1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
|
/**************************************************************************
*
* Copyright 2009 VMware, Inc.
* Copyright 2007 VMware, Inc.
* All Rights Reserved.
*
* 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, sub license, 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 NON-INFRINGEMENT.
* IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS 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
* Code generate the whole fragment pipeline.
*
* The fragment pipeline consists of the following stages:
* - early depth test
* - fragment shader
* - alpha test
* - depth/stencil test
* - blending
*
* This file has only the glue to assemble the fragment pipeline. The actual
* plumbing of converting Gallium state into LLVM IR is done elsewhere, in the
* lp_bld_*.[ch] files, and in a complete generic and reusable way. Here we
* muster the LLVM JIT execution engine to create a function that follows an
* established binary interface and that can be called from C directly.
*
* A big source of complexity here is that we often want to run different
* stages with different precisions and data types and precisions. For example,
* the fragment shader needs typically to be done in floats, but the
* depth/stencil test and blending is better done in the type that most closely
* matches the depth/stencil and color buffer respectively.
*
* Since the width of a SIMD vector register stays the same regardless of the
* element type, different types imply different number of elements, so we must
* code generate more instances of the stages with larger types to be able to
* feed/consume the stages with smaller types.
*
* @author Jose Fonseca <jfonseca@vmware.com>
*/
#include <limits.h>
#include "pipe/p_defines.h"
#include "util/u_inlines.h"
#include "util/u_memory.h"
#include "util/u_pointer.h"
#include "util/format/u_format.h"
#include "util/u_dump.h"
#include "util/u_string.h"
#include "util/simple_list.h"
#include "util/u_dual_blend.h"
#include "util/os_time.h"
#include "pipe/p_shader_tokens.h"
#include "draw/draw_context.h"
#include "tgsi/tgsi_dump.h"
#include "tgsi/tgsi_scan.h"
#include "tgsi/tgsi_parse.h"
#include "gallivm/lp_bld_type.h"
#include "gallivm/lp_bld_const.h"
#include "gallivm/lp_bld_conv.h"
#include "gallivm/lp_bld_init.h"
#include "gallivm/lp_bld_intr.h"
#include "gallivm/lp_bld_logic.h"
#include "gallivm/lp_bld_tgsi.h"
#include "gallivm/lp_bld_swizzle.h"
#include "gallivm/lp_bld_flow.h"
#include "gallivm/lp_bld_debug.h"
#include "gallivm/lp_bld_arit.h"
#include "gallivm/lp_bld_bitarit.h"
#include "gallivm/lp_bld_pack.h"
#include "gallivm/lp_bld_format.h"
#include "gallivm/lp_bld_quad.h"
#include "lp_bld_alpha.h"
#include "lp_bld_blend.h"
#include "lp_bld_depth.h"
#include "lp_bld_interp.h"
#include "lp_context.h"
#include "lp_debug.h"
#include "lp_perf.h"
#include "lp_setup.h"
#include "lp_state.h"
#include "lp_tex_sample.h"
#include "lp_flush.h"
#include "lp_state_fs.h"
#include "lp_rast.h"
/** Fragment shader number (for debugging) */
static unsigned fs_no = 0;
/**
* Expand the relevant bits of mask_input to a n*4-dword mask for the
* n*four pixels in n 2x2 quads. This will set the n*four elements of the
* quad mask vector to 0 or ~0.
* Grouping is 01, 23 for 2 quad mode hence only 0 and 2 are valid
* quad arguments with fs length 8.
*
* \param first_quad which quad(s) of the quad group to test, in [0,3]
* \param mask_input bitwise mask for the whole 4x4 stamp
*/
static LLVMValueRef
generate_quad_mask(struct gallivm_state *gallivm,
struct lp_type fs_type,
unsigned first_quad,
LLVMValueRef mask_input) /* int32 */
{
LLVMBuilderRef builder = gallivm->builder;
struct lp_type mask_type;
LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context);
LLVMValueRef bits[16];
LLVMValueRef mask, bits_vec;
int shift, i;
/*
* XXX: We'll need a different path for 16 x u8
*/
assert(fs_type.width == 32);
assert(fs_type.length <= ARRAY_SIZE(bits));
mask_type = lp_int_type(fs_type);
/*
* mask_input >>= (quad * 4)
*/
switch (first_quad) {
case 0:
shift = 0;
break;
case 1:
assert(fs_type.length == 4);
shift = 2;
break;
case 2:
shift = 8;
break;
case 3:
assert(fs_type.length == 4);
shift = 10;
break;
default:
assert(0);
shift = 0;
}
mask_input = LLVMBuildLShr(builder,
mask_input,
LLVMConstInt(i32t, shift, 0),
"");
/*
* mask = { mask_input & (1 << i), for i in [0,3] }
*/
mask = lp_build_broadcast(gallivm,
lp_build_vec_type(gallivm, mask_type),
mask_input);
for (i = 0; i < fs_type.length / 4; i++) {
unsigned j = 2 * (i % 2) + (i / 2) * 8;
bits[4*i + 0] = LLVMConstInt(i32t, 1ULL << (j + 0), 0);
bits[4*i + 1] = LLVMConstInt(i32t, 1ULL << (j + 1), 0);
bits[4*i + 2] = LLVMConstInt(i32t, 1ULL << (j + 4), 0);
bits[4*i + 3] = LLVMConstInt(i32t, 1ULL << (j + 5), 0);
}
bits_vec = LLVMConstVector(bits, fs_type.length);
mask = LLVMBuildAnd(builder, mask, bits_vec, "");
/*
* mask = mask == bits ? ~0 : 0
*/
mask = lp_build_compare(gallivm,
mask_type, PIPE_FUNC_EQUAL,
mask, bits_vec);
return mask;
}
#define EARLY_DEPTH_TEST 0x1
#define LATE_DEPTH_TEST 0x2
#define EARLY_DEPTH_WRITE 0x4
#define LATE_DEPTH_WRITE 0x8
static int
find_output_by_semantic( const struct tgsi_shader_info *info,
unsigned semantic,
unsigned index )
{
int i;
for (i = 0; i < info->num_outputs; i++)
if (info->output_semantic_name[i] == semantic &&
info->output_semantic_index[i] == index)
return i;
return -1;
}
/**
* Fetch the specified lp_jit_viewport structure for a given viewport_index.
*/
static LLVMValueRef
lp_llvm_viewport(LLVMValueRef context_ptr,
struct gallivm_state *gallivm,
LLVMValueRef viewport_index)
{
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef ptr;
LLVMValueRef res;
struct lp_type viewport_type =
lp_type_float_vec(32, 32 * LP_JIT_VIEWPORT_NUM_FIELDS);
ptr = lp_jit_context_viewports(gallivm, context_ptr);
ptr = LLVMBuildPointerCast(builder, ptr,
LLVMPointerType(lp_build_vec_type(gallivm, viewport_type), 0), "");
res = lp_build_pointer_get(builder, ptr, viewport_index);
return res;
}
static LLVMValueRef
lp_build_depth_clamp(struct gallivm_state *gallivm,
LLVMBuilderRef builder,
struct lp_type type,
LLVMValueRef context_ptr,
LLVMValueRef thread_data_ptr,
LLVMValueRef z)
{
LLVMValueRef viewport, min_depth, max_depth;
LLVMValueRef viewport_index;
struct lp_build_context f32_bld;
assert(type.floating);
lp_build_context_init(&f32_bld, gallivm, type);
/*
* Assumes clamping of the viewport index will occur in setup/gs. Value
* is passed through the rasterization stage via lp_rast_shader_inputs.
*
* See: draw_clamp_viewport_idx and lp_clamp_viewport_idx for clamping
* semantics.
*/
viewport_index = lp_jit_thread_data_raster_state_viewport_index(gallivm,
thread_data_ptr);
/*
* Load the min and max depth from the lp_jit_context.viewports
* array of lp_jit_viewport structures.
*/
viewport = lp_llvm_viewport(context_ptr, gallivm, viewport_index);
/* viewports[viewport_index].min_depth */
min_depth = LLVMBuildExtractElement(builder, viewport,
lp_build_const_int32(gallivm, LP_JIT_VIEWPORT_MIN_DEPTH), "");
min_depth = lp_build_broadcast_scalar(&f32_bld, min_depth);
/* viewports[viewport_index].max_depth */
max_depth = LLVMBuildExtractElement(builder, viewport,
lp_build_const_int32(gallivm, LP_JIT_VIEWPORT_MAX_DEPTH), "");
max_depth = lp_build_broadcast_scalar(&f32_bld, max_depth);
/*
* Clamp to the min and max depth values for the given viewport.
*/
return lp_build_clamp(&f32_bld, z, min_depth, max_depth);
}
/**
* Generate the fragment shader, depth/stencil test, and alpha tests.
*/
static void
generate_fs_loop(struct gallivm_state *gallivm,
struct lp_fragment_shader *shader,
const struct lp_fragment_shader_variant_key *key,
LLVMBuilderRef builder,
struct lp_type type,
LLVMValueRef context_ptr,
LLVMValueRef num_loop,
struct lp_build_interp_soa_context *interp,
const struct lp_build_sampler_soa *sampler,
const struct lp_build_image_soa *image,
LLVMValueRef mask_store,
LLVMValueRef (*out_color)[4],
LLVMValueRef depth_ptr,
LLVMValueRef depth_stride,
LLVMValueRef facing,
LLVMValueRef thread_data_ptr)
{
const struct util_format_description *zs_format_desc = NULL;
const struct tgsi_token *tokens = shader->base.tokens;
struct lp_type int_type = lp_int_type(type);
LLVMTypeRef vec_type, int_vec_type;
LLVMValueRef mask_ptr, mask_val;
LLVMValueRef consts_ptr, num_consts_ptr;
LLVMValueRef ssbo_ptr, num_ssbo_ptr;
LLVMValueRef z;
LLVMValueRef z_value, s_value;
LLVMValueRef z_fb, s_fb;
LLVMValueRef stencil_refs[2];
LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][TGSI_NUM_CHANNELS];
struct lp_build_for_loop_state loop_state;
struct lp_build_mask_context mask;
/*
* TODO: figure out if simple_shader optimization is really worthwile to
* keep. Disabled because it may hide some real bugs in the (depth/stencil)
* code since tests tend to take another codepath than real shaders.
*/
boolean simple_shader = (shader->info.base.file_count[TGSI_FILE_SAMPLER] == 0 &&
shader->info.base.num_inputs < 3 &&
shader->info.base.num_instructions < 8) && 0;
const boolean dual_source_blend = key->blend.rt[0].blend_enable &&
util_blend_state_is_dual(&key->blend, 0);
unsigned attrib;
unsigned chan;
unsigned cbuf;
unsigned depth_mode;
struct lp_bld_tgsi_system_values system_values;
memset(&system_values, 0, sizeof(system_values));
if (key->depth.enabled ||
key->stencil[0].enabled) {
zs_format_desc = util_format_description(key->zsbuf_format);
assert(zs_format_desc);
if (shader->info.base.properties[TGSI_PROPERTY_FS_EARLY_DEPTH_STENCIL])
depth_mode = EARLY_DEPTH_TEST | EARLY_DEPTH_WRITE;
else if (!shader->info.base.writes_z && !shader->info.base.writes_stencil) {
if (shader->info.base.writes_memory)
depth_mode = LATE_DEPTH_TEST | LATE_DEPTH_WRITE;
else if (key->alpha.enabled ||
key->blend.alpha_to_coverage ||
shader->info.base.uses_kill ||
shader->info.base.writes_samplemask) {
/* With alpha test and kill, can do the depth test early
* and hopefully eliminate some quads. But need to do a
* special deferred depth write once the final mask value
* is known. This only works though if there's either no
* stencil test or the stencil value isn't written.
*/
if (key->stencil[0].enabled && (key->stencil[0].writemask ||
(key->stencil[1].enabled &&
key->stencil[1].writemask)))
depth_mode = LATE_DEPTH_TEST | LATE_DEPTH_WRITE;
else
depth_mode = EARLY_DEPTH_TEST | LATE_DEPTH_WRITE;
}
else
depth_mode = EARLY_DEPTH_TEST | EARLY_DEPTH_WRITE;
}
else {
depth_mode = LATE_DEPTH_TEST | LATE_DEPTH_WRITE;
}
if (!(key->depth.enabled && key->depth.writemask) &&
!(key->stencil[0].enabled && (key->stencil[0].writemask ||
(key->stencil[1].enabled &&
key->stencil[1].writemask))))
depth_mode &= ~(LATE_DEPTH_WRITE | EARLY_DEPTH_WRITE);
}
else {
depth_mode = 0;
}
vec_type = lp_build_vec_type(gallivm, type);
int_vec_type = lp_build_vec_type(gallivm, int_type);
stencil_refs[0] = lp_jit_context_stencil_ref_front_value(gallivm, context_ptr);
stencil_refs[1] = lp_jit_context_stencil_ref_back_value(gallivm, context_ptr);
/* convert scalar stencil refs into vectors */
stencil_refs[0] = lp_build_broadcast(gallivm, int_vec_type, stencil_refs[0]);
stencil_refs[1] = lp_build_broadcast(gallivm, int_vec_type, stencil_refs[1]);
consts_ptr = lp_jit_context_constants(gallivm, context_ptr);
num_consts_ptr = lp_jit_context_num_constants(gallivm, context_ptr);
ssbo_ptr = lp_jit_context_ssbos(gallivm, context_ptr);
num_ssbo_ptr = lp_jit_context_num_ssbos(gallivm, context_ptr);
lp_build_for_loop_begin(&loop_state, gallivm,
lp_build_const_int32(gallivm, 0),
LLVMIntULT,
num_loop,
lp_build_const_int32(gallivm, 1));
mask_ptr = LLVMBuildGEP(builder, mask_store,
&loop_state.counter, 1, "mask_ptr");
mask_val = LLVMBuildLoad(builder, mask_ptr, "");
memset(outputs, 0, sizeof outputs);
for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++) {
for(chan = 0; chan < TGSI_NUM_CHANNELS; ++chan) {
out_color[cbuf][chan] = lp_build_array_alloca(gallivm,
lp_build_vec_type(gallivm,
type),
num_loop, "color");
}
}
if (dual_source_blend) {
assert(key->nr_cbufs <= 1);
for(chan = 0; chan < TGSI_NUM_CHANNELS; ++chan) {
out_color[1][chan] = lp_build_array_alloca(gallivm,
lp_build_vec_type(gallivm,
type),
num_loop, "color1");
}
}
/* 'mask' will control execution based on quad's pixel alive/killed state */
lp_build_mask_begin(&mask, gallivm, type, mask_val);
if (!(depth_mode & EARLY_DEPTH_TEST) && !simple_shader)
lp_build_mask_check(&mask);
lp_build_interp_soa_update_pos_dyn(interp, gallivm, loop_state.counter);
z = interp->pos[2];
if (depth_mode & EARLY_DEPTH_TEST) {
/*
* Clamp according to ARB_depth_clamp semantics.
*/
if (key->depth_clamp) {
z = lp_build_depth_clamp(gallivm, builder, type, context_ptr,
thread_data_ptr, z);
}
lp_build_depth_stencil_load_swizzled(gallivm, type,
zs_format_desc, key->resource_1d,
depth_ptr, depth_stride,
&z_fb, &s_fb, loop_state.counter);
lp_build_depth_stencil_test(gallivm,
&key->depth,
key->stencil,
type,
zs_format_desc,
&mask,
stencil_refs,
z, z_fb, s_fb,
facing,
&z_value, &s_value,
!simple_shader);
if (depth_mode & EARLY_DEPTH_WRITE) {
lp_build_depth_stencil_write_swizzled(gallivm, type,
zs_format_desc, key->resource_1d,
NULL, NULL, NULL, loop_state.counter,
depth_ptr, depth_stride,
z_value, s_value);
}
/*
* Note mask check if stencil is enabled must be after ds write not after
* stencil test otherwise new stencil values may not get written if all
* fragments got killed by depth/stencil test.
