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
|
/* SPDX-License-Identifier: Apache-2.0
* Copyright 2021-2022 Blender Foundation */
#ifdef WITH_METAL
# include "device/metal/device_impl.h"
# include "device/metal/device.h"
# include "util/debug.h"
# include "util/md5.h"
# include "util/path.h"
# include "util/time.h"
CCL_NAMESPACE_BEGIN
class MetalDevice;
BVHLayoutMask MetalDevice::get_bvh_layout_mask() const
{
return use_metalrt ? BVH_LAYOUT_METAL : BVH_LAYOUT_BVH2;
}
void MetalDevice::set_error(const string &error)
{
static std::mutex s_error_mutex;
std::lock_guard<std::mutex> lock(s_error_mutex);
Device::set_error(error);
if (first_error) {
fprintf(stderr, "\nRefer to the Cycles GPU rendering documentation for possible solutions:\n");
fprintf(stderr,
"https://docs.blender.org/manual/en/latest/render/cycles/gpu_rendering.html\n\n");
first_error = false;
}
}
MetalDevice::MetalDevice(const DeviceInfo &info, Stats &stats, Profiler &profiler)
: Device(info, stats, profiler), texture_info(this, "texture_info", MEM_GLOBAL)
{
mtlDevId = info.num;
/* select chosen device */
auto usable_devices = MetalInfo::get_usable_devices();
assert(mtlDevId < usable_devices.size());
mtlDevice = usable_devices[mtlDevId];
device_vendor = MetalInfo::get_device_vendor(mtlDevice);
assert(device_vendor != METAL_GPU_UNKNOWN);
metal_printf("Creating new Cycles device for Metal: %s\n", info.description.c_str());
/* determine default storage mode based on whether UMA is supported */
default_storage_mode = MTLResourceStorageModeManaged;
if (@available(macos 11.0, *)) {
if ([mtlDevice hasUnifiedMemory]) {
default_storage_mode = MTLResourceStorageModeShared;
init_host_memory();
}
}
texture_bindings_2d = [mtlDevice newBufferWithLength:4096 options:default_storage_mode];
texture_bindings_3d = [mtlDevice newBufferWithLength:4096 options:default_storage_mode];
stats.mem_alloc(texture_bindings_2d.allocatedSize + texture_bindings_3d.allocatedSize);
switch (device_vendor) {
default:
break;
case METAL_GPU_INTEL: {
max_threads_per_threadgroup = 64;
break;
}
case METAL_GPU_AMD: {
max_threads_per_threadgroup = 128;
break;
}
case METAL_GPU_APPLE: {
max_threads_per_threadgroup = 512;
use_metalrt = info.use_metalrt;
break;
}
}
if (auto metalrt = getenv("CYCLES_METALRT")) {
use_metalrt = (atoi(metalrt) != 0);
}
if (getenv("CYCLES_DEBUG_METAL_CAPTURE_KERNEL")) {
capture_enabled = true;
}
MTLArgumentDescriptor *arg_desc_params = [[MTLArgumentDescriptor alloc] init];
arg_desc_params.dataType = MTLDataTypePointer;
arg_desc_params.access = MTLArgumentAccessReadOnly;
arg_desc_params.arrayLength = sizeof(KernelParamsMetal) / sizeof(device_ptr);
mtlBufferKernelParamsEncoder = [mtlDevice newArgumentEncoderWithArguments:@[ arg_desc_params ]];
MTLArgumentDescriptor *arg_desc_texture = [[MTLArgumentDescriptor alloc] init];
arg_desc_texture.dataType = MTLDataTypeTexture;
arg_desc_texture.access = MTLArgumentAccessReadOnly;
mtlTextureArgEncoder = [mtlDevice newArgumentEncoderWithArguments:@[ arg_desc_texture ]];
/* command queue for non-tracing work on the GPU */
mtlGeneralCommandQueue = [mtlDevice newCommandQueue];
/* Acceleration structure arg encoder, if needed */
if (@available(macos 12.0, *)) {
if (use_metalrt) {
