1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
|
/*
* Copyright 2011-2013 Blender Foundation
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifdef WITH_OPENCL
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "clew.h"
#include "device.h"
#include "device_intern.h"
#include "buffers.h"
#include "util_foreach.h"
#include "util_logging.h"
#include "util_map.h"
#include "util_math.h"
#include "util_md5.h"
#include "util_opengl.h"
#include "util_path.h"
#include "util_time.h"
CCL_NAMESPACE_BEGIN
#define CL_MEM_PTR(p) ((cl_mem)(uintptr_t)(p))
/* Macro declarations used with split kernel */
/* Macro to enable/disable work-stealing */
#define __WORK_STEALING__
#define SPLIT_KERNEL_LOCAL_SIZE_X 64
#define SPLIT_KERNEL_LOCAL_SIZE_Y 1
/* This value may be tuned according to the scene we are rendering.
*
* Modifying PATH_ITER_INC_FACTOR value proportional to number of expected
* ray-bounces will improve performance.
*/
#define PATH_ITER_INC_FACTOR 8
/* When allocate global memory in chunks. We may not be able to
* allocate exactly "CL_DEVICE_MAX_MEM_ALLOC_SIZE" bytes in chunks;
* Since some bytes may be needed for aligning chunks of memory;
* This is the amount of memory that we dedicate for that purpose.
*/
#define DATA_ALLOCATION_MEM_FACTOR 5000000 //5MB
struct OpenCLPlatformDevice {
OpenCLPlatformDevice(cl_platform_id platform_id,
const string& platform_name,
cl_device_id device_id,
cl_device_type device_type,
const string& device_name)
: platform_id(platform_id),
platform_name(platform_name),
device_id(device_id),
device_type(device_type),
device_name(device_name) {}
cl_platform_id platform_id;
string platform_name;
cl_device_id device_id;
cl_device_type device_type;
string device_name;
};
namespace {
cl_device_type opencl_device_type()
{
char *device = getenv("CYCLES_OPENCL_TEST");
if(device) {
if(strcmp(device, "NONE") == 0)
return 0;
if(strcmp(device, "ALL") == 0)
return CL_DEVICE_TYPE_ALL;
else if(strcmp(device, "DEFAULT") == 0)
return CL_DEVICE_TYPE_DEFAULT;
else if(strcmp(device, "CPU") == 0)
return CL_DEVICE_TYPE_CPU;
else if(strcmp(device, "GPU") == 0)
return CL_DEVICE_TYPE_GPU;
else if(strcmp(device, "ACCELERATOR") == 0)
return CL_DEVICE_TYPE_ACCELERATOR;
}
return CL_DEVICE_TYPE_ALL;
}
bool opencl_kernel_use_debug()
{
return (getenv("CYCLES_OPENCL_DEBUG") != NULL);
}
bool opencl_kernel_use_advanced_shading(const string& platform)
{
/* keep this in sync with kernel_types.h! */
if(platform == "NVIDIA CUDA")
return true;
else if(platform == "Apple")
return true;
else if(platform == "AMD Accelerated Parallel Processing")
return true;
else if(platform == "Intel(R) OpenCL")
return true;
/* Make sure officially unsupported OpenCL platforms
* does not set up to use advanced shading.
*/
return false;
}
bool opencl_kernel_use_split(const string& platform_name,
const cl_device_type device_type)
{
if(getenv("CYCLES_OPENCL_SPLIT_KERNEL_TEST") != NULL) {
VLOG(1) << "Forcing split kernel to use.";
return true;
}
if(getenv("CYCLES_OPENCL_MEGA_KERNEL_TEST") != NULL) {
VLOG(1) << "Forcing mega kernel to use.";
return false;
}
/* TODO(sergey): Replace string lookups with more enum-like API,
* similar to device/vendor checks blender's gpu.
*/
if(platform_name == "AMD Accelerated Parallel Processing" &&
device_type == CL_DEVICE_TYPE_GPU)
{
return true;
}
return false;
}
bool opencl_device_supported(const string& platform_name,
const cl_device_id device_id)
{
cl_device_type device_type;
clGetDeviceInfo(device_id,
CL_DEVICE_TYPE,
sizeof(cl_device_type),
&device_type,
NULL);
if(platform_name == "AMD Accelerated Parallel Processing" &&
device_type == CL_DEVICE_TYPE_GPU)
{
return true;
}
if(platform_name == "Apple" && device_type == CL_DEVICE_TYPE_GPU) {
return true;
}
return false;
}
bool opencl_platform_version_check(cl_platform_id platform,
string *error = NULL)
{
const int req_major = 1, req_minor = 1;
int major, minor;
char version[256];
clGetPlatformInfo(platform,
CL_PLATFORM_VERSION,
sizeof(version),
&version,
NULL);
if(sscanf(version, "OpenCL %d.%d", &major, &minor) < 2) {
if(error != NULL) {
*error = string_printf("OpenCL: failed to parse platform version string (%s).", version);
}
return false;
}
if(!((major == req_major && minor >= req_minor) || (major > req_major))) {
if(error != NULL) {
*error = string_printf("OpenCL: platform version 1.1 or later required, found %d.%d", major, minor);
}
return false;
}
if(error != NULL) {
*error = "";
}
return true;
}
bool opencl_device_version_check(cl_device_id device,
string *error = NULL)
{
const int req_major = 1, req_minor = 1;
int major, minor;
char version[256];
clGetDeviceInfo(device,
CL_DEVICE_OPENCL_C_VERSION,
sizeof(version),
&version,
NULL);
if(sscanf(version, "OpenCL C %d.%d", &major, &minor) < 2) {
if(error != NULL) {
*error = string_printf("OpenCL: failed to parse OpenCL C version string (%s).", version);
}
return false;
}
if(!((major == req_major && minor >= req_minor) || (major > req_major))) {
if(error != NULL) {
*error = string_printf("OpenCL: C version 1.1 or later required, found %d.%d", major, minor);
}
return false;
}
if(error != NULL) {
*error = "";
}
return true;
}
void opencl_get_usable_devices(vector<OpenCLPlatformDevice> *usable_devices)
{
const bool force_all_platforms =
(getenv("CYCLES_OPENCL_TEST") != NULL) ||
(getenv("CYCLES_OPENCL_SPLIT_KERNEL_TEST")) != NULL;
const cl_device_type device_type = opencl_device_type();
static bool first_time = true;
#define FIRST_VLOG(severity) if(first_time) VLOG(severity)
usable_devices->clear();
if(device_type == 0) {
FIRST_VLOG(2) << "OpenCL devices are forced to be disabled.";
first_time = false;
return;
}
vector<cl_device_id> device_ids;
cl_uint num_devices = 0;
vector<cl_platform_id> platform_ids;
cl_uint num_platforms = 0;
/* Get devices. */
if(clGetPlatformIDs(0, NULL, &num_platforms) != CL_SUCCESS ||
num_platforms == 0)
{
FIRST_VLOG(2) << "No OpenCL platforms were found.";
first_time = false;
return;
}
platform_ids.resize(num_platforms);
if(clGetPlatformIDs(num_platforms, &platform_ids[0], NULL) != CL_SUCCESS) {
FIRST_VLOG(2) << "Failed to fetch platform IDs from the driver..";
first_time = false;
return;
}
/* Devices are numbered consecutively across platforms. */
for(int platform = 0; platform < num_platforms; platform++) {
cl_platform_id platform_id = platform_ids[platform];
char pname[256];
if(clGetPlatformInfo(platform_id,
CL_PLATFORM_NAME,
sizeof(pname),
&pname,
NULL) != CL_SUCCESS)
{
FIRST_VLOG(2) << "Failed to get platform name, ignoring.";
continue;
}
string platform_name = pname;
FIRST_VLOG(2) << "Enumerating devices for platform "
<< platform_name << ".";
if(!opencl_platform_version_check(platform_id)) {
FIRST_VLOG(2) << "Ignoring platform " << platform_name
<< " due to too old compiler version.";
continue;
}
num_devices = 0;
if(clGetDeviceIDs(platform_id,
device_type,
0,
NULL,
&num_devices) != CL_SUCCESS || num_devices == 0)
{
FIRST_VLOG(2) << "Ignoring platform " << platform_name
<< ", failed to fetch number of devices.";
continue;
}
device_ids.resize(num_devices);
if(clGetDeviceIDs(platform_id,
device_type,
num_devices,
&device_ids[0],
NULL) != CL_SUCCESS)
{
FIRST_VLOG(2) << "Ignoring platform " << platform_name
<< ", failed to fetch devices list.";
continue;
}
for(int num = 0; num < num_devices; num++) {
cl_device_id device_id = device_ids[num];
char device_name[1024] = "\0";
if(clGetDeviceInfo(device_id,
CL_DEVICE_NAME,
sizeof(device_name),
&device_name,
NULL) != CL_SUCCESS)
{
FIRST_VLOG(2) << "Failed to fetch device name, ignoring.";
continue;
}
if(!opencl_device_version_check(device_id)) {
FIRST_VLOG(2) << "Ignoring device " << device_name
<< " due to old compiler version.";
continue;
}
if(force_all_platforms ||
opencl_device_supported(platform_name, device_id))
{
cl_device_type device_type;
if(clGetDeviceInfo(device_id,
CL_DEVICE_TYPE,
sizeof(cl_device_type),
&device_type,
NULL) != CL_SUCCESS)
{
FIRST_VLOG(2) << "Ignoring device " << device_name
<< ", failed to fetch device type.";
continue;
}
FIRST_VLOG(2) << "Adding new device " << device_name << ".";
usable_devices->push_back(OpenCLPlatformDevice(platform_id,
platform_name,
device_id,
device_type,
device_name));
}
else {
FIRST_VLOG(2) << "Ignoring device " << device_name
<< ", not officially supported yet.";
}
}
}
first_time = false;
}
} /* namespace */
/* Thread safe cache for contexts and programs.
*
* TODO(sergey): Make it more generous, so it can contain any type of program
* without hardcoding possible program types in the slot.
*/
class OpenCLCache
{
struct Slot
{
thread_mutex *mutex;
cl_context context;
/* cl_program for shader, bake, film_convert kernels (used in OpenCLDeviceBase) */
cl_program ocl_dev_base_program;
/* cl_program for megakernel (used in OpenCLDeviceMegaKernel) */
cl_program ocl_dev_megakernel_program;
Slot() : mutex(NULL),
context(NULL),
ocl_dev_base_program(NULL),
ocl_dev_megakernel_program(NULL) {}
Slot(const Slot& rhs)
: mutex(rhs.mutex),
context(rhs.context),
ocl_dev_base_program(rhs.ocl_dev_base_program),
ocl_dev_megakernel_program(rhs.ocl_dev_megakernel_program)
{
/* copy can only happen in map insert, assert that */
assert(mutex == NULL);
}
~Slot()
{
delete mutex;
mutex = NULL;
}
};
/* key is combination of platform ID and device ID */
typedef pair<cl_platform_id, cl_device_id> PlatformDevicePair;
/* map of Slot objects */
typedef map<PlatformDevicePair, Slot> CacheMap;
CacheMap cache;
thread_mutex cache_lock;
/* lazy instantiate */
static OpenCLCache &global_instance()
{
static OpenCLCache instance;
return instance;
}
OpenCLCache()
{
}
~OpenCLCache()
{
/* Intel OpenCL bug raises SIGABRT due to pure virtual call
* so this is disabled. It's not necessary to free objects
* at process exit anyway.
* http://software.intel.com/en-us/forums/topic/370083#comments */
//flush();
}
/* lookup something in the cache. If this returns NULL, slot_locker
* will be holding a lock for the cache. slot_locker should refer to a
* default constructed thread_scoped_lock */
template<typename T>
static T get_something(cl_platform_id platform,
cl_device_id device,
T Slot::*member,
thread_scoped_lock& slot_locker)
{
assert(platform != NULL);
OpenCLCache& self = global_instance();
thread_scoped_lock cache_lock(self.cache_lock);
pair<CacheMap::iterator,bool> ins = self.cache.insert(
CacheMap::value_type(PlatformDevicePair(platform, device), Slot()));
Slot &slot = ins.first->second;
/* create slot lock only while holding cache lock */
if(!slot.mutex)
slot.mutex = new thread_mutex;
/* need to unlock cache before locking slot, to allow store to complete */
cache_lock.unlock();
/* lock the slot */
slot_locker = thread_scoped_lock(*slot.mutex);
/* If the thing isn't cached */
if(slot.*member == NULL) {
/* return with the caller's lock holder holding the slot lock */
return NULL;
}
/* the item was already cached, release the slot lock */
slot_locker.unlock();
return slot.*member;
}
/* store something in the cache. you MUST have tried to get the item before storing to it */
template<typename T>
static void store_something(cl_platform_id platform,
cl_device_id device,
T thing,
T Slot::*member,
thread_scoped_lock& slot_locker)
{
assert(platform != NULL);
assert(device != NULL);
assert(thing != NULL);
OpenCLCache &self = global_instance();
thread_scoped_lock cache_lock(self.cache_lock);
CacheMap::iterator i = self.cache.find(PlatformDevicePair(platform, device));
cache_lock.unlock();
Slot &slot = i->second;
/* sanity check */
assert(i != self.cache.end());
assert(slot.*member == NULL);
slot.*member = thing;
/* unlock the slot */
slot_locker.unlock();
}
public:
enum ProgramName {
OCL_DEV_BASE_PROGRAM,
OCL_DEV_MEGAKERNEL_PROGRAM,
};
/* see get_something comment */
static cl_context get_context(cl_platform_id platform,
cl_device_id device,
thread_scoped_lock& slot_locker)
{
cl_context context = get_something<cl_context>(platform,
device,
&Slot::context,
slot_locker);
if(!context)
return NULL;
/* caller is going to release it when done with it, so retain it */
cl_int ciErr = clRetainContext(context);
assert(ciErr == CL_SUCCESS);
(void)ciErr;
return context;
}
/* see get_something comment */
static cl_program get_program(cl_platform_id platform,
cl_device_id device,
ProgramName program_name,
thread_scoped_lock& slot_locker)
{
cl_program program = NULL;
switch(program_name) {
case OCL_DEV_BASE_PROGRAM:
/* Get program related to OpenCLDeviceBase */
program = get_something<cl_program>(platform,
device,
&Slot::ocl_dev_base_program,
slot_locker);
break;
case OCL_DEV_MEGAKERNEL_PROGRAM:
/* Get program related to megakernel */
program = get_something<cl_program>(platform,
device,
&Slot::ocl_dev_megakernel_program,
slot_locker);
break;
default:
assert(!"Invalid program name");
}
if(!program)
return NULL;
/* caller is going to release it when done with it, so retain it */
cl_int ciErr = clRetainProgram(program);
assert(ciErr == CL_SUCCESS);
(void)ciErr;
return program;
}
/* see store_something comment */
static void store_context(cl_platform_id platform,
cl_device_id device,
cl_context context,
thread_scoped_lock& slot_locker)
{
store_something<cl_context>(platform,
device,
context,
&Slot::context,
slot_locker);
/* increment reference count in OpenCL.
