forked from intel/llvm
-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathmemory_manager.cpp
1679 lines (1506 loc) · 72.6 KB
/
memory_manager.cpp
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
//==-------------- memory_manager.cpp --------------------------------------==//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#include "ur_api.h"
#include <detail/context_impl.hpp>
#include <detail/device_image_impl.hpp>
#include <detail/event_impl.hpp>
#include <detail/mem_alloc_helper.hpp>
#include <detail/memory_manager.hpp>
#include <detail/queue_impl.hpp>
#include <detail/ur_utils.hpp>
#include <detail/xpti_registry.hpp>
#include <sycl/detail/ur.hpp>
#include <sycl/ext/oneapi/bindless_images_memory.hpp>
#include <sycl/usm/usm_enums.hpp>
#include <sycl/usm/usm_pointer_info.hpp>
#include <algorithm>
#include <cassert>
#include <cstring>
#include <vector>
#ifdef XPTI_ENABLE_INSTRUMENTATION
#include <xpti/xpti_data_types.h>
#include <xpti/xpti_trace_framework.hpp>
#endif
namespace sycl {
inline namespace _V1 {
namespace detail {
#ifdef XPTI_ENABLE_INSTRUMENTATION
uint8_t GMemAllocStreamID;
xpti::trace_event_data_t *GMemAllocEvent;
#endif
uint64_t emitMemAllocBeginTrace(uintptr_t ObjHandle, size_t AllocSize,
size_t GuardZone) {
(void)ObjHandle;
(void)AllocSize;
(void)GuardZone;
uint64_t CorrelationID = 0;
#ifdef XPTI_ENABLE_INSTRUMENTATION
constexpr uint16_t NotificationTraceType =
static_cast<uint16_t>(xpti::trace_point_type_t::mem_alloc_begin);
if (xptiCheckTraceEnabled(GMemAllocStreamID, NotificationTraceType)) {
xpti::mem_alloc_data_t MemAlloc{ObjHandle, 0 /* alloc ptr */, AllocSize,
GuardZone};
CorrelationID = xptiGetUniqueId();
xptiNotifySubscribers(GMemAllocStreamID, NotificationTraceType,
GMemAllocEvent, nullptr, CorrelationID, &MemAlloc);
}
#endif
return CorrelationID;
}
void emitMemAllocEndTrace(uintptr_t ObjHandle, uintptr_t AllocPtr,
size_t AllocSize, size_t GuardZone,
uint64_t CorrelationID) {
(void)ObjHandle;
(void)AllocPtr;
(void)AllocSize;
(void)GuardZone;
(void)CorrelationID;
#ifdef XPTI_ENABLE_INSTRUMENTATION
constexpr uint16_t NotificationTraceType =
static_cast<uint16_t>(xpti::trace_point_type_t::mem_alloc_end);
if (xptiCheckTraceEnabled(GMemAllocStreamID, NotificationTraceType)) {
xpti::mem_alloc_data_t MemAlloc{ObjHandle, AllocPtr, AllocSize, GuardZone};
xptiNotifySubscribers(GMemAllocStreamID, NotificationTraceType,
GMemAllocEvent, nullptr, CorrelationID, &MemAlloc);
}
#endif
}
uint64_t emitMemReleaseBeginTrace(uintptr_t ObjHandle, uintptr_t AllocPtr) {
(void)ObjHandle;
(void)AllocPtr;
uint64_t CorrelationID = 0;
#ifdef XPTI_ENABLE_INSTRUMENTATION
constexpr uint16_t NotificationTraceType =
static_cast<uint16_t>(xpti::trace_point_type_t::mem_release_begin);
if (xptiCheckTraceEnabled(GMemAllocStreamID, NotificationTraceType)) {
xpti::mem_alloc_data_t MemAlloc{ObjHandle, AllocPtr, 0 /* alloc size */,
0 /* guard zone */};
CorrelationID = xptiGetUniqueId();
xptiNotifySubscribers(GMemAllocStreamID, NotificationTraceType,
GMemAllocEvent, nullptr, CorrelationID, &MemAlloc);
}
#endif
return CorrelationID;
}
void emitMemReleaseEndTrace(uintptr_t ObjHandle, uintptr_t AllocPtr,
uint64_t CorrelationID) {
(void)ObjHandle;
(void)AllocPtr;
(void)CorrelationID;
#ifdef XPTI_ENABLE_INSTRUMENTATION
constexpr uint16_t NotificationTraceType =
static_cast<uint16_t>(xpti::trace_point_type_t::mem_release_end);
if (xptiCheckTraceEnabled(GMemAllocStreamID, NotificationTraceType)) {
xpti::mem_alloc_data_t MemAlloc{ObjHandle, AllocPtr, 0 /* alloc size */,
0 /* guard zone */};
xptiNotifySubscribers(GMemAllocStreamID, NotificationTraceType,
GMemAllocEvent, nullptr, CorrelationID, &MemAlloc);
}
#endif
}
static void waitForEvents(const std::vector<EventImplPtr> &Events) {
// Assuming all events will be on the same device or
// devices associated with the same Backend.
if (!Events.empty()) {
const AdapterPtr &Adapter = Events[0]->getAdapter();
std::vector<ur_event_handle_t> UrEvents(Events.size());
std::transform(
Events.begin(), Events.end(), UrEvents.begin(),
[](const EventImplPtr &EventImpl) { return EventImpl->getHandle(); });
if (!UrEvents.empty() && UrEvents[0]) {
Adapter->call<UrApiKind::urEventWait>(UrEvents.size(), &UrEvents[0]);
}
}
}
void memBufferCreateHelper(const AdapterPtr &Adapter, ur_context_handle_t Ctx,
ur_mem_flags_t Flags, size_t Size,
ur_mem_handle_t *RetMem,
const ur_buffer_properties_t *Props) {
#ifdef XPTI_ENABLE_INSTRUMENTATION
uint64_t CorrID = 0;
#endif
// We only want to instrument urMemBufferCreate
{
#ifdef XPTI_ENABLE_INSTRUMENTATION
CorrID =
emitMemAllocBeginTrace(0 /* mem object */, Size, 0 /* guard zone */);
xpti::utils::finally _{[&] {
// C-style cast is required for MSVC
uintptr_t MemObjID = (uintptr_t)(*RetMem);
ur_native_handle_t Ptr = 0;
// Always use call_nocheck here, because call may throw an exception,
// and this lambda will be called from destructor, which in combination
// rewards us with UB.