MTLArgumentDescriptor *arg_desc_as = [[MTLArgumentDescriptor alloc] init];
arg_desc_as.dataType = MTLDataTypeInstanceAccelerationStructure;
arg_desc_as.access = MTLArgumentAccessReadOnly;
mtlASArgEncoder = [mtlDevice newArgumentEncoderWithArguments:@[ arg_desc_as ]];
[arg_desc_as release];
}
}
/* Build the arg encoder for the ancillary bindings */
{
NSMutableArray *ancillary_desc = [[NSMutableArray alloc] init];
int index = 0;
MTLArgumentDescriptor *arg_desc_tex = [[MTLArgumentDescriptor alloc] init];
arg_desc_tex.dataType = MTLDataTypePointer;
arg_desc_tex.access = MTLArgumentAccessReadOnly;
arg_desc_tex.index = index++;
[ancillary_desc addObject:[arg_desc_tex copy]]; /* metal_tex_2d */
arg_desc_tex.index = index++;
[ancillary_desc addObject:[arg_desc_tex copy]]; /* metal_tex_3d */
[arg_desc_tex release];
if (@available(macos 12.0, *)) {
if (use_metalrt) {
MTLArgumentDescriptor *arg_desc_as = [[MTLArgumentDescriptor alloc] init];
arg_desc_as.dataType = MTLDataTypeInstanceAccelerationStructure;
arg_desc_as.access = MTLArgumentAccessReadOnly;
MTLArgumentDescriptor *arg_desc_ift = [[MTLArgumentDescriptor alloc] init];
arg_desc_ift.dataType = MTLDataTypeIntersectionFunctionTable;
arg_desc_ift.access = MTLArgumentAccessReadOnly;
arg_desc_as.index = index++;
[ancillary_desc addObject:[arg_desc_as copy]]; /* accel_struct */
arg_desc_ift.index = index++;
[ancillary_desc addObject:[arg_desc_ift copy]]; /* ift_default */
arg_desc_ift.index = index++;
[ancillary_desc addObject:[arg_desc_ift copy]]; /* ift_shadow */
arg_desc_ift.index = index++;
[ancillary_desc addObject:[arg_desc_ift copy]]; /* ift_local */
[arg_desc_ift release];
[arg_desc_as release];
}
}
mtlAncillaryArgEncoder = [mtlDevice newArgumentEncoderWithArguments:ancillary_desc];
for (int i = 0; i < ancillary_desc.count; i++) {
[ancillary_desc[i] release];
}
[ancillary_desc release];
}
[arg_desc_params release];
[arg_desc_texture release];
}
MetalDevice::~MetalDevice()
{
for (auto &tex : texture_slot_map) {
if (tex) {
[tex release];
tex = nil;
}
}
flush_delayed_free_list();
if (texture_bindings_2d) {
stats.mem_free(texture_bindings_2d.allocatedSize + texture_bindings_3d.allocatedSize);
[texture_bindings_2d release];
[texture_bindings_3d release];
}
[mtlTextureArgEncoder release];
[mtlBufferKernelParamsEncoder release];
[mtlASArgEncoder release];
[mtlAncillaryArgEncoder release];
[mtlGeneralCommandQueue release];
[mtlDevice release];
texture_info.free();
}
bool MetalDevice::support_device(const uint kernel_features /*requested_features*/)
{
return true;
}
bool MetalDevice::check_peer_access(Device *peer_device)
{
assert(0);
/* does peer access make sense? */
return false;
}
bool MetalDevice::use_adaptive_compilation()
{
return DebugFlags().metal.adaptive_compile;
}
string MetalDevice::get_source(const uint kernel_features)
{
string build_options;
if (use_adaptive_compilation()) {
build_options += " -D__KERNEL_FEATURES__=" + to_string(kernel_features);
}
if (use_metalrt) {
build_options += "-D__METALRT__ ";
if (motion_blur) {
build_options += "-D__METALRT_MOTION__ ";
}
}
# ifdef WITH_CYCLES_DEBUG
build_options += "-D__KERNEL_DEBUG__ ";
# endif
switch (device_vendor) {
default:
break;
case METAL_GPU_INTEL:
