LLVM: lib/Analysis/ConstantFolding.cpp Source File (original) (raw)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
31#include "llvm/Config/config.h"
45#include "llvm/IR/IntrinsicsAArch64.h"
46#include "llvm/IR/IntrinsicsAMDGPU.h"
47#include "llvm/IR/IntrinsicsARM.h"
48#include "llvm/IR/IntrinsicsNVPTX.h"
49#include "llvm/IR/IntrinsicsWebAssembly.h"
50#include "llvm/IR/IntrinsicsX86.h"
59#include
60#include
61#include
62#include
63#include
64
65using namespace llvm;
66
68 "disable-fp-call-folding",
69 cl::desc("Disable constant-folding of FP intrinsics and libcalls."),
71
72namespace {
73
74
75
76
77
78static Constant *foldConstVectorToAPInt(APInt &Result, Type *DestTy,
80 unsigned NumSrcElts,
82
83
84 unsigned BitShift = DL.getTypeSizeInBits(SrcEltTy);
85 for (unsigned i = 0; i != NumSrcElts; ++i) {
87 if (DL.isLittleEndian())
88 Element = C->getAggregateElement(NumSrcElts - i - 1);
89 else
90 Element = C->getAggregateElement(i);
91
93 Result <<= BitShift;
94 continue;
95 }
96
98 if (!ElementCI)
100
102 Result |= ElementCI->getValue().zext(Result.getBitWidth());
103 }
104
105 return nullptr;
106}
107
108
109
110
113 "Invalid constantexpr bitcast!");
114
115
117 return Res;
118
120
123 Type *SrcEltTy = VTy->getElementType();
124
125
126
131
133 }
134
136 if (Constant *CE = foldConstVectorToAPInt(Result, DestTy, C,
137 SrcEltTy, NumSrcElts, DL))
138 return CE;
139
141 return ConstantInt::get(DestTy, Result);
142
144 return ConstantFP::get(DestTy->getContext(), FP);
145 }
146 }
147
148
150 if (!DestVTy)
152
153
154
159 }
160
161
162
165
166
170
171
174 if (NumDstElt == NumSrcElt)
176
178 Type *DstEltTy = DestVTy->getElementType();
179
180
181
182
183
184
185
186
187
188
189
191
195
197
198
200 }
201
202
203
208
212 "Constant folding cannot fail for plain fp->int bitcast!");
213 }
214
215
216
217
218
219 bool isLittleEndian = DL.isLittleEndian();
220
222 if (NumDstElt < NumSrcElt) {
223
225 unsigned Ratio = NumSrcElt/NumDstElt;
227 unsigned SrcElt = 0;
228 for (unsigned i = 0; i != NumDstElt; ++i) {
229
231 unsigned ShiftAmt = isLittleEndian ? 0 : SrcBitSize*(Ratio-1);
232 for (unsigned j = 0; j != Ratio; ++j) {
233 Constant *Src = C->getAggregateElement(SrcElt++);
237 else
239 if (!Src)
241
242
245 assert(Src && "Constant folding cannot fail on plain integers");
246
247
249 Instruction::Shl, Src, ConstantInt::get(Src->getType(), ShiftAmt),
251 assert(Src && "Constant folding cannot fail on plain integers");
252
253 ShiftAmt += isLittleEndian ? SrcBitSize : -SrcBitSize;
254
255
257 assert(Elt && "Constant folding cannot fail on plain integers");
258 }
259 Result.push_back(Elt);
260 }
262 }
263
264
265 unsigned Ratio = NumDstElt/NumSrcElt;
266 unsigned DstBitSize = DL.getTypeSizeInBits(DstEltTy);
267
268
269 for (unsigned i = 0; i != NumSrcElt; ++i) {
270 auto *Element = C->getAggregateElement(i);
271
272 if (!Element)
274
276
278 continue;
279 }
280
282 if (!Src)
284
285 unsigned ShiftAmt = isLittleEndian ? 0 : DstBitSize*(Ratio-1);
286 for (unsigned j = 0; j != Ratio; ++j) {
287
288
289 APInt Elt = Src->getValue().lshr(ShiftAmt);
290 ShiftAmt += isLittleEndian ? DstBitSize : -DstBitSize;
291
292
293 Result.push_back(ConstantInt::get(DstEltTy, Elt.trunc(DstBitSize)));
294 }
295 }
296
298}
299
300}
301
302
303
307 if (DSOEquiv)
308 *DSOEquiv = nullptr;
309
310
314 return true;
315 }
316
318 if (DSOEquiv)
319 *DSOEquiv = FoundDSOEquiv;
320 GV = FoundDSOEquiv->getGlobalValue();
323 return true;
324 }
325
326
328 if (!CE) return false;
329
330
331 if (CE->getOpcode() == Instruction::PtrToInt ||
332 CE->getOpcode() == Instruction::PtrToAddr ||
333 CE->getOpcode() == Instruction::BitCast)
335 DSOEquiv);
336
337
339 if ()
340 return false;
341
342 unsigned BitWidth = DL.getIndexTypeSizeInBits(GEP->getType());
344
345
347 DSOEquiv))
348 return false;
349
350
351 if (->accumulateConstantOffset(DL, TmpOffset))
352 return false;
353
355 return true;
356}
357
360 do {
361 Type *SrcTy = C->getType();
362 if (SrcTy == DestTy)
363 return C;
364
365 TypeSize DestSize = DL.getTypeSizeInBits(DestTy);
366 TypeSize SrcSize = DL.getTypeSizeInBits(SrcTy);
368 return nullptr;
369
370
371
373 return Res;
374
375
376
377
378 if (SrcSize == DestSize &&
379 DL.isNonIntegralPointerType(SrcTy->getScalarType()) ==
382
383
384 if (SrcTy->isIntegerTy() && DestTy->isPointerTy())
385 Cast = Instruction::IntToPtr;
386 else if (SrcTy->isPointerTy() && DestTy->isIntegerTy())
387 Cast = Instruction::PtrToInt;
388
391 }
392
393
394
395 if (!SrcTy->isAggregateType() && !SrcTy->isVectorTy())
396 return nullptr;
397
398
399
400
401
402 if (SrcTy->isStructTy()) {
403
404
405 unsigned Elem = 0;
407 do {
408 ElemC = C->getAggregateElement(Elem++);
409 } while (ElemC && DL.getTypeSizeInBits(ElemC->getType()).isZero());
410 C = ElemC;
411 } else {
412
413
415 if (.typeSizeEqualsStoreSize(VT->getElementType()))
416 return nullptr;
417
418 C = C->getAggregateElement(0u);
419 }
420 } while (C);
421
422 return nullptr;
423}
424
425namespace {
426
427
428
429
430
431bool ReadDataFromGlobal(Constant *C, uint64_t ByteOffset, unsigned char *CurPtr,
433 assert(ByteOffset <= DL.getTypeAllocSize(C->getType()) &&
434 "Out of range access");
435
436
437 if (ByteOffset >= DL.getTypeStoreSize(C->getType()))
438 return true;
439
440
441
443 return true;
444
446 if ((CI->getBitWidth() & 7) != 0)
447 return false;
448 const APInt &Val = CI->getValue();
449 unsigned IntBytes = unsigned(CI->getBitWidth()/8);
450
451 for (unsigned i = 0; i != BytesLeft && ByteOffset != IntBytes; ++i) {
452 unsigned n = ByteOffset;
453 if (.isLittleEndian())
454 n = IntBytes - n - 1;
456 ++ByteOffset;
457 }
458 return true;
459 }
460
462 if (CFP->getType()->isDoubleTy()) {
464 return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, DL);
465 }
466 if (CFP->getType()->isFloatTy()){
468 return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, DL);
469 }
470 if (CFP->getType()->isHalfTy()){
472 return ReadDataFromGlobal(C, ByteOffset, CurPtr, BytesLeft, DL);
473 }
474 return false;
475 }
476
478 const StructLayout *SL = DL.getStructLayout(CS->getType());
481 ByteOffset -= CurEltOffset;
482
483 while (true) {
484
485
486 uint64_t EltSize = DL.getTypeAllocSize(CS->getOperand(Index)->getType());
487
488 if (ByteOffset < EltSize &&
489 !ReadDataFromGlobal(CS->getOperand(Index), ByteOffset, CurPtr,
490 BytesLeft, DL))
491 return false;
492
493 ++Index;
494
495
496 if (Index == CS->getType()->getNumElements())
497 return true;
498
499
501
502 if (BytesLeft <= NextEltOffset - CurEltOffset - ByteOffset)
503 return true;
504
505
506 CurPtr += NextEltOffset - CurEltOffset - ByteOffset;
507 BytesLeft -= NextEltOffset - CurEltOffset - ByteOffset;
508 ByteOffset = 0;
509 CurEltOffset = NextEltOffset;
510 }
511
512 }
513
519 NumElts = AT->getNumElements();
520 EltTy = AT->getElementType();
521 EltSize = DL.getTypeAllocSize(EltTy);
522 } else {
525
526
527 if (.typeSizeEqualsStoreSize(EltTy))
528 return false;
529
530 EltSize = DL.getTypeStoreSize(EltTy);
531 }
532 uint64_t Index = ByteOffset / EltSize;
534
535 for (; Index != NumElts; ++Index) {
536 if (!ReadDataFromGlobal(C->getAggregateElement(Index), Offset, CurPtr,
537 BytesLeft, DL))
538 return false;
539
541 assert(BytesWritten <= EltSize && "Not indexing into this element?");
542 if (BytesWritten >= BytesLeft)
543 return true;
544
546 BytesLeft -= BytesWritten;
547 CurPtr += BytesWritten;
548 }
549 return true;
550 }
551
553 if (CE->getOpcode() == Instruction::IntToPtr &&
554 CE->getOperand(0)->getType() == DL.getIntPtrType(CE->getType())) {
555 return ReadDataFromGlobal(CE->getOperand(0), ByteOffset, CurPtr,
556 BytesLeft, DL);
557 }
558 }
559
560
561 return false;
562}
563
566
568 return nullptr;
569
571
572
573 if (!IntType) {
574
575
576
577
580 return nullptr;
581
583 DL.getTypeSizeInBits(LoadTy).getFixedValue());
584 if (Constant *Res = FoldReinterpretLoadFromConst(C, MapTy, Offset, DL)) {
585 if (Res->isNullValue() && !LoadTy->isX86_AMXTy())
586
591
592 if (Res->isNullValue() && !LoadTy->isX86_AMXTy())
594 if (DL.isNonIntegralPointerType(LoadTy->getScalarType()))
595
596 return nullptr;
598 }
599 return Res;
600 }
601 return nullptr;
602 }
603
604 unsigned BytesLoaded = (IntType->getBitWidth() + 7) / 8;
605 if (BytesLoaded > 32 || BytesLoaded == 0)
606 return nullptr;
607
608
609 if (Offset <= -1 * static_cast<int64_t>(BytesLoaded))
611
612
613 TypeSize InitializerSize = DL.getTypeAllocSize(C->getType());
615 return nullptr;
616
617
620
621 unsigned char RawBytes[32] = {0};
622 unsigned char *CurPtr = RawBytes;
623 unsigned BytesLeft = BytesLoaded;
624
625
630 }
631
632 if (!ReadDataFromGlobal(C, Offset, CurPtr, BytesLeft, DL))
633 return nullptr;
634
635 APInt ResultVal = APInt(IntType->getBitWidth(), 0);
636 if (DL.isLittleEndian()) {
637 ResultVal = RawBytes[BytesLoaded - 1];
638 for (unsigned i = 1; i != BytesLoaded; ++i) {
639 ResultVal <<= 8;
640 ResultVal |= RawBytes[BytesLoaded - 1 - i];
641 }
642 } else {
643 ResultVal = RawBytes[0];
644 for (unsigned i = 1; i != BytesLoaded; ++i) {
645 ResultVal <<= 8;
646 ResultVal |= RawBytes[i];
647 }
648 }
649
650 return ConstantInt::get(IntType->getContext(), ResultVal);
651}
652
653}
654
655
656
657
661 return nullptr;
662
665 TypeSize InitSize = DL.getTypeAllocSize(Init->getType());
666 if (InitSize < Offset)
667 return nullptr;
668
670 if (NBytes > UINT16_MAX)
671
672
673
674
675 return nullptr;
676
678 unsigned char *CurPtr = RawBytes.data();
679
680 if (!ReadDataFromGlobal(Init, Offset, CurPtr, NBytes, DL))
681 return nullptr;
682
684}
685
686
687
692
694 return nullptr;
695
696 Type *ElemTy = Base->getType();
698 if (.isZero() || !Indices[0].isZero())
699 return nullptr;
700
703 if (Index.isNegative() || Index.getActiveBits() >= 32)
704 return nullptr;
705
706 C = C->getAggregateElement(Index.getZExtValue());
707 if ()
708 return nullptr;
709 }
710
711 return C;
712}
713
719 return Result;
720
721
722
724 if (.isScalable() && Offset.sge(Size.getFixedValue()))
726
727
729 return Result;
730
731
732 if (Offset.getSignificantBits() <= 64)
734 FoldReinterpretLoadFromConst(C, Ty, Offset.getSExtValue(), DL))
735 return Result;
736
737 return nullptr;
738}
739
744
748
749
751 if (!GV || !GV->isConstant() || !GV->hasDefinitiveInitializer())
752 return nullptr;
753
756
757 if (C == GV)
760 return Result;
761
762
763
765}
766
772
779
780
781 if (.typeSizeEqualsStoreSize(C->getType()))
782 return nullptr;
783 if (C->isNullValue() && !Ty->isX86_AMXTy())
785 if (C->isAllOnesValue() &&
786 (Ty->isIntOrIntVectorTy() || Ty->isFPOrFPVectorTy()))
788 return nullptr;
789}
790
791namespace {
792
793
794
795
796
799
800
801
802
803
804
805 if (Opc == Instruction::And) {
808 if ((Known1.One | Known0.Zero).isAllOnes()) {
809
810 return Op0;
811 }
812 if ((Known0.One | Known1.Zero).isAllOnes()) {
813
814 return Op1;
815 }
816
817 Known0 &= Known1;
820 }
821
822
823
824 if (Opc == Instruction::Sub) {
826 APInt Offs1, Offs2;
827
830 unsigned OpSize = DL.getTypeSizeInBits(Op0->getType());
831
832
833
834
837 }
838 }
839
840 return nullptr;
841}
842
843
844
847 std::optional InRange,
849 Type *IntIdxTy = DL.getIndexType(ResultTy);
