forked from KhronosGroup/SPIRV-LLVM-Translator
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathSPIRVToOCL.cpp
1243 lines (1142 loc) · 43.8 KB
/
SPIRVToOCL.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
//===- SPIRVToOCL.cpp - Transform SPIR-V builtins to OCL builtins------===//
//
// The LLVM/SPIRV Translator
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
// Copyright (c) 2014 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a
// copy of this software and associated documentation files (the "Software"),
// to deal with the Software without restriction, including without limitation
// the rights to use, copy, modify, merge, publish, distribute, sublicense,
// and/or sell copies of the Software, and to permit persons to whom the
// Software is furnished to do so, subject to the following conditions:
//
// Redistributions of source code must retain the above copyright notice,
// this list of conditions and the following disclaimers.
// Redistributions in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimers in the documentation
// and/or other materials provided with the distribution.
// Neither the names of Advanced Micro Devices, Inc., nor the names of its
// contributors may be used to endorse or promote products derived from this
// Software without specific prior written permission.
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS WITH
// THE SOFTWARE.
//
//===----------------------------------------------------------------------===//
//
// This file implements common transform of SPIR-V builtins to OCL builtins.
//
// Some of the visit functions are translations to OCL2.0 builtins, but they
// are currently used also for OCL1.2, so theirs implementations are placed
// in this pass as a common functionality for both versions.
//
//===----------------------------------------------------------------------===//
#include "SPIRVToOCL.h"
#include "llvm/IR/TypedPointerType.h"
#include "llvm/IR/Verifier.h"
#include "llvm/Support/CommandLine.h"
#define DEBUG_TYPE "spvtocl"
namespace SPIRV {
void SPIRVToOCLBase::visitCallInst(CallInst &CI) {
LLVM_DEBUG(dbgs() << "[visistCallInst] " << CI << '\n');
auto *F = CI.getCalledFunction();
if (!F)
return;
OCLExtOpKind ExtOp;
if (isSPIRVOCLExtInst(&CI, &ExtOp)) {
switch (ExtOp) {
case OpenCLLIB::Vloadn:
case OpenCLLIB::Vloada_halfn:
case OpenCLLIB::Vload_halfn:
visitCallSPIRVVLoadn(&CI, ExtOp);
break;
case OpenCLLIB::Vstoren:
case OpenCLLIB::Vstore_halfn:
case OpenCLLIB::Vstorea_halfn:
case OpenCLLIB::Vstore_half_r:
case OpenCLLIB::Vstore_halfn_r:
case OpenCLLIB::Vstorea_halfn_r:
visitCallSPIRVVStore(&CI, ExtOp);
break;
case OpenCLLIB::Printf: {
// TODO: Lower the printf instruction with the non-constant address space
// format string to suitable for OpenCL representation
auto *PT = dyn_cast<PointerType>(CI.getOperand(0)->getType());
if (PT && PT->getAddressSpace() == SPIR::TypeAttributeEnum::ATTR_CONST)
visitCallSPIRVPrintf(&CI, ExtOp);
break;
}
default:
visitCallSPIRVOCLExt(&CI, ExtOp);
break;
}
return;
}
auto MangledName = F->getName();
StringRef DemangledName;
Op OC = OpNop;
SPIRVBuiltinVariableKind BuiltinKind = SPIRVBuiltinVariableKind::BuiltInMax;
if (!oclIsBuiltin(MangledName, DemangledName) ||
((OC = getSPIRVFuncOC(DemangledName)) == OpNop &&
!getSPIRVBuiltin(DemangledName.str(), BuiltinKind)))
return;
LLVM_DEBUG(dbgs() << "DemangledName = " << DemangledName.str() << '\n'
<< "OpCode = " << OC << '\n'
<< "BuiltinKind = " << BuiltinKind << '\n');
if (BuiltinKind != SPIRVBuiltinVariableKind::BuiltInMax) {
if (static_cast<uint32_t>(BuiltinKind) >=
internal::BuiltInSubDeviceIDINTEL &&
static_cast<uint32_t>(BuiltinKind) <=
internal::BuiltInGlobalHWThreadIDINTEL)
return;
visitCallSPIRVBuiltin(&CI, BuiltinKind);
return;
}
if (OC == OpImageQuerySize || OC == OpImageQuerySizeLod) {
visitCallSPIRVImageQuerySize(&CI);
return;
}
if (OC == OpMemoryBarrier) {
