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
|
/*
* 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 "device.h"
#include "device_intern.h"
#include "buffers.h"
#include "util_foreach.h"
#include "util_map.h"
#include "util_math.h"
#include "util_md5.h"
#include "util_opencl.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))
static cl_device_type opencl_device_type()
{
char *device = getenv("CYCLES_OPENCL_TEST");
if(device) {
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;
}
static bool opencl_kernel_use_debug()
{
return (getenv("CYCLES_OPENCL_DEBUG") != NULL);
}
static 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 false;
else if(platform == "AMD Accelerated Parallel Processing")
return false;
else if(platform == "Intel(R) OpenCL")
return true;
return false;
}
static string opencl_kernel_build_options(const string& platform, const string *debug_src = NULL)
{
string build_options = " -cl-fast-relaxed-math ";
if(platform == "NVIDIA CUDA")
build_options += "-D__KERNEL_OPENCL_NVIDIA__ -cl-nv-maxrregcount=32 -cl-nv-verbose ";
else if(platform == "Apple")
build_options += "-D__KERNEL_OPENCL_APPLE__ -Wno-missing-prototypes ";
else if(platform == "AMD Accelerated Parallel Processing")
build_options += "-D__KERNEL_OPENCL_AMD__ ";
else if(platform == "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__ ";
if(opencl_kernel_use_advanced_shading(platform))
build_options += "-D__KERNEL_OPENCL_NEED_ADVANCED_SHADING__ ";
return build_options;
}
/* thread safe cache for contexts and programs */
class OpenCLCache
{
struct Slot
{
thread_mutex *mutex;
cl_context context;
cl_program program;
Slot() : mutex(NULL), context(NULL), program(NULL) {}
Slot(const Slot &rhs)
: mutex(rhs.mutex)
, context(rhs.context)
, program(rhs.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:
/* 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,
thread_scoped_lock &slot_locker)
{
cl_program program = get_something<cl_program>(platform, device, &Slot::program, slot_locker);
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,
thread_scoped_lock &slot_locker)
{
store_something<cl_program>(platform, device, program, &Slot::program, slot_locker);
/* 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
* the parameter is a temporary workaround. See OpenCLCache::~OpenCLCache */
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.program != NULL)
clReleaseProgram(item.second.program);
if(item.second.context != NULL)
clReleaseContext(item.second.context);
}
self.cache.clear();
}
};
class OpenCLDevice : 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 ckPathTraceKernel;
cl_kernel ckFilmConvertByteKernel;
cl_kernel ckFilmConvertHalfFloatKernel;
cl_kernel ckShaderKernel;
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;
const char *opencl_error_string(cl_int err)
{
switch (err) {
case CL_SUCCESS: return "Success!";
case CL_DEVICE_NOT_FOUND: return "Device not found.";
case CL_DEVICE_NOT_AVAILABLE: return "Device not available";
case CL_COMPILER_NOT_AVAILABLE: return "Compiler not available";
case CL_MEM_OBJECT_ALLOCATION_FAILURE: return "Memory object allocation failure";
case CL_OUT_OF_RESOURCES: return "Out of resources";
case CL_OUT_OF_HOST_MEMORY: return "Out of host memory";
case CL_PROFILING_INFO_NOT_AVAILABLE: return "Profiling information not available";
case CL_MEM_COPY_OVERLAP: return "Memory copy overlap";
case CL_IMAGE_FORMAT_MISMATCH: return "Image format mismatch";
case CL_IMAGE_FORMAT_NOT_SUPPORTED: return "Image format not supported";
case CL_BUILD_PROGRAM_FAILURE: return "Program build failure";
case CL_MAP_FAILURE: return "Map failure";
case CL_INVALID_VALUE: return "Invalid value";
case CL_INVALID_DEVICE_TYPE: return "Invalid device type";
case CL_INVALID_PLATFORM: return "Invalid platform";