851
852 bool Any = false;
854 for (unsigned i = 1, e = Ops.size(); i != e; ++i) {
855 if ((i == 1 ||
857 SrcElemTy, Ops.slice(1, i - 1)))) &&
858 Ops[i]->getType()->getScalarType() != IntIdxScalarTy) {
859 Any = true;
860 Type *NewType =
861 Ops[i]->getType()->isVectorTy() ? IntIdxTy : IntIdxScalarTy;
865 if (!NewIdx)
866 return nullptr;
868 } else
870 }
871
872 if ()
873 return nullptr;
874
878}
879
880
885 Type *SrcElemTy = GEP->getSourceElementType();
886 Type *ResTy = GEP->getType();
888 return nullptr;
889
890 if (Constant *C = CastGEPIndices(SrcElemTy, Ops, ResTy, GEP->getNoWrapFlags(),
891 GEP->getInRange(), DL, TLI))
892 return C;
893
896 return nullptr;
897
898 Type *IntIdxTy = DL.getIndexType(Ptr->getType());
899
900 for (unsigned i = 1, e = Ops.size(); i != e; ++i)
902 return nullptr;
903
904 unsigned BitWidth = DL.getTypeSizeInBits(IntIdxTy);
907 DL.getIndexedOffsetInType(
909 true, true);
910
911 std::optional InRange = GEP->getInRange();
914
915
917 bool Overflow = false;
919 NW &= GEP->getNoWrapFlags();
920
922
923
924 bool AllConstantInt = true;
925 for (Value *NestedOp : NestedOps)
927 AllConstantInt = false;
928 break;
929 }
930 if (!AllConstantInt)
931 break;
932
933
934 if (auto GEPRange = GEP->getInRange()) {
935 auto AdjustedGEPRange = GEPRange->sextOrTrunc(BitWidth).subtract(Offset);
937 InRange ? InRange->intersectWith(AdjustedGEPRange) : AdjustedGEPRange;
938 }
939
941 SrcElemTy = GEP->getSourceElementType();
943 APInt(BitWidth, DL.getIndexedOffsetInType(SrcElemTy, NestedOps),
944 true, true),
945 Overflow);
946 }
947
948
949
952
953
954
955 APInt BaseIntVal(DL.getPointerTypeSizeInBits(Ptr->getType()), 0);
957 if (CE->getOpcode() == Instruction::IntToPtr) {
959 BaseIntVal = Base->getValue().zextOrTrunc(BaseIntVal.getBitWidth());
960 }
961 }
962
963 if ((Ptr->isNullValue() || BaseIntVal != 0) &&
964 .mustNotIntroduceIntToPtr(Ptr->getType())) {
965
966
967
968 BaseIntVal.insertBits(BaseIntVal.trunc(BitWidth) + Offset, 0);
971 }
972
973
975 bool CanBeNull, CanBeFreed;
978 if (DerefBytes != 0 && !CanBeNull && Offset.sle(DerefBytes))
980 }
981
982
985
986
989 ConstantInt::get(Ctx, Offset), NW,
991}
992
993
994
995
996
997
998Constant *ConstantFoldInstOperandsImpl(const Value *InstOrCE, unsigned Opcode,
1002 bool AllowNonDeterministic) {
1004
1007
1009 switch (Opcode) {
1010 default:
1011 break;
1012 case Instruction::FAdd:
1013 case Instruction::FSub:
1014 case Instruction::FMul:
1015 case Instruction::FDiv:
1016 case Instruction::FRem:
1017
1018
1019
1022 AllowNonDeterministic);
1023 }
1024 }
1026 }
1027
1030
1032 Type *SrcElemTy = GEP->getSourceElementType();
1034 return nullptr;
1035
1037 return C;
1038
1040 GEP->getNoWrapFlags(),
1041 GEP->getInRange());
1042 }
1043
1045 return CE->getWithOperands(Ops);
1046
1047 switch (Opcode) {
1048 default: return nullptr;
1049 case Instruction::ICmp:
1050 case Instruction::FCmp: {
1054 }
1055 case Instruction::Freeze:
1057 case Instruction::Call:
1062 AllowNonDeterministic);
1063 }
1064 return nullptr;
1065 case Instruction::Select:
1067 case Instruction::ExtractElement:
1069 case Instruction::ExtractValue:
1072 case Instruction::InsertElement:
1074 case Instruction::InsertValue:
1077 case Instruction::ShuffleVector:
1080 case Instruction::Load: {
1082 if (LI->isVolatile())
1083 return nullptr;
1085 }
1086 }
1087}
1088
1089}
1090
1091
1092
1093
1094
1095namespace {
1096
1102 return const_cast<Constant *>(C);
1103
1105 for (const Use &OldU : C->operands()) {
1108
1109
1111 auto It = FoldedOps.find(OldC);
1112 if (It == FoldedOps.end()) {
1113 NewC = ConstantFoldConstantImpl(OldC, DL, TLI, FoldedOps);
1114 FoldedOps.insert({OldC, NewC});
1115 } else {
1116 NewC = It->second;
1117 }
1118 }
1119 Ops.push_back(NewC);
1120 }
1121
1123 if (Constant *Res = ConstantFoldInstOperandsImpl(
1124 CE, CE->getOpcode(), Ops, DL, TLI, true))
1125 return Res;
1126 return const_cast<Constant *>(C);
1127 }
1128
1131}
1132
1133}
1134
1138
1140 Constant *CommonValue = nullptr;
1141
1144
1145
1146
1147
1149 continue;
1150
1152 if ()
1153 return nullptr;
1154
1155 C = ConstantFoldConstantImpl(C, DL, TLI, FoldedOps);
1156
1157
1158 if (CommonValue && C != CommonValue)
1159 return nullptr;
1160 CommonValue = C;
1161 }
1162
1163
1164 return CommonValue ? CommonValue : UndefValue::get(PN->getType());
1165 }
1166
1167
1168
1169 if ((I->operands(), [](const Use &U) { return isa(U); }))
1170 return nullptr;
1171
1174 for (const Use &OpU : I->operands()) {
1176
1177 Op = ConstantFoldConstantImpl(Op, DL, TLI, FoldedOps);
1179 }
1180
1182}
1183
1187 return ConstantFoldConstantImpl(C, DL, TLI, FoldedOps);
1188}
1189
1194 bool AllowNonDeterministic) {
1195 return ConstantFoldInstOperandsImpl(I, I->getOpcode(), Ops, DL, TLI,
1196 AllowNonDeterministic);
1197}
1198
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1215 if (CE0->getOpcode() == Instruction::IntToPtr) {
1216 Type *IntPtrTy = DL.getIntPtrType(CE0->getType());
1217
1218
1220 false, DL)) {
1223 }
1224 }
1225
1226
1227
1228 if (CE0->getOpcode() == Instruction::PtrToInt) {
1229 Type *IntPtrTy = DL.getIntPtrType(CE0->getOperand(0)->getType());
1230 if (CE0->getType() == IntPtrTy) {
1231 Constant *C = CE0->getOperand(0);
1234 }
1235 }
1236 }
1237
1239 if (CE0->getOpcode() == CE1->getOpcode()) {
1240 if (CE0->getOpcode() == Instruction::IntToPtr) {
1241 Type *IntPtrTy = DL.getIntPtrType(CE0->getType());
1242
1243
1244
1246 false, DL);
1248 false, DL);
1249 if (C0 && C1)
1251 }
1252
1253
1254
1255 if (CE0->getOpcode() == Instruction::PtrToInt) {
1256 Type *IntPtrTy = DL.getIntPtrType(CE0->getOperand(0)->getType());
1257 if (CE0->getType() == IntPtrTy &&
1258 CE0->getOperand(0)->getType() == CE1->getOperand(0)->getType()) {
1260 Predicate, CE0->getOperand(0), CE1->getOperand(0), DL, TLI);
1261 }
1262 }
1263 }
1264 }
1265
1266
1267
1268
1269
1270
1272 unsigned IndexWidth = DL.getIndexTypeSizeInBits(Ops0->getType());
1273 APInt Offset0(IndexWidth, 0);
1276 DL, Offset0, IsEqPred,
1277 false, nullptr,
1278 IsEqPred);
1279 APInt Offset1(IndexWidth, 0);
1281 DL, Offset1, IsEqPred,
1282 false, nullptr,
1283 IsEqPred);
1284 if (Stripped0 == Stripped1)
1289 }
1291
1292
1295 }
1296
1298
1299
1301 if (!Ops0)
1302 return nullptr;
1304 if (!Ops1)
1305 return nullptr;
1306 }
1307
1309}
1310
1317
1323 if (Constant *C = SymbolicallyEvaluateBinop(Opcode, LHS, RHS, DL))
1324 return C;
1325
1329}
1330
1333 switch (Mode) {
1335 return nullptr;
1337 return ConstantFP::get(Ty->getContext(), APF);
1339 return ConstantFP::get(
1340 Ty->getContext(),
1343 return ConstantFP::get(Ty->getContext(),
1345 default:
1346 break;
1347 }
1348
1350}
1351
1352
1353
1359
1362 bool IsOutput) {
1365 return CFP;
1366
1369 IsOutput ? Mode.Output : Mode.Input);
1370}
1371
1373 bool IsOutput) {
1376
1378 return Operand;
1379
1382 if (VecTy) {
1385 if (!Folded)
1386 return nullptr;
1388 }
1389
1391 }
1392
1394 return Operand;
1395
1398 for (unsigned i = 0, e = CV->getNumOperands(); i != e; ++i) {
1402 continue;
1403 }
1404
1406 if (!CFP)
1407 return nullptr;
1408
1410 if (!Folded)
1411 return nullptr;
1413 }
1414
1416 }
1417
1420 for (unsigned I = 0, E = CDV->getNumElements(); I < E; ++I) {
1421 const APFloat &Elt = CDV->getElementAsAPFloat(I);
1423 NewElts.push_back(ConstantFP::get(Ty, Elt));
1424 } else {
1428 if (!Folded)
1429 return nullptr;
1431 }
1432 }
1433
1435 }
1436
1437 return nullptr;
1438}
1439
1443 bool AllowNonDeterministic) {
1445
1447 if (!Op0)
1448 return nullptr;
1450 if (!Op1)
1451 return nullptr;
1452
1453
1454
1455
1456 if (!AllowNonDeterministic)
1458 if (FP->hasNoSignedZeros() || FP->hasAllowReassoc() ||
1459 FP->hasAllowContract() || FP->hasAllowReciprocal())
1460 return nullptr;
1461
1462
1464 if ()
1465 return nullptr;
1466
1467
1469 if ()
1470 return nullptr;
1471
1472
1473 if (!AllowNonDeterministic && C->isNaN())
1474 return nullptr;
1475
1476 return C;
1477 }
1478
1479
1481}
1482
1486
1488 if (CE->isCast())
1492 C->getType(), DestTy, &DL))
1494
1495 switch (Opcode) {
1496 default:
1498 case Instruction::PtrToAddr:
1499 case Instruction::PtrToInt:
1501 Constant *FoldedValue = nullptr;
1502
1503
1504 if (CE->getOpcode() == Instruction::IntToPtr) {
1505
1506 Type *MidTy = Opcode == Instruction::PtrToInt
1507 ? DL.getAddressType(CE->getType())
1508 : DL.getIntPtrType(CE->getType());
1510 false, DL);
1512
1513
1514
1515 unsigned BitWidth = DL.getIndexTypeSizeInBits(GEP->getType());
1518 DL, BaseOffset, true));
1519 if (Base->isNullValue()) {
1520 FoldedValue = ConstantInt::get(CE->getContext(), BaseOffset);
1521 } else {
1522
1523
1524 if (GEP->getNumIndices() == 1 &&
1525 GEP->getSourceElementType()->isIntegerTy(8)) {
1528 Type *IntIdxTy = DL.getIndexType(Ptr->getType());
1529 if (Sub && Sub->getType() == IntIdxTy &&
1530 Sub->getOpcode() == Instruction::Sub &&
1531 Sub->getOperand(0)->isNullValue())
1534 Sub->getOperand(1));
1535 }
1536 }
1537 }
1538 if (FoldedValue) {
1539
1541 DL);
1542 }
1543 }
1544 break;
1545 case Instruction::IntToPtr:
1546
1547
1548
1549
1551 if (CE->getOpcode() == Instruction::PtrToInt) {
1552 Constant *SrcPtr = CE->getOperand(0);
1553 unsigned SrcPtrSize = DL.getPointerTypeSizeInBits(SrcPtr->getType());
1554 unsigned MidIntSize = CE->getType()->getScalarSizeInBits();
1555
1556 if (MidIntSize >= SrcPtrSize) {
1559 return FoldBitCast(CE->getOperand(0), DestTy, DL);
1560 }
1561 }
1562 }
1563 break;
1564 case Instruction::Trunc:
1565 case Instruction::ZExt:
1566 case Instruction::SExt:
1567 case Instruction::FPTrunc:
1568 case Instruction::FPExt:
1569 case Instruction::UIToFP:
1570 case Instruction::SIToFP:
1571 case Instruction::FPToUI:
1572 case Instruction::FPToSI:
1573 case Instruction::AddrSpaceCast:
1574 break;
1575 case Instruction::BitCast:
1577 }
1578
1582}
1583
1586 Type *SrcTy = C->getType();
1587 if (SrcTy == DestTy)
1588 return C;
1591 if (IsSigned)
1594}
1595
1596
1597
1598
1599
1601 if (Call->isNoBuiltin())
1602 return false;
1603 if (Call->getFunctionType() != F->getFunctionType())
1604 return false;
1605
1606
1607
1608
1609
1612 return Arg.getType()->isFloatingPointTy();
1613 })))
1614 return false;
1615
1616 switch (F->getIntrinsicID()) {
1617
1618
1619 case Intrinsic::bswap:
1620 case Intrinsic::ctpop:
1621 case Intrinsic::ctlz:
1622 case Intrinsic::cttz:
1623 case Intrinsic::fshl:
1624 case Intrinsic::fshr:
1625 case Intrinsic::launder_invariant_group:
1626 case Intrinsic::strip_invariant_group:
1627 case Intrinsic::masked_load:
1628 case Intrinsic::get_active_lane_mask:
1629 case Intrinsic::abs:
1630 case Intrinsic::smax:
1631 case Intrinsic::smin:
1632 case Intrinsic::umax:
1633 case Intrinsic::umin:
1634 case Intrinsic::scmp:
1635 case Intrinsic::ucmp:
1636 case Intrinsic::sadd_with_overflow:
1637 case Intrinsic::uadd_with_overflow:
1638 case Intrinsic::ssub_with_overflow:
1639 case Intrinsic::usub_with_overflow:
1640 case Intrinsic::smul_with_overflow:
1641 case Intrinsic::umul_with_overflow:
1642 case Intrinsic::sadd_sat:
1643 case Intrinsic::uadd_sat:
1644 case Intrinsic::ssub_sat:
1645 case Intrinsic::usub_sat:
1646 case Intrinsic::smul_fix:
1647 case Intrinsic::smul_fix_sat:
1648 case Intrinsic::bitreverse:
1649 case Intrinsic::is_constant:
1650 case Intrinsic::vector_reduce_add:
1651 case Intrinsic::vector_reduce_mul:
1652 case Intrinsic::vector_reduce_and:
1653 case Intrinsic::vector_reduce_or:
1654 case Intrinsic::vector_reduce_xor:
1655 case Intrinsic::vector_reduce_smin:
1656 case Intrinsic::vector_reduce_smax:
1657 case Intrinsic::vector_reduce_umin:
1658 case Intrinsic::vector_reduce_umax:
1659 case Intrinsic::vector_extract:
1660 case Intrinsic::vector_insert:
1661 case Intrinsic::vector_interleave2:
1662 case Intrinsic::vector_interleave3:
1663 case Intrinsic::vector_interleave4:
1664 case Intrinsic::vector_interleave5:
1665 case Intrinsic::vector_interleave6:
1666 case Intrinsic::vector_interleave7:
1667 case Intrinsic::vector_interleave8:
1668 case Intrinsic::vector_deinterleave2:
1669 case Intrinsic::vector_deinterleave3:
1670 case Intrinsic::vector_deinterleave4:
1671 case Intrinsic::vector_deinterleave5:
1672 case Intrinsic::vector_deinterleave6:
1673 case Intrinsic::vector_deinterleave7:
1674 case Intrinsic::vector_deinterleave8:
1675
1676 case Intrinsic::amdgcn_perm:
1677 case Intrinsic::amdgcn_wave_reduce_umin:
1678 case Intrinsic::amdgcn_wave_reduce_umax:
1679 case Intrinsic::amdgcn_wave_reduce_max:
1680 case Intrinsic::amdgcn_wave_reduce_min:
1681 case Intrinsic::amdgcn_wave_reduce_add:
1682 case Intrinsic::amdgcn_wave_reduce_sub:
1683 case Intrinsic::amdgcn_wave_reduce_and:
1684 case Intrinsic::amdgcn_wave_reduce_or:
1685 case Intrinsic::amdgcn_wave_reduce_xor:
1686 case Intrinsic::amdgcn_s_wqm:
1687 case Intrinsic::amdgcn_s_quadmask:
1688 case Intrinsic::amdgcn_s_bitreplicate:
1689 case Intrinsic::arm_mve_vctp8:
1690 case Intrinsic::arm_mve_vctp16:
1691 case Intrinsic::arm_mve_vctp32:
1692 case Intrinsic::arm_mve_vctp64:
1693 case Intrinsic::aarch64_sve_convert_from_svbool:
1694 case Intrinsic::wasm_alltrue:
1695 case Intrinsic::wasm_anytrue:
1696 case Intrinsic::wasm_dot:
1697
1698 case Intrinsic::wasm_trunc_signed:
1699 case Intrinsic::wasm_trunc_unsigned:
1700 return true;
1701
1702
1703
1704 case Intrinsic::minnum:
1705 case Intrinsic::maxnum:
1706 case Intrinsic::minimum:
1707 case Intrinsic::maximum:
1708 case Intrinsic::minimumnum:
1709 case Intrinsic::maximumnum:
1710 case Intrinsic:🪵
1711 case Intrinsic::log2:
1712 case Intrinsic::log10:
1713 case Intrinsic::exp:
1714 case Intrinsic::exp2:
1715 case Intrinsic::exp10:
1716 case Intrinsic::sqrt:
1717 case Intrinsic::sin:
1718 case Intrinsic::cos:
1719 case Intrinsic::sincos:
1720 case Intrinsic::sinh:
1721 case Intrinsic::cosh:
1722 case Intrinsic::atan:
1723 case Intrinsic::pow:
1724 case Intrinsic::powi:
1725 case Intrinsic::ldexp:
1726 case Intrinsic::fma:
1727 case Intrinsic::fmuladd:
1728 case Intrinsic::frexp:
1729 case Intrinsic::fptoui_sat:
1730 case Intrinsic::fptosi_sat:
1731 case Intrinsic::convert_from_fp16:
1732 case Intrinsic::convert_to_fp16:
1733 case Intrinsic::amdgcn_cos:
1734 case Intrinsic::amdgcn_cubeid:
1735 case Intrinsic::amdgcn_cubema:
1736 case Intrinsic::amdgcn_cubesc:
1737 case Intrinsic::amdgcn_cubetc:
1738 case Intrinsic::amdgcn_fmul_legacy:
1739 case Intrinsic::amdgcn_fma_legacy:
1740 case Intrinsic::amdgcn_fract:
1741 case Intrinsic::amdgcn_sin:
1742
1743 case Intrinsic::x86_sse_cvtss2si:
1744 case Intrinsic::x86_sse_cvtss2si64:
1745 case Intrinsic::x86_sse_cvttss2si:
1746 case Intrinsic::x86_sse_cvttss2si64:
1747 case Intrinsic::x86_sse2_cvtsd2si:
1748 case Intrinsic::x86_sse2_cvtsd2si64:
1749 case Intrinsic::x86_sse2_cvttsd2si:
1750 case Intrinsic::x86_sse2_cvttsd2si64:
1751 case Intrinsic::x86_avx512_vcvtss2si32:
1752 case Intrinsic::x86_avx512_vcvtss2si64:
1753 case Intrinsic::x86_avx512_cvttss2si:
1754 case Intrinsic::x86_avx512_cvttss2si64:
1755 case Intrinsic::x86_avx512_vcvtsd2si32:
1756 case Intrinsic::x86_avx512_vcvtsd2si64:
1757 case Intrinsic::x86_avx512_cvttsd2si:
1758 case Intrinsic::x86_avx512_cvttsd2si64:
1759 case Intrinsic::x86_avx512_vcvtss2usi32:
1760 case Intrinsic::x86_avx512_vcvtss2usi64:
1761 case Intrinsic::x86_avx512_cvttss2usi:
1762 case Intrinsic::x86_avx512_cvttss2usi64:
1763 case Intrinsic::x86_avx512_vcvtsd2usi32:
1764 case Intrinsic::x86_avx512_vcvtsd2usi64:
1765 case Intrinsic::x86_avx512_cvttsd2usi:
1766 case Intrinsic::x86_avx512_cvttsd2usi64:
1767
1768
1769 case Intrinsic::nvvm_fmax_d:
1770 case Intrinsic::nvvm_fmax_f:
1771 case Intrinsic::nvvm_fmax_ftz_f:
1772 case Intrinsic::nvvm_fmax_ftz_nan_f:
1773 case Intrinsic::nvvm_fmax_ftz_nan_xorsign_abs_f:
1774 case Intrinsic::nvvm_fmax_ftz_xorsign_abs_f:
1775 case Intrinsic::nvvm_fmax_nan_f:
1776 case Intrinsic::nvvm_fmax_nan_xorsign_abs_f:
1777 case Intrinsic::nvvm_fmax_xorsign_abs_f:
1778
1779
1780 case Intrinsic::nvvm_fmin_d:
1781 case Intrinsic::nvvm_fmin_f:
1782 case Intrinsic::nvvm_fmin_ftz_f:
1783 case Intrinsic::nvvm_fmin_ftz_nan_f:
1784 case Intrinsic::nvvm_fmin_ftz_nan_xorsign_abs_f:
1785 case Intrinsic::nvvm_fmin_ftz_xorsign_abs_f:
1786 case Intrinsic::nvvm_fmin_nan_f:
1787 case Intrinsic::nvvm_fmin_nan_xorsign_abs_f:
1788 case Intrinsic::nvvm_fmin_xorsign_abs_f:
1789
1790
1791 case Intrinsic::nvvm_f2i_rm:
1792 case Intrinsic::nvvm_f2i_rn:
1793 case Intrinsic::nvvm_f2i_rp:
1794 case Intrinsic::nvvm_f2i_rz:
1795 case Intrinsic::nvvm_f2i_rm_ftz:
1796 case Intrinsic::nvvm_f2i_rn_ftz:
1797 case Intrinsic::nvvm_f2i_rp_ftz:
1798 case Intrinsic::nvvm_f2i_rz_ftz:
1799 case Intrinsic::nvvm_f2ui_rm:
1800 case Intrinsic::nvvm_f2ui_rn:
1801 case Intrinsic::nvvm_f2ui_rp:
1802 case Intrinsic::nvvm_f2ui_rz:
1803 case Intrinsic::nvvm_f2ui_rm_ftz:
1804 case Intrinsic::nvvm_f2ui_rn_ftz:
1805 case Intrinsic::nvvm_f2ui_rp_ftz:
1806 case Intrinsic::nvvm_f2ui_rz_ftz:
1807 case Intrinsic::nvvm_d2i_rm:
1808 case Intrinsic::nvvm_d2i_rn:
1809 case Intrinsic::nvvm_d2i_rp:
1810 case Intrinsic::nvvm_d2i_rz:
1811 case Intrinsic::nvvm_d2ui_rm:
1812 case Intrinsic::nvvm_d2ui_rn:
1813 case Intrinsic::nvvm_d2ui_rp:
1814 case Intrinsic::nvvm_d2ui_rz:
1815
1816
1817 case Intrinsic::nvvm_f2ll_rm:
1818 case Intrinsic::nvvm_f2ll_rn:
1819 case Intrinsic::nvvm_f2ll_rp:
1820 case Intrinsic::nvvm_f2ll_rz:
1821 case Intrinsic::nvvm_f2ll_rm_ftz:
1822 case Intrinsic::nvvm_f2ll_rn_ftz:
1823 case Intrinsic::nvvm_f2ll_rp_ftz:
1824 case Intrinsic::nvvm_f2ll_rz_ftz:
1825 case Intrinsic::nvvm_f2ull_rm:
1826 case Intrinsic::nvvm_f2ull_rn:
1827 case Intrinsic::nvvm_f2ull_rp:
1828 case Intrinsic::nvvm_f2ull_rz:
1829 case Intrinsic::nvvm_f2ull_rm_ftz:
1830 case Intrinsic::nvvm_f2ull_rn_ftz:
1831 case Intrinsic::nvvm_f2ull_rp_ftz:
1832 case Intrinsic::nvvm_f2ull_rz_ftz:
1833 case Intrinsic::nvvm_d2ll_rm:
1834 case Intrinsic::nvvm_d2ll_rn:
1835 case Intrinsic::nvvm_d2ll_rp:
1836 case Intrinsic::nvvm_d2ll_rz:
1837 case Intrinsic::nvvm_d2ull_rm:
1838 case Intrinsic::nvvm_d2ull_rn:
1839 case Intrinsic::nvvm_d2ull_rp:
1840 case Intrinsic::nvvm_d2ull_rz:
1841
1842
1843 case Intrinsic::nvvm_ceil_d:
1844 case Intrinsic::nvvm_ceil_f:
1845 case Intrinsic::nvvm_ceil_ftz_f:
1846
1847 case Intrinsic::nvvm_fabs:
1848 case Intrinsic::nvvm_fabs_ftz:
1849
1850 case Intrinsic::nvvm_floor_d:
1851 case Intrinsic::nvvm_floor_f:
1852 case Intrinsic::nvvm_floor_ftz_f:
1853
1854 case Intrinsic::nvvm_rcp_rm_d:
1855 case Intrinsic::nvvm_rcp_rm_f:
1856 case Intrinsic::nvvm_rcp_rm_ftz_f:
1857 case Intrinsic::nvvm_rcp_rn_d:
1858 case Intrinsic::nvvm_rcp_rn_f:
1859 case Intrinsic::nvvm_rcp_rn_ftz_f:
1860 case Intrinsic::nvvm_rcp_rp_d:
1861 case Intrinsic::nvvm_rcp_rp_f:
1862 case Intrinsic::nvvm_rcp_rp_ftz_f:
1863 case Intrinsic::nvvm_rcp_rz_d:
1864 case Intrinsic::nvvm_rcp_rz_f:
1865 case Intrinsic::nvvm_rcp_rz_ftz_f:
1866
1867 case Intrinsic::nvvm_round_d:
1868 case Intrinsic::nvvm_round_f:
1869 case Intrinsic::nvvm_round_ftz_f:
1870
1871 case Intrinsic::nvvm_saturate_d:
1872 case Intrinsic::nvvm_saturate_f:
1873 case Intrinsic::nvvm_saturate_ftz_f:
1874
1875 case Intrinsic::nvvm_sqrt_f:
1876 case Intrinsic::nvvm_sqrt_rn_d:
1877 case Intrinsic::nvvm_sqrt_rn_f:
1878 case Intrinsic::nvvm_sqrt_rn_ftz_f:
1879 return ->isStrictFP();
1880
1881
1882 case Intrinsic::nvvm_add_rm_d:
1883 case Intrinsic::nvvm_add_rn_d:
1884 case Intrinsic::nvvm_add_rp_d:
1885 case Intrinsic::nvvm_add_rz_d:
1886 case Intrinsic::nvvm_add_rm_f:
1887 case Intrinsic::nvvm_add_rn_f:
1888 case Intrinsic::nvvm_add_rp_f:
1889 case Intrinsic::nvvm_add_rz_f:
1890 case Intrinsic::nvvm_add_rm_ftz_f:
1891 case Intrinsic::nvvm_add_rn_ftz_f:
1892 case Intrinsic::nvvm_add_rp_ftz_f:
1893 case Intrinsic::nvvm_add_rz_ftz_f:
1894
1895
1896 case Intrinsic::nvvm_div_rm_d:
1897 case Intrinsic::nvvm_div_rn_d:
1898 case Intrinsic::nvvm_div_rp_d:
1899 case Intrinsic::nvvm_div_rz_d:
1900 case Intrinsic::nvvm_div_rm_f:
1901 case Intrinsic::nvvm_div_rn_f:
1902 case Intrinsic::nvvm_div_rp_f:
1903 case Intrinsic::nvvm_div_rz_f:
1904 case Intrinsic::nvvm_div_rm_ftz_f:
1905 case Intrinsic::nvvm_div_rn_ftz_f:
1906 case Intrinsic::nvvm_div_rp_ftz_f:
1907 case Intrinsic::nvvm_div_rz_ftz_f:
1908
1909
1910 case Intrinsic::nvvm_mul_rm_d:
1911 case Intrinsic::nvvm_mul_rn_d:
1912 case Intrinsic::nvvm_mul_rp_d:
1913 case Intrinsic::nvvm_mul_rz_d:
1914 case Intrinsic::nvvm_mul_rm_f:
1915 case Intrinsic::nvvm_mul_rn_f:
1916 case Intrinsic::nvvm_mul_rp_f:
1917 case Intrinsic::nvvm_mul_rz_f:
1918 case Intrinsic::nvvm_mul_rm_ftz_f:
1919 case Intrinsic::nvvm_mul_rn_ftz_f:
1920 case Intrinsic::nvvm_mul_rp_ftz_f:
1921 case Intrinsic::nvvm_mul_rz_ftz_f:
1922
1923
1924 case Intrinsic::nvvm_fma_rm_d:
1925 case Intrinsic::nvvm_fma_rn_d:
1926 case Intrinsic::nvvm_fma_rp_d:
1927 case Intrinsic::nvvm_fma_rz_d:
1928 case Intrinsic::nvvm_fma_rm_f:
1929 case Intrinsic::nvvm_fma_rn_f:
1930 case Intrinsic::nvvm_fma_rp_f:
1931 case Intrinsic::nvvm_fma_rz_f:
1932 case Intrinsic::nvvm_fma_rm_ftz_f:
1933 case Intrinsic::nvvm_fma_rn_ftz_f:
1934 case Intrinsic::nvvm_fma_rp_ftz_f:
1935 case Intrinsic::nvvm_fma_rz_ftz_f:
1936
1937
1938
1939 case Intrinsic::fabs:
1940 case Intrinsic::copysign:
1941 case Intrinsic::is_fpclass:
1942
1943
1944 case Intrinsic::ceil:
1945 case Intrinsic:🤣
1946 case Intrinsic::round:
1947 case Intrinsic::roundeven:
1948 case Intrinsic::trunc:
1949 case Intrinsic::nearbyint:
1950 case Intrinsic::rint:
1951 case Intrinsic::canonicalize:
1952
1953
1954
1955 case Intrinsic::experimental_constrained_fma:
1956 case Intrinsic::experimental_constrained_fmuladd:
1957 case Intrinsic::experimental_constrained_fadd:
1958 case Intrinsic::experimental_constrained_fsub:
1959 case Intrinsic::experimental_constrained_fmul:
1960 case Intrinsic::experimental_constrained_fdiv:
1961 case Intrinsic::experimental_constrained_frem:
1962 case Intrinsic::experimental_constrained_ceil:
1963 case Intrinsic::experimental_constrained_floor:
1964 case Intrinsic::experimental_constrained_round:
1965 case Intrinsic::experimental_constrained_roundeven:
1966 case Intrinsic::experimental_constrained_trunc:
1967 case Intrinsic::experimental_constrained_nearbyint:
1968 case Intrinsic::experimental_constrained_rint:
1969 case Intrinsic::experimental_constrained_fcmp:
1970 case Intrinsic::experimental_constrained_fcmps:
1971 return true;
1972 default:
1973 return false;
1975 }
1976
1977 if (->hasName() || Call->isStrictFP())