visitCallSPIRVMemoryBarrier(&CI);
return;
}
if (OC == OpControlBarrier) {
visitCallSPIRVControlBarrier(&CI);
}
if (isSplitBarrierINTELOpCode(OC)) {
visitCallSPIRVSplitBarrierINTEL(&CI, OC);
return;
}
if (isAtomicOpCode(OC)) {
visitCallSPIRVAtomicBuiltin(&CI, OC);
return;
}
if (isGroupOpCode(OC) || isGroupNonUniformOpcode(OC)) {
visitCallSPIRVGroupBuiltin(&CI, OC);
return;
}
if (isPipeOpCode(OC)) {
visitCallSPIRVPipeBuiltin(&CI, OC);
return;
}
if (isMediaBlockINTELOpcode(OC)) {
visitCallSPIRVImageMediaBlockBuiltin(&CI, OC);
return;
}
if (isIntelSubgroupOpCode(OC)) {
visitCallSPIRVSubgroupINTELBuiltIn(&CI, OC);
return;
}
if (isSubgroupAvcINTELEvaluateOpcode(OC)) {
visitCallSPIRVAvcINTELEvaluateBuiltIn(&CI, OC);
return;
}
if (isSubgroupAvcINTELInstructionOpCode(OC)) {
visitCallSPIRVAvcINTELInstructionBuiltin(&CI, OC);
return;
}
if (OC == OpBuildNDRange) {
visitCallBuildNDRangeBuiltIn(&CI, OC, DemangledName);
return;
}
if (OC == OpGenericCastToPtr) {
visitCallGenericCastToPtrBuiltIn(&CI, OC);
return;
}
if (OC == OpGenericCastToPtrExplicit) {
visitCallGenericCastToPtrExplicitBuiltIn(&CI, OC);
return;
}
if (isCvtOpCode(OC)) {
visitCallSPIRVCvtBuiltin(&CI, OC, DemangledName);
return;
}
if (OC == OpGroupAsyncCopy) {
visitCallAsyncWorkGroupCopy(&CI, OC);
return;
}
if (OC == OpGroupWaitEvents) {
visitCallGroupWaitEvents(&CI, OC);
return;
}
if (OC == OpImageSampleExplicitLod) {
visitCallSPIRVImageSampleExplicitLodBuiltIn(&CI, OC);
return;
}
if (OC == OpImageWrite) {
visitCallSPIRVImageWriteBuiltIn(&CI, OC);
return;
}
if (OC == OpImageRead) {
visitCallSPIRVImageReadBuiltIn(&CI, OC);
return;
}
if (OC == OpImageQueryOrder || OC == OpImageQueryFormat) {
visitCallSPIRVImageQueryBuiltIn(&CI, OC);
return;
}
if (OC == OpEnqueueKernel) {
visitCallSPIRVEnqueueKernel(&CI, OC);
return;
}
if (OC == OpGenericPtrMemSemantics) {
visitCallSPIRVGenericPtrMemSemantics(&CI);
return;
}
// Check if OC is OpenCL relational builtin except bitselect and select.
auto IsOclRelationalOp = [](Op OC) {
return isUnaryPredicateOpCode(OC) || OC == OpOrdered || OC == OpUnordered ||
OC == OpFOrdEqual || OC == OpFUnordNotEqual ||
OC == OpFOrdGreaterThan || OC == OpFOrdGreaterThanEqual ||
OC == OpFOrdLessThan || OC == OpFOrdLessThanEqual ||
OC == OpFOrdNotEqual;
};
if (IsOclRelationalOp(OC)) {
if (OC == OpAny || OC == OpAll)
visitCallSPIRVAnyAll(&CI, OC);
else
visitCallSPIRVRelational(&CI, OC);
return;
}
if (OC == OpReadClockKHR) {
visitCallSPIRVReadClockKHR(&CI);
return;
}
if (OC == OpConvertFToBF16INTEL || OC == OpConvertBF16ToFINTEL) {
visitCallSPIRVBFloat16Conversions(&CI, OC);
return;
}
if (OC == OpSDot || OC == OpUDot || OC == OpSUDot || OC == OpSDotAccSat ||
OC == OpUDotAccSat || OC == OpSUDotAccSat) {
visitCallSPIRVDot(&CI, OC, DemangledName);
return;
}
if (OCLSPIRVBuiltinMap::rfind(OC))
visitCallSPIRVBuiltin(&CI, OC);
}
void SPIRVToOCLBase::visitCastInst(CastInst &Cast) {
if (!isa<ZExtInst>(Cast) && !isa<SExtInst>(Cast) && !isa<TruncInst>(Cast) &&
!isa<FPTruncInst>(Cast) && !isa<FPExtInst>(Cast) &&
!isa<FPToUIInst>(Cast) && !isa<FPToSIInst>(Cast) &&
!isa<UIToFPInst>(Cast) && !isa<SIToFPInst>(Cast))
return;
Type const *SrcTy = Cast.getSrcTy();
Type *DstVecTy = Cast.getDestTy();
// Leave scalar casts as is. Skip boolean vector casts becase there
// are no suitable OCL built-ins.
if (!DstVecTy->isVectorTy() || SrcTy->getScalarSizeInBits() == 1 ||
DstVecTy->getScalarSizeInBits() == 1)
return;
// Assemble built-in name -> convert_gentypeN
std::string CastBuiltInName(kOCLBuiltinName::ConvertPrefix);
// Check if this is 'floating point -> unsigned integer' cast
CastBuiltInName += mapLLVMTypeToOCLType(DstVecTy, !isa<FPToUIInst>(Cast));
// Replace LLVM conversion instruction with call to conversion built-in
BuiltinFuncMangleInfo Mangle;
// It does matter if the source is unsigned integer or not. SExt is for
// signed source, ZExt and UIToFPInst are for unsigned source.