// When doing buffer interop we don't know what device the memory should
// be resident on, so pass nullptr for Device param. Buffer interop may
// not be supported by all backends.
Adapter->call_nocheck<UrApiKind::urMemGetNativeHandle>(
*RetMem, /*Dev*/ nullptr, &Ptr);
emitMemAllocEndTrace(MemObjID, (uintptr_t)(Ptr), Size, 0 /* guard zone */,
CorrID);
}};
#endif
if (Size)
Adapter->call<UrApiKind::urMemBufferCreate>(Ctx, Flags, Size, Props,
RetMem);
}
}
void memReleaseHelper(const AdapterPtr &Adapter, ur_mem_handle_t Mem) {
// FIXME urMemRelease does not guarante memory release. It is only true if
// reference counter is 1. However, SYCL runtime currently only calls
// urMemRetain only for OpenCL interop
#ifdef XPTI_ENABLE_INSTRUMENTATION
uint64_t CorrID = 0;
// C-style cast is required for MSVC
uintptr_t MemObjID = (uintptr_t)(Mem);
uintptr_t Ptr = 0;
// Do not make unnecessary UR calls without instrumentation enabled
if (xptiTraceEnabled()) {
ur_native_handle_t PtrHandle = 0;
// When doing buffer interop we don't know what device the memory should be
// resident on, so pass nullptr for Device param. Buffer interop may not be
// supported by all backends.
Adapter->call_nocheck<UrApiKind::urMemGetNativeHandle>(Mem, /*Dev*/ nullptr,
&PtrHandle);
Ptr = (uintptr_t)(PtrHandle);
}
#endif
// We only want to instrument urMemRelease
{
#ifdef XPTI_ENABLE_INSTRUMENTATION
CorrID = emitMemReleaseBeginTrace(MemObjID, Ptr);
xpti::utils::finally _{
[&] { emitMemReleaseEndTrace(MemObjID, Ptr, CorrID); }};
#endif
Adapter->call<UrApiKind::urMemRelease>(Mem);
}
}
void memBufferMapHelper(const AdapterPtr &Adapter, ur_queue_handle_t Queue,
ur_mem_handle_t Buffer, bool Blocking,
ur_map_flags_t Flags, size_t Offset, size_t Size,
uint32_t NumEvents, const ur_event_handle_t *WaitList,
ur_event_handle_t *Event, void **RetMap) {
#ifdef XPTI_ENABLE_INSTRUMENTATION
uint64_t CorrID = 0;
uintptr_t MemObjID = (uintptr_t)(Buffer);
#endif
// We only want to instrument urEnqueueMemBufferMap
#ifdef XPTI_ENABLE_INSTRUMENTATION
CorrID = emitMemAllocBeginTrace(MemObjID, Size, 0 /* guard zone */);
xpti::utils::finally _{[&] {
emitMemAllocEndTrace(MemObjID, (uintptr_t)(*RetMap), Size,
0 /* guard zone */, CorrID);
}};
#endif
Adapter->call<UrApiKind::urEnqueueMemBufferMap>(
Queue, Buffer, Blocking, Flags, Offset, Size, NumEvents, WaitList, Event,
RetMap);
}
void memUnmapHelper(const AdapterPtr &Adapter, ur_queue_handle_t Queue,
ur_mem_handle_t Mem, void *MappedPtr, uint32_t NumEvents,
const ur_event_handle_t *WaitList,
ur_event_handle_t *Event) {
#ifdef XPTI_ENABLE_INSTRUMENTATION
uint64_t CorrID = 0;
uintptr_t MemObjID = (uintptr_t)(Mem);
uintptr_t Ptr = (uintptr_t)(MappedPtr);
#endif
// We only want to instrument urEnqueueMemUnmap
{
#ifdef XPTI_ENABLE_INSTRUMENTATION
CorrID = emitMemReleaseBeginTrace(MemObjID, Ptr);
xpti::utils::finally _{[&] {
// There's no way for SYCL to know, when the pointer is freed, so we have
// to explicitly wait for the end of data transfers here in order to
// report correct events.
// Always use call_nocheck here, because call may throw an exception,
// and this lambda will be called from destructor, which in combination
// rewards us with UB.
Adapter->call_nocheck<UrApiKind::urEventWait>(1, Event);
emitMemReleaseEndTrace(MemObjID, Ptr, CorrID);
}};
#endif
Adapter->call<UrApiKind::urEnqueueMemUnmap>(Queue, Mem, MappedPtr,
NumEvents, WaitList, Event);
}
}
void MemoryManager::release(ContextImplPtr TargetContext, SYCLMemObjI *MemObj,
void *MemAllocation,
std::vector<EventImplPtr> DepEvents,
ur_event_handle_t &OutEvent) {
// There is no async API for memory releasing. Explicitly wait for all
// dependency events and return empty event.
waitForEvents(DepEvents);
OutEvent = nullptr;
XPTIRegistry::bufferReleaseNotification(MemObj, MemAllocation);
MemObj->releaseMem(TargetContext, MemAllocation);
}
void MemoryManager::releaseMemObj(ContextImplPtr TargetContext,
SYCLMemObjI *MemObj, void *MemAllocation,
void *UserPtr) {
if (UserPtr == MemAllocation) {
// Do nothing as it's user provided memory.