build_options += "-D__KERNEL_METAL_INTEL__ ";
break;
case METAL_GPU_AMD:
build_options += "-D__KERNEL_METAL_AMD__ ";
break;
case METAL_GPU_APPLE:
build_options += "-D__KERNEL_METAL_APPLE__ ";
break;
}
/* reformat -D defines list into compilable form */
vector<string> components;
string_replace(build_options, "-D", "");
string_split(components, build_options, " ");
string globalDefines;
for (const string &component : components) {
vector<string> assignments;
string_split(assignments, component, "=");
if (assignments.size() == 2)
globalDefines += string_printf(
"#define %s %s\n", assignments[0].c_str(), assignments[1].c_str());
else
globalDefines += string_printf("#define %s\n", assignments[0].c_str());
}
string source = globalDefines + "\n#include \"kernel/device/metal/kernel.metal\"\n";
source = path_source_replace_includes(source, path_get("source"));
metal_printf("Global defines:\n%s\n", globalDefines.c_str());
return source;
}
bool MetalDevice::load_kernels(const uint _kernel_features)
{
kernel_features = _kernel_features;
/* check if GPU is supported */
if (!support_device(kernel_features))
return false;
/* Keep track of whether motion blur is enabled, so to enable/disable motion in BVH builds
* This is necessary since objects may be reported to have motion if the Vector pass is
* active, but may still need to be rendered without motion blur if that isn't active as well. */
motion_blur = kernel_features & KERNEL_FEATURE_OBJECT_MOTION;
source[PSO_GENERIC] = get_source(kernel_features);
mtlLibrary[PSO_GENERIC] = compile(source[PSO_GENERIC]);
MD5Hash md5;
md5.append(source[PSO_GENERIC]);
source_md5[PSO_GENERIC] = md5.get_hex();
metal_printf("Front-end compilation finished (generic)\n");
bool result = MetalDeviceKernels::load(this, false);
reserve_local_memory(kernel_features);
return result;
}
id<MTLLibrary> MetalDevice::compile(string const &source)
{
MTLCompileOptions *options = [[MTLCompileOptions alloc] init];
options.fastMathEnabled = YES;
if (@available(macOS 12.0, *)) {
options.languageVersion = MTLLanguageVersion2_4;
}
NSError *error = NULL;
id<MTLLibrary> mtlLibrary = [mtlDevice newLibraryWithSource:@(source.c_str())
options:options
error:&error];
if (!mtlLibrary) {
NSString *err = [error localizedDescription];
set_error(string_printf("Failed to compile library:\n%s", [err UTF8String]));
}
[options release];
return mtlLibrary;
}
void MetalDevice::reserve_local_memory(const uint kernel_features)
{
/* METAL_WIP - implement this */
}
void MetalDevice::init_host_memory()
{
/* METAL_WIP - implement this */
}
void MetalDevice::load_texture_info()
{
if (need_texture_info) {
/* Unset flag before copying. */
need_texture_info = false;
texture_info.copy_to_device();
int num_textures = texture_info.size();
for (int tex = 0; tex < num_textures; tex++) {
uint64_t offset = tex * sizeof(void *);
id<MTLTexture> metal_texture = texture_slot_map[tex];
if (!metal_texture) {
[mtlTextureArgEncoder setArgumentBuffer:texture_bindings_2d offset:offset];
[mtlTextureArgEncoder setTexture:nil atIndex:0];
[mtlTextureArgEncoder setArgumentBuffer:texture_bindings_3d offset:offset];
[mtlTextureArgEncoder setTexture:nil atIndex:0];
}
else {
MTLTextureType type = metal_texture.textureType;