*/
if (!simple_shader && key->stencil[0].enabled)
lp_build_mask_check(&mask);
}
lp_build_interp_soa_update_inputs_dyn(interp, gallivm, loop_state.counter);
struct lp_build_tgsi_params params;
memset(¶ms, 0, sizeof(params));
params.type = type;
params.mask = &mask;
params.consts_ptr = consts_ptr;
params.const_sizes_ptr = num_consts_ptr;
params.system_values = &system_values;
params.inputs = interp->inputs;
params.context_ptr = context_ptr;
params.thread_data_ptr = thread_data_ptr;
params.sampler = sampler;
params.info = &shader->info.base;
params.ssbo_ptr = ssbo_ptr;
params.ssbo_sizes_ptr = num_ssbo_ptr;
params.image = image;
/* Build the actual shader */
lp_build_tgsi_soa(gallivm, tokens, ¶ms,
outputs);
/* Alpha test */
if (key->alpha.enabled) {
int color0 = find_output_by_semantic(&shader->info.base,
TGSI_SEMANTIC_COLOR,
0);
if (color0 != -1 && outputs[color0][3]) {
const struct util_format_description *cbuf_format_desc;
LLVMValueRef alpha = LLVMBuildLoad(builder, outputs[color0][3], "alpha");
LLVMValueRef alpha_ref_value;
alpha_ref_value = lp_jit_context_alpha_ref_value(gallivm, context_ptr);
alpha_ref_value = lp_build_broadcast(gallivm, vec_type, alpha_ref_value);
cbuf_format_desc = util_format_description(key->cbuf_format[0]);
lp_build_alpha_test(gallivm, key->alpha.func, type, cbuf_format_desc,
&mask, alpha, alpha_ref_value,
(depth_mode & LATE_DEPTH_TEST) != 0);
}
}
/* Emulate Alpha to Coverage with Alpha test */
if (key->blend.alpha_to_coverage) {
int color0 = find_output_by_semantic(&shader->info.base,
TGSI_SEMANTIC_COLOR,
0);
if (color0 != -1 && outputs[color0][3]) {
LLVMValueRef alpha = LLVMBuildLoad(builder, outputs[color0][3], "alpha");
lp_build_alpha_to_coverage(gallivm, type,
&mask, alpha,
(depth_mode & LATE_DEPTH_TEST) != 0);
}
}
if (shader->info.base.writes_samplemask) {
int smaski = find_output_by_semantic(&shader->info.base,
TGSI_SEMANTIC_SAMPLEMASK,
0);
LLVMValueRef smask;
struct lp_build_context smask_bld;
lp_build_context_init(&smask_bld, gallivm, int_type);
assert(smaski >= 0);
smask = LLVMBuildLoad(builder, outputs[smaski][0], "smask");
/*
* Pixel is alive according to the first sample in the mask.
*/
smask = LLVMBuildBitCast(builder, smask, smask_bld.vec_type, "");
smask = lp_build_and(&smask_bld, smask, smask_bld.one);
smask = lp_build_cmp(&smask_bld, PIPE_FUNC_NOTEQUAL, smask, smask_bld.zero);
lp_build_mask_update(&mask, smask);
}
/* Late Z test */
if (depth_mode & LATE_DEPTH_TEST) {
int pos0 = find_output_by_semantic(&shader->info.base,
TGSI_SEMANTIC_POSITION,
0);
int s_out = find_output_by_semantic(&shader->info.base,
TGSI_SEMANTIC_STENCIL,
0);
if (pos0 != -1 && outputs[pos0][2]) {
z = LLVMBuildLoad(builder, outputs[pos0][2], "output.z");
}
/*
* Clamp according to ARB_depth_clamp semantics.
*/
if (key->depth_clamp) {
z = lp_build_depth_clamp(gallivm, builder, type, context_ptr,
thread_data_ptr, z);
}
if (s_out != -1 && outputs[s_out][1]) {
/* there's only one value, and spec says to discard additional bits */
LLVMValueRef s_max_mask = lp_build_const_int_vec(gallivm, int_type, 255);
stencil_refs[0] = LLVMBuildLoad(builder, outputs[s_out][1], "output.s");
stencil_refs[0] = LLVMBuildBitCast(builder, stencil_refs[0], int_vec_type, "");
stencil_refs[0] = LLVMBuildAnd(builder, stencil_refs[0], s_max_mask, "");
stencil_refs[1] = stencil_refs[0];
}
lp_build_depth_stencil_load_swizzled(gallivm, type,
zs_format_desc, key->resource_1d,
depth_ptr, depth_stride,
&z_fb, &s_fb, loop_state.counter);
lp_build_depth_stencil_test(gallivm,
&key->depth,
key->stencil,
type,
zs_format_desc,
&mask,
stencil_refs,
z, z_fb, s_fb,
facing,
&z_value, &s_value,
!simple_shader);
/* Late Z write */
if (depth_mode & LATE_DEPTH_WRITE) {
lp_build_depth_stencil_write_swizzled(gallivm, type,
zs_format_desc, key->resource_1d,
NULL, NULL, NULL, loop_state.counter,
depth_ptr, depth_stride,
z_value, s_value);
}
}
else if ((depth_mode & EARLY_DEPTH_TEST) &&
(depth_mode & LATE_DEPTH_WRITE))
{
/* Need to apply a reduced mask to the depth write. Reload the
* depth value, update from zs_value with the new mask value and
* write that out.
*/
lp_build_depth_stencil_write_swizzled(gallivm, type,
zs_format_desc, key->resource_1d,
&mask, z_fb, s_fb, loop_state.counter,
depth_ptr, depth_stride,
z_value, s_value);
}
/* Color write */
for (attrib = 0; attrib < shader->info.base.num_outputs; ++attrib)
{
unsigned cbuf = shader->info.base.output_semantic_index[attrib];
if ((shader->info.base.output_semantic_name[attrib] == TGSI_SEMANTIC_COLOR) &&
((cbuf < key->nr_cbufs) || (cbuf == 1 && dual_source_blend)))
{
for(chan = 0; chan < TGSI_NUM_CHANNELS; ++chan) {
if(outputs[attrib][chan]) {
/* XXX: just initialize outputs to point at colors[] and
* skip this.
*/
LLVMValueRef out = LLVMBuildLoad(builder, outputs[attrib][chan], "");
LLVMValueRef color_ptr;
color_ptr = LLVMBuildGEP(builder, out_color[cbuf][chan],
&loop_state.counter, 1, "");
lp_build_name(out, "color%u.%c", attrib, "rgba"[chan]);
LLVMBuildStore(builder, out, color_ptr);
}
}
}
}
if (key->occlusion_count) {
LLVMValueRef counter = lp_jit_thread_data_counter(gallivm, thread_data_ptr);
lp_build_name(counter, "counter");
lp_build_occlusion_count(gallivm, type,
lp_build_mask_value(&mask), counter);
}
mask_val = lp_build_mask_end(&mask);
LLVMBuildStore(builder, mask_val, mask_ptr);
lp_build_for_loop_end(&loop_state);
}
/**
* This function will reorder pixels from the fragment shader SoA to memory layout AoS
*
* Fragment Shader outputs pixels in small 2x2 blocks
* e.g. (0, 0), (1, 0), (0, 1), (1, 1) ; (2, 0) ...
*
* However in memory pixels are stored in rows
* e.g. (0, 0), (1, 0), (2, 0), (3, 0) ; (0, 1) ...
*
* @param type fragment shader type (4x or 8x float)
* @param num_fs number of fs_src
* @param is_1d whether we're outputting to a 1d resource
* @param dst_channels number of output channels
* @param fs_src output from fragment shader
* @param dst pointer to store result
* @param pad_inline is channel padding inline or at end of row
* @return the number of dsts
*/
static int
generate_fs_twiddle(struct gallivm_state *gallivm,
struct lp_type type,
unsigned num_fs,
unsigned dst_channels,
LLVMValueRef fs_src[][4],
LLVMValueRef* dst,
bool pad_inline)
{
LLVMValueRef src[16];
bool swizzle_pad;
bool twiddle;
bool split;
unsigned pixels = type.length / 4;
unsigned reorder_group;
unsigned src_channels;
unsigned src_count;
unsigned i;
src_channels = dst_channels < 3 ? dst_channels : 4;
src_count = num_fs * src_channels;
assert(pixels == 2 || pixels == 1);
assert(num_fs * src_channels <= ARRAY_SIZE(src));
/*
* Transpose from SoA -> AoS
*/
for (i = 0; i < num_fs; ++i) {
lp_build_transpose_aos_n(gallivm, type, &fs_src[i][0], src_channels, &src[i * src_channels]);
}
/*
* Pick transformation options
*/
swizzle_pad = false;
twiddle = false;
split = false;
reorder_group = 0;
if (dst_channels == 1) {
twiddle = true;
if (pixels == 2) {
split = true;
}
} else if (dst_channels == 2) {
if (pixels == 1) {
reorder_group = 1;
}
} else if (dst_channels > 2) {
if (pixels == 1) {
reorder_group = 2;
} else {
twiddle = true;
}
if (!pad_inline && dst_channels == 3 && pixels > 1) {
swizzle_pad = true;
}
}
/*
* Split the src in half
*/
if (split) {
for (i = num_fs; i > 0; --i) {
src[(i - 1)*2 + 1] = lp_build_extract_range(gallivm, src[i - 1], 4, 4);
src[(i - 1)*2 + 0] = lp_build_extract_range(gallivm, src[i - 1], 0, 4);
}
src_count *= 2;
type.length = 4;
}
/*
* Ensure pixels are in memory order
*/
if (reorder_group) {
/* Twiddle pixels by reordering the array, e.g.:
*
* src_count = 8 -> 0 2 1 3 4 6 5 7
* src_count = 16 -> 0 1 4 5 2 3 6 7 8 9 12 13 10 11 14 15
*/
const unsigned reorder_sw[] = { 0, 2, 1, 3 };
for (i = 0; i < src_count; ++i) {
unsigned group = i / reorder_group;
unsigned block = (group / 4) * 4 * reorder_group;
unsigned j = block + (reorder_sw[group % 4] * reorder_group) + (i % reorder_group);
dst[i] = src[j];
}
} else if (twiddle) {
/* Twiddle pixels across elements of array */
/*
* XXX: we should avoid this in some cases, but would need to tell
* lp_build_conv to reorder (or deal with it ourselves).
*/
lp_bld_quad_twiddle(gallivm, type, src, src_count, dst);
} else {
/* Do nothing */
memcpy(dst, src, sizeof(LLVMValueRef) * src_count);
}
/*
* Moves any padding between pixels to the end
* e.g. RGBXRGBX -> RGBRGBXX
*/
if (swizzle_pad) {
unsigned char swizzles[16];
unsigned elems = pixels * dst_channels;
for (i = 0; i < type.length; ++i) {
if (i < elems)
swizzles[i] = i % dst_channels + (i / dst_channels) * 4;
else
swizzles[i] = LP_BLD_SWIZZLE_DONTCARE;
}
for (i = 0; i < src_count; ++i) {
dst[i] = lp_build_swizzle_aos_n(gallivm, dst[i], swizzles, type.length, type.length);
}
}
return src_count;
}
/*
* Untwiddle and transpose, much like the above.
* However, this is after conversion, so we get packed vectors.
* At this time only handle 4x16i8 rgba / 2x16i8 rg / 1x16i8 r data,
* the vectors will look like:
* r0r1r4r5r2r3r6r7r8r9r12... (albeit color channels may
* be swizzled here). Extending to 16bit should be trivial.
* Should also be extended to handle twice wide vectors with AVX2...
*/
static void
fs_twiddle_transpose(struct gallivm_state *gallivm,
struct lp_type type,
LLVMValueRef *src,
unsigned src_count,
LLVMValueRef *dst)
{
unsigned i, j;
struct lp_type type64, type16, type32;
LLVMTypeRef type64_t, type8_t, type16_t, type32_t;
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef tmp[4], shuf[8];
for (j = 0; j < 2; j++) {
shuf[j*4 + 0] = lp_build_const_int32(gallivm, j*4 + 0);
shuf[j*4 + 1] = lp_build_const_int32(gallivm, j*4 + 2);
shuf[j*4 + 2] = lp_build_const_int32(gallivm, j*4 + 1);
shuf[j*4 + 3] = lp_build_const_int32(gallivm, j*4 + 3);
}
assert(src_count == 4 || src_count == 2 || src_count == 1);
assert(type.width == 8);
assert(type.length == 16);
type8_t = lp_build_vec_type(gallivm, type);
type64 = type;
type64.length /= 8;
type64.width *= 8;
type64_t = lp_build_vec_type(gallivm, type64);
type16 = type;
type16.length /= 2;
type16.width *= 2;
type16_t = lp_build_vec_type(gallivm, type16);
type32 = type;
type32.length /= 4;
type32.width *= 4;
type32_t = lp_build_vec_type(gallivm, type32);
lp_build_transpose_aos_n(gallivm, type, src, src_count, tmp);
if (src_count == 1) {
/* transpose was no-op, just untwiddle */
LLVMValueRef shuf_vec;
shuf_vec = LLVMConstVector(shuf, 8);
tmp[0] = LLVMBuildBitCast(builder, src[0], type16_t, "");
tmp[0] = LLVMBuildShuffleVector(builder, tmp[0], tmp[0], shuf_vec, "");
dst[0] = LLVMBuildBitCast(builder, tmp[0], type8_t, "");
} else if (src_count == 2) {
LLVMValueRef shuf_vec;
shuf_vec = LLVMConstVector(shuf, 4);
for (i = 0; i < 2; i++) {
tmp[i] = LLVMBuildBitCast(builder, tmp[i], type32_t, "");
tmp[i] = LLVMBuildShuffleVector(builder, tmp[i], tmp[i], shuf_vec, "");
dst[i] = LLVMBuildBitCast(builder, tmp[i], type8_t, "");
}
} else {
for (j = 0; j < 2; j++) {
LLVMValueRef lo, hi, lo2, hi2;
/*
* Note that if we only really have 3 valid channels (rgb)
* and we don't need alpha we could substitute a undef here
* for the respective channel (causing llvm to drop conversion
* for alpha).
