_reg vec4_result = fs_reg(GRF, virtual_grf_alloc(4 * scale), dst.type);
inst = new(mem_ctx) fs_inst(op, vec4_result, surf_index, vec4_offset);
inst->regs_written = 4 * scale;
instructions.push_tail(inst);
if (brw->gen < 7) {
inst->base_mrf = 13;
inst->header_present = true;
if (brw->gen == 4)
inst->mlen = 3;
else
inst->mlen = 1 + dispatch_width / 8;
}
vec4_result.reg_offset += (const_offset & 3) * scale;
instructions.push_tail(MOV(dst, vec4_result));
return instructions;
}
/**
* A helper for MOV generation for fixing up broken hardware SEND dependency
* handling.
*/
fs_inst *
fs_visitor::DEP_RESOLVE_MOV(int grf)
{
fs_inst *inst = MOV(brw_null_reg(), fs_reg(GRF, grf, BRW_REGISTER_TYPE_F));
inst->ir = NULL;
inst->annotation = "send dependency resolve";
/* The caller always wants uncompressed to emit the minimal extra
* dependencies, and to avoid having to deal with aligning its regs to 2.
*/
inst->force_uncompressed = true;
return inst;
}
bool
fs_inst::equals(fs_inst *inst)
{
return (opcode == inst->opcode &&
dst.equals(inst->dst) &&
src[0].equals(inst->src[0]) &&
src[1].equals(inst->src[1]) &&
src[2].equals(inst->src[2]) &&
saturate == inst->saturate &&
predicate == inst->predicate &&
conditional_mod == inst->conditional_mod &&
mlen == inst->mlen &&
base_mrf == inst->base_mrf &&
sampler == inst->sampler &&
target == inst->target &&
eot == inst->eot &&
header_present == inst->header_present &&
shadow_compare == inst->shadow_compare &&
offset == inst->offset);
}
bool
fs_inst::overwrites_reg(const fs_reg ®)
{
return (reg.file == dst.file &&
reg.reg == dst.reg &&
reg.reg_offset >= dst.reg_offset &&
reg.reg_offset < dst.reg_offset + regs_written);
}
bool
fs_inst::is_send_from_grf()
{
return (opcode == FS_OPCODE_VARYING_PULL_CONSTANT_LOAD_GEN7 ||
opcode == SHADER_OPCODE_SHADER_TIME_ADD ||
(opcode == FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD &&
src[1].file == GRF));
}
bool
fs_visitor::can_do_source_mods(fs_inst *inst)
{
if (brw->gen == 6 && inst->is_math())
return false;
if (inst->is_send_from_grf())
return false;
return true;
}
void
fs_reg::init()
{
memset(this, 0, sizeof(*this));
this->smear = -1;
}
/** Generic unset register constructor. */
fs_reg::fs_reg()
{
init();
this->file = BAD_FILE;
}
/** Immediate value constructor. */
fs_reg::fs_reg(float f)
{
init();
this->file = IMM;
this->type = BRW_REGISTER_TYPE_F;
this->imm.f = f;
}
/** Immediate value constructor. */
fs_reg::fs_reg(int32_t i)
{
init();
this->file = IMM;
this->type = BRW_REGISTER_TYPE_D;
this->imm.i = i;
}
/** Immediate value constructor. */
fs_reg::fs_reg(uint32_t u)
{
init();
this->file = IMM;
this->type = BRW_REGISTER_TYPE_UD;
this->imm.u = u;
}
/** Fixed brw_reg Immediate value constructor. */
fs_reg::fs_reg(struct brw_reg fixed_hw_reg)
{
init();
this->file = HW_REG;
this->fixed_hw_reg = fixed_hw_reg;
this->type = fixed_hw_reg.type;
}
bool
fs_reg::equals(const fs_reg &r) const
{
return (file == r.file &&
reg == r.reg &&
reg_offset == r.reg_offset &&
type == r.type &&
negate == r.negate &&
abs == r.abs &&
!reladdr && !r.reladdr &&
memcmp(&fixed_hw_reg, &r.fixed_hw_reg,
sizeof(fixed_hw_reg)) == 0 &&
smear == r.smear &&
imm.u == r.imm.u);
}
bool
fs_reg::is_zero() const
{
if (file != IMM)
return false;
return type == BRW_REGISTER_TYPE_F ? imm.f == 0.0 : imm.i == 0;
}
bool
fs_reg::is_one() const
{
if (file != IMM)
return false;
return type == BRW_REGISTER_TYPE_F ? imm.f == 1.0 : imm.i == 1;
}
bool
fs_reg::is_valid_3src() const
{
return file == GRF || file == UNIFORM;
}
int
fs_visitor::type_size(const struct glsl_type *type)
{
unsigned int size, i;
switch (type->base_type) {
case GLSL_TYPE_UINT:
case GLSL_TYPE_INT:
case GLSL_TYPE_FLOAT:
case GLSL_TYPE_BOOL:
return type->components();
case GLSL_TYPE_ARRAY:
return type_size(type->fields.array) * type->length;
case GLSL_TYPE_STRUCT:
size = 0;
for (i = 0; i < type->length; i++) {
size += type_size(type->fields.structure[i].type);
}
return size;
case GLSL_TYPE_SAMPLER:
/* Samplers take up no register space, since they're baked in at
* link time.
*/
return 0;
case GLSL_TYPE_VOID:
case GLSL_TYPE_ERROR:
case GLSL_TYPE_INTERFACE:
assert(!"not reached");
break;
}
return 0;
}
fs_reg
fs_visitor::get_timestamp()
{
assert(brw->gen >= 7);
fs_reg ts = fs_reg(retype(brw_vec1_reg(BRW_ARCHITECTURE_REGISTER_FILE,
BRW_ARF_TIMESTAMP,
0),
BRW_REGISTER_TYPE_UD));
fs_reg dst = fs_reg(this, glsl_type::uint_type);
fs_inst *mov = emit(MOV(dst, ts));
/* We want to read the 3 fields we care about (mostly field 0, but also 2)
* even if it's not enabled in the dispatch.
*/
mov->force_writemask_all = true;
mov->force_uncompressed = true;
/* The caller wants the low 32 bits of the timestamp. Since it's running
* at the GPU clock rate of ~1.2ghz, it will roll over every ~3 seconds,
* which is plenty of time for our purposes. It is identical across the
* EUs, but since it's tracking GPU core speed it will increment at a
* varying rate as render P-states change.
*
* The caller could also check if render P-states have changed (or anything
* else that might disrupt timing) by setting smear to 2 and checking if
* that field is != 0.
*/
dst.smear = 0;
return dst;
}
void
fs_visitor::emit_shader_time_begin()
{
current_annotation = "shader time start";
shader_start_time = get_timestamp();
}
void
fs_visitor::emit_shader_time_end()
{
current_annotation = "shader time end";
enum shader_time_shader_type type, written_type, reset_type;
if (dispatch_width == 8) {
type = ST_FS8;
written_type = ST_FS8_WRITTEN;
reset_type = ST_FS8_RESET;
} else {
assert(dispatch_width == 16);
type = ST_FS16;
written_type = ST_FS16_WRITTEN;
reset_type = ST_FS16_RESET;
}
fs_reg shader_end_time = get_timestamp();
/* Check that there weren't any timestamp reset events (assuming these
* were the only two timestamp reads that happened).
*/
fs_reg reset = shader_end_time;
reset.smear = 2;
fs_inst *test = emit(AND(reg_null_d, reset, fs_reg(1u)));
test->conditional_mod = BRW_CONDITIONAL_Z;
emit(IF(BRW_PREDICATE_NORMAL));
push_force_uncompressed();
fs_reg start = shader_start_time;
start.negate = true;
fs_reg diff = fs_reg(this, glsl_type::uint_type);
emit(ADD(diff, start, shader_end_time));
/* If there were no instructions between the two timestamp gets, the diff
* is 2 cycles. Remove that overhead, so I can forget about that when
* trying to determine the time taken for single instructions.
*/
emit(ADD(diff, diff, fs_reg(-2u)));
emit_shader_time_write(type, diff);
emit_shader_time_write(written_type, fs_reg(1u));
emit(BRW_OPCODE_ELSE);
emit_shader_time_write(reset_type, fs_reg(1u));
emit(BRW_OPCODE_ENDIF);
pop_force_uncompressed();
}
void
fs_visitor::emit_shader_time_write(enum shader_time_shader_type type,
fs_reg value)
{
int shader_time_index =
brw_get_shader_time_index(brw, shader_prog, &fp->Base, type);
fs_reg offset = fs_reg(shader_time_index * SHADER_TIME_STRIDE);
fs_reg payload;
if (dispatch_width == 8)
payload = fs_reg(this, glsl_type::uvec2_type);
else
payload = fs_reg(this, glsl_type::uint_type);
emit(fs_inst(SHADER_OPCODE_SHADER_TIME_ADD,
fs_reg(), payload, offset, value));
}
void
fs_visitor::fail(const char *format, ...)
{
va_list va;
char *msg;
if (failed)
return;
failed = true;
va_start(va, format);
msg = ralloc_vasprintf(mem_ctx, format, va);
va_end(va);
msg = ralloc_asprintf(mem_ctx, "FS compile failed: %s\n", msg);
this->fail_msg = msg;
if (INTEL_DEBUG & DEBUG_WM) {
fprintf(stderr, "%s", msg);
}
}
fs_inst *
fs_visitor::emit(enum opcode opcode)
{
return emit(fs_inst(opcode));
}
fs_inst *
fs_visitor::emit(enum opcode opcode, fs_reg dst)
{
return emit(fs_inst(opcode, dst));
}
fs_inst *
fs_visitor::emit(enum opcode opcode, fs_reg dst, fs_reg src0)
{
return emit(fs_inst(opcode, dst, src0));
}
fs_inst *
fs_visitor::emit(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1)
{
return emit(fs_inst(opcode, dst, src0, src1));
}
fs_inst *
fs_visitor::emit(enum opcode opcode, fs_reg dst,
fs_reg src0, fs_reg src1, fs_reg src2)
{
return emit(fs_inst(opcode, dst, src0, src1, src2));
}
void
fs_visitor::push_force_uncompressed()
{
force_uncompressed_stack++;
}
void
fs_visitor::pop_force_uncompressed()
{
force_uncompressed_stack--;
assert(force_uncompressed_stack >= 0);
}
void
fs_visitor::push_force_sechalf()
{
force_sechalf_stack++;
}
void
fs_visitor::pop_force_sechalf()
{
force_sechalf_stack--;
assert(force_sechalf_stack >= 0);
}
/**
* Returns true if the instruction has a flag that means it won't
* update an entire destination register.