[mtlTextureArgEncoder setArgumentBuffer:texture_bindings_2d offset:offset];
[mtlTextureArgEncoder setTexture:type == MTLTextureType2D ? metal_texture : nil atIndex:0];
[mtlTextureArgEncoder setArgumentBuffer:texture_bindings_3d offset:offset];
[mtlTextureArgEncoder setTexture:type == MTLTextureType3D ? metal_texture : nil atIndex:0];
}
}
if (default_storage_mode == MTLResourceStorageModeManaged) {
[texture_bindings_2d didModifyRange:NSMakeRange(0, num_textures * sizeof(void *))];
[texture_bindings_3d didModifyRange:NSMakeRange(0, num_textures * sizeof(void *))];
}
}
}
void MetalDevice::erase_allocation(device_memory &mem)
{
stats.mem_free(mem.device_size);
mem.device_pointer = 0;
mem.device_size = 0;
auto it = metal_mem_map.find(&mem);
if (it != metal_mem_map.end()) {
MetalMem *mmem = it->second.get();
/* blank out reference to MetalMem* in the launch params (fixes crash T94736) */
if (mmem->pointer_index >= 0) {
device_ptr *pointers = (device_ptr *)&launch_params;
pointers[mmem->pointer_index] = 0;
}
metal_mem_map.erase(it);
}
}
MetalDevice::MetalMem *MetalDevice::generic_alloc(device_memory &mem)
{
size_t size = mem.memory_size();
mem.device_pointer = 0;
id<MTLBuffer> metal_buffer = nil;
MTLResourceOptions options = default_storage_mode;
/* Workaround for "bake" unit tests which fail if RenderBuffers is allocated with
* MTLResourceStorageModeShared. */
if (strstr(mem.name, "RenderBuffers")) {
options = MTLResourceStorageModeManaged;
}
if (size > 0) {
if (mem.type == MEM_DEVICE_ONLY && !capture_enabled) {
options = MTLResourceStorageModePrivate;
}
metal_buffer = [mtlDevice newBufferWithLength:size options:options];
if (!metal_buffer) {
set_error("System is out of GPU memory");
return nullptr;
}
}
if (mem.name) {
VLOG_WORK << "Buffer allocate: " << mem.name << ", "
<< string_human_readable_number(mem.memory_size()) << " bytes. ("
<< string_human_readable_size(mem.memory_size()) << ")";
}
mem.device_size = metal_buffer.allocatedSize;
stats.mem_alloc(mem.device_size);
metal_buffer.label = [[NSString alloc] initWithFormat:@"%s", mem.name];
std::lock_guard<std::recursive_mutex> lock(metal_mem_map_mutex);
assert(metal_mem_map.count(&mem) == 0); /* assert against double-alloc */
MetalMem *mmem = new MetalMem;
metal_mem_map[&mem] = std::unique_ptr<MetalMem>(mmem);
mmem->mem = &mem;
mmem->mtlBuffer = metal_buffer;
mmem->offset = 0;
mmem->size = size;
if (options != MTLResourceStorageModePrivate) {
mmem->hostPtr = [metal_buffer contents];
}
else {
mmem->hostPtr = nullptr;
}
/* encode device_pointer as (MetalMem*) in order to handle resource relocation and device pointer
* recalculation */
mem.device_pointer = device_ptr(mmem);
if (metal_buffer.storageMode == MTLResourceStorageModeShared) {
/* Replace host pointer with our host allocation. */
if (mem.host_pointer && mem.host_pointer != mmem->hostPtr) {
memcpy(mmem->hostPtr, mem.host_pointer, size);
mem.host_free();
mem.host_pointer = mmem->hostPtr;
}
mem.shared_pointer = mmem->hostPtr;
mem.shared_counter++;
mmem->use_UMA = true;
}
else {
mmem->use_UMA = false;
}
return mmem;
}
void MetalDevice::generic_copy_to(device_memory &mem)
{
if (!mem.host_pointer || !mem.device_pointer) {
return;
}
std::lock_guard<std::recursive_mutex> lock(metal_mem_map_mutex);
