forked from llvm/llvm-project
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathomptarget.cpp
1552 lines (1381 loc) · 62.7 KB
/
omptarget.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
//===------ omptarget.cpp - Target independent OpenMP target RTL -- C++ -*-===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// Implementation of the interface to be used by Clang during the codegen of a
// target region.
//
//===----------------------------------------------------------------------===//
#include "omptarget.h"
#include "OffloadPolicy.h"
#include "OpenMP/OMPT/Callback.h"
#include "OpenMP/OMPT/Interface.h"
#include "PluginManager.h"
#include "Shared/Debug.h"
#include "Shared/EnvironmentVar.h"
#include "Shared/Utils.h"
#include "device.h"
#include "private.h"
#include "rtl.h"
#include "Shared/Profile.h"
#include "OpenMP/Mapping.h"
#include "OpenMP/omp.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/bit.h"
#include "llvm/Frontend/OpenMP/OMPConstants.h"
#include "llvm/Object/ObjectFile.h"
#include <cassert>
#include <cstdint>
#include <vector>
using llvm::SmallVector;
#ifdef OMPT_SUPPORT
using namespace llvm::omp::target::ompt;
#endif
int AsyncInfoTy::synchronize() {
int Result = OFFLOAD_SUCCESS;
if (!isQueueEmpty()) {
switch (SyncType) {
case SyncTy::BLOCKING:
// If we have a queue we need to synchronize it now.
Result = Device.synchronize(*this);
assert(AsyncInfo.Queue == nullptr &&
"The device plugin should have nulled the queue to indicate there "
"are no outstanding actions!");
break;
case SyncTy::NON_BLOCKING:
Result = Device.queryAsync(*this);
break;
}
}
// Run any pending post-processing function registered on this async object.
if (Result == OFFLOAD_SUCCESS && isQueueEmpty())
Result = runPostProcessing();
return Result;
}
void *&AsyncInfoTy::getVoidPtrLocation() {
BufferLocations.push_back(nullptr);
return BufferLocations.back();
}
bool AsyncInfoTy::isDone() const { return isQueueEmpty(); }
int32_t AsyncInfoTy::runPostProcessing() {
size_t Size = PostProcessingFunctions.size();
for (size_t I = 0; I < Size; ++I) {
const int Result = PostProcessingFunctions[I]();
if (Result != OFFLOAD_SUCCESS)
return Result;
}
// Clear the vector up until the last known function, since post-processing
// procedures might add new procedures themselves.
const auto *PrevBegin = PostProcessingFunctions.begin();
PostProcessingFunctions.erase(PrevBegin, PrevBegin + Size);
return OFFLOAD_SUCCESS;
}
bool AsyncInfoTy::isQueueEmpty() const { return AsyncInfo.Queue == nullptr; }
/* All begin addresses for partially mapped structs must be aligned, up to 16,
* in order to ensure proper alignment of members. E.g.
*
* struct S {
* int a; // 4-aligned
* int b; // 4-aligned
* int *p; // 8-aligned
* } s1;
* ...
* #pragma omp target map(tofrom: s1.b, s1.p[0:N])
* {
* s1.b = 5;
* for (int i...) s1.p[i] = ...;
* }
*
* Here we are mapping s1 starting from member b, so BaseAddress=&s1=&s1.a and
* BeginAddress=&s1.b. Let's assume that the struct begins at address 0x100,
* then &s1.a=0x100, &s1.b=0x104, &s1.p=0x108. Each member obeys the alignment
* requirements for its type. Now, when we allocate memory on the device, in
* CUDA's case cuMemAlloc() returns an address which is at least 256-aligned.
* This means that the chunk of the struct on the device will start at a
* 256-aligned address, let's say 0x200. Then the address of b will be 0x200 and
* address of p will be a misaligned 0x204 (on the host there was no need to add
* padding between b and p, so p comes exactly 4 bytes after b). If the device
* kernel tries to access s1.p, a misaligned address error occurs (as reported
* by the CUDA plugin). By padding the begin address down to a multiple of 8 and
* extending the size of the allocated chuck accordingly, the chuck on the
* device will start at 0x200 with the padding (4 bytes), then &s1.b=0x204 and
* &s1.p=0x208, as they should be to satisfy the alignment requirements.
*/
static const int64_t MaxAlignment = 16;
/// Return the alignment requirement of partially mapped structs, see
/// MaxAlignment above.