*
* For example, dead code elimination and live variable analysis want to know
* when a write to a variable screens off any preceding values that were in
* it.
*/
bool
fs_inst::is_partial_write()
{
return ((this->predicate && this->opcode != BRW_OPCODE_SEL) ||
this->force_uncompressed ||
this->force_sechalf);
}
/**
* Returns how many MRFs an FS opcode will write over.
*
* Note that this is not the 0 or 1 implied writes in an actual gen
* instruction -- the FS opcodes often generate MOVs in addition.
*/
int
fs_visitor::implied_mrf_writes(fs_inst *inst)
{
if (inst->mlen == 0)
return 0;
switch (inst->opcode) {
case SHADER_OPCODE_RCP:
case SHADER_OPCODE_RSQ:
case SHADER_OPCODE_SQRT:
case SHADER_OPCODE_EXP2:
case SHADER_OPCODE_LOG2:
case SHADER_OPCODE_SIN:
case SHADER_OPCODE_COS:
return 1 * dispatch_width / 8;
case SHADER_OPCODE_POW:
case SHADER_OPCODE_INT_QUOTIENT:
case SHADER_OPCODE_INT_REMAINDER:
return 2 * dispatch_width / 8;
case SHADER_OPCODE_TEX:
case FS_OPCODE_TXB:
case SHADER_OPCODE_TXD:
case SHADER_OPCODE_TXF:
case SHADER_OPCODE_TXF_MS:
case SHADER_OPCODE_TG4:
case SHADER_OPCODE_TXL:
case SHADER_OPCODE_TXS:
case SHADER_OPCODE_LOD:
return 1;
case FS_OPCODE_FB_WRITE:
return 2;
case FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD:
case FS_OPCODE_UNSPILL:
return 1;
case FS_OPCODE_VARYING_PULL_CONSTANT_LOAD:
return inst->mlen;
case FS_OPCODE_SPILL:
return 2;
default:
assert(!"not reached");
return inst->mlen;
}
}
int
fs_visitor::virtual_grf_alloc(int size)
{
if (virtual_grf_array_size <= virtual_grf_count) {
if (virtual_grf_array_size == 0)
virtual_grf_array_size = 16;
else
virtual_grf_array_size *= 2;
virtual_grf_sizes = reralloc(mem_ctx, virtual_grf_sizes, int,
virtual_grf_array_size);
}
virtual_grf_sizes[virtual_grf_count] = size;
return virtual_grf_count++;
}
/** Fixed HW reg constructor. */
fs_reg::fs_reg(enum register_file file, int reg)
{
init();
this->file = file;
this->reg = reg;
this->type = BRW_REGISTER_TYPE_F;
}
/** Fixed HW reg constructor. */
fs_reg::fs_reg(enum register_file file, int reg, uint32_t type)
{
init();
this->file = file;
this->reg = reg;
this->type = type;
}
/** Automatic reg constructor. */
fs_reg::fs_reg(class fs_visitor *v, const struct glsl_type *type)
{
init();
this->file = GRF;
this->reg = v->virtual_grf_alloc(v->type_size(type));
this->reg_offset = 0;
this->type = brw_type_for_base_type(type);
}
fs_reg *
fs_visitor::variable_storage(ir_variable *var)
{
return (fs_reg *)hash_table_find(this->variable_ht, var);
}
void
import_uniforms_callback(const void *key,
void *data,
void *closure)
{
struct hash_table *dst_ht = (struct hash_table *)closure;
const fs_reg *reg = (const fs_reg *)data;
if (reg->file != UNIFORM)
return;
hash_table_insert(dst_ht, data, key);
}
/* For 16-wide, we need to follow from the uniform setup of 8-wide dispatch.
* This brings in those uniform definitions
*/
void
fs_visitor::import_uniforms(fs_visitor *v)
{
hash_table_call_foreach(v->variable_ht,
import_uniforms_callback,
variable_ht);
this->params_remap = v->params_remap;
this->nr_params_remap = v->nr_params_remap;
}
/* Our support for uniforms is piggy-backed on the struct
* gl_fragment_program, because that's where the values actually
* get stored, rather than in some global gl_shader_program uniform
* store.
*/
void
fs_visitor::setup_uniform_values(ir_variable *ir)
{
int namelen = strlen(ir->name);
/* The data for our (non-builtin) uniforms is stored in a series of
* gl_uniform_driver_storage structs for each subcomponent that
* glGetUniformLocation() could name. We know it's been set up in the same
* order we'd walk the type, so walk the list of storage and find anything
* with our name, or the prefix of a component that starts with our name.
*/
unsigned params_before = c->prog_data.nr_params;
for (unsigned u = 0; u < shader_prog->NumUserUniformStorage; u++) {
struct gl_uniform_storage *storage = &shader_prog->UniformStorage[u];
if (strncmp(ir->name, storage->name, namelen) != 0 ||
(storage->name[namelen] != 0 &&
storage->name[namelen] != '.' &&
storage->name[namelen] != '[')) {
continue;
}
unsigned slots = storage->type->component_slots();
if (storage->array_elements)
slots *= storage->array_elements;
for (unsigned i = 0; i < slots; i++) {
c->prog_data.param[c->prog_data.nr_params++] =
&storage->storage[i].f;
}
}
/* Make sure we actually initialized the right amount of stuff here. */
assert(params_before + ir->type->component_slots() ==
c->prog_data.nr_params);
(void)params_before;
}
/* Our support for builtin uniforms is even scarier than non-builtin.
* It sits on top of the PROG_STATE_VAR parameters that are
* automatically updated from GL context state.
*/
void
fs_visitor::setup_builtin_uniform_values(ir_variable *ir)
{
const ir_state_slot *const slots = ir->state_slots;
assert(ir->state_slots != NULL);
for (unsigned int i = 0; i < ir->num_state_slots; i++) {
/* This state reference has already been setup by ir_to_mesa, but we'll
* get the same index back here.
*/
int index = _mesa_add_state_reference(this->fp->Base.Parameters,
(gl_state_index *)slots[i].tokens);
/* Add each of the unique swizzles of the element as a parameter.
* This'll end up matching the expected layout of the
* array/matrix/structure we're trying to fill in.
*/
int last_swiz = -1;
for (unsigned int j = 0; j < 4; j++) {
int swiz = GET_SWZ(slots[i].swizzle, j);
if (swiz == last_swiz)
break;
last_swiz = swiz;
c->prog_data.param[c->prog_data.nr_params++] =
&fp->Base.Parameters->ParameterValues[index][swiz].f;
}
}
}
fs_reg *
fs_visitor::emit_fragcoord_interpolation(ir_variable *ir)
{
fs_reg *reg = new(this->mem_ctx) fs_reg(this, ir->type);
fs_reg wpos = *reg;
bool flip = !ir->origin_upper_left ^ c->key.render_to_fbo;
/* gl_FragCoord.x */
if (ir->pixel_center_integer) {
emit(MOV(wpos, this->pixel_x));
} else {
emit(ADD(wpos, this->pixel_x, fs_reg(0.5f)));
}
wpos.reg_offset++;
/* gl_FragCoord.y */
if (!flip && ir->pixel_center_integer) {
emit(MOV(wpos, this->pixel_y));
} else {
fs_reg pixel_y = this->pixel_y;
float offset = (ir->pixel_center_integer ? 0.0 : 0.5);
if (flip) {
pixel_y.negate = true;
offset += c->key.drawable_height - 1.0;
}
emit(ADD(wpos, pixel_y, fs_reg(offset)));
}
wpos.reg_offset++;
/* gl_FragCoord.z */
if (brw->gen >= 6) {
emit(MOV(wpos, fs_reg(brw_vec8_grf(c->source_depth_reg, 0))));
} else {
emit(FS_OPCODE_LINTERP, wpos,
this->delta_x[BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC],
this->delta_y[BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC],
interp_reg(VARYING_SLOT_POS, 2));
}
wpos.reg_offset++;
/* gl_FragCoord.w: Already set up in emit_interpolation */
emit(BRW_OPCODE_MOV, wpos, this->wpos_w);
return reg;
}
fs_inst *
fs_visitor::emit_linterp(const fs_reg &attr, const fs_reg &interp,
glsl_interp_qualifier interpolation_mode,
bool is_centroid)
{
brw_wm_barycentric_interp_mode barycoord_mode;
if (brw->gen >= 6) {
if (is_centroid) {
if (interpolation_mode == INTERP_QUALIFIER_SMOOTH)
barycoord_mode = BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
else
barycoord_mode = BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
} else {
if (interpolation_mode == INTERP_QUALIFIER_SMOOTH)
barycoord_mode = BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
else
barycoord_mode = BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
}
} else {
/* On Ironlake and below, there is only one interpolation mode.
* Centroid interpolation doesn't mean anything on this hardware --
* there is no multisampling.
*/
barycoord_mode = BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
}
return emit(FS_OPCODE_LINTERP, attr,
this->delta_x[barycoord_mode],
this->delta_y[barycoord_mode], interp);
}
fs_reg *
fs_visitor::emit_general_interpolation(ir_variable *ir)
{
fs_reg *reg = new(this->mem_ctx) fs_reg(this, ir->type);
reg->type = brw_type_for_base_type(ir->type->get_scalar_type());
fs_reg attr = *reg;
unsigned int array_elements;
const glsl_type *type;
if (ir->type->is_array()) {
array_elements = ir->type->length;
if (array_elements == 0) {
fail("dereferenced array '%s' has length 0\n", ir->name);
}
type = ir->type->fields.array;
} else {
array_elements = 1;
type = ir->type;
}
glsl_interp_qualifier interpolation_mode =
ir->determine_interpolation_mode(c->key.flat_shade);
int location = ir->location;
for (unsigned int i = 0; i < array_elements; i++) {
for (unsigned int j = 0; j < type->matrix_columns; j++) {
if (c->prog_data.urb_setup[location] == -1) {
/* If there's no incoming setup data for this slot, don't
* emit interpolation for it.
*/
attr.reg_offset += type->vector_elements;
location++;
continue;
}
if (interpolation_mode == INTERP_QUALIFIER_FLAT) {
/* Constant interpolation (flat shading) case. The SF has
* handed us defined values in only the constant offset
* field of the setup reg.
*/
for (unsigned int k = 0; k < type->vector_elements; k++) {
struct brw_reg interp = interp_reg(location, k);
interp = suboffset(interp, 3);
interp.type = reg->type;
emit(FS_OPCODE_CINTERP, attr, fs_reg(interp));
attr.reg_offset++;
}
} else {
/* Smooth/noperspective interpolation case. */
for (unsigned int k = 0; k < type->vector_elements; k++) {
/* FINISHME: At some point we probably want to push
* this farther by giving similar treatment to the
* other potentially constant components of the
* attribute, as well as making brw_vs_constval.c
* handle varyings other than gl_TexCoord.