* The caller is going to release the object when done with it. */
cl_int ciErr = clRetainContext(context);
assert(ciErr == CL_SUCCESS);
(void)ciErr;
}
/* see store_something comment */
static void store_program(cl_platform_id platform,
cl_device_id device,
cl_program program,
ProgramName program_name,
thread_scoped_lock& slot_locker)
{
switch(program_name) {
case OCL_DEV_BASE_PROGRAM:
store_something<cl_program>(platform,
device,
program,
&Slot::ocl_dev_base_program,
slot_locker);
break;
case OCL_DEV_MEGAKERNEL_PROGRAM:
store_something<cl_program>(platform,
device,
program,
&Slot::ocl_dev_megakernel_program,
slot_locker);
break;
default:
assert(!"Invalid program name\n");
return;
}
/* Increment reference count in OpenCL.
* The caller is going to release the object when done with it.
*/
cl_int ciErr = clRetainProgram(program);
assert(ciErr == CL_SUCCESS);
(void)ciErr;
}
/* Discard all cached contexts and programs. */
static void flush()
{
OpenCLCache &self = global_instance();
thread_scoped_lock cache_lock(self.cache_lock);
foreach(CacheMap::value_type &item, self.cache) {
if(item.second.ocl_dev_base_program != NULL)
clReleaseProgram(item.second.ocl_dev_base_program);
if(item.second.ocl_dev_megakernel_program != NULL)
clReleaseProgram(item.second.ocl_dev_megakernel_program);
if(item.second.context != NULL)
clReleaseContext(item.second.context);
}
self.cache.clear();
}
};
class OpenCLDeviceBase : public Device
{
public:
DedicatedTaskPool task_pool;
cl_context cxContext;
cl_command_queue cqCommandQueue;
cl_platform_id cpPlatform;
cl_device_id cdDevice;
cl_program cpProgram;
cl_kernel ckFilmConvertByteKernel;
cl_kernel ckFilmConvertHalfFloatKernel;
cl_kernel ckShaderKernel;
cl_kernel ckBakeKernel;
cl_int ciErr;
typedef map<string, device_vector<uchar>*> ConstMemMap;
typedef map<string, device_ptr> MemMap;
ConstMemMap const_mem_map;
MemMap mem_map;
device_ptr null_mem;
bool device_initialized;
string platform_name;
bool opencl_error(cl_int err)
{
if(err != CL_SUCCESS) {
string message = string_printf("OpenCL error (%d): %s", err, clewErrorString(err));
if(error_msg == "")
error_msg = message;
fprintf(stderr, "%s\n", message.c_str());
return true;
}
return false;
}
void opencl_error(const string& message)
{
if(error_msg == "")
error_msg = message;
fprintf(stderr, "%s\n", message.c_str());
}
#define opencl_assert(stmt) \
{ \
cl_int err = stmt; \
\
if(err != CL_SUCCESS) { \
string message = string_printf("OpenCL error: %s in %s", clewErrorString(err), #stmt); \
if(error_msg == "") \
error_msg = message; \
fprintf(stderr, "%s\n", message.c_str()); \
} \
} (void)0
void opencl_assert_err(cl_int err, const char* where)
{
if(err != CL_SUCCESS) {
string message = string_printf("OpenCL error (%d): %s in %s", err, clewErrorString(err), where);
if(error_msg == "")
error_msg = message;
fprintf(stderr, "%s\n", message.c_str());
#ifndef NDEBUG
abort();
#endif
}
}
OpenCLDeviceBase(DeviceInfo& info, Stats &stats, bool background_)
: Device(info, stats, background_)
{
cpPlatform = NULL;
cdDevice = NULL;
cxContext = NULL;
cqCommandQueue = NULL;
cpProgram = NULL;
ckFilmConvertByteKernel = NULL;
ckFilmConvertHalfFloatKernel = NULL;
ckShaderKernel = NULL;
ckBakeKernel = NULL;
null_mem = 0;
device_initialized = false;
vector<OpenCLPlatformDevice> usable_devices;
opencl_get_usable_devices(&usable_devices);
if(usable_devices.size() == 0) {
opencl_error("OpenCL: no devices found.");
return;
}
assert(info.num < usable_devices.size());
OpenCLPlatformDevice& platform_device = usable_devices[info.num];
cpPlatform = platform_device.platform_id;
cdDevice = platform_device.device_id;
platform_name = platform_device.platform_name;
VLOG(2) << "Creating new Cycles device for OpenCL platform "
<< platform_name << ", device "
<< platform_device.device_name << ".";
{
/* try to use cached context */
thread_scoped_lock cache_locker;
cxContext = OpenCLCache::get_context(cpPlatform, cdDevice, cache_locker);
if(cxContext == NULL) {
/* create context properties array to specify platform */
const cl_context_properties context_props[] = {
CL_CONTEXT_PLATFORM, (cl_context_properties)cpPlatform,
0, 0
};
/* create context */
cxContext = clCreateContext(context_props, 1, &cdDevice,
context_notify_callback, cdDevice, &ciErr);
if(opencl_error(ciErr)) {
opencl_error("OpenCL: clCreateContext failed");
return;
}
/* cache it */
OpenCLCache::store_context(cpPlatform, cdDevice, cxContext, cache_locker);
}
}
cqCommandQueue = clCreateCommandQueue(cxContext, cdDevice, 0, &ciErr);
if(opencl_error(ciErr))
return;
null_mem = (device_ptr)clCreateBuffer(cxContext, CL_MEM_READ_ONLY, 1, NULL, &ciErr);
if(opencl_error(ciErr))
return;
fprintf(stderr, "Device init success\n");
device_initialized = true;
}
static void CL_CALLBACK context_notify_callback(const char *err_info,
const void * /*private_info*/, size_t /*cb*/, void *user_data)
{
char name[256];
clGetDeviceInfo((cl_device_id)user_data, CL_DEVICE_NAME, sizeof(name), &name, NULL);
fprintf(stderr, "OpenCL error (%s): %s\n", name, err_info);
}
bool opencl_version_check()
{
string error;
if(!opencl_platform_version_check(cpPlatform, &error)) {
opencl_error(error);
return false;
}
if(!opencl_device_version_check(cdDevice, &error)) {
opencl_error(error);
return false;
}
return true;
}
bool load_binary(const string& /*kernel_path*/,
const string& clbin,
string custom_kernel_build_options,
cl_program *program,
const string *debug_src = NULL)
{
/* read binary into memory */
vector<uint8_t> binary;
if(!path_read_binary(clbin, binary)) {
opencl_error(string_printf("OpenCL failed to read cached binary %s.", clbin.c_str()));
return false;
}
/* create program */
cl_int status;
size_t size = binary.size();
const uint8_t *bytes = &binary[0];
*program = clCreateProgramWithBinary(cxContext, 1, &cdDevice,
&size, &bytes, &status, &ciErr);
if(opencl_error(status) || opencl_error(ciErr)) {
opencl_error(string_printf("OpenCL failed create program from cached binary %s.", clbin.c_str()));
return false;
}
if(!build_kernel(program, custom_kernel_build_options, debug_src))
return false;
return true;
}
bool save_binary(cl_program *program, const string& clbin)
{
size_t size = 0;
clGetProgramInfo(*program, CL_PROGRAM_BINARY_SIZES, sizeof(size_t), &size, NULL);
if(!size)
return false;
vector<uint8_t> binary(size);
uint8_t *bytes = &binary[0];
clGetProgramInfo(*program, CL_PROGRAM_BINARIES, sizeof(uint8_t*), &bytes, NULL);
if(!path_write_binary(clbin, binary)) {
opencl_error(string_printf("OpenCL failed to write cached binary %s.", clbin.c_str()));
return false;
}
return true;
}
bool build_kernel(cl_program *kernel_program,
string custom_kernel_build_options,
const string *debug_src = NULL)
{
string build_options;
build_options = kernel_build_options(debug_src) + custom_kernel_build_options;
ciErr = clBuildProgram(*kernel_program, 0, NULL, build_options.c_str(), NULL, NULL);
/* show warnings even if build is successful */
size_t ret_val_size = 0;
clGetProgramBuildInfo(*kernel_program, cdDevice, CL_PROGRAM_BUILD_LOG, 0, NULL, &ret_val_size);
if(ret_val_size > 1) {
vector<char> build_log(ret_val_size + 1);
clGetProgramBuildInfo(*kernel_program, cdDevice, CL_PROGRAM_BUILD_LOG, ret_val_size, &build_log[0], NULL);
build_log[ret_val_size] = '\0';
/* Skip meaningless empty output from the NVidia compiler. */
if(!(ret_val_size == 2 && build_log[0] == '\n')) {
fprintf(stderr, "OpenCL kernel build output:\n");
fprintf(stderr, "%s\n", &build_log[0]);
}
}
if(ciErr != CL_SUCCESS) {
opencl_error("OpenCL build failed: errors in console");
return false;
}
return true;
}
bool compile_kernel(const string& kernel_path,
string source,
string custom_kernel_build_options,
cl_program *kernel_program,
const string *debug_src = NULL)
{
/* we compile kernels consisting of many files. unfortunately opencl
* kernel caches do not seem to recognize changes in included files.
* so we force recompile on changes by adding the md5 hash of all files */
source = path_source_replace_includes(source, kernel_path);
if(debug_src)
path_write_text(*debug_src, source);
size_t source_len = source.size();
const char *source_str = source.c_str();
*kernel_program = clCreateProgramWithSource(cxContext, 1, &source_str, &source_len, &ciErr);
if(opencl_error(ciErr))
return false;
double starttime = time_dt();
printf("Compiling OpenCL kernel ...\n");
/* TODO(sergey): Report which kernel is being compiled
* as well (megakernel or which of split kernels etc..).
*/
printf("Build flags: %s\n", custom_kernel_build_options.c_str());
if(!build_kernel(kernel_program, custom_kernel_build_options, debug_src))
return false;
printf("Kernel compilation finished in %.2lfs.\n", time_dt() - starttime);
return true;
}
string device_md5_hash(string kernel_custom_build_options = "")
{
MD5Hash md5;
char version[256], driver[256], name[256], vendor[256];
clGetPlatformInfo(cpPlatform, CL_PLATFORM_VENDOR, sizeof(vendor), &vendor, NULL);
clGetDeviceInfo(cdDevice, CL_DEVICE_VERSION, sizeof(version), &version, NULL);
clGetDeviceInfo(cdDevice, CL_DEVICE_NAME, sizeof(name), &name, NULL);
clGetDeviceInfo(cdDevice, CL_DRIVER_VERSION, sizeof(driver), &driver, NULL);
md5.append((uint8_t*)vendor, strlen(vendor));
md5.append((uint8_t*)version, strlen(version));
md5.append((uint8_t*)name, strlen(name));
md5.append((uint8_t*)driver, strlen(driver));
string options = kernel_build_options();
options += kernel_custom_build_options;
md5.append((uint8_t*)options.c_str(), options.size());
return md5.get_hex();
}
bool load_kernels(const DeviceRequestedFeatures& requested_features)
{
/* Verify if device was initialized. */
if(!device_initialized) {
fprintf(stderr, "OpenCL: failed to initialize device.\n");
return false;
}
/* Try to use cached kernel. */
thread_scoped_lock cache_locker;
cpProgram = load_cached_kernel(requested_features,
OpenCLCache::OCL_DEV_BASE_PROGRAM,
cache_locker);
if(!cpProgram) {
VLOG(2) << "No cached OpenCL kernel.";
/* Verify we have right opencl version. */
if(!opencl_version_check())
return false;
string build_flags = build_options_for_base_program(requested_features);
/* Calculate md5 hashes to detect changes. */
string kernel_path = path_get("kernel");
string kernel_md5 = path_files_md5_hash(kernel_path);
string device_md5 = device_md5_hash(build_flags);
/* Path to cached binary.
*
* TODO(sergey): Seems we could de-duplicate all this string_printf()
* calls with some utility function which will give file name for a
* given hashes..
*/
string clbin = string_printf("cycles_kernel_%s_%s.clbin",
device_md5.c_str(),
kernel_md5.c_str());
clbin = path_user_get(path_join("cache", clbin));
/* path to preprocessed source for debugging */
string clsrc, *debug_src = NULL;
if(opencl_kernel_use_debug()) {
clsrc = string_printf("cycles_kernel_%s_%s.cl",
device_md5.c_str(),
kernel_md5.c_str());
clsrc = path_user_get(path_join("cache", clsrc));
debug_src = &clsrc;
}
/* If binary kernel exists already, try use it. */
if(path_exists(clbin) && load_binary(kernel_path,
clbin,
build_flags,
&cpProgram)) {
/* Kernel loaded from binary, nothing to do. */
VLOG(2) << "Loaded kernel from " << clbin << ".";
}
else {
VLOG(2) << "Kernel file " << clbin << " either doesn't exist or failed to be loaded by driver.";
string init_kernel_source = "#include \"kernels/opencl/kernel.cl\" // " + kernel_md5 + "\n";
/* If does not exist or loading binary failed, compile kernel. */
if(!compile_kernel(kernel_path,
init_kernel_source,
build_flags,
&cpProgram,
debug_src))
{
return false;
}
/* Save binary for reuse. */
if(!save_binary(&cpProgram, clbin)) {
return false;
}
}
/* Cache the program. */
store_cached_kernel(cpPlatform,
cdDevice,
cpProgram,
OpenCLCache::OCL_DEV_BASE_PROGRAM,
cache_locker);
}
else {
VLOG(2) << "Found cached OpenCL kernel.";
}
/* Find kernels. */
#define FIND_KERNEL(kernel_var, kernel_name) \
do { \
kernel_var = clCreateKernel(cpProgram, "kernel_ocl_" kernel_name, &ciErr); \
if(opencl_error(ciErr)) \
return false; \
} while(0)
FIND_KERNEL(ckFilmConvertByteKernel, "convert_to_byte");
FIND_KERNEL(ckFilmConvertHalfFloatKernel, "convert_to_half_float");
FIND_KERNEL(ckShaderKernel, "shader");
FIND_KERNEL(ckBakeKernel, "bake");
#undef FIND_KERNEL
return true;
}
~OpenCLDeviceBase()
{
task_pool.stop();
if(null_mem)
clReleaseMemObject(CL_MEM_PTR(null_mem));
ConstMemMap::iterator mt;
for(mt = const_mem_map.begin(); mt != const_mem_map.end(); mt++) {
mem_free(*(mt->second));
delete mt->second;
}
if(ckFilmConvertByteKernel)
clReleaseKernel(ckFilmConvertByteKernel);
if(ckFilmConvertHalfFloatKernel)
clReleaseKernel(ckFilmConvertHalfFloatKernel);
if(ckShaderKernel)
clReleaseKernel(ckShaderKernel);
if(ckBakeKernel)
clReleaseKernel(ckBakeKernel);
if(cpProgram)
clReleaseProgram(cpProgram);
if(cqCommandQueue)
clReleaseCommandQueue(cqCommandQueue);
if(cxContext)
clReleaseContext(cxContext);
}
void mem_alloc(device_memory& mem, MemoryType type)
{
size_t size = mem.memory_size();
cl_mem_flags mem_flag;
void *mem_ptr = NULL;
if(type == MEM_READ_ONLY)
mem_flag = CL_MEM_READ_ONLY;
else if(type == MEM_WRITE_ONLY)
mem_flag = CL_MEM_WRITE_ONLY;
else
mem_flag = CL_MEM_READ_WRITE;
/* Zero-size allocation might be invoked by render, but not really
* supported by OpenCL. Using NULL as device pointer also doesn't really
* work for some reason, so for the time being we'll use special case
* will null_mem buffer.