*/
/* we now have rgba0rgba1rgba4rgba5 etc, untwiddle */
lo2 = LLVMBuildBitCast(builder, tmp[j*2], type64_t, "");
hi2 = LLVMBuildBitCast(builder, tmp[j*2 + 1], type64_t, "");
lo = lp_build_interleave2(gallivm, type64, lo2, hi2, 0);
hi = lp_build_interleave2(gallivm, type64, lo2, hi2, 1);
dst[j*2] = LLVMBuildBitCast(builder, lo, type8_t, "");
dst[j*2 + 1] = LLVMBuildBitCast(builder, hi, type8_t, "");
}
}
}
/**
* Load an unswizzled block of pixels from memory
*/
static void
load_unswizzled_block(struct gallivm_state *gallivm,
LLVMValueRef base_ptr,
LLVMValueRef stride,
unsigned block_width,
unsigned block_height,
LLVMValueRef* dst,
struct lp_type dst_type,
unsigned dst_count,
unsigned dst_alignment)
{
LLVMBuilderRef builder = gallivm->builder;
unsigned row_size = dst_count / block_height;
unsigned i;
/* Ensure block exactly fits into dst */
assert((block_width * block_height) % dst_count == 0);
for (i = 0; i < dst_count; ++i) {
unsigned x = i % row_size;
unsigned y = i / row_size;
LLVMValueRef bx = lp_build_const_int32(gallivm, x * (dst_type.width / 8) * dst_type.length);
LLVMValueRef by = LLVMBuildMul(builder, lp_build_const_int32(gallivm, y), stride, "");
LLVMValueRef gep[2];
LLVMValueRef dst_ptr;
gep[0] = lp_build_const_int32(gallivm, 0);
gep[1] = LLVMBuildAdd(builder, bx, by, "");
dst_ptr = LLVMBuildGEP(builder, base_ptr, gep, 2, "");
dst_ptr = LLVMBuildBitCast(builder, dst_ptr,
LLVMPointerType(lp_build_vec_type(gallivm, dst_type), 0), "");
dst[i] = LLVMBuildLoad(builder, dst_ptr, "");
LLVMSetAlignment(dst[i], dst_alignment);
}
}
/**
* Store an unswizzled block of pixels to memory
*/
static void
store_unswizzled_block(struct gallivm_state *gallivm,
LLVMValueRef base_ptr,
LLVMValueRef stride,
unsigned block_width,
unsigned block_height,
LLVMValueRef* src,
struct lp_type src_type,
unsigned src_count,
unsigned src_alignment)
{
LLVMBuilderRef builder = gallivm->builder;
unsigned row_size = src_count / block_height;
unsigned i;
/* Ensure src exactly fits into block */
assert((block_width * block_height) % src_count == 0);
for (i = 0; i < src_count; ++i) {
unsigned x = i % row_size;
unsigned y = i / row_size;
LLVMValueRef bx = lp_build_const_int32(gallivm, x * (src_type.width / 8) * src_type.length);
LLVMValueRef by = LLVMBuildMul(builder, lp_build_const_int32(gallivm, y), stride, "");
LLVMValueRef gep[2];
LLVMValueRef src_ptr;
gep[0] = lp_build_const_int32(gallivm, 0);
gep[1] = LLVMBuildAdd(builder, bx, by, "");
src_ptr = LLVMBuildGEP(builder, base_ptr, gep, 2, "");
src_ptr = LLVMBuildBitCast(builder, src_ptr,
LLVMPointerType(lp_build_vec_type(gallivm, src_type), 0), "");
src_ptr = LLVMBuildStore(builder, src[i], src_ptr);
LLVMSetAlignment(src_ptr, src_alignment);
}
}
/**
* Checks if a format description is an arithmetic format
*
* A format which has irregular channel sizes such as R3_G3_B2 or R5_G6_B5.
*/
static inline boolean
is_arithmetic_format(const struct util_format_description *format_desc)
{
boolean arith = false;
unsigned i;
for (i = 0; i < format_desc->nr_channels; ++i) {
arith |= format_desc->channel[i].size != format_desc->channel[0].size;
arith |= (format_desc->channel[i].size % 8) != 0;
}
return arith;
}
/**
* Checks if this format requires special handling due to required expansion
* to floats for blending, and furthermore has "natural" packed AoS -> unpacked
* SoA conversion.
*/
static inline boolean
format_expands_to_float_soa(const struct util_format_description *format_desc)
{
if (format_desc->format == PIPE_FORMAT_R11G11B10_FLOAT ||
format_desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB) {
return true;
}
return false;
}
/**
* Retrieves the type representing the memory layout for a format
*
* e.g. RGBA16F = 4x half-float and R3G3B2 = 1x byte
*/
static inline void
lp_mem_type_from_format_desc(const struct util_format_description *format_desc,
struct lp_type* type)
{
unsigned i;
unsigned chan;
if (format_expands_to_float_soa(format_desc)) {
/* just make this a uint with width of block */
type->floating = false;
type->fixed = false;
type->sign = false;
type->norm = false;
type->width = format_desc->block.bits;
type->length = 1;
return;
}
for (i = 0; i < 4; i++)
if (format_desc->channel[i].type != UTIL_FORMAT_TYPE_VOID)
break;
chan = i;
memset(type, 0, sizeof(struct lp_type));
type->floating = format_desc->channel[chan].type == UTIL_FORMAT_TYPE_FLOAT;
type->fixed = format_desc->channel[chan].type == UTIL_FORMAT_TYPE_FIXED;
type->sign = format_desc->channel[chan].type != UTIL_FORMAT_TYPE_UNSIGNED;
type->norm = format_desc->channel[chan].normalized;
if (is_arithmetic_format(format_desc)) {
type->width = 0;
type->length = 1;
for (i = 0; i < format_desc->nr_channels; ++i) {
type->width += format_desc->channel[i].size;
}
} else {
type->width = format_desc->channel[chan].size;
type->length = format_desc->nr_channels;
}
}
/**
* Retrieves the type for a format which is usable in the blending code.
*
* e.g. RGBA16F = 4x float, R3G3B2 = 3x byte
*/
static inline void
lp_blend_type_from_format_desc(const struct util_format_description *format_desc,
struct lp_type* type)
{
unsigned i;
unsigned chan;
if (format_expands_to_float_soa(format_desc)) {
/* always use ordinary floats for blending */
type->floating = true;
type->fixed = false;
type->sign = true;
type->norm = false;
type->width = 32;
type->length = 4;
return;
}
for (i = 0; i < 4; i++)
if (format_desc->channel[i].type != UTIL_FORMAT_TYPE_VOID)
break;
chan = i;
memset(type, 0, sizeof(struct lp_type));
type->floating = format_desc->channel[chan].type == UTIL_FORMAT_TYPE_FLOAT;
type->fixed = format_desc->channel[chan].type == UTIL_FORMAT_TYPE_FIXED;
type->sign = format_desc->channel[chan].type != UTIL_FORMAT_TYPE_UNSIGNED;
type->norm = format_desc->channel[chan].normalized;
type->width = format_desc->channel[chan].size;
type->length = format_desc->nr_channels;
for (i = 1; i < format_desc->nr_channels; ++i) {
if (format_desc->channel[i].size > type->width)
type->width = format_desc->channel[i].size;
}
if (type->floating) {
type->width = 32;
} else {
if (type->width <= 8) {
type->width = 8;
} else if (type->width <= 16) {
type->width = 16;
} else {
type->width = 32;
}
}
if (is_arithmetic_format(format_desc) && type->length == 3) {
type->length = 4;
}
}
/**
* Scale a normalized value from src_bits to dst_bits.
*
* The exact calculation is
*
* dst = iround(src * dst_mask / src_mask)
*
* or with integer rounding
*
* dst = src * (2*dst_mask + sign(src)*src_mask) / (2*src_mask)
*
* where
*
* src_mask = (1 << src_bits) - 1
* dst_mask = (1 << dst_bits) - 1
*
* but we try to avoid division and multiplication through shifts.
*/
static inline LLVMValueRef
scale_bits(struct gallivm_state *gallivm,
int src_bits,
int dst_bits,
LLVMValueRef src,
struct lp_type src_type)
{
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef result = src;
if (dst_bits < src_bits) {
int delta_bits = src_bits - dst_bits;
if (delta_bits <= dst_bits) {
/*
* Approximate the rescaling with a single shift.
*
* This gives the wrong rounding.
*/
result = LLVMBuildLShr(builder,
src,
lp_build_const_int_vec(gallivm, src_type, delta_bits),
"");
} else {
/*
* Try more accurate rescaling.
*/
/*
* Drop the least significant bits to make space for the multiplication.
*
* XXX: A better approach would be to use a wider integer type as intermediate. But
* this is enough to convert alpha from 16bits -> 2 when rendering to
* PIPE_FORMAT_R10G10B10A2_UNORM.
*/
result = LLVMBuildLShr(builder,
src,
lp_build_const_int_vec(gallivm, src_type, dst_bits),
"");
result = LLVMBuildMul(builder,
result,
lp_build_const_int_vec(gallivm, src_type, (1LL << dst_bits) - 1),
"");
/*
* Add a rounding term before the division.
*
* TODO: Handle signed integers too.
*/
if (!src_type.sign) {
result = LLVMBuildAdd(builder,
result,
lp_build_const_int_vec(gallivm, src_type, (1LL << (delta_bits - 1))),
"");
}
/*
* Approximate the division by src_mask with a src_bits shift.
*
* Given the src has already been shifted by dst_bits, all we need
* to do is to shift by the difference.
*/
result = LLVMBuildLShr(builder,
result,
lp_build_const_int_vec(gallivm, src_type, delta_bits),
"");
}
} else if (dst_bits > src_bits) {
/* Scale up bits */
int db = dst_bits - src_bits;
/* Shift left by difference in bits */
result = LLVMBuildShl(builder,
src,
lp_build_const_int_vec(gallivm, src_type, db),
"");
if (db <= src_bits) {
/* Enough bits in src to fill the remainder */
LLVMValueRef lower = LLVMBuildLShr(builder,
src,
lp_build_const_int_vec(gallivm, src_type, src_bits - db),
"");
result = LLVMBuildOr(builder, result, lower, "");
} else if (db > src_bits) {
/* Need to repeatedly copy src bits to fill remainder in dst */
unsigned n;
for (n = src_bits; n < dst_bits; n *= 2) {
LLVMValueRef shuv = lp_build_const_int_vec(gallivm, src_type, n);
result = LLVMBuildOr(builder,
result,
LLVMBuildLShr(builder, result, shuv, ""),
"");
}
}
}
return result;
}
/**
* If RT is a smallfloat (needing denorms) format
*/
static inline int
have_smallfloat_format(struct lp_type dst_type,
enum pipe_format format)
{
return ((dst_type.floating && dst_type.width != 32) ||
/* due to format handling hacks this format doesn't have floating set
* here (and actually has width set to 32 too) so special case this. */
(format == PIPE_FORMAT_R11G11B10_FLOAT));
}
/**
* Convert from memory format to blending format
*
* e.g. GL_R3G3B2 is 1 byte in memory but 3 bytes for blending
*/
static void
convert_to_blend_type(struct gallivm_state *gallivm,
unsigned block_size,
const struct util_format_description *src_fmt,
struct lp_type src_type,
struct lp_type dst_type,
LLVMValueRef* src, // and dst
unsigned num_srcs)
{
LLVMValueRef *dst = src;
LLVMBuilderRef builder = gallivm->builder;
struct lp_type blend_type;
struct lp_type mem_type;
unsigned i, j;
unsigned pixels = block_size / num_srcs;
bool is_arith;
/*
* full custom path for packed floats and srgb formats - none of the later
* functions would do anything useful, and given the lp_type representation they
* can't be fixed. Should really have some SoA blend path for these kind of
* formats rather than hacking them in here.
*/
if (format_expands_to_float_soa(src_fmt)) {
LLVMValueRef tmpsrc[4];
/*
* This is pretty suboptimal for this case blending in SoA would be much
* better, since conversion gets us SoA values so need to convert back.
*/
assert(src_type.width == 32 || src_type.width == 16);
assert(dst_type.floating);
assert(dst_type.width == 32);
assert(dst_type.length % 4 == 0);
assert(num_srcs % 4 == 0);
if (src_type.width == 16) {
/* expand 4x16bit values to 4x32bit */
struct lp_type type32x4 = src_type;
LLVMTypeRef ltype32x4;
unsigned num_fetch = dst_type.length == 8 ? num_srcs / 2 : num_srcs / 4;
type32x4.width = 32;
ltype32x4 = lp_build_vec_type(gallivm, type32x4);
for (i = 0; i < num_fetch; i++) {
src[i] = LLVMBuildZExt(builder, src[i], ltype32x4, "");
}
src_type.width = 32;
}
for (i = 0; i < 4; i++) {
tmpsrc[i] = src[i];
}
for (i = 0; i < num_srcs / 4; i++) {
LLVMValueRef tmpsoa[4];
LLVMValueRef tmps = tmpsrc[i];
if (dst_type.length == 8) {
LLVMValueRef shuffles[8];
unsigned j;
/* fetch was 4 values but need 8-wide output values */
tmps = lp_build_concat(gallivm, &tmpsrc[i * 2], src_type, 2);
/*
* for 8-wide aos transpose would give us wrong order not matching
* incoming converted fs values and mask. ARGH.
*/
for (j = 0; j < 4; j++) {
shuffles[j] = lp_build_const_int32(gallivm, j * 2);
shuffles[j + 4] = lp_build_const_int32(gallivm, j * 2 + 1);
}
tmps = LLVMBuildShuffleVector(builder, tmps, tmps,
LLVMConstVector(shuffles, 8), "");
}
if (src_fmt->format == PIPE_FORMAT_R11G11B10_FLOAT) {
lp_build_r11g11b10_to_float(gallivm, tmps, tmpsoa);
}
else {
lp_build_unpack_rgba_soa(gallivm, src_fmt, dst_type, tmps, tmpsoa);
}
lp_build_transpose_aos(gallivm, dst_type, tmpsoa, &src[i * 4]);
}
return;
}
lp_mem_type_from_format_desc(src_fmt, &mem_type);
lp_blend_type_from_format_desc(src_fmt, &blend_type);
/* Is the format arithmetic */
is_arith = blend_type.length * blend_type.width != mem_type.width * mem_type.length;
is_arith &= !(mem_type.width == 16 && mem_type.floating);
/* Pad if necessary */
if (!is_arith && src_type.length < dst_type.length) {
for (i = 0; i < num_srcs; ++i) {
dst[i] = lp_build_pad_vector(gallivm, src[i], dst_type.length);
}
src_type.length = dst_type.length;
}
/* Special case for half-floats */
if (mem_type.width == 16 && mem_type.floating) {
assert(blend_type.width == 32 && blend_type.floating);
lp_build_conv_auto(gallivm, src_type, &dst_type, dst, num_srcs, dst);
is_arith = false;
}
if (!is_arith) {
return;
}
src_type.width = blend_type.width * blend_type.length;
blend_type.length *= pixels;
src_type.length *= pixels / (src_type.length / mem_type.length);
for (i = 0; i < num_srcs; ++i) {
LLVMValueRef chans[4];
LLVMValueRef res = NULL;
dst[i] = LLVMBuildZExt(builder, src[i], lp_build_vec_type(gallivm, src_type), "");
for (j = 0; j < src_fmt->nr_channels; ++j) {
unsigned mask = 0;
unsigned sa = src_fmt->channel[j].shift;
#if UTIL_ARCH_LITTLE_ENDIAN
unsigned from_lsb = j;
#else
unsigned from_lsb = src_fmt->nr_channels - j - 1;
#endif
mask = (1 << src_fmt->channel[j].size) - 1;
/* Extract bits from source */
chans[j] = LLVMBuildLShr(builder,
dst[i],
lp_build_const_int_vec(gallivm, src_type, sa),
"");
chans[j] = LLVMBuildAnd(builder,
chans[j],
lp_build_const_int_vec(gallivm, src_type, mask),
"");
/* Scale bits */
if (src_type.norm) {
chans[j] = scale_bits(gallivm, src_fmt->channel[j].size,
blend_type.width, chans[j], src_type);
}
/* Insert bits into correct position */
chans[j] = LLVMBuildShl(builder,
chans[j],
lp_build_const_int_vec(gallivm, src_type, from_lsb * blend_type.width),
"");
if (j == 0) {
res = chans[j];
} else {
res = LLVMBuildOr(builder, res, chans[j], "");
}
}
dst[i] = LLVMBuildBitCast(builder, res, lp_build_vec_type(gallivm, blend_type), "");
}
}
/**
* Convert from blending format to memory format
*
* e.g. GL_R3G3B2 is 3 bytes for blending but 1 byte in memory
*/
static void
convert_from_blend_type(struct gallivm_state *gallivm,
unsigned block_size,
const struct util_format_description *src_fmt,
struct lp_type src_type,
struct lp_type dst_type,
LLVMValueRef* src, // and dst
unsigned num_srcs)
{
LLVMValueRef* dst = src;
unsigned i, j, k;
struct lp_type mem_type;
struct lp_type blend_type;
LLVMBuilderRef builder = gallivm->builder;
unsigned pixels = block_size / num_srcs;
bool is_arith;
/*
* full custom path for packed floats and srgb formats - none of the later
* functions would do anything useful, and given the lp_type representation they
* can't be fixed. Should really have some SoA blend path for these kind of
* formats rather than hacking them in here.