*/
struct brw_reg interp = interp_reg(location, k);
emit_linterp(attr, fs_reg(interp), interpolation_mode,
ir->centroid);
if (brw->needs_unlit_centroid_workaround && ir->centroid) {
/* Get the pixel/sample mask into f0 so that we know
* which pixels are lit. Then, for each channel that is
* unlit, replace the centroid data with non-centroid
* data.
*/
emit(FS_OPCODE_MOV_DISPATCH_TO_FLAGS);
fs_inst *inst = emit_linterp(attr, fs_reg(interp),
interpolation_mode, false);
inst->predicate = BRW_PREDICATE_NORMAL;
inst->predicate_inverse = true;
}
if (brw->gen < 6 && interpolation_mode == INTERP_QUALIFIER_SMOOTH) {
emit(BRW_OPCODE_MUL, attr, attr, this->pixel_w);
}
attr.reg_offset++;
}
}
location++;
}
}
return reg;
}
fs_reg *
fs_visitor::emit_frontfacing_interpolation(ir_variable *ir)
{
fs_reg *reg = new(this->mem_ctx) fs_reg(this, ir->type);
/* The frontfacing comes in as a bit in the thread payload. */
if (brw->gen >= 6) {
emit(BRW_OPCODE_ASR, *reg,
fs_reg(retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_D)),
fs_reg(15));
emit(BRW_OPCODE_NOT, *reg, *reg);
emit(BRW_OPCODE_AND, *reg, *reg, fs_reg(1));
} else {
struct brw_reg r1_6ud = retype(brw_vec1_grf(1, 6), BRW_REGISTER_TYPE_UD);
/* bit 31 is "primitive is back face", so checking < (1 << 31) gives
* us front face
*/
emit(CMP(*reg, fs_reg(r1_6ud), fs_reg(1u << 31), BRW_CONDITIONAL_L));
emit(BRW_OPCODE_AND, *reg, *reg, fs_reg(1u));
}
return reg;
}
fs_reg
fs_visitor::fix_math_operand(fs_reg src)
{
/* Can't do hstride == 0 args on gen6 math, so expand it out. We
* might be able to do better by doing execsize = 1 math and then
* expanding that result out, but we would need to be careful with
* masking.
*
* The hardware ignores source modifiers (negate and abs) on math
* instructions, so we also move to a temp to set those up.
*/
if (brw->gen == 6 && src.file != UNIFORM && src.file != IMM &&
!src.abs && !src.negate)
return src;
/* Gen7 relaxes most of the above restrictions, but still can't use IMM
* operands to math
*/
if (brw->gen >= 7 && src.file != IMM)
return src;
fs_reg expanded = fs_reg(this, glsl_type::float_type);
expanded.type = src.type;
emit(BRW_OPCODE_MOV, expanded, src);
return expanded;
}
fs_inst *
fs_visitor::emit_math(enum opcode opcode, fs_reg dst, fs_reg src)
{
switch (opcode) {
case SHADER_OPCODE_RCP:
case SHADER_OPCODE_RSQ:
case SHADER_OPCODE_SQRT:
case SHADER_OPCODE_EXP2:
case SHADER_OPCODE_LOG2:
case SHADER_OPCODE_SIN:
case SHADER_OPCODE_COS:
break;
default:
assert(!"not reached: bad math opcode");
return NULL;
}
/* Can't do hstride == 0 args to gen6 math, so expand it out. We
* might be able to do better by doing execsize = 1 math and then
* expanding that result out, but we would need to be careful with
* masking.
*
* Gen 6 hardware ignores source modifiers (negate and abs) on math
* instructions, so we also move to a temp to set those up.
*/
if (brw->gen >= 6)
src = fix_math_operand(src);
fs_inst *inst = emit(opcode, dst, src);
if (brw->gen < 6) {
inst->base_mrf = 2;
inst->mlen = dispatch_width / 8;
}
return inst;
}
fs_inst *
fs_visitor::emit_math(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1)
{
int base_mrf = 2;
fs_inst *inst;
switch (opcode) {
case SHADER_OPCODE_INT_QUOTIENT:
case SHADER_OPCODE_INT_REMAINDER:
if (brw->gen >= 7 && dispatch_width == 16)
fail("16-wide INTDIV unsupported\n");
break;
case SHADER_OPCODE_POW:
break;
default:
assert(!"not reached: unsupported binary math opcode.");
return NULL;
}
if (brw->gen >= 6) {
src0 = fix_math_operand(src0);
src1 = fix_math_operand(src1);
inst = emit(opcode, dst, src0, src1);
} else {
/* From the Ironlake PRM, Volume 4, Part 1, Section 6.1.13
* "Message Payload":
*
* "Operand0[7]. For the INT DIV functions, this operand is the
* denominator."
* ...
* "Operand1[7]. For the INT DIV functions, this operand is the
* numerator."
*/
bool is_int_div = opcode != SHADER_OPCODE_POW;
fs_reg &op0 = is_int_div ? src1 : src0;
fs_reg &op1 = is_int_div ? src0 : src1;
emit(BRW_OPCODE_MOV, fs_reg(MRF, base_mrf + 1, op1.type), op1);
inst = emit(opcode, dst, op0, reg_null_f);
inst->base_mrf = base_mrf;
inst->mlen = 2 * dispatch_width / 8;
}
return inst;
}
void
fs_visitor::assign_curb_setup()
{
c->prog_data.curb_read_length = ALIGN(c->prog_data.nr_params, 8) / 8;
if (dispatch_width == 8) {
c->prog_data.first_curbe_grf = c->nr_payload_regs;
} else {
c->prog_data.first_curbe_grf_16 = c->nr_payload_regs;
}
/* Map the offsets in the UNIFORM file to fixed HW regs. */
foreach_list(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
for (unsigned int i = 0; i < 3; i++) {
if (inst->src[i].file == UNIFORM) {
int constant_nr = inst->src[i].reg + inst->src[i].reg_offset;
struct brw_reg brw_reg = brw_vec1_grf(c->nr_payload_regs +
constant_nr / 8,
constant_nr % 8);
inst->src[i].file = HW_REG;
inst->src[i].fixed_hw_reg = retype(brw_reg, inst->src[i].type);
}
}
}
}
void
fs_visitor::calculate_urb_setup()
{
for (unsigned int i = 0; i < VARYING_SLOT_MAX; i++) {
c->prog_data.urb_setup[i] = -1;
}
int urb_next = 0;
/* Figure out where each of the incoming setup attributes lands. */
if (brw->gen >= 6) {
if (_mesa_bitcount_64(fp->Base.InputsRead &
BRW_FS_VARYING_INPUT_MASK) <= 16) {
/* The SF/SBE pipeline stage can do arbitrary rearrangement of the
* first 16 varying inputs, so we can put them wherever we want.
* Just put them in order.
*
* This is useful because it means that (a) inputs not used by the
* fragment shader won't take up valuable register space, and (b) we
* won't have to recompile the fragment shader if it gets paired with
* a different vertex (or geometry) shader.
*/
for (unsigned int i = 0; i < VARYING_SLOT_MAX; i++) {
if (fp->Base.InputsRead & BRW_FS_VARYING_INPUT_MASK &
BITFIELD64_BIT(i)) {
c->prog_data.urb_setup[i] = urb_next++;
}
}
} else {
/* We have enough input varyings that the SF/SBE pipeline stage can't
* arbitrarily rearrange them to suit our whim; we have to put them
* in an order that matches the output of the previous pipeline stage
* (geometry or vertex shader).
*/
struct brw_vue_map prev_stage_vue_map;
brw_compute_vue_map(brw, &prev_stage_vue_map,
c->key.input_slots_valid);
int first_slot = 2 * BRW_SF_URB_ENTRY_READ_OFFSET;
assert(prev_stage_vue_map.num_slots <= first_slot + 32);
for (int slot = first_slot; slot < prev_stage_vue_map.num_slots;
slot++) {
int varying = prev_stage_vue_map.slot_to_varying[slot];
/* Note that varying == BRW_VARYING_SLOT_COUNT when a slot is
* unused.
*/
if (varying != BRW_VARYING_SLOT_COUNT &&
(fp->Base.InputsRead & BRW_FS_VARYING_INPUT_MASK &
BITFIELD64_BIT(varying))) {
c->prog_data.urb_setup[varying] = slot - first_slot;
}
}
urb_next = prev_stage_vue_map.num_slots - first_slot;
}
} else {
/* FINISHME: The sf doesn't map VS->FS inputs for us very well. */
for (unsigned int i = 0; i < VARYING_SLOT_MAX; i++) {
/* Point size is packed into the header, not as a general attribute */
if (i == VARYING_SLOT_PSIZ)
continue;
if (c->key.input_slots_valid & BITFIELD64_BIT(i)) {
/* The back color slot is skipped when the front color is
* also written to. In addition, some slots can be
* written in the vertex shader and not read in the
* fragment shader. So the register number must always be
* incremented, mapped or not.
*/
if (_mesa_varying_slot_in_fs((gl_varying_slot) i))
c->prog_data.urb_setup[i] = urb_next;
urb_next++;
}
}
/*
* It's a FS only attribute, and we did interpolation for this attribute
* in SF thread. So, count it here, too.
*
* See compile_sf_prog() for more info.
*/
if (fp->Base.InputsRead & BITFIELD64_BIT(VARYING_SLOT_PNTC))
c->prog_data.urb_setup[VARYING_SLOT_PNTC] = urb_next++;
}
c->prog_data.num_varying_inputs = urb_next;
}
void
fs_visitor::assign_urb_setup()
{
int urb_start = c->nr_payload_regs + c->prog_data.curb_read_length;
/* Offset all the urb_setup[] index by the actual position of the
* setup regs, now that the location of the constants has been chosen.
*/
foreach_list(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
if (inst->opcode == FS_OPCODE_LINTERP) {
assert(inst->src[2].file == HW_REG);
inst->src[2].fixed_hw_reg.nr += urb_start;
}
if (inst->opcode == FS_OPCODE_CINTERP) {
assert(inst->src[0].file == HW_REG);
inst->src[0].fixed_hw_reg.nr += urb_start;
}
}
/* Each attribute is 4 setup channels, each of which is half a reg. */
this->first_non_payload_grf =
urb_start + c->prog_data.num_varying_inputs * 2;
}
/**
* Split large virtual GRFs into separate components if we can.
*
* This is mostly duplicated with what brw_fs_vector_splitting does,
* but that's really conservative because it's afraid of doing
* splitting that doesn't result in real progress after the rest of
* the optimization phases, which would cause infinite looping in
* optimization. We can do it once here, safely. This also has the
* opportunity to split interpolated values, or maybe even uniforms,
* which we don't have at the IR level.