if (!metal_mem_map.at(&mem)->use_UMA || mem.host_pointer != mem.shared_pointer) {
MetalMem &mmem = *metal_mem_map.at(&mem);
memcpy(mmem.hostPtr, mem.host_pointer, mem.memory_size());
if (mmem.mtlBuffer.storageMode == MTLStorageModeManaged) {
[mmem.mtlBuffer didModifyRange:NSMakeRange(0, mem.memory_size())];
}
}
}
void MetalDevice::generic_free(device_memory &mem)
{
if (mem.device_pointer) {
std::lock_guard<std::recursive_mutex> lock(metal_mem_map_mutex);
MetalMem &mmem = *metal_mem_map.at(&mem);
size_t size = mmem.size;
/* If mmem.use_uma is true, reference counting is used
* to safely free memory. */
bool free_mtlBuffer = false;
if (mmem.use_UMA) {
assert(mem.shared_pointer);
if (mem.shared_pointer) {
assert(mem.shared_counter > 0);
if (--mem.shared_counter == 0) {
free_mtlBuffer = true;
}
}
}
else {
free_mtlBuffer = true;
}
if (free_mtlBuffer) {
if (mem.host_pointer && mem.host_pointer == mem.shared_pointer) {
/* Safely move the device-side data back to the host before it is freed. */
mem.host_pointer = mem.host_alloc(size);
memcpy(mem.host_pointer, mem.shared_pointer, size);
mmem.use_UMA = false;
}
mem.shared_pointer = 0;
/* Free device memory. */
delayed_free_list.push_back(mmem.mtlBuffer);
mmem.mtlBuffer = nil;
}
erase_allocation(mem);
}
}
void MetalDevice::mem_alloc(device_memory &mem)
{
if (mem.type == MEM_TEXTURE) {
assert(!"mem_alloc not supported for textures.");
}
else if (mem.type == MEM_GLOBAL) {
generic_alloc(mem);
}
else {
generic_alloc(mem);
}
}
void MetalDevice::mem_copy_to(device_memory &mem)
{
if (mem.type == MEM_GLOBAL) {
global_free(mem);
global_alloc(mem);
}
else if (mem.type == MEM_TEXTURE) {
tex_free((device_texture &)mem);
tex_alloc((device_texture &)mem);
}
else {
if (!mem.device_pointer) {
generic_alloc(mem);
}
generic_copy_to(mem);
}
}
void MetalDevice::mem_copy_from(device_memory &mem, size_t y, size_t w, size_t h, size_t elem)
{
if (mem.host_pointer) {
bool subcopy = (w >= 0 && h >= 0);
const size_t size = subcopy ? (elem * w * h) : mem.memory_size();
const size_t offset = subcopy ? (elem * y * w) : 0;
if (mem.device_pointer) {
std::lock_guard<std::recursive_mutex> lock(metal_mem_map_mutex);
MetalMem &mmem = *metal_mem_map.at(&mem);
if ([mmem.mtlBuffer storageMode] == MTLStorageModeManaged) {
id<MTLCommandBuffer> cmdBuffer = [mtlGeneralCommandQueue commandBuffer];
id<MTLBlitCommandEncoder> blitEncoder = [cmdBuffer blitCommandEncoder];
[blitEncoder synchronizeResource:mmem.mtlBuffer];
[blitEncoder endEncoding];
[cmdBuffer commit];
[cmdBuffer waitUntilCompleted];
}
if (mem.host_pointer != mmem.hostPtr) {
memcpy((uchar *)mem.host_pointer + offset, (uchar *)mmem.hostPtr + offset, size);
}
}
else {
memset((char *)mem.host_pointer + offset, 0, size);
}
}
}
void MetalDevice::mem_zero(device_memory &mem)
{
if (!mem.device_pointer) {
mem_alloc(mem);
}
if (!mem.device_pointer) {
return;
}
size_t size = mem.memory_size();
std::lock_guard<std::recursive_mutex> lock(metal_mem_map_mutex);
MetalMem &mmem = *metal_mem_map.at(&mem);
memset(mmem.hostPtr, 0, size);
if ([mmem.mtlBuffer storageMode] == MTLStorageModeManaged) {
[mmem.mtlBuffer didModifyRange:NSMakeRange(0, size)];
}
}
void MetalDevice::mem_free(device_memory &mem)
{