if (isa<ZExtInst>(Cast) || isa<UIToFPInst>(Cast))
Mangle.addUnsignedArg(0);
AttributeList Attributes;
CallInst *Call =
addCallInst(M, CastBuiltInName, DstVecTy, Cast.getOperand(0), &Attributes,
&Cast, &Mangle, Cast.getName(), false);
Cast.replaceAllUsesWith(Call);
Cast.eraseFromParent();
}
void SPIRVToOCLBase::visitCallSPIRVImageQuerySize(CallInst *CI) {
// Get image type
Type *ImgTy = getCallValueType(CI, 0);
auto Desc = getImageDescriptor(ImgTy);
unsigned ImgDim = getImageDimension(Desc.Dim);
bool ImgArray = Desc.Arrayed;
AttributeList Attributes = CI->getCalledFunction()->getAttributes();
BuiltinFuncMangleInfo Mangle;
Mangle.getTypeMangleInfo(0).PointerTy = ImgTy;
Type *Int32Ty = Type::getInt32Ty(*Ctx);
Instruction *GetImageSize = nullptr;
if (ImgDim == 1) {
// OpImageQuerySize from non-arrayed 1d image is always translated
// into get_image_width returning scalar argument
GetImageSize = addCallInst(M, kOCLBuiltinName::GetImageWidth, Int32Ty,
CI->getArgOperand(0), &Attributes, CI, &Mangle,
CI->getName(), false);
// The width of integer type returning by OpImageQuerySize[Lod] may
// differ from i32
if (CI->getType()->getScalarType() != Int32Ty) {
GetImageSize = CastInst::CreateIntegerCast(
GetImageSize, CI->getType()->getScalarType(), false, CI->getName(),
CI->getIterator());
}
} else {
assert((ImgDim == 2 || ImgDim == 3) && "invalid image type");
assert(CI->getType()->isVectorTy() &&
"this code can handle vector result type only");
// get_image_dim returns int2 and int4 for 2d and 3d images respecitvely.
const unsigned ImgDimRetEls = ImgDim == 2 ? 2 : 4;
VectorType *RetTy = FixedVectorType::get(Int32Ty, ImgDimRetEls);
GetImageSize = addCallInst(M, kOCLBuiltinName::GetImageDim, RetTy,
CI->getArgOperand(0), &Attributes, CI, &Mangle,
CI->getName(), false);
// The width of integer type returning by OpImageQuerySize[Lod] may
// differ from i32
if (CI->getType()->getScalarType() != Int32Ty) {
GetImageSize = CastInst::CreateIntegerCast(
GetImageSize,
FixedVectorType::get(
CI->getType()->getScalarType(),
cast<FixedVectorType>(GetImageSize->getType())->getNumElements()),
false, CI->getName(), CI->getIterator());
}
}
if (ImgArray || ImgDim == 3) {
auto *VecTy = cast<FixedVectorType>(CI->getType());
const unsigned ImgQuerySizeRetEls = VecTy->getNumElements();
if (ImgDim == 1) {
// get_image_width returns scalar result while OpImageQuerySize
// for image1d_array_t returns <2 x i32> vector.
assert(ImgQuerySizeRetEls == 2 &&
"OpImageQuerySize[Lod] must return <2 x iN> vector type");
GetImageSize = InsertElementInst::Create(
PoisonValue::get(VecTy), GetImageSize, ConstantInt::get(Int32Ty, 0),
CI->getName(), CI->getIterator());
} else {
// get_image_dim and OpImageQuerySize returns different vector
// types for arrayed and 3d images.
SmallVector<Constant *, 4> MaskEls;
for (unsigned Idx = 0; Idx < ImgQuerySizeRetEls; ++Idx)
MaskEls.push_back(ConstantInt::get(Int32Ty, Idx));
Constant *Mask = ConstantVector::get(MaskEls);
GetImageSize = new ShuffleVectorInst(
GetImageSize, PoisonValue::get(GetImageSize->getType()), Mask,
CI->getName(), CI->getIterator());
}
}
if (ImgArray) {
assert((ImgDim == 1 || ImgDim == 2) && "invalid image array type");
// Insert get_image_array_size to the last position of the resulting vector.
auto *VecTy = cast<FixedVectorType>(CI->getType());
Type *SizeTy =
Type::getIntNTy(*Ctx, M->getDataLayout().getPointerSizeInBits(0));
Instruction *GetImageArraySize = addCallInst(
M, kOCLBuiltinName::GetImageArraySize, SizeTy, CI->getArgOperand(0),
&Attributes, CI, &Mangle, CI->getName(), false);
// The width of integer type returning by OpImageQuerySize[Lod] may
// differ from size_t which is returned by get_image_array_size
if (GetImageArraySize->getType() != VecTy->getElementType()) {
GetImageArraySize = CastInst::CreateIntegerCast(
GetImageArraySize, VecTy->getElementType(), false, CI->getName(),
CI->getIterator());
}
GetImageSize = InsertElementInst::Create(
GetImageSize, GetImageArraySize,
ConstantInt::get(Int32Ty, VecTy->getNumElements() - 1), CI->getName(),
CI->getIterator());
}
assert(GetImageSize && "must not be null");
CI->replaceAllUsesWith(GetImageSize);
CI->eraseFromParent();
}