return;
}
if (!TargetContext) {
MemObj->releaseHostMem(MemAllocation);
return;
}
const AdapterPtr &Adapter = TargetContext->getAdapter();
memReleaseHelper(Adapter, ur::cast<ur_mem_handle_t>(MemAllocation));
}
void *MemoryManager::allocate(ContextImplPtr TargetContext, SYCLMemObjI *MemObj,
bool InitFromUserData, void *HostPtr,
std::vector<EventImplPtr> DepEvents,
ur_event_handle_t &OutEvent) {
// There is no async API for memory allocation. Explicitly wait for all
// dependency events and return empty event.
waitForEvents(DepEvents);
OutEvent = nullptr;
return MemObj->allocateMem(TargetContext, InitFromUserData, HostPtr,
OutEvent);
}
void *MemoryManager::allocateHostMemory(SYCLMemObjI *MemObj, void *UserPtr,
bool HostPtrReadOnly, size_t Size,
const sycl::property_list &) {
std::ignore = HostPtrReadOnly;
std::ignore = Size;
// Can return user pointer directly if it is not a nullptr.
if (UserPtr)
return UserPtr;
return MemObj->allocateHostMem();
}
void *MemoryManager::allocateInteropMemObject(
ContextImplPtr TargetContext, void *UserPtr,
const EventImplPtr &InteropEvent, const ContextImplPtr &InteropContext,
const sycl::property_list &, ur_event_handle_t &OutEventToWait) {
(void)TargetContext;
(void)InteropContext;
// If memory object is created with interop c'tor return cl_mem as is.
assert(TargetContext == InteropContext && "Expected matching contexts");
OutEventToWait = InteropEvent->getHandle();
// Retain the event since it will be released during alloca command
// destruction
if (nullptr != OutEventToWait) {
const AdapterPtr &Adapter = InteropEvent->getAdapter();
Adapter->call<UrApiKind::urEventRetain>(OutEventToWait);
}
return UserPtr;
}
static ur_mem_flags_t getMemObjCreationFlags(void *UserPtr,
bool HostPtrReadOnly) {
// Create read_write mem object to handle arbitrary uses.
ur_mem_flags_t Result =
HostPtrReadOnly ? UR_MEM_FLAG_READ_ONLY : UR_MEM_FLAG_READ_WRITE;
if (UserPtr)
Result |= UR_MEM_FLAG_USE_HOST_POINTER;
return Result;
}
void *MemoryManager::allocateImageObject(ContextImplPtr TargetContext,
void *UserPtr, bool HostPtrReadOnly,
const ur_image_desc_t &Desc,
const ur_image_format_t &Format,
const sycl::property_list &) {
ur_mem_flags_t CreationFlags =
getMemObjCreationFlags(UserPtr, HostPtrReadOnly);
ur_mem_handle_t NewMem = nullptr;
const AdapterPtr &Adapter = TargetContext->getAdapter();
Adapter->call<UrApiKind::urMemImageCreate>(TargetContext->getHandleRef(),
CreationFlags, &Format, &Desc,
UserPtr, &NewMem);
return NewMem;
}
void *
MemoryManager::allocateBufferObject(ContextImplPtr TargetContext, void *UserPtr,
bool HostPtrReadOnly, const size_t Size,
const sycl::property_list &PropsList) {
ur_mem_flags_t CreationFlags =
getMemObjCreationFlags(UserPtr, HostPtrReadOnly);
if (PropsList.has_property<
sycl::ext::oneapi::property::buffer::use_pinned_host_memory>())
CreationFlags |= UR_MEM_FLAG_ALLOC_HOST_POINTER;
ur_mem_handle_t NewMem = nullptr;
const AdapterPtr &Adapter = TargetContext->getAdapter();
ur_buffer_properties_t AllocProps = {UR_STRUCTURE_TYPE_BUFFER_PROPERTIES,
nullptr, UserPtr};
void **Next = &AllocProps.pNext;
ur_buffer_alloc_location_properties_t LocationProperties = {
UR_STRUCTURE_TYPE_BUFFER_ALLOC_LOCATION_PROPERTIES, nullptr, 0};
if (PropsList.has_property<property::buffer::detail::buffer_location>() &&
TargetContext->isBufferLocationSupported()) {
LocationProperties.location =
PropsList.get_property<property::buffer::detail::buffer_location>()
.get_buffer_location();
*Next = &LocationProperties;
Next = &LocationProperties.pNext;
}
ur_buffer_channel_properties_t ChannelProperties = {
UR_STRUCTURE_TYPE_BUFFER_CHANNEL_PROPERTIES, nullptr, 0};
if (PropsList.has_property<property::buffer::mem_channel>()) {
ChannelProperties.channel =
PropsList.get_property<property::buffer::mem_channel>().get_channel();
*Next = &ChannelProperties;
}
memBufferCreateHelper(Adapter, TargetContext->getHandleRef(), CreationFlags,
Size, &NewMem, &AllocProps);
return NewMem;
}
void *MemoryManager::allocateMemBuffer(ContextImplPtr TargetContext,
SYCLMemObjI *MemObj, void *UserPtr,
bool HostPtrReadOnly, size_t Size,
const EventImplPtr &InteropEvent,
const ContextImplPtr &InteropContext,
const sycl::property_list &PropsList,
ur_event_handle_t &OutEventToWait) {
void *MemPtr;
if (!TargetContext)
MemPtr =
allocateHostMemory(MemObj, UserPtr, HostPtrReadOnly, Size, PropsList);
else if (UserPtr && InteropContext)
MemPtr =
allocateInteropMemObject(TargetContext, UserPtr, InteropEvent,
InteropContext, PropsList, OutEventToWait);
else