*
* We want to split, because virtual GRFs are what we register
* allocate and spill (due to contiguousness requirements for some
* instructions), and they're what we naturally generate in the
* codegen process, but most virtual GRFs don't actually need to be
* contiguous sets of GRFs. If we split, we'll end up with reduced
* live intervals and better dead code elimination and coalescing.
*/
void
fs_visitor::split_virtual_grfs()
{
int num_vars = this->virtual_grf_count;
bool split_grf[num_vars];
int new_virtual_grf[num_vars];
/* Try to split anything > 0 sized. */
for (int i = 0; i < num_vars; i++) {
if (this->virtual_grf_sizes[i] != 1)
split_grf[i] = true;
else
split_grf[i] = false;
}
if (brw->has_pln &&
this->delta_x[BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC].file == GRF) {
/* PLN opcodes rely on the delta_xy being contiguous. We only have to
* check this for BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC, because prior to
* Gen6, that was the only supported interpolation mode, and since Gen6,
* delta_x and delta_y are in fixed hardware registers.
*/
split_grf[this->delta_x[BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC].reg] =
false;
}
foreach_list(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
/* If there's a SEND message that requires contiguous destination
* registers, no splitting is allowed.
*/
if (inst->regs_written > 1) {
split_grf[inst->dst.reg] = false;
}
/* If we're sending from a GRF, don't split it, on the assumption that
* the send is reading the whole thing.
*/
if (inst->is_send_from_grf()) {
for (int i = 0; i < 3; i++) {
if (inst->src[i].file == GRF) {
split_grf[inst->src[i].reg] = false;
}
}
}
}
/* Allocate new space for split regs. Note that the virtual
* numbers will be contiguous.
*/
for (int i = 0; i < num_vars; i++) {
if (split_grf[i]) {
new_virtual_grf[i] = virtual_grf_alloc(1);
for (int j = 2; j < this->virtual_grf_sizes[i]; j++) {
int reg = virtual_grf_alloc(1);
assert(reg == new_virtual_grf[i] + j - 1);
(void) reg;
}
this->virtual_grf_sizes[i] = 1;
}
}
foreach_list(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
if (inst->dst.file == GRF &&
split_grf[inst->dst.reg] &&
inst->dst.reg_offset != 0) {
inst->dst.reg = (new_virtual_grf[inst->dst.reg] +
inst->dst.reg_offset - 1);
inst->dst.reg_offset = 0;
}
for (int i = 0; i < 3; i++) {
if (inst->src[i].file == GRF &&
split_grf[inst->src[i].reg] &&
inst->src[i].reg_offset != 0) {
inst->src[i].reg = (new_virtual_grf[inst->src[i].reg] +
inst->src[i].reg_offset - 1);
inst->src[i].reg_offset = 0;
}
}
}
invalidate_live_intervals();
}
/**
* Remove unused virtual GRFs and compact the virtual_grf_* arrays.
*
* During code generation, we create tons of temporary variables, many of
* which get immediately killed and are never used again. Yet, in later
* optimization and analysis passes, such as compute_live_intervals, we need
* to loop over all the virtual GRFs. Compacting them can save a lot of
* overhead.
*/
void
fs_visitor::compact_virtual_grfs()
{
/* Mark which virtual GRFs are used, and count how many. */
int remap_table[this->virtual_grf_count];
memset(remap_table, -1, sizeof(remap_table));
foreach_list(node, &this->instructions) {
const fs_inst *inst = (const fs_inst *) node;
if (inst->dst.file == GRF)
remap_table[inst->dst.reg] = 0;
for (int i = 0; i < 3; i++) {
if (inst->src[i].file == GRF)
remap_table[inst->src[i].reg] = 0;
}
}
/* In addition to registers used in instructions, fs_visitor keeps
* direct references to certain special values which must be patched:
*/
fs_reg *special[] = {
&frag_depth, &pixel_x, &pixel_y, &pixel_w, &wpos_w, &dual_src_output,
&outputs[0], &outputs[1], &outputs[2], &outputs[3],
&outputs[4], &outputs[5], &outputs[6], &outputs[7],
&delta_x[0], &delta_x[1], &delta_x[2],
&delta_x[3], &delta_x[4], &delta_x[5],
&delta_y[0], &delta_y[1], &delta_y[2],
&delta_y[3], &delta_y[4], &delta_y[5],
};
STATIC_ASSERT(BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT == 6);
STATIC_ASSERT(BRW_MAX_DRAW_BUFFERS == 8);
/* Treat all special values as used, to be conservative */
for (unsigned i = 0; i < ARRAY_SIZE(special); i++) {
if (special[i]->file == GRF)
remap_table[special[i]->reg] = 0;
}
/* Compact the GRF arrays. */
int new_index = 0;
for (int i = 0; i < this->virtual_grf_count; i++) {
if (remap_table[i] != -1) {
remap_table[i] = new_index;
virtual_grf_sizes[new_index] = virtual_grf_sizes[i];
invalidate_live_intervals();
++new_index;
}
}
this->virtual_grf_count = new_index;
/* Patch all the instructions to use the newly renumbered registers */
foreach_list(node, &this->instructions) {
fs_inst *inst = (fs_inst *) node;
if (inst->dst.file == GRF)
inst->dst.reg = remap_table[inst->dst.reg];
for (int i = 0; i < 3; i++) {
if (inst->src[i].file == GRF)
inst->src[i].reg = remap_table[inst->src[i].reg];
}
}
/* Patch all the references to special values */
for (unsigned i = 0; i < ARRAY_SIZE(special); i++) {
if (special[i]->file == GRF && remap_table[special[i]->reg] != -1)
special[i]->reg = remap_table[special[i]->reg];
}
}
bool
fs_visitor::remove_dead_constants()
{
if (dispatch_width == 8) {
this->params_remap = ralloc_array(mem_ctx, int, c->prog_data.nr_params);
this->nr_params_remap = c->prog_data.nr_params;
for (unsigned int i = 0; i < c->prog_data.nr_params; i++)
this->params_remap[i] = -1;
/* Find which params are still in use. */
foreach_list(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
for (int i = 0; i < 3; i++) {
int constant_nr = inst->src[i].reg + inst->src[i].reg_offset;
if (inst->src[i].file != UNIFORM)
continue;
/* Section 5.11 of the OpenGL 4.3 spec says:
*
* "Out-of-bounds reads return undefined values, which include
* values from other variables of the active program or zero."
*/
if (constant_nr < 0 || constant_nr >= (int)c->prog_data.nr_params) {
constant_nr = 0;
}
/* For now, set this to non-negative. We'll give it the
* actual new number in a moment, in order to keep the
* register numbers nicely ordered.
*/
this->params_remap[constant_nr] = 0;
}
}
/* Figure out what the new numbers for the params will be. At some
* point when we're doing uniform array access, we're going to want
* to keep the distinction between .reg and .reg_offset, but for
* now we don't care.
*/
unsigned int new_nr_params = 0;
for (unsigned int i = 0; i < c->prog_data.nr_params; i++) {
if (this->params_remap[i] != -1) {
this->params_remap[i] = new_nr_params++;
}
}
/* Update the list of params to be uploaded to match our new numbering. */
for (unsigned int i = 0; i < c->prog_data.nr_params; i++) {
int remapped = this->params_remap[i];
if (remapped == -1)
continue;
c->prog_data.param[remapped] = c->prog_data.param[i];
}
c->prog_data.nr_params = new_nr_params;
} else {
/* This should have been generated in the 8-wide pass already. */
assert(this->params_remap);
}
/* Now do the renumbering of the shader to remove unused params. */
foreach_list(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
for (int i = 0; i < 3; i++) {
int constant_nr = inst->src[i].reg + inst->src[i].reg_offset;
if (inst->src[i].file != UNIFORM)
continue;
/* as above alias to 0 */
if (constant_nr < 0 || constant_nr >= (int)this->nr_params_remap) {
constant_nr = 0;
}
assert(this->params_remap[constant_nr] != -1);
inst->src[i].reg = this->params_remap[constant_nr];
inst->src[i].reg_offset = 0;
}
}
return true;
}
/*
* Implements array access of uniforms by inserting a
* PULL_CONSTANT_LOAD instruction.
*
* Unlike temporary GRF array access (where we don't support it due to
* the difficulty of doing relative addressing on instruction
* destinations), we could potentially do array access of uniforms
* that were loaded in GRF space as push constants. In real-world
* usage we've seen, though, the arrays being used are always larger
* than we could load as push constants, so just always move all
* uniform array access out to a pull constant buffer.
*/
void
fs_visitor::move_uniform_array_access_to_pull_constants()
{
int pull_constant_loc[c->prog_data.nr_params];
for (unsigned int i = 0; i < c->prog_data.nr_params; i++) {
pull_constant_loc[i] = -1;
}
/* Walk through and find array access of uniforms. Put a copy of that
* uniform in the pull constant buffer.
*
* Note that we don't move constant-indexed accesses to arrays. No
* testing has been done of the performance impact of this choice.
*/
foreach_list_safe(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
for (int i = 0 ; i < 3; i++) {
if (inst->src[i].file != UNIFORM || !inst->src[i].reladdr)
continue;
int uniform = inst->src[i].reg;
/* If this array isn't already present in the pull constant buffer,
* add it.
*/
if (pull_constant_loc[uniform] == -1) {
const float **values = &c->prog_data.param[uniform];
pull_constant_loc[uniform] = c->prog_data.nr_pull_params;
assert(param_size[uniform]);
for (int j = 0; j < param_size[uniform]; j++) {
c->prog_data.pull_param[c->prog_data.nr_pull_params++] =
values[j];
}
}
/* Set up the annotation tracking for new generated instructions. */
base_ir = inst->ir;
current_annotation = inst->annotation;
fs_reg surf_index = fs_reg((unsigned)SURF_INDEX_FRAG_CONST_BUFFER);
fs_reg temp = fs_reg(this, glsl_type::float_type);
exec_list list = VARYING_PULL_CONSTANT_LOAD(temp,
surf_index,
*inst->src[i].reladdr,
pull_constant_loc[uniform] +
inst->src[i].reg_offset);
inst->insert_before(&list);
inst->src[i].file = temp.file;
inst->src[i].reg = temp.reg;
inst->src[i].reg_offset = temp.reg_offset;
inst->src[i].reladdr = NULL;
}
}
}
/**
* Choose accesses from the UNIFORM file to demote to using the pull
* constant buffer.
*
* We allow a fragment shader to have more than the specified minimum
* maximum number of fragment shader uniform components (64). If
* there are too many of these, they'd fill up all of register space.
* So, this will push some of them out to the pull constant buffer and
* update the program to load them.