if (mem.type == MEM_GLOBAL) {
global_free(mem);
}
else if (mem.type == MEM_TEXTURE) {
tex_free((device_texture &)mem);
}
else {
generic_free(mem);
}
}
device_ptr MetalDevice::mem_alloc_sub_ptr(device_memory &mem, size_t offset, size_t /*size*/)
{
/* METAL_WIP - revive if necessary */
assert(0);
return 0;
}
void MetalDevice::const_copy_to(const char *name, void *host, size_t size)
{
if (strcmp(name, "data") == 0) {
assert(size == sizeof(KernelData));
memcpy((uint8_t *)&launch_params.data, host, sizeof(KernelData));
return;
}
auto update_launch_pointers =
[&](size_t offset, void *data, size_t data_size, size_t pointers_size) {
memcpy((uint8_t *)&launch_params + offset, data, data_size);
MetalMem **mmem = (MetalMem **)data;
int pointer_count = pointers_size / sizeof(device_ptr);
int pointer_index = offset / sizeof(device_ptr);
for (int i = 0; i < pointer_count; i++) {
if (mmem[i]) {
mmem[i]->pointer_index = pointer_index + i;
}
}
};
/* Update data storage pointers in launch parameters. */
if (strcmp(name, "integrator_state") == 0) {
/* IntegratorStateGPU is contiguous pointers */
const size_t pointer_block_size = sizeof(IntegratorStateGPU);
update_launch_pointers(
offsetof(KernelParamsMetal, integrator_state), host, size, pointer_block_size);
}
# define KERNEL_DATA_ARRAY(data_type, tex_name) \
else if (strcmp(name, #tex_name) == 0) \
{ \
update_launch_pointers(offsetof(KernelParamsMetal, tex_name), host, size, size); \
}
# include "kernel/data_arrays.h"
# undef KERNEL_DATA_ARRAY
}
void MetalDevice::global_alloc(device_memory &mem)
{
if (mem.is_resident(this)) {
generic_alloc(mem);
generic_copy_to(mem);
}
const_copy_to(mem.name, &mem.device_pointer, sizeof(mem.device_pointer));
}
void MetalDevice::global_free(device_memory &mem)
{
if (mem.is_resident(this) && mem.device_pointer) {
generic_free(mem);
}
}
void MetalDevice::tex_alloc_as_buffer(device_texture &mem)
{
generic_alloc(mem);
generic_copy_to(mem);
/* Resize once */
const uint slot = mem.slot;
if (slot >= texture_info.size()) {
/* Allocate some slots in advance, to reduce amount
* of re-allocations. */
texture_info.resize(round_up(slot + 1, 128));
}
mem.info.data = (uint64_t)mem.device_pointer;
/* Set Mapping and tag that we need to (re-)upload to device */
texture_info[slot] = mem.info;
need_texture_info = true;
}
void MetalDevice::tex_alloc(device_texture &mem)
{
/* Check that dimensions fit within maximum allowable size.
* See: https://developer.apple.com/metal/Metal-Feature-Set-Tables.pdf */
if (mem.data_width > 16384 || mem.data_height > 16384) {
set_error(string_printf(
"Texture exceeds maximum allowed size of 16384 x 16384 (requested: %zu x %zu)",
mem.data_width,
mem.data_height));
return;
}
MTLStorageMode storage_mode = MTLStorageModeManaged;
if (@available(macos 10.15, *)) {
if ([mtlDevice hasUnifiedMemory] &&
device_vendor !=
METAL_GPU_INTEL) { /* Intel GPUs don't support MTLStorageModeShared for MTLTextures */
storage_mode = MTLStorageModeShared;
}
}
/* General variables for both architectures */
string bind_name = mem.name;
size_t dsize = datatype_size(mem.data_type);
size_t size = mem.memory_size();
/* sampler_index maps into the GPU's constant 'metal_samplers' array */
uint64_t sampler_index = mem.info.extension;