MemPtr = allocateBufferObject(TargetContext, UserPtr, HostPtrReadOnly, Size,
PropsList);
XPTIRegistry::bufferAssociateNotification(MemObj, MemPtr);
return MemPtr;
}
void *MemoryManager::allocateMemImage(
ContextImplPtr TargetContext, SYCLMemObjI *MemObj, void *UserPtr,
bool HostPtrReadOnly, size_t Size, const ur_image_desc_t &Desc,
const ur_image_format_t &Format, const EventImplPtr &InteropEvent,
const ContextImplPtr &InteropContext, const sycl::property_list &PropsList,
ur_event_handle_t &OutEventToWait) {
if (!TargetContext)
return allocateHostMemory(MemObj, UserPtr, HostPtrReadOnly, Size,
PropsList);
if (UserPtr && InteropContext)
return allocateInteropMemObject(TargetContext, UserPtr, InteropEvent,
InteropContext, PropsList, OutEventToWait);
return allocateImageObject(TargetContext, UserPtr, HostPtrReadOnly, Desc,
Format, PropsList);
}
void *MemoryManager::allocateMemSubBuffer(ContextImplPtr TargetContext,
void *ParentMemObj, size_t ElemSize,
size_t Offset, range<3> Range,
std::vector<EventImplPtr> DepEvents,
ur_event_handle_t &OutEvent) {
waitForEvents(DepEvents);
OutEvent = nullptr;
if (!TargetContext)
return static_cast<void *>(static_cast<char *>(ParentMemObj) + Offset);
size_t SizeInBytes = ElemSize;
for (size_t I = 0; I < 3; ++I)
SizeInBytes *= Range[I];
ur_result_t Error = UR_RESULT_SUCCESS;
ur_buffer_region_t Region = {UR_STRUCTURE_TYPE_BUFFER_REGION, nullptr, Offset,
SizeInBytes};
ur_mem_handle_t NewMem;
const AdapterPtr &Adapter = TargetContext->getAdapter();
Error = Adapter->call_nocheck<UrApiKind::urMemBufferPartition>(
ur::cast<ur_mem_handle_t>(ParentMemObj), UR_MEM_FLAG_READ_WRITE,
UR_BUFFER_CREATE_TYPE_REGION, &Region, &NewMem);
if (Error == UR_RESULT_ERROR_MISALIGNED_SUB_BUFFER_OFFSET)
throw detail::set_ur_error(
exception(make_error_code(errc::invalid),
"Specified offset of the sub-buffer being constructed is not "
"a multiple of the memory base address alignment"),
Error);
Adapter->checkUrResult(Error);
return NewMem;
}
struct TermPositions {
int XTerm;
int YTerm;
int ZTerm;
};
void prepTermPositions(TermPositions &pos, int Dimensions,
detail::SYCLMemObjI::MemObjType Type) {
// For buffers, the offsets/ranges coming from accessor are always
// id<3>/range<3> But their organization varies by dimension:
// 1 ==> {width, 1, 1}
// 2 ==> {height, width, 1}
// 3 ==> {depth, height, width}
// Some callers schedule 0 as DimDst/DimSrc.
if (Type == detail::SYCLMemObjI::MemObjType::Buffer) {
if (Dimensions == 3) {
pos.XTerm = 2, pos.YTerm = 1, pos.ZTerm = 0;
} else if (Dimensions == 2) {
pos.XTerm = 1, pos.YTerm = 0, pos.ZTerm = 2;
} else { // Dimension is 1 or 0
pos.XTerm = 0, pos.YTerm = 1, pos.ZTerm = 2;
}
} else { // While range<>/id<> use by images is different than buffers, it's
// consistent with their accessors.
pos.XTerm = 0;
pos.YTerm = 1;
pos.ZTerm = 2;
}
}
void copyH2D(SYCLMemObjI *SYCLMemObj, char *SrcMem, QueueImplPtr,
unsigned int DimSrc, sycl::range<3> SrcSize,
sycl::range<3> SrcAccessRange, sycl::id<3> SrcOffset,
unsigned int SrcElemSize, ur_mem_handle_t DstMem,
QueueImplPtr TgtQueue, unsigned int DimDst, sycl::range<3> DstSize,
sycl::range<3> DstAccessRange, sycl::id<3> DstOffset,
unsigned int DstElemSize, std::vector<ur_event_handle_t> DepEvents,
ur_event_handle_t &OutEvent,
const detail::EventImplPtr &OutEventImpl) {
(void)SrcAccessRange;
assert(SYCLMemObj && "The SYCLMemObj is nullptr");
assert(TgtQueue && "Destination mem object queue must be not nullptr");
const ur_queue_handle_t Queue = TgtQueue->getHandleRef();
const AdapterPtr &Adapter = TgtQueue->getAdapter();
detail::SYCLMemObjI::MemObjType MemType = SYCLMemObj->getType();
TermPositions SrcPos, DstPos;
prepTermPositions(SrcPos, DimSrc, MemType);
prepTermPositions(DstPos, DimDst, MemType);
size_t DstXOffBytes = DstOffset[DstPos.XTerm] * DstElemSize;
size_t SrcXOffBytes = SrcOffset[SrcPos.XTerm] * SrcElemSize;
size_t DstAccessRangeWidthBytes = DstAccessRange[DstPos.XTerm] * DstElemSize;
size_t DstSzWidthBytes = DstSize[DstPos.XTerm] * DstElemSize;
size_t SrcSzWidthBytes = SrcSize[SrcPos.XTerm] * SrcElemSize;
if (MemType == detail::SYCLMemObjI::MemObjType::Buffer) {
if (1 == DimDst && 1 == DimSrc) {
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
Adapter->call<UrApiKind::urEnqueueMemBufferWrite>(
Queue, DstMem,
/*blocking_write=*/false, DstXOffBytes, DstAccessRangeWidthBytes,
SrcMem + SrcXOffBytes, DepEvents.size(), DepEvents.data(), &OutEvent);
} else {
size_t BufferRowPitch = (1 == DimDst) ? 0 : DstSzWidthBytes;
size_t BufferSlicePitch =
(3 == DimDst) ? DstSzWidthBytes * DstSize[DstPos.YTerm] : 0;