*/
void
fs_visitor::setup_pull_constants()
{
/* Only allow 16 registers (128 uniform components) as push constants. */
unsigned int max_uniform_components = 16 * 8;
if (c->prog_data.nr_params <= max_uniform_components)
return;
if (dispatch_width == 16) {
fail("Pull constants not supported in 16-wide\n");
return;
}
/* Just demote the end of the list. We could probably do better
* here, demoting things that are rarely used in the program first.
*/
unsigned int pull_uniform_base = max_uniform_components;
int pull_constant_loc[c->prog_data.nr_params];
for (unsigned int i = 0; i < c->prog_data.nr_params; i++) {
if (i < pull_uniform_base) {
pull_constant_loc[i] = -1;
} else {
pull_constant_loc[i] = -1;
/* If our constant is already being uploaded for reladdr purposes,
* reuse it.
*/
for (unsigned int j = 0; j < c->prog_data.nr_pull_params; j++) {
if (c->prog_data.pull_param[j] == c->prog_data.param[i]) {
pull_constant_loc[i] = j;
break;
}
}
if (pull_constant_loc[i] == -1) {
int pull_index = c->prog_data.nr_pull_params++;
c->prog_data.pull_param[pull_index] = c->prog_data.param[i];
pull_constant_loc[i] = pull_index;;
}
}
}
c->prog_data.nr_params = pull_uniform_base;
foreach_list(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
for (int i = 0; i < 3; i++) {
if (inst->src[i].file != UNIFORM)
continue;
int pull_index = pull_constant_loc[inst->src[i].reg +
inst->src[i].reg_offset];
if (pull_index == -1)
continue;
assert(!inst->src[i].reladdr);
fs_reg dst = fs_reg(this, glsl_type::float_type);
fs_reg index = fs_reg((unsigned)SURF_INDEX_FRAG_CONST_BUFFER);
fs_reg offset = fs_reg((unsigned)(pull_index * 4) & ~15);
fs_inst *pull =
new(mem_ctx) fs_inst(FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD,
dst, index, offset);
pull->ir = inst->ir;
pull->annotation = inst->annotation;
inst->insert_before(pull);
inst->src[i].file = GRF;
inst->src[i].reg = dst.reg;
inst->src[i].reg_offset = 0;
inst->src[i].smear = pull_index & 3;
}
}
}
bool
fs_visitor::opt_algebraic()
{
bool progress = false;
foreach_list(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
switch (inst->opcode) {
case BRW_OPCODE_MUL:
if (inst->src[1].file != IMM)
continue;
/* a * 1.0 = a */
if (inst->src[1].is_one()) {
inst->opcode = BRW_OPCODE_MOV;
inst->src[1] = reg_undef;
progress = true;
break;
}
/* a * 0.0 = 0.0 */
if (inst->src[1].is_zero()) {
inst->opcode = BRW_OPCODE_MOV;
inst->src[0] = inst->src[1];
inst->src[1] = reg_undef;
progress = true;
break;
}
break;
case BRW_OPCODE_ADD:
if (inst->src[1].file != IMM)
continue;
/* a + 0.0 = a */
if (inst->src[1].is_zero()) {
inst->opcode = BRW_OPCODE_MOV;
inst->src[1] = reg_undef;
progress = true;
break;
}
break;
default:
break;
}
}
return progress;
}
/**
* Removes any instructions writing a VGRF where that VGRF is not used by any
* later instruction.
*/
bool
fs_visitor::dead_code_eliminate()
{
bool progress = false;
int pc = 0;
calculate_live_intervals();
foreach_list_safe(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
if (inst->dst.file == GRF) {
assert(this->virtual_grf_end[inst->dst.reg] >= pc);
if (this->virtual_grf_end[inst->dst.reg] == pc) {
/* Don't dead code eliminate instructions that write to the
* accumulator as a side-effect. Instead just set the destination
* to the null register to free it.
*/
switch (inst->opcode) {
case BRW_OPCODE_ADDC:
case BRW_OPCODE_SUBB:
case BRW_OPCODE_MACH:
inst->dst = fs_reg(retype(brw_null_reg(), inst->dst.type));
break;
default:
inst->remove();
break;
}
progress = true;
}
}
pc++;
}
if (progress)
invalidate_live_intervals();
return progress;
}
struct dead_code_hash_key
{
int vgrf;
int reg_offset;
};
static bool
dead_code_hash_compare(const void *a, const void *b)
{
return memcmp(a, b, sizeof(struct dead_code_hash_key)) == 0;
}
static void
clear_dead_code_hash(struct hash_table *ht)
{
struct hash_entry *entry;
hash_table_foreach(ht, entry) {
_mesa_hash_table_remove(ht, entry);
}
}
static void
insert_dead_code_hash(struct hash_table *ht,
int vgrf, int reg_offset, fs_inst *inst)
{
/* We don't bother freeing keys, because they'll be GCed with the ht. */
struct dead_code_hash_key *key = ralloc(ht, struct dead_code_hash_key);
key->vgrf = vgrf;
key->reg_offset = reg_offset;
_mesa_hash_table_insert(ht, _mesa_hash_data(key, sizeof(*key)), key, inst);
}
static struct hash_entry *
get_dead_code_hash_entry(struct hash_table *ht, int vgrf, int reg_offset)
{
struct dead_code_hash_key key;
key.vgrf = vgrf;
key.reg_offset = reg_offset;
return _mesa_hash_table_search(ht, _mesa_hash_data(&key, sizeof(key)), &key);
}
static void
remove_dead_code_hash(struct hash_table *ht,
int vgrf, int reg_offset)
{
struct hash_entry *entry = get_dead_code_hash_entry(ht, vgrf, reg_offset);
if (!entry)
return;
_mesa_hash_table_remove(ht, entry);
}
/**
* Walks basic blocks, removing any regs that are written but not read before
* being redefined.
*
* The dead_code_eliminate() function implements a global dead code
* elimination, but it only handles the removing the last write to a register
* if it's never read. This one can handle intermediate writes, but only
* within a basic block.
*/
bool
fs_visitor::dead_code_eliminate_local()
{
struct hash_table *ht;
bool progress = false;
ht = _mesa_hash_table_create(mem_ctx, dead_code_hash_compare);
foreach_list_safe(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
/* At a basic block, empty the HT since we don't understand dataflow
* here.
*/
if (inst->is_control_flow()) {
clear_dead_code_hash(ht);
continue;
}
/* Clear the HT of any instructions that got read. */
for (int i = 0; i < 3; i++) {
fs_reg src = inst->src[i];
if (src.file != GRF)
continue;
int read = 1;
if (inst->is_send_from_grf())
read = virtual_grf_sizes[src.reg] - src.reg_offset;
for (int reg_offset = src.reg_offset;
reg_offset < src.reg_offset + read;
reg_offset++) {
remove_dead_code_hash(ht, src.reg, reg_offset);
}
}
/* Add any update of a GRF to the HT, removing a previous write if it
* wasn't read.
*/
if (inst->dst.file == GRF) {
if (inst->regs_written > 1) {
/* We don't know how to trim channels from an instruction's
* writes, so we can't incrementally remove unread channels from
* it. Just remove whatever it overwrites from the table
*/
for (int i = 0; i < inst->regs_written; i++) {
remove_dead_code_hash(ht,
inst->dst.reg,
inst->dst.reg_offset + i);
}
} else {
struct hash_entry *entry =
get_dead_code_hash_entry(ht, inst->dst.reg,
inst->dst.reg_offset);
if (inst->is_partial_write()) {
/* For a partial write, we can't remove any previous dead code
* candidate, since we're just modifying their result, but we can
* be dead code eliminiated ourselves.
*/
if (entry) {
entry->data = inst;
} else {
insert_dead_code_hash(ht, inst->dst.reg, inst->dst.reg_offset,
inst);
}
} else {
if (entry) {
/* We're completely updating a channel, and there was a
* previous write to the channel that wasn't read. Kill it!
*/
fs_inst *inst = (fs_inst *)entry->data;
inst->remove();
progress = true;
_mesa_hash_table_remove(ht, entry);
}
insert_dead_code_hash(ht, inst->dst.reg, inst->dst.reg_offset,
inst);
}
}
}
}
_mesa_hash_table_destroy(ht, NULL);
if (progress)
invalidate_live_intervals();
return progress;
}
/**
* Implements a second type of register coalescing: This one checks if
* the two regs involved in a raw move don't interfere, in which case
* they can both by stored in the same place and the MOV removed.
*/
bool
fs_visitor::register_coalesce_2()
{
bool progress = false;
calculate_live_intervals();
foreach_list_safe(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
if (inst->opcode != BRW_OPCODE_MOV ||
inst->is_partial_write() ||
inst->saturate ||
inst->src[0].file != GRF ||
inst->src[0].negate ||
inst->src[0].abs ||
inst->src[0].smear != -1 ||
inst->dst.file != GRF ||
inst->dst.type != inst->src[0].type ||
virtual_grf_sizes[inst->src[0].reg] != 1 ||
virtual_grf_interferes(inst->dst.reg, inst->src[0].reg)) {
continue;
}
int reg_from = inst->src[0].reg;
assert(inst->src[0].reg_offset == 0);
int reg_to = inst->dst.reg;
int reg_to_offset = inst->dst.reg_offset;
foreach_list(node, &this->instructions) {
fs_inst *scan_inst = (fs_inst *)node;
if (scan_inst->dst.file == GRF &&
scan_inst->dst.reg == reg_from) {
scan_inst->dst.reg = reg_to;
scan_inst->dst.reg_offset = reg_to_offset;
}
for (int i = 0; i < 3; i++) {
if (scan_inst->src[i].file == GRF &&
scan_inst->src[i].reg == reg_from) {
scan_inst->src[i].reg = reg_to;
scan_inst->src[i].reg_offset = reg_to_offset;
}
}
}
inst->remove();
/* We don't need to recalculate live intervals inside the loop despite
* invalidating them; we only use them for the interferes test, and we
* must have had a situation where the intervals were:
*
* from to
* ^
* |
* v
* ^
* |
* v
*
* Some register R that might get coalesced with one of these two could
* only be referencing "to", otherwise "from"'s range would have been
* longer. R's range could also only start at the end of "to" or later,
* otherwise it will conflict with "to" when we try to coalesce "to"
* into Rw anyway.
*/
invalidate_live_intervals();
progress = true;
continue;
}
return progress;
}
bool
fs_visitor::register_coalesce()
{
bool progress = false;
int if_depth = 0;
int loop_depth = 0;
foreach_list_safe(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
/* Make sure that we dominate the instructions we're going to
* scan for interfering with our coalescing, or we won't have
* scanned enough to see if anything interferes with our
* coalescing. We don't dominate the following instructions if
* we're in a loop or an if block.