case CL_INVALID_DEVICE: return "Invalid device";
case CL_INVALID_CONTEXT: return "Invalid context";
case CL_INVALID_QUEUE_PROPERTIES: return "Invalid queue properties";
case CL_INVALID_COMMAND_QUEUE: return "Invalid command queue";
case CL_INVALID_HOST_PTR: return "Invalid host pointer";
case CL_INVALID_MEM_OBJECT: return "Invalid memory object";
case CL_INVALID_IMAGE_FORMAT_DESCRIPTOR: return "Invalid image format descriptor";
case CL_INVALID_IMAGE_SIZE: return "Invalid image size";
case CL_INVALID_SAMPLER: return "Invalid sampler";
case CL_INVALID_BINARY: return "Invalid binary";
case CL_INVALID_BUILD_OPTIONS: return "Invalid build options";
case CL_INVALID_PROGRAM: return "Invalid program";
case CL_INVALID_PROGRAM_EXECUTABLE: return "Invalid program executable";
case CL_INVALID_KERNEL_NAME: return "Invalid kernel name";
case CL_INVALID_KERNEL_DEFINITION: return "Invalid kernel definition";
case CL_INVALID_KERNEL: return "Invalid kernel";
case CL_INVALID_ARG_INDEX: return "Invalid argument index";
case CL_INVALID_ARG_VALUE: return "Invalid argument value";
case CL_INVALID_ARG_SIZE: return "Invalid argument size";
case CL_INVALID_KERNEL_ARGS: return "Invalid kernel arguments";
case CL_INVALID_WORK_DIMENSION: return "Invalid work dimension";
case CL_INVALID_WORK_GROUP_SIZE: return "Invalid work group size";
case CL_INVALID_WORK_ITEM_SIZE: return "Invalid work item size";
case CL_INVALID_GLOBAL_OFFSET: return "Invalid global offset";
case CL_INVALID_EVENT_WAIT_LIST: return "Invalid event wait list";
case CL_INVALID_EVENT: return "Invalid event";
case CL_INVALID_OPERATION: return "Invalid operation";
case CL_INVALID_GL_OBJECT: return "Invalid OpenGL object";
case CL_INVALID_BUFFER_SIZE: return "Invalid buffer size";
case CL_INVALID_MIP_LEVEL: return "Invalid mip-map level";
default: return "Unknown";
}
}
bool opencl_error(cl_int err)
{
if(err != CL_SUCCESS) {
string message = string_printf("OpenCL error (%d): %s", err, opencl_error_string(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());
}
void opencl_assert(cl_int err)
{
if(err != CL_SUCCESS) {
string message = string_printf("OpenCL error (%d): %s", err, opencl_error_string(err));
if(error_msg == "")
error_msg = message;
fprintf(stderr, "%s\n", message.c_str());
#ifndef NDEBUG
abort();
#endif
}
}
OpenCLDevice(DeviceInfo& info, Stats &stats, bool background_)
: Device(info, stats, background_)
{
cpPlatform = NULL;
cdDevice = NULL;
cxContext = NULL;
cqCommandQueue = NULL;
cpProgram = NULL;
ckPathTraceKernel = NULL;
ckFilmConvertByteKernel = NULL;
ckFilmConvertHalfFloatKernel = NULL;
ckShaderKernel = NULL;
null_mem = 0;
device_initialized = false;
/* setup platform */
cl_uint num_platforms;
ciErr = clGetPlatformIDs(0, NULL, &num_platforms);
if(opencl_error(ciErr))
return;
if(num_platforms == 0) {
opencl_error("OpenCL: no platforms found.");
return;
}
vector<cl_platform_id> platforms(num_platforms, NULL);
ciErr = clGetPlatformIDs(num_platforms, &platforms[0], NULL);
if(opencl_error(ciErr))
return;
int num_base = 0;
int total_devices = 0;
for (int platform = 0; platform < num_platforms; platform++) {
cl_uint num_devices;
if(opencl_error(clGetDeviceIDs(platforms[platform], opencl_device_type(), 0, NULL, &num_devices)))
return;
total_devices += num_devices;
if(info.num - num_base >= num_devices) {
/* num doesn't refer to a device in this platform */
num_base += num_devices;
continue;
}
/* device is in this platform */
cpPlatform = platforms[platform];
/* get devices */
vector<cl_device_id> device_ids(num_devices, NULL);
if(opencl_error(clGetDeviceIDs(cpPlatform, opencl_device_type(), num_devices, &device_ids[0], NULL)))
return;
cdDevice = device_ids[info.num - num_base];
char name[256];
clGetPlatformInfo(cpPlatform, CL_PLATFORM_NAME, sizeof(name), &name, NULL);
platform_name = name;
break;
}
if(total_devices == 0) {