size_t HostRowPitch = (1 == DimSrc) ? 0 : SrcSzWidthBytes;
size_t HostSlicePitch =
(3 == DimSrc) ? SrcSzWidthBytes * SrcSize[SrcPos.YTerm] : 0;
ur_rect_offset_t BufferOffset{DstXOffBytes, DstOffset[DstPos.YTerm],
DstOffset[DstPos.ZTerm]};
ur_rect_offset_t HostOffset{SrcXOffBytes, SrcOffset[SrcPos.YTerm],
SrcOffset[SrcPos.ZTerm]};
ur_rect_region_t RectRegion{DstAccessRangeWidthBytes,
DstAccessRange[DstPos.YTerm],
DstAccessRange[DstPos.ZTerm]};
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
Adapter->call<UrApiKind::urEnqueueMemBufferWriteRect>(
Queue, DstMem,
/*blocking_write=*/false, BufferOffset, HostOffset, RectRegion,
BufferRowPitch, BufferSlicePitch, HostRowPitch, HostSlicePitch,
SrcMem, DepEvents.size(), DepEvents.data(), &OutEvent);
}
} else {
size_t InputRowPitch = (1 == DimDst) ? 0 : DstSzWidthBytes;
size_t InputSlicePitch =
(3 == DimDst) ? DstSzWidthBytes * DstSize[DstPos.YTerm] : 0;
ur_rect_offset_t Origin{DstOffset[DstPos.XTerm], DstOffset[DstPos.YTerm],
DstOffset[DstPos.ZTerm]};
ur_rect_region_t Region{DstAccessRange[DstPos.XTerm],
DstAccessRange[DstPos.YTerm],
DstAccessRange[DstPos.ZTerm]};
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
Adapter->call<UrApiKind::urEnqueueMemImageWrite>(
Queue, DstMem,
/*blocking_write=*/false, Origin, Region, InputRowPitch,
InputSlicePitch, SrcMem, DepEvents.size(), DepEvents.data(), &OutEvent);
}
}
void copyD2H(SYCLMemObjI *SYCLMemObj, ur_mem_handle_t SrcMem,
QueueImplPtr SrcQueue, unsigned int DimSrc, sycl::range<3> SrcSize,
sycl::range<3> SrcAccessRange, sycl::id<3> SrcOffset,
unsigned int SrcElemSize, char *DstMem, QueueImplPtr,
unsigned int DimDst, sycl::range<3> DstSize,
sycl::range<3> DstAccessRange, sycl::id<3> DstOffset,
unsigned int DstElemSize, std::vector<ur_event_handle_t> DepEvents,
ur_event_handle_t &OutEvent,
const detail::EventImplPtr &OutEventImpl) {
(void)DstAccessRange;
assert(SYCLMemObj && "The SYCLMemObj is nullptr");
assert(SrcQueue && "Source mem object queue is expected to be not nullptr");
const ur_queue_handle_t Queue = SrcQueue->getHandleRef();
const AdapterPtr &Adapter = SrcQueue->getAdapter();
detail::SYCLMemObjI::MemObjType MemType = SYCLMemObj->getType();
TermPositions SrcPos, DstPos;
prepTermPositions(SrcPos, DimSrc, MemType);
prepTermPositions(DstPos, DimDst, MemType);
// For a given buffer, the various mem copy routines (copyD2H, copyH2D,
// copyD2D) will usually have the same values for AccessRange, Size,
// Dimension, Offset, etc. EXCEPT when the dtor for ~SYCLMemObjT is called.
// Essentially, it schedules a copyBack of chars thus in copyD2H the
// Dimension will then be 1 and DstAccessRange[0] and DstSize[0] will be
// sized to bytes with a DstElemSize of 1.
size_t DstXOffBytes = DstOffset[DstPos.XTerm] * DstElemSize;
size_t SrcXOffBytes = SrcOffset[SrcPos.XTerm] * SrcElemSize;
size_t SrcAccessRangeWidthBytes = SrcAccessRange[SrcPos.XTerm] * SrcElemSize;
size_t DstSzWidthBytes = DstSize[DstPos.XTerm] * DstElemSize;
size_t SrcSzWidthBytes = SrcSize[SrcPos.XTerm] * SrcElemSize;
if (MemType == detail::SYCLMemObjI::MemObjType::Buffer) {
if (1 == DimDst && 1 == DimSrc) {
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
Adapter->call<UrApiKind::urEnqueueMemBufferRead>(
Queue, SrcMem,
/*blocking_read=*/false, SrcXOffBytes, SrcAccessRangeWidthBytes,
DstMem + DstXOffBytes, DepEvents.size(), DepEvents.data(), &OutEvent);
} else {
size_t BufferRowPitch = (1 == DimSrc) ? 0 : SrcSzWidthBytes;
size_t BufferSlicePitch =
(3 == DimSrc) ? SrcSzWidthBytes * SrcSize[SrcPos.YTerm] : 0;
size_t HostRowPitch = (1 == DimDst) ? 0 : DstSzWidthBytes;
size_t HostSlicePitch =
(3 == DimDst) ? DstSzWidthBytes * DstSize[DstPos.YTerm] : 0;
ur_rect_offset_t BufferOffset{SrcXOffBytes, SrcOffset[SrcPos.YTerm],
SrcOffset[SrcPos.ZTerm]};
ur_rect_offset_t HostOffset{DstXOffBytes, DstOffset[DstPos.YTerm],
DstOffset[DstPos.ZTerm]};
ur_rect_region_t RectRegion{SrcAccessRangeWidthBytes,
SrcAccessRange[SrcPos.YTerm],
SrcAccessRange[SrcPos.ZTerm]};
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
Adapter->call<UrApiKind::urEnqueueMemBufferReadRect>(
Queue, SrcMem,
/*blocking_read=*/false, BufferOffset, HostOffset, RectRegion,
BufferRowPitch, BufferSlicePitch, HostRowPitch, HostSlicePitch,
DstMem, DepEvents.size(), DepEvents.data(), &OutEvent);
}
} else {
size_t RowPitch = (1 == DimSrc) ? 0 : SrcSzWidthBytes;
size_t SlicePitch =
(3 == DimSrc) ? SrcSzWidthBytes * SrcSize[SrcPos.YTerm] : 0;
ur_rect_offset_t Offset{SrcOffset[SrcPos.XTerm], SrcOffset[SrcPos.YTerm],
SrcOffset[SrcPos.ZTerm]};