*/
if (format_expands_to_float_soa(src_fmt)) {
/*
* This is pretty suboptimal for this case blending in SoA would be much
* better - we need to transpose the AoS values back to SoA values for
* conversion/packing.
*/
assert(src_type.floating);
assert(src_type.width == 32);
assert(src_type.length % 4 == 0);
assert(dst_type.width == 32 || dst_type.width == 16);
for (i = 0; i < num_srcs / 4; i++) {
LLVMValueRef tmpsoa[4], tmpdst;
lp_build_transpose_aos(gallivm, src_type, &src[i * 4], tmpsoa);
/* really really need SoA here */
if (src_fmt->format == PIPE_FORMAT_R11G11B10_FLOAT) {
tmpdst = lp_build_float_to_r11g11b10(gallivm, tmpsoa);
}
else {
tmpdst = lp_build_float_to_srgb_packed(gallivm, src_fmt,
src_type, tmpsoa);
}
if (src_type.length == 8) {
LLVMValueRef tmpaos, shuffles[8];
unsigned j;
/*
* for 8-wide aos transpose has given us wrong order not matching
* output order. HMPF. Also need to split the output values manually.
*/
for (j = 0; j < 4; j++) {
shuffles[j * 2] = lp_build_const_int32(gallivm, j);
shuffles[j * 2 + 1] = lp_build_const_int32(gallivm, j + 4);
}
tmpaos = LLVMBuildShuffleVector(builder, tmpdst, tmpdst,
LLVMConstVector(shuffles, 8), "");
src[i * 2] = lp_build_extract_range(gallivm, tmpaos, 0, 4);
src[i * 2 + 1] = lp_build_extract_range(gallivm, tmpaos, 4, 4);
}
else {
src[i] = tmpdst;
}
}
if (dst_type.width == 16) {
struct lp_type type16x8 = dst_type;
struct lp_type type32x4 = dst_type;
LLVMTypeRef ltype16x4, ltypei64, ltypei128;
unsigned num_fetch = src_type.length == 8 ? num_srcs / 2 : num_srcs / 4;
type16x8.length = 8;
type32x4.width = 32;
ltypei128 = LLVMIntTypeInContext(gallivm->context, 128);
ltypei64 = LLVMIntTypeInContext(gallivm->context, 64);
ltype16x4 = lp_build_vec_type(gallivm, dst_type);
/* We could do vector truncation but it doesn't generate very good code */
for (i = 0; i < num_fetch; i++) {
src[i] = lp_build_pack2(gallivm, type32x4, type16x8,
src[i], lp_build_zero(gallivm, type32x4));
src[i] = LLVMBuildBitCast(builder, src[i], ltypei128, "");
src[i] = LLVMBuildTrunc(builder, src[i], ltypei64, "");
src[i] = LLVMBuildBitCast(builder, src[i], ltype16x4, "");
}
}
return;
}
lp_mem_type_from_format_desc(src_fmt, &mem_type);
lp_blend_type_from_format_desc(src_fmt, &blend_type);
is_arith = (blend_type.length * blend_type.width != mem_type.width * mem_type.length);
/* Special case for half-floats */
if (mem_type.width == 16 && mem_type.floating) {
int length = dst_type.length;
assert(blend_type.width == 32 && blend_type.floating);
dst_type.length = src_type.length;
lp_build_conv_auto(gallivm, src_type, &dst_type, dst, num_srcs, dst);
dst_type.length = length;
is_arith = false;
}
/* Remove any padding */
if (!is_arith && (src_type.length % mem_type.length)) {
src_type.length -= (src_type.length % mem_type.length);
for (i = 0; i < num_srcs; ++i) {
dst[i] = lp_build_extract_range(gallivm, dst[i], 0, src_type.length);
}
}
/* No bit arithmetic to do */
if (!is_arith) {
return;
}
src_type.length = pixels;
src_type.width = blend_type.length * blend_type.width;
dst_type.length = pixels;
for (i = 0; i < num_srcs; ++i) {
LLVMValueRef chans[4];
LLVMValueRef res = NULL;
dst[i] = LLVMBuildBitCast(builder, src[i], lp_build_vec_type(gallivm, src_type), "");
for (j = 0; j < src_fmt->nr_channels; ++j) {
unsigned mask = 0;
unsigned sa = src_fmt->channel[j].shift;
#if UTIL_ARCH_LITTLE_ENDIAN
unsigned from_lsb = j;
#else
unsigned from_lsb = src_fmt->nr_channels - j - 1;
#endif
assert(blend_type.width > src_fmt->channel[j].size);
for (k = 0; k < blend_type.width; ++k) {
mask |= 1 << k;
}
/* Extract bits */
chans[j] = LLVMBuildLShr(builder,
dst[i],
lp_build_const_int_vec(gallivm, src_type,
from_lsb * blend_type.width),
"");
chans[j] = LLVMBuildAnd(builder,
chans[j],
lp_build_const_int_vec(gallivm, src_type, mask),
"");
/* Scale down bits */
if (src_type.norm) {
chans[j] = scale_bits(gallivm, blend_type.width,
src_fmt->channel[j].size, chans[j], src_type);
}
/* Insert bits */
chans[j] = LLVMBuildShl(builder,
chans[j],
lp_build_const_int_vec(gallivm, src_type, sa),
"");
sa += src_fmt->channel[j].size;
if (j == 0) {
res = chans[j];
} else {
res = LLVMBuildOr(builder, res, chans[j], "");
}
}
assert (dst_type.width != 24);
dst[i] = LLVMBuildTrunc(builder, res, lp_build_vec_type(gallivm, dst_type), "");
}
}
/**
* Convert alpha to same blend type as src
*/
static void
convert_alpha(struct gallivm_state *gallivm,
struct lp_type row_type,
struct lp_type alpha_type,
const unsigned block_size,
const unsigned block_height,
const unsigned src_count,
const unsigned dst_channels,
const bool pad_inline,
LLVMValueRef* src_alpha)
{
LLVMBuilderRef builder = gallivm->builder;
unsigned i, j;
unsigned length = row_type.length;
row_type.length = alpha_type.length;
/* Twiddle the alpha to match pixels */
lp_bld_quad_twiddle(gallivm, alpha_type, src_alpha, block_height, src_alpha);
/*
* TODO this should use single lp_build_conv call for
* src_count == 1 && dst_channels == 1 case (dropping the concat below)
*/
for (i = 0; i < block_height; ++i) {
lp_build_conv(gallivm, alpha_type, row_type, &src_alpha[i], 1, &src_alpha[i], 1);
}
alpha_type = row_type;
row_type.length = length;
/* If only one channel we can only need the single alpha value per pixel */
if (src_count == 1 && dst_channels == 1) {
lp_build_concat_n(gallivm, alpha_type, src_alpha, block_height, src_alpha, src_count);
} else {
/* If there are more srcs than rows then we need to split alpha up */
if (src_count > block_height) {
for (i = src_count; i > 0; --i) {
unsigned pixels = block_size / src_count;
unsigned idx = i - 1;
src_alpha[idx] = lp_build_extract_range(gallivm, src_alpha[(idx * pixels) / 4],
(idx * pixels) % 4, pixels);
}
}
/* If there is a src for each pixel broadcast the alpha across whole row */
if (src_count == block_size) {
for (i = 0; i < src_count; ++i) {
src_alpha[i] = lp_build_broadcast(gallivm,
lp_build_vec_type(gallivm, row_type), src_alpha[i]);
}
} else {
unsigned pixels = block_size / src_count;
unsigned channels = pad_inline ? TGSI_NUM_CHANNELS : dst_channels;
unsigned alpha_span = 1;
LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH];
/* Check if we need 2 src_alphas for our shuffles */
if (pixels > alpha_type.length) {
alpha_span = 2;
}
/* Broadcast alpha across all channels, e.g. a1a2 to a1a1a1a1a2a2a2a2 */
for (j = 0; j < row_type.length; ++j) {
if (j < pixels * channels) {
shuffles[j] = lp_build_const_int32(gallivm, j / channels);
} else {
shuffles[j] = LLVMGetUndef(LLVMInt32TypeInContext(gallivm->context));
}
}
for (i = 0; i < src_count; ++i) {
unsigned idx1 = i, idx2 = i;
if (alpha_span > 1){
idx1 *= alpha_span;
idx2 = idx1 + 1;
}
src_alpha[i] = LLVMBuildShuffleVector(builder,
src_alpha[idx1],
src_alpha[idx2],
LLVMConstVector(shuffles, row_type.length),
"");
}
}
}
}
/**
* Generates the blend function for unswizzled colour buffers
* Also generates the read & write from colour buffer
*/
static void
generate_unswizzled_blend(struct gallivm_state *gallivm,
unsigned rt,
struct lp_fragment_shader_variant *variant,
enum pipe_format out_format,
unsigned int num_fs,
struct lp_type fs_type,
LLVMValueRef* fs_mask,
LLVMValueRef fs_out_color[PIPE_MAX_COLOR_BUFS][TGSI_NUM_CHANNELS][4],
LLVMValueRef context_ptr,
LLVMValueRef color_ptr,
LLVMValueRef stride,
unsigned partial_mask,
boolean do_branch)
{
const unsigned alpha_channel = 3;
const unsigned block_width = LP_RASTER_BLOCK_SIZE;
const unsigned block_height = LP_RASTER_BLOCK_SIZE;
const unsigned block_size = block_width * block_height;
const unsigned lp_integer_vector_width = 128;
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef fs_src[4][TGSI_NUM_CHANNELS];
LLVMValueRef fs_src1[4][TGSI_NUM_CHANNELS];
LLVMValueRef src_alpha[4 * 4];
LLVMValueRef src1_alpha[4 * 4] = { NULL };
LLVMValueRef src_mask[4 * 4];
LLVMValueRef src[4 * 4];
LLVMValueRef src1[4 * 4];
LLVMValueRef dst[4 * 4];
LLVMValueRef blend_color;
LLVMValueRef blend_alpha;
LLVMValueRef i32_zero;
LLVMValueRef check_mask;
LLVMValueRef undef_src_val;
struct lp_build_mask_context mask_ctx;
struct lp_type mask_type;
struct lp_type blend_type;
struct lp_type row_type;
struct lp_type dst_type;
struct lp_type ls_type;
unsigned char swizzle[TGSI_NUM_CHANNELS];
unsigned vector_width;
unsigned src_channels = TGSI_NUM_CHANNELS;
unsigned dst_channels;
unsigned dst_count;
unsigned src_count;
unsigned i, j;
const struct util_format_description* out_format_desc = util_format_description(out_format);
unsigned dst_alignment;
bool pad_inline = is_arithmetic_format(out_format_desc);
bool has_alpha = false;
const boolean dual_source_blend = variant->key.blend.rt[0].blend_enable &&
util_blend_state_is_dual(&variant->key.blend, 0);
const boolean is_1d = variant->key.resource_1d;
boolean twiddle_after_convert = FALSE;
unsigned num_fullblock_fs = is_1d ? 2 * num_fs : num_fs;
LLVMValueRef fpstate = 0;
/* Get type from output format */
lp_blend_type_from_format_desc(out_format_desc, &row_type);
lp_mem_type_from_format_desc(out_format_desc, &dst_type);
/*
* Technically this code should go into lp_build_smallfloat_to_float
* and lp_build_float_to_smallfloat but due to the
* http://llvm.org/bugs/show_bug.cgi?id=6393
* llvm reorders the mxcsr intrinsics in a way that breaks the code.
* So the ordering is important here and there shouldn't be any
* llvm ir instrunctions in this function before
* this, otherwise half-float format conversions won't work
* (again due to llvm bug #6393).
*/
if (have_smallfloat_format(dst_type, out_format)) {
/* We need to make sure that denorms are ok for half float
conversions */
fpstate = lp_build_fpstate_get(gallivm);
lp_build_fpstate_set_denorms_zero(gallivm, FALSE);
}
mask_type = lp_int32_vec4_type();
mask_type.length = fs_type.length;
for (i = num_fs; i < num_fullblock_fs; i++) {
fs_mask[i] = lp_build_zero(gallivm, mask_type);
}
/* Do not bother executing code when mask is empty.. */
if (do_branch) {
check_mask = LLVMConstNull(lp_build_int_vec_type(gallivm, mask_type));
for (i = 0; i < num_fullblock_fs; ++i) {
check_mask = LLVMBuildOr(builder, check_mask, fs_mask[i], "");
}
lp_build_mask_begin(&mask_ctx, gallivm, mask_type, check_mask);
lp_build_mask_check(&mask_ctx);
}
partial_mask |= !variant->opaque;
i32_zero = lp_build_const_int32(gallivm, 0);
undef_src_val = lp_build_undef(gallivm, fs_type);
row_type.length = fs_type.length;
vector_width = dst_type.floating ? lp_native_vector_width : lp_integer_vector_width;
/* Compute correct swizzle and count channels */
memset(swizzle, LP_BLD_SWIZZLE_DONTCARE, TGSI_NUM_CHANNELS);
dst_channels = 0;
for (i = 0; i < TGSI_NUM_CHANNELS; ++i) {
/* Ensure channel is used */
if (out_format_desc->swizzle[i] >= TGSI_NUM_CHANNELS) {
continue;
}
/* Ensure not already written to (happens in case with GL_ALPHA) */
if (swizzle[out_format_desc->swizzle[i]] < TGSI_NUM_CHANNELS) {
continue;
}
/* Ensure we havn't already found all channels */
if (dst_channels >= out_format_desc->nr_channels) {
continue;
}
swizzle[out_format_desc->swizzle[i]] = i;
++dst_channels;
if (i == alpha_channel) {
has_alpha = true;
}
}
if (format_expands_to_float_soa(out_format_desc)) {
/*
* the code above can't work for layout_other
* for srgb it would sort of work but we short-circuit swizzles, etc.
* as that is done as part of unpack / pack.
*/
dst_channels = 4; /* HACK: this is fake 4 really but need it due to transpose stuff later */
has_alpha = true;
swizzle[0] = 0;
swizzle[1] = 1;
swizzle[2] = 2;
swizzle[3] = 3;
pad_inline = true; /* HACK: prevent rgbxrgbx->rgbrgbxx conversion later */
}
/* If 3 channels then pad to include alpha for 4 element transpose */
if (dst_channels == 3) {
assert (!has_alpha);
for (i = 0; i < TGSI_NUM_CHANNELS; i++) {
if (swizzle[i] > TGSI_NUM_CHANNELS)
swizzle[i] = 3;
}
if (out_format_desc->nr_channels == 4) {
dst_channels = 4;
/*
* We use alpha from the color conversion, not separate one.