std::string SPIRVToOCLBase::getUniformArithmeticBuiltinName(CallInst *CI,
Op OC) {
assert(isUniformArithmeticOpCode(OC) &&
"Not intended to handle other than uniform arithmetic opcodes!");
auto FuncName = OCLSPIRVBuiltinMap::rmap(OC);
std::string Prefix = getGroupBuiltinPrefix(CI);
std::string Op = FuncName;
Op.erase(0, strlen(kSPIRVName::GroupPrefix));
// unsigned prefix cannot be removed yet, as it is necessary to properly
// mangle the function
bool Unsigned = Op.front() == 'u';
if (!Unsigned)
Op = Op.erase(0, 1);
std::string GroupOp;
auto GO = getArgAs<spv::GroupOperation>(CI, 1);
switch (GO) {
case GroupOperationReduce:
GroupOp = "reduce";
break;
case GroupOperationInclusiveScan:
GroupOp = "scan_inclusive";
break;
case GroupOperationExclusiveScan:
GroupOp = "scan_exclusive";
break;
default:
llvm_unreachable("Unsupported group operation!");
break;
}
return Prefix + kSPIRVName::GroupPrefix + GroupOp + "_" + Op;
}
std::string SPIRVToOCLBase::getNonUniformArithmeticBuiltinName(CallInst *CI,
Op OC) {
assert(isNonUniformArithmeticOpCode(OC) &&
"Not intended to handle other than non uniform arithmetic opcodes!");
std::string Prefix = getGroupBuiltinPrefix(CI);
assert((Prefix == kOCLBuiltinName::SubPrefix) &&
"Workgroup scope is not supported for OpGroupNonUniform opcodes");
auto FuncName = OCLSPIRVBuiltinMap::rmap(OC);
std::string Op = FuncName;
Op.erase(0, strlen(kSPIRVName::GroupNonUniformPrefix));
if (!isGroupLogicalOpCode(OC)) {
// unsigned prefix cannot be removed yet, as it is necessary to properly
// mangle the function
const char Sign = Op.front();
bool Signed = (Sign == 'i' || Sign == 'f' || Sign == 's');
if (Signed)
Op = Op.erase(0, 1);
else
assert((Sign == 'u') && "Incorrect sign!");
} else { // LogicalOpcode
assert(
(Op == "logical_iand" || Op == "logical_ior" || Op == "logical_ixor") &&
"Incorrect logical operation");
Op = Op.erase(8, 1);
}
std::string GroupOp;
std::string GroupPrefix = kSPIRVName::GroupNonUniformPrefix;
auto GO = getArgAs<spv::GroupOperation>(CI, 1);
switch (GO) {
case GroupOperationReduce:
GroupOp = "reduce";
break;
case GroupOperationInclusiveScan:
GroupOp = "scan_inclusive";
break;
case GroupOperationExclusiveScan:
GroupOp = "scan_exclusive";
break;
case GroupOperationClusteredReduce:
GroupOp = "clustered_reduce";
// OpenCL clustered builtin has no non_uniform prefix, ex.
// sub_group_reduce_clustered_logical_and
GroupPrefix = kSPIRVName::GroupPrefix;
break;
default:
llvm_unreachable("Unsupported group operation!");
break;
}
return Prefix + GroupPrefix + GroupOp + "_" + Op;
}
std::string SPIRVToOCLBase::getBallotBuiltinName(CallInst *CI, Op OC) {
assert((OC == OpGroupNonUniformBallotBitCount) &&
"Not inteded to handle other opcodes than "
"OpGroupNonUniformBallotBitCount!");
std::string Prefix = getGroupBuiltinPrefix(CI);
assert(
(Prefix == kOCLBuiltinName::SubPrefix) &&
"Workgroup scope is not supported for OpGroupNonUniformBallotBitCount");
std::string GroupOp;
auto GO = getArgAs<spv::GroupOperation>(CI, 1);
switch (GO) {
case GroupOperationReduce:
GroupOp = "bit_count";
break;
case GroupOperationInclusiveScan:
GroupOp = "inclusive_scan";
break;
case GroupOperationExclusiveScan:
GroupOp = "exclusive_scan";
break;
default:
llvm_unreachable("Unsupported group operation!");
break;
}
return Prefix + kSPIRVName::GroupPrefix + "ballot_" + GroupOp;
}
std::string SPIRVToOCLBase::getRotateBuiltinName(CallInst *CI, Op OC) {
assert((OC == OpGroupNonUniformRotateKHR) &&
"Not intended to handle other opcodes");
std::string Prefix = getGroupBuiltinPrefix(CI);
assert((Prefix == kOCLBuiltinName::SubPrefix) &&
"Workgroup scope is not supported for OpGroupNonUniformRotateKHR");
std::string OptionalClustered;
if (CI->arg_size() == 4)
OptionalClustered = "clustered_";
return Prefix + kSPIRVName::GroupPrefix + OptionalClustered + "rotate";
}
std::string SPIRVToOCLBase::groupOCToOCLBuiltinName(CallInst *CI, Op OC) {
if (OC == OpGroupNonUniformRotateKHR)
return getRotateBuiltinName(CI, OC);
auto FuncName = OCLSPIRVBuiltinMap::rmap(OC);
assert(FuncName.find(kSPIRVName::GroupPrefix) == 0);
if (!hasGroupOperation(OC)) {
/// Transform OpenCL group builtin function names from group_
/// to work_group_ and sub_group_.