if (mem.info.interpolation != INTERPOLATION_CLOSEST) {
sampler_index += 3;
}
/* Image Texture Storage */
MTLPixelFormat format;
switch (mem.data_type) {
case TYPE_UCHAR: {
MTLPixelFormat formats[] = {MTLPixelFormatR8Unorm,
MTLPixelFormatRG8Unorm,
MTLPixelFormatInvalid,
MTLPixelFormatRGBA8Unorm};
format = formats[mem.data_elements - 1];
} break;
case TYPE_UINT16: {
MTLPixelFormat formats[] = {MTLPixelFormatR16Unorm,
MTLPixelFormatRG16Unorm,
MTLPixelFormatInvalid,
MTLPixelFormatRGBA16Unorm};
format = formats[mem.data_elements - 1];
} break;
case TYPE_UINT: {
MTLPixelFormat formats[] = {MTLPixelFormatR32Uint,
MTLPixelFormatRG32Uint,
MTLPixelFormatInvalid,
MTLPixelFormatRGBA32Uint};
format = formats[mem.data_elements - 1];
} break;
case TYPE_INT: {
MTLPixelFormat formats[] = {MTLPixelFormatR32Sint,
MTLPixelFormatRG32Sint,
MTLPixelFormatInvalid,
MTLPixelFormatRGBA32Sint};
format = formats[mem.data_elements - 1];
} break;
case TYPE_FLOAT: {
MTLPixelFormat formats[] = {MTLPixelFormatR32Float,
MTLPixelFormatRG32Float,
MTLPixelFormatInvalid,
MTLPixelFormatRGBA32Float};
format = formats[mem.data_elements - 1];
} break;
case TYPE_HALF: {
MTLPixelFormat formats[] = {MTLPixelFormatR16Float,
MTLPixelFormatRG16Float,
MTLPixelFormatInvalid,
MTLPixelFormatRGBA16Float};
format = formats[mem.data_elements - 1];
} break;
default:
assert(0);
return;
}
assert(format != MTLPixelFormatInvalid);
id<MTLTexture> mtlTexture = nil;
size_t src_pitch = mem.data_width * dsize * mem.data_elements;
if (mem.data_depth > 1) {
/* 3D texture using array */
MTLTextureDescriptor *desc;
desc = [MTLTextureDescriptor texture2DDescriptorWithPixelFormat:format
width:mem.data_width
height:mem.data_height
mipmapped:NO];
desc.storageMode = storage_mode;
desc.usage = MTLTextureUsageShaderRead;
desc.textureType = MTLTextureType3D;
desc.depth = mem.data_depth;
VLOG_WORK << "Texture 3D allocate: " << mem.name << ", "
<< string_human_readable_number(mem.memory_size()) << " bytes. ("
<< string_human_readable_size(mem.memory_size()) << ")";
mtlTexture = [mtlDevice newTextureWithDescriptor:desc];
assert(mtlTexture);
if (!mtlTexture) {
return;
}
const size_t imageBytes = src_pitch * mem.data_height;
for (size_t d = 0; d < mem.data_depth; d++) {
const size_t offset = d * imageBytes;
[mtlTexture replaceRegion:MTLRegionMake3D(0, 0, d, mem.data_width, mem.data_height, 1)
mipmapLevel:0
slice:0
withBytes:(uint8_t *)mem.host_pointer + offset
bytesPerRow:src_pitch
bytesPerImage:0];
}
}
else if (mem.data_height > 0) {
/* 2D texture */
MTLTextureDescriptor *desc;
desc = [MTLTextureDescriptor texture2DDescriptorWithPixelFormat:format
width:mem.data_width
height:mem.data_height
mipmapped:NO];
desc.storageMode = storage_mode;
desc.usage = MTLTextureUsageShaderRead;
VLOG_WORK << "Texture 2D allocate: " << mem.name << ", "
<< string_human_readable_number(mem.memory_size()) << " bytes. ("
<< string_human_readable_size(mem.memory_size()) << ")";
mtlTexture = [mtlDevice newTextureWithDescriptor:desc];
assert(mtlTexture);
[mtlTexture replaceRegion:MTLRegionMake2D(0, 0, mem.data_width, mem.data_height)
mipmapLevel:0
withBytes:mem.host_pointer
bytesPerRow:src_pitch];
}
else {
assert(0);
/* 1D texture, using linear memory. */
}