1978 return false;
1979
1980
1981
1982
1984 switch (Name[0]) {
1985 default:
1986 return false;
1987
1988 case 'a':
1989 return Name == "acos" || Name == "acosf" ||
1990 Name == "asin" || Name == "asinf" ||
1991 Name == "atan" || Name == "atanf" ||
1992 Name == "atan2" || Name == "atan2f";
1993 case 'c':
1994 return Name == "ceil" || Name == "ceilf" ||
1995 Name == "cos" || Name == "cosf" ||
1996 Name == "cosh" || Name == "coshf";
1997 case 'e':
1998 return Name == "exp" || Name == "expf" || Name == "exp2" ||
1999 Name == "exp2f" || Name == "erf" || Name == "erff";
2000 case 'f':
2001 return Name == "fabs" || Name == "fabsf" ||
2002 Name == "floor" || Name == "floorf" ||
2003 Name == "fmod" || Name == "fmodf";
2004 case 'i':
2005 return Name == "ilogb" || Name == "ilogbf";
2006 case 'l':
2007 return Name == "log" || Name == "logf" || Name == "logl" ||
2008 Name == "log2" || Name == "log2f" || Name == "log10" ||
2009 Name == "log10f" || Name == "logb" || Name == "logbf" ||
2010 Name == "log1p" || Name == "log1pf";
2011 case 'n':
2012 return Name == "nearbyint" || Name == "nearbyintf";
2013 case 'p':
2014 return Name == "pow" || Name == "powf";
2015 case 'r':
2016 return Name == "remainder" || Name == "remainderf" ||
2017 Name == "rint" || Name == "rintf" ||
2018 Name == "round" || Name == "roundf" ||
2019 Name == "roundeven" || Name == "roundevenf";
2020 case 's':
2021 return Name == "sin" || Name == "sinf" ||
2022 Name == "sinh" || Name == "sinhf" ||
2023 Name == "sqrt" || Name == "sqrtf";
2024 case 't':
2025 return Name == "tan" || Name == "tanf" ||
2026 Name == "tanh" || Name == "tanhf" ||
2027 Name == "trunc" || Name == "truncf";
2028 case '_':
2029
2030
2031
2032
2033
2034
2035 if (Name.size() < 12 || Name[1] != '_')
2036 return false;
2037 switch (Name[2]) {
2038 default:
2039 return false;
2040 case 'a':
2041 return Name == "__acos_finite" || Name == "__acosf_finite" ||
2042 Name == "__asin_finite" || Name == "__asinf_finite" ||
2043 Name == "__atan2_finite" || Name == "__atan2f_finite";
2044 case 'c':
2045 return Name == "__cosh_finite" || Name == "__coshf_finite";
2046 case 'e':
2047 return Name == "__exp_finite" || Name == "__expf_finite" ||
2048 Name == "__exp2_finite" || Name == "__exp2f_finite";
2049 case 'l':
2050 return Name == "__log_finite" || Name == "__logf_finite" ||
2051 Name == "__log10_finite" || Name == "__log10f_finite";
2052 case 'p':
2053 return Name == "__pow_finite" || Name == "__powf_finite";
2054 case 's':
2055 return Name == "__sinh_finite" || Name == "__sinhf_finite";
2056 }
2057
2058 }
2059}
2060
2061namespace {
2062
2063Constant *GetConstantFoldFPValue(double V, Type *Ty) {
2064 if (Ty->isHalfTy() || Ty->isFloatTy()) {
2066 bool unused;
2068 return ConstantFP::get(Ty->getContext(), APF);
2069 }
2070 if (Ty->isDoubleTy())
2071 return ConstantFP::get(Ty->getContext(), APFloat(V));
2072 llvm_unreachable("Can only constant fold half/float/double");
2073}
2074
2075#if defined(HAS_IEE754_FLOAT128) && defined(HAS_LOGF128)
2076Constant *GetConstantFoldFPValue128(float128 V, Type *Ty) {
2077 if (Ty->isFP128Ty())
2078 return ConstantFP::get(Ty, V);
2080}
2081#endif
2082
2083
2084inline void llvm_fenv_clearexcept() {
2085#if HAVE_DECL_FE_ALL_EXCEPT
2086 feclearexcept(FE_ALL_EXCEPT);
2087#endif
2088 errno = 0;
2089}
2090
2091
2092inline bool llvm_fenv_testexcept() {
2093 int errno_val = errno;
2094 if (errno_val == ERANGE || errno_val == EDOM)
2095 return true;
2096#if HAVE_DECL_FE_ALL_EXCEPT && HAVE_DECL_FE_INEXACT
2097 if (fetestexcept(FE_ALL_EXCEPT & ~FE_INEXACT))
2098 return true;
2099#endif
2100 return false;
2101}
2102
2104 if (V.isDenormal())
2106 return V;
2107}
2108
2109static APFloat FlushToPositiveZero(const APFloat &V) {
2110 if (V.isDenormal())
2112 return V;
2113}
2114
2119 switch (DenormKind) {
2121 return V;
2123 return FTZPreserveSign(V);
2125 return FlushToPositiveZero(V);
2126 default:
2128 }
2129}
2130
2131Constant *ConstantFoldFP(double (*NativeFP)(double), const APFloat &V, Type *Ty,
2133 if (!DenormMode.isValid() ||
2136 return nullptr;
2137
2138 llvm_fenv_clearexcept();
2139 auto Input = FlushWithDenormKind(V, DenormMode.Input);
2140 double Result = NativeFP(Input.convertToDouble());
2141 if (llvm_fenv_testexcept()) {
2142 llvm_fenv_clearexcept();
2143 return nullptr;
2144 }
2145
2146 Constant *Output = GetConstantFoldFPValue(Result, Ty);
2148 return Output;
2149 const auto *CFP = static_cast<ConstantFP *>(Output);
2150 const auto Res = FlushWithDenormKind(CFP->getValueAPF(), DenormMode.Output);
2151 return ConstantFP::get(Ty->getContext(), Res);
2152}
2153
2154#if defined(HAS_IEE754_FLOAT128) && defined(HAS_LOGF128)
2155Constant *ConstantFoldFP128(float128 (*NativeFP)(float128), const APFloat &V,
2157 llvm_fenv_clearexcept();
2158 float128 Result = NativeFP(V.convertToQuad());
2159 if (llvm_fenv_testexcept()) {
2160 llvm_fenv_clearexcept();
2161 return nullptr;
2162 }
2163
2164 return GetConstantFoldFPValue128(Result, Ty);
2165}
2166#endif
2167
2168Constant *ConstantFoldBinaryFP(double (*NativeFP)(double, double),
2170 llvm_fenv_clearexcept();
2171 double Result = NativeFP(V.convertToDouble(), W.convertToDouble());
2172 if (llvm_fenv_testexcept()) {
2173 llvm_fenv_clearexcept();
2174 return nullptr;
2175 }
2176
2177 return GetConstantFoldFPValue(Result, Ty);
2178}
2179
2182
2183
2184 if (Op->containsPoisonElement())
2186
2187
2188 if (Constant *SplatVal = Op->getSplatValue()) {
2189 switch (IID) {
2190 case Intrinsic::vector_reduce_and:
2191 case Intrinsic::vector_reduce_or:
2192 case Intrinsic::vector_reduce_smin:
2193 case Intrinsic::vector_reduce_smax:
2194 case Intrinsic::vector_reduce_umin:
2195 case Intrinsic::vector_reduce_umax:
2196 return SplatVal;
2197 case Intrinsic::vector_reduce_add:
2198 if (SplatVal->isNullValue())
2199 return SplatVal;
2200 break;
2201 case Intrinsic::vector_reduce_mul:
2202 if (SplatVal->isNullValue() || SplatVal->isOneValue())
2203 return SplatVal;
2204 break;
2205 case Intrinsic::vector_reduce_xor:
2206 if (SplatVal->isNullValue())
2207 return SplatVal;
2208 if (OpVT->getElementCount().isKnownMultipleOf(2))
2210 break;
2211 }
2212 }
2213
2215 if (!VT)
2216 return nullptr;
2217
2218
2220 if (!EltC)
2221 return nullptr;
2222
2223 APInt Acc = EltC->getValue();
2226 return nullptr;
2227 const APInt &X = EltC->getValue();
2228 switch (IID) {
2229 case Intrinsic::vector_reduce_add:
2230 Acc = Acc + X;
2231 break;
2232 case Intrinsic::vector_reduce_mul:
2233 Acc = Acc * X;
2234 break;
2235 case Intrinsic::vector_reduce_and:
2236 Acc = Acc & X;
2237 break;
2238 case Intrinsic::vector_reduce_or:
2239 Acc = Acc | X;
2240 break;
2241 case Intrinsic::vector_reduce_xor:
2242 Acc = Acc ^ X;
2243 break;
2244 case Intrinsic::vector_reduce_smin:
2246 break;
2247 case Intrinsic::vector_reduce_smax:
2249 break;
2250 case Intrinsic::vector_reduce_umin:
2252 break;
2253 case Intrinsic::vector_reduce_umax:
2255 break;
2256 }
2257 }
2258
2259 return ConstantInt::get(Op->getContext(), Acc);
2260}
2261
2262
2263
2264
2265
2266
2267
2268
2269Constant *ConstantFoldSSEConvertToInt(const APFloat &Val, bool roundTowardZero,
2270 Type *Ty, bool IsSigned) {
2271
2272 unsigned ResultWidth = Ty->getIntegerBitWidth();
2273 assert(ResultWidth <= 64 &&
2274 "Can only constant fold conversions to 64 and 32 bit ints");
2275
2277 bool isExact = false;
2282 IsSigned, mode, &isExact);
2285 return nullptr;
2286 return ConstantInt::get(Ty, UIntVal, IsSigned);
2287}
2288
2290 Type *Ty = Op->getType();
2291
2292 if (Ty->isBFloatTy() || Ty->isHalfTy() || Ty->isFloatTy() || Ty->isDoubleTy())
2293 return Op->getValueAPF().convertToDouble();
2294
2295 bool unused;
2296 APFloat APF = Op->getValueAPF();
2299}
2300
2301static bool getConstIntOrUndef(Value *Op, const APInt *&C) {
2303 C = &CI->getValue();
2304 return true;
2305 }
2307 C = nullptr;
2308 return true;
2309 }
2310 return false;
2311}
2312
2313
2314
2315
2316
2317
2320 std::optional ORM = CI->getRoundingMode();
2322
2323
2324
2326 return true;
2327
2328
2329
2331 return false;
2332
2333
2334
2336 return true;
2337
2338
2339
2340 return false;
2341}
2342
2343
2346 std::optional ORM = CI->getRoundingMode();
2348
2349
2350
2351
2353 return *ORM;
2354}
2355
2356
2357static Constant *constantFoldCanonicalize(const Type *Ty, const CallBase *CI,
2359
2360 if (Src.isZero()) {
2361
2362 return ConstantFP::get(
2365 }
2366
2367 if (!Ty->isIEEELikeFPTy())
2368 return nullptr;
2369
2370
2371
2372
2373
2374 if (Src.isNormal() || Src.isInfinity())
2375 return ConstantFP::get(CI->getContext(), Src);
2376
2380
2382 return ConstantFP::get(CI->getContext(), Src);
2383
2385 return nullptr;
2386
2387
2392 return nullptr;
2393
2394 bool IsPositive =
2398
2399 return ConstantFP::get(CI->getContext(),
2401 }
2402
2403 return nullptr;
2404}
2405
2412 assert(Operands.size() == 1 && "Wrong number of operands.");
2413
2414 if (IntrinsicID == Intrinsic::is_constant) {
2415
2416
2417
2418 if (Operands[0]->isManifestConstant())
2420 return nullptr;
2421 }
2422
2424
2425
2426
2427 if (IntrinsicID == Intrinsic::cos ||
2428 IntrinsicID == Intrinsic::ctpop ||
2429 IntrinsicID == Intrinsic::fptoui_sat ||
2430 IntrinsicID == Intrinsic::fptosi_sat ||
2431 IntrinsicID == Intrinsic::canonicalize)
2433 if (IntrinsicID == Intrinsic::bswap ||
2434 IntrinsicID == Intrinsic::bitreverse ||
2435 IntrinsicID == Intrinsic::launder_invariant_group ||
2436 IntrinsicID == Intrinsic::strip_invariant_group)
2437 return Operands[0];
2438 }
2439
2441
2442 if (IntrinsicID == Intrinsic::launder_invariant_group ||
2443 IntrinsicID == Intrinsic::strip_invariant_group) {
2444
2445
2446
2448 Call->getParent() ? Call->getCaller() : nullptr;
2449 if (Caller &&
2452 return Operands[0];
2453 }
2454 return nullptr;
2455 }
2456 }
2457
2459 if (IntrinsicID == Intrinsic::convert_to_fp16) {
2461
2462 bool lost = false;
2464
2465 return ConstantInt::get(Ty->getContext(), Val.bitcastToAPInt());
2466 }
2467
2469
2470 if (IntrinsicID == Intrinsic::wasm_trunc_signed ||
2471 IntrinsicID == Intrinsic::wasm_trunc_unsigned) {
2472 bool Signed = IntrinsicID == Intrinsic::wasm_trunc_signed;
2473
2474 if (U.isNaN())
2475 return nullptr;
2476
2477 unsigned Width = Ty->getIntegerBitWidth();
2479 bool IsExact = false;
2482
2484 return ConstantInt::get(Ty, Int);
2485
2486 return nullptr;
2487 }
2488
2489 if (IntrinsicID == Intrinsic::fptoui_sat ||
2490 IntrinsicID == Intrinsic::fptosi_sat) {
2491
2492 APSInt Int(Ty->getIntegerBitWidth(),
2493 IntrinsicID == Intrinsic::fptoui_sat);
2494 bool IsExact;
2496 return ConstantInt::get(Ty, Int);
2497 }
2498
2499 if (IntrinsicID == Intrinsic::canonicalize)
2500 return constantFoldCanonicalize(Ty, Call, U);
2501
2502#if defined(HAS_IEE754_FLOAT128) && defined(HAS_LOGF128)
2503 if (Ty->isFP128Ty()) {
2504 if (IntrinsicID == Intrinsic::log) {
2505 float128 Result = logf128(Op->getValueAPF().convertToQuad());
2506 return GetConstantFoldFPValue128(Result, Ty);
2507 }
2508
2509 LibFunc Fp128Func = NotLibFunc;
2510 if (TLI && TLI->getLibFunc(Name, Fp128Func) && TLI->has(Fp128Func) &&
2511 Fp128Func == LibFunc_logl)
2512 return ConstantFoldFP128(logf128, Op->getValueAPF(), Ty);
2513 }
2514#endif