*/
if(size != 0) {
mem.device_pointer = (device_ptr)clCreateBuffer(cxContext,
mem_flag,
size,
mem_ptr,
&ciErr);
opencl_assert_err(ciErr, "clCreateBuffer");
}
else {
mem.device_pointer = null_mem;
}
stats.mem_alloc(size);
mem.device_size = size;
}
void mem_copy_to(device_memory& mem)
{
/* this is blocking */
size_t size = mem.memory_size();
if(size != 0) {
opencl_assert(clEnqueueWriteBuffer(cqCommandQueue,
CL_MEM_PTR(mem.device_pointer),
CL_TRUE,
0,
size,
(void*)mem.data_pointer,
0,
NULL, NULL));
}
}
void mem_copy_from(device_memory& mem, int y, int w, int h, int elem)
{
size_t offset = elem*y*w;
size_t size = elem*w*h;
assert(size != 0);
opencl_assert(clEnqueueReadBuffer(cqCommandQueue,
CL_MEM_PTR(mem.device_pointer),
CL_TRUE,
offset,
size,
(uchar*)mem.data_pointer + offset,
0,
NULL, NULL));
}
void mem_zero(device_memory& mem)
{
if(mem.device_pointer) {
memset((void*)mem.data_pointer, 0, mem.memory_size());
mem_copy_to(mem);
}
}
void mem_free(device_memory& mem)
{
if(mem.device_pointer) {
if(mem.device_pointer != null_mem) {
opencl_assert(clReleaseMemObject(CL_MEM_PTR(mem.device_pointer)));
}
mem.device_pointer = 0;
stats.mem_free(mem.device_size);
mem.device_size = 0;
}
}
void const_copy_to(const char *name, void *host, size_t size)
{
ConstMemMap::iterator i = const_mem_map.find(name);
if(i == const_mem_map.end()) {
device_vector<uchar> *data = new device_vector<uchar>();
data->copy((uchar*)host, size);
mem_alloc(*data, MEM_READ_ONLY);
i = const_mem_map.insert(ConstMemMap::value_type(name, data)).first;
}
else {
device_vector<uchar> *data = i->second;
data->copy((uchar*)host, size);
}
mem_copy_to(*i->second);
}
void tex_alloc(const char *name,
device_memory& mem,
InterpolationType /*interpolation*/,
ExtensionType /*extension*/)
{
VLOG(1) << "Texture allocate: " << name << ", " << mem.memory_size() << " bytes.";
mem_alloc(mem, MEM_READ_ONLY);
mem_copy_to(mem);
assert(mem_map.find(name) == mem_map.end());
mem_map.insert(MemMap::value_type(name, mem.device_pointer));
}
void tex_free(device_memory& mem)
{
if(mem.device_pointer) {
foreach(const MemMap::value_type& value, mem_map) {
if(value.second == mem.device_pointer) {
mem_map.erase(value.first);
break;
}
}
mem_free(mem);
}
}
size_t global_size_round_up(int group_size, int global_size)
{
int r = global_size % group_size;
return global_size + ((r == 0)? 0: group_size - r);
}
void enqueue_kernel(cl_kernel kernel, size_t w, size_t h)
{
size_t workgroup_size, max_work_items[3];
clGetKernelWorkGroupInfo(kernel, cdDevice,
CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &workgroup_size, NULL);
clGetDeviceInfo(cdDevice,
CL_DEVICE_MAX_WORK_ITEM_SIZES, sizeof(size_t)*3, max_work_items, NULL);
/* try to divide evenly over 2 dimensions */
size_t sqrt_workgroup_size = max((size_t)sqrt((double)workgroup_size), 1);
size_t local_size[2] = {sqrt_workgroup_size, sqrt_workgroup_size};
/* some implementations have max size 1 on 2nd dimension */
if(local_size[1] > max_work_items[1]) {
local_size[0] = workgroup_size/max_work_items[1];
local_size[1] = max_work_items[1];
}
size_t global_size[2] = {global_size_round_up(local_size[0], w), global_size_round_up(local_size[1], h)};
/* run kernel */
opencl_assert(clEnqueueNDRangeKernel(cqCommandQueue, kernel, 2, NULL, global_size, NULL, 0, NULL, NULL));
opencl_assert(clFlush(cqCommandQueue));
}
void set_kernel_arg_mem(cl_kernel kernel, cl_uint *narg, const char *name)
{
cl_mem ptr;
MemMap::iterator i = mem_map.find(name);
if(i != mem_map.end()) {
ptr = CL_MEM_PTR(i->second);
}
else {
/* work around NULL not working, even though the spec says otherwise */
ptr = CL_MEM_PTR(null_mem);
}
opencl_assert(clSetKernelArg(kernel, (*narg)++, sizeof(ptr), (void*)&ptr));
}
void film_convert(DeviceTask& task, device_ptr buffer, device_ptr rgba_byte, device_ptr rgba_half)
{
/* cast arguments to cl types */
cl_mem d_data = CL_MEM_PTR(const_mem_map["__data"]->device_pointer);
cl_mem d_rgba = (rgba_byte)? CL_MEM_PTR(rgba_byte): CL_MEM_PTR(rgba_half);
cl_mem d_buffer = CL_MEM_PTR(buffer);
cl_int d_x = task.x;
cl_int d_y = task.y;
cl_int d_w = task.w;
cl_int d_h = task.h;
cl_float d_sample_scale = 1.0f/(task.sample + 1);
cl_int d_offset = task.offset;
cl_int d_stride = task.stride;
cl_kernel ckFilmConvertKernel = (rgba_byte)? ckFilmConvertByteKernel: ckFilmConvertHalfFloatKernel;
cl_uint start_arg_index =
kernel_set_args(ckFilmConvertKernel,
0,
d_data,
d_rgba,
d_buffer);
#define KERNEL_TEX(type, ttype, name) \
set_kernel_arg_mem(ckFilmConvertKernel, &start_arg_index, #name);
#include "kernel_textures.h"
#undef KERNEL_TEX
start_arg_index += kernel_set_args(ckFilmConvertKernel,
start_arg_index,
d_sample_scale,
d_x,
d_y,
d_w,
d_h,
d_offset,
d_stride);
enqueue_kernel(ckFilmConvertKernel, d_w, d_h);
}
void shader(DeviceTask& task)
{
/* cast arguments to cl types */
cl_mem d_data = CL_MEM_PTR(const_mem_map["__data"]->device_pointer);
cl_mem d_input = CL_MEM_PTR(task.shader_input);
cl_mem d_output = CL_MEM_PTR(task.shader_output);
cl_int d_shader_eval_type = task.shader_eval_type;
cl_int d_shader_x = task.shader_x;
cl_int d_shader_w = task.shader_w;
cl_int d_offset = task.offset;
cl_kernel kernel;
if(task.shader_eval_type >= SHADER_EVAL_BAKE)
kernel = ckBakeKernel;
else
kernel = ckShaderKernel;
for(int sample = 0; sample < task.num_samples; sample++) {
if(task.get_cancel())
break;
cl_int d_sample = sample;
cl_uint start_arg_index =
kernel_set_args(kernel,
0,
d_data,
d_input,
d_output);
#define KERNEL_TEX(type, ttype, name) \
set_kernel_arg_mem(kernel, &start_arg_index, #name);
#include "kernel_textures.h"
#undef KERNEL_TEX
start_arg_index += kernel_set_args(kernel,
start_arg_index,
d_shader_eval_type,
d_shader_x,
d_shader_w,
d_offset,
d_sample);
enqueue_kernel(kernel, task.shader_w, 1);
task.update_progress(NULL);
}
}
class OpenCLDeviceTask : public DeviceTask {
public:
OpenCLDeviceTask(OpenCLDeviceBase *device, DeviceTask& task)
: DeviceTask(task)
{
run = function_bind(&OpenCLDeviceBase::thread_run,
device,
this);
}
};
int get_split_task_count(DeviceTask& /*task*/)
{
return 1;
}
void task_add(DeviceTask& task)
{
task_pool.push(new OpenCLDeviceTask(this, task));
}
void task_wait()
{
task_pool.wait();
}
void task_cancel()
{
task_pool.cancel();
}
virtual void thread_run(DeviceTask * /*task*/) = 0;
protected:
string kernel_build_options(const string *debug_src = NULL)
{
string build_options = "-cl-fast-relaxed-math ";
if(platform_name == "NVIDIA CUDA") {
build_options += "-D__KERNEL_OPENCL_NVIDIA__ "
"-cl-nv-maxrregcount=32 "
"-cl-nv-verbose ";
uint compute_capability_major, compute_capability_minor;
clGetDeviceInfo(cdDevice, CL_DEVICE_COMPUTE_CAPABILITY_MAJOR_NV,
sizeof(cl_uint), &compute_capability_major, NULL);
clGetDeviceInfo(cdDevice, CL_DEVICE_COMPUTE_CAPABILITY_MINOR_NV,
sizeof(cl_uint), &compute_capability_minor, NULL);
build_options += string_printf("-D__COMPUTE_CAPABILITY__=%u ",
compute_capability_major * 100 +
compute_capability_minor * 10);
}
else if(platform_name == "Apple")
build_options += "-D__KERNEL_OPENCL_APPLE__ ";
else if(platform_name == "AMD Accelerated Parallel Processing")
build_options += "-D__KERNEL_OPENCL_AMD__ ";
else if(platform_name == "Intel(R) OpenCL") {
build_options += "-D__KERNEL_OPENCL_INTEL_CPU__ ";
/* Options for gdb source level kernel debugging.
* this segfaults on linux currently.
*/
if(opencl_kernel_use_debug() && debug_src)
build_options += "-g -s \"" + *debug_src + "\" ";
}
if(opencl_kernel_use_debug())
build_options += "-D__KERNEL_OPENCL_DEBUG__ ";
#ifdef WITH_CYCLES_DEBUG
build_options += "-D__KERNEL_DEBUG__ ";
#endif
return build_options;
}
class ArgumentWrapper {
public:
ArgumentWrapper() : size(0), pointer(NULL) {}
template <typename T>
ArgumentWrapper(T& argument) : size(sizeof(argument)),
pointer(&argument) { }
size_t size;
void *pointer;
};
/* TODO(sergey): In the future we can use variadic templates, once
* C++0x is allowed. Should allow to clean this up a bit.
*/
int kernel_set_args(cl_kernel kernel,
int start_argument_index,
const ArgumentWrapper& arg1 = ArgumentWrapper(),
const ArgumentWrapper& arg2 = ArgumentWrapper(),
const ArgumentWrapper& arg3 = ArgumentWrapper(),
const ArgumentWrapper& arg4 = ArgumentWrapper(),
const ArgumentWrapper& arg5 = ArgumentWrapper(),
const ArgumentWrapper& arg6 = ArgumentWrapper(),
const ArgumentWrapper& arg7 = ArgumentWrapper(),
const ArgumentWrapper& arg8 = ArgumentWrapper(),
const ArgumentWrapper& arg9 = ArgumentWrapper(),
const ArgumentWrapper& arg10 = ArgumentWrapper(),
const ArgumentWrapper& arg11 = ArgumentWrapper(),
const ArgumentWrapper& arg12 = ArgumentWrapper(),
const ArgumentWrapper& arg13 = ArgumentWrapper(),
const ArgumentWrapper& arg14 = ArgumentWrapper(),
const ArgumentWrapper& arg15 = ArgumentWrapper(),
const ArgumentWrapper& arg16 = ArgumentWrapper(),
const ArgumentWrapper& arg17 = ArgumentWrapper(),
const ArgumentWrapper& arg18 = ArgumentWrapper(),
const ArgumentWrapper& arg19 = ArgumentWrapper(),
const ArgumentWrapper& arg20 = ArgumentWrapper(),
const ArgumentWrapper& arg21 = ArgumentWrapper(),
const ArgumentWrapper& arg22 = ArgumentWrapper(),
const ArgumentWrapper& arg23 = ArgumentWrapper(),
const ArgumentWrapper& arg24 = ArgumentWrapper(),
const ArgumentWrapper& arg25 = ArgumentWrapper(),
const ArgumentWrapper& arg26 = ArgumentWrapper(),
const ArgumentWrapper& arg27 = ArgumentWrapper(),
const ArgumentWrapper& arg28 = ArgumentWrapper(),
const ArgumentWrapper& arg29 = ArgumentWrapper(),
const ArgumentWrapper& arg30 = ArgumentWrapper(),
const ArgumentWrapper& arg31 = ArgumentWrapper(),
const ArgumentWrapper& arg32 = ArgumentWrapper(),
const ArgumentWrapper& arg33 = ArgumentWrapper())
{
int current_arg_index = 0;
#define FAKE_VARARG_HANDLE_ARG(arg) \
do { \
if(arg.pointer != NULL) { \
opencl_assert(clSetKernelArg( \
kernel, \
start_argument_index + current_arg_index, \
arg.size, arg.pointer)); \
++current_arg_index; \
} \
else { \
return current_arg_index; \
} \
} while(false)
FAKE_VARARG_HANDLE_ARG(arg1);
FAKE_VARARG_HANDLE_ARG(arg2);
FAKE_VARARG_HANDLE_ARG(arg3);
FAKE_VARARG_HANDLE_ARG(arg4);
FAKE_VARARG_HANDLE_ARG(arg5);
FAKE_VARARG_HANDLE_ARG(arg6);
FAKE_VARARG_HANDLE_ARG(arg7);
FAKE_VARARG_HANDLE_ARG(arg8);
FAKE_VARARG_HANDLE_ARG(arg9);
FAKE_VARARG_HANDLE_ARG(arg10);
FAKE_VARARG_HANDLE_ARG(arg11);
FAKE_VARARG_HANDLE_ARG(arg12);
FAKE_VARARG_HANDLE_ARG(arg13);
FAKE_VARARG_HANDLE_ARG(arg14);
FAKE_VARARG_HANDLE_ARG(arg15);
FAKE_VARARG_HANDLE_ARG(arg16);
FAKE_VARARG_HANDLE_ARG(arg17);
FAKE_VARARG_HANDLE_ARG(arg18);
FAKE_VARARG_HANDLE_ARG(arg19);
FAKE_VARARG_HANDLE_ARG(arg20);
FAKE_VARARG_HANDLE_ARG(arg21);
FAKE_VARARG_HANDLE_ARG(arg22);
FAKE_VARARG_HANDLE_ARG(arg23);
FAKE_VARARG_HANDLE_ARG(arg24);
FAKE_VARARG_HANDLE_ARG(arg25);
FAKE_VARARG_HANDLE_ARG(arg26);
FAKE_VARARG_HANDLE_ARG(arg27);
FAKE_VARARG_HANDLE_ARG(arg28);
FAKE_VARARG_HANDLE_ARG(arg29);
FAKE_VARARG_HANDLE_ARG(arg30);
FAKE_VARARG_HANDLE_ARG(arg31);
FAKE_VARARG_HANDLE_ARG(arg32);
FAKE_VARARG_HANDLE_ARG(arg33);
#undef FAKE_VARARG_HANDLE_ARG
return current_arg_index;
}
inline void release_kernel_safe(cl_kernel kernel)
{
if(kernel) {
clReleaseKernel(kernel);
}
}
inline void release_mem_object_safe(cl_mem mem)
{
if(mem != NULL) {
clReleaseMemObject(mem);
}
}
inline void release_program_safe(cl_program program)
{
if(program) {
clReleaseProgram(program);
}
}
string build_options_from_requested_features(
const DeviceRequestedFeatures& requested_features)
{
string build_options = "";
if(requested_features.experimental) {
build_options += "-D__KERNEL_EXPERIMENTAL__ ";
}
build_options += "-D__NODES_MAX_GROUP__=" +
string_printf("%d", requested_features.max_nodes_group);
build_options += " -D__NODES_FEATURES__=" +
string_printf("%d", requested_features.nodes_features);
build_options += string_printf(" -D__MAX_CLOSURE__=%d",
requested_features.max_closure);
if(!requested_features.use_hair) {
build_options += " -D__NO_HAIR__";
}
if(!requested_features.use_object_motion) {
build_options += " -D__NO_OBJECT_MOTION__";
}
if(!requested_features.use_camera_motion) {
build_options += " -D__NO_CAMERA_MOTION__";
}
if(!requested_features.use_baking) {
build_options += " -D__NO_BAKING__";
}
return build_options;
}
/* ** Those guys are for workign around some compiler-specific bugs ** */
virtual cl_program load_cached_kernel(
const DeviceRequestedFeatures& /*requested_features*/,
OpenCLCache::ProgramName program_name,
thread_scoped_lock& cache_locker)
{
return OpenCLCache::get_program(cpPlatform,
cdDevice,
program_name,
cache_locker);
}
virtual void store_cached_kernel(cl_platform_id platform,
cl_device_id device,
cl_program program,
OpenCLCache::ProgramName program_name,
thread_scoped_lock& cache_locker)
{
OpenCLCache::store_program(platform,
device,
program,
program_name,
cache_locker);
}
virtual string build_options_for_base_program(
const DeviceRequestedFeatures& /*requested_features*/)
{
/* TODO(sergey): By default we compile all features, meaning
* mega kernel is not getting feature-based optimizations.