* We had to include it for transpose, hence it will get converted
* too (albeit when doing transpose after conversion, that would
* no longer be the case necessarily).
* (It works only with 4 channel dsts, e.g. rgbx formats, because
* otherwise we really have padding, not alpha, included.)
*/
has_alpha = true;
}
}
/*
* Load shader output
*/
for (i = 0; i < num_fullblock_fs; ++i) {
/* Always load alpha for use in blending */
LLVMValueRef alpha;
if (i < num_fs) {
alpha = LLVMBuildLoad(builder, fs_out_color[rt][alpha_channel][i], "");
}
else {
alpha = undef_src_val;
}
/* Load each channel */
for (j = 0; j < dst_channels; ++j) {
assert(swizzle[j] < 4);
if (i < num_fs) {
fs_src[i][j] = LLVMBuildLoad(builder, fs_out_color[rt][swizzle[j]][i], "");
}
else {
fs_src[i][j] = undef_src_val;
}
}
/* If 3 channels then pad to include alpha for 4 element transpose */
/*
* XXX If we include that here maybe could actually use it instead of
* separate alpha for blending?
* (Difficult though we actually convert pad channels, not alpha.)
*/
if (dst_channels == 3 && !has_alpha) {
fs_src[i][3] = alpha;
}
/* We split the row_mask and row_alpha as we want 128bit interleave */
if (fs_type.length == 8) {
src_mask[i*2 + 0] = lp_build_extract_range(gallivm, fs_mask[i],
0, src_channels);
src_mask[i*2 + 1] = lp_build_extract_range(gallivm, fs_mask[i],
src_channels, src_channels);
src_alpha[i*2 + 0] = lp_build_extract_range(gallivm, alpha, 0, src_channels);
src_alpha[i*2 + 1] = lp_build_extract_range(gallivm, alpha,
src_channels, src_channels);
} else {
src_mask[i] = fs_mask[i];
src_alpha[i] = alpha;
}
}
if (dual_source_blend) {
/* same as above except different src/dst, skip masks and comments... */
for (i = 0; i < num_fullblock_fs; ++i) {
LLVMValueRef alpha;
if (i < num_fs) {
alpha = LLVMBuildLoad(builder, fs_out_color[1][alpha_channel][i], "");
}
else {
alpha = undef_src_val;
}
for (j = 0; j < dst_channels; ++j) {
assert(swizzle[j] < 4);
if (i < num_fs) {
fs_src1[i][j] = LLVMBuildLoad(builder, fs_out_color[1][swizzle[j]][i], "");
}
else {
fs_src1[i][j] = undef_src_val;
}
}
if (dst_channels == 3 && !has_alpha) {
fs_src1[i][3] = alpha;
}
if (fs_type.length == 8) {
src1_alpha[i*2 + 0] = lp_build_extract_range(gallivm, alpha, 0, src_channels);
src1_alpha[i*2 + 1] = lp_build_extract_range(gallivm, alpha,
src_channels, src_channels);
} else {
src1_alpha[i] = alpha;
}
}
}
if (util_format_is_pure_integer(out_format)) {
/*
* In this case fs_type was really ints or uints disguised as floats,
* fix that up now.
*/
fs_type.floating = 0;
fs_type.sign = dst_type.sign;
for (i = 0; i < num_fullblock_fs; ++i) {
for (j = 0; j < dst_channels; ++j) {
fs_src[i][j] = LLVMBuildBitCast(builder, fs_src[i][j],
lp_build_vec_type(gallivm, fs_type), "");
}
if (dst_channels == 3 && !has_alpha) {
fs_src[i][3] = LLVMBuildBitCast(builder, fs_src[i][3],
lp_build_vec_type(gallivm, fs_type), "");
}
}
}
/*
* We actually should generally do conversion first (for non-1d cases)
* when the blend format is 8 or 16 bits. The reason is obvious,
* there's 2 or 4 times less vectors to deal with for the interleave...
* Albeit for the AVX (not AVX2) case there's no benefit with 16 bit
* vectors (as it can do 32bit unpack with 256bit vectors, but 8/16bit
* unpack only with 128bit vectors).
* Note: for 16bit sizes really need matching pack conversion code
*/
if (!is_1d && dst_channels != 3 && dst_type.width == 8) {
twiddle_after_convert = TRUE;
}
/*
* Pixel twiddle from fragment shader order to memory order
*/
if (!twiddle_after_convert) {
src_count = generate_fs_twiddle(gallivm, fs_type, num_fullblock_fs,
dst_channels, fs_src, src, pad_inline);
if (dual_source_blend) {
generate_fs_twiddle(gallivm, fs_type, num_fullblock_fs, dst_channels,
fs_src1, src1, pad_inline);
}
} else {
src_count = num_fullblock_fs * dst_channels;
/*
* We reorder things a bit here, so the cases for 4-wide and 8-wide
* (AVX) turn out the same later when untwiddling/transpose (albeit
* for true AVX2 path untwiddle needs to be different).
* For now just order by colors first (so we can use unpack later).
*/
for (j = 0; j < num_fullblock_fs; j++) {
for (i = 0; i < dst_channels; i++) {
src[i*num_fullblock_fs + j] = fs_src[j][i];
if (dual_source_blend) {
src1[i*num_fullblock_fs + j] = fs_src1[j][i];
}
}
}
}
src_channels = dst_channels < 3 ? dst_channels : 4;
if (src_count != num_fullblock_fs * src_channels) {
unsigned ds = src_count / (num_fullblock_fs * src_channels);
row_type.length /= ds;
fs_type.length = row_type.length;
}
blend_type = row_type;
mask_type.length = 4;
/* Convert src to row_type */
if (dual_source_blend) {
struct lp_type old_row_type = row_type;
lp_build_conv_auto(gallivm, fs_type, &row_type, src, src_count, src);
src_count = lp_build_conv_auto(gallivm, fs_type, &old_row_type, src1, src_count, src1);
}
else {
src_count = lp_build_conv_auto(gallivm, fs_type, &row_type, src, src_count, src);
}
/* If the rows are not an SSE vector, combine them to become SSE size! */
if ((row_type.width * row_type.length) % 128) {
unsigned bits = row_type.width * row_type.length;
unsigned combined;
assert(src_count >= (vector_width / bits));
dst_count = src_count / (vector_width / bits);
combined = lp_build_concat_n(gallivm, row_type, src, src_count, src, dst_count);
if (dual_source_blend) {
lp_build_concat_n(gallivm, row_type, src1, src_count, src1, dst_count);
}
row_type.length *= combined;
src_count /= combined;
bits = row_type.width * row_type.length;
assert(bits == 128 || bits == 256);
}
if (twiddle_after_convert) {
fs_twiddle_transpose(gallivm, row_type, src, src_count, src);
if (dual_source_blend) {
fs_twiddle_transpose(gallivm, row_type, src1, src_count, src1);
}
}
/*
* Blend Colour conversion
*/
blend_color = lp_jit_context_f_blend_color(gallivm, context_ptr);
blend_color = LLVMBuildPointerCast(builder, blend_color,
LLVMPointerType(lp_build_vec_type(gallivm, fs_type), 0), "");
blend_color = LLVMBuildLoad(builder, LLVMBuildGEP(builder, blend_color,
&i32_zero, 1, ""), "");
/* Convert */
lp_build_conv(gallivm, fs_type, blend_type, &blend_color, 1, &blend_color, 1);
if (out_format_desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB) {
/*
* since blending is done with floats, there was no conversion.
* However, the rules according to fixed point renderbuffers still
* apply, that is we must clamp inputs to 0.0/1.0.
* (This would apply to separate alpha conversion too but we currently
* force has_alpha to be true.)
* TODO: should skip this with "fake" blend, since post-blend conversion
* will clamp anyway.
* TODO: could also skip this if fragment color clamping is enabled. We
* don't support it natively so it gets baked into the shader however, so
* can't really tell here.
*/
struct lp_build_context f32_bld;
assert(row_type.floating);
lp_build_context_init(&f32_bld, gallivm, row_type);
for (i = 0; i < src_count; i++) {
src[i] = lp_build_clamp_zero_one_nanzero(&f32_bld, src[i]);
}
if (dual_source_blend) {
for (i = 0; i < src_count; i++) {
src1[i] = lp_build_clamp_zero_one_nanzero(&f32_bld, src1[i]);
}
}
/* probably can't be different than row_type but better safe than sorry... */
lp_build_context_init(&f32_bld, gallivm, blend_type);
blend_color = lp_build_clamp(&f32_bld, blend_color, f32_bld.zero, f32_bld.one);
}
/* Extract alpha */
blend_alpha = lp_build_extract_broadcast(gallivm, blend_type, row_type, blend_color, lp_build_const_int32(gallivm, 3));
/* Swizzle to appropriate channels, e.g. from RGBA to BGRA BGRA */
pad_inline &= (dst_channels * (block_size / src_count) * row_type.width) != vector_width;
if (pad_inline) {
/* Use all 4 channels e.g. from RGBA RGBA to RGxx RGxx */
blend_color = lp_build_swizzle_aos_n(gallivm, blend_color, swizzle, TGSI_NUM_CHANNELS, row_type.length);
} else {
/* Only use dst_channels e.g. RGBA RGBA to RG RG xxxx */
blend_color = lp_build_swizzle_aos_n(gallivm, blend_color, swizzle, dst_channels, row_type.length);
}
/*
* Mask conversion
*/
lp_bld_quad_twiddle(gallivm, mask_type, &src_mask[0], block_height, &src_mask[0]);
if (src_count < block_height) {
lp_build_concat_n(gallivm, mask_type, src_mask, 4, src_mask, src_count);
} else if (src_count > block_height) {
for (i = src_count; i > 0; --i) {
unsigned pixels = block_size / src_count;
unsigned idx = i - 1;
src_mask[idx] = lp_build_extract_range(gallivm, src_mask[(idx * pixels) / 4],
(idx * pixels) % 4, pixels);
}
}
assert(mask_type.width == 32);
for (i = 0; i < src_count; ++i) {
unsigned pixels = block_size / src_count;
unsigned pixel_width = row_type.width * dst_channels;
if (pixel_width == 24) {
mask_type.width = 8;
mask_type.length = vector_width / mask_type.width;
} else {
mask_type.length = pixels;
mask_type.width = row_type.width * dst_channels;
/*
* If mask_type width is smaller than 32bit, this doesn't quite
* generate the most efficient code (could use some pack).
*/
src_mask[i] = LLVMBuildIntCast(builder, src_mask[i],
lp_build_int_vec_type(gallivm, mask_type), "");
mask_type.length *= dst_channels;
mask_type.width /= dst_channels;
}
src_mask[i] = LLVMBuildBitCast(builder, src_mask[i],
lp_build_int_vec_type(gallivm, mask_type), "");
src_mask[i] = lp_build_pad_vector(gallivm, src_mask[i], row_type.length);
}
/*
* Alpha conversion
*/
if (!has_alpha) {
struct lp_type alpha_type = fs_type;
alpha_type.length = 4;
convert_alpha(gallivm, row_type, alpha_type,
block_size, block_height,
src_count, dst_channels,
pad_inline, src_alpha);
if (dual_source_blend) {
convert_alpha(gallivm, row_type, alpha_type,
block_size, block_height,
src_count, dst_channels,
pad_inline, src1_alpha);
}
}
/*
* Load dst from memory
*/
if (src_count < block_height) {
dst_count = block_height;
} else {
dst_count = src_count;
}
dst_type.length *= block_size / dst_count;
if (format_expands_to_float_soa(out_format_desc)) {
/*
* we need multiple values at once for the conversion, so can as well
* load them vectorized here too instead of concatenating later.
* (Still need concatenation later for 8-wide vectors).
*/
dst_count = block_height;
dst_type.length = block_width;
}
/*
* Compute the alignment of the destination pointer in bytes
* We fetch 1-4 pixels, if the format has pot alignment then those fetches
* are always aligned by MIN2(16, fetch_width) except for buffers (not
* 1d tex but can't distinguish here) so need to stick with per-pixel
* alignment in this case.
*/
if (is_1d) {
dst_alignment = (out_format_desc->block.bits + 7)/(out_format_desc->block.width * 8);
}
else {
dst_alignment = dst_type.length * dst_type.width / 8;
}
/* Force power-of-two alignment by extracting only the least-significant-bit */
dst_alignment = 1 << (ffs(dst_alignment) - 1);
/*
* Resource base and stride pointers are aligned to 16 bytes, so that's
* the maximum alignment we can guarantee
*/
dst_alignment = MIN2(16, dst_alignment);
ls_type = dst_type;
if (dst_count > src_count) {
if ((dst_type.width == 8 || dst_type.width == 16) &&
util_is_power_of_two_or_zero(dst_type.length) &&
dst_type.length * dst_type.width < 128) {
/*
* Never try to load values as 4xi8 which we will then
* concatenate to larger vectors. This gives llvm a real
* headache (the problem is the type legalizer (?) will
* try to load that as 4xi8 zext to 4xi32 to fill the vector,
* then the shuffles to concatenate are more or less impossible
* - llvm is easily capable of generating a sequence of 32
* pextrb/pinsrb instructions for that. Albeit it appears to
* be fixed in llvm 4.0. So, load and concatenate with 32bit
* width to avoid the trouble (16bit seems not as bad, llvm
* probably recognizes the load+shuffle as only one shuffle
* is necessary, but we can do just the same anyway).
*/
ls_type.length = dst_type.length * dst_type.width / 32;
ls_type.width = 32;
}
}
if (is_1d) {
load_unswizzled_block(gallivm, color_ptr, stride, block_width, 1,
dst, ls_type, dst_count / 4, dst_alignment);
for (i = dst_count / 4; i < dst_count; i++) {
dst[i] = lp_build_undef(gallivm, ls_type);
}
}
else {
load_unswizzled_block(gallivm, color_ptr, stride, block_width, block_height,
dst, ls_type, dst_count, dst_alignment);
}
/*
* Convert from dst/output format to src/blending format.
*
* This is necessary as we can only read 1 row from memory at a time,
* so the minimum dst_count will ever be at this point is 4.
*
* With, for example, R8 format you can have all 16 pixels in a 128 bit vector,
* this will take the 4 dsts and combine them into 1 src so we can perform blending
* on all 16 pixels in that single vector at once.
*/
if (dst_count > src_count) {
if (ls_type.length != dst_type.length && ls_type.length == 1) {
LLVMTypeRef elem_type = lp_build_elem_type(gallivm, ls_type);
LLVMTypeRef ls_vec_type = LLVMVectorType(elem_type, 1);
for (i = 0; i < dst_count; i++) {
dst[i] = LLVMBuildBitCast(builder, dst[i], ls_vec_type, "");
}
}
lp_build_concat_n(gallivm, ls_type, dst, 4, dst, src_count);
if (ls_type.length != dst_type.length) {
struct lp_type tmp_type = dst_type;
tmp_type.length = dst_type.length * 4 / src_count;
for (i = 0; i < src_count; i++) {
dst[i] = LLVMBuildBitCast(builder, dst[i],
lp_build_vec_type(gallivm, tmp_type), "");
}
}
}
/*
* Blending
*/
/* XXX this is broken for RGB8 formats -
* they get expanded from 12 to 16 elements (to include alpha)
* by convert_to_blend_type then reduced to 15 instead of 12
* by convert_from_blend_type (a simple fix though breaks A8...).
* R16G16B16 also crashes differently however something going wrong
* inside llvm handling npot vector sizes seemingly.