2515
2516 if (!Ty->isHalfTy() && !Ty->isFloatTy() && !Ty->isDoubleTy() &&
2517 !Ty->isIntegerTy())
2518 return nullptr;
2519
2520
2521
2522 if (IntrinsicID == Intrinsic::nearbyint || IntrinsicID == Intrinsic::rint ||
2523 IntrinsicID == Intrinsic::roundeven) {
2525 return ConstantFP::get(Ty->getContext(), U);
2526 }
2527
2528 if (IntrinsicID == Intrinsic::round) {
2530 return ConstantFP::get(Ty->getContext(), U);
2531 }
2532
2533 if (IntrinsicID == Intrinsic::roundeven) {
2535 return ConstantFP::get(Ty->getContext(), U);
2536 }
2537
2538 if (IntrinsicID == Intrinsic::ceil) {
2540 return ConstantFP::get(Ty->getContext(), U);
2541 }
2542
2543 if (IntrinsicID == Intrinsic::floor) {
2545 return ConstantFP::get(Ty->getContext(), U);
2546 }
2547
2548 if (IntrinsicID == Intrinsic::trunc) {
2550 return ConstantFP::get(Ty->getContext(), U);
2551 }
2552
2553 if (IntrinsicID == Intrinsic::fabs) {
2554 U.clearSign();
2555 return ConstantFP::get(Ty->getContext(), U);
2556 }
2557
2558 if (IntrinsicID == Intrinsic::amdgcn_fract) {
2559
2560
2561
2562
2565 APFloat FractU(U - FloorU);
2566 APFloat AlmostOne(U.getSemantics(), 1);
2567 AlmostOne.next( true);
2568 return ConstantFP::get(Ty->getContext(), minimum(FractU, AlmostOne));
2569 }
2570
2571
2572
2573
2574 std::optionalAPFloat::roundingMode RM;
2575 switch (IntrinsicID) {
2576 default:
2577 break;
2578 case Intrinsic::experimental_constrained_nearbyint:
2579 case Intrinsic::experimental_constrained_rint: {
2581 RM = CI->getRoundingMode();
2583 return nullptr;
2584 break;
2585 }
2586 case Intrinsic::experimental_constrained_round:
2588 break;
2589 case Intrinsic::experimental_constrained_ceil:
2591 break;
2592 case Intrinsic::experimental_constrained_floor:
2594 break;
2595 case Intrinsic::experimental_constrained_trunc:
2597 break;
2598 }
2599 if (RM) {
2601 if (U.isFinite()) {
2603 if (IntrinsicID == Intrinsic::experimental_constrained_rint &&
2605 std::optionalfp::ExceptionBehavior EB = CI->getExceptionBehavior();
2607 return nullptr;
2608 }
2609 } else if (U.isSignaling()) {
2610 std::optionalfp::ExceptionBehavior EB = CI->getExceptionBehavior();
2612 return nullptr;
2614 }
2615 return ConstantFP::get(Ty->getContext(), U);
2616 }
2617
2618
2619 switch (IntrinsicID) {
2620
2621 case Intrinsic::nvvm_f2i_rm:
2622 case Intrinsic::nvvm_f2i_rn:
2623 case Intrinsic::nvvm_f2i_rp:
2624 case Intrinsic::nvvm_f2i_rz:
2625 case Intrinsic::nvvm_f2i_rm_ftz:
2626 case Intrinsic::nvvm_f2i_rn_ftz:
2627 case Intrinsic::nvvm_f2i_rp_ftz:
2628 case Intrinsic::nvvm_f2i_rz_ftz:
2629
2630 case Intrinsic::nvvm_f2ui_rm:
2631 case Intrinsic::nvvm_f2ui_rn:
2632 case Intrinsic::nvvm_f2ui_rp:
2633 case Intrinsic::nvvm_f2ui_rz:
2634 case Intrinsic::nvvm_f2ui_rm_ftz:
2635 case Intrinsic::nvvm_f2ui_rn_ftz:
2636 case Intrinsic::nvvm_f2ui_rp_ftz:
2637 case Intrinsic::nvvm_f2ui_rz_ftz:
2638
2639 case Intrinsic::nvvm_d2i_rm:
2640 case Intrinsic::nvvm_d2i_rn:
2641 case Intrinsic::nvvm_d2i_rp:
2642 case Intrinsic::nvvm_d2i_rz:
2643
2644 case Intrinsic::nvvm_d2ui_rm:
2645 case Intrinsic::nvvm_d2ui_rn:
2646 case Intrinsic::nvvm_d2ui_rp:
2647 case Intrinsic::nvvm_d2ui_rz:
2648
2649 case Intrinsic::nvvm_f2ll_rm:
2650 case Intrinsic::nvvm_f2ll_rn:
2651 case Intrinsic::nvvm_f2ll_rp:
2652 case Intrinsic::nvvm_f2ll_rz:
2653 case Intrinsic::nvvm_f2ll_rm_ftz:
2654 case Intrinsic::nvvm_f2ll_rn_ftz:
2655 case Intrinsic::nvvm_f2ll_rp_ftz:
2656 case Intrinsic::nvvm_f2ll_rz_ftz:
2657
2658 case Intrinsic::nvvm_f2ull_rm:
2659 case Intrinsic::nvvm_f2ull_rn:
2660 case Intrinsic::nvvm_f2ull_rp:
2661 case Intrinsic::nvvm_f2ull_rz:
2662 case Intrinsic::nvvm_f2ull_rm_ftz:
2663 case Intrinsic::nvvm_f2ull_rn_ftz:
2664 case Intrinsic::nvvm_f2ull_rp_ftz:
2665 case Intrinsic::nvvm_f2ull_rz_ftz:
2666
2667 case Intrinsic::nvvm_d2ll_rm:
2668 case Intrinsic::nvvm_d2ll_rn:
2669 case Intrinsic::nvvm_d2ll_rp:
2670 case Intrinsic::nvvm_d2ll_rz:
2671
2672 case Intrinsic::nvvm_d2ull_rm:
2673 case Intrinsic::nvvm_d2ull_rn:
2674 case Intrinsic::nvvm_d2ull_rp:
2675 case Intrinsic::nvvm_d2ull_rz: {
2676
2677 if (U.isNaN()) {
2678
2679
2681 return ConstantInt::get(Ty, 0);
2682
2683
2684 unsigned BitWidth = Ty->getIntegerBitWidth();
2686 return ConstantInt::get(Ty, APInt(BitWidth, Val, false));
2687 }
2688
2693
2694 APSInt ResInt(Ty->getIntegerBitWidth(), !IsSigned);
2695 auto FloatToRound = IsFTZ ? FTZPreserveSign(U) : U;
2696
2697
2698
2699 bool IsExact = false;
2700 FloatToRound.convertToInteger(ResInt, RMode, &IsExact);
2701 return ConstantInt::get(Ty, ResInt);
2702 }
2703 }
2704
2705
2706
2707
2708 if (!U.isFinite())
2709 return nullptr;
2710
2711
2712
2713
2714
2715 const APFloat &APF = Op->getValueAPF();
2716
2717 switch (IntrinsicID) {
2718 default: break;
2719 case Intrinsic:🪵
2720 return ConstantFoldFP(log, APF, Ty);
2721 case Intrinsic::log2:
2722
2723 return ConstantFoldFP(log2, APF, Ty);
2724 case Intrinsic::log10:
2725
2726 return ConstantFoldFP(log10, APF, Ty);
2727 case Intrinsic::exp:
2728 return ConstantFoldFP(exp, APF, Ty);
2729 case Intrinsic::exp2:
2730
2731 return ConstantFoldBinaryFP(pow, APFloat(2.0), APF, Ty);
2732 case Intrinsic::exp10:
2733
2734 return ConstantFoldBinaryFP(pow, APFloat(10.0), APF, Ty);
2735 case Intrinsic::sin:
2736 return ConstantFoldFP(sin, APF, Ty);
2737 case Intrinsic::cos:
2738 return ConstantFoldFP(cos, APF, Ty);
2739 case Intrinsic::sinh:
2740 return ConstantFoldFP(sinh, APF, Ty);
2741 case Intrinsic::cosh:
2742 return ConstantFoldFP(cosh, APF, Ty);
2743 case Intrinsic::atan:
2744
2745 if (U.isZero())
2746 return ConstantFP::get(Ty->getContext(), U);
2747 return ConstantFoldFP(atan, APF, Ty);
2748 case Intrinsic::sqrt:
2749 return ConstantFoldFP(sqrt, APF, Ty);
2750
2751
2752 case Intrinsic::nvvm_ceil_ftz_f:
2753 case Intrinsic::nvvm_ceil_f:
2754 case Intrinsic::nvvm_ceil_d:
2755 return ConstantFoldFP(
2756 ceil, APF, Ty,
2759
2760 case Intrinsic::nvvm_fabs_ftz:
2761 case Intrinsic::nvvm_fabs:
2762 return ConstantFoldFP(
2763 fabs, APF, Ty,
2766
2767 case Intrinsic::nvvm_floor_ftz_f:
2768 case Intrinsic::nvvm_floor_f:
2769 case Intrinsic::nvvm_floor_d:
2770 return ConstantFoldFP(
2771 floor, APF, Ty,
2774
2775 case Intrinsic::nvvm_rcp_rm_ftz_f:
2776 case Intrinsic::nvvm_rcp_rn_ftz_f:
2777 case Intrinsic::nvvm_rcp_rp_ftz_f:
2778 case Intrinsic::nvvm_rcp_rz_ftz_f:
2779 case Intrinsic::nvvm_rcp_rm_d:
2780 case Intrinsic::nvvm_rcp_rm_f:
2781 case Intrinsic::nvvm_rcp_rn_d:
2782 case Intrinsic::nvvm_rcp_rn_f:
2783 case Intrinsic::nvvm_rcp_rp_d:
2784 case Intrinsic::nvvm_rcp_rp_f:
2785 case Intrinsic::nvvm_rcp_rz_d:
2786 case Intrinsic::nvvm_rcp_rz_f: {
2789
2790 auto Denominator = IsFTZ ? FTZPreserveSign(APF) : APF;
2793
2795 if (IsFTZ)
2796 Res = FTZPreserveSign(Res);
2797 return ConstantFP::get(Ty->getContext(), Res);
2798 }
2799 return nullptr;
2800 }
2801
2802 case Intrinsic::nvvm_round_ftz_f:
2803 case Intrinsic::nvvm_round_f:
2804 case Intrinsic::nvvm_round_d: {
2805
2806
2807
2809 auto V = IsFTZ ? FTZPreserveSign(APF) : APF;
2811 return ConstantFP::get(Ty->getContext(), V);
2812 }
2813
2814 case Intrinsic::nvvm_saturate_ftz_f:
2815 case Intrinsic::nvvm_saturate_d:
2816 case Intrinsic::nvvm_saturate_f: {
2818 auto V = IsFTZ ? FTZPreserveSign(APF) : APF;
2819 if (V.isNegative() || V.isZero() || V.isNaN())
2822 if (V > One)
2823 return ConstantFP::get(Ty->getContext(), One);
2824 return ConstantFP::get(Ty->getContext(), APF);
2825 }
2826
2827 case Intrinsic::nvvm_sqrt_rn_ftz_f:
2828 case Intrinsic::nvvm_sqrt_f:
2829 case Intrinsic::nvvm_sqrt_rn_d:
2830 case Intrinsic::nvvm_sqrt_rn_f:
2832 return nullptr;
2833 return ConstantFoldFP(
2834 sqrt, APF, Ty,
2837
2838
2839 case Intrinsic::amdgcn_cos:
2840 case Intrinsic::amdgcn_sin: {
2841 double V = getValueAsDouble(Op);
2842 if (V < -256.0 || V > 256.0)
2843
2844
2845
2846 return nullptr;
2847 bool IsCos = IntrinsicID == Intrinsic::amdgcn_cos;
2848 double V4 = V * 4.0;
2849 if (V4 == floor(V4)) {
2850
2851 const double SinVals[4] = { 0.0, 1.0, 0.0, -1.0 };
2852 V = SinVals[((int)V4 + (IsCos ? 1 : 0)) & 3];
2853 } else {
2854 if (IsCos)
2856 else
2858 }
2859 return GetConstantFoldFPValue(V, Ty);
2860 }
2861 }
2862
2863 if (!TLI)
2864 return nullptr;
2865
2866 LibFunc Func = NotLibFunc;
2868 return nullptr;
2869
2870 switch (Func) {
2871 default:
2872 break;
2873 case LibFunc_acos:
2874 case LibFunc_acosf:
2875 case LibFunc_acos_finite:
2876 case LibFunc_acosf_finite:
2877 if (TLI->has(Func))
2878 return ConstantFoldFP(acos, APF, Ty);
2879 break;
2880 case LibFunc_asin:
2881 case LibFunc_asinf:
2882 case LibFunc_asin_finite:
2883 case LibFunc_asinf_finite:
2884 if (TLI->has(Func))
2885 return ConstantFoldFP(asin, APF, Ty);
2886 break;
2887 case LibFunc_atan:
2888 case LibFunc_atanf:
2889
2890 if (U.isZero())
2891 return ConstantFP::get(Ty->getContext(), U);
2892 if (TLI->has(Func))
2893 return ConstantFoldFP(atan, APF, Ty);
2894 break;
2895 case LibFunc_ceil:
2896 case LibFunc_ceilf:
2897 if (TLI->has(Func)) {
2899 return ConstantFP::get(Ty->getContext(), U);
2900 }
2901 break;
2902 case LibFunc_cos:
2903 case LibFunc_cosf:
2904 if (TLI->has(Func))
2905 return ConstantFoldFP(cos, APF, Ty);
2906 break;
2907 case LibFunc_cosh:
2908 case LibFunc_coshf:
2909 case LibFunc_cosh_finite:
2910 case LibFunc_coshf_finite:
2911 if (TLI->has(Func))
2912 return ConstantFoldFP(cosh, APF, Ty);
2913 break;
2914 case LibFunc_exp:
2915 case LibFunc_expf:
2916 case LibFunc_exp_finite:
2917 case LibFunc_expf_finite:
2918 if (TLI->has(Func))
2919 return ConstantFoldFP(exp, APF, Ty);
2920 break;
2921 case LibFunc_exp2:
2922 case LibFunc_exp2f:
2923 case LibFunc_exp2_finite:
2924 case LibFunc_exp2f_finite:
2925 if (TLI->has(Func))
2926
2927 return ConstantFoldBinaryFP(pow, APFloat(2.0), APF, Ty);
2928 break;
2929 case LibFunc_fabs:
2930 case LibFunc_fabsf:
2931 if (TLI->has(Func)) {
2932 U.clearSign();
2933 return ConstantFP::get(Ty->getContext(), U);
2934 }
2935 break;
2936 case LibFunc_floor:
2937 case LibFunc_floorf:
2938 if (TLI->has(Func)) {
2940 return ConstantFP::get(Ty->getContext(), U);
2941 }
2942 break;
2943 case LibFunc_log:
2944 case LibFunc_logf:
2945 case LibFunc_log_finite:
2946 case LibFunc_logf_finite:
2948 return ConstantFoldFP(log, APF, Ty);
2949 break;
2950 case LibFunc_log2:
2951 case LibFunc_log2f:
2952 case LibFunc_log2_finite:
2953 case LibFunc_log2f_finite:
2955
2956 return ConstantFoldFP(log2, APF, Ty);
2957 break;
2958 case LibFunc_log10:
2959 case LibFunc_log10f:
2960 case LibFunc_log10_finite:
2961 case LibFunc_log10f_finite:
2963
2964 return ConstantFoldFP(log10, APF, Ty);
2965 break;
2966 case LibFunc_ilogb:
2967 case LibFunc_ilogbf:
2968 if (!APF.isZero() && TLI->has(Func))
2969 return ConstantInt::get(Ty, ilogb(APF), true);
2970 break;
2971 case LibFunc_logb:
2972 case LibFunc_logbf:
2973 if (!APF.isZero() && TLI->has(Func))
2974 return ConstantFoldFP(logb, APF, Ty);
2975 break;
2976 case LibFunc_log1p:
2977 case LibFunc_log1pf:
2978
2979 if (U.isZero())
2980 return ConstantFP::get(Ty->getContext(), U);
2982 return ConstantFoldFP(log1p, APF, Ty);
2983 break;
2984 case LibFunc_logl:
2985 return nullptr;
2986 case LibFunc_erf:
2987 case LibFunc_erff:
2988 if (TLI->has(Func))