opencl_error("OpenCL: no devices found.");
return;
}
else if(!cdDevice) {
opencl_error("OpenCL: specified device not found.");
return;
}
{
/* 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))
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;
device_initialized = true;
}
static void 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()
{
char version[256];
int major, minor, req_major = 1, req_minor = 1;
clGetPlatformInfo(cpPlatform, CL_PLATFORM_VERSION, sizeof(version), &version, NULL);
if(sscanf(version, "OpenCL %d.%d", &major, &minor) < 2) {
opencl_error(string_printf("OpenCL: failed to parse platform version string (%s).", version));
return false;
}
if(!((major == req_major && minor >= req_minor) || (major > req_major))) {
opencl_error(string_printf("OpenCL: platform version 1.1 or later required, found %d.%d", major, minor));
return false;
}
clGetDeviceInfo(cdDevice, CL_DEVICE_OPENCL_C_VERSION, sizeof(version), &version, NULL);
if(sscanf(version, "OpenCL C %d.%d", &major, &minor) < 2) {
opencl_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))) {
opencl_error(string_printf("OpenCL: C version 1.1 or later required, found %d.%d", major, minor));
return false;
}
return true;
}
bool load_binary(const string& kernel_path, const string& clbin, 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];
cpProgram = 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(kernel_path, debug_src))
return false;
return true;
}
bool save_binary(const string& clbin)
{
size_t size = 0;
clGetProgramInfo(cpProgram, CL_PROGRAM_BINARY_SIZES, sizeof(size_t), &size, NULL);
if(!size)
return false;
vector<uint8_t> binary(size);
uint8_t *bytes = &binary[0];
clGetProgramInfo(cpProgram, 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(const string& kernel_path, const string *debug_src = NULL)
{
string build_options = opencl_kernel_build_options(platform_name, debug_src);
ciErr = clBuildProgram(cpProgram, 0, NULL, build_options.c_str(), NULL, NULL);
/* show warnings even if build is successful */
size_t ret_val_size = 0;
clGetProgramBuildInfo(cpProgram, cdDevice, CL_PROGRAM_BUILD_LOG, 0, NULL, &ret_val_size);
if(ret_val_size > 1) {
vector<char> build_log(ret_val_size+1);
clGetProgramBuildInfo(cpProgram, cdDevice, CL_PROGRAM_BUILD_LOG, ret_val_size, &build_log[0], NULL);
build_log[ret_val_size] = '\0';
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, const string& kernel_md5, 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 */
string source = "#include \"kernel.cl\" // " + kernel_md5 + "\n";
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();
cpProgram = clCreateProgramWithSource(cxContext, 1, &source_str, &source_len, &ciErr);
if(opencl_error(ciErr))
return false;
double starttime = time_dt();
printf("Compiling OpenCL kernel ...\n");
if(!build_kernel(kernel_path, debug_src))
return false;
printf("Kernel compilation finished in %.2lfs.\n", time_dt() - starttime);
return true;
}
string device_md5_hash()
{
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 = opencl_kernel_build_options(platform_name);
md5.append((uint8_t*)options.c_str(), options.size());
return md5.get_hex();
}
bool load_kernels(bool experimental)
{
/* 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 = OpenCLCache::get_program(cpPlatform, cdDevice, cache_locker);
if(!cpProgram) {
/* verify we have right opencl version */
if(!opencl_version_check())
return false;
/* md5 hash to detect changes */
string kernel_path = path_get("kernel");
string kernel_md5 = path_files_md5_hash(kernel_path);
string device_md5 = device_md5_hash();
/* 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, debug_src)) {
/* kernel loaded from binary */
}
else {
/* if does not exist or loading binary failed, compile kernel */
if(!compile_kernel(kernel_path, kernel_md5, debug_src))