*
* Ideally we need always compile kernel with as less features
* enabled as possible to keep performance at it's max.
*/
return "";
}
};
class OpenCLDeviceMegaKernel : public OpenCLDeviceBase
{
public:
cl_kernel ckPathTraceKernel;
cl_program path_trace_program;
OpenCLDeviceMegaKernel(DeviceInfo& info, Stats &stats, bool background_)
: OpenCLDeviceBase(info, stats, background_)
{
ckPathTraceKernel = NULL;
path_trace_program = NULL;
}
bool load_kernels(const DeviceRequestedFeatures& requested_features)
{
/* Get Shader, bake and film convert kernels.
* It'll also do verification of OpenCL actually initialized.
*/
if(!OpenCLDeviceBase::load_kernels(requested_features)) {
return false;
}
/* Try to use cached kernel. */
thread_scoped_lock cache_locker;
path_trace_program = OpenCLCache::get_program(cpPlatform,
cdDevice,
OpenCLCache::OCL_DEV_MEGAKERNEL_PROGRAM,
cache_locker);
if(!path_trace_program) {
/* Verify we have right opencl version. */
if(!opencl_version_check())
return false;
/* Calculate md5 hash to detect changes. */
string kernel_path = path_get("kernel");
string kernel_md5 = path_files_md5_hash(kernel_path);
string custom_kernel_build_options = "-D__COMPILE_ONLY_MEGAKERNEL__ ";
string device_md5 = device_md5_hash(custom_kernel_build_options);
/* Path to cached binary. */
string clbin = string_printf("cycles_kernel_%s_%s.clbin",
device_md5.c_str(),
kernel_md5.c_str());
clbin = path_user_get(path_join("cache", clbin));
/* Path to preprocessed source for debugging. */
string clsrc, *debug_src = NULL;
if(opencl_kernel_use_debug()) {
clsrc = string_printf("cycles_kernel_%s_%s.cl",
device_md5.c_str(),
kernel_md5.c_str());
clsrc = path_user_get(path_join("cache", clsrc));
debug_src = &clsrc;
}
/* If exists already, try use it. */
if(path_exists(clbin) && load_binary(kernel_path,
clbin,
custom_kernel_build_options,
&path_trace_program,
debug_src)) {
/* Kernel loaded from binary, nothing to do. */
}
else {
string init_kernel_source = "#include \"kernels/opencl/kernel.cl\" // " +
kernel_md5 + "\n";
/* If does not exist or loading binary failed, compile kernel. */
if(!compile_kernel(kernel_path,
init_kernel_source,
custom_kernel_build_options,
&path_trace_program,
debug_src))
{
return false;
}
/* Save binary for reuse. */
if(!save_binary(&path_trace_program, clbin)) {
return false;
}
}
/* Cache the program. */
OpenCLCache::store_program(cpPlatform,
cdDevice,
path_trace_program,
OpenCLCache::OCL_DEV_MEGAKERNEL_PROGRAM,
cache_locker);
}
/* Find kernels. */
ckPathTraceKernel = clCreateKernel(path_trace_program,
"kernel_ocl_path_trace",
&ciErr);
if(opencl_error(ciErr))
return false;
return true;
}
~OpenCLDeviceMegaKernel()
{
task_pool.stop();
release_kernel_safe(ckPathTraceKernel);
release_program_safe(path_trace_program);
}
void path_trace(RenderTile& rtile, int sample)
{
/* Cast arguments to cl types. */
cl_mem d_data = CL_MEM_PTR(const_mem_map["__data"]->device_pointer);
cl_mem d_buffer = CL_MEM_PTR(rtile.buffer);
cl_mem d_rng_state = CL_MEM_PTR(rtile.rng_state);
cl_int d_x = rtile.x;
cl_int d_y = rtile.y;
cl_int d_w = rtile.w;
cl_int d_h = rtile.h;
cl_int d_offset = rtile.offset;
cl_int d_stride = rtile.stride;
/* Sample arguments. */
cl_int d_sample = sample;
cl_uint start_arg_index =
kernel_set_args(ckPathTraceKernel,
0,
d_data,
d_buffer,
d_rng_state);
#define KERNEL_TEX(type, ttype, name) \
set_kernel_arg_mem(ckPathTraceKernel, &start_arg_index, #name);
#include "kernel_textures.h"
#undef KERNEL_TEX
start_arg_index += kernel_set_args(ckPathTraceKernel,
start_arg_index,
d_sample,
d_x,
d_y,
d_w,
d_h,
d_offset,
d_stride);
enqueue_kernel(ckPathTraceKernel, d_w, d_h);
}
void thread_run(DeviceTask *task)
{
if(task->type == DeviceTask::FILM_CONVERT) {
film_convert(*task, task->buffer, task->rgba_byte, task->rgba_half);
}
else if(task->type == DeviceTask::SHADER) {
shader(*task);
}
else if(task->type == DeviceTask::PATH_TRACE) {
RenderTile tile;
/* Keep rendering tiles until done. */
while(task->acquire_tile(this, tile)) {
int start_sample = tile.start_sample;
int end_sample = tile.start_sample + tile.num_samples;
for(int sample = start_sample; sample < end_sample; sample++) {
if(task->get_cancel()) {
if(task->need_finish_queue == false)
break;
}
path_trace(tile, sample);
tile.sample = sample + 1;
task->update_progress(&tile);
}
/* Complete kernel execution before release tile */
/* This helps in multi-device render;
* The device that reaches the critical-section function
* release_tile waits (stalling other devices from entering
* release_tile) for all kernels to complete. If device1 (a
* slow-render device) reaches release_tile first then it would
* stall device2 (a fast-render device) from proceeding to render
* next tile.
*/
clFinish(cqCommandQueue);
task->release_tile(tile);
}
}
}
};
/* TODO(sergey): This is to keep tile split on OpenCL level working
* for now, since without this view-port render does not work as it
* should.
*
* Ideally it'll be done on the higher level, but we need to get ready
* for merge rather soon, so let's keep split logic private here in
* the file.
*/
class SplitRenderTile : public RenderTile {
public:
SplitRenderTile()
: RenderTile(),
buffer_offset_x(0),
buffer_offset_y(0),
rng_state_offset_x(0),
rng_state_offset_y(0),
buffer_rng_state_stride(0) {}
explicit SplitRenderTile(RenderTile& tile)
: RenderTile(),
buffer_offset_x(0),
buffer_offset_y(0),
rng_state_offset_x(0),
rng_state_offset_y(0),
buffer_rng_state_stride(0)
{
x = tile.x;
y = tile.y;
w = tile.w;
h = tile.h;
start_sample = tile.start_sample;
num_samples = tile.num_samples;
sample = tile.sample;
resolution = tile.resolution;
offset = tile.offset;
stride = tile.stride;
buffer = tile.buffer;
rng_state = tile.rng_state;
buffers = tile.buffers;
}
/* Split kernel is device global memory constrained;
* hence split kernel cant render big tile size's in
* one go. If the user sets a big tile size (big tile size
* is a term relative to the available device global memory),
* we split the tile further and then call path_trace on
* each of those split tiles. The following variables declared,
* assist in achieving that purpose
*/
int buffer_offset_x;
int buffer_offset_y;
int rng_state_offset_x;
int rng_state_offset_y;
int buffer_rng_state_stride;
};
/* OpenCLDeviceSplitKernel's declaration/definition. */
class OpenCLDeviceSplitKernel : public OpenCLDeviceBase
{
public:
/* Kernel declaration. */
cl_kernel ckPathTraceKernel_data_init;
cl_kernel ckPathTraceKernel_scene_intersect;
cl_kernel ckPathTraceKernel_lamp_emission;
cl_kernel ckPathTraceKernel_queue_enqueue;
cl_kernel ckPathTraceKernel_background_buffer_update;
cl_kernel ckPathTraceKernel_shader_eval;
cl_kernel ckPathTraceKernel_holdout_emission_blurring_pathtermination_ao;
cl_kernel ckPathTraceKernel_direct_lighting;
cl_kernel ckPathTraceKernel_shadow_blocked;
cl_kernel ckPathTraceKernel_next_iteration_setup;
cl_kernel ckPathTraceKernel_sum_all_radiance;
/* cl_program declaration. */
cl_program data_init_program;
cl_program scene_intersect_program;
cl_program lamp_emission_program;
cl_program queue_enqueue_program;
cl_program background_buffer_update_program;
cl_program shader_eval_program;
cl_program holdout_emission_blurring_pathtermination_ao_program;
cl_program direct_lighting_program;
cl_program shadow_blocked_program;
cl_program next_iteration_setup_program;
cl_program sum_all_radiance_program;
/* Global memory variables [porting]; These memory is used for
* co-operation between different kernels; Data written by one
* kernel will be available to another kernel via this global
* memory.
*/
cl_mem rng_coop;
cl_mem throughput_coop;
cl_mem L_transparent_coop;
cl_mem PathRadiance_coop;
cl_mem Ray_coop;
cl_mem PathState_coop;
cl_mem Intersection_coop;
cl_mem kgbuffer; /* KernelGlobals buffer. */
/* Global buffers for ShaderData. */
cl_mem sd; /* ShaderData used in the main path-iteration loop. */
cl_mem sd_DL_shadow; /* ShaderData used in Direct Lighting and
* shadow_blocked kernel.
*/
/* Global buffers of each member of ShaderData. */
cl_mem P_sd;
cl_mem P_sd_DL_shadow;
cl_mem N_sd;
cl_mem N_sd_DL_shadow;
cl_mem Ng_sd;
cl_mem Ng_sd_DL_shadow;
cl_mem I_sd;
cl_mem I_sd_DL_shadow;
cl_mem shader_sd;
cl_mem shader_sd_DL_shadow;
cl_mem flag_sd;
cl_mem flag_sd_DL_shadow;
cl_mem prim_sd;
cl_mem prim_sd_DL_shadow;
cl_mem type_sd;
cl_mem type_sd_DL_shadow;
cl_mem u_sd;
cl_mem u_sd_DL_shadow;
cl_mem v_sd;
cl_mem v_sd_DL_shadow;
cl_mem object_sd;
cl_mem object_sd_DL_shadow;
cl_mem time_sd;
cl_mem time_sd_DL_shadow;
cl_mem ray_length_sd;
cl_mem ray_length_sd_DL_shadow;
cl_mem ray_depth_sd;
cl_mem ray_depth_sd_DL_shadow;
cl_mem transparent_depth_sd;
cl_mem transparent_depth_sd_DL_shadow;
/* Ray differentials. */
cl_mem dP_sd, dI_sd;
cl_mem dP_sd_DL_shadow, dI_sd_DL_shadow;
cl_mem du_sd, dv_sd;
cl_mem du_sd_DL_shadow, dv_sd_DL_shadow;
/* Dp/Du */
cl_mem dPdu_sd, dPdv_sd;
cl_mem dPdu_sd_DL_shadow, dPdv_sd_DL_shadow;
/* Object motion. */
cl_mem ob_tfm_sd, ob_itfm_sd;
cl_mem ob_tfm_sd_DL_shadow, ob_itfm_sd_DL_shadow;
cl_mem closure_sd;
cl_mem closure_sd_DL_shadow;
cl_mem num_closure_sd;
cl_mem num_closure_sd_DL_shadow;
cl_mem randb_closure_sd;
cl_mem randb_closure_sd_DL_shadow;
cl_mem ray_P_sd;
cl_mem ray_P_sd_DL_shadow;
cl_mem ray_dP_sd;
cl_mem ray_dP_sd_DL_shadow;
/* Global memory required for shadow blocked and accum_radiance. */
cl_mem BSDFEval_coop;
cl_mem ISLamp_coop;
cl_mem LightRay_coop;
cl_mem AOAlpha_coop;
cl_mem AOBSDF_coop;
cl_mem AOLightRay_coop;
cl_mem Intersection_coop_AO;
cl_mem Intersection_coop_DL;
#ifdef WITH_CYCLES_DEBUG
/* DebugData memory */
cl_mem debugdata_coop;
#endif
/* Global state array that tracks ray state. */
cl_mem ray_state;
/* Per sample buffers. */
cl_mem per_sample_output_buffers;
/* Denotes which sample each ray is being processed for. */
cl_mem work_array;
/* Queue */
cl_mem Queue_data; /* Array of size queuesize * num_queues * sizeof(int). */
cl_mem Queue_index; /* Array of size num_queues * sizeof(int);
* Tracks the size of each queue.
*/
/* Flag to make sceneintersect and lampemission kernel use queues. */
cl_mem use_queues_flag;
/* Amount of memory in output buffer associated with one pixel/thread. */
size_t per_thread_output_buffer_size;
/* Total allocatable available device memory. */
size_t total_allocatable_memory;
/* host version of ray_state; Used in checking host path-iteration
* termination.