* It seems some cleanup could be done here (like skipping conversion/blend
* when not needed).
*/
convert_to_blend_type(gallivm, block_size, out_format_desc, dst_type,
row_type, dst, src_count);
/*
* FIXME: Really should get logic ops / masks out of generic blend / row
* format. Logic ops will definitely not work on the blend float format
* used for SRGB here and I think OpenGL expects this to work as expected
* (that is incoming values converted to srgb then logic op applied).
*/
for (i = 0; i < src_count; ++i) {
dst[i] = lp_build_blend_aos(gallivm,
&variant->key.blend,
out_format,
row_type,
rt,
src[i],
has_alpha ? NULL : src_alpha[i],
src1[i],
has_alpha ? NULL : src1_alpha[i],
dst[i],
partial_mask ? src_mask[i] : NULL,
blend_color,
has_alpha ? NULL : blend_alpha,
swizzle,
pad_inline ? 4 : dst_channels);
}
convert_from_blend_type(gallivm, block_size, out_format_desc,
row_type, dst_type, dst, src_count);
/* Split the blend rows back to memory rows */
if (dst_count > src_count) {
row_type.length = dst_type.length * (dst_count / src_count);
if (src_count == 1) {
dst[1] = lp_build_extract_range(gallivm, dst[0], row_type.length / 2, row_type.length / 2);
dst[0] = lp_build_extract_range(gallivm, dst[0], 0, row_type.length / 2);
row_type.length /= 2;
src_count *= 2;
}
dst[3] = lp_build_extract_range(gallivm, dst[1], row_type.length / 2, row_type.length / 2);
dst[2] = lp_build_extract_range(gallivm, dst[1], 0, row_type.length / 2);
dst[1] = lp_build_extract_range(gallivm, dst[0], row_type.length / 2, row_type.length / 2);
dst[0] = lp_build_extract_range(gallivm, dst[0], 0, row_type.length / 2);
row_type.length /= 2;
src_count *= 2;
}
/*
* Store blend result to memory
*/
if (is_1d) {
store_unswizzled_block(gallivm, color_ptr, stride, block_width, 1,
dst, dst_type, dst_count / 4, dst_alignment);
}
else {
store_unswizzled_block(gallivm, color_ptr, stride, block_width, block_height,
dst, dst_type, dst_count, dst_alignment);
}
if (have_smallfloat_format(dst_type, out_format)) {
lp_build_fpstate_set(gallivm, fpstate);
}
if (do_branch) {
lp_build_mask_end(&mask_ctx);
}
}
/**
* Generate the runtime callable function for the whole fragment pipeline.
* Note that the function which we generate operates on a block of 16
* pixels at at time. The block contains 2x2 quads. Each quad contains
* 2x2 pixels.
*/
static void
generate_fragment(struct llvmpipe_context *lp,
struct lp_fragment_shader *shader,
struct lp_fragment_shader_variant *variant,
unsigned partial_mask)
{
struct gallivm_state *gallivm = variant->gallivm;
struct lp_fragment_shader_variant_key *key = &variant->key;
struct lp_shader_input inputs[PIPE_MAX_SHADER_INPUTS];
char func_name[64];
struct lp_type fs_type;
struct lp_type blend_type;
LLVMTypeRef fs_elem_type;
LLVMTypeRef blend_vec_type;
LLVMTypeRef arg_types[13];
LLVMTypeRef func_type;
LLVMTypeRef int32_type = LLVMInt32TypeInContext(gallivm->context);
LLVMTypeRef int8_type = LLVMInt8TypeInContext(gallivm->context);
LLVMValueRef context_ptr;
LLVMValueRef x;
LLVMValueRef y;
LLVMValueRef a0_ptr;
LLVMValueRef dadx_ptr;
LLVMValueRef dady_ptr;
LLVMValueRef color_ptr_ptr;
LLVMValueRef stride_ptr;
LLVMValueRef depth_ptr;
LLVMValueRef depth_stride;
LLVMValueRef mask_input;
LLVMValueRef thread_data_ptr;
LLVMBasicBlockRef block;
LLVMBuilderRef builder;
struct lp_build_sampler_soa *sampler;
struct lp_build_image_soa *image;
struct lp_build_interp_soa_context interp;
LLVMValueRef fs_mask[16 / 4];
LLVMValueRef fs_out_color[PIPE_MAX_COLOR_BUFS][TGSI_NUM_CHANNELS][16 / 4];
LLVMValueRef function;
LLVMValueRef facing;
unsigned num_fs;
unsigned i;
unsigned chan;
unsigned cbuf;
boolean cbuf0_write_all;
const boolean dual_source_blend = key->blend.rt[0].blend_enable &&
util_blend_state_is_dual(&key->blend, 0);
assert(lp_native_vector_width / 32 >= 4);
/* Adjust color input interpolation according to flatshade state:
*/
memcpy(inputs, shader->inputs, shader->info.base.num_inputs * sizeof inputs[0]);
for (i = 0; i < shader->info.base.num_inputs; i++) {
if (inputs[i].interp == LP_INTERP_COLOR) {
if (key->flatshade)
inputs[i].interp = LP_INTERP_CONSTANT;
else
inputs[i].interp = LP_INTERP_PERSPECTIVE;
}
}
/* check if writes to cbuf[0] are to be copied to all cbufs */
cbuf0_write_all =
shader->info.base.properties[TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS];
/* TODO: actually pick these based on the fs and color buffer
* characteristics. */
memset(&fs_type, 0, sizeof fs_type);
fs_type.floating = TRUE; /* floating point values */
fs_type.sign = TRUE; /* values are signed */
fs_type.norm = FALSE; /* values are not limited to [0,1] or [-1,1] */
fs_type.width = 32; /* 32-bit float */
fs_type.length = MIN2(lp_native_vector_width / 32, 16); /* n*4 elements per vector */
memset(&blend_type, 0, sizeof blend_type);
blend_type.floating = FALSE; /* values are integers */
blend_type.sign = FALSE; /* values are unsigned */
blend_type.norm = TRUE; /* values are in [0,1] or [-1,1] */
blend_type.width = 8; /* 8-bit ubyte values */
blend_type.length = 16; /* 16 elements per vector */
/*
* Generate the function prototype. Any change here must be reflected in
* lp_jit.h's lp_jit_frag_func function pointer type, and vice-versa.
*/
fs_elem_type = lp_build_elem_type(gallivm, fs_type);
blend_vec_type = lp_build_vec_type(gallivm, blend_type);
snprintf(func_name, sizeof(func_name), "fs%u_variant%u_%s",
shader->no, variant->no, partial_mask ? "partial" : "whole");
arg_types[0] = variant->jit_context_ptr_type; /* context */
arg_types[1] = int32_type; /* x */
arg_types[2] = int32_type; /* y */
arg_types[3] = int32_type; /* facing */
arg_types[4] = LLVMPointerType(fs_elem_type, 0); /* a0 */
arg_types[5] = LLVMPointerType(fs_elem_type, 0); /* dadx */
arg_types[6] = LLVMPointerType(fs_elem_type, 0); /* dady */
arg_types[7] = LLVMPointerType(LLVMPointerType(blend_vec_type, 0), 0); /* color */
arg_types[8] = LLVMPointerType(int8_type, 0); /* depth */
arg_types[9] = int32_type; /* mask_input */
arg_types[10] = variant->jit_thread_data_ptr_type; /* per thread data */
arg_types[11] = LLVMPointerType(int32_type, 0); /* stride */
arg_types[12] = int32_type; /* depth_stride */
func_type = LLVMFunctionType(LLVMVoidTypeInContext(gallivm->context),
arg_types, ARRAY_SIZE(arg_types), 0);
function = LLVMAddFunction(gallivm->module, func_name, func_type);
LLVMSetFunctionCallConv(function, LLVMCCallConv);
variant->function[partial_mask] = function;
/* XXX: need to propagate noalias down into color param now we are
* passing a pointer-to-pointer?
*/
for(i = 0; i < ARRAY_SIZE(arg_types); ++i)
if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
lp_add_function_attr(function, i + 1, LP_FUNC_ATTR_NOALIAS);
context_ptr = LLVMGetParam(function, 0);
x = LLVMGetParam(function, 1);
y = LLVMGetParam(function, 2);
facing = LLVMGetParam(function, 3);
a0_ptr = LLVMGetParam(function, 4);
dadx_ptr = LLVMGetParam(function, 5);
dady_ptr = LLVMGetParam(function, 6);
color_ptr_ptr = LLVMGetParam(function, 7);
depth_ptr = LLVMGetParam(function, 8);
mask_input = LLVMGetParam(function, 9);
thread_data_ptr = LLVMGetParam(function, 10);
stride_ptr = LLVMGetParam(function, 11);
depth_stride = LLVMGetParam(function, 12);
lp_build_name(context_ptr, "context");
lp_build_name(x, "x");
lp_build_name(y, "y");
lp_build_name(a0_ptr, "a0");
lp_build_name(dadx_ptr, "dadx");
lp_build_name(dady_ptr, "dady");
lp_build_name(color_ptr_ptr, "color_ptr_ptr");
lp_build_name(depth_ptr, "depth");
lp_build_name(mask_input, "mask_input");
lp_build_name(thread_data_ptr, "thread_data");
lp_build_name(stride_ptr, "stride_ptr");
lp_build_name(depth_stride, "depth_stride");
/*
* Function body
*/
block = LLVMAppendBasicBlockInContext(gallivm->context, function, "entry");
builder = gallivm->builder;
assert(builder);
LLVMPositionBuilderAtEnd(builder, block);
/*
* Must not count ps invocations if there's a null shader.
* (It would be ok to count with null shader if there's d/s tests,
* but only if there's d/s buffers too, which is different
* to implicit rasterization disable which must not depend
* on the d/s buffers.)
* Could use popcount on mask, but pixel accuracy is not required.
* Could disable if there's no stats query, but maybe not worth it.
*/
if (shader->info.base.num_instructions > 1) {
LLVMValueRef invocs, val;
invocs = lp_jit_thread_data_invocations(gallivm, thread_data_ptr);
val = LLVMBuildLoad(builder, invocs, "");
val = LLVMBuildAdd(builder, val,
LLVMConstInt(LLVMInt64TypeInContext(gallivm->context), 1, 0),
"invoc_count");
LLVMBuildStore(builder, val, invocs);
}
/* code generated texture sampling */
sampler = lp_llvm_sampler_soa_create(key->samplers);
image = lp_llvm_image_soa_create(lp_fs_variant_key_images(key));
num_fs = 16 / fs_type.length; /* number of loops per 4x4 stamp */
/* for 1d resources only run "upper half" of stamp */
if (key->resource_1d)
num_fs /= 2;
{
LLVMValueRef num_loop = lp_build_const_int32(gallivm, num_fs);
LLVMTypeRef mask_type = lp_build_int_vec_type(gallivm, fs_type);
LLVMValueRef mask_store = lp_build_array_alloca(gallivm, mask_type,
num_loop, "mask_store");
LLVMValueRef color_store[PIPE_MAX_COLOR_BUFS][TGSI_NUM_CHANNELS];
boolean pixel_center_integer =
shader->info.base.properties[TGSI_PROPERTY_FS_COORD_PIXEL_CENTER];
/*
* The shader input interpolation info is not explicitely baked in the
* shader key, but everything it derives from (TGSI, and flatshade) is
* already included in the shader key.
*/
lp_build_interp_soa_init(&interp,
gallivm,
shader->info.base.num_inputs,
inputs,
pixel_center_integer,
key->depth_clamp,
builder, fs_type,
a0_ptr, dadx_ptr, dady_ptr,
x, y);
for (i = 0; i < num_fs; i++) {
LLVMValueRef mask;
LLVMValueRef indexi = lp_build_const_int32(gallivm, i);
LLVMValueRef mask_ptr = LLVMBuildGEP(builder, mask_store,
&indexi, 1, "mask_ptr");
if (partial_mask) {
mask = generate_quad_mask(gallivm, fs_type,
i*fs_type.length/4, mask_input);
}
else {
mask = lp_build_const_int_vec(gallivm, fs_type, ~0);
}
LLVMBuildStore(builder, mask, mask_ptr);
}
generate_fs_loop(gallivm,
shader, key,
builder,
fs_type,
context_ptr,
num_loop,
&interp,
sampler,
image,
mask_store, /* output */
color_store,
depth_ptr,
depth_stride,
facing,
thread_data_ptr);
for (i = 0; i < num_fs; i++) {
LLVMValueRef indexi = lp_build_const_int32(gallivm, i);
LLVMValueRef ptr = LLVMBuildGEP(builder, mask_store,
&indexi, 1, "");
fs_mask[i] = LLVMBuildLoad(builder, ptr, "mask");
/* This is fucked up need to reorganize things */
for (cbuf = 0; cbuf < key->nr_cbufs; cbuf++) {
for (chan = 0; chan < TGSI_NUM_CHANNELS; ++chan) {
ptr = LLVMBuildGEP(builder,
color_store[cbuf * !cbuf0_write_all][chan],
&indexi, 1, "");
fs_out_color[cbuf][chan][i] = ptr;
}
}
if (dual_source_blend) {
/* only support one dual source blend target hence always use output 1 */
for (chan = 0; chan < TGSI_NUM_CHANNELS; ++chan) {
ptr = LLVMBuildGEP(builder,
color_store[1][chan],
&indexi, 1, "");
fs_out_color[1][chan][i] = ptr;
}
}
}
}
sampler->destroy(sampler);
image->destroy(image);
/* Loop over color outputs / color buffers to do blending.