FuncName = getGroupBuiltinPrefix(CI) + FuncName;
} else { // Opcodes with group operation parameter
if (isUniformArithmeticOpCode(OC))
FuncName = getUniformArithmeticBuiltinName(CI, OC);
else if (isNonUniformArithmeticOpCode(OC))
FuncName = getNonUniformArithmeticBuiltinName(CI, OC);
else if (OC == OpGroupNonUniformBallotBitCount)
FuncName = getBallotBuiltinName(CI, OC);
else
llvm_unreachable("Unsupported opcode!");
}
return FuncName;
}
/// Return true if the original boolean return type needs to be changed to i32
/// when mapping the SPIR-V op to an OpenCL builtin.
static bool needsInt32RetTy(Op OC) {
return OC == OpGroupAny || OC == OpGroupAll || OC == OpGroupNonUniformAny ||
OC == OpGroupNonUniformAll || OC == OpGroupNonUniformAllEqual ||
OC == OpGroupNonUniformElect || OC == OpGroupNonUniformInverseBallot ||
OC == OpGroupNonUniformBallotBitExtract || isGroupLogicalOpCode(OC);
}
void SPIRVToOCLBase::visitCallSPIRVGroupBuiltin(CallInst *CI, Op OC) {
auto FuncName = groupOCToOCLBuiltinName(CI, OC);
auto Mutator = mutateCallInst(CI, FuncName);
/// Remove Group Operation argument,
/// as in OpenCL representation this is included in the function name
Mutator.removeArgs(0, (hasGroupOperation(OC) ? 2 : 1));
Type *Int32Ty = Type::getInt32Ty(*Ctx);
bool HasArg0ExtendedToi32 =
OC == OpGroupAny || OC == OpGroupAll || OC == OpGroupNonUniformAny ||
OC == OpGroupNonUniformAll || OC == OpGroupNonUniformBallot ||
isGroupLogicalOpCode(OC);
// Handle function arguments
if (OC == OpGroupBroadcast) {
Value *VecArg = Mutator.getArg(1);
if (auto *VT = dyn_cast<FixedVectorType>(VecArg->getType())) {
unsigned NumElements = VT->getNumElements();
for (unsigned I = 0; I < NumElements; I++)
Mutator.insertArg(1 + I, Mutator.Builder.CreateExtractElement(
VecArg, Mutator.Builder.getInt32(I)));
Mutator.removeArg(1 + NumElements);
}
} else if (HasArg0ExtendedToi32)
Mutator.mapArg(0, [](IRBuilder<> &Builder, Value *V) {
return Builder.CreateZExt(V, Builder.getInt32Ty());
});
// Handle function return type
if (needsInt32RetTy(OC))
Mutator.changeReturnType(Int32Ty, [](IRBuilder<> &Builder, CallInst *CI) {
// The OpenCL builtin returns a non-zero integer value. Convert to a
// boolean value.
return Builder.CreateICmpNE(CI, Builder.getInt32(0));
});
}
void SPIRVToOCLBase::visitCallSPIRVPipeBuiltin(CallInst *CI, Op OC) {
auto DemangledName = OCLSPIRVBuiltinMap::rmap(OC);
bool HasScope = DemangledName.find(kSPIRVName::GroupPrefix) == 0;
if (HasScope)
DemangledName = getGroupBuiltinPrefix(CI) + DemangledName;
assert(CI->getCalledFunction() && "Unexpected indirect call");
auto Mutator = mutateCallInst(CI, DemangledName);
if (HasScope)
Mutator.removeArg(0);
if (OC == OpReadPipe || OC == OpWritePipe || OC == OpReservedReadPipe ||
OC == OpReservedWritePipe || OC == OpReadPipeBlockingINTEL ||
OC == OpWritePipeBlockingINTEL) {
Mutator.mapArg(Mutator.arg_size() - 3, [](IRBuilder<> &Builder, Value *P) {
Type *T = P->getType();
assert(isa<PointerType>(T));
auto *NewTy = Builder.getPtrTy(SPIRAS_Generic);
if (T != NewTy) {
P = Builder.CreateAddrSpaceCast(P, NewTy);
}
return std::make_pair(
P, TypedPointerType::get(Builder.getInt8Ty(), SPIRAS_Generic));
});
}
}
void SPIRVToOCLBase::visitCallSPIRVImageMediaBlockBuiltin(CallInst *CI, Op OC) {
Type *RetType = CI->getType();
if (OC == OpSubgroupImageMediaBlockWriteINTEL) {
assert(CI->arg_size() >= 5 && "Wrong media block write signature");
RetType = CI->getArgOperand(4)->getType(); // texel type
}
unsigned int BitWidth = RetType->getScalarSizeInBits();
std::string FuncPostfix;
if (BitWidth == 8)
FuncPostfix = "_uc";
else if (BitWidth == 16)
FuncPostfix = "_us";
else if (BitWidth == 32)
FuncPostfix = "_ui";
else
assert(0 && "Unsupported texel type!");
if (auto *VecTy = dyn_cast<FixedVectorType>(RetType)) {
unsigned int NumEl = VecTy->getNumElements();
assert((NumEl == 2 || NumEl == 4 || NumEl == 8 || NumEl == 16) &&
"Wrong function type!");
FuncPostfix += std::to_string(NumEl);
}
mutateCallInst(CI, OCLSPIRVBuiltinMap::rmap(OC) + FuncPostfix)
.moveArg(0, CI->arg_size() - 1);
}
void SPIRVToOCLBase::visitCallBuildNDRangeBuiltIn(CallInst *CI, Op OC,
StringRef DemangledName) {
assert(CI->getCalledFunction() && "Unexpected indirect call");
// __spirv_BuildNDRange_nD, drop __spirv_
StringRef S = DemangledName;
S = S.drop_front(strlen(kSPIRVName::Prefix));
SmallVector<StringRef, 8> Split;
// BuildNDRange_nD
S.split(Split, kSPIRVPostfix::Divider,
/*MaxSplit=*/-1, /*KeepEmpty=*/false);
assert(Split.size() >= 2 && "Invalid SPIRV function name");
// Cut _nD and add it to function name.
mutateCallInst(CI, std::string(kOCLBuiltinName::NDRangePrefix) +
Split[1].substr(0, 3).str())
// OpenCL built-in has another order of parameters.