*/
char *hostRayStateArray;
/* Number of path-iterations to be done in one shot. */
unsigned int PathIteration_times;
#ifdef __WORK_STEALING__
/* Work pool with respect to each work group. */
cl_mem work_pool_wgs;
/* Denotes the maximum work groups possible w.r.t. current tile size. */
unsigned int max_work_groups;
#endif
/* clos_max value for which the kernels have been loaded currently. */
int current_max_closure;
/* Marked True in constructor and marked false at the end of path_trace(). */
bool first_tile;
OpenCLDeviceSplitKernel(DeviceInfo& info, Stats &stats, bool background_)
: OpenCLDeviceBase(info, stats, background_)
{
background = background_;
/* Initialize kernels. */
ckPathTraceKernel_data_init = NULL;
ckPathTraceKernel_scene_intersect = NULL;
ckPathTraceKernel_lamp_emission = NULL;
ckPathTraceKernel_background_buffer_update = NULL;
ckPathTraceKernel_shader_eval = NULL;
ckPathTraceKernel_holdout_emission_blurring_pathtermination_ao = NULL;
ckPathTraceKernel_direct_lighting = NULL;
ckPathTraceKernel_shadow_blocked = NULL;
ckPathTraceKernel_next_iteration_setup = NULL;
ckPathTraceKernel_sum_all_radiance = NULL;
ckPathTraceKernel_queue_enqueue = NULL;
/* Initialize program. */
data_init_program = NULL;
scene_intersect_program = NULL;
lamp_emission_program = NULL;
queue_enqueue_program = NULL;
background_buffer_update_program = NULL;
shader_eval_program = NULL;
holdout_emission_blurring_pathtermination_ao_program = NULL;
direct_lighting_program = NULL;
shadow_blocked_program = NULL;
next_iteration_setup_program = NULL;
sum_all_radiance_program = NULL;
/* Initialize cl_mem variables. */
kgbuffer = NULL;
sd = NULL;
sd_DL_shadow = NULL;
P_sd = NULL;
P_sd_DL_shadow = NULL;
N_sd = NULL;
N_sd_DL_shadow = NULL;
Ng_sd = NULL;
Ng_sd_DL_shadow = NULL;
I_sd = NULL;
I_sd_DL_shadow = NULL;
shader_sd = NULL;
shader_sd_DL_shadow = NULL;
flag_sd = NULL;
flag_sd_DL_shadow = NULL;
prim_sd = NULL;
prim_sd_DL_shadow = NULL;
type_sd = NULL;
type_sd_DL_shadow = NULL;
u_sd = NULL;
u_sd_DL_shadow = NULL;
v_sd = NULL;
v_sd_DL_shadow = NULL;
object_sd = NULL;
object_sd_DL_shadow = NULL;
time_sd = NULL;
time_sd_DL_shadow = NULL;
ray_length_sd = NULL;
ray_length_sd_DL_shadow = NULL;
ray_depth_sd = NULL;
ray_depth_sd_DL_shadow = NULL;
transparent_depth_sd = NULL;
transparent_depth_sd_DL_shadow = NULL;
/* Ray differentials. */
dP_sd = NULL;
dI_sd = NULL;
dP_sd_DL_shadow = NULL;
dI_sd_DL_shadow = NULL;
du_sd = NULL;
dv_sd = NULL;
du_sd_DL_shadow = NULL;
dv_sd_DL_shadow = NULL;
/* Dp/Du */
dPdu_sd = NULL;
dPdv_sd = NULL;
dPdu_sd_DL_shadow = NULL;
dPdv_sd_DL_shadow = NULL;
/* Object motion. */
ob_tfm_sd = NULL;
ob_itfm_sd = NULL;
ob_tfm_sd_DL_shadow = NULL;
ob_itfm_sd_DL_shadow = NULL;
closure_sd = NULL;
closure_sd_DL_shadow = NULL;
num_closure_sd = NULL;
num_closure_sd_DL_shadow = NULL;
randb_closure_sd = NULL;
randb_closure_sd_DL_shadow = NULL;
ray_P_sd = NULL;
ray_P_sd_DL_shadow = NULL;
ray_dP_sd = NULL;
ray_dP_sd_DL_shadow = NULL;
rng_coop = NULL;
throughput_coop = NULL;
L_transparent_coop = NULL;
PathRadiance_coop = NULL;
Ray_coop = NULL;
PathState_coop = NULL;
Intersection_coop = NULL;
ray_state = NULL;
AOAlpha_coop = NULL;
AOBSDF_coop = NULL;
AOLightRay_coop = NULL;
BSDFEval_coop = NULL;
ISLamp_coop = NULL;
LightRay_coop = NULL;
Intersection_coop_AO = NULL;
Intersection_coop_DL = NULL;
#ifdef WITH_CYCLES_DEBUG
debugdata_coop = NULL;
#endif
work_array = NULL;
/* Queue. */
Queue_data = NULL;
Queue_index = NULL;
use_queues_flag = NULL;
per_sample_output_buffers = NULL;
per_thread_output_buffer_size = 0;
hostRayStateArray = NULL;
PathIteration_times = PATH_ITER_INC_FACTOR;
#ifdef __WORK_STEALING__
work_pool_wgs = NULL;
max_work_groups = 0;
#endif
current_max_closure = -1;
first_tile = true;
/* Get device's maximum memory that can be allocated. */
ciErr = clGetDeviceInfo(cdDevice,
CL_DEVICE_MAX_MEM_ALLOC_SIZE,
sizeof(size_t),
&total_allocatable_memory,
NULL);
assert(ciErr == CL_SUCCESS);
if(platform_name == "AMD Accelerated Parallel Processing") {
/* This value is tweak-able; AMD platform does not seem to
* give maximum performance when all of CL_DEVICE_MAX_MEM_ALLOC_SIZE
* is considered for further computation.
*/
total_allocatable_memory /= 2;
}
}
/* TODO(sergey): Seems really close to load_kernel(),
* could it be de-duplicated?
*/
bool load_split_kernel(string kernel_path,
string kernel_init_source,
string clbin,
string custom_kernel_build_options,
cl_program *program,
const string *debug_src = NULL)
{
if(!opencl_version_check())
return false;
clbin = path_user_get(path_join("cache", clbin));
/* If exists already, try use it. */
if(path_exists(clbin) && load_binary(kernel_path,
clbin,
custom_kernel_build_options,
program,
debug_src)) {
/* Kernel loaded from binary. */
}
else {
/* If does not exist or loading binary failed, compile kernel. */
if(!compile_kernel(kernel_path,
kernel_init_source,
custom_kernel_build_options,
program,
debug_src))
{
return false;
}
/* Save binary for reuse. */
if(!save_binary(program, clbin)) {
return false;
}
}
return true;
}
/* Split kernel utility functions. */
size_t get_tex_size(const char *tex_name)
{
cl_mem ptr;
size_t ret_size = 0;
MemMap::iterator i = mem_map.find(tex_name);
if(i != mem_map.end()) {
ptr = CL_MEM_PTR(i->second);
ciErr = clGetMemObjectInfo(ptr,
CL_MEM_SIZE,
sizeof(ret_size),
&ret_size,
NULL);
assert(ciErr == CL_SUCCESS);
}
return ret_size;
}
size_t get_shader_closure_size(int max_closure)
{
return (sizeof(ShaderClosure) * max_closure);
}
size_t get_shader_data_size(size_t shader_closure_size)
{
/* ShaderData size without accounting for ShaderClosure array. */
size_t shader_data_size =
sizeof(ShaderData) - (sizeof(ShaderClosure) * MAX_CLOSURE);
return (shader_data_size + shader_closure_size);
}
/* Returns size of KernelGlobals structure associated with OpenCL. */
size_t get_KernelGlobals_size()
{
/* Copy dummy KernelGlobals related to OpenCL from kernel_globals.h to
* fetch its size.
*/
typedef struct KernelGlobals {
ccl_constant KernelData *data;
#define KERNEL_TEX(type, ttype, name) \
ccl_global type *name;
#include "kernel_textures.h"
#undef KERNEL_TEX
} KernelGlobals;
return sizeof(KernelGlobals);
}
/* Returns size of Structure of arrays implementation of. */
size_t get_shaderdata_soa_size()
{
size_t shader_soa_size = 0;
#define SD_VAR(type, what) shader_soa_size += sizeof(void *);
#define SD_CLOSURE_VAR(type, what, max_closure) shader_soa_size += sizeof(void *);
#include "kernel_shaderdata_vars.h"
#undef SD_VAR
#undef SD_CLOSURE_VAR
return shader_soa_size;
}
bool load_kernels(const DeviceRequestedFeatures& requested_features)
{
/* Get Shader, bake and film_convert kernels.
* It'll also do verification of OpenCL actually initialized.
*/
if(!OpenCLDeviceBase::load_kernels(requested_features)) {
return false;
}
string kernel_path = path_get("kernel");
string kernel_md5 = path_files_md5_hash(kernel_path);
string device_md5;
string kernel_init_source;
string clbin;
string clsrc, *debug_src = NULL;
string build_options = "-D__SPLIT_KERNEL__";
#ifdef __WORK_STEALING__
build_options += " -D__WORK_STEALING__";
#endif
build_options += build_options_from_requested_features(requested_features);
/* Set compute device build option. */
cl_device_type device_type;
ciErr = clGetDeviceInfo(cdDevice,
CL_DEVICE_TYPE,
sizeof(cl_device_type),
&device_type,
NULL);
assert(ciErr == CL_SUCCESS);
if(device_type == CL_DEVICE_TYPE_GPU) {
build_options += " -D__COMPUTE_DEVICE_GPU__";
}
#define GLUE(a, b) a ## b
#define LOAD_KERNEL(name) \
do { \
kernel_init_source = "#include \"kernels/opencl/kernel_" #name ".cl\" // " + \
kernel_md5 + "\n"; \
device_md5 = device_md5_hash(build_options); \
clbin = string_printf("cycles_kernel_%s_%s_" #name ".clbin", \
device_md5.c_str(), kernel_md5.c_str()); \
if(opencl_kernel_use_debug()) { \
clsrc = string_printf("cycles_kernel_%s_%s_" #name ".cl", \
device_md5.c_str(), kernel_md5.c_str()); \
clsrc = path_user_get(path_join("cache", clsrc)); \
debug_src = &clsrc; \
} \
if(!load_split_kernel(kernel_path, kernel_init_source, clbin, \
build_options, \
&GLUE(name, _program), \
debug_src)) \
{ \
fprintf(stderr, "Faled to compile %s\n", #name); \
return false; \
} \
} while(false)
LOAD_KERNEL(data_init);
LOAD_KERNEL(scene_intersect);
LOAD_KERNEL(lamp_emission);
LOAD_KERNEL(queue_enqueue);
LOAD_KERNEL(background_buffer_update);
LOAD_KERNEL(shader_eval);
LOAD_KERNEL(holdout_emission_blurring_pathtermination_ao);
LOAD_KERNEL(direct_lighting);
LOAD_KERNEL(shadow_blocked);
LOAD_KERNEL(next_iteration_setup);
LOAD_KERNEL(sum_all_radiance);
#undef LOAD_KERNEL
#define FIND_KERNEL(name) \
do { \
GLUE(ckPathTraceKernel_, name) = \
clCreateKernel(GLUE(name, _program), \
"kernel_ocl_path_trace_" #name, &ciErr); \
if(opencl_error(ciErr)) { \
fprintf(stderr,"Missing kernel kernel_ocl_path_trace_%s\n", #name); \
return false; \
} \
} while(false)
FIND_KERNEL(data_init);
FIND_KERNEL(scene_intersect);
FIND_KERNEL(lamp_emission);
FIND_KERNEL(queue_enqueue);
FIND_KERNEL(background_buffer_update);
FIND_KERNEL(shader_eval);
FIND_KERNEL(holdout_emission_blurring_pathtermination_ao);
FIND_KERNEL(direct_lighting);
FIND_KERNEL(shadow_blocked);
FIND_KERNEL(next_iteration_setup);
FIND_KERNEL(sum_all_radiance);
#undef FIND_KERNEL
#undef GLUE
current_max_closure = requested_features.max_closure;
return true;
}
~OpenCLDeviceSplitKernel()
{
task_pool.stop();
/* Release kernels */
release_kernel_safe(ckPathTraceKernel_data_init);
release_kernel_safe(ckPathTraceKernel_scene_intersect);
release_kernel_safe(ckPathTraceKernel_lamp_emission);
release_kernel_safe(ckPathTraceKernel_queue_enqueue);
release_kernel_safe(ckPathTraceKernel_background_buffer_update);
release_kernel_safe(ckPathTraceKernel_shader_eval);
release_kernel_safe(ckPathTraceKernel_holdout_emission_blurring_pathtermination_ao);
release_kernel_safe(ckPathTraceKernel_direct_lighting);
release_kernel_safe(ckPathTraceKernel_shadow_blocked);
release_kernel_safe(ckPathTraceKernel_next_iteration_setup);
release_kernel_safe(ckPathTraceKernel_sum_all_radiance);
/* Release global memory */
release_mem_object_safe(P_sd);
release_mem_object_safe(P_sd_DL_shadow);
release_mem_object_safe(N_sd);
release_mem_object_safe(N_sd_DL_shadow);
release_mem_object_safe(Ng_sd);
release_mem_object_safe(Ng_sd_DL_shadow);
release_mem_object_safe(I_sd);
release_mem_object_safe(I_sd_DL_shadow);
release_mem_object_safe(shader_sd);
release_mem_object_safe(shader_sd_DL_shadow);
release_mem_object_safe(flag_sd);
release_mem_object_safe(flag_sd_DL_shadow);
release_mem_object_safe(prim_sd);
release_mem_object_safe(prim_sd_DL_shadow);
release_mem_object_safe(type_sd);
release_mem_object_safe(type_sd_DL_shadow);
release_mem_object_safe(u_sd);
release_mem_object_safe(u_sd_DL_shadow);
release_mem_object_safe(v_sd);
release_mem_object_safe(v_sd_DL_shadow);
release_mem_object_safe(object_sd);
release_mem_object_safe(object_sd_DL_shadow);
release_mem_object_safe(time_sd);
release_mem_object_safe(time_sd_DL_shadow);
release_mem_object_safe(ray_length_sd);
release_mem_object_safe(ray_length_sd_DL_shadow);
release_mem_object_safe(ray_depth_sd);
release_mem_object_safe(ray_depth_sd_DL_shadow);
release_mem_object_safe(transparent_depth_sd);
release_mem_object_safe(transparent_depth_sd_DL_shadow);
/* Ray differentials. */
release_mem_object_safe(dP_sd);
release_mem_object_safe(dP_sd_DL_shadow);
release_mem_object_safe(dI_sd);
release_mem_object_safe(dI_sd_DL_shadow);
release_mem_object_safe(du_sd);
release_mem_object_safe(du_sd_DL_shadow);
release_mem_object_safe(dv_sd);
release_mem_object_safe(dv_sd_DL_shadow);
/* Dp/Du */
release_mem_object_safe(dPdu_sd);
release_mem_object_safe(dPdu_sd_DL_shadow);
release_mem_object_safe(dPdv_sd);
release_mem_object_safe(dPdv_sd_DL_shadow);
/* Object motion. */
release_mem_object_safe(ob_tfm_sd);
release_mem_object_safe(ob_itfm_sd);
release_mem_object_safe(ob_tfm_sd_DL_shadow);
release_mem_object_safe(ob_itfm_sd_DL_shadow);
release_mem_object_safe(closure_sd);
release_mem_object_safe(closure_sd_DL_shadow);
release_mem_object_safe(num_closure_sd);
release_mem_object_safe(num_closure_sd_DL_shadow);
release_mem_object_safe(randb_closure_sd);
release_mem_object_safe(randb_closure_sd_DL_shadow);
release_mem_object_safe(ray_P_sd);
release_mem_object_safe(ray_P_sd_DL_shadow);
release_mem_object_safe(ray_dP_sd);
release_mem_object_safe(ray_dP_sd_DL_shadow);
release_mem_object_safe(rng_coop);
release_mem_object_safe(throughput_coop);
release_mem_object_safe(L_transparent_coop);
release_mem_object_safe(PathRadiance_coop);
release_mem_object_safe(Ray_coop);
release_mem_object_safe(PathState_coop);
release_mem_object_safe(Intersection_coop);
release_mem_object_safe(kgbuffer);
release_mem_object_safe(sd);
release_mem_object_safe(sd_DL_shadow);
release_mem_object_safe(ray_state);
release_mem_object_safe(AOAlpha_coop);
release_mem_object_safe(AOBSDF_coop);
release_mem_object_safe(AOLightRay_coop);
release_mem_object_safe(BSDFEval_coop);
release_mem_object_safe(ISLamp_coop);
release_mem_object_safe(LightRay_coop);
release_mem_object_safe(Intersection_coop_AO);
release_mem_object_safe(Intersection_coop_DL);
#ifdef WITH_CYCLES_DEBUG
release_mem_object_safe(debugdata_coop);
#endif
release_mem_object_safe(use_queues_flag);
release_mem_object_safe(Queue_data);
release_mem_object_safe(Queue_index);
release_mem_object_safe(work_array);
#ifdef __WORK_STEALING__
release_mem_object_safe(work_pool_wgs);
#endif
release_mem_object_safe(per_sample_output_buffers);
/* Release programs */
release_program_safe(data_init_program);
release_program_safe(scene_intersect_program);
release_program_safe(lamp_emission_program);
release_program_safe(queue_enqueue_program);
release_program_safe(background_buffer_update_program);
release_program_safe(shader_eval_program);
release_program_safe(holdout_emission_blurring_pathtermination_ao_program);
release_program_safe(direct_lighting_program);
release_program_safe(shadow_blocked_program);
release_program_safe(next_iteration_setup_program);
release_program_safe(sum_all_radiance_program);
if(hostRayStateArray != NULL) {
free(hostRayStateArray);
}
}
void path_trace(SplitRenderTile& rtile, int2 max_render_feasible_tile_size)
{
/* cast arguments to cl types */
cl_mem d_data = CL_MEM_PTR(const_mem_map["__data"]->device_pointer);
cl_mem d_buffer = CL_MEM_PTR(rtile.buffer);
cl_mem d_rng_state = CL_MEM_PTR(rtile.rng_state);
cl_int d_x = rtile.x;
cl_int d_y = rtile.y;
cl_int d_w = rtile.w;
cl_int d_h = rtile.h;
cl_int d_offset = rtile.offset;
cl_int d_stride = rtile.stride;
/* Make sure that set render feasible tile size is a multiple of local
* work size dimensions.