static uint64_t getPartialStructRequiredAlignment(void *HstPtrBase) {
int LowestOneBit = __builtin_ffsl(reinterpret_cast<uintptr_t>(HstPtrBase));
uint64_t BaseAlignment = 1 << (LowestOneBit - 1);
return MaxAlignment < BaseAlignment ? MaxAlignment : BaseAlignment;
}
void handleTargetOutcome(bool Success, ident_t *Loc) {
switch (OffloadPolicy::get(*PM).Kind) {
case OffloadPolicy::DISABLED:
if (Success) {
FATAL_MESSAGE0(1, "expected no offloading while offloading is disabled");
}
break;
case OffloadPolicy::MANDATORY:
if (!Success) {
if (getInfoLevel() & OMP_INFOTYPE_DUMP_TABLE) {
auto ExclusiveDevicesAccessor = PM->getExclusiveDevicesAccessor();
for (auto &Device : PM->devices(ExclusiveDevicesAccessor))
dumpTargetPointerMappings(Loc, Device);
} else
FAILURE_MESSAGE("Consult https://openmp.llvm.org/design/Runtimes.html "
"for debugging options.\n");
if (!PM->getNumActivePlugins()) {
FAILURE_MESSAGE(
"No images found compatible with the installed hardware. ");
llvm::SmallVector<llvm::StringRef> Archs;
for (auto &Image : PM->deviceImages()) {
const char *Start = reinterpret_cast<const char *>(
Image.getExecutableImage().ImageStart);
uint64_t Length =
utils::getPtrDiff(Start, Image.getExecutableImage().ImageEnd);
llvm::MemoryBufferRef Buffer(llvm::StringRef(Start, Length),
/*Identifier=*/"");
auto ObjectOrErr = llvm::object::ObjectFile::createObjectFile(Buffer);
if (auto Err = ObjectOrErr.takeError()) {
llvm::consumeError(std::move(Err));
continue;
}
if (auto CPU = (*ObjectOrErr)->tryGetCPUName())
Archs.push_back(*CPU);
}
fprintf(stderr, "Found %zu image(s): (%s)\n", Archs.size(),
llvm::join(Archs, ",").c_str());
}
SourceInfo Info(Loc);
if (Info.isAvailible())
fprintf(stderr, "%s:%d:%d: ", Info.getFilename(), Info.getLine(),
Info.getColumn());
else
FAILURE_MESSAGE("Source location information not present. Compile with "
"-g or -gline-tables-only.\n");
FATAL_MESSAGE0(
1, "failure of target construct while offloading is mandatory");
} else {
if (getInfoLevel() & OMP_INFOTYPE_DUMP_TABLE) {
auto ExclusiveDevicesAccessor = PM->getExclusiveDevicesAccessor();
for (auto &Device : PM->devices(ExclusiveDevicesAccessor))
dumpTargetPointerMappings(Loc, Device);
}
}
break;
}
}
static int32_t getParentIndex(int64_t Type) {
return ((Type & OMP_TGT_MAPTYPE_MEMBER_OF) >> 48) - 1;
}
void *targetAllocExplicit(size_t Size, int DeviceNum, int Kind,
const char *Name) {
DP("Call to %s for device %d requesting %zu bytes\n", Name, DeviceNum, Size);
if (Size <= 0) {
DP("Call to %s with non-positive length\n", Name);
return NULL;
}
void *Rc = NULL;
if (DeviceNum == omp_get_initial_device()) {
Rc = malloc(Size);
DP("%s returns host ptr " DPxMOD "\n", Name, DPxPTR(Rc));
return Rc;
}
auto DeviceOrErr = PM->getDevice(DeviceNum);
if (!DeviceOrErr)
FATAL_MESSAGE(DeviceNum, "%s", toString(DeviceOrErr.takeError()).c_str());
Rc = DeviceOrErr->allocData(Size, nullptr, Kind);
DP("%s returns device ptr " DPxMOD "\n", Name, DPxPTR(Rc));
return Rc;
}
void targetFreeExplicit(void *DevicePtr, int DeviceNum, int Kind,
const char *Name) {
DP("Call to %s for device %d and address " DPxMOD "\n", Name, DeviceNum,
DPxPTR(DevicePtr));
if (!DevicePtr) {
DP("Call to %s with NULL ptr\n", Name);
return;
}
if (DeviceNum == omp_get_initial_device()) {
free(DevicePtr);
DP("%s deallocated host ptr\n", Name);
return;
}
auto DeviceOrErr = PM->getDevice(DeviceNum);
if (!DeviceOrErr)
FATAL_MESSAGE(DeviceNum, "%s", toString(DeviceOrErr.takeError()).c_str());
if (DeviceOrErr->deleteData(DevicePtr, Kind) == OFFLOAD_FAIL)
FATAL_MESSAGE(DeviceNum, "%s",
"Failed to deallocate device ptr. Set "
"OFFLOAD_TRACK_ALLOCATION_TRACES=1 to track allocations.");
DP("omp_target_free deallocated device ptr\n");
}
void *targetLockExplicit(void *HostPtr, size_t Size, int DeviceNum,
const char *Name) {
DP("Call to %s for device %d locking %zu bytes\n", Name, DeviceNum, Size);
if (Size <= 0) {
DP("Call to %s with non-positive length\n", Name);
return NULL;
}
void *RC = NULL;
auto DeviceOrErr = PM->getDevice(DeviceNum);
if (!DeviceOrErr)
FATAL_MESSAGE(DeviceNum, "%s", toString(DeviceOrErr.takeError()).c_str());
int32_t Err = 0;
Err = DeviceOrErr->RTL->data_lock(DeviceNum, HostPtr, Size, &RC);
if (Err) {
DP("Could not lock ptr %p\n", HostPtr);
return nullptr;
}
DP("%s returns device ptr " DPxMOD "\n", Name, DPxPTR(RC));
return RC;
}
void targetUnlockExplicit(void *HostPtr, int DeviceNum, const char *Name) {
DP("Call to %s for device %d unlocking\n", Name, DeviceNum);
auto DeviceOrErr = PM->getDevice(DeviceNum);
if (!DeviceOrErr)
FATAL_MESSAGE(DeviceNum, "%s", toString(DeviceOrErr.takeError()).c_str());
DeviceOrErr->RTL->data_unlock(DeviceNum, HostPtr);
DP("%s returns\n", Name);
}
/// Call the user-defined mapper function followed by the appropriate
// targetData* function (targetData{Begin,End,Update}).
int targetDataMapper(ident_t *Loc, DeviceTy &Device, void *ArgBase, void *Arg,
int64_t ArgSize, int64_t ArgType, map_var_info_t ArgNames,
void *ArgMapper, AsyncInfoTy &AsyncInfo,
TargetDataFuncPtrTy TargetDataFunction) {
DP("Calling the mapper function " DPxMOD "\n", DPxPTR(ArgMapper));
// The mapper function fills up Components.