return false;
/* save binary for reuse */
if(!save_binary(clbin))
return false;
}
/* cache the program */
OpenCLCache::store_program(cpPlatform, cdDevice, cpProgram, cache_locker);
}
/* find kernels */
ckPathTraceKernel = clCreateKernel(cpProgram, "kernel_ocl_path_trace", &ciErr);
if(opencl_error(ciErr))
return false;
ckFilmConvertByteKernel = clCreateKernel(cpProgram, "kernel_ocl_convert_to_byte", &ciErr);
if(opencl_error(ciErr))
return false;
ckFilmConvertHalfFloatKernel = clCreateKernel(cpProgram, "kernel_ocl_convert_to_half_float", &ciErr);
if(opencl_error(ciErr))
return false;
ckShaderKernel = clCreateKernel(cpProgram, "kernel_ocl_shader", &ciErr);
if(opencl_error(ciErr))
return false;
return true;
}
~OpenCLDevice()
{
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(ckPathTraceKernel)
clReleaseKernel(ckPathTraceKernel);
if(ckFilmConvertByteKernel)
clReleaseKernel(ckFilmConvertByteKernel);
if(ckFilmConvertHalfFloatKernel)
clReleaseKernel(ckFilmConvertHalfFloatKernel);
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;
mem.device_pointer = (device_ptr)clCreateBuffer(cxContext, mem_flag, size, mem_ptr, &ciErr);
opencl_assert(ciErr);
stats.mem_alloc(size);
}
void mem_copy_to(device_memory& mem)
{
/* this is blocking */
size_t size = mem.memory_size();
ciErr = clEnqueueWriteBuffer(cqCommandQueue, CL_MEM_PTR(mem.device_pointer), CL_TRUE, 0, size, (void*)mem.data_pointer, 0, NULL, NULL);
opencl_assert(ciErr);
}
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;
ciErr = clEnqueueReadBuffer(cqCommandQueue, CL_MEM_PTR(mem.device_pointer), CL_TRUE, offset, size, (uchar*)mem.data_pointer + offset, 0, NULL, NULL);
opencl_assert(ciErr);
}
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) {
ciErr = clReleaseMemObject(CL_MEM_PTR(mem.device_pointer));
mem.device_pointer = 0;
opencl_assert(ciErr);
stats.mem_free(mem.memory_size());
}
}
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, bool interpolation, bool periodic)
{
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(sqrt((double)workgroup_size), 1.0);
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 */
ciErr = clEnqueueNDRangeKernel(cqCommandQueue, kernel, 2, NULL, global_size, local_size, 0, NULL, NULL);
opencl_assert(ciErr);
opencl_assert(clFlush(cqCommandQueue));
}
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_sample = sample;
cl_int d_offset = rtile.offset;
cl_int d_stride = rtile.stride;
/* sample arguments */
cl_uint narg = 0;
ciErr = 0;
ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_data), (void*)&d_data);
ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_buffer), (void*)&d_buffer);
ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_rng_state), (void*)&d_rng_state);
#define KERNEL_TEX(type, ttype, name) \
ciErr |= set_kernel_arg_mem(ckPathTraceKernel, &narg, #name);
#include "kernel_textures.h"
ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_sample), (void*)&d_sample);
ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_x), (void*)&d_x);
ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_y), (void*)&d_y);
ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_w), (void*)&d_w);
ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_h), (void*)&d_h);
ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_offset), (void*)&d_offset);
ciErr |= clSetKernelArg(ckPathTraceKernel, narg++, sizeof(d_stride), (void*)&d_stride);
opencl_assert(ciErr);
enqueue_kernel(ckPathTraceKernel, d_w, d_h);
}
cl_int set_kernel_arg_mem(cl_kernel kernel, cl_uint *narg, const char *name)
{
cl_mem ptr;
cl_int err = 0;
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);
}
err |= clSetKernelArg(kernel, (*narg)++, sizeof(ptr), (void*)&ptr);
opencl_assert(err);
return err;
}
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;
/* sample arguments */