*/
assert(max_render_feasible_tile_size.x % SPLIT_KERNEL_LOCAL_SIZE_X == 0);
assert(max_render_feasible_tile_size.y % SPLIT_KERNEL_LOCAL_SIZE_Y == 0);
size_t global_size[2];
size_t local_size[2] = {SPLIT_KERNEL_LOCAL_SIZE_X,
SPLIT_KERNEL_LOCAL_SIZE_Y};
/* Set the range of samples to be processed for every ray in
* path-regeneration logic.
*/
cl_int start_sample = rtile.start_sample;
cl_int end_sample = rtile.start_sample + rtile.num_samples;
cl_int num_samples = rtile.num_samples;
#ifdef __WORK_STEALING__
global_size[0] = (((d_w - 1) / local_size[0]) + 1) * local_size[0];
global_size[1] = (((d_h - 1) / local_size[1]) + 1) * local_size[1];
unsigned int num_parallel_samples = 1;
#else
global_size[1] = (((d_h - 1) / local_size[1]) + 1) * local_size[1];
unsigned int num_threads = max_render_feasible_tile_size.x *
max_render_feasible_tile_size.y;
unsigned int num_tile_columns_possible = num_threads / global_size[1];
/* Estimate number of parallel samples that can be
* processed in parallel.
*/
unsigned int num_parallel_samples = min(num_tile_columns_possible / d_w,
rtile.num_samples);
/* Wavefront size in AMD is 64.
* TODO(sergey): What about other platforms?
*/
if(num_parallel_samples >= 64) {
/* TODO(sergey): Could use generic round-up here. */
num_parallel_samples = (num_parallel_samples / 64) * 64;
}
assert(num_parallel_samples != 0);
global_size[0] = d_w * num_parallel_samples;
#endif /* __WORK_STEALING__ */
assert(global_size[0] * global_size[1] <=
max_render_feasible_tile_size.x * max_render_feasible_tile_size.y);
/* Allocate all required global memory once. */
if(first_tile) {
size_t num_global_elements = max_render_feasible_tile_size.x *
max_render_feasible_tile_size.y;
/* TODO(sergey): This will actually over-allocate if
* particular kernel does not support multiclosure.
*/
size_t ShaderClosure_size = get_shader_closure_size(current_max_closure);
#ifdef __WORK_STEALING__
/* Calculate max groups */
size_t max_global_size[2];
size_t tile_x = max_render_feasible_tile_size.x;
size_t tile_y = max_render_feasible_tile_size.y;
max_global_size[0] = (((tile_x - 1) / local_size[0]) + 1) * local_size[0];
max_global_size[1] = (((tile_y - 1) / local_size[1]) + 1) * local_size[1];
max_work_groups = (max_global_size[0] * max_global_size[1]) /
(local_size[0] * local_size[1]);
/* Allocate work_pool_wgs memory. */
work_pool_wgs = mem_alloc(max_work_groups * sizeof(unsigned int));
#endif /* __WORK_STEALING__ */
/* Allocate queue_index memory only once. */
Queue_index = mem_alloc(NUM_QUEUES * sizeof(int));
use_queues_flag = mem_alloc(sizeof(char));
kgbuffer = mem_alloc(get_KernelGlobals_size());
/* Create global buffers for ShaderData. */
sd = mem_alloc(get_shaderdata_soa_size());
sd_DL_shadow = mem_alloc(get_shaderdata_soa_size());
P_sd = mem_alloc(num_global_elements * sizeof(float3));
P_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(float3));
N_sd = mem_alloc(num_global_elements * sizeof(float3));
N_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(float3));
Ng_sd = mem_alloc(num_global_elements * sizeof(float3));
Ng_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(float3));
I_sd = mem_alloc(num_global_elements * sizeof(float3));
I_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(float3));
shader_sd = mem_alloc(num_global_elements * sizeof(int));
shader_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(int));
flag_sd = mem_alloc(num_global_elements * sizeof(int));
flag_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(int));
prim_sd = mem_alloc(num_global_elements * sizeof(int));
prim_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(int));
type_sd = mem_alloc(num_global_elements * sizeof(int));
type_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(int));
u_sd = mem_alloc(num_global_elements * sizeof(float));
u_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(float));
v_sd = mem_alloc(num_global_elements * sizeof(float));
v_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(float));
object_sd = mem_alloc(num_global_elements * sizeof(int));
object_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(int));
time_sd = mem_alloc(num_global_elements * sizeof(float));
time_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(float));
ray_length_sd = mem_alloc(num_global_elements * sizeof(float));
ray_length_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(float));
ray_depth_sd = mem_alloc(num_global_elements * sizeof(int));
ray_depth_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(int));
transparent_depth_sd = mem_alloc(num_global_elements * sizeof(int));
transparent_depth_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(int));
/* Ray differentials. */
dP_sd = mem_alloc(num_global_elements * sizeof(differential3));
dP_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(differential3));
dI_sd = mem_alloc(num_global_elements * sizeof(differential3));
dI_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(differential3));
du_sd = mem_alloc(num_global_elements * sizeof(differential));
du_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(differential));
dv_sd = mem_alloc(num_global_elements * sizeof(differential));
dv_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(differential));
/* Dp/Du */
dPdu_sd = mem_alloc(num_global_elements * sizeof(float3));
dPdu_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(float3));
dPdv_sd = mem_alloc(num_global_elements * sizeof(float3));
dPdv_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(float3));
/* Object motion. */
ob_tfm_sd = mem_alloc(num_global_elements * sizeof(Transform));
ob_tfm_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(Transform));
ob_itfm_sd = mem_alloc(num_global_elements * sizeof(Transform));
ob_itfm_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(Transform));
closure_sd = mem_alloc(num_global_elements * ShaderClosure_size);
closure_sd_DL_shadow = mem_alloc(num_global_elements * 2 * ShaderClosure_size);
num_closure_sd = mem_alloc(num_global_elements * sizeof(int));
num_closure_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(int));
randb_closure_sd = mem_alloc(num_global_elements * sizeof(float));
randb_closure_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(float));
ray_P_sd = mem_alloc(num_global_elements * sizeof(float3));
ray_P_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(float3));
ray_dP_sd = mem_alloc(num_global_elements * sizeof(differential3));
ray_dP_sd_DL_shadow = mem_alloc(num_global_elements * 2 * sizeof(differential3));
/* Creation of global memory buffers which are shared among
* the kernels.
*/
rng_coop = mem_alloc(num_global_elements * sizeof(RNG));
throughput_coop = mem_alloc(num_global_elements * sizeof(float3));
L_transparent_coop = mem_alloc(num_global_elements * sizeof(float));
PathRadiance_coop = mem_alloc(num_global_elements * sizeof(PathRadiance));
Ray_coop = mem_alloc(num_global_elements * sizeof(Ray));
PathState_coop = mem_alloc(num_global_elements * sizeof(PathState));
Intersection_coop = mem_alloc(num_global_elements * sizeof(Intersection));
AOAlpha_coop = mem_alloc(num_global_elements * sizeof(float3));
AOBSDF_coop = mem_alloc(num_global_elements * sizeof(float3));
AOLightRay_coop = mem_alloc(num_global_elements * sizeof(Ray));
BSDFEval_coop = mem_alloc(num_global_elements * sizeof(BsdfEval));
ISLamp_coop = mem_alloc(num_global_elements * sizeof(int));
LightRay_coop = mem_alloc(num_global_elements * sizeof(Ray));
Intersection_coop_AO = mem_alloc(num_global_elements * sizeof(Intersection));
Intersection_coop_DL = mem_alloc(num_global_elements * sizeof(Intersection));
#ifdef WITH_CYCLES_DEBUG
debugdata_coop = mem_alloc(num_global_elements * sizeof(DebugData));
#endif
ray_state = mem_alloc(num_global_elements * sizeof(char));
hostRayStateArray = (char *)calloc(num_global_elements, sizeof(char));
assert(hostRayStateArray != NULL && "Can't create hostRayStateArray memory");
Queue_data = mem_alloc(num_global_elements * (NUM_QUEUES * sizeof(int)+sizeof(int)));
work_array = mem_alloc(num_global_elements * sizeof(unsigned int));
per_sample_output_buffers = mem_alloc(num_global_elements *
per_thread_output_buffer_size);
}
cl_int dQueue_size = global_size[0] * global_size[1];
cl_int total_num_rays = global_size[0] * global_size[1];
cl_uint start_arg_index =
kernel_set_args(ckPathTraceKernel_data_init,
0,
kgbuffer,
sd,
sd_DL_shadow,
P_sd,
P_sd_DL_shadow,
N_sd,
N_sd_DL_shadow,
Ng_sd,
Ng_sd_DL_shadow,
I_sd,
I_sd_DL_shadow,
shader_sd,
shader_sd_DL_shadow,
flag_sd,
flag_sd_DL_shadow,
prim_sd,
prim_sd_DL_shadow,
type_sd,
type_sd_DL_shadow,
u_sd,
u_sd_DL_shadow,
v_sd,
v_sd_DL_shadow,
object_sd,
object_sd_DL_shadow,
time_sd,
time_sd_DL_shadow,
ray_length_sd,
ray_length_sd_DL_shadow,
ray_depth_sd,
ray_depth_sd_DL_shadow,
transparent_depth_sd,
transparent_depth_sd_DL_shadow);
/* Ray differentials. */
start_arg_index +=
kernel_set_args(ckPathTraceKernel_data_init,
start_arg_index,
dP_sd,
dP_sd_DL_shadow,
dI_sd,
dI_sd_DL_shadow,
du_sd,
du_sd_DL_shadow,
dv_sd,
dv_sd_DL_shadow);
/* Dp/Du */
start_arg_index +=
kernel_set_args(ckPathTraceKernel_data_init,
start_arg_index,
dPdu_sd,
dPdu_sd_DL_shadow,
dPdv_sd,
dPdv_sd_DL_shadow);
/* Object motion. */
start_arg_index +=
kernel_set_args(ckPathTraceKernel_data_init,
start_arg_index,
ob_tfm_sd,
ob_tfm_sd_DL_shadow,
ob_itfm_sd,
ob_itfm_sd_DL_shadow);
start_arg_index +=
kernel_set_args(ckPathTraceKernel_data_init,
start_arg_index,
closure_sd,
closure_sd_DL_shadow,
num_closure_sd,
num_closure_sd_DL_shadow,
randb_closure_sd,
randb_closure_sd_DL_shadow,
ray_P_sd,
ray_P_sd_DL_shadow,
ray_dP_sd,
ray_dP_sd_DL_shadow,
d_data,
per_sample_output_buffers,
d_rng_state,
rng_coop,
throughput_coop,
L_transparent_coop,
PathRadiance_coop,
Ray_coop,
PathState_coop,
ray_state);
/* TODO(segrey): Avoid map lookup here. */
#define KERNEL_TEX(type, ttype, name) \
set_kernel_arg_mem(ckPathTraceKernel_data_init, &start_arg_index, #name);
#include "kernel_textures.h"
#undef KERNEL_TEX
start_arg_index +=
kernel_set_args(ckPathTraceKernel_data_init,
start_arg_index,
start_sample,
d_x,
d_y,
d_w,
d_h,
d_offset,
d_stride,
rtile.rng_state_offset_x,
rtile.rng_state_offset_y,
rtile.buffer_rng_state_stride,
Queue_data,
Queue_index,
dQueue_size,
use_queues_flag,
work_array,
#ifdef __WORK_STEALING__
work_pool_wgs,
num_samples,
#endif
#ifdef WITH_CYCLES_DEBUG
debugdata_coop,
#endif
num_parallel_samples);
kernel_set_args(ckPathTraceKernel_scene_intersect,
0,
kgbuffer,
d_data,
rng_coop,
Ray_coop,
PathState_coop,
Intersection_coop,
ray_state,
d_w,
d_h,
Queue_data,
Queue_index,
dQueue_size,
use_queues_flag,
#ifdef WITH_CYCLES_DEBUG
debugdata_coop,
#endif
num_parallel_samples);
kernel_set_args(ckPathTraceKernel_lamp_emission,
0,
kgbuffer,
d_data,
sd,
throughput_coop,
PathRadiance_coop,
Ray_coop,
PathState_coop,
Intersection_coop,
ray_state,
d_w,
d_h,
Queue_data,
Queue_index,
dQueue_size,
use_queues_flag,
num_parallel_samples);
kernel_set_args(ckPathTraceKernel_queue_enqueue,
0,
Queue_data,
Queue_index,
ray_state,
dQueue_size);
kernel_set_args(ckPathTraceKernel_background_buffer_update,
0,
kgbuffer,
d_data,
sd,
per_sample_output_buffers,
d_rng_state,
rng_coop,
throughput_coop,
PathRadiance_coop,
Ray_coop,
PathState_coop,
L_transparent_coop,
ray_state,
d_w,
d_h,
d_x,
d_y,
d_stride,
rtile.rng_state_offset_x,
rtile.rng_state_offset_y,
rtile.buffer_rng_state_stride,
work_array,
Queue_data,
Queue_index,
dQueue_size,
end_sample,
start_sample,
#ifdef __WORK_STEALING__
work_pool_wgs,
num_samples,
#endif
#ifdef WITH_CYCLES_DEBUG
debugdata_coop,
#endif
num_parallel_samples);
kernel_set_args(ckPathTraceKernel_shader_eval,
0,
kgbuffer,
d_data,
sd,
rng_coop,
Ray_coop,
PathState_coop,
Intersection_coop,
ray_state,
Queue_data,
Queue_index,
dQueue_size);
kernel_set_args(ckPathTraceKernel_holdout_emission_blurring_pathtermination_ao,
0,
kgbuffer,
d_data,
sd,
per_sample_output_buffers,
rng_coop,
throughput_coop,
L_transparent_coop,
PathRadiance_coop,
PathState_coop,
Intersection_coop,
AOAlpha_coop,
AOBSDF_coop,
AOLightRay_coop,
d_w,
d_h,
d_x,
d_y,
d_stride,
ray_state,
work_array,
Queue_data,
Queue_index,
dQueue_size,
#ifdef __WORK_STEALING__
start_sample,
#endif
num_parallel_samples);
kernel_set_args(ckPathTraceKernel_direct_lighting,
0,
kgbuffer,
d_data,
sd,
sd_DL_shadow,
rng_coop,
PathState_coop,
ISLamp_coop,
LightRay_coop,
BSDFEval_coop,
ray_state,
Queue_data,
Queue_index,
dQueue_size);
kernel_set_args(ckPathTraceKernel_shadow_blocked,
0,
kgbuffer,
d_data,
sd_DL_shadow,
PathState_coop,
LightRay_coop,
AOLightRay_coop,
Intersection_coop_AO,
Intersection_coop_DL,
ray_state,
Queue_data,
Queue_index,
dQueue_size,
total_num_rays);
kernel_set_args(ckPathTraceKernel_next_iteration_setup,
0,
kgbuffer,
d_data,
sd,
rng_coop,
throughput_coop,
PathRadiance_coop,
Ray_coop,
PathState_coop,
LightRay_coop,
ISLamp_coop,
BSDFEval_coop,
AOLightRay_coop,
AOBSDF_coop,
AOAlpha_coop,
ray_state,
Queue_data,
Queue_index,
dQueue_size,
use_queues_flag);
kernel_set_args(ckPathTraceKernel_sum_all_radiance,
0,
d_data,
d_buffer,
per_sample_output_buffers,
num_parallel_samples,
d_w,
d_h,
d_stride,
rtile.buffer_offset_x,
rtile.buffer_offset_y,
rtile.buffer_rng_state_stride,
start_sample);
/* Macro for Enqueuing split kernels. */
#define GLUE(a, b) a ## b
#define ENQUEUE_SPLIT_KERNEL(kernelName, globalSize, localSize) \
{ \
ciErr = clEnqueueNDRangeKernel(cqCommandQueue, \
GLUE(ckPathTraceKernel_, \
kernelName), \
2, \
NULL, \
globalSize, \
localSize, \
0, \
NULL, \
NULL); \
opencl_assert_err(ciErr, "clEnqueueNDRangeKernel"); \
if(ciErr != CL_SUCCESS) { \
string message = string_printf("OpenCL error: %s in clEnqueueNDRangeKernel()", \
clewErrorString(ciErr)); \
opencl_error(message); \
return; \
} \
} (void) 0
/* Enqueue ckPathTraceKernel_data_init kernel. */
ENQUEUE_SPLIT_KERNEL(data_init, global_size, local_size);
bool activeRaysAvailable = true;
/* Record number of time host intervention has been made */
unsigned int numHostIntervention = 0;
unsigned int numNextPathIterTimes = PathIteration_times;
while(activeRaysAvailable) {
/* Twice the global work size of other kernels for
* ckPathTraceKernel_shadow_blocked_direct_lighting. */
size_t global_size_shadow_blocked[2];
global_size_shadow_blocked[0] = global_size[0] * 2;
global_size_shadow_blocked[1] = global_size[1];
/* Do path-iteration in host [Enqueue Path-iteration kernels. */
for(int PathIter = 0; PathIter < PathIteration_times; PathIter++) {
ENQUEUE_SPLIT_KERNEL(scene_intersect, global_size, local_size);
ENQUEUE_SPLIT_KERNEL(lamp_emission, global_size, local_size);
ENQUEUE_SPLIT_KERNEL(queue_enqueue, global_size, local_size);
ENQUEUE_SPLIT_KERNEL(background_buffer_update, global_size, local_size);
ENQUEUE_SPLIT_KERNEL(shader_eval, global_size, local_size);
ENQUEUE_SPLIT_KERNEL(holdout_emission_blurring_pathtermination_ao, global_size, local_size);
ENQUEUE_SPLIT_KERNEL(direct_lighting, global_size, local_size);
ENQUEUE_SPLIT_KERNEL(shadow_blocked, global_size_shadow_blocked, local_size);
ENQUEUE_SPLIT_KERNEL(next_iteration_setup, global_size, local_size);
}
/* Read ray-state into Host memory to decide if we should exit
* path-iteration in host.