MapperComponentsTy MapperComponents;
MapperFuncPtrTy MapperFuncPtr = (MapperFuncPtrTy)(ArgMapper);
(*MapperFuncPtr)((void *)&MapperComponents, ArgBase, Arg, ArgSize, ArgType,
ArgNames);
// Construct new arrays for args_base, args, arg_sizes and arg_types
// using the information in MapperComponents and call the corresponding
// targetData* function using these new arrays.
SmallVector<void *> MapperArgsBase(MapperComponents.Components.size());
SmallVector<void *> MapperArgs(MapperComponents.Components.size());
SmallVector<int64_t> MapperArgSizes(MapperComponents.Components.size());
SmallVector<int64_t> MapperArgTypes(MapperComponents.Components.size());
SmallVector<void *> MapperArgNames(MapperComponents.Components.size());
for (unsigned I = 0, E = MapperComponents.Components.size(); I < E; ++I) {
auto &C = MapperComponents.Components[I];
MapperArgsBase[I] = C.Base;
MapperArgs[I] = C.Begin;
MapperArgSizes[I] = C.Size;
MapperArgTypes[I] = C.Type;
MapperArgNames[I] = C.Name;
}
int Rc = TargetDataFunction(Loc, Device, MapperComponents.Components.size(),
MapperArgsBase.data(), MapperArgs.data(),
MapperArgSizes.data(), MapperArgTypes.data(),
MapperArgNames.data(), /*arg_mappers*/ nullptr,
AsyncInfo, /*FromMapper=*/true);
return Rc;
}
/// Internal function to do the mapping and transfer the data to the device
int targetDataBegin(ident_t *Loc, DeviceTy &Device, int32_t ArgNum,
void **ArgsBase, void **Args, int64_t *ArgSizes,
int64_t *ArgTypes, map_var_info_t *ArgNames,
void **ArgMappers, AsyncInfoTy &AsyncInfo,
bool FromMapper) {
// process each input.
for (int32_t I = 0; I < ArgNum; ++I) {
// Ignore private variables and arrays - there is no mapping for them.
if ((ArgTypes[I] & OMP_TGT_MAPTYPE_LITERAL) ||
(ArgTypes[I] & OMP_TGT_MAPTYPE_PRIVATE))
continue;
TIMESCOPE_WITH_DETAILS_AND_IDENT(
"HostToDev", "Size=" + std::to_string(ArgSizes[I]) + "B", Loc);
if (ArgMappers && ArgMappers[I]) {
// Instead of executing the regular path of targetDataBegin, call the
// targetDataMapper variant which will call targetDataBegin again
// with new arguments.
DP("Calling targetDataMapper for the %dth argument\n", I);
map_var_info_t ArgName = (!ArgNames) ? nullptr : ArgNames[I];
int Rc = targetDataMapper(Loc, Device, ArgsBase[I], Args[I], ArgSizes[I],
ArgTypes[I], ArgName, ArgMappers[I], AsyncInfo,
targetDataBegin);
if (Rc != OFFLOAD_SUCCESS) {
REPORT("Call to targetDataBegin via targetDataMapper for custom mapper"
" failed.\n");
return OFFLOAD_FAIL;
}
// Skip the rest of this function, continue to the next argument.
continue;
}
void *HstPtrBegin = Args[I];
void *HstPtrBase = ArgsBase[I];
int64_t DataSize = ArgSizes[I];
map_var_info_t HstPtrName = (!ArgNames) ? nullptr : ArgNames[I];
// Adjust for proper alignment if this is a combined entry (for structs).
// Look at the next argument - if that is MEMBER_OF this one, then this one
// is a combined entry.
int64_t TgtPadding = 0;
const int NextI = I + 1;
if (getParentIndex(ArgTypes[I]) < 0 && NextI < ArgNum &&
getParentIndex(ArgTypes[NextI]) == I) {
int64_t Alignment = getPartialStructRequiredAlignment(HstPtrBase);
TgtPadding = (int64_t)HstPtrBegin % Alignment;
if (TgtPadding) {
DP("Using a padding of %" PRId64 " bytes for begin address " DPxMOD
"\n",
TgtPadding, DPxPTR(HstPtrBegin));
}
}
// Address of pointer on the host and device, respectively.
void *PointerHstPtrBegin, *PointerTgtPtrBegin;
TargetPointerResultTy PointerTpr;
bool IsHostPtr = false;
bool IsImplicit = ArgTypes[I] & OMP_TGT_MAPTYPE_IMPLICIT;
// Force the creation of a device side copy of the data when:
// a close map modifier was associated with a map that contained a to.
bool HasCloseModifier = ArgTypes[I] & OMP_TGT_MAPTYPE_CLOSE;
bool HasPresentModifier = ArgTypes[I] & OMP_TGT_MAPTYPE_PRESENT;
bool HasHoldModifier = ArgTypes[I] & OMP_TGT_MAPTYPE_OMPX_HOLD;
// UpdateRef is based on MEMBER_OF instead of TARGET_PARAM because if we
// have reached this point via __tgt_target_data_begin and not __tgt_target
// then no argument is marked as TARGET_PARAM ("omp target data map" is not
// associated with a target region, so there are no target parameters). This
// may be considered a hack, we could revise the scheme in the future.
bool UpdateRef =
!(ArgTypes[I] & OMP_TGT_MAPTYPE_MEMBER_OF) && !(FromMapper && I == 0);
MappingInfoTy::HDTTMapAccessorTy HDTTMap =
Device.getMappingInfo().HostDataToTargetMap.getExclusiveAccessor();
if (ArgTypes[I] & OMP_TGT_MAPTYPE_PTR_AND_OBJ) {
DP("Has a pointer entry: \n");
// Base is address of pointer.