cl_uint narg = 0;
ciErr = 0;
cl_kernel ckFilmConvertKernel = (rgba_byte)? ckFilmConvertByteKernel: ckFilmConvertHalfFloatKernel;
ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_data), (void*)&d_data);
ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_rgba), (void*)&d_rgba);
ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_buffer), (void*)&d_buffer);
#define KERNEL_TEX(type, ttype, name) \
ciErr |= set_kernel_arg_mem(ckFilmConvertKernel, &narg, #name);
#include "kernel_textures.h"
ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_sample_scale), (void*)&d_sample_scale);
ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_x), (void*)&d_x);
ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_y), (void*)&d_y);
ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_w), (void*)&d_w);
ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_h), (void*)&d_h);
ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_offset), (void*)&d_offset);
ciErr |= clSetKernelArg(ckFilmConvertKernel, narg++, sizeof(d_stride), (void*)&d_stride);
opencl_assert(ciErr);
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;
/* sample arguments */
cl_uint narg = 0;
ciErr = 0;
ciErr |= clSetKernelArg(ckShaderKernel, narg++, sizeof(d_data), (void*)&d_data);
ciErr |= clSetKernelArg(ckShaderKernel, narg++, sizeof(d_input), (void*)&d_input);
ciErr |= clSetKernelArg(ckShaderKernel, narg++, sizeof(d_output), (void*)&d_output);
#define KERNEL_TEX(type, ttype, name) \
ciErr |= set_kernel_arg_mem(ckShaderKernel, &narg, #name);
#include "kernel_textures.h"
ciErr |= clSetKernelArg(ckShaderKernel, narg++, sizeof(d_shader_eval_type), (void*)&d_shader_eval_type);
ciErr |= clSetKernelArg(ckShaderKernel, narg++, sizeof(d_shader_x), (void*)&d_shader_x);
ciErr |= clSetKernelArg(ckShaderKernel, narg++, sizeof(d_shader_w), (void*)&d_shader_w);
opencl_assert(ciErr);
enqueue_kernel(ckShaderKernel, task.shader_w, 1);
}
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);
}
task->release_tile(tile);
}
}
}
class OpenCLDeviceTask : public DeviceTask {
public:
OpenCLDeviceTask(OpenCLDevice *device, DeviceTask& task)
: DeviceTask(task)
{
run = function_bind(&OpenCLDevice::thread_run, device, this);
}
};
void task_add(DeviceTask& task)
{
task_pool.push(new OpenCLDeviceTask(this, task));
}
void task_wait()
{
task_pool.wait();
}
void task_cancel()
{
task_pool.cancel();
}
};
Device *device_opencl_create(DeviceInfo& info, Stats &stats, bool background)
{
return new OpenCLDevice(info, stats, background);
}
void device_opencl_info(vector<DeviceInfo>& devices)
{
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)
return;
platform_ids.resize(num_platforms);
if(clGetPlatformIDs(num_platforms, &platform_ids[0], NULL) != CL_SUCCESS)
return;
/* devices are numbered consecutively across platforms */
int num_base = 0;
for (int platform = 0; platform < num_platforms; platform++, num_base += num_devices) {
num_devices = 0;
if(clGetDeviceIDs(platform_ids[platform], opencl_device_type(), 0, NULL, &num_devices) != CL_SUCCESS || num_devices == 0)
continue;
device_ids.resize(num_devices);
if(clGetDeviceIDs(platform_ids[platform], opencl_device_type(), num_devices, &device_ids[0], NULL) != CL_SUCCESS)
continue;
char pname[256];
clGetPlatformInfo(platform_ids[platform], CL_PLATFORM_NAME, sizeof(pname), &pname, NULL);
string platform_name = pname;
/* add devices */
for(int num = 0; num < num_devices; num++) {
cl_device_id device_id = device_ids[num];
char name[1024] = "\0";
if(clGetDeviceInfo(device_id, CL_DEVICE_NAME, sizeof(name), &name, NULL) != CL_SUCCESS)
continue;
DeviceInfo info;
info.type = DEVICE_OPENCL;
info.description = string(name);
info.num = num_base + num;
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;
devices.push_back(info);
}
}
}
CCL_NAMESPACE_END
#endif /* WITH_OPENCL */
|