.moveArg(2, 0);
}
void SPIRVToOCLBase::visitCallGenericCastToPtrBuiltIn(CallInst *CI, Op OC) {
assert(CI->getCalledFunction() && "Unexpected indirect call");
IRBuilder<> Builder(CI);
Value *PtrArg = CI->getArgOperand(0);
auto AddrSpace =
static_cast<SPIRAddressSpace>(CI->getType()->getPointerAddressSpace());
Type *NewTy = PointerType::get(PtrArg->getType(), AddrSpace);
Value *ASC = Builder.CreateAddrSpaceCast(PtrArg, NewTy);
CI->replaceAllUsesWith(ASC);
CI->eraseFromParent();
}
void SPIRVToOCLBase::visitCallGenericCastToPtrExplicitBuiltIn(CallInst *CI,
Op OC) {
assert(CI->getCalledFunction() && "Unexpected indirect call");
StringRef Name;
auto AddrSpace =
static_cast<SPIRAddressSpace>(CI->getType()->getPointerAddressSpace());
switch (AddrSpace) {
case SPIRAS_Global:
Name = kOCLBuiltinName::ToGlobal;
break;
case SPIRAS_Local:
Name = kOCLBuiltinName::ToLocal;
break;
case SPIRAS_Private:
Name = kOCLBuiltinName::ToPrivate;
break;
default:
llvm_unreachable("Invalid address space");
}
mutateCallInst(CI, Name.str())
// The instruction has two arguments, whereas ocl built-in has only one
// argument.
.removeArg(1);
}
void SPIRVToOCLBase::visitCallSPIRVCvtBuiltin(CallInst *CI, Op OC,
StringRef DemangledName) {
std::string CastBuiltInName;
if (isCvtFromUnsignedOpCode(OC))
CastBuiltInName = "u";
CastBuiltInName += kOCLBuiltinName::ConvertPrefix;
Type *DstTy = CI->getType();
CastBuiltInName += mapLLVMTypeToOCLType(DstTy, !isCvtToUnsignedOpCode(OC));
if (DemangledName.find("_sat") != StringRef::npos || isSatCvtOpCode(OC))
CastBuiltInName += "_sat";
Value *Src = CI->getOperand(0);
assert(Src && "Invalid SPIRV convert builtin call");
Type *SrcTy = Src->getType();
auto Loc = DemangledName.find("_rt");
if (Loc != StringRef::npos &&
!(isa<IntegerType>(SrcTy) && isa<IntegerType>(DstTy)))
CastBuiltInName += DemangledName.substr(Loc, 4).str();
mutateCallInst(CI, CastBuiltInName);
}
void SPIRVToOCLBase::visitCallAsyncWorkGroupCopy(CallInst *CI, Op OC) {
// First argument of AsyncWorkGroupCopy instruction is Scope, OCL
// built-in async_work_group_strided_copy doesn't have this argument
mutateCallInst(CI, OCLSPIRVBuiltinMap::rmap(OC)).removeArg(0);
}
void SPIRVToOCLBase::visitCallGroupWaitEvents(CallInst *CI, Op OC) {
// First argument of GroupWaitEvents instruction is Scope, OCL
// built-in wait_group_events doesn't have this argument
mutateCallInst(CI, OCLSPIRVBuiltinMap::rmap(OC)).removeArg(0);
}
static std::string getTypeSuffix(Type *T, bool IsSigned) {
std::string Suffix;
Type *ST = T->getScalarType();
if (ST->isHalfTy())
Suffix = "h";
else if (ST->isFloatTy())
Suffix = "f";
else if (IsSigned)
Suffix = "i";
else
Suffix = "ui";
return Suffix;
}
BuiltinCallMutator
SPIRVToOCLBase::mutateCallImageOperands(CallInst *CI, StringRef NewFuncName,
Type *T, unsigned ImOpArgIndex) {
// Default to signed.
bool IsSigned = true;
uint64_t ImOpValue = 0;
if (CI->arg_size() > ImOpArgIndex) {
ConstantInt *ImOp = dyn_cast<ConstantInt>(CI->getArgOperand(ImOpArgIndex));
if (ImOp)
ImOpValue = ImOp->getZExtValue();
unsigned SignZeroExtMasks = ImageOperandsMask::ImageOperandsSignExtendMask |
ImageOperandsMask::ImageOperandsZeroExtendMask;
// If one of the SPIR-V 1.4 SignExtend/ZeroExtend operands is present, take
// it into account and drop the mask.
if (ImOpValue & SignZeroExtMasks) {
if (ImOpValue & ImageOperandsMask::ImageOperandsZeroExtendMask)
IsSigned = false;
ImOpValue &= ~SignZeroExtMasks;
}
}
auto Mutator =
mutateCallInst(CI, NewFuncName.str() + getTypeSuffix(T, IsSigned));
if (ImOpArgIndex < Mutator.arg_size()) {
// Drop "Image Operands" argument.