//
// Usually, the pointer is already allocated by this time. For example:
//
// #pragma omp target map(s.p[0:N])
//
// The map entry for s comes first, and the PTR_AND_OBJ entry comes
// afterward, so the pointer is already allocated by the time the
// PTR_AND_OBJ entry is handled below, and PointerTgtPtrBegin is thus
// non-null. However, "declare target link" can produce a PTR_AND_OBJ
// entry for a global that might not already be allocated by the time the
// PTR_AND_OBJ entry is handled below, and so the allocation might fail
// when HasPresentModifier.
PointerTpr = Device.getMappingInfo().getTargetPointer(
HDTTMap, HstPtrBase, HstPtrBase, /*TgtPadding=*/0, sizeof(void *),
/*HstPtrName=*/nullptr,
/*HasFlagTo=*/false, /*HasFlagAlways=*/false, IsImplicit, UpdateRef,
HasCloseModifier, HasPresentModifier, HasHoldModifier, AsyncInfo,
/*OwnedTPR=*/nullptr, /*ReleaseHDTTMap=*/false);
PointerTgtPtrBegin = PointerTpr.TargetPointer;
IsHostPtr = PointerTpr.Flags.IsHostPointer;
if (!PointerTgtPtrBegin) {
REPORT("Call to getTargetPointer returned null pointer (%s).\n",
HasPresentModifier ? "'present' map type modifier"
: "device failure or illegal mapping");
return OFFLOAD_FAIL;
}
DP("There are %zu bytes allocated at target address " DPxMOD " - is%s new"
"\n",
sizeof(void *), DPxPTR(PointerTgtPtrBegin),
(PointerTpr.Flags.IsNewEntry ? "" : " not"));
PointerHstPtrBegin = HstPtrBase;
// modify current entry.
HstPtrBase = *(void **)HstPtrBase;
// No need to update pointee ref count for the first element of the
// subelement that comes from mapper.
UpdateRef =
(!FromMapper || I != 0); // subsequently update ref count of pointee
}
const bool HasFlagTo = ArgTypes[I] & OMP_TGT_MAPTYPE_TO;
const bool HasFlagAlways = ArgTypes[I] & OMP_TGT_MAPTYPE_ALWAYS;
// Note that HDTTMap will be released in getTargetPointer.
auto TPR = Device.getMappingInfo().getTargetPointer(
HDTTMap, HstPtrBegin, HstPtrBase, TgtPadding, DataSize, HstPtrName,
HasFlagTo, HasFlagAlways, IsImplicit, UpdateRef, HasCloseModifier,
HasPresentModifier, HasHoldModifier, AsyncInfo, PointerTpr.getEntry());
void *TgtPtrBegin = TPR.TargetPointer;
IsHostPtr = TPR.Flags.IsHostPointer;
// If data_size==0, then the argument could be a zero-length pointer to
// NULL, so getOrAlloc() returning NULL is not an error.
if (!TgtPtrBegin && (DataSize || HasPresentModifier)) {
REPORT("Call to getTargetPointer returned null pointer (%s).\n",
HasPresentModifier ? "'present' map type modifier"
: "device failure or illegal mapping");
return OFFLOAD_FAIL;
}
DP("There are %" PRId64 " bytes allocated at target address " DPxMOD
" - is%s new\n",
DataSize, DPxPTR(TgtPtrBegin), (TPR.Flags.IsNewEntry ? "" : " not"));
if (ArgTypes[I] & OMP_TGT_MAPTYPE_RETURN_PARAM) {
uintptr_t Delta = (uintptr_t)HstPtrBegin - (uintptr_t)HstPtrBase;
void *TgtPtrBase = (void *)((uintptr_t)TgtPtrBegin - Delta);
DP("Returning device pointer " DPxMOD "\n", DPxPTR(TgtPtrBase));
ArgsBase[I] = TgtPtrBase;
}
if (ArgTypes[I] & OMP_TGT_MAPTYPE_PTR_AND_OBJ && !IsHostPtr) {
uint64_t Delta = (uint64_t)HstPtrBegin - (uint64_t)HstPtrBase;
void *ExpectedTgtPtrBase = (void *)((uint64_t)TgtPtrBegin - Delta);
if (PointerTpr.getEntry()->addShadowPointer(ShadowPtrInfoTy{
(void **)PointerHstPtrBegin, HstPtrBase,
(void **)PointerTgtPtrBegin, ExpectedTgtPtrBase})) {
DP("Update pointer (" DPxMOD ") -> [" DPxMOD "]\n",
DPxPTR(PointerTgtPtrBegin), DPxPTR(TgtPtrBegin));
void *&TgtPtrBase = AsyncInfo.getVoidPtrLocation();
TgtPtrBase = ExpectedTgtPtrBase;
int Ret =
Device.submitData(PointerTgtPtrBegin, &TgtPtrBase, sizeof(void *),
AsyncInfo, PointerTpr.getEntry());
if (Ret != OFFLOAD_SUCCESS) {
REPORT("Copying data to device failed.\n");
return OFFLOAD_FAIL;
}
if (PointerTpr.getEntry()->addEventIfNecessary(Device, AsyncInfo) !=
OFFLOAD_SUCCESS)
return OFFLOAD_FAIL;
}
}
// Check if variable can be used on the device:
bool IsStructMember = ArgTypes[I] & OMP_TGT_MAPTYPE_MEMBER_OF;
if (getInfoLevel() & OMP_INFOTYPE_EMPTY_MAPPING && ArgTypes[I] != 0 &&
!IsStructMember && !IsImplicit && !TPR.isPresent() &&
!TPR.isContained() && !TPR.isHostPointer())
INFO(OMP_INFOTYPE_EMPTY_MAPPING, Device.DeviceID,
"variable %s does not have a valid device counterpart\n",
(HstPtrName) ? getNameFromMapping(HstPtrName).c_str() : "unknown");
}
return OFFLOAD_SUCCESS;
}
namespace {
/// This structure contains information to deallocate a target pointer, aka.