2989 return ConstantFoldFP(erf, APF, Ty);
2990 break;
2991 case LibFunc_nearbyint:
2992 case LibFunc_nearbyintf:
2993 case LibFunc_rint:
2994 case LibFunc_rintf:
2995 case LibFunc_roundeven:
2996 case LibFunc_roundevenf:
2997 if (TLI->has(Func)) {
2999 return ConstantFP::get(Ty->getContext(), U);
3000 }
3001 break;
3002 case LibFunc_round:
3003 case LibFunc_roundf:
3004 if (TLI->has(Func)) {
3006 return ConstantFP::get(Ty->getContext(), U);
3007 }
3008 break;
3009 case LibFunc_sin:
3010 case LibFunc_sinf:
3011 if (TLI->has(Func))
3012 return ConstantFoldFP(sin, APF, Ty);
3013 break;
3014 case LibFunc_sinh:
3015 case LibFunc_sinhf:
3016 case LibFunc_sinh_finite:
3017 case LibFunc_sinhf_finite:
3018 if (TLI->has(Func))
3019 return ConstantFoldFP(sinh, APF, Ty);
3020 break;
3021 case LibFunc_sqrt:
3022 case LibFunc_sqrtf:
3024 return ConstantFoldFP(sqrt, APF, Ty);
3025 break;
3026 case LibFunc_tan:
3027 case LibFunc_tanf:
3028 if (TLI->has(Func))
3029 return ConstantFoldFP(tan, APF, Ty);
3030 break;
3031 case LibFunc_tanh:
3032 case LibFunc_tanhf:
3033 if (TLI->has(Func))
3034 return ConstantFoldFP(tanh, APF, Ty);
3035 break;
3036 case LibFunc_trunc:
3037 case LibFunc_truncf:
3038 if (TLI->has(Func)) {
3040 return ConstantFP::get(Ty->getContext(), U);
3041 }
3042 break;
3043 }
3044 return nullptr;
3045 }
3046
3048 switch (IntrinsicID) {
3049 case Intrinsic::bswap:
3050 return ConstantInt::get(Ty->getContext(), Op->getValue().byteSwap());
3051 case Intrinsic::ctpop:
3052 return ConstantInt::get(Ty, Op->getValue().popcount());
3053 case Intrinsic::bitreverse:
3054 return ConstantInt::get(Ty->getContext(), Op->getValue().reverseBits());
3055 case Intrinsic::convert_from_fp16: {
3057
3058 bool lost = false;
3061
3062
3063 (void)status;
3065 "Precision lost during fp16 constfolding");
3066
3067 return ConstantFP::get(Ty->getContext(), Val);
3068 }
3069
3070 case Intrinsic::amdgcn_s_wqm: {
3072 Val |= (Val & 0x5555555555555555ULL) << 1 |
3073 ((Val >> 1) & 0x5555555555555555ULL);
3074 Val |= (Val & 0x3333333333333333ULL) << 2 |
3075 ((Val >> 2) & 0x3333333333333333ULL);
3076 return ConstantInt::get(Ty, Val);
3077 }
3078
3079 case Intrinsic::amdgcn_s_quadmask: {
3082 for (unsigned I = 0; I < Op->getBitWidth() / 4; ++I, Val >>= 4) {
3083 if (!(Val & 0xF))
3084 continue;
3085
3086 QuadMask |= (1ULL << I);
3087 }
3088 return ConstantInt::get(Ty, QuadMask);
3089 }
3090
3091 case Intrinsic::amdgcn_s_bitreplicate: {
3093 Val = (Val & 0x000000000000FFFFULL) | (Val & 0x00000000FFFF0000ULL) << 16;
3094 Val = (Val & 0x000000FF000000FFULL) | (Val & 0x0000FF000000FF00ULL) << 8;
3095 Val = (Val & 0x000F000F000F000FULL) | (Val & 0x00F000F000F000F0ULL) << 4;
3096 Val = (Val & 0x0303030303030303ULL) | (Val & 0x0C0C0C0C0C0C0C0CULL) << 2;
3097 Val = (Val & 0x1111111111111111ULL) | (Val & 0x2222222222222222ULL) << 1;
3098 Val = Val | Val << 1;
3099 return ConstantInt::get(Ty, Val);
3100 }
3101 }
3102 }
3103
3104 if (Operands[0]->getType()->isVectorTy()) {
3106 switch (IntrinsicID) {
3107 default: break;
3108 case Intrinsic::vector_reduce_add:
3109 case Intrinsic::vector_reduce_mul:
3110 case Intrinsic::vector_reduce_and:
3111 case Intrinsic::vector_reduce_or:
3112 case Intrinsic::vector_reduce_xor:
3113 case Intrinsic::vector_reduce_smin:
3114 case Intrinsic::vector_reduce_smax:
3115 case Intrinsic::vector_reduce_umin:
3116 case Intrinsic::vector_reduce_umax:
3117 if (Constant *C = constantFoldVectorReduce(IntrinsicID, Operands[0]))
3118 return C;
3119 break;
3120 case Intrinsic::x86_sse_cvtss2si:
3121 case Intrinsic::x86_sse_cvtss2si64:
3122 case Intrinsic::x86_sse2_cvtsd2si:
3123 case Intrinsic::x86_sse2_cvtsd2si64:
3126 return ConstantFoldSSEConvertToInt(FPOp->getValueAPF(),
3127 false, Ty,
3128 true);
3129 break;
3130 case Intrinsic::x86_sse_cvttss2si:
3131 case Intrinsic::x86_sse_cvttss2si64:
3132 case Intrinsic::x86_sse2_cvttsd2si:
3133 case Intrinsic::x86_sse2_cvttsd2si64:
3136 return ConstantFoldSSEConvertToInt(FPOp->getValueAPF(),
3137 true, Ty,
3138 true);
3139 break;
3140
3141 case Intrinsic::wasm_anytrue:
3142 return Op->isZeroValue() ? ConstantInt::get(Ty, 0)
3144
3145 case Intrinsic::wasm_alltrue:
3146
3148 for (unsigned I = 0; I != E; ++I) {
3149 Constant *Elt = Op->getAggregateElement(I);
3150
3152 return ConstantInt::get(Ty, 0);
3153
3155 return nullptr;
3156 }
3157
3158 return ConstantInt::get(Ty, 1);
3159 }
3160 }
3161
3162 return nullptr;
3163}
3164
3170 if (FCmp->isSignaling()) {
3173 } else {
3176 }
3179 return ConstantInt::get(Call->getType()->getScalarType(), Result);
3180 return nullptr;
3181}
3182
3186 if (!TLI)
3187 return nullptr;
3188
3189 LibFunc Func = NotLibFunc;
3191 return nullptr;
3192
3194 if (!Op1)
3195 return nullptr;
3196
3198 if (!Op2)
3199 return nullptr;
3200
3201 const APFloat &Op1V = Op1->getValueAPF();
3202 const APFloat &Op2V = Op2->getValueAPF();
3203
3204 switch (Func) {
3205 default:
3206 break;
3207 case LibFunc_pow:
3208 case LibFunc_powf:
3209 case LibFunc_pow_finite:
3210 case LibFunc_powf_finite:
3211 if (TLI->has(Func))
3212 return ConstantFoldBinaryFP(pow, Op1V, Op2V, Ty);
3213 break;
3214 case LibFunc_fmod:
3215 case LibFunc_fmodf:
3216 if (TLI->has(Func)) {
3217 APFloat V = Op1->getValueAPF();
3219 return ConstantFP::get(Ty->getContext(), V);
3220 }
3221 break;
3222 case LibFunc_remainder:
3223 case LibFunc_remainderf:
3224 if (TLI->has(Func)) {
3225 APFloat V = Op1->getValueAPF();
3227 return ConstantFP::get(Ty->getContext(), V);
3228 }
3229 break;
3230 case LibFunc_atan2:
3231 case LibFunc_atan2f:
3232
3233
3235 return nullptr;
3236 [[fallthrough]];
3237 case LibFunc_atan2_finite:
3238 case LibFunc_atan2f_finite:
3239 if (TLI->has(Func))
3240 return ConstantFoldBinaryFP(atan2, Op1V, Op2V, Ty);
3241 break;
3242 }
3243
3244 return nullptr;
3245}
3246
3250 assert(Operands.size() == 2 && "Wrong number of operands.");
3251
3252 if (Ty->isFloatingPointTy()) {
3253
3254
3257 switch (IntrinsicID) {
3258 case Intrinsic::maxnum:
3259 case Intrinsic::minnum:
3260 case Intrinsic::maximum:
3261 case Intrinsic::minimum:
3262 case Intrinsic::maximumnum:
3263 case Intrinsic::minimumnum:
3264 case Intrinsic::nvvm_fmax_d:
3265 case Intrinsic::nvvm_fmin_d:
3266
3267 if (IsOp0Undef)
3268 return Operands[1];
3269 if (IsOp1Undef)
3270 return Operands[0];
3271 break;
3272
3273 case Intrinsic::nvvm_fmax_f:
3274 case Intrinsic::nvvm_fmax_ftz_f:
3275 case Intrinsic::nvvm_fmax_ftz_nan_f:
3276 case Intrinsic::nvvm_fmax_ftz_nan_xorsign_abs_f:
3277 case Intrinsic::nvvm_fmax_ftz_xorsign_abs_f:
3278 case Intrinsic::nvvm_fmax_nan_f:
3279 case Intrinsic::nvvm_fmax_nan_xorsign_abs_f:
3280 case Intrinsic::nvvm_fmax_xorsign_abs_f:
3281
3282 case Intrinsic::nvvm_fmin_f:
3283 case Intrinsic::nvvm_fmin_ftz_f:
3284 case Intrinsic::nvvm_fmin_ftz_nan_f:
3285 case Intrinsic::nvvm_fmin_ftz_nan_xorsign_abs_f:
3286 case Intrinsic::nvvm_fmin_ftz_xorsign_abs_f:
3287 case Intrinsic::nvvm_fmin_nan_f:
3288 case Intrinsic::nvvm_fmin_nan_xorsign_abs_f:
3289 case Intrinsic::nvvm_fmin_xorsign_abs_f:
3290
3291
3292
3293 if (!IsOp0Undef && !IsOp1Undef)
3294 break;
3296 if (Op->isNaN()) {
3297 APInt NVCanonicalNaN(32, 0x7fffffff);
3298 return ConstantFP::get(
3299 Ty, APFloat(Ty->getFltSemantics(), NVCanonicalNaN));
3300 }
3302 return ConstantFP::get(Ty, FTZPreserveSign(Op->getValueAPF()));
3303 else
3304 return Op;
3305 }
3306 break;
3307 }
3308 }
3309
3311 const APFloat &Op1V = Op1->getValueAPF();
3312
3314 if (Op2->getType() != Op1->getType())
3315 return nullptr;
3316 const APFloat &Op2V = Op2->getValueAPF();
3317
3318 if (const auto *ConstrIntr =
3320 RoundingMode RM = getEvaluationRoundingMode(ConstrIntr);
3323 switch (IntrinsicID) {
3324 default:
3325 return nullptr;
3326 case Intrinsic::experimental_constrained_fadd:
3327 St = Res.add(Op2V, RM);
3328 break;
3329 case Intrinsic::experimental_constrained_fsub:
3330 St = Res.subtract(Op2V, RM);
3331 break;
3332 case Intrinsic::experimental_constrained_fmul:
3333 St = Res.multiply(Op2V, RM);
3334 break;
3335 case Intrinsic::experimental_constrained_fdiv:
3336 St = Res.divide(Op2V, RM);
3337 break;
3338 case Intrinsic::experimental_constrained_frem:
3339 St = Res.mod(Op2V);
3340 break;
3341 case Intrinsic::experimental_constrained_fcmp:
3342 case Intrinsic::experimental_constrained_fcmps:
3343 return evaluateCompare(Op1V, Op2V, ConstrIntr);
3344 }
3346 St))
3347 return ConstantFP::get(Ty->getContext(), Res);
3348 return nullptr;
3349 }
3350
3351 switch (IntrinsicID) {
3352 default:
3353 break;
3354 case Intrinsic::copysign:
3355 return ConstantFP::get(Ty->getContext(), APFloat::copySign(Op1V, Op2V));
3356 case Intrinsic::minnum:
3358 return nullptr;
3359 return ConstantFP::get(Ty->getContext(), minnum(Op1V, Op2V));
3360 case Intrinsic::maxnum:
3362 return nullptr;
3363 return ConstantFP::get(Ty->getContext(), maxnum(Op1V, Op2V));
3364 case Intrinsic::minimum:
3365 return ConstantFP::get(Ty->getContext(), minimum(Op1V, Op2V));
3366 case Intrinsic::maximum:
3367 return ConstantFP::get(Ty->getContext(), maximum(Op1V, Op2V));
3368 case Intrinsic::minimumnum:
3369 return ConstantFP::get(Ty->getContext(), minimumnum(Op1V, Op2V));
3370 case Intrinsic::maximumnum:
3371 return ConstantFP::get(Ty->getContext(), maximumnum(Op1V, Op2V));
3372
3373 case Intrinsic::nvvm_fmax_d:
3374 case Intrinsic::nvvm_fmax_f:
3375 case Intrinsic::nvvm_fmax_ftz_f:
3376 case Intrinsic::nvvm_fmax_ftz_nan_f:
3377 case Intrinsic::nvvm_fmax_ftz_nan_xorsign_abs_f:
3378 case Intrinsic::nvvm_fmax_ftz_xorsign_abs_f:
3379 case Intrinsic::nvvm_fmax_nan_f:
3380 case Intrinsic::nvvm_fmax_nan_xorsign_abs_f:
3381 case Intrinsic::nvvm_fmax_xorsign_abs_f:
3382
3383 case Intrinsic::nvvm_fmin_d:
3384 case Intrinsic::nvvm_fmin_f:
3385 case Intrinsic::nvvm_fmin_ftz_f:
3386 case Intrinsic::nvvm_fmin_ftz_nan_f:
3387 case Intrinsic::nvvm_fmin_ftz_nan_xorsign_abs_f:
3388 case Intrinsic::nvvm_fmin_ftz_xorsign_abs_f:
3389 case Intrinsic::nvvm_fmin_nan_f:
3390 case Intrinsic::nvvm_fmin_nan_xorsign_abs_f:
3391 case Intrinsic::nvvm_fmin_xorsign_abs_f: {
3392
3393 bool ShouldCanonicalizeNaNs = !(IntrinsicID == Intrinsic::nvvm_fmax_d ||
3394 IntrinsicID == Intrinsic::nvvm_fmin_d);
3398
3399 APFloat A = IsFTZ ? FTZPreserveSign(Op1V) : Op1V;
3400 APFloat B = IsFTZ ? FTZPreserveSign(Op2V) : Op2V;
3401
3402 bool XorSign = false;
3403 if (IsXorSignAbs) {
3404 XorSign = A.isNegative() ^ B.isNegative();
3407 }
3408
3409 bool IsFMax = false;
3410 switch (IntrinsicID) {
3411 case Intrinsic::nvvm_fmax_d:
3412 case Intrinsic::nvvm_fmax_f:
3413 case Intrinsic::nvvm_fmax_ftz_f:
3414 case Intrinsic::nvvm_fmax_ftz_nan_f:
3415 case Intrinsic::nvvm_fmax_ftz_nan_xorsign_abs_f:
3416 case Intrinsic::nvvm_fmax_ftz_xorsign_abs_f:
3417 case Intrinsic::nvvm_fmax_nan_f:
3418 case Intrinsic::nvvm_fmax_nan_xorsign_abs_f:
3419 case Intrinsic::nvvm_fmax_xorsign_abs_f:
3420 IsFMax = true;
3421 break;
3422 }
3424
3425 if (ShouldCanonicalizeNaNs) {
3427 if (A.isNaN() && B.isNaN())
3428 return ConstantFP::get(Ty, NVCanonicalNaN);
3429 else if (IsNaNPropagating && (A.isNaN() || B.isNaN()))
3430 return ConstantFP::get(Ty, NVCanonicalNaN);
3431 }
3432
3433 if (A.isNaN() && B.isNaN())