Mutator.removeArg(ImOpArgIndex);
if (ImOpArgIndex < Mutator.arg_size()) {
ConstantFP *LodVal = dyn_cast<ConstantFP>(Mutator.getArg(ImOpArgIndex));
// If the image operand is LOD and its value is zero, drop it too.
if (LodVal && LodVal->isNullValue() &&
ImOpValue == ImageOperandsMask::ImageOperandsLodMask)
Mutator.removeArgs(ImOpArgIndex, Mutator.arg_size() - ImOpArgIndex);
}
}
return Mutator;
}
void SPIRVToOCLBase::visitCallSPIRVImageSampleExplicitLodBuiltIn(CallInst *CI,
Op OC) {
Type *T = CI->getType();
if (auto *VT = dyn_cast<VectorType>(T))
T = VT->getElementType();
auto Mutator =
mutateCallImageOperands(CI, kOCLBuiltinName::SampledReadImage, T, 2);
CallInst *CallSampledImg = cast<CallInst>(CI->getArgOperand(0));
auto Img = getCallValue(CallSampledImg, 0);
auto Sampler = getCallValue(CallSampledImg, 1);
bool IsDepthImage = false;
Mutator.mapArg(0, [&](Value *SampledImg) {
StringRef ImageTypeName;
if (isOCLImageType(Img.second, &ImageTypeName))
IsDepthImage = ImageTypeName.contains("_depth_");
if (CallSampledImg->hasOneUse()) {
CallSampledImg->replaceAllUsesWith(
PoisonValue::get(CallSampledImg->getType()));
CallSampledImg->dropAllReferences();
CallSampledImg->eraseFromParent();
}
return Img;
});
Mutator.insertArg(1, Sampler);
if (IsDepthImage)
Mutator.changeReturnType(T, [&](IRBuilder<> &Builder, CallInst *NewCI) {
return Builder.CreateInsertElement(
FixedVectorType::get(NewCI->getType(), 4), NewCI, uint64_t(0));
});
}
void SPIRVToOCLBase::visitCallSPIRVImageWriteBuiltIn(CallInst *CI, Op OC) {
auto Mutator = mutateCallImageOperands(CI, kOCLBuiltinName::WriteImage,
CI->getArgOperand(2)->getType(), 3);
if (Mutator.arg_size() > 3)
Mutator.moveArg(3, 2);
}
void SPIRVToOCLBase::visitCallSPIRVImageReadBuiltIn(CallInst *CI, Op OC) {
mutateCallImageOperands(CI, kOCLBuiltinName::ReadImage, CI->getType(), 2);
}
void SPIRVToOCLBase::visitCallSPIRVImageQueryBuiltIn(CallInst *CI, Op OC) {
mutateCallInst(CI, OCLSPIRVBuiltinMap::rmap(OC))
.changeReturnType(CI->getType(), [=](IRBuilder<> &Builder, CallInst *CI) {
unsigned int Offset = 0;
if (OC == OpImageQueryFormat)
Offset = OCLImageChannelDataTypeOffset;
else if (OC == OpImageQueryOrder)
Offset = OCLImageChannelOrderOffset;
else
llvm_unreachable("Unsupported opcode");
return Builder.CreateSub(CI, Builder.getInt32(Offset));
});
}
void SPIRVToOCLBase::visitCallSPIRVSubgroupINTELBuiltIn(CallInst *CI, Op OC) {
std::stringstream Name;
Type *DataTy = nullptr;
switch (OC) {
case OpSubgroupBlockReadINTEL:
case OpSubgroupImageBlockReadINTEL:
Name << "intel_sub_group_block_read";
DataTy = CI->getType();
break;
case OpSubgroupBlockWriteINTEL:
Name << "intel_sub_group_block_write";
DataTy = CI->getOperand(1)->getType();
break;
case OpSubgroupImageBlockWriteINTEL:
Name << "intel_sub_group_block_write";
DataTy = CI->getOperand(2)->getType();
break;
default:
Name << OCLSPIRVBuiltinMap::rmap(OC);
break;
}
if (DataTy) {
unsigned VectorNumElements = 1;
if (FixedVectorType *VT = dyn_cast<FixedVectorType>(DataTy))
VectorNumElements = VT->getNumElements();
unsigned ElementBitSize = DataTy->getScalarSizeInBits();
Name << getIntelSubgroupBlockDataPostfix(ElementBitSize, VectorNumElements);
}
mutateCallInst(CI, Name.str());
}
void SPIRVToOCLBase::visitCallSPIRVAvcINTELEvaluateBuiltIn(CallInst *CI,
Op OC) {
// There are three types of AVC Intel Evaluate opcodes:
// 1. With multi reference images - does not use OpVmeImageINTEL opcode
// for reference images
// 2. With dual reference images - uses two OpVmeImageINTEL opcodes for
// reference image
// 3. With single reference image - uses one OpVmeImageINTEL opcode for
// reference image
StringRef FnName = CI->getCalledFunction()->getName();
int NumImages = 0;
if (FnName.contains("SingleReference"))
NumImages = 2;
else if (FnName.contains("DualReference"))
NumImages = 3;
else if (FnName.contains("MultiReference"))
NumImages = 1;