/// used to fix up the shadow map and potentially delete the entry from the
/// mapping table via \p DeviceTy::deallocTgtPtr.
struct PostProcessingInfo {
/// Host pointer used to look up into the map table
void *HstPtrBegin;
/// Size of the data
int64_t DataSize;
/// The mapping type (bitfield).
int64_t ArgType;
/// The target pointer information.
TargetPointerResultTy TPR;
PostProcessingInfo(void *HstPtr, int64_t Size, int64_t ArgType,
TargetPointerResultTy &&TPR)
: HstPtrBegin(HstPtr), DataSize(Size), ArgType(ArgType),
TPR(std::move(TPR)) {}
};
} // namespace
/// Applies the necessary post-processing procedures to entries listed in \p
/// EntriesInfo after the execution of all device side operations from a target
/// data end. This includes the update of pointers at the host and removal of
/// device buffer when needed. It returns OFFLOAD_FAIL or OFFLOAD_SUCCESS
/// according to the successfulness of the operations.
[[nodiscard]] static int
postProcessingTargetDataEnd(DeviceTy *Device,
SmallVector<PostProcessingInfo> &EntriesInfo) {
int Ret = OFFLOAD_SUCCESS;
for (auto &[HstPtrBegin, DataSize, ArgType, TPR] : EntriesInfo) {
bool DelEntry = !TPR.isHostPointer();
// If the last element from the mapper (for end transfer args comes in
// reverse order), do not remove the partial entry, the parent struct still
// exists.
if ((ArgType & OMP_TGT_MAPTYPE_MEMBER_OF) &&
!(ArgType & OMP_TGT_MAPTYPE_PTR_AND_OBJ)) {
DelEntry = false; // protect parent struct from being deallocated
}
// If we marked the entry to be deleted we need to verify no other
// thread reused it by now. If deletion is still supposed to happen by
// this thread LR will be set and exclusive access to the HDTT map
// will avoid another thread reusing the entry now. Note that we do
// not request (exclusive) access to the HDTT map if DelEntry is
// not set.
MappingInfoTy::HDTTMapAccessorTy HDTTMap =
Device->getMappingInfo().HostDataToTargetMap.getExclusiveAccessor();
// We cannot use a lock guard because we may end up delete the mutex.
// We also explicitly unlocked the entry after it was put in the EntriesInfo
// so it can be reused.
TPR.getEntry()->lock();
auto *Entry = TPR.getEntry();
const bool IsNotLastUser = Entry->decDataEndThreadCount() != 0;
if (DelEntry && (Entry->getTotalRefCount() != 0 || IsNotLastUser)) {
// The thread is not in charge of deletion anymore. Give up access
// to the HDTT map and unset the deletion flag.
HDTTMap.destroy();
DelEntry = false;
}
// If we copied back to the host a struct/array containing pointers,
// we need to restore the original host pointer values from their
// shadow copies. If the struct is going to be deallocated, remove any
// remaining shadow pointer entries for this struct.
const bool HasFrom = ArgType & OMP_TGT_MAPTYPE_FROM;
if (HasFrom) {
Entry->foreachShadowPointerInfo([&](const ShadowPtrInfoTy &ShadowPtr) {
*ShadowPtr.HstPtrAddr = ShadowPtr.HstPtrVal;
DP("Restoring original host pointer value " DPxMOD " for host "
"pointer " DPxMOD "\n",
DPxPTR(ShadowPtr.HstPtrVal), DPxPTR(ShadowPtr.HstPtrAddr));
return OFFLOAD_SUCCESS;
});
}
// Give up the lock as we either don't need it anymore (e.g., done with
// TPR), or erase TPR.
TPR.setEntry(nullptr);
if (!DelEntry)
continue;
Ret = Device->getMappingInfo().eraseMapEntry(HDTTMap, Entry, DataSize);
// Entry is already remove from the map, we can unlock it now.
HDTTMap.destroy();
Ret |= Device->getMappingInfo().deallocTgtPtrAndEntry(Entry, DataSize);
if (Ret != OFFLOAD_SUCCESS) {
REPORT("Deallocating data from device failed.\n");
break;
}
}
delete &EntriesInfo;
return Ret;
}
/// Internal function to undo the mapping and retrieve the data from the device.
int targetDataEnd(ident_t *Loc, DeviceTy &Device, int32_t ArgNum,
void **ArgBases, void **Args, int64_t *ArgSizes,
int64_t *ArgTypes, map_var_info_t *ArgNames,
void **ArgMappers, AsyncInfoTy &AsyncInfo, bool FromMapper) {
int Ret = OFFLOAD_SUCCESS;
auto *PostProcessingPtrs = new SmallVector<PostProcessingInfo>();
// process each input.
for (int32_t I = ArgNum - 1; I >= 0; --I) {
// Ignore private variables and arrays - there is no mapping for them.