ur_rect_region_t Region{SrcAccessRange[SrcPos.XTerm],
SrcAccessRange[SrcPos.YTerm],
SrcAccessRange[SrcPos.ZTerm]};
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
Adapter->call<UrApiKind::urEnqueueMemImageRead>(
Queue, SrcMem, false, Offset, Region, RowPitch, SlicePitch, DstMem,
DepEvents.size(), DepEvents.data(), &OutEvent);
}
}
void copyD2D(SYCLMemObjI *SYCLMemObj, ur_mem_handle_t SrcMem,
QueueImplPtr SrcQueue, unsigned int DimSrc, sycl::range<3> SrcSize,
sycl::range<3> SrcAccessRange, sycl::id<3> SrcOffset,
unsigned int SrcElemSize, ur_mem_handle_t DstMem, QueueImplPtr,
unsigned int DimDst, sycl::range<3> DstSize, sycl::range<3>,
sycl::id<3> DstOffset, unsigned int DstElemSize,
std::vector<ur_event_handle_t> DepEvents,
ur_event_handle_t &OutEvent,
const detail::EventImplPtr &OutEventImpl) {
assert(SYCLMemObj && "The SYCLMemObj is nullptr");
assert(SrcQueue && "Source mem object and target mem object queues are "
"expected to be not nullptr");
const ur_queue_handle_t Queue = SrcQueue->getHandleRef();
const AdapterPtr &Adapter = SrcQueue->getAdapter();
detail::SYCLMemObjI::MemObjType MemType = SYCLMemObj->getType();
TermPositions SrcPos, DstPos;
prepTermPositions(SrcPos, DimSrc, MemType);
prepTermPositions(DstPos, DimDst, MemType);
size_t DstXOffBytes = DstOffset[DstPos.XTerm] * DstElemSize;
size_t SrcXOffBytes = SrcOffset[SrcPos.XTerm] * SrcElemSize;
size_t SrcAccessRangeWidthBytes = SrcAccessRange[SrcPos.XTerm] * SrcElemSize;
size_t DstSzWidthBytes = DstSize[DstPos.XTerm] * DstElemSize;
size_t SrcSzWidthBytes = SrcSize[SrcPos.XTerm] * SrcElemSize;
if (MemType == detail::SYCLMemObjI::MemObjType::Buffer) {
if (1 == DimDst && 1 == DimSrc) {
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
Adapter->call<UrApiKind::urEnqueueMemBufferCopy>(
Queue, SrcMem, DstMem, SrcXOffBytes, DstXOffBytes,
SrcAccessRangeWidthBytes, DepEvents.size(), DepEvents.data(),
&OutEvent);
} else {
// passing 0 for pitches not allowed. Because clEnqueueCopyBufferRect will
// calculate both src and dest pitch using region[0], which is not correct
// if src and dest are not the same size.
size_t SrcRowPitch = SrcSzWidthBytes;
size_t SrcSlicePitch = (DimSrc <= 1)
? SrcSzWidthBytes
: SrcSzWidthBytes * SrcSize[SrcPos.YTerm];
size_t DstRowPitch = DstSzWidthBytes;
size_t DstSlicePitch = (DimDst <= 1)
? DstSzWidthBytes
: DstSzWidthBytes * DstSize[DstPos.YTerm];
ur_rect_offset_t SrcOrigin{SrcXOffBytes, SrcOffset[SrcPos.YTerm],
SrcOffset[SrcPos.ZTerm]};
ur_rect_offset_t DstOrigin{DstXOffBytes, DstOffset[DstPos.YTerm],
DstOffset[DstPos.ZTerm]};
ur_rect_region_t Region{SrcAccessRangeWidthBytes,
SrcAccessRange[SrcPos.YTerm],
SrcAccessRange[SrcPos.ZTerm]};
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
Adapter->call<UrApiKind::urEnqueueMemBufferCopyRect>(
Queue, SrcMem, DstMem, SrcOrigin, DstOrigin, Region, SrcRowPitch,
SrcSlicePitch, DstRowPitch, DstSlicePitch, DepEvents.size(),
DepEvents.data(), &OutEvent);
}
} else {
ur_rect_offset_t SrcOrigin{SrcOffset[SrcPos.XTerm], SrcOffset[SrcPos.YTerm],
SrcOffset[SrcPos.ZTerm]};
ur_rect_offset_t DstOrigin{DstOffset[DstPos.XTerm], DstOffset[DstPos.YTerm],
DstOffset[DstPos.ZTerm]};
ur_rect_region_t Region{SrcAccessRange[SrcPos.XTerm],
SrcAccessRange[SrcPos.YTerm],
SrcAccessRange[SrcPos.ZTerm]};
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
Adapter->call<UrApiKind::urEnqueueMemImageCopy>(
Queue, SrcMem, DstMem, SrcOrigin, DstOrigin, Region, DepEvents.size(),
DepEvents.data(), &OutEvent);
}
}
static void copyH2H(SYCLMemObjI *, char *SrcMem, QueueImplPtr,
unsigned int DimSrc, sycl::range<3> SrcSize,
sycl::range<3> SrcAccessRange, sycl::id<3> SrcOffset,
unsigned int SrcElemSize, char *DstMem, QueueImplPtr,
unsigned int DimDst, sycl::range<3> DstSize,
sycl::range<3> DstAccessRange, sycl::id<3> DstOffset,
unsigned int DstElemSize, std::vector<ur_event_handle_t>,
ur_event_handle_t &, const detail::EventImplPtr &) {
if ((DimSrc != 1 || DimDst != 1) &&
(SrcOffset != id<3>{0, 0, 0} || DstOffset != id<3>{0, 0, 0} ||
SrcSize != SrcAccessRange || DstSize != DstAccessRange)) {
throw exception(make_error_code(errc::feature_not_supported),
"Not supported configuration of memcpy requested");
}
SrcMem += SrcOffset[0] * SrcElemSize;
DstMem += DstOffset[0] * DstElemSize;
if (SrcMem == DstMem)
return;
size_t BytesToCopy =
SrcAccessRange[0] * SrcElemSize * SrcAccessRange[1] * SrcAccessRange[2];
std::memcpy(DstMem, SrcMem, BytesToCopy);
}
// Copies memory between: host and device, host and host,
// device and device if memory objects bound to the one context.