3434 return Operands[1];
3436 Res = B;
3438 Res = A;
3439
3440 if (IsXorSignAbs && XorSign != Res.isNegative())
3442
3443 return ConstantFP::get(Ty->getContext(), Res);
3444 }
3445
3446 case Intrinsic::nvvm_add_rm_f:
3447 case Intrinsic::nvvm_add_rn_f:
3448 case Intrinsic::nvvm_add_rp_f:
3449 case Intrinsic::nvvm_add_rz_f:
3450 case Intrinsic::nvvm_add_rm_d:
3451 case Intrinsic::nvvm_add_rn_d:
3452 case Intrinsic::nvvm_add_rp_d:
3453 case Intrinsic::nvvm_add_rz_d:
3454 case Intrinsic::nvvm_add_rm_ftz_f:
3455 case Intrinsic::nvvm_add_rn_ftz_f:
3456 case Intrinsic::nvvm_add_rp_ftz_f:
3457 case Intrinsic::nvvm_add_rz_ftz_f: {
3458
3460 APFloat A = IsFTZ ? FTZPreserveSign(Op1V) : Op1V;
3461 APFloat B = IsFTZ ? FTZPreserveSign(Op2V) : Op2V;
3462
3465
3468
3469 if (!Res.isNaN() &&
3471 Res = IsFTZ ? FTZPreserveSign(Res) : Res;
3472 return ConstantFP::get(Ty->getContext(), Res);
3473 }
3474 return nullptr;
3475 }
3476
3477 case Intrinsic::nvvm_mul_rm_f:
3478 case Intrinsic::nvvm_mul_rn_f:
3479 case Intrinsic::nvvm_mul_rp_f:
3480 case Intrinsic::nvvm_mul_rz_f:
3481 case Intrinsic::nvvm_mul_rm_d:
3482 case Intrinsic::nvvm_mul_rn_d:
3483 case Intrinsic::nvvm_mul_rp_d:
3484 case Intrinsic::nvvm_mul_rz_d:
3485 case Intrinsic::nvvm_mul_rm_ftz_f:
3486 case Intrinsic::nvvm_mul_rn_ftz_f:
3487 case Intrinsic::nvvm_mul_rp_ftz_f:
3488 case Intrinsic::nvvm_mul_rz_ftz_f: {
3489
3491 APFloat A = IsFTZ ? FTZPreserveSign(Op1V) : Op1V;
3492 APFloat B = IsFTZ ? FTZPreserveSign(Op2V) : Op2V;
3493
3496
3499
3500 if (!Res.isNaN() &&
3502 Res = IsFTZ ? FTZPreserveSign(Res) : Res;
3503 return ConstantFP::get(Ty->getContext(), Res);
3504 }
3505 return nullptr;
3506 }
3507
3508 case Intrinsic::nvvm_div_rm_f:
3509 case Intrinsic::nvvm_div_rn_f:
3510 case Intrinsic::nvvm_div_rp_f:
3511 case Intrinsic::nvvm_div_rz_f:
3512 case Intrinsic::nvvm_div_rm_d:
3513 case Intrinsic::nvvm_div_rn_d:
3514 case Intrinsic::nvvm_div_rp_d:
3515 case Intrinsic::nvvm_div_rz_d:
3516 case Intrinsic::nvvm_div_rm_ftz_f:
3517 case Intrinsic::nvvm_div_rn_ftz_f:
3518 case Intrinsic::nvvm_div_rp_ftz_f:
3519 case Intrinsic::nvvm_div_rz_ftz_f: {
3521 APFloat A = IsFTZ ? FTZPreserveSign(Op1V) : Op1V;
3522 APFloat B = IsFTZ ? FTZPreserveSign(Op2V) : Op2V;
3525
3528 if (!Res.isNaN() &&
3530 Res = IsFTZ ? FTZPreserveSign(Res) : Res;
3531 return ConstantFP::get(Ty->getContext(), Res);
3532 }
3533 return nullptr;
3534 }
3535 }
3536
3537 if (!Ty->isHalfTy() && !Ty->isFloatTy() && !Ty->isDoubleTy())
3538 return nullptr;
3539
3540 switch (IntrinsicID) {
3541 default:
3542 break;
3543 case Intrinsic::pow:
3544 return ConstantFoldBinaryFP(pow, Op1V, Op2V, Ty);
3545 case Intrinsic::amdgcn_fmul_legacy:
3546
3547
3550 return ConstantFP::get(Ty->getContext(), Op1V * Op2V);
3551 }
3552
3554 switch (IntrinsicID) {
3555 case Intrinsic::ldexp: {
3556 return ConstantFP::get(
3557 Ty->getContext(),
3559 }
3560 case Intrinsic::is_fpclass: {
3573 return ConstantInt::get(Ty, Result);
3574 }
3575 case Intrinsic::powi: {
3576 int Exp = static_cast<int>(Op2C->getSExtValue());
3577 switch (Ty->getTypeID()) {
3581 if (Ty->isHalfTy()) {
3584 &Unused);
3585 }
3586 return ConstantFP::get(Ty->getContext(), Res);
3587 }
3589 return ConstantFP::get(Ty, std::pow(Op1V.convertToDouble(), Exp));
3590 default:
3591 return nullptr;
3592 }
3593 }
3594 default:
3595 break;
3596 }
3597 }
3598 return nullptr;
3599 }
3600
3601 if (Operands[0]->getType()->isIntegerTy() &&
3602 Operands[1]->getType()->isIntegerTy()) {
3603 const APInt *C0, *C1;
3604 if (!getConstIntOrUndef(Operands[0], C0) ||
3605 !getConstIntOrUndef(Operands[1], C1))
3606 return nullptr;
3607
3608 switch (IntrinsicID) {
3609 default: break;
3610 case Intrinsic::smax:
3611 case Intrinsic::smin:
3612 case Intrinsic::umax:
3613 case Intrinsic::umin:
3614 if (!C0 && !C1)
3616 if (!C0 || !C1)
3618 return ConstantInt::get(
3621 ? *C0
3622 : *C1);
3623
3624 case Intrinsic::scmp:
3625 case Intrinsic::ucmp:
3626 if (!C0 || !C1)
3627 return ConstantInt::get(Ty, 0);
3628
3629 int Res;
3630 if (IntrinsicID == Intrinsic::scmp)
3631 Res = C0->sgt(*C1) ? 1 : C0->slt(*C1) ? -1 : 0;
3632 else
3633 Res = C0->ugt(*C1) ? 1 : C0->ult(*C1) ? -1 : 0;
3634 return ConstantInt::get(Ty, Res, true);
3635
3636 case Intrinsic::usub_with_overflow:
3637 case Intrinsic::ssub_with_overflow:
3638
3639
3640 if (!C0 || !C1)
3642 [[fallthrough]];
3643 case Intrinsic::uadd_with_overflow:
3644 case Intrinsic::sadd_with_overflow:
3645
3646
3647 if (!C0 || !C1) {
3652 }
3653 [[fallthrough]];
3654 case Intrinsic::smul_with_overflow:
3655 case Intrinsic::umul_with_overflow: {
3656
3657
3658 if (!C0 || !C1)
3660
3662 bool Overflow;
3663 switch (IntrinsicID) {
3665 case Intrinsic::sadd_with_overflow:
3666 Res = C0->sadd_ov(*C1, Overflow);
3667 break;
3668 case Intrinsic::uadd_with_overflow:
3669 Res = C0->uadd_ov(*C1, Overflow);
3670 break;
3671 case Intrinsic::ssub_with_overflow:
3672 Res = C0->ssub_ov(*C1, Overflow);
3673 break;
3674 case Intrinsic::usub_with_overflow:
3675 Res = C0->usub_ov(*C1, Overflow);
3676 break;
3677 case Intrinsic::smul_with_overflow:
3678 Res = C0->smul_ov(*C1, Overflow);
3679 break;
3680 case Intrinsic::umul_with_overflow:
3681 Res = C0->umul_ov(*C1, Overflow);
3682 break;
3683 }
3685 ConstantInt::get(Ty->getContext(), Res),
3686 ConstantInt::get(Type::getInt1Ty(Ty->getContext()), Overflow)
3687 };
3689 }
3690 case Intrinsic::uadd_sat:
3691 case Intrinsic::sadd_sat:
3692 if (!C0 && !C1)
3694 if (!C0 || !C1)
3696 if (IntrinsicID == Intrinsic::uadd_sat)
3697 return ConstantInt::get(Ty, C0->uadd_sat(*C1));
3698 else
3699 return ConstantInt::get(Ty, C0->sadd_sat(*C1));
3700 case Intrinsic::usub_sat:
3701 case Intrinsic::ssub_sat:
3702 if (!C0 && !C1)
3704 if (!C0 || !C1)
3706 if (IntrinsicID == Intrinsic::usub_sat)
3707 return ConstantInt::get(Ty, C0->usub_sat(*C1));
3708 else
3709 return ConstantInt::get(Ty, C0->ssub_sat(*C1));
3710 case Intrinsic::cttz:
3711 case Intrinsic::ctlz:
3712 assert(C1 && "Must be constant int");
3713
3714
3715 if (C1->isOne() && (!C0 || C0->isZero()))
3717 if (!C0)
3719 if (IntrinsicID == Intrinsic::cttz)
3720 return ConstantInt::get(Ty, C0->countr_zero());
3721 else
3722 return ConstantInt::get(Ty, C0->countl_zero());
3723
3724 case Intrinsic::abs:
3725 assert(C1 && "Must be constant int");
3727
3728
3731
3732
3733 if (!C0)
3735
3736 return ConstantInt::get(Ty, C0->abs());
3737 case Intrinsic::amdgcn_wave_reduce_umin:
3738 case Intrinsic::amdgcn_wave_reduce_umax:
3739 case Intrinsic::amdgcn_wave_reduce_max:
3740 case Intrinsic::amdgcn_wave_reduce_min:
3741 case Intrinsic::amdgcn_wave_reduce_add:
3742 case Intrinsic::amdgcn_wave_reduce_sub:
3743 case Intrinsic::amdgcn_wave_reduce_and:
3744 case Intrinsic::amdgcn_wave_reduce_or:
3745 case Intrinsic::amdgcn_wave_reduce_xor:
3747 }
3748
3749 return nullptr;
3750 }
3751
3752
3755
3756
3760 switch (IntrinsicID) {
3761 default: break;
3762 case Intrinsic::x86_avx512_vcvtss2si32:
3763 case Intrinsic::x86_avx512_vcvtss2si64:
3764 case Intrinsic::x86_avx512_vcvtsd2si32:
3765 case Intrinsic::x86_avx512_vcvtsd2si64:
3768 return ConstantFoldSSEConvertToInt(FPOp->getValueAPF(),
3769 false, Ty,
3770 true);
3771 break;
3772 case Intrinsic::x86_avx512_vcvtss2usi32:
3773 case Intrinsic::x86_avx512_vcvtss2usi64:
3774 case Intrinsic::x86_avx512_vcvtsd2usi32:
3775 case Intrinsic::x86_avx512_vcvtsd2usi64:
3778 return ConstantFoldSSEConvertToInt(FPOp->getValueAPF(),
3779 false, Ty,
3780 false);
3781 break;
3782 case Intrinsic::x86_avx512_cvttss2si:
3783 case Intrinsic::x86_avx512_cvttss2si64:
3784 case Intrinsic::x86_avx512_cvttsd2si:
3785 case Intrinsic::x86_avx512_cvttsd2si64:
3788 return ConstantFoldSSEConvertToInt(FPOp->getValueAPF(),
3789 true, Ty,
3790 true);
3791 break;
3792 case Intrinsic::x86_avx512_cvttss2usi:
3793 case Intrinsic::x86_avx512_cvttss2usi64:
3794 case Intrinsic::x86_avx512_cvttsd2usi:
3795 case Intrinsic::x86_avx512_cvttsd2usi64:
3798 return ConstantFoldSSEConvertToInt(FPOp->getValueAPF(),
3799 true, Ty,
3800 false);
3801 break;
3802 }
3803 }
3804 return nullptr;
3805}
3806
3811 unsigned ID;
3813 APFloat MA(Sem), SC(Sem), TC(Sem);
3816
3817 ID = 5;
3818 SC = -S0;
3819 } else {
3820 ID = 4;
3821 SC = S0;
3822 }
3823 MA = S2;
3824 TC = -S1;
3825 } else if (abs(S1) >= abs(S0)) {
3826 if (S1.isNegative() && S1.isNonZero() && .isNaN()) {
3827
3828 ID = 3;
3829 TC = -S2;
3830 } else {
3831 ID = 2;
3832 TC = S2;
3833 }
3834 MA = S1;
3835 SC = S0;
3836 } else {
3838
3839 ID = 1;
3840 SC = S2;
3841 } else {
3842 ID = 0;
3843 SC = -S2;
3844 }
3845 MA = S0;
3846 TC = -S1;
3847 }
3848 switch (IntrinsicID) {
3849 default:
3851 case Intrinsic::amdgcn_cubeid:
3853 case Intrinsic::amdgcn_cubema:
3854 return MA + MA;
3855 case Intrinsic::amdgcn_cubesc:
3856 return SC;
3857 case Intrinsic::amdgcn_cubetc:
3858 return TC;
3859 }
3860}
3861
3864 const APInt *C0, *C1, *C2;
3865 if (!getConstIntOrUndef(Operands[0], C0) ||
3866 !getConstIntOrUndef(Operands[1], C1) ||
3867 !getConstIntOrUndef(Operands[2], C2))
3868 return nullptr;
3869
3870 if (!C2)
3872
3873 APInt Val(32, 0);
3874 unsigned NumUndefBytes = 0;
3875 for (unsigned I = 0; I < 32; I += 8) {
3877 unsigned B = 0;
3878
3879 if (Sel >= 13)
3880 B = 0xff;
3881 else if (Sel == 12)
3882 B = 0x00;
3883 else {
3884 const APInt *Src = ((Sel & 10) == 10 || (Sel & 12) == 4) ? C0 : C1;
3885 if (!Src)
3886 ++NumUndefBytes;
3887 else if (Sel < 8)
3888 B = Src->extractBitsAsZExtValue(8, (Sel & 3) * 8);
3889 else
3890 B = Src->extractBitsAsZExtValue(1, (Sel & 1) ? 31 : 15) * 0xff;
3891 }
3892
3894 }
3895
3896 if (NumUndefBytes == 4)
3898
3899 return ConstantInt::get(Ty, Val);
3900}
3901
3908 assert(Operands.size() == 3 && "Wrong number of operands.");
3909
3913 const APFloat &C1 = Op1->getValueAPF();
3914 const APFloat &C2 = Op2->getValueAPF();
3915 const APFloat &C3 = Op3->getValueAPF();
3916
3918 RoundingMode RM = getEvaluationRoundingMode(ConstrIntr);
3921 switch (IntrinsicID) {
3922 default:
3923 return nullptr;
3924 case Intrinsic::experimental_constrained_fma:
3925 case Intrinsic::experimental_constrained_fmuladd:
3927 break;
3928 }
3929 if (mayFoldConstrained(
3931 return ConstantFP::get(Ty->getContext(), Res);
3932 return nullptr;
3933 }
3934
3935 switch (IntrinsicID) {
3936 default: break;
3937 case Intrinsic::amdgcn_fma_legacy: {
3938
3939
3941
3942
3943 return ConstantFP::get(Ty->getContext(), APFloat(0.0f) + C3);
3944 }
3945 [[fallthrough]];
3946 }
3947 case Intrinsic::fma:
3948 case Intrinsic::fmuladd: {
3951 return ConstantFP::get(Ty->getContext(), V);
3952 }
3953
3954 case Intrinsic::nvvm_fma_rm_f:
3955 case Intrinsic::nvvm_fma_rn_f:
3956 case Intrinsic::nvvm_fma_rp_f:
3957 case Intrinsic::nvvm_fma_rz_f:
3958 case Intrinsic::nvvm_fma_rm_d:
3959 case Intrinsic::nvvm_fma_rn_d:
3960 case Intrinsic::nvvm_fma_rp_d:
3961 case Intrinsic::nvvm_fma_rz_d:
3962 case Intrinsic::nvvm_fma_rm_ftz_f:
3963 case Intrinsic::nvvm_fma_rn_ftz_f:
3964 case Intrinsic::nvvm_fma_rp_ftz_f:
3965 case Intrinsic::nvvm_fma_rz_ftz_f: {
3967 APFloat A = IsFTZ ? FTZPreserveSign(C1) : C1;
3968 APFloat B = IsFTZ ? FTZPreserveSign(C2) : C2;
3969 APFloat C = IsFTZ ? FTZPreserveSign(C3) : C3;
3970
3973
3976
3977 if (!Res.isNaN() &&