// Also, ignore the use_device_ptr directive, it has no effect here.
if ((ArgTypes[I] & OMP_TGT_MAPTYPE_LITERAL) ||
(ArgTypes[I] & OMP_TGT_MAPTYPE_PRIVATE))
continue;
if (ArgMappers && ArgMappers[I]) {
// Instead of executing the regular path of targetDataEnd, call the
// targetDataMapper variant which will call targetDataEnd again
// with new arguments.
DP("Calling targetDataMapper for the %dth argument\n", I);
map_var_info_t ArgName = (!ArgNames) ? nullptr : ArgNames[I];
Ret = targetDataMapper(Loc, Device, ArgBases[I], Args[I], ArgSizes[I],
ArgTypes[I], ArgName, ArgMappers[I], AsyncInfo,
targetDataEnd);
if (Ret != OFFLOAD_SUCCESS) {
REPORT("Call to targetDataEnd via targetDataMapper for custom mapper"
" failed.\n");
return OFFLOAD_FAIL;
}
// Skip the rest of this function, continue to the next argument.
continue;
}
void *HstPtrBegin = Args[I];
int64_t DataSize = ArgSizes[I];
bool IsImplicit = ArgTypes[I] & OMP_TGT_MAPTYPE_IMPLICIT;
bool UpdateRef = (!(ArgTypes[I] & OMP_TGT_MAPTYPE_MEMBER_OF) ||
(ArgTypes[I] & OMP_TGT_MAPTYPE_PTR_AND_OBJ)) &&
!(FromMapper && I == 0);
bool ForceDelete = ArgTypes[I] & OMP_TGT_MAPTYPE_DELETE;
bool HasPresentModifier = ArgTypes[I] & OMP_TGT_MAPTYPE_PRESENT;
bool HasHoldModifier = ArgTypes[I] & OMP_TGT_MAPTYPE_OMPX_HOLD;
// If PTR_AND_OBJ, HstPtrBegin is address of pointee
TargetPointerResultTy TPR = Device.getMappingInfo().getTgtPtrBegin(
HstPtrBegin, DataSize, UpdateRef, HasHoldModifier, !IsImplicit,
ForceDelete, /*FromDataEnd=*/true);
void *TgtPtrBegin = TPR.TargetPointer;
if (!TPR.isPresent() && !TPR.isHostPointer() &&
(DataSize || HasPresentModifier)) {
DP("Mapping does not exist (%s)\n",
(HasPresentModifier ? "'present' map type modifier" : "ignored"));
if (HasPresentModifier) {
// OpenMP 5.1, sec. 2.21.7.1 "map Clause", p. 350 L10-13:
// "If a map clause appears on a target, target data, target enter data
// or target exit data construct with a present map-type-modifier then
// on entry to the region if the corresponding list item does not appear
// in the device data environment then an error occurs and the program
// terminates."
//
// This should be an error upon entering an "omp target exit data". It
// should not be an error upon exiting an "omp target data" or "omp
// target". For "omp target data", Clang thus doesn't include present
// modifiers for end calls. For "omp target", we have not found a valid
// OpenMP program for which the error matters: it appears that, if a
// program can guarantee that data is present at the beginning of an
// "omp target" region so that there's no error there, that data is also
// guaranteed to be present at the end.
MESSAGE("device mapping required by 'present' map type modifier does "
"not exist for host address " DPxMOD " (%" PRId64 " bytes)",
DPxPTR(HstPtrBegin), DataSize);
return OFFLOAD_FAIL;
}
} else {
DP("There are %" PRId64 " bytes allocated at target address " DPxMOD
" - is%s last\n",
DataSize, DPxPTR(TgtPtrBegin), (TPR.Flags.IsLast ? "" : " not"));
}
// OpenMP 5.1, sec. 2.21.7.1 "map Clause", p. 351 L14-16:
// "If the map clause appears on a target, target data, or target exit data
// construct and a corresponding list item of the original list item is not
// present in the device data environment on exit from the region then the
// list item is ignored."
if (!TPR.isPresent())
continue;
// Move data back to the host
const bool HasAlways = ArgTypes[I] & OMP_TGT_MAPTYPE_ALWAYS;
const bool HasFrom = ArgTypes[I] & OMP_TGT_MAPTYPE_FROM;
if (HasFrom && (HasAlways || TPR.Flags.IsLast) &&
!TPR.Flags.IsHostPointer && DataSize != 0) {
DP("Moving %" PRId64 " bytes (tgt:" DPxMOD ") -> (hst:" DPxMOD ")\n",
DataSize, DPxPTR(TgtPtrBegin), DPxPTR(HstPtrBegin));
TIMESCOPE_WITH_DETAILS_AND_IDENT(
"DevToHost", "Size=" + std::to_string(DataSize) + "B", Loc);
// Wait for any previous transfer if an event is present.
if (void *Event = TPR.getEntry()->getEvent()) {
if (Device.waitEvent(Event, AsyncInfo) != OFFLOAD_SUCCESS) {
REPORT("Failed to wait for event " DPxMOD ".\n", DPxPTR(Event));
return OFFLOAD_FAIL;
}
}
Ret = Device.retrieveData(HstPtrBegin, TgtPtrBegin, DataSize, AsyncInfo,
TPR.getEntry());
if (Ret != OFFLOAD_SUCCESS) {
REPORT("Copying data from device failed.\n");
return OFFLOAD_FAIL;
}
// As we are expecting to delete the entry the d2h copy might race
// with another one that also tries to delete the entry. This happens
// as the entry can be reused and the reuse might happen after the
// copy-back was issued but before it completed. Since the reuse might
// also copy-back a value we would race.
if (TPR.Flags.IsLast) {
if (TPR.getEntry()->addEventIfNecessary(Device, AsyncInfo) !=
OFFLOAD_SUCCESS)
return OFFLOAD_FAIL;
}
}
// Add pointer to the buffer for post-synchronize processing.