*/
for(cbuf = 0; cbuf < key->nr_cbufs; cbuf++) {
if (key->cbuf_format[cbuf] != PIPE_FORMAT_NONE) {
LLVMValueRef color_ptr;
LLVMValueRef stride;
LLVMValueRef index = lp_build_const_int32(gallivm, cbuf);
boolean do_branch = ((key->depth.enabled
|| key->stencil[0].enabled
|| key->alpha.enabled)
&& !shader->info.base.uses_kill);
color_ptr = LLVMBuildLoad(builder,
LLVMBuildGEP(builder, color_ptr_ptr,
&index, 1, ""),
"");
lp_build_name(color_ptr, "color_ptr%d", cbuf);
stride = LLVMBuildLoad(builder,
LLVMBuildGEP(builder, stride_ptr, &index, 1, ""),
"");
generate_unswizzled_blend(gallivm, cbuf, variant,
key->cbuf_format[cbuf],
num_fs, fs_type, fs_mask, fs_out_color,
context_ptr, color_ptr, stride,
partial_mask, do_branch);
}
}
LLVMBuildRetVoid(builder);
gallivm_verify_function(gallivm, function);
}
static void
dump_fs_variant_key(struct lp_fragment_shader_variant_key *key)
{
unsigned i;
debug_printf("fs variant %p:\n", (void *) key);
if (key->flatshade) {
debug_printf("flatshade = 1\n");
}
for (i = 0; i < key->nr_cbufs; ++i) {
debug_printf("cbuf_format[%u] = %s\n", i, util_format_name(key->cbuf_format[i]));
}
if (key->depth.enabled || key->stencil[0].enabled) {
debug_printf("depth.format = %s\n", util_format_name(key->zsbuf_format));
}
if (key->depth.enabled) {
debug_printf("depth.func = %s\n", util_str_func(key->depth.func, TRUE));
debug_printf("depth.writemask = %u\n", key->depth.writemask);
}
for (i = 0; i < 2; ++i) {
if (key->stencil[i].enabled) {
debug_printf("stencil[%u].func = %s\n", i, util_str_func(key->stencil[i].func, TRUE));
debug_printf("stencil[%u].fail_op = %s\n", i, util_str_stencil_op(key->stencil[i].fail_op, TRUE));
debug_printf("stencil[%u].zpass_op = %s\n", i, util_str_stencil_op(key->stencil[i].zpass_op, TRUE));
debug_printf("stencil[%u].zfail_op = %s\n", i, util_str_stencil_op(key->stencil[i].zfail_op, TRUE));
debug_printf("stencil[%u].valuemask = 0x%x\n", i, key->stencil[i].valuemask);
debug_printf("stencil[%u].writemask = 0x%x\n", i, key->stencil[i].writemask);
}
}
if (key->alpha.enabled) {
debug_printf("alpha.func = %s\n", util_str_func(key->alpha.func, TRUE));
}
if (key->occlusion_count) {
debug_printf("occlusion_count = 1\n");
}
if (key->blend.logicop_enable) {
debug_printf("blend.logicop_func = %s\n", util_str_logicop(key->blend.logicop_func, TRUE));
}
else if (key->blend.rt[0].blend_enable) {
debug_printf("blend.rgb_func = %s\n", util_str_blend_func (key->blend.rt[0].rgb_func, TRUE));
debug_printf("blend.rgb_src_factor = %s\n", util_str_blend_factor(key->blend.rt[0].rgb_src_factor, TRUE));
debug_printf("blend.rgb_dst_factor = %s\n", util_str_blend_factor(key->blend.rt[0].rgb_dst_factor, TRUE));
debug_printf("blend.alpha_func = %s\n", util_str_blend_func (key->blend.rt[0].alpha_func, TRUE));
debug_printf("blend.alpha_src_factor = %s\n", util_str_blend_factor(key->blend.rt[0].alpha_src_factor, TRUE));
debug_printf("blend.alpha_dst_factor = %s\n", util_str_blend_factor(key->blend.rt[0].alpha_dst_factor, TRUE));
}
debug_printf("blend.colormask = 0x%x\n", key->blend.rt[0].colormask);
if (key->blend.alpha_to_coverage) {
debug_printf("blend.alpha_to_coverage is enabled\n");
}
for (i = 0; i < key->nr_samplers; ++i) {
const struct lp_static_sampler_state *sampler = &key->samplers[i].sampler_state;
debug_printf("sampler[%u] = \n", i);
debug_printf(" .wrap = %s %s %s\n",
util_str_tex_wrap(sampler->wrap_s, TRUE),
util_str_tex_wrap(sampler->wrap_t, TRUE),
util_str_tex_wrap(sampler->wrap_r, TRUE));
debug_printf(" .min_img_filter = %s\n",
util_str_tex_filter(sampler->min_img_filter, TRUE));
debug_printf(" .min_mip_filter = %s\n",
util_str_tex_mipfilter(sampler->min_mip_filter, TRUE));
debug_printf(" .mag_img_filter = %s\n",
util_str_tex_filter(sampler->mag_img_filter, TRUE));
if (sampler->compare_mode != PIPE_TEX_COMPARE_NONE)
debug_printf(" .compare_func = %s\n", util_str_func(sampler->compare_func, TRUE));
debug_printf(" .normalized_coords = %u\n", sampler->normalized_coords);
debug_printf(" .min_max_lod_equal = %u\n", sampler->min_max_lod_equal);
debug_printf(" .lod_bias_non_zero = %u\n", sampler->lod_bias_non_zero);
debug_printf(" .apply_min_lod = %u\n", sampler->apply_min_lod);
debug_printf(" .apply_max_lod = %u\n", sampler->apply_max_lod);
}
for (i = 0; i < key->nr_sampler_views; ++i) {
const struct lp_static_texture_state *texture = &key->samplers[i].texture_state;
debug_printf("texture[%u] = \n", i);
debug_printf(" .format = %s\n",
util_format_name(texture->format));
debug_printf(" .target = %s\n",
util_str_tex_target(texture->target, TRUE));
debug_printf(" .level_zero_only = %u\n",
texture->level_zero_only);
debug_printf(" .pot = %u %u %u\n",
texture->pot_width,
texture->pot_height,
texture->pot_depth);
}
struct lp_image_static_state *images = lp_fs_variant_key_images(key);
for (i = 0; i < key->nr_images; ++i) {
const struct lp_static_texture_state *image = &images[i].image_state;
debug_printf("image[%u] = \n", i);
debug_printf(" .format = %s\n",
util_format_name(image->format));
debug_printf(" .target = %s\n",
util_str_tex_target(image->target, TRUE));
debug_printf(" .level_zero_only = %u\n",
image->level_zero_only);
debug_printf(" .pot = %u %u %u\n",
image->pot_width,
image->pot_height,
image->pot_depth);
}
}
void
lp_debug_fs_variant(struct lp_fragment_shader_variant *variant)
{
debug_printf("llvmpipe: Fragment shader #%u variant #%u:\n",
variant->shader->no, variant->no);
tgsi_dump(variant->shader->base.tokens, 0);
dump_fs_variant_key(&variant->key);
debug_printf("variant->opaque = %u\n", variant->opaque);
debug_printf("\n");
}
/**
* Generate a new fragment shader variant from the shader code and
* other state indicated by the key.
*/
static struct lp_fragment_shader_variant *
generate_variant(struct llvmpipe_context *lp,
struct lp_fragment_shader *shader,
const struct lp_fragment_shader_variant_key *key)
{
struct lp_fragment_shader_variant *variant;
const struct util_format_description *cbuf0_format_desc = NULL;
boolean fullcolormask;
char module_name[64];
variant = MALLOC(sizeof *variant + shader->variant_key_size - sizeof variant->key);
if (!variant)
return NULL;
memset(variant, 0, sizeof(*variant));
snprintf(module_name, sizeof(module_name), "fs%u_variant%u",
shader->no, shader->variants_created);
variant->gallivm = gallivm_create(module_name, lp->context);
if (!variant->gallivm) {
FREE(variant);
return NULL;
}
variant->shader = shader;
variant->list_item_global.base = variant;
variant->list_item_local.base = variant;
variant->no = shader->variants_created++;
memcpy(&variant->key, key, shader->variant_key_size);
/*
* Determine whether we are touching all channels in the color buffer.
*/
fullcolormask = FALSE;
if (key->nr_cbufs == 1) {
cbuf0_format_desc = util_format_description(key->cbuf_format[0]);
fullcolormask = util_format_colormask_full(cbuf0_format_desc, key->blend.rt[0].colormask);
}
variant->opaque =
!key->blend.logicop_enable &&
!key->blend.rt[0].blend_enable &&
fullcolormask &&
!key->stencil[0].enabled &&
!key->alpha.enabled &&
!key->blend.alpha_to_coverage &&
!key->depth.enabled &&
!shader->info.base.uses_kill &&
!shader->info.base.writes_samplemask
? TRUE : FALSE;
if ((LP_DEBUG & DEBUG_FS) || (gallivm_debug & GALLIVM_DEBUG_IR)) {
lp_debug_fs_variant(variant);
}
lp_jit_init_types(variant);
if (variant->jit_function[RAST_EDGE_TEST] == NULL)
generate_fragment(lp, shader, variant, RAST_EDGE_TEST);
if (variant->jit_function[RAST_WHOLE] == NULL) {
if (variant->opaque) {
/* Specialized shader, which doesn't need to read the color buffer. */
generate_fragment(lp, shader, variant, RAST_WHOLE);
}
}
/*
* Compile everything
*/
gallivm_compile_module(variant->gallivm);
variant->nr_instrs += lp_build_count_ir_module(variant->gallivm->module);
if (variant->function[RAST_EDGE_TEST]) {
variant->jit_function[RAST_EDGE_TEST] = (lp_jit_frag_func)
gallivm_jit_function(variant->gallivm,
variant->function[RAST_EDGE_TEST]);
}
if (variant->function[RAST_WHOLE]) {
variant->jit_function[RAST_WHOLE] = (lp_jit_frag_func)
gallivm_jit_function(variant->gallivm,
variant->function[RAST_WHOLE]);
} else if (!variant->jit_function[RAST_WHOLE]) {
variant->jit_function[RAST_WHOLE] = variant->jit_function[RAST_EDGE_TEST];
}
gallivm_free_ir(variant->gallivm);
return variant;
}
static void *
llvmpipe_create_fs_state(struct pipe_context *pipe,
const struct pipe_shader_state *templ)
{
struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
struct lp_fragment_shader *shader;
int nr_samplers;
int nr_sampler_views;
int nr_images;
int i;
shader = CALLOC_STRUCT(lp_fragment_shader);
if (!shader)
return NULL;
shader->no = fs_no++;
make_empty_list(&shader->variants);
/* get/save the summary info for this shader */
lp_build_tgsi_info(templ->tokens, &shader->info);
/* we need to keep a local copy of the tokens */
shader->base.tokens = tgsi_dup_tokens(templ->tokens);
shader->draw_data = draw_create_fragment_shader(llvmpipe->draw, templ);
if (shader->draw_data == NULL) {
FREE((void *) shader->base.tokens);
FREE(shader);
return NULL;
}
nr_samplers = shader->info.base.file_max[TGSI_FILE_SAMPLER] + 1;
nr_sampler_views = shader->info.base.file_max[TGSI_FILE_SAMPLER_VIEW] + 1;
nr_images = shader->info.base.file_max[TGSI_FILE_IMAGE] + 1;
shader->variant_key_size = lp_fs_variant_key_size(MAX2(nr_samplers, nr_sampler_views), nr_images);
for (i = 0; i < shader->info.base.num_inputs; i++) {
shader->inputs[i].usage_mask = shader->info.base.input_usage_mask[i];
shader->inputs[i].cyl_wrap = shader->info.base.input_cylindrical_wrap[i];
switch (shader->info.base.input_interpolate[i]) {
case TGSI_INTERPOLATE_CONSTANT:
shader->inputs[i].interp = LP_INTERP_CONSTANT;
break;
case TGSI_INTERPOLATE_LINEAR:
shader->inputs[i].interp = LP_INTERP_LINEAR;
break;
case TGSI_INTERPOLATE_PERSPECTIVE:
shader->inputs[i].interp = LP_INTERP_PERSPECTIVE;
break;
case TGSI_INTERPOLATE_COLOR:
shader->inputs[i].interp = LP_INTERP_COLOR;
break;
default:
assert(0);
break;
}
switch (shader->info.base.input_semantic_name[i]) {
case TGSI_SEMANTIC_FACE:
shader->inputs[i].interp = LP_INTERP_FACING;
break;
case TGSI_SEMANTIC_POSITION:
/* Position was already emitted above
*/
shader->inputs[i].interp = LP_INTERP_POSITION;
shader->inputs[i].src_index = 0;
continue;
}
/* XXX this is a completely pointless index map... */
shader->inputs[i].src_index = i+1;
}
if (LP_DEBUG & DEBUG_TGSI) {
unsigned attrib;
debug_printf("llvmpipe: Create fragment shader #%u %p:\n",
shader->no, (void *) shader);
tgsi_dump(templ->tokens, 0);
debug_printf("usage masks:\n");
for (attrib = 0; attrib < shader->info.base.num_inputs; ++attrib) {
unsigned usage_mask = shader->info.base.input_usage_mask[attrib];
debug_printf(" IN[%u].%s%s%s%s\n",
attrib,
usage_mask & TGSI_WRITEMASK_X ? "x" : "",
usage_mask & TGSI_WRITEMASK_Y ? "y" : "",
usage_mask & TGSI_WRITEMASK_Z ? "z" : "",
usage_mask & TGSI_WRITEMASK_W ? "w" : "");
}
debug_printf("\n");
}
return shader;
}
static void
llvmpipe_bind_fs_state(struct pipe_context *pipe, void *fs)
{
struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
if (llvmpipe->fs == fs)
return;
llvmpipe->fs = (struct lp_fragment_shader *) fs;
draw_bind_fragment_shader(llvmpipe->draw,
(llvmpipe->fs ? llvmpipe->fs->draw_data : NULL));
llvmpipe->dirty |= LP_NEW_FS;
}
/**
* Remove shader variant from two lists: the shader's variant list
* and the context's variant list.
*/
static void
llvmpipe_remove_shader_variant(struct llvmpipe_context *lp,
struct lp_fragment_shader_variant *variant)
{
if ((LP_DEBUG & DEBUG_FS) || (gallivm_debug & GALLIVM_DEBUG_IR)) {
debug_printf("llvmpipe: del fs #%u var %u v created %u v cached %u "
"v total cached %u inst %u total inst %u\n",
variant->shader->no, variant->no,
variant->shader->variants_created,
variant->shader->variants_cached,
lp->nr_fs_variants, variant->nr_instrs, lp->nr_fs_instrs);
}
gallivm_destroy(variant->gallivm);
/* remove from shader's list */
remove_from_list(&variant->list_item_local);
variant->shader->variants_cached--;
/* remove from context's list */
remove_from_list(&variant->list_item_global);
lp->nr_fs_variants--;
lp->nr_fs_instrs -= variant->nr_instrs;
FREE(variant);
}
static void
llvmpipe_delete_fs_state(struct pipe_context *pipe, void *fs)
{
struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
struct lp_fragment_shader *shader = fs;
struct lp_fs_variant_list_item *li;
assert(fs != llvmpipe->fs);
/*
* XXX: we need to flush the context until we have some sort of reference
* counting in fragment shaders as they may still be binned
* Flushing alone might not sufficient we need to wait on it too.