mem.device_pointer = (device_ptr)mtlTexture;
mem.device_size = size;
stats.mem_alloc(size);
std::lock_guard<std::recursive_mutex> lock(metal_mem_map_mutex);
MetalMem *mmem = new MetalMem;
metal_mem_map[&mem] = std::unique_ptr<MetalMem>(mmem);
mmem->mem = &mem;
mmem->mtlTexture = mtlTexture;
/* Resize once */
const uint slot = mem.slot;
if (slot >= texture_info.size()) {
/* Allocate some slots in advance, to reduce amount
* of re-allocations. */
texture_info.resize(slot + 128);
texture_slot_map.resize(slot + 128);
ssize_t min_buffer_length = sizeof(void *) * texture_info.size();
if (!texture_bindings_2d || (texture_bindings_2d.length < min_buffer_length)) {
if (texture_bindings_2d) {
delayed_free_list.push_back(texture_bindings_2d);
delayed_free_list.push_back(texture_bindings_3d);
stats.mem_free(texture_bindings_2d.allocatedSize + texture_bindings_3d.allocatedSize);
}
texture_bindings_2d = [mtlDevice newBufferWithLength:min_buffer_length
options:default_storage_mode];
texture_bindings_3d = [mtlDevice newBufferWithLength:min_buffer_length
options:default_storage_mode];
stats.mem_alloc(texture_bindings_2d.allocatedSize + texture_bindings_3d.allocatedSize);
}
}
if (@available(macos 10.14, *)) {
/* Optimize the texture for GPU access. */
id<MTLCommandBuffer> commandBuffer = [mtlGeneralCommandQueue commandBuffer];
id<MTLBlitCommandEncoder> blitCommandEncoder = [commandBuffer blitCommandEncoder];
[blitCommandEncoder optimizeContentsForGPUAccess:mtlTexture];
[blitCommandEncoder endEncoding];
[commandBuffer commit];
}
/* Set Mapping and tag that we need to (re-)upload to device */
texture_slot_map[slot] = mtlTexture;
texture_info[slot] = mem.info;
need_texture_info = true;
texture_info[slot].data = uint64_t(slot) | (sampler_index << 32);
}
void MetalDevice::tex_free(device_texture &mem)
{
if (metal_mem_map.count(&mem)) {
std::lock_guard<std::recursive_mutex> lock(metal_mem_map_mutex);
MetalMem &mmem = *metal_mem_map.at(&mem);
assert(texture_slot_map[mem.slot] == mmem.mtlTexture);
texture_slot_map[mem.slot] = nil;
if (mmem.mtlTexture) {
/* Free bindless texture. */
delayed_free_list.push_back(mmem.mtlTexture);
mmem.mtlTexture = nil;
}
erase_allocation(mem);
}
}
unique_ptr<DeviceQueue> MetalDevice::gpu_queue_create()
{
return make_unique<MetalDeviceQueue>(this);
}
bool MetalDevice::should_use_graphics_interop()
{
/* METAL_WIP - provide fast interop */
return false;
}
void MetalDevice::flush_delayed_free_list()
{
/* free any Metal buffers that may have been freed by host while a command
* buffer was being generated. This function should be called after each
* completion of a command buffer */
std::lock_guard<std::recursive_mutex> lock(metal_mem_map_mutex);
for (auto &it : delayed_free_list) {
[it release];
}
delayed_free_list.clear();
}
void MetalDevice::build_bvh(BVH *bvh, Progress &progress, bool refit)
{
if (bvh->params.bvh_layout == BVH_LAYOUT_BVH2) {
Device::build_bvh(bvh, progress, refit);
return;
}
BVHMetal *bvh_metal = static_cast<BVHMetal *>(bvh);
bvh_metal->motion_blur = motion_blur;
if (bvh_metal->build(progress, mtlDevice, mtlGeneralCommandQueue, refit)) {
if (@available(macos 11.0, *)) {
if (bvh->params.top_level) {
bvhMetalRT = bvh_metal;
}
}
}
}
CCL_NAMESPACE_END
#endif
|