*/
ciErr = clEnqueueReadBuffer(cqCommandQueue,
ray_state,
CL_TRUE,
0,
global_size[0] * global_size[1] * sizeof(char),
hostRayStateArray,
0,
NULL,
NULL);
assert(ciErr == CL_SUCCESS);
activeRaysAvailable = false;
for(int rayStateIter = 0;
rayStateIter < global_size[0] * global_size[1];
++rayStateIter)
{
if(int8_t(hostRayStateArray[rayStateIter]) != RAY_INACTIVE) {
/* Not all rays are RAY_INACTIVE. */
activeRaysAvailable = true;
break;
}
}
if(activeRaysAvailable) {
numHostIntervention++;
PathIteration_times = PATH_ITER_INC_FACTOR;
/* Host intervention done before all rays become RAY_INACTIVE;
* Set do more initial iterations for the next tile.
*/
numNextPathIterTimes += PATH_ITER_INC_FACTOR;
}
}
/* Execute SumALLRadiance kernel to accumulate radiance calculated in
* per_sample_output_buffers into RenderTile's output buffer.
*/
size_t sum_all_radiance_local_size[2] = {16, 16};
size_t sum_all_radiance_global_size[2];
sum_all_radiance_global_size[0] =
(((d_w - 1) / sum_all_radiance_local_size[0]) + 1) *
sum_all_radiance_local_size[0];
sum_all_radiance_global_size[1] =
(((d_h - 1) / sum_all_radiance_local_size[1]) + 1) *
sum_all_radiance_local_size[1];
ENQUEUE_SPLIT_KERNEL(sum_all_radiance,
sum_all_radiance_global_size,
sum_all_radiance_local_size);
#undef ENQUEUE_SPLIT_KERNEL
#undef GLUE
if(numHostIntervention == 0) {
/* This means that we are executing kernel more than required
* Must avoid this for the next sample/tile.
*/
PathIteration_times = ((numNextPathIterTimes - PATH_ITER_INC_FACTOR) <= 0) ?
PATH_ITER_INC_FACTOR : numNextPathIterTimes - PATH_ITER_INC_FACTOR;
}
else {
/* Number of path-iterations done for this tile is set as
* Initial path-iteration times for the next tile
*/
PathIteration_times = numNextPathIterTimes;
}
first_tile = false;
}
/* Calculates the amount of memory that has to be always
* allocated in order for the split kernel to function.
* This memory is tile/scene-property invariant (meaning,
* the value returned by this function does not depend
* on the user set tile size or scene properties.
*/
size_t get_invariable_mem_allocated()
{
size_t total_invariable_mem_allocated = 0;
size_t KernelGlobals_size = 0;
size_t ShaderData_SOA_size = 0;
KernelGlobals_size = get_KernelGlobals_size();
ShaderData_SOA_size = get_shaderdata_soa_size();
total_invariable_mem_allocated += KernelGlobals_size; /* KernelGlobals size */
total_invariable_mem_allocated += NUM_QUEUES * sizeof(unsigned int); /* Queue index size */
total_invariable_mem_allocated += sizeof(char); /* use_queues_flag size */
total_invariable_mem_allocated += ShaderData_SOA_size; /* sd size */
total_invariable_mem_allocated += ShaderData_SOA_size; /* sd_DL_shadow size */
return total_invariable_mem_allocated;
}
/* Calculate the memory that has-to-be/has-been allocated for
* the split kernel to function.
*/
size_t get_tile_specific_mem_allocated(const int2 tile_size)
{
size_t tile_specific_mem_allocated = 0;
/* Get required tile info */
unsigned int user_set_tile_w = tile_size.x;
unsigned int user_set_tile_h = tile_size.y;
#ifdef __WORK_STEALING__
/* Calculate memory to be allocated for work_pools in
* case of work_stealing.
*/
size_t max_global_size[2];
size_t max_num_work_pools = 0;
max_global_size[0] =
(((user_set_tile_w - 1) / SPLIT_KERNEL_LOCAL_SIZE_X) + 1) *
SPLIT_KERNEL_LOCAL_SIZE_X;
max_global_size[1] =
(((user_set_tile_h - 1) / SPLIT_KERNEL_LOCAL_SIZE_Y) + 1) *
SPLIT_KERNEL_LOCAL_SIZE_Y;
max_num_work_pools =
(max_global_size[0] * max_global_size[1]) /
(SPLIT_KERNEL_LOCAL_SIZE_X * SPLIT_KERNEL_LOCAL_SIZE_Y);
tile_specific_mem_allocated += max_num_work_pools * sizeof(unsigned int);
#endif
tile_specific_mem_allocated +=
user_set_tile_w * user_set_tile_h * per_thread_output_buffer_size;
tile_specific_mem_allocated +=
user_set_tile_w * user_set_tile_h * sizeof(RNG);
return tile_specific_mem_allocated;
}
/* Calculates the texture memories and KernelData (d_data) memory
* that has been allocated.
*/
size_t get_scene_specific_mem_allocated(cl_mem d_data)
{
size_t scene_specific_mem_allocated = 0;
/* Calculate texture memories. */
#define KERNEL_TEX(type, ttype, name) \
scene_specific_mem_allocated += get_tex_size(#name);
#include "kernel_textures.h"
#undef KERNEL_TEX
size_t d_data_size;
ciErr = clGetMemObjectInfo(d_data,
CL_MEM_SIZE,
sizeof(d_data_size),
&d_data_size,
NULL);
assert(ciErr == CL_SUCCESS && "Can't get d_data mem object info");
scene_specific_mem_allocated += d_data_size;
return scene_specific_mem_allocated;
}
/* Calculate the memory required for one thread in split kernel. */
size_t get_per_thread_memory()
{
size_t shader_closure_size = 0;
size_t shaderdata_volume = 0;
shader_closure_size = get_shader_closure_size(current_max_closure);
/* TODO(sergey): This will actually over-allocate if
* particular kernel does not support multiclosure.
*/
shaderdata_volume = get_shader_data_size(shader_closure_size);
size_t retval = sizeof(RNG)
+ sizeof(float3) /* Throughput size */
+ sizeof(float) /* L transparent size */
+ sizeof(char) /* Ray state size */
+ sizeof(unsigned int) /* Work element size */
+ sizeof(int) /* ISLamp_size */
+ sizeof(PathRadiance) + sizeof(Ray) + sizeof(PathState)
+ sizeof(Intersection) /* Overall isect */
+ sizeof(Intersection) /* Instersection_coop_AO */
+ sizeof(Intersection) /* Intersection coop DL */
+ shaderdata_volume /* Overall ShaderData */
+ (shaderdata_volume * 2) /* ShaderData : DL and shadow */
+ sizeof(Ray) + sizeof(BsdfEval)
+ sizeof(float3) /* AOAlpha size */
+ sizeof(float3) /* AOBSDF size */
+ sizeof(Ray)
+ (sizeof(int) * NUM_QUEUES)
+ per_thread_output_buffer_size;
return retval;
}
/* Considers the total memory available in the device and
* and returns the maximum global work size possible.
*/
size_t get_feasible_global_work_size(int2 tile_size, cl_mem d_data)
{
/* Calculate invariably allocated memory. */
size_t invariable_mem_allocated = get_invariable_mem_allocated();
/* Calculate tile specific allocated memory. */
size_t tile_specific_mem_allocated =
get_tile_specific_mem_allocated(tile_size);
/* Calculate scene specific allocated memory. */
size_t scene_specific_mem_allocated =
get_scene_specific_mem_allocated(d_data);
/* Calculate total memory available for the threads in global work size. */
size_t available_memory = total_allocatable_memory
- invariable_mem_allocated
- tile_specific_mem_allocated
- scene_specific_mem_allocated
- DATA_ALLOCATION_MEM_FACTOR;
size_t per_thread_memory_required = get_per_thread_memory();
return (available_memory / per_thread_memory_required);
}
/* Checks if the device has enough memory to render the whole tile;
* If not, we should split single tile into multiple tiles of small size
* and process them all.
*/
bool need_to_split_tile(unsigned int d_w,
unsigned int d_h,
int2 max_render_feasible_tile_size)
{
size_t global_size_estimate[2];
/* TODO(sergey): Such round-ups are in quite few places, need to replace
* them with an utility macro.
*/
global_size_estimate[0] =
(((d_w - 1) / SPLIT_KERNEL_LOCAL_SIZE_X) + 1) *
SPLIT_KERNEL_LOCAL_SIZE_X;
global_size_estimate[1] =
(((d_h - 1) / SPLIT_KERNEL_LOCAL_SIZE_Y) + 1) *
SPLIT_KERNEL_LOCAL_SIZE_Y;
if((global_size_estimate[0] * global_size_estimate[1]) >
(max_render_feasible_tile_size.x * max_render_feasible_tile_size.y))
{
return true;
}
else {
return false;
}
}
/* Considers the scene properties, global memory available in the device
* and returns a rectanglular tile dimension (approx the maximum)
* that should render on split kernel.
*/
int2 get_max_render_feasible_tile_size(size_t feasible_global_work_size)
{
int2 max_render_feasible_tile_size;
int square_root_val = (int)sqrt(feasible_global_work_size);
max_render_feasible_tile_size.x = square_root_val;
max_render_feasible_tile_size.y = square_root_val;
/* Ciel round-off max_render_feasible_tile_size. */
int2 ceil_render_feasible_tile_size;
ceil_render_feasible_tile_size.x =
(((max_render_feasible_tile_size.x - 1) / SPLIT_KERNEL_LOCAL_SIZE_X) + 1) *
SPLIT_KERNEL_LOCAL_SIZE_X;
ceil_render_feasible_tile_size.y =
(((max_render_feasible_tile_size.y - 1) / SPLIT_KERNEL_LOCAL_SIZE_Y) + 1) *
SPLIT_KERNEL_LOCAL_SIZE_Y;
if(ceil_render_feasible_tile_size.x * ceil_render_feasible_tile_size.y <=
feasible_global_work_size)
{
return ceil_render_feasible_tile_size;
}
/* Floor round-off max_render_feasible_tile_size. */
int2 floor_render_feasible_tile_size;
floor_render_feasible_tile_size.x =
(max_render_feasible_tile_size.x / SPLIT_KERNEL_LOCAL_SIZE_X) *
SPLIT_KERNEL_LOCAL_SIZE_X;
floor_render_feasible_tile_size.y =
(max_render_feasible_tile_size.y / SPLIT_KERNEL_LOCAL_SIZE_Y) *
SPLIT_KERNEL_LOCAL_SIZE_Y;
return floor_render_feasible_tile_size;
}
/* Try splitting the current tile into multiple smaller
* almost-square-tiles.