*/
switch (inst->opcode) {
case BRW_OPCODE_DO:
loop_depth++;
break;
case BRW_OPCODE_WHILE:
loop_depth--;
break;
case BRW_OPCODE_IF:
if_depth++;
break;
case BRW_OPCODE_ENDIF:
if_depth--;
break;
default:
break;
}
if (loop_depth || if_depth)
continue;
if (inst->opcode != BRW_OPCODE_MOV ||
inst->is_partial_write() ||
inst->saturate ||
inst->dst.file != GRF || (inst->src[0].file != GRF &&
inst->src[0].file != UNIFORM)||
inst->dst.type != inst->src[0].type)
continue;
bool has_source_modifiers = (inst->src[0].abs ||
inst->src[0].negate ||
inst->src[0].smear != -1 ||
inst->src[0].file == UNIFORM);
/* Found a move of a GRF to a GRF. Let's see if we can coalesce
* them: check for no writes to either one until the exit of the
* program.
*/
bool interfered = false;
for (fs_inst *scan_inst = (fs_inst *)inst->next;
!scan_inst->is_tail_sentinel();
scan_inst = (fs_inst *)scan_inst->next) {
if (scan_inst->dst.file == GRF) {
if (scan_inst->overwrites_reg(inst->dst) ||
scan_inst->overwrites_reg(inst->src[0])) {
interfered = true;
break;
}
}
if (has_source_modifiers) {
for (int i = 0; i < 3; i++) {
if (scan_inst->src[i].file == GRF &&
scan_inst->src[i].reg == inst->dst.reg &&
scan_inst->src[i].reg_offset == inst->dst.reg_offset &&
inst->dst.type != scan_inst->src[i].type)
{
interfered = true;
break;
}
}
}
/* The gen6 MATH instruction can't handle source modifiers or
* unusual register regions, so avoid coalescing those for
* now. We should do something more specific.
*/
if (has_source_modifiers && !can_do_source_mods(scan_inst)) {
interfered = true;
break;
}
/* The accumulator result appears to get used for the
* conditional modifier generation. When negating a UD
* value, there is a 33rd bit generated for the sign in the
* accumulator value, so now you can't check, for example,
* equality with a 32-bit value. See piglit fs-op-neg-uint.
*/
if (scan_inst->conditional_mod &&
inst->src[0].negate &&
inst->src[0].type == BRW_REGISTER_TYPE_UD) {
interfered = true;
break;
}
}
if (interfered) {
continue;
}
/* Rewrite the later usage to point at the source of the move to
* be removed.
*/
for (fs_inst *scan_inst = inst;
!scan_inst->is_tail_sentinel();
scan_inst = (fs_inst *)scan_inst->next) {
for (int i = 0; i < 3; i++) {
if (scan_inst->src[i].file == GRF &&
scan_inst->src[i].reg == inst->dst.reg &&
scan_inst->src[i].reg_offset == inst->dst.reg_offset) {
fs_reg new_src = inst->src[0];
if (scan_inst->src[i].abs) {
new_src.negate = 0;
new_src.abs = 1;
}
new_src.negate ^= scan_inst->src[i].negate;
new_src.sechalf = scan_inst->src[i].sechalf;
scan_inst->src[i] = new_src;
}
}
}
inst->remove();
progress = true;
}
if (progress)
invalidate_live_intervals();
return progress;
}
bool
fs_visitor::compute_to_mrf()
{
bool progress = false;
int next_ip = 0;
calculate_live_intervals();
foreach_list_safe(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
int ip = next_ip;
next_ip++;
if (inst->opcode != BRW_OPCODE_MOV ||
inst->is_partial_write() ||
inst->dst.file != MRF || inst->src[0].file != GRF ||
inst->dst.type != inst->src[0].type ||
inst->src[0].abs || inst->src[0].negate || inst->src[0].smear != -1)
continue;
/* Work out which hardware MRF registers are written by this
* instruction.
*/
int mrf_low = inst->dst.reg & ~BRW_MRF_COMPR4;
int mrf_high;
if (inst->dst.reg & BRW_MRF_COMPR4) {
mrf_high = mrf_low + 4;
} else if (dispatch_width == 16 &&
(!inst->force_uncompressed && !inst->force_sechalf)) {
mrf_high = mrf_low + 1;
} else {
mrf_high = mrf_low;
}
/* Can't compute-to-MRF this GRF if someone else was going to
* read it later.
*/
if (this->virtual_grf_end[inst->src[0].reg] > ip)
continue;
/* Found a move of a GRF to a MRF. Let's see if we can go
* rewrite the thing that made this GRF to write into the MRF.
*/
fs_inst *scan_inst;
for (scan_inst = (fs_inst *)inst->prev;
scan_inst->prev != NULL;
scan_inst = (fs_inst *)scan_inst->prev) {
if (scan_inst->dst.file == GRF &&
scan_inst->dst.reg == inst->src[0].reg) {
/* Found the last thing to write our reg we want to turn
* into a compute-to-MRF.
*/
/* If this one instruction didn't populate all the
* channels, bail. We might be able to rewrite everything
* that writes that reg, but it would require smarter
* tracking to delay the rewriting until complete success.
*/
if (scan_inst->is_partial_write())
break;
/* Things returning more than one register would need us to
* understand coalescing out more than one MOV at a time.
*/
if (scan_inst->regs_written > 1)
break;
/* SEND instructions can't have MRF as a destination. */
if (scan_inst->mlen)
break;
if (brw->gen == 6) {
/* gen6 math instructions must have the destination be
* GRF, so no compute-to-MRF for them.
*/
if (scan_inst->is_math()) {
break;
}
}
if (scan_inst->dst.reg_offset == inst->src[0].reg_offset) {
/* Found the creator of our MRF's source value. */
scan_inst->dst.file = MRF;
scan_inst->dst.reg = inst->dst.reg;
scan_inst->saturate |= inst->saturate;
inst->remove();
progress = true;
}
break;
}
/* We don't handle control flow here. Most computation of
* values that end up in MRFs are shortly before the MRF
* write anyway.
*/
if (scan_inst->is_control_flow() && scan_inst->opcode != BRW_OPCODE_IF)
break;
/* You can't read from an MRF, so if someone else reads our
* MRF's source GRF that we wanted to rewrite, that stops us.
*/
bool interfered = false;
for (int i = 0; i < 3; i++) {
if (scan_inst->src[i].file == GRF &&
scan_inst->src[i].reg == inst->src[0].reg &&
scan_inst->src[i].reg_offset == inst->src[0].reg_offset) {
interfered = true;
}
}
if (interfered)
break;
if (scan_inst->dst.file == MRF) {
/* If somebody else writes our MRF here, we can't
* compute-to-MRF before that.
*/
int scan_mrf_low = scan_inst->dst.reg & ~BRW_MRF_COMPR4;
int scan_mrf_high;
if (scan_inst->dst.reg & BRW_MRF_COMPR4) {
scan_mrf_high = scan_mrf_low + 4;
} else if (dispatch_width == 16 &&
(!scan_inst->force_uncompressed &&
!scan_inst->force_sechalf)) {
scan_mrf_high = scan_mrf_low + 1;
} else {
scan_mrf_high = scan_mrf_low;
}
if (mrf_low == scan_mrf_low ||
mrf_low == scan_mrf_high ||
mrf_high == scan_mrf_low ||
mrf_high == scan_mrf_high) {
break;
}
}
if (scan_inst->mlen > 0) {
/* Found a SEND instruction, which means that there are
* live values in MRFs from base_mrf to base_mrf +
* scan_inst->mlen - 1. Don't go pushing our MRF write up
* above it.
*/
if (mrf_low >= scan_inst->base_mrf &&
mrf_low < scan_inst->base_mrf + scan_inst->mlen) {
break;
}
if (mrf_high >= scan_inst->base_mrf &&
mrf_high < scan_inst->base_mrf + scan_inst->mlen) {
break;
}
}
}
}
if (progress)
invalidate_live_intervals();
return progress;
}
/**
* Walks through basic blocks, looking for repeated MRF writes and
* removing the later ones.
*/
bool
fs_visitor::remove_duplicate_mrf_writes()
{
fs_inst *last_mrf_move[16];
bool progress = false;
/* Need to update the MRF tracking for compressed instructions. */
if (dispatch_width == 16)
return false;
memset(last_mrf_move, 0, sizeof(last_mrf_move));
foreach_list_safe(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
if (inst->is_control_flow()) {
memset(last_mrf_move, 0, sizeof(last_mrf_move));
}
if (inst->opcode == BRW_OPCODE_MOV &&
inst->dst.file == MRF) {
fs_inst *prev_inst = last_mrf_move[inst->dst.reg];
if (prev_inst && inst->equals(prev_inst)) {
inst->remove();
progress = true;
continue;
}
}
/* Clear out the last-write records for MRFs that were overwritten. */
if (inst->dst.file == MRF) {
last_mrf_move[inst->dst.reg] = NULL;
}
if (inst->mlen > 0) {
/* Found a SEND instruction, which will include two or fewer
* implied MRF writes. We could do better here.
*/
for (int i = 0; i < implied_mrf_writes(inst); i++) {
last_mrf_move[inst->base_mrf + i] = NULL;
}
}
/* Clear out any MRF move records whose sources got overwritten. */
if (inst->dst.file == GRF) {
for (unsigned int i = 0; i < Elements(last_mrf_move); i++) {
if (last_mrf_move[i] &&
last_mrf_move[i]->src[0].reg == inst->dst.reg) {
last_mrf_move[i] = NULL;
}
}
}
if (inst->opcode == BRW_OPCODE_MOV &&
inst->dst.file == MRF &&
inst->src[0].file == GRF &&
!inst->is_partial_write()) {
last_mrf_move[inst->dst.reg] = inst;
}
}
if (progress)
invalidate_live_intervals();
return progress;
}
static void
clear_deps_for_inst_src(fs_inst *inst, int dispatch_width, bool *deps,
int first_grf, int grf_len)
{
bool inst_16wide = (dispatch_width > 8 &&
!inst->force_uncompressed &&
!inst->force_sechalf);
/* Clear the flag for registers that actually got read (as expected). */
for (int i = 0; i < 3; i++) {
int grf;
if (inst->src[i].file == GRF) {
grf = inst->src[i].reg;
} else if (inst->src[i].file == HW_REG &&
inst->src[i].fixed_hw_reg.file == BRW_GENERAL_REGISTER_FILE) {
grf = inst->src[i].fixed_hw_reg.nr;
} else {
continue;
}
if (grf >= first_grf &&
grf < first_grf + grf_len) {
deps[grf - first_grf] = false;
if (inst_16wide)
deps[grf - first_grf + 1] = false;
}
}
}
/**
* Implements this workaround for the original 965:
*
* "[DevBW, DevCL] Implementation Restrictions: As the hardware does not
* check for post destination dependencies on this instruction, software
* must ensure that there is no destination hazard for the case of ‘write
* followed by a posted write’ shown in the following example.