else if (FnName.contains("EvaluateIpe"))
NumImages = 1;
auto EraseVmeImageCall = [](CallInst *CI) {
if (CI->hasOneUse()) {
CI->replaceAllUsesWith(PoisonValue::get(CI->getType()));
CI->dropAllReferences();
CI->eraseFromParent();
}
};
auto Mutator =
mutateCallInst(CI, OCLSPIRVSubgroupAVCIntelBuiltinMap::rmap(OC));
if (NumImages) {
CallInst *SrcImage = cast<CallInst>(Mutator.getArg(0));
if (NumImages == 1) {
// Multi reference opcode - remove src image OpVmeImageINTEL opcode
// and replace it with corresponding OpImage and OpSampler arguments
size_t SamplerPos = Mutator.arg_size() - 1;
Mutator.replaceArg(0, getCallValue(SrcImage, 0));
Mutator.insertArg(SamplerPos, getCallValue(SrcImage, 1));
} else {
CallInst *FwdRefImage = cast<CallInst>(Mutator.getArg(1));
CallInst *BwdRefImage =
NumImages == 3 ? cast<CallInst>(Mutator.getArg(2)) : nullptr;
// Single reference opcode - remove src and ref image
// OpVmeImageINTEL opcodes and replace them with src and ref OpImage
// opcodes and OpSampler
Mutator.removeArgs(0, NumImages);
// insert source OpImage and OpSampler
Mutator.insertArg(0, getCallValue(SrcImage, 0));
Mutator.insertArg(1, getCallValue(SrcImage, 1));
// insert reference OpImage
Mutator.insertArg(1, getCallValue(FwdRefImage, 0));
EraseVmeImageCall(SrcImage);
EraseVmeImageCall(FwdRefImage);
if (BwdRefImage) {
// Dual reference opcode - insert second reference OpImage argument
Mutator.insertArg(2, getCallValue(BwdRefImage, 0));
EraseVmeImageCall(BwdRefImage);
}
}
} else
llvm_unreachable("invalid avc instruction");
}
void SPIRVToOCLBase::visitCallSPIRVGenericPtrMemSemantics(CallInst *CI) {
mutateCallInst(CI, OCLSPIRVBuiltinMap::rmap(OpGenericPtrMemSemantics))
.changeReturnType(CI->getType(),
[](IRBuilder<> &Builder, CallInst *NewCI) {
return Builder.CreateShl(NewCI, Builder.getInt32(8));
});
}
void SPIRVToOCLBase::visitCallSPIRVBFloat16Conversions(CallInst *CI, Op OC) {
Type *ArgTy = CI->getOperand(0)->getType();
std::string N =
ArgTy->isVectorTy()
? std::to_string(cast<FixedVectorType>(ArgTy)->getNumElements())
: "";
std::string Name;
switch (static_cast<uint32_t>(OC)) {
case OpConvertFToBF16INTEL:
Name = "intel_convert_bfloat16" + N + "_as_ushort" + N;
break;
case OpConvertBF16ToFINTEL:
Name = "intel_convert_as_bfloat16" + N + "_float" + N;
break;
default:
break; // do nothing
}
mutateCallInst(CI, Name);
}
void SPIRVToOCLBase::visitCallSPIRVDot(CallInst *CI, Op OC,
StringRef DemangledName) {
// OpenCL only supports integer dot product builtins that have return types
// of int and uint.
if (!(DemangledName.contains("_Rint") || DemangledName.contains("_Ruint")))
return;
bool IsPacked = !CI->getOperand(0)->getType()->isVectorTy();
std::stringstream Name;
switch (OC) {
case OpSDot:
if (IsPacked)
Name << kOCLBuiltinName::Dot4x8PackedPrefix << "ss_int";
else
// Add an extra suffix to help determine signed/unsigned arguments
Name << kOCLBuiltinName::Dot << "_ss";
break;
case OpUDot:
if (IsPacked)
Name << kOCLBuiltinName::Dot4x8PackedPrefix << "uu_uint";
else
Name << kOCLBuiltinName::Dot << "_uu";
break;
case OpSUDot:
if (IsPacked)
Name << kOCLBuiltinName::Dot4x8PackedPrefix << "su_int";
else
Name << kOCLBuiltinName::Dot << "_su";
break;
case OpSDotAccSat:
if (IsPacked)
Name << kOCLBuiltinName::DotAccSat4x8PackedPrefix << "ss_int";
else
Name << kOCLBuiltinName::DotAccSat << "_ss";
break;
case OpUDotAccSat:
if (IsPacked)
Name << kOCLBuiltinName::DotAccSat4x8PackedPrefix << "uu_uint";
else
Name << kOCLBuiltinName::DotAccSat << "_uu";
break;
case OpSUDotAccSat:
if (IsPacked)
Name << kOCLBuiltinName::DotAccSat4x8PackedPrefix << "su_int";
else
Name << kOCLBuiltinName::DotAccSat << "_su";
break;
default:
break; // do nothing
}
auto Mutator = mutateCallInst(CI, Name.str());
if (IsPacked)
Mutator.removeArg(CI->arg_size() - 1);
}
void SPIRVToOCLBase::visitCallSPIRVBuiltin(CallInst *CI, Op OC) {
mutateCallInst(CI, OCLSPIRVBuiltinMap::rmap(OC));
}