void MemoryManager::copy(
SYCLMemObjI *SYCLMemObj, void *SrcMem, QueueImplPtr SrcQueue,
unsigned int DimSrc, sycl::range<3> SrcSize, sycl::range<3> SrcAccessRange,
sycl::id<3> SrcOffset, unsigned int SrcElemSize, void *DstMem,
QueueImplPtr TgtQueue, unsigned int DimDst, sycl::range<3> DstSize,
sycl::range<3> DstAccessRange, sycl::id<3> DstOffset,
unsigned int DstElemSize, std::vector<ur_event_handle_t> DepEvents,
ur_event_handle_t &OutEvent, const detail::EventImplPtr &OutEventImpl) {
if (!SrcQueue) {
if (!TgtQueue)
copyH2H(SYCLMemObj, (char *)SrcMem, nullptr, DimSrc, SrcSize,
SrcAccessRange, SrcOffset, SrcElemSize, (char *)DstMem, nullptr,
DimDst, DstSize, DstAccessRange, DstOffset, DstElemSize,
std::move(DepEvents), OutEvent, OutEventImpl);
else
copyH2D(SYCLMemObj, (char *)SrcMem, nullptr, DimSrc, SrcSize,
SrcAccessRange, SrcOffset, SrcElemSize,
ur::cast<ur_mem_handle_t>(DstMem), std::move(TgtQueue), DimDst,
DstSize, DstAccessRange, DstOffset, DstElemSize,
std::move(DepEvents), OutEvent, OutEventImpl);
} else {
if (!TgtQueue)
copyD2H(SYCLMemObj, ur::cast<ur_mem_handle_t>(SrcMem),
std::move(SrcQueue), DimSrc, SrcSize, SrcAccessRange, SrcOffset,
SrcElemSize, (char *)DstMem, nullptr, DimDst, DstSize,
DstAccessRange, DstOffset, DstElemSize, std::move(DepEvents),
OutEvent, OutEventImpl);
else
copyD2D(SYCLMemObj, ur::cast<ur_mem_handle_t>(SrcMem),
std::move(SrcQueue), DimSrc, SrcSize, SrcAccessRange, SrcOffset,
SrcElemSize, ur::cast<ur_mem_handle_t>(DstMem),
std::move(TgtQueue), DimDst, DstSize, DstAccessRange, DstOffset,
DstElemSize, std::move(DepEvents), OutEvent, OutEventImpl);
}
}
void MemoryManager::fill(SYCLMemObjI *SYCLMemObj, void *Mem, QueueImplPtr Queue,
size_t PatternSize, const unsigned char *Pattern,
unsigned int Dim, sycl::range<3> MemRange,
sycl::range<3> AccRange, sycl::id<3> Offset,
unsigned int ElementSize,
std::vector<ur_event_handle_t> DepEvents,
ur_event_handle_t &OutEvent,
const detail::EventImplPtr &OutEventImpl) {
assert(SYCLMemObj && "The SYCLMemObj is nullptr");
assert(Queue && "Fill should be called only with a valid device queue");
const AdapterPtr &Adapter = Queue->getAdapter();
if (SYCLMemObj->getType() == detail::SYCLMemObjI::MemObjType::Buffer) {
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
// 2D and 3D buffers accessors can't have custom range or the data will
// likely be discontiguous.
bool RangesUsable = (Dim <= 1) || (MemRange == AccRange);
// For 2D and 3D buffers, the offset must be 0, or the data will be
// discontiguous.
bool OffsetUsable = (Dim <= 1) || (Offset == sycl::id<3>{0, 0, 0});
size_t RangeMultiplier = AccRange[0] * AccRange[1] * AccRange[2];
if (RangesUsable && OffsetUsable) {
Adapter->call<UrApiKind::urEnqueueMemBufferFill>(
Queue->getHandleRef(), ur::cast<ur_mem_handle_t>(Mem), Pattern,
PatternSize, Offset[0] * ElementSize, RangeMultiplier * ElementSize,
DepEvents.size(), DepEvents.data(), &OutEvent);
return;
}
// The sycl::handler uses a parallel_for kernel in the case of unusable
// Range or Offset, not CG:Fill. So we should not be here.
throw exception(make_error_code(errc::runtime),
"Not supported configuration of fill requested");
} else {
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
// We don't have any backend implementations that support enqueueing a fill
// on non-buffer mem objects like this. The old UR function was a stub with
// an abort.