*/
int2 get_split_tile_size(RenderTile rtile,
int2 max_render_feasible_tile_size)
{
int2 split_tile_size;
int num_global_threads = max_render_feasible_tile_size.x *
max_render_feasible_tile_size.y;
int d_w = rtile.w;
int d_h = rtile.h;
/* Ceil round off d_w and d_h */
d_w = (((d_w - 1) / SPLIT_KERNEL_LOCAL_SIZE_X) + 1) *
SPLIT_KERNEL_LOCAL_SIZE_X;
d_h = (((d_h - 1) / SPLIT_KERNEL_LOCAL_SIZE_Y) + 1) *
SPLIT_KERNEL_LOCAL_SIZE_Y;
while(d_w * d_h > num_global_threads) {
/* Halve the longer dimension. */
if(d_w >= d_h) {
d_w = d_w / 2;
d_w = (((d_w - 1) / SPLIT_KERNEL_LOCAL_SIZE_X) + 1) *
SPLIT_KERNEL_LOCAL_SIZE_X;
}
else {
d_h = d_h / 2;
d_h = (((d_h - 1) / SPLIT_KERNEL_LOCAL_SIZE_Y) + 1) *
SPLIT_KERNEL_LOCAL_SIZE_Y;
}
}
split_tile_size.x = d_w;
split_tile_size.y = d_h;
return split_tile_size;
}
/* Splits existing tile into multiple tiles of tile size split_tile_size. */
vector<SplitRenderTile> split_tiles(RenderTile rtile, int2 split_tile_size)
{
vector<SplitRenderTile> to_path_trace_rtile;
int d_w = rtile.w;
int d_h = rtile.h;
int num_tiles_x = (((d_w - 1) / split_tile_size.x) + 1);
int num_tiles_y = (((d_h - 1) / split_tile_size.y) + 1);
/* Buffer and rng_state offset calc. */
size_t offset_index = rtile.offset + (rtile.x + rtile.y * rtile.stride);
size_t offset_x = offset_index % rtile.stride;
size_t offset_y = offset_index / rtile.stride;
/* Resize to_path_trace_rtile. */
to_path_trace_rtile.resize(num_tiles_x * num_tiles_y);
for(int tile_iter_y = 0; tile_iter_y < num_tiles_y; tile_iter_y++) {
for(int tile_iter_x = 0; tile_iter_x < num_tiles_x; tile_iter_x++) {
int rtile_index = tile_iter_y * num_tiles_x + tile_iter_x;
to_path_trace_rtile[rtile_index].rng_state_offset_x = offset_x + tile_iter_x * split_tile_size.x;
to_path_trace_rtile[rtile_index].rng_state_offset_y = offset_y + tile_iter_y * split_tile_size.y;
to_path_trace_rtile[rtile_index].buffer_offset_x = offset_x + tile_iter_x * split_tile_size.x;
to_path_trace_rtile[rtile_index].buffer_offset_y = offset_y + tile_iter_y * split_tile_size.y;
to_path_trace_rtile[rtile_index].start_sample = rtile.start_sample;
to_path_trace_rtile[rtile_index].num_samples = rtile.num_samples;
to_path_trace_rtile[rtile_index].sample = rtile.sample;
to_path_trace_rtile[rtile_index].resolution = rtile.resolution;
to_path_trace_rtile[rtile_index].offset = rtile.offset;
to_path_trace_rtile[rtile_index].buffers = rtile.buffers;
to_path_trace_rtile[rtile_index].buffer = rtile.buffer;
to_path_trace_rtile[rtile_index].rng_state = rtile.rng_state;
to_path_trace_rtile[rtile_index].x = rtile.x + (tile_iter_x * split_tile_size.x);
to_path_trace_rtile[rtile_index].y = rtile.y + (tile_iter_y * split_tile_size.y);
to_path_trace_rtile[rtile_index].buffer_rng_state_stride = rtile.stride;
/* Fill width and height of the new render tile. */
to_path_trace_rtile[rtile_index].w = (tile_iter_x == (num_tiles_x - 1)) ?
(d_w - (tile_iter_x * split_tile_size.x)) /* Border tile */
: split_tile_size.x;
to_path_trace_rtile[rtile_index].h = (tile_iter_y == (num_tiles_y - 1)) ?
(d_h - (tile_iter_y * split_tile_size.y)) /* Border tile */
: split_tile_size.y;
to_path_trace_rtile[rtile_index].stride = to_path_trace_rtile[rtile_index].w;
}
}
return to_path_trace_rtile;
}
void thread_run(DeviceTask *task)
{
if(task->type == DeviceTask::FILM_CONVERT) {
film_convert(*task, task->buffer, task->rgba_byte, task->rgba_half);
}
else if(task->type == DeviceTask::SHADER) {
shader(*task);
}
else if(task->type == DeviceTask::PATH_TRACE) {
RenderTile tile;
bool initialize_data_and_check_render_feasibility = false;
bool need_to_split_tiles_further = false;
int2 max_render_feasible_tile_size;
size_t feasible_global_work_size;
const int2 tile_size = task->requested_tile_size;
/* Keep rendering tiles until done. */
while(task->acquire_tile(this, tile)) {
if(!initialize_data_and_check_render_feasibility) {
/* Initialize data. */
/* Calculate per_thread_output_buffer_size. */
size_t output_buffer_size = 0;
ciErr = clGetMemObjectInfo((cl_mem)tile.buffer,
CL_MEM_SIZE,
sizeof(output_buffer_size),
&output_buffer_size,
NULL);
assert(ciErr == CL_SUCCESS && "Can't get tile.buffer mem object info");
/* This value is different when running on AMD and NV. */
if(background) {
/* In offline render the number of buffer elements
* associated with tile.buffer is the current tile size.
*/
per_thread_output_buffer_size =
output_buffer_size / (tile.w * tile.h);
}
else {
/* interactive rendering, unlike offline render, the number of buffer elements
* associated with tile.buffer is the entire viewport size.
*/
per_thread_output_buffer_size =
output_buffer_size / (tile.buffers->params.width *
tile.buffers->params.height);
}
/* Check render feasibility. */
feasible_global_work_size = get_feasible_global_work_size(
tile_size,
CL_MEM_PTR(const_mem_map["__data"]->device_pointer));
max_render_feasible_tile_size =
get_max_render_feasible_tile_size(
feasible_global_work_size);
need_to_split_tiles_further =
need_to_split_tile(tile_size.x,
tile_size.y,
max_render_feasible_tile_size);
initialize_data_and_check_render_feasibility = true;
}
if(need_to_split_tiles_further) {
int2 split_tile_size =
get_split_tile_size(tile,
max_render_feasible_tile_size);
vector<SplitRenderTile> to_path_trace_render_tiles =
split_tiles(tile, split_tile_size);
/* Print message to console */
if(background && (to_path_trace_render_tiles.size() > 1)) {
fprintf(stderr, "Message : Tiles need to be split "
"further inside path trace (due to insufficient "
"device-global-memory for split kernel to "
"function) \n"
"The current tile of dimensions %dx%d is split "
"into tiles of dimension %dx%d for render \n",
tile.w, tile.h,
split_tile_size.x,
split_tile_size.y);
}
/* Process all split tiles. */
for(int tile_iter = 0;
tile_iter < to_path_trace_render_tiles.size();
++tile_iter)
{
path_trace(to_path_trace_render_tiles[tile_iter],
max_render_feasible_tile_size);
}
}
else {
/* No splitting required; process the entire tile at once. */
/* Render feasible tile size is user-set-tile-size itself. */
max_render_feasible_tile_size.x =
(((tile_size.x - 1) / SPLIT_KERNEL_LOCAL_SIZE_X) + 1) *
SPLIT_KERNEL_LOCAL_SIZE_X;
max_render_feasible_tile_size.y =
(((tile_size.y - 1) / SPLIT_KERNEL_LOCAL_SIZE_Y) + 1) *
SPLIT_KERNEL_LOCAL_SIZE_Y;
/* buffer_rng_state_stride is stride itself. */
SplitRenderTile split_tile(tile);
split_tile.buffer_rng_state_stride = tile.stride;
path_trace(split_tile, max_render_feasible_tile_size);
}
tile.sample = tile.start_sample + tile.num_samples;
/* Complete kernel execution before release tile. */
/* This helps in multi-device render;
* The device that reaches the critical-section function
* release_tile waits (stalling other devices from entering
* release_tile) for all kernels to complete. If device1 (a
* slow-render device) reaches release_tile first then it would
* stall device2 (a fast-render device) from proceeding to render
* next tile.
*/
clFinish(cqCommandQueue);
task->release_tile(tile);
}
}
}
protected:
cl_mem mem_alloc(size_t bufsize, cl_mem_flags mem_flag = CL_MEM_READ_WRITE)
{
cl_mem ptr;
assert(bufsize != 0);
ptr = clCreateBuffer(cxContext, mem_flag, bufsize, NULL, &ciErr);
opencl_assert_err(ciErr, "clCreateBuffer");
return ptr;
}
/* ** Those guys are for workign around some compiler-specific bugs ** */
cl_program load_cached_kernel(
const DeviceRequestedFeatures& /*requested_features*/,
OpenCLCache::ProgramName /*program_name*/,
thread_scoped_lock /*cache_locker*/)
{
VLOG(2) << "Skip loading kernel from cache, "
<< "not supported by split kernel.";
return NULL;
}
void store_cached_kernel(cl_platform_id /*platform*/,
cl_device_id /*device*/,
cl_program /*program*/,
OpenCLCache::ProgramName /*program_name*/,
thread_scoped_lock& /*slot_locker*/)
{
VLOG(2) << "Skip storing kernel in cache, "
<< "not supported by split kernel.";
}
string build_options_for_base_program(
const DeviceRequestedFeatures& requested_features)
{
return build_options_from_requested_features(requested_features);
}
};
Device *device_opencl_create(DeviceInfo& info, Stats &stats, bool background)
{
vector<OpenCLPlatformDevice> usable_devices;
opencl_get_usable_devices(&usable_devices);
assert(info.num < usable_devices.size());
const OpenCLPlatformDevice& platform_device = usable_devices[info.num];
const string& platform_name = platform_device.platform_name;
const cl_device_type device_type = platform_device.device_type;
if(opencl_kernel_use_split(platform_name, device_type)) {
VLOG(1) << "Using split kernel.";
return new OpenCLDeviceSplitKernel(info, stats, background);
} else {
VLOG(1) << "Using mega kernel.";
return new OpenCLDeviceMegaKernel(info, stats, background);
}
}
bool device_opencl_init(void)
{
static bool initialized = false;
static bool result = false;
if(initialized)
return result;
initialized = true;
if(opencl_device_type() != 0) {
int clew_result = clewInit();
if(clew_result == CLEW_SUCCESS) {
VLOG(1) << "CLEW initialization succeeded.";
result = true;
}
else {
VLOG(1) << "CLEW initialization failed: "
<< ((clew_result == CLEW_ERROR_ATEXIT_FAILED)
? "Error setting up atexit() handler"
: "Error opening the library");
}
}
else {
VLOG(1) << "Skip initializing CLEW, platform is force disabled.";
result = false;
}
return result;
}
void device_opencl_info(vector<DeviceInfo>& devices)
{
vector<OpenCLPlatformDevice> usable_devices;
opencl_get_usable_devices(&usable_devices);
/* Devices are numbered consecutively across platforms. */
int num_devices = 0;
foreach(OpenCLPlatformDevice& platform_device, usable_devices) {
const string& platform_name = platform_device.platform_name;
const cl_device_type device_type = platform_device.device_type;
const string& device_name = platform_device.device_name;
DeviceInfo info;
info.type = DEVICE_OPENCL;
info.description = string_remove_trademark(string(device_name));
info.num = num_devices;
info.id = string_printf("OPENCL_%d", info.num);
/* We don't know if it's used for display, but assume it is. */
info.display_device = true;
info.advanced_shading = opencl_kernel_use_advanced_shading(platform_name);
info.pack_images = true;
info.use_split_kernel = opencl_kernel_use_split(platform_name,
device_type);
devices.push_back(info);
num_devices++;
}
}
string device_opencl_capabilities(void)
{
if(opencl_device_type() == 0) {
return "All OpenCL devices are forced to be OFF";
}
string result = "";
string error_msg = ""; /* Only used by opencl_assert(), but in the future
* it could also be nicely reported to the console.
*/
cl_uint num_platforms = 0;
opencl_assert(clGetPlatformIDs(0, NULL, &num_platforms));
if(num_platforms == 0) {
return "No OpenCL platforms found\n";
}
result += string_printf("Number of platforms: %u\n", num_platforms);
vector<cl_platform_id> platform_ids;
platform_ids.resize(num_platforms);
opencl_assert(clGetPlatformIDs(num_platforms, &platform_ids[0], NULL));
#define APPEND_STRING_INFO(func, id, name, what) \
do { \
char data[1024] = "\0"; \
opencl_assert(func(id, what, sizeof(data), &data, NULL)); \
result += string_printf("%s: %s\n", name, data); \
} while(false)
#define APPEND_PLATFORM_STRING_INFO(id, name, what) \
APPEND_STRING_INFO(clGetPlatformInfo, id, "\tPlatform " name, what)
#define APPEND_DEVICE_STRING_INFO(id, name, what) \
APPEND_STRING_INFO(clGetDeviceInfo, id, "\t\t\tDevice " name, what)
vector<cl_device_id> device_ids;
for(cl_uint platform = 0; platform < num_platforms; ++platform) {
cl_platform_id platform_id = platform_ids[platform];
result += string_printf("Platform #%u\n", platform);
APPEND_PLATFORM_STRING_INFO(platform_id, "Name", CL_PLATFORM_NAME);
APPEND_PLATFORM_STRING_INFO(platform_id, "Vendor", CL_PLATFORM_VENDOR);
APPEND_PLATFORM_STRING_INFO(platform_id, "Version", CL_PLATFORM_VERSION);
APPEND_PLATFORM_STRING_INFO(platform_id, "Profile", CL_PLATFORM_PROFILE);
APPEND_PLATFORM_STRING_INFO(platform_id, "Extensions", CL_PLATFORM_EXTENSIONS);
cl_uint num_devices = 0;
opencl_assert(clGetDeviceIDs(platform_ids[platform],
CL_DEVICE_TYPE_ALL,
0,
NULL,
&num_devices));
result += string_printf("\tNumber of devices: %u\n", num_devices);
device_ids.resize(num_devices);
opencl_assert(clGetDeviceIDs(platform_ids[platform],
CL_DEVICE_TYPE_ALL,
num_devices,
&device_ids[0],
NULL));
for(cl_uint device = 0; device < num_devices; ++device) {
cl_device_id device_id = device_ids[device];
result += string_printf("\t\tDevice: #%u\n", device);
APPEND_DEVICE_STRING_INFO(device_id, "Name", CL_DEVICE_NAME);
APPEND_DEVICE_STRING_INFO(device_id, "Vendor", CL_DEVICE_VENDOR);
APPEND_DEVICE_STRING_INFO(device_id, "OpenCL C Version", CL_DEVICE_OPENCL_C_VERSION);
APPEND_DEVICE_STRING_INFO(device_id, "Profile", CL_DEVICE_PROFILE);
APPEND_DEVICE_STRING_INFO(device_id, "Version", CL_DEVICE_VERSION);
APPEND_DEVICE_STRING_INFO(device_id, "Extensions", CL_DEVICE_EXTENSIONS);
}
}
#undef APPEND_STRING_INFO
#undef APPEND_PLATFORM_STRING_INFO
#undef APPEND_DEVICE_STRING_INFO
return result;
}
CCL_NAMESPACE_END
#endif /* WITH_OPENCL */
|