3979 Res = IsFTZ ? FTZPreserveSign(Res) : Res;
3980 return ConstantFP::get(Ty->getContext(), Res);
3981 }
3982 return nullptr;
3983 }
3984
3985 case Intrinsic::amdgcn_cubeid:
3986 case Intrinsic::amdgcn_cubema:
3987 case Intrinsic::amdgcn_cubesc:
3988 case Intrinsic::amdgcn_cubetc: {
3989 APFloat V = ConstantFoldAMDGCNCubeIntrinsic(IntrinsicID, C1, C2, C3);
3990 return ConstantFP::get(Ty->getContext(), V);
3991 }
3992 }
3993 }
3994 }
3995 }
3996
3997 if (IntrinsicID == Intrinsic::smul_fix ||
3998 IntrinsicID == Intrinsic::smul_fix_sat) {
3999 const APInt *C0, *C1;
4000 if (!getConstIntOrUndef(Operands[0], C0) ||
4001 !getConstIntOrUndef(Operands[1], C1))
4002 return nullptr;
4003
4004
4005
4006 if (!C0 || !C1)
4008
4009
4010
4011
4012
4013
4014
4015 unsigned Scale = cast(Operands[2])->getZExtValue();
4017 assert(Scale < Width && "Illegal scale.");
4018 unsigned ExtendedWidth = Width * 2;
4020 (C0->sext(ExtendedWidth) * C1->sext(ExtendedWidth)).ashr(Scale);
4021 if (IntrinsicID == Intrinsic::smul_fix_sat) {
4026 }
4027 return ConstantInt::get(Ty->getContext(), Product.sextOrTrunc(Width));
4028 }
4029
4030 if (IntrinsicID == Intrinsic::fshl || IntrinsicID == Intrinsic::fshr) {
4031 const APInt *C0, *C1, *C2;
4032 if (!getConstIntOrUndef(Operands[0], C0) ||
4033 !getConstIntOrUndef(Operands[1], C1) ||
4034 !getConstIntOrUndef(Operands[2], C2))
4035 return nullptr;
4036
4037 bool IsRight = IntrinsicID == Intrinsic::fshr;
4038 if (!C2)
4039 return Operands[IsRight ? 1 : 0];
4040 if (!C0 && !C1)
4042
4043
4044
4047 if (!ShAmt)
4048 return Operands[IsRight ? 1 : 0];
4049
4050
4051 unsigned LshrAmt = IsRight ? ShAmt : BitWidth - ShAmt;
4052 unsigned ShlAmt = !IsRight ? ShAmt : BitWidth - ShAmt;
4053 if (!C0)
4054 return ConstantInt::get(Ty, C1->lshr(LshrAmt));
4055 if (!C1)
4056 return ConstantInt::get(Ty, C0->shl(ShlAmt));
4057 return ConstantInt::get(Ty, C0->shl(ShlAmt) | C1->lshr(LshrAmt));
4058 }
4059
4060 if (IntrinsicID == Intrinsic::amdgcn_perm)
4061 return ConstantFoldAMDGCNPermIntrinsic(Operands, Ty);
4062
4063 return nullptr;
4064}
4065
4076
4077 if (Operands.size() == 1)
4078 return ConstantFoldScalarCall1(Name, IntrinsicID, Ty, Operands, TLI, Call);
4079
4080 if (Operands.size() == 2) {
4081 if (Constant *FoldedLibCall =
4082 ConstantFoldLibCall2(Name, Ty, Operands, TLI)) {
4083 return FoldedLibCall;
4084 }
4085 return ConstantFoldIntrinsicCall2(IntrinsicID, Ty, Operands, Call);
4086 }
4087
4088 if (Operands.size() == 3)
4089 return ConstantFoldScalarCall3(Name, IntrinsicID, Ty, Operands, TLI, Call);
4090
4091 return nullptr;
4092}
4093
4094static Constant *ConstantFoldFixedVectorCall(
4101
4102 switch (IntrinsicID) {
4103 case Intrinsic::masked_load: {
4104 auto *SrcPtr = Operands[0];
4105 auto *Mask = Operands[1];
4106 auto *Passthru = Operands[2];
4107
4109
4112 auto *MaskElt = Mask->getAggregateElement(I);
4113 if (!MaskElt)
4114 break;
4115 auto *PassthruElt = Passthru->getAggregateElement(I);
4118 if (PassthruElt)
4119 NewElements.push_back(PassthruElt);
4120 else if (VecElt)
4122 else
4123 return nullptr;
4124 }
4125 if (MaskElt->isNullValue()) {
4126 if (!PassthruElt)
4127 return nullptr;
4128 NewElements.push_back(PassthruElt);
4129 } else if (MaskElt->isOneValue()) {
4130 if (!VecElt)
4131 return nullptr;
4133 } else {
4134 return nullptr;
4135 }
4136 }
4138 return nullptr;
4140 }
4141 case Intrinsic::arm_mve_vctp8:
4142 case Intrinsic::arm_mve_vctp16:
4143 case Intrinsic::arm_mve_vctp32:
4144 case Intrinsic::arm_mve_vctp64: {
4147 uint64_t Limit = Op->getZExtValue();
4148
4150 for (unsigned i = 0; i < Lanes; i++) {
4151 if (i < Limit)
4153 else
4155 }
4157 }
4158 return nullptr;
4159 }
4160 case Intrinsic::get_active_lane_mask: {
4163 if (Op0 && Op1) {
4166 uint64_t Limit = Op1->getZExtValue();
4167
4169 for (unsigned i = 0; i < Lanes; i++) {
4170 if (Base + i < Limit)
4172 else
4174 }
4176 }
4177 return nullptr;
4178 }
4179 case Intrinsic::vector_extract: {
4181 Constant *Vec = Operands[0];
4183 return nullptr;
4184
4186 unsigned VecNumElements =
4188 unsigned StartingIndex = Idx->getZExtValue();
4189
4190
4191 if (NumElements == VecNumElements && StartingIndex == 0)
4192 return Vec;
4193
4194 for (unsigned I = StartingIndex, E = StartingIndex + NumElements; I < E;
4195 ++I) {
4197 if (!Elt)
4198 return nullptr;
4199 Result[I - StartingIndex] = Elt;
4200 }
4201
4203 }
4204 case Intrinsic::vector_insert: {
4205 Constant *Vec = Operands[0];
4206 Constant *SubVec = Operands[1];
4209 return nullptr;
4210
4211 unsigned SubVecNumElements =
4213 unsigned VecNumElements =
4215 unsigned IdxN = Idx->getZExtValue();
4216
4217 if (SubVecNumElements == VecNumElements && IdxN == 0)
4218 return SubVec;
4219
4220 for (unsigned I = 0; I < VecNumElements; ++I) {
4222 if (I < IdxN + SubVecNumElements)
4224 else
4226 if (!Elt)
4227 return nullptr;
4229 }
4231 }
4232 case Intrinsic::vector_interleave2:
4233 case Intrinsic::vector_interleave3:
4234 case Intrinsic::vector_interleave4:
4235 case Intrinsic::vector_interleave5:
4236 case Intrinsic::vector_interleave6:
4237 case Intrinsic::vector_interleave7:
4238 case Intrinsic::vector_interleave8: {
4239 unsigned NumElements =
4241 unsigned NumOperands = Operands.size();
4242 for (unsigned I = 0; I < NumElements; ++I) {
4243 for (unsigned J = 0; J < NumOperands; ++J) {
4244 Constant *Elt = Operands[J]->getAggregateElement(I);
4245 if (!Elt)
4246 return nullptr;
4247 Result[NumOperands * I + J] = Elt;
4248 }
4249 }
4251 }
4252 case Intrinsic::wasm_dot: {
4253 unsigned NumElements =
4255
4256 assert(NumElements == 8 && Result.size() == 4 &&
4257 "wasm dot takes i16x8 and produces i32x4");
4258 assert(Ty->isIntegerTy());
4259 int32_t MulVector[8];
4260
4261 for (unsigned I = 0; I < NumElements; ++I) {
4266
4268 }
4269 for (unsigned I = 0; I < Result.size(); I++) {
4270 int64_t IAdd = (int64_t)MulVector[I * 2] + (int64_t)MulVector[I * 2 + 1];
4271 Result[I] = ConstantInt::get(Ty, IAdd);
4272 }
4273
4275 }
4276 default:
4277 break;
4278 }
4279
4281
4282 for (unsigned J = 0, JE = Operands.size(); J != JE; ++J) {
4283
4285 Lane[J] = Operands[J];
4286 continue;
4287 }
4288
4289 Constant *Agg = Operands[J]->getAggregateElement(I);
4290 if (!Agg)
4291 return nullptr;
4292
4293 Lane[J] = Agg;
4294 }
4295
4296
4298 ConstantFoldScalarCall(Name, IntrinsicID, Ty, Lane, TLI, Call);
4299 if (!Folded)
4300 return nullptr;
4302 }
4303
4305}
4306
4307static Constant *ConstantFoldScalableVectorCall(
4311 switch (IntrinsicID) {
4312 case Intrinsic::aarch64_sve_convert_from_svbool: {
4314 if (!Src || !Src->isNullValue())
4315 break;
4316
4318 }
4319 case Intrinsic::get_active_lane_mask: {
4322 if (Op0 && Op1 && Op0->getValue().uge(Op1->getValue()))
4324 break;
4325 }
4326 case Intrinsic::vector_interleave2:
4327 case Intrinsic::vector_interleave3:
4328 case Intrinsic::vector_interleave4:
4329 case Intrinsic::vector_interleave5:
4330 case Intrinsic::vector_interleave6:
4331 case Intrinsic::vector_interleave7:
4332 case Intrinsic::vector_interleave8: {
4333 Constant *SplatVal = Operands[0]->getSplatValue();
4334 if (!SplatVal)
4335 return nullptr;
4336
4338 return nullptr;
4339
4341 }
4342 default:
4343 break;
4344 }
4345
4346
4347
4348
4349
4351 return nullptr;
4352
4357 continue;
4358 }
4361 return nullptr;
4363 }
4364 Constant *Folded = ConstantFoldScalarCall(
4366 if (!Folded)
4367 return nullptr;
4369}
4370
4371static std::pair<Constant *, Constant *>
4372ConstantFoldScalarFrexpCall(Constant *Op, Type *IntTy) {
4375
4377 if (!ConstFP)
4378 return {};
4379
4380 const APFloat &U = ConstFP->getValueAPF();
4381 int FrexpExp;
4383 Constant *Result0 = ConstantFP::get(ConstFP->getType(), FrexpMant);
4384
4385
4386
4390 return {Result0, Result1};
4391}
4392
4393
4399
4400 switch (IntrinsicID) {
4401 case Intrinsic::frexp: {
4404
4408
4409 for (unsigned I = 0, E = FVTy0->getNumElements(); I != E; ++I) {
4410 Constant *Lane = Operands[0]->getAggregateElement(I);
4411 std::tie(Results0[I], Results1[I]) =
4412 ConstantFoldScalarFrexpCall(Lane, Ty1);
4413 if (!Results0[I])
4414 return nullptr;
4415 }
4416
4419 }
4420
4421 auto [Result0, Result1] = ConstantFoldScalarFrexpCall(Operands[0], Ty1);
4422 if (!Result0)
4423 return nullptr;
4425 }
4426 case Intrinsic::sincos: {
4429
4430 auto ConstantFoldScalarSincosCall =
4431 [&](Constant *Op) -> std::pair<Constant *, Constant *> {
4433 ConstantFoldScalarCall(Name, Intrinsic::sin, TyScalar, Op, TLI, Call);
4435 ConstantFoldScalarCall(Name, Intrinsic::cos, TyScalar, Op, TLI, Call);
4436 return std::make_pair(SinResult, CosResult);
4437 };
4438
4442
4444 Constant *Lane = Operands[0]->getAggregateElement(I);
4445 std::tie(SinResults[I], CosResults[I]) =
4446 ConstantFoldScalarSincosCall(Lane);
4447 if (!SinResults[I] || !CosResults[I])
4448 return nullptr;
4449 }
4450
4453 }
4454
4455 auto [SinResult, CosResult] = ConstantFoldScalarSincosCall(Operands[0]);
4456 if (!SinResult || !CosResult)
4457 return nullptr;
4459 }
4460 case Intrinsic::vector_deinterleave2:
4461 case Intrinsic::vector_deinterleave3:
4462 case Intrinsic::vector_deinterleave4:
4463 case Intrinsic::vector_deinterleave5:
4464 case Intrinsic::vector_deinterleave6:
4465 case Intrinsic::vector_deinterleave7:
4466 case Intrinsic::vector_deinterleave8: {
4468 auto *Vec = Operands[0];
4470
4473
4478 }
4479
4480 if (!ResultEC.isFixed())
4481 return nullptr;
4482
4483 unsigned NumElements = ResultEC.getFixedValue();
4486 for (unsigned I = 0; I != NumResults; ++I) {
4487 for (unsigned J = 0; J != NumElements; ++J) {
4489 if (!Elt)
4490 return nullptr;
4492 }
4494 }
4496 }
4497 default:
4498
4499
4500 return ConstantFoldScalarCall(Name, IntrinsicID, StTy, Operands, TLI, Call);
4501 }
4502
4503 return nullptr;
4504}
4505
4506}
4507
4512
4513
4515 return nullptr;
4516 return ConstantFoldIntrinsicCall2(ID, Ty, {LHS, RHS}, Call);
4517}
4518
4522 bool AllowNonDeterministic) {
4523 if (Call->isNoBuiltin())
4524 return nullptr;
4525 if (->hasName())
4526 return nullptr;
4527
4528
4531 if (!TLI)
4532 return nullptr;
4533 LibFunc LibF;
4535 return nullptr;
4536 }
4537
4538
4539
4540 Type *Ty = F->getReturnType();
4541 if (!AllowNonDeterministic && Ty->isFPOrFPVectorTy())
4542 return nullptr;
4543
4546 return ConstantFoldFixedVectorCall(
4547 Name, IID, FVTy, Operands, F->getDataLayout(), TLI, Call);
4548
4550 return ConstantFoldScalableVectorCall(
4551 Name, IID, SVTy, Operands, F->getDataLayout(), TLI, Call);
4552
4554 return ConstantFoldStructCall(Name, IID, StTy, Operands,
4555 F->getDataLayout(), TLI, Call);
4556
4557
4558
4559
4560 return ConstantFoldScalarCall(Name, IID, Ty, Operands, TLI, Call);
4561}
4562
4565
4566
4567 if (Call->isNoBuiltin() || Call->isStrictFP())
4568 return false;
4570 if ()
4571 return false;
4572
4573 LibFunc Func;
4575 return false;
4576
4577 if (Call->arg_size() == 1) {
4579 const APFloat &Op = OpC->getValueAPF();
4580 switch (Func) {
4581 case LibFunc_logl:
4582 case LibFunc_log:
4583 case LibFunc_logf:
4584 case LibFunc_log2l:
4585 case LibFunc_log2:
4586 case LibFunc_log2f:
4587 case LibFunc_log10l:
4588 case LibFunc_log10:
4589 case LibFunc_log10f:
4590 return Op.isNaN() || (.isZero() &&