throw exception(make_error_code(errc::runtime),
"Fill operation not supported for the given mem object");
}
}
void *MemoryManager::map(SYCLMemObjI *, void *Mem, QueueImplPtr Queue,
access::mode AccessMode, unsigned int, sycl::range<3>,
sycl::range<3> AccessRange, sycl::id<3> AccessOffset,
unsigned int ElementSize,
std::vector<ur_event_handle_t> DepEvents,
ur_event_handle_t &OutEvent) {
if (!Queue) {
throw exception(make_error_code(errc::runtime),
"Not supported configuration of map requested");
}
ur_map_flags_t Flags = 0;
switch (AccessMode) {
case access::mode::read:
Flags |= UR_MAP_FLAG_READ;
break;
case access::mode::write:
Flags |= UR_MAP_FLAG_WRITE;
break;
case access::mode::read_write:
case access::mode::atomic:
Flags = UR_MAP_FLAG_WRITE | UR_MAP_FLAG_READ;
break;
case access::mode::discard_write:
case access::mode::discard_read_write:
Flags |= UR_MAP_FLAG_WRITE_INVALIDATE_REGION;
break;
}
AccessOffset[0] *= ElementSize;
AccessRange[0] *= ElementSize;
// TODO: Handle offset
assert(AccessOffset[0] == 0 && "Handle offset");
void *MappedPtr = nullptr;
const size_t BytesToMap = AccessRange[0] * AccessRange[1] * AccessRange[2];
const AdapterPtr &Adapter = Queue->getAdapter();
memBufferMapHelper(Adapter, Queue->getHandleRef(),
ur::cast<ur_mem_handle_t>(Mem), false, Flags,
AccessOffset[0], BytesToMap, DepEvents.size(),
DepEvents.data(), &OutEvent, &MappedPtr);
return MappedPtr;
}
void MemoryManager::unmap(SYCLMemObjI *, void *Mem, QueueImplPtr Queue,
void *MappedPtr,
std::vector<ur_event_handle_t> DepEvents,
ur_event_handle_t &OutEvent) {
// Execution on host is not supported here.
if (!Queue) {
throw exception(make_error_code(errc::runtime),
"Not supported configuration of unmap requested");
}
// All DepEvents are to the same Context.
// Using the adapter of the Queue.
const AdapterPtr &Adapter = Queue->getAdapter();
memUnmapHelper(Adapter, Queue->getHandleRef(), ur::cast<ur_mem_handle_t>(Mem),
MappedPtr, DepEvents.size(), DepEvents.data(), &OutEvent);
}
void MemoryManager::copy_usm(const void *SrcMem, QueueImplPtr SrcQueue,
size_t Len, void *DstMem,
std::vector<ur_event_handle_t> DepEvents,
ur_event_handle_t *OutEvent,
const detail::EventImplPtr &OutEventImpl) {
assert(SrcQueue && "USM copy must be called with a valid device queue");
if (!Len) { // no-op, but ensure DepEvents will still be waited on
if (!DepEvents.empty()) {
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
SrcQueue->getAdapter()->call<UrApiKind::urEnqueueEventsWait>(
SrcQueue->getHandleRef(), DepEvents.size(), DepEvents.data(),
OutEvent);
}
return;
}
if (!SrcMem || !DstMem)
throw exception(make_error_code(errc::invalid),
"NULL pointer argument in memory copy operation.");
const AdapterPtr &Adapter = SrcQueue->getAdapter();
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
Adapter->call<UrApiKind::urEnqueueUSMMemcpy>(SrcQueue->getHandleRef(),
/* blocking */ false, DstMem,
SrcMem, Len, DepEvents.size(),
DepEvents.data(), OutEvent);
}
void MemoryManager::fill_usm(void *Mem, QueueImplPtr Queue, size_t Length,
const std::vector<unsigned char> &Pattern,
std::vector<ur_event_handle_t> DepEvents,
ur_event_handle_t *OutEvent,
const detail::EventImplPtr &OutEventImpl) {
assert(Queue && "USM fill must be called with a valid device queue");
if (!Length) { // no-op, but ensure DepEvents will still be waited on
if (!DepEvents.empty()) {
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
Queue->getAdapter()->call<UrApiKind::urEnqueueEventsWait>(
Queue->getHandleRef(), DepEvents.size(), DepEvents.data(), OutEvent);
}
return;
}
if (!Mem)
throw exception(make_error_code(errc::invalid),
"NULL pointer argument in memory fill operation.");
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
const AdapterPtr &Adapter = Queue->getAdapter();
Adapter->call<UrApiKind::urEnqueueUSMFill>(
Queue->getHandleRef(), Mem, Pattern.size(), Pattern.data(), Length,
DepEvents.size(), DepEvents.data(), OutEvent);
}
void MemoryManager::prefetch_usm(void *Mem, QueueImplPtr Queue, size_t Length,
std::vector<ur_event_handle_t> DepEvents,
ur_event_handle_t *OutEvent,
const detail::EventImplPtr &OutEventImpl) {
assert(Queue && "USM prefetch must be called with a valid device queue");
const AdapterPtr &Adapter = Queue->getAdapter();
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
Adapter->call<UrApiKind::urEnqueueUSMPrefetch>(Queue->getHandleRef(), Mem,
Length, 0, DepEvents.size(),
DepEvents.data(), OutEvent);
}
void MemoryManager::advise_usm(const void *Mem, QueueImplPtr Queue,
size_t Length, ur_usm_advice_flags_t Advice,
std::vector<ur_event_handle_t> /*DepEvents*/,
ur_event_handle_t *OutEvent,
const detail::EventImplPtr &OutEventImpl) {
assert(Queue && "USM advise must be called with a valid device queue");
const AdapterPtr &Adapter = Queue->getAdapter();
if (OutEventImpl != nullptr)
OutEventImpl->setHostEnqueueTime();
Adapter->call<UrApiKind::urEnqueueUSMAdvise>(Queue->getHandleRef(), Mem,
Length, Advice, OutEvent);
}
void MemoryManager::copy_2d_usm(const void *SrcMem, size_t SrcPitch,
QueueImplPtr Queue, void *DstMem,
size_t DstPitch, size_t Width, size_t Height,
std::vector<ur_event_handle_t> DepEvents,
ur_event_handle_t *OutEvent,
const detail::EventImplPtr &OutEventImpl) {
assert(Queue && "USM copy 2d must be called with a valid device queue");
if (Width == 0 || Height == 0) {