PostProcessingPtrs->emplace_back(HstPtrBegin, DataSize, ArgTypes[I],
std::move(TPR));
PostProcessingPtrs->back().TPR.getEntry()->unlock();
}
// Add post-processing functions
// TODO: We might want to remove `mutable` in the future by not changing the
// captured variables somehow.
AsyncInfo.addPostProcessingFunction([=, Device = &Device]() mutable -> int {
return postProcessingTargetDataEnd(Device, *PostProcessingPtrs);
});
return Ret;
}
static int targetDataContiguous(ident_t *Loc, DeviceTy &Device, void *ArgsBase,
void *HstPtrBegin, int64_t ArgSize,
int64_t ArgType, AsyncInfoTy &AsyncInfo) {
TargetPointerResultTy TPR = Device.getMappingInfo().getTgtPtrBegin(
HstPtrBegin, ArgSize, /*UpdateRefCount=*/false,
/*UseHoldRefCount=*/false, /*MustContain=*/true);
void *TgtPtrBegin = TPR.TargetPointer;
if (!TPR.isPresent()) {
DP("hst data:" DPxMOD " not found, becomes a noop\n", DPxPTR(HstPtrBegin));
if (ArgType & OMP_TGT_MAPTYPE_PRESENT) {
MESSAGE("device mapping required by 'present' motion modifier does not "
"exist for host address " DPxMOD " (%" PRId64 " bytes)",
DPxPTR(HstPtrBegin), ArgSize);
return OFFLOAD_FAIL;
}
return OFFLOAD_SUCCESS;
}
if (TPR.Flags.IsHostPointer) {
DP("hst data:" DPxMOD " unified and shared, becomes a noop\n",
DPxPTR(HstPtrBegin));
return OFFLOAD_SUCCESS;
}
if (ArgType & OMP_TGT_MAPTYPE_TO) {
DP("Moving %" PRId64 " bytes (hst:" DPxMOD ") -> (tgt:" DPxMOD ")\n",
ArgSize, DPxPTR(HstPtrBegin), DPxPTR(TgtPtrBegin));
int Ret = Device.submitData(TgtPtrBegin, HstPtrBegin, ArgSize, AsyncInfo,
TPR.getEntry());
if (Ret != OFFLOAD_SUCCESS) {
REPORT("Copying data to device failed.\n");
return OFFLOAD_FAIL;
}
if (TPR.getEntry()) {
int Ret = TPR.getEntry()->foreachShadowPointerInfo(
[&](ShadowPtrInfoTy &ShadowPtr) {
DP("Restoring original target pointer value " DPxMOD " for target "
"pointer " DPxMOD "\n",
DPxPTR(ShadowPtr.TgtPtrVal), DPxPTR(ShadowPtr.TgtPtrAddr));
Ret = Device.submitData(ShadowPtr.TgtPtrAddr,
(void *)&ShadowPtr.TgtPtrVal,
sizeof(void *), AsyncInfo);
if (Ret != OFFLOAD_SUCCESS) {
REPORT("Copying data to device failed.\n");
return OFFLOAD_FAIL;
}
return OFFLOAD_SUCCESS;
});
if (Ret != OFFLOAD_SUCCESS) {
DP("Updating shadow map failed\n");
return Ret;
}
}
}
if (ArgType & OMP_TGT_MAPTYPE_FROM) {
DP("Moving %" PRId64 " bytes (tgt:" DPxMOD ") -> (hst:" DPxMOD ")\n",
ArgSize, DPxPTR(TgtPtrBegin), DPxPTR(HstPtrBegin));
int Ret = Device.retrieveData(HstPtrBegin, TgtPtrBegin, ArgSize, AsyncInfo,
TPR.getEntry());
if (Ret != OFFLOAD_SUCCESS) {
REPORT("Copying data from device failed.\n");
return OFFLOAD_FAIL;
}
// Wait for device-to-host memcopies for whole struct to complete,
// before restoring the correct host pointer.
if (auto *Entry = TPR.getEntry()) {
AsyncInfo.addPostProcessingFunction([=]() -> int {
int Ret = Entry->foreachShadowPointerInfo(
[&](const ShadowPtrInfoTy &ShadowPtr) {
*ShadowPtr.HstPtrAddr = ShadowPtr.HstPtrVal;
DP("Restoring original host pointer value " DPxMOD
" for host pointer " DPxMOD "\n",
DPxPTR(ShadowPtr.HstPtrVal), DPxPTR(ShadowPtr.HstPtrAddr));
return OFFLOAD_SUCCESS;
});
Entry->unlock();
if (Ret != OFFLOAD_SUCCESS) {
DP("Updating shadow map failed\n");
return Ret;
}
return OFFLOAD_SUCCESS;
});
}
}
return OFFLOAD_SUCCESS;
}
static int targetDataNonContiguous(ident_t *Loc, DeviceTy &Device,
void *ArgsBase,
__tgt_target_non_contig *NonContig,
uint64_t Size, int64_t ArgType,
int CurrentDim, int DimSize, uint64_t Offset,
AsyncInfoTy &AsyncInfo) {
int Ret = OFFLOAD_SUCCESS;
if (CurrentDim < DimSize) {
for (unsigned int I = 0; I < NonContig[CurrentDim].Count; ++I) {
uint64_t CurOffset =
(NonContig[CurrentDim].Offset + I) * NonContig[CurrentDim].Stride;
// we only need to transfer the first element for the last dimension
// since we've already got a contiguous piece.