*/
llvmpipe_finish(pipe, __FUNCTION__);
/* Delete all the variants */
li = first_elem(&shader->variants);
while(!at_end(&shader->variants, li)) {
struct lp_fs_variant_list_item *next = next_elem(li);
llvmpipe_remove_shader_variant(llvmpipe, li->base);
li = next;
}
/* Delete draw module's data */
draw_delete_fragment_shader(llvmpipe->draw, shader->draw_data);
assert(shader->variants_cached == 0);
FREE((void *) shader->base.tokens);
FREE(shader);
}
static void
llvmpipe_set_constant_buffer(struct pipe_context *pipe,
enum pipe_shader_type shader, uint index,
const struct pipe_constant_buffer *cb)
{
struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
struct pipe_resource *constants = cb ? cb->buffer : NULL;
assert(shader < PIPE_SHADER_TYPES);
assert(index < ARRAY_SIZE(llvmpipe->constants[shader]));
/* note: reference counting */
util_copy_constant_buffer(&llvmpipe->constants[shader][index], cb);
if (constants) {
if (!(constants->bind & PIPE_BIND_CONSTANT_BUFFER)) {
debug_printf("Illegal set constant without bind flag\n");
constants->bind |= PIPE_BIND_CONSTANT_BUFFER;
}
}
if (shader == PIPE_SHADER_VERTEX ||
shader == PIPE_SHADER_GEOMETRY) {
/* Pass the constants to the 'draw' module */
const unsigned size = cb ? cb->buffer_size : 0;
const ubyte *data;
if (constants) {
data = (ubyte *) llvmpipe_resource_data(constants);
}
else if (cb && cb->user_buffer) {
data = (ubyte *) cb->user_buffer;
}
else {
data = NULL;
}
if (data)
data += cb->buffer_offset;
draw_set_mapped_constant_buffer(llvmpipe->draw, shader,
index, data, size);
}
else if (shader == PIPE_SHADER_COMPUTE)
llvmpipe->cs_dirty |= LP_CSNEW_CONSTANTS;
else
llvmpipe->dirty |= LP_NEW_FS_CONSTANTS;
if (cb && cb->user_buffer) {
pipe_resource_reference(&constants, NULL);
}
}
static void
llvmpipe_set_shader_buffers(struct pipe_context *pipe,
enum pipe_shader_type shader, unsigned start_slot,
unsigned count, const struct pipe_shader_buffer *buffers,
unsigned writable_bitmask)
{
struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
unsigned i, idx;
for (i = start_slot, idx = 0; i < start_slot + count; i++, idx++) {
const struct pipe_shader_buffer *buffer = buffers ? &buffers[idx] : NULL;
util_copy_shader_buffer(&llvmpipe->ssbos[shader][i], buffer);
if (shader == PIPE_SHADER_VERTEX ||
shader == PIPE_SHADER_GEOMETRY) {
const unsigned size = buffer ? buffer->buffer_size : 0;
const ubyte *data = NULL;
if (buffer && buffer->buffer)
data = (ubyte *) llvmpipe_resource_data(buffer->buffer);
if (data)
data += buffer->buffer_offset;
draw_set_mapped_shader_buffer(llvmpipe->draw, shader,
i, data, size);
} else if (shader == PIPE_SHADER_COMPUTE) {
llvmpipe->cs_dirty |= LP_CSNEW_SSBOS;
} else if (shader == PIPE_SHADER_FRAGMENT) {
llvmpipe->dirty |= LP_NEW_FS_SSBOS;
}
}
}
static void
llvmpipe_set_shader_images(struct pipe_context *pipe,
enum pipe_shader_type shader, unsigned start_slot,
unsigned count, const struct pipe_image_view *images)
{
struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
unsigned i, idx;
draw_flush(llvmpipe->draw);
for (i = start_slot, idx = 0; i < start_slot + count; i++, idx++) {
const struct pipe_image_view *image = images ? &images[idx] : NULL;
util_copy_image_view(&llvmpipe->images[shader][i], image);
}
llvmpipe->num_images[shader] = start_slot + count;
if (shader == PIPE_SHADER_VERTEX ||
shader == PIPE_SHADER_GEOMETRY) {
draw_set_images(llvmpipe->draw,
shader,
llvmpipe->images[shader],
start_slot + count);
} else if (shader == PIPE_SHADER_COMPUTE)
llvmpipe->cs_dirty |= LP_CSNEW_IMAGES;
else
llvmpipe->dirty |= LP_NEW_FS_IMAGES;
}
/**
* Return the blend factor equivalent to a destination alpha of one.
*/
static inline unsigned
force_dst_alpha_one(unsigned factor, boolean clamped_zero)
{
switch(factor) {
case PIPE_BLENDFACTOR_DST_ALPHA:
return PIPE_BLENDFACTOR_ONE;
case PIPE_BLENDFACTOR_INV_DST_ALPHA:
return PIPE_BLENDFACTOR_ZERO;
case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
if (clamped_zero)
return PIPE_BLENDFACTOR_ZERO;
else
return PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE;
}
return factor;
}
/**
* We need to generate several variants of the fragment pipeline to match
* all the combinations of the contributing state atoms.
*
* TODO: there is actually no reason to tie this to context state -- the
* generated code could be cached globally in the screen.
*/
static struct lp_fragment_shader_variant_key *
make_variant_key(struct llvmpipe_context *lp,
struct lp_fragment_shader *shader,
char *store)
{
unsigned i;
struct lp_fragment_shader_variant_key *key;
key = (struct lp_fragment_shader_variant_key *)store;
memset(key, 0, offsetof(struct lp_fragment_shader_variant_key, samplers[1]));
if (lp->framebuffer.zsbuf) {
enum pipe_format zsbuf_format = lp->framebuffer.zsbuf->format;
const struct util_format_description *zsbuf_desc =
util_format_description(zsbuf_format);
if (lp->depth_stencil->depth.enabled &&
util_format_has_depth(zsbuf_desc)) {
key->zsbuf_format = zsbuf_format;
memcpy(&key->depth, &lp->depth_stencil->depth, sizeof key->depth);
}
if (lp->depth_stencil->stencil[0].enabled &&
util_format_has_stencil(zsbuf_desc)) {
key->zsbuf_format = zsbuf_format;
memcpy(&key->stencil, &lp->depth_stencil->stencil, sizeof key->stencil);
}
if (llvmpipe_resource_is_1d(lp->framebuffer.zsbuf->texture)) {
key->resource_1d = TRUE;
}
}
/*
* Propagate the depth clamp setting from the rasterizer state.
* depth_clip == 0 implies depth clamping is enabled.
*
* When clip_halfz is enabled, then always clamp the depth values.
*
* XXX: This is incorrect for GL, but correct for d3d10 (depth
* clamp is always active in d3d10, regardless if depth clip is
* enabled or not).
* (GL has an always-on [0,1] clamp on fs depth output instead
* to ensure the depth values stay in range. Doesn't look like
* we do that, though...)
*/
if (lp->rasterizer->clip_halfz) {
key->depth_clamp = 1;
} else {
key->depth_clamp = (lp->rasterizer->depth_clip_near == 0) ? 1 : 0;
}
/* alpha test only applies if render buffer 0 is non-integer (or does not exist) */
if (!lp->framebuffer.nr_cbufs ||
!lp->framebuffer.cbufs[0] ||
!util_format_is_pure_integer(lp->framebuffer.cbufs[0]->format)) {
key->alpha.enabled = lp->depth_stencil->alpha.enabled;
}
if(key->alpha.enabled)
key->alpha.func = lp->depth_stencil->alpha.func;
/* alpha.ref_value is passed in jit_context */
key->flatshade = lp->rasterizer->flatshade;
if (lp->active_occlusion_queries) {
key->occlusion_count = TRUE;
}
if (lp->framebuffer.nr_cbufs) {
memcpy(&key->blend, lp->blend, sizeof key->blend);
}
key->nr_cbufs = lp->framebuffer.nr_cbufs;
if (!key->blend.independent_blend_enable) {
/* we always need independent blend otherwise the fixups below won't work */
for (i = 1; i < key->nr_cbufs; i++) {
memcpy(&key->blend.rt[i], &key->blend.rt[0], sizeof(key->blend.rt[0]));
}
key->blend.independent_blend_enable = 1;
}
for (i = 0; i < lp->framebuffer.nr_cbufs; i++) {
struct pipe_rt_blend_state *blend_rt = &key->blend.rt[i];
if (lp->framebuffer.cbufs[i]) {
enum pipe_format format = lp->framebuffer.cbufs[i]->format;
const struct util_format_description *format_desc;
key->cbuf_format[i] = format;
/*
* Figure out if this is a 1d resource. Note that OpenGL allows crazy
* mixing of 2d textures with height 1 and 1d textures, so make sure
* we pick 1d if any cbuf or zsbuf is 1d.
*/
if (llvmpipe_resource_is_1d(lp->framebuffer.cbufs[i]->texture)) {
key->resource_1d = TRUE;
}
format_desc = util_format_description(format);
assert(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_RGB ||
format_desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB);
/*
* Mask out color channels not present in the color buffer.
*/
blend_rt->colormask &= util_format_colormask(format_desc);
/*
* Disable blend for integer formats.
*/
if (util_format_is_pure_integer(format)) {
blend_rt->blend_enable = 0;
}
/*
* Our swizzled render tiles always have an alpha channel, but the
* linear render target format often does not, so force here the dst
* alpha to be one.
*
* This is not a mere optimization. Wrong results will be produced if
* the dst alpha is used, the dst format does not have alpha, and the
* previous rendering was not flushed from the swizzled to linear
* buffer. For example, NonPowTwo DCT.
*
* TODO: This should be generalized to all channels for better
* performance, but only alpha causes correctness issues.
*
* Also, force rgb/alpha func/factors match, to make AoS blending
* easier.
*/
if (format_desc->swizzle[3] > PIPE_SWIZZLE_W ||
format_desc->swizzle[3] == format_desc->swizzle[0]) {
/* Doesn't cover mixed snorm/unorm but can't render to them anyway */
boolean clamped_zero = !util_format_is_float(format) &&
!util_format_is_snorm(format);
blend_rt->rgb_src_factor =
force_dst_alpha_one(blend_rt->rgb_src_factor, clamped_zero);
blend_rt->rgb_dst_factor =
force_dst_alpha_one(blend_rt->rgb_dst_factor, clamped_zero);
blend_rt->alpha_func = blend_rt->rgb_func;
blend_rt->alpha_src_factor = blend_rt->rgb_src_factor;
blend_rt->alpha_dst_factor = blend_rt->rgb_dst_factor;
}
}
else {
/* no color buffer for this fragment output */
key->cbuf_format[i] = PIPE_FORMAT_NONE;
blend_rt->colormask = 0x0;
blend_rt->blend_enable = 0;
}
}
/* This value will be the same for all the variants of a given shader:
*/
key->nr_samplers = shader->info.base.file_max[TGSI_FILE_SAMPLER] + 1;
struct lp_sampler_static_state *fs_sampler;
fs_sampler = key->samplers;
memset(fs_sampler, 0, MAX2(key->nr_samplers, key->nr_sampler_views) * sizeof *fs_sampler);
for(i = 0; i < key->nr_samplers; ++i) {
if(shader->info.base.file_mask[TGSI_FILE_SAMPLER] & (1 << i)) {
lp_sampler_static_sampler_state(&fs_sampler[i].sampler_state,
lp->samplers[PIPE_SHADER_FRAGMENT][i]);
}
}
/*
* XXX If TGSI_FILE_SAMPLER_VIEW exists assume all texture opcodes
* are dx10-style? Can't really have mixed opcodes, at least not
* if we want to skip the holes here (without rescanning tgsi).
*/
if (shader->info.base.file_max[TGSI_FILE_SAMPLER_VIEW] != -1) {
key->nr_sampler_views = shader->info.base.file_max[TGSI_FILE_SAMPLER_VIEW] + 1;
for(i = 0; i < key->nr_sampler_views; ++i) {
/*
* Note sview may exceed what's representable by file_mask.
* This will still work, the only downside is that not actually
* used views may be included in the shader key.
*/
if(shader->info.base.file_mask[TGSI_FILE_SAMPLER_VIEW] & (1u << (i & 31))) {
lp_sampler_static_texture_state(&fs_sampler[i].texture_state,
lp->sampler_views[PIPE_SHADER_FRAGMENT][i]);
}
}
}
else {
key->nr_sampler_views = key->nr_samplers;
for(i = 0; i < key->nr_sampler_views; ++i) {
if(shader->info.base.file_mask[TGSI_FILE_SAMPLER] & (1 << i)) {
lp_sampler_static_texture_state(&fs_sampler[i].texture_state,
lp->sampler_views[PIPE_SHADER_FRAGMENT][i]);
}
}
}
struct lp_image_static_state *lp_image;
lp_image = lp_fs_variant_key_images(key);
key->nr_images = shader->info.base.file_max[TGSI_FILE_IMAGE] + 1;
for (i = 0; i < key->nr_images; ++i) {
if (shader->info.base.file_mask[TGSI_FILE_IMAGE] & (1 << i)) {
lp_sampler_static_texture_state_image(&lp_image[i].image_state,
&lp->images[PIPE_SHADER_FRAGMENT][i]);
}
}
return key;
}
/**
* Update fragment shader state. This is called just prior to drawing
* something when some fragment-related state has changed.
*/
void
llvmpipe_update_fs(struct llvmpipe_context *lp)
{
struct lp_fragment_shader *shader = lp->fs;
struct lp_fragment_shader_variant_key *key;
struct lp_fragment_shader_variant *variant = NULL;
struct lp_fs_variant_list_item *li;
char store[LP_FS_MAX_VARIANT_KEY_SIZE];
key = make_variant_key(lp, shader, store);
/* Search the variants for one which matches the key */
li = first_elem(&shader->variants);
while(!at_end(&shader->variants, li)) {
if(memcmp(&li->base->key, key, shader->variant_key_size) == 0) {
variant = li->base;
break;
}
li = next_elem(li);
}
if (variant) {
/* Move this variant to the head of the list to implement LRU
* deletion of shader's when we have too many.
*/
move_to_head(&lp->fs_variants_list, &variant->list_item_global);
}
else {
/* variant not found, create it now */
int64_t t0, t1, dt;
unsigned i;
unsigned variants_to_cull;
if (LP_DEBUG & DEBUG_FS) {
debug_printf("%u variants,\t%u instrs,\t%u instrs/variant\n",
lp->nr_fs_variants,
lp->nr_fs_instrs,
lp->nr_fs_variants ? lp->nr_fs_instrs / lp->nr_fs_variants : 0);
}
/* First, check if we've exceeded the max number of shader variants.
* If so, free 6.25% of them (the least recently used ones).
*/
variants_to_cull = lp->nr_fs_variants >= LP_MAX_SHADER_VARIANTS ? LP_MAX_SHADER_VARIANTS / 16 : 0;
if (variants_to_cull ||
lp->nr_fs_instrs >= LP_MAX_SHADER_INSTRUCTIONS) {
struct pipe_context *pipe = &lp->pipe;
if (gallivm_debug & GALLIVM_DEBUG_PERF) {
debug_printf("Evicting FS: %u fs variants,\t%u total variants,"
"\t%u instrs,\t%u instrs/variant\n",
shader->variants_cached,
lp->nr_fs_variants, lp->nr_fs_instrs,
lp->nr_fs_instrs / lp->nr_fs_variants);
}
/*
* XXX: we need to flush the context until we have some sort of
* reference counting in fragment shaders as they may still be binned
* Flushing alone might not be sufficient we need to wait on it too.
*/
llvmpipe_finish(pipe, __FUNCTION__);
/*
* We need to re-check lp->nr_fs_variants because an arbitrarliy large
* number of shader variants (potentially all of them) could be
* pending for destruction on flush.
*/
for (i = 0; i < variants_to_cull || lp->nr_fs_instrs >= LP_MAX_SHADER_INSTRUCTIONS; i++) {
struct lp_fs_variant_list_item *item;
if (is_empty_list(&lp->fs_variants_list)) {
break;
}
item = last_elem(&lp->fs_variants_list);
assert(item);
assert(item->base);
llvmpipe_remove_shader_variant(lp, item->base);
}
}
/*
* Generate the new variant.
*/
t0 = os_time_get();
variant = generate_variant(lp, shader, key);
t1 = os_time_get();
dt = t1 - t0;
LP_COUNT_ADD(llvm_compile_time, dt);
LP_COUNT_ADD(nr_llvm_compiles, 2); /* emit vs. omit in/out test */
/* Put the new variant into the list */
if (variant) {
insert_at_head(&shader->variants, &variant->list_item_local);
insert_at_head(&lp->fs_variants_list, &variant->list_item_global);
lp->nr_fs_variants++;
lp->nr_fs_instrs += variant->nr_instrs;
shader->variants_cached++;
}
}
/* Bind this variant */
lp_setup_set_fs_variant(lp->setup, variant);
}
void
llvmpipe_init_fs_funcs(struct llvmpipe_context *llvmpipe)
{
llvmpipe->pipe.create_fs_state = llvmpipe_create_fs_state;
llvmpipe->pipe.bind_fs_state = llvmpipe_bind_fs_state;
llvmpipe->pipe.delete_fs_state = llvmpipe_delete_fs_state;
llvmpipe->pipe.set_constant_buffer = llvmpipe_set_constant_buffer;
llvmpipe->pipe.set_shader_buffers = llvmpipe_set_shader_buffers;
llvmpipe->pipe.set_shader_images = llvmpipe_set_shader_images;
}
|