*
* 1. mov r3 0
* 2. send r3.xy <rest of send instruction>
* 3. mov r2 r3
*
* Due to no post-destination dependency check on the ‘send’, the above
* code sequence could have two instructions (1 and 2) in flight at the
* same time that both consider ‘r3’ as the target of their final writes.
*/
void
fs_visitor::insert_gen4_pre_send_dependency_workarounds(fs_inst *inst)
{
int reg_size = dispatch_width / 8;
int write_len = inst->regs_written * reg_size;
int first_write_grf = inst->dst.reg;
bool needs_dep[BRW_MAX_MRF];
assert(write_len < (int)sizeof(needs_dep) - 1);
memset(needs_dep, false, sizeof(needs_dep));
memset(needs_dep, true, write_len);
clear_deps_for_inst_src(inst, dispatch_width,
needs_dep, first_write_grf, write_len);
/* Walk backwards looking for writes to registers we're writing which
* aren't read since being written. If we hit the start of the program,
* we assume that there are no outstanding dependencies on entry to the
* program.
*/
for (fs_inst *scan_inst = (fs_inst *)inst->prev;
scan_inst != NULL;
scan_inst = (fs_inst *)scan_inst->prev) {
/* If we hit control flow, assume that there *are* outstanding
* dependencies, and force their cleanup before our instruction.
*/
if (scan_inst->is_control_flow()) {
for (int i = 0; i < write_len; i++) {
if (needs_dep[i]) {
inst->insert_before(DEP_RESOLVE_MOV(first_write_grf + i));
}
}
return;
}
bool scan_inst_16wide = (dispatch_width > 8 &&
!scan_inst->force_uncompressed &&
!scan_inst->force_sechalf);
/* We insert our reads as late as possible on the assumption that any
* instruction but a MOV that might have left us an outstanding
* dependency has more latency than a MOV.
*/
if (scan_inst->dst.file == GRF) {
for (int i = 0; i < scan_inst->regs_written; i++) {
int reg = scan_inst->dst.reg + i * reg_size;
if (reg >= first_write_grf &&
reg < first_write_grf + write_len &&
needs_dep[reg - first_write_grf]) {
inst->insert_before(DEP_RESOLVE_MOV(reg));
needs_dep[reg - first_write_grf] = false;
if (scan_inst_16wide)
needs_dep[reg - first_write_grf + 1] = false;
}
}
}
/* Clear the flag for registers that actually got read (as expected). */
clear_deps_for_inst_src(scan_inst, dispatch_width,
needs_dep, first_write_grf, write_len);
/* Continue the loop only if we haven't resolved all the dependencies */
int i;
for (i = 0; i < write_len; i++) {
if (needs_dep[i])
break;
}
if (i == write_len)
return;
}
}
/**
* Implements this workaround for the original 965:
*
* "[DevBW, DevCL] Errata: A destination register from a send can not be
* used as a destination register until after it has been sourced by an
* instruction with a different destination register.
*/
void
fs_visitor::insert_gen4_post_send_dependency_workarounds(fs_inst *inst)
{
int write_len = inst->regs_written * dispatch_width / 8;
int first_write_grf = inst->dst.reg;
bool needs_dep[BRW_MAX_MRF];
assert(write_len < (int)sizeof(needs_dep) - 1);
memset(needs_dep, false, sizeof(needs_dep));
memset(needs_dep, true, write_len);
/* Walk forwards looking for writes to registers we're writing which aren't
* read before being written.
*/
for (fs_inst *scan_inst = (fs_inst *)inst->next;
!scan_inst->is_tail_sentinel();
scan_inst = (fs_inst *)scan_inst->next) {
/* If we hit control flow, force resolve all remaining dependencies. */
if (scan_inst->is_control_flow()) {
for (int i = 0; i < write_len; i++) {
if (needs_dep[i])
scan_inst->insert_before(DEP_RESOLVE_MOV(first_write_grf + i));
}
return;
}
/* Clear the flag for registers that actually got read (as expected). */
clear_deps_for_inst_src(scan_inst, dispatch_width,
needs_dep, first_write_grf, write_len);
/* We insert our reads as late as possible since they're reading the
* result of a SEND, which has massive latency.
*/
if (scan_inst->dst.file == GRF &&
scan_inst->dst.reg >= first_write_grf &&
scan_inst->dst.reg < first_write_grf + write_len &&
needs_dep[scan_inst->dst.reg - first_write_grf]) {
scan_inst->insert_before(DEP_RESOLVE_MOV(scan_inst->dst.reg));
needs_dep[scan_inst->dst.reg - first_write_grf] = false;
}
/* Continue the loop only if we haven't resolved all the dependencies */
int i;
for (i = 0; i < write_len; i++) {
if (needs_dep[i])
break;
}
if (i == write_len)
return;
}
/* If we hit the end of the program, resolve all remaining dependencies out
* of paranoia.
*/
fs_inst *last_inst = (fs_inst *)this->instructions.get_tail();
assert(last_inst->eot);
for (int i = 0; i < write_len; i++) {
if (needs_dep[i])
last_inst->insert_before(DEP_RESOLVE_MOV(first_write_grf + i));
}
}
void
fs_visitor::insert_gen4_send_dependency_workarounds()
{
if (brw->gen != 4 || brw->is_g4x)
return;
/* Note that we're done with register allocation, so GRF fs_regs always
* have a .reg_offset of 0.
*/
foreach_list_safe(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
if (inst->mlen != 0 && inst->dst.file == GRF) {
insert_gen4_pre_send_dependency_workarounds(inst);
insert_gen4_post_send_dependency_workarounds(inst);
}
}
}
/**
* Turns the generic expression-style uniform pull constant load instruction
* into a hardware-specific series of instructions for loading a pull
* constant.
*
* The expression style allows the CSE pass before this to optimize out
* repeated loads from the same offset, and gives the pre-register-allocation
* scheduling full flexibility, while the conversion to native instructions
* allows the post-register-allocation scheduler the best information
* possible.
*
* Note that execution masking for setting up pull constant loads is special:
* the channels that need to be written are unrelated to the current execution
* mask, since a later instruction will use one of the result channels as a
* source operand for all 8 or 16 of its channels.
*/
void
fs_visitor::lower_uniform_pull_constant_loads()
{
foreach_list(node, &this->instructions) {
fs_inst *inst = (fs_inst *)node;
if (inst->opcode != FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD)
continue;
if (brw->gen >= 7) {
/* The offset arg before was a vec4-aligned byte offset. We need to
* turn it into a dword offset.
*/
fs_reg const_offset_reg = inst->src[1];
assert(const_offset_reg.file == IMM &&
const_offset_reg.type == BRW_REGISTER_TYPE_UD);
const_offset_reg.imm.u /= 4;
fs_reg payload = fs_reg(this, glsl_type::uint_type);
/* This is actually going to be a MOV, but since only the first dword
* is accessed, we have a special opcode to do just that one. Note
* that this needs to be an operation that will be considered a def
* by live variable analysis, or register allocation will explode.
*/
fs_inst *setup = new(mem_ctx) fs_inst(FS_OPCODE_SET_SIMD4X2_OFFSET,
payload, const_offset_reg);
setup->force_writemask_all = true;
setup->ir = inst->ir;
setup->annotation = inst->annotation;
inst->insert_before(setup);
/* Similarly, this will only populate the first 4 channels of the
* result register (since we only use smear values from 0-3), but we
* don't tell the optimizer.
*/
inst->opcode = FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD_GEN7;
inst->src[1] = payload;
invalidate_live_intervals();
} else {
/* Before register allocation, we didn't tell the scheduler about the
* MRF we use. We know it's safe to use this MRF because nothing
* else does except for register spill/unspill, which generates and
* uses its MRF within a single IR instruction.
*/
inst->base_mrf = 14;
inst->mlen = 1;
}
}
}
void
fs_visitor::dump_instruction(backend_instruction *be_inst)
{
fs_inst *inst = (fs_inst *)be_inst;
if (inst->predicate) {
printf("(%cf0.%d) ",
inst->predicate_inverse ? '-' : '+',
inst->flag_subreg);
}
printf("%s", brw_instruction_name(inst->opcode));
if (inst->saturate)
printf(".sat");
if (inst->conditional_mod) {
printf(".cmod");
if (!inst->predicate &&
(brw->gen < 5 || (inst->opcode != BRW_OPCODE_SEL &&
inst->opcode != BRW_OPCODE_IF &&
inst->opcode != BRW_OPCODE_WHILE))) {
printf(".f0.%d", inst->flag_subreg);
}
}
printf(" ");
switch (inst->dst.file) {
case GRF:
printf("vgrf%d", inst->dst.reg);
if (inst->dst.reg_offset)
printf("+%d", inst->dst.reg_offset);
break;
case MRF:
printf("m%d", inst->dst.reg);
break;
case BAD_FILE:
printf("(null)");
break;
case UNIFORM:
printf("***u%d***", inst->dst.reg);
break;
default:
printf("???");
break;
}
printf(", ");
for (int i = 0; i < 3; i++) {
if (inst->src[i].negate)
printf("-");
if (inst->src[i].abs)
printf("|");
switch (inst->src[i].file) {
case GRF:
printf("vgrf%d", inst->src[i].reg);
if (inst->src[i].reg_offset)
printf("+%d", inst->src[i].reg_offset);
break;
case MRF:
printf("***m%d***", inst->src[i].reg);
break;
case UNIFORM:
printf("u%d", inst->src[i].reg);
if (inst->src[i].reg_offset)
printf(".%d", inst->src[i].reg_offset);
break;
case BAD_FILE:
printf("(null)");
break;
case IMM:
switch (inst->src[i].type) {
case BRW_REGISTER_TYPE_F:
printf("%ff", inst->src[i].imm.f);
break;
case BRW_REGISTER_TYPE_D:
printf("%dd", inst->src[i].imm.i);
break;
case BRW_REGISTER_TYPE_UD:
printf("%uu", inst->src[i].imm.u);
break;
default:
printf("???");
break;
}
break;
default:
printf("???");
break;
}
if (inst->src[i].abs)
printf("|");
if (i < 3)
printf(", ");
}
printf(" ");
if (inst->force_uncompressed)
printf("1sthalf ");
if (inst->force_sechalf)
printf("2ndhalf ");
printf("\n");
}
/**
* Possibly returns an instruction that set up @param reg.