if (CurrentDim != DimSize - 1 || I == 0) {
Ret = targetDataNonContiguous(Loc, Device, ArgsBase, NonContig, Size,
ArgType, CurrentDim + 1, DimSize,
Offset + CurOffset, AsyncInfo);
// Stop the whole process if any contiguous piece returns anything
// other than OFFLOAD_SUCCESS.
if (Ret != OFFLOAD_SUCCESS)
return Ret;
}
}
} else {
char *Ptr = (char *)ArgsBase + Offset;
DP("Transfer of non-contiguous : host ptr " DPxMOD " offset %" PRIu64
" len %" PRIu64 "\n",
DPxPTR(Ptr), Offset, Size);
Ret = targetDataContiguous(Loc, Device, ArgsBase, Ptr, Size, ArgType,
AsyncInfo);
}
return Ret;
}
static int getNonContigMergedDimension(__tgt_target_non_contig *NonContig,
int32_t DimSize) {
int RemovedDim = 0;
for (int I = DimSize - 1; I > 0; --I) {
if (NonContig[I].Count * NonContig[I].Stride == NonContig[I - 1].Stride)
RemovedDim++;
}
return RemovedDim;
}
/// Internal function to pass data to/from the target.
int targetDataUpdate(ident_t *Loc, DeviceTy &Device, int32_t ArgNum,
void **ArgsBase, void **Args, int64_t *ArgSizes,
int64_t *ArgTypes, map_var_info_t *ArgNames,
void **ArgMappers, AsyncInfoTy &AsyncInfo, bool) {
// process each input.
for (int32_t I = 0; I < ArgNum; ++I) {
if ((ArgTypes[I] & OMP_TGT_MAPTYPE_LITERAL) ||
(ArgTypes[I] & OMP_TGT_MAPTYPE_PRIVATE))
continue;
if (ArgMappers && ArgMappers[I]) {
// Instead of executing the regular path of targetDataUpdate, call the
// targetDataMapper variant which will call targetDataUpdate again
// with new arguments.
DP("Calling targetDataMapper for the %dth argument\n", I);
map_var_info_t ArgName = (!ArgNames) ? nullptr : ArgNames[I];
int Ret = targetDataMapper(Loc, Device, ArgsBase[I], Args[I], ArgSizes[I],
ArgTypes[I], ArgName, ArgMappers[I], AsyncInfo,
targetDataUpdate);
if (Ret != OFFLOAD_SUCCESS) {
REPORT("Call to targetDataUpdate via targetDataMapper for custom mapper"
" failed.\n");
return OFFLOAD_FAIL;
}
// Skip the rest of this function, continue to the next argument.
continue;
}
int Ret = OFFLOAD_SUCCESS;
if (ArgTypes[I] & OMP_TGT_MAPTYPE_NON_CONTIG) {
__tgt_target_non_contig *NonContig = (__tgt_target_non_contig *)Args[I];
int32_t DimSize = ArgSizes[I];
uint64_t Size =
NonContig[DimSize - 1].Count * NonContig[DimSize - 1].Stride;
int32_t MergedDim = getNonContigMergedDimension(NonContig, DimSize);
Ret = targetDataNonContiguous(
Loc, Device, ArgsBase[I], NonContig, Size, ArgTypes[I],
/*current_dim=*/0, DimSize - MergedDim, /*offset=*/0, AsyncInfo);
} else {
Ret = targetDataContiguous(Loc, Device, ArgsBase[I], Args[I], ArgSizes[I],
ArgTypes[I], AsyncInfo);
}
if (Ret == OFFLOAD_FAIL)
return OFFLOAD_FAIL;
}
return OFFLOAD_SUCCESS;
}
static const unsigned LambdaMapping = OMP_TGT_MAPTYPE_PTR_AND_OBJ |
OMP_TGT_MAPTYPE_LITERAL |
OMP_TGT_MAPTYPE_IMPLICIT;
static bool isLambdaMapping(int64_t Mapping) {
return (Mapping & LambdaMapping) == LambdaMapping;
}
namespace {
/// Find the table information in the map or look it up in the translation
/// tables.
TableMap *getTableMap(void *HostPtr) {
std::lock_guard<std::mutex> TblMapLock(PM->TblMapMtx);
HostPtrToTableMapTy::iterator TableMapIt =
PM->HostPtrToTableMap.find(HostPtr);
if (TableMapIt != PM->HostPtrToTableMap.end())
return &TableMapIt->second;
// We don't have a map. So search all the registered libraries.
TableMap *TM = nullptr;
std::lock_guard<std::mutex> TrlTblLock(PM->TrlTblMtx);
for (HostEntriesBeginToTransTableTy::iterator Itr =
PM->HostEntriesBeginToTransTable.begin();
Itr != PM->HostEntriesBeginToTransTable.end(); ++Itr) {
// get the translation table (which contains all the good info).
TranslationTable *TransTable = &Itr->second;
// iterate over all the host table entries to see if we can locate the
// host_ptr.
__tgt_offload_entry *Cur = TransTable->HostTable.EntriesBegin;
for (uint32_t I = 0; Cur < TransTable->HostTable.EntriesEnd; ++Cur, ++I) {
if (Cur->addr != HostPtr)
continue;
// we got a match, now fill the HostPtrToTableMap so that we
// may avoid this search next time.
TM = &(PM->HostPtrToTableMap)[HostPtr];
TM->Table = TransTable;
TM->Index = I;
return TM;
}
}
return nullptr;
}
/// A class manages private arguments in a target region.
class PrivateArgumentManagerTy {
/// A data structure for the information of first-private arguments. We can
/// use this information to optimize data transfer by packing all
/// first-private arguments and transfer them all at once.