*
* Sometimes we want to take the result of some expression/variable
* dereference tree and rewrite the instruction generating the result
* of the tree. When processing the tree, we know that the
* instructions generated are all writing temporaries that are dead
* outside of this tree. So, if we have some instructions that write
* a temporary, we're free to point that temp write somewhere else.
*
* Note that this doesn't guarantee that the instruction generated
* only reg -- it might be the size=4 destination of a texture instruction.
*/
fs_inst *
fs_visitor::get_instruction_generating_reg(fs_inst *start,
fs_inst *end,
fs_reg reg)
{
if (end == start ||
end->is_partial_write() ||
reg.reladdr ||
!reg.equals(end->dst)) {
return NULL;
} else {
return end;
}
}
void
fs_visitor::setup_payload_gen6()
{
bool uses_depth =
(fp->Base.InputsRead & (1 << VARYING_SLOT_POS)) != 0;
unsigned barycentric_interp_modes = c->prog_data.barycentric_interp_modes;
assert(brw->gen >= 6);
/* R0-1: masks, pixel X/Y coordinates. */
c->nr_payload_regs = 2;
/* R2: only for 32-pixel dispatch.*/
/* R3-26: barycentric interpolation coordinates. These appear in the
* same order that they appear in the brw_wm_barycentric_interp_mode
* enum. Each set of coordinates occupies 2 registers if dispatch width
* == 8 and 4 registers if dispatch width == 16. Coordinates only
* appear if they were enabled using the "Barycentric Interpolation
* Mode" bits in WM_STATE.
*/
for (int i = 0; i < BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT; ++i) {
if (barycentric_interp_modes & (1 << i)) {
c->barycentric_coord_reg[i] = c->nr_payload_regs;
c->nr_payload_regs += 2;
if (dispatch_width == 16) {
c->nr_payload_regs += 2;
}
}
}
/* R27: interpolated depth if uses source depth */
if (uses_depth) {
c->source_depth_reg = c->nr_payload_regs;
c->nr_payload_regs++;
if (dispatch_width == 16) {
/* R28: interpolated depth if not 8-wide. */
c->nr_payload_regs++;
}
}
/* R29: interpolated W set if GEN6_WM_USES_SOURCE_W. */
if (uses_depth) {
c->source_w_reg = c->nr_payload_regs;
c->nr_payload_regs++;
if (dispatch_width == 16) {
/* R30: interpolated W if not 8-wide. */
c->nr_payload_regs++;
}
}
/* R31: MSAA position offsets. */
/* R32-: bary for 32-pixel. */
/* R58-59: interp W for 32-pixel. */
if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
c->source_depth_to_render_target = true;
}
}
bool
fs_visitor::run()
{
sanity_param_count = fp->Base.Parameters->NumParameters;
uint32_t orig_nr_params = c->prog_data.nr_params;
if (brw->gen >= 6)
setup_payload_gen6();
else
setup_payload_gen4();
if (0) {
emit_dummy_fs();
} else {
if (INTEL_DEBUG & DEBUG_SHADER_TIME)
emit_shader_time_begin();
calculate_urb_setup();
if (brw->gen < 6)
emit_interpolation_setup_gen4();
else
emit_interpolation_setup_gen6();
/* We handle discards by keeping track of the still-live pixels in f0.1.
* Initialize it with the dispatched pixels.
*/
if (fp->UsesKill) {
fs_inst *discard_init = emit(FS_OPCODE_MOV_DISPATCH_TO_FLAGS);
discard_init->flag_subreg = 1;
}
/* Generate FS IR for main(). (the visitor only descends into
* functions called "main").
*/
if (shader) {
foreach_list(node, &*shader->ir) {
ir_instruction *ir = (ir_instruction *)node;
base_ir = ir;
this->result = reg_undef;
ir->accept(this);
}
} else {
emit_fragment_program_code();
}
base_ir = NULL;
if (failed)
return false;
emit(FS_OPCODE_PLACEHOLDER_HALT);
emit_fb_writes();
split_virtual_grfs();
move_uniform_array_access_to_pull_constants();
setup_pull_constants();
bool progress;
do {
progress = false;
compact_virtual_grfs();
progress = remove_duplicate_mrf_writes() || progress;
progress = opt_algebraic() || progress;
progress = opt_cse() || progress;
progress = opt_copy_propagate() || progress;
progress = dead_code_eliminate() || progress;
progress = dead_code_eliminate_local() || progress;
progress = register_coalesce() || progress;
progress = register_coalesce_2() || progress;
progress = compute_to_mrf() || progress;
} while (progress);
remove_dead_constants();
schedule_instructions(false);
lower_uniform_pull_constant_loads();
assign_curb_setup();
assign_urb_setup();
if (0) {
/* Debug of register spilling: Go spill everything. */
for (int i = 0; i < virtual_grf_count; i++) {
spill_reg(i);
}
}
if (0)
assign_regs_trivial();
else {
while (!assign_regs()) {
if (failed)
break;
}
}
}
assert(force_uncompressed_stack == 0);
assert(force_sechalf_stack == 0);
/* This must come after all optimization and register allocation, since
* it inserts dead code that happens to have side effects, and it does
* so based on the actual physical registers in use.
*/
insert_gen4_send_dependency_workarounds();
if (failed)
return false;
schedule_instructions(true);
if (dispatch_width == 8) {
c->prog_data.reg_blocks = brw_register_blocks(grf_used);
} else {
c->prog_data.reg_blocks_16 = brw_register_blocks(grf_used);
/* Make sure we didn't try to sneak in an extra uniform */
assert(orig_nr_params == c->prog_data.nr_params);
(void) orig_nr_params;
}
/* If any state parameters were appended, then ParameterValues could have
* been realloced, in which case the driver uniform storage set up by
* _mesa_associate_uniform_storage() would point to freed memory. Make
* sure that didn't happen.
*/
assert(sanity_param_count == fp->Base.Parameters->NumParameters);
return !failed;
}
const unsigned *
brw_wm_fs_emit(struct brw_context *brw, struct brw_wm_compile *c,
struct gl_fragment_program *fp,
struct gl_shader_program *prog,
unsigned *final_assembly_size)
{
bool start_busy = false;
float start_time = 0;
if (unlikely(brw->perf_debug)) {
start_busy = (brw->batch.last_bo &&
drm_intel_bo_busy(brw->batch.last_bo));
start_time = get_time();
}
struct brw_shader *shader = NULL;
if (prog)
shader = (brw_shader *) prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
if (unlikely(INTEL_DEBUG & DEBUG_WM)) {
if (prog) {
printf("GLSL IR for native fragment shader %d:\n", prog->Name);
_mesa_print_ir(shader->ir, NULL);
printf("\n\n");
} else {
printf("ARB_fragment_program %d ir for native fragment shader\n",
fp->Base.Id);
_mesa_print_program(&fp->Base);
}
}
/* Now the main event: Visit the shader IR and generate our FS IR for it.
*/
fs_visitor v(brw, c, prog, fp, 8);
if (!v.run()) {
if (prog) {
prog->LinkStatus = false;
ralloc_strcat(&prog->InfoLog, v.fail_msg);
}
_mesa_problem(NULL, "Failed to compile fragment shader: %s\n",
v.fail_msg);
return NULL;
}
exec_list *simd16_instructions = NULL;
fs_visitor v2(brw, c, prog, fp, 16);
if (brw->gen >= 5 && likely(!(INTEL_DEBUG & DEBUG_NO16))) {
if (c->prog_data.nr_pull_params == 0) {
/* Try a 16-wide compile */
v2.import_uniforms(&v);
if (!v2.run()) {
perf_debug("16-wide shader failed to compile, falling back to "
"8-wide at a 10-20%% performance cost: %s", v2.fail_msg);
} else {
simd16_instructions = &v2.instructions;
}
} else {
perf_debug("Skipping 16-wide due to pull parameters.\n");
}
}
c->prog_data.dispatch_width = 8;
fs_generator g(brw, c, prog, fp, v.dual_src_output.file != BAD_FILE);
const unsigned *generated = g.generate_assembly(&v.instructions,
simd16_instructions,
final_assembly_size);
if (unlikely(brw->perf_debug) && shader) {
if (shader->compiled_once)
brw_wm_debug_recompile(brw, prog, &c->key);
shader->compiled_once = true;
if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
perf_debug("FS compile took %.03f ms and stalled the GPU\n",
(get_time() - start_time) * 1000);
}
}
return generated;
}
bool
brw_fs_precompile(struct gl_context *ctx, struct gl_shader_program *prog)
{
struct brw_context *brw = brw_context(ctx);
struct brw_wm_prog_key key;
if (!prog->_LinkedShaders[MESA_SHADER_FRAGMENT])
return true;
struct gl_fragment_program *fp = (struct gl_fragment_program *)
prog->_LinkedShaders[MESA_SHADER_FRAGMENT]->Program;
struct brw_fragment_program *bfp = brw_fragment_program(fp);
bool program_uses_dfdy = fp->UsesDFdy;
memset(&key, 0, sizeof(key));
if (brw->gen < 6) {
if (fp->UsesKill)
key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
/* Just assume depth testing. */
key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
}
if (brw->gen < 6 || _mesa_bitcount_64(fp->Base.InputsRead &
BRW_FS_VARYING_INPUT_MASK) > 16)
key.input_slots_valid = fp->Base.InputsRead | VARYING_BIT_POS;
key.clamp_fragment_color = ctx->API == API_OPENGL_COMPAT;
unsigned sampler_count = _mesa_fls(fp->Base.SamplersUsed);
for (unsigned i = 0; i < sampler_count; i++) {
if (fp->Base.ShadowSamplers & (1 << i)) {
/* Assume DEPTH_TEXTURE_MODE is the default: X, X, X, 1 */
key.tex.swizzles[i] =
MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_ONE);
} else {
/* Color sampler: assume no swizzling. */
key.tex.swizzles[i] = SWIZZLE_XYZW;
}
}
if (fp->Base.InputsRead & VARYING_BIT_POS) {
key.drawable_height = ctx->DrawBuffer->Height;
}
if ((fp->Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
key.render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
}
key.nr_color_regions = 1;
/* GL_FRAGMENT_SHADER_DERIVATIVE_HINT is almost always GL_DONT_CARE. The
* quality of the derivatives is likely to be determined by the driconf
* option.
*/
key.high_quality_derivatives = brw->disable_derivative_optimization;
key.program_string_id = bfp->id;
uint32_t old_prog_offset = brw->wm.base.prog_offset;
struct brw_wm_prog_data *old_prog_data = brw->wm.prog_data;
bool success = do_wm_prog(brw, prog, bfp, &key);
brw->wm.base.prog_offset = old_prog_offset;
brw->wm.prog_data = old_prog_data;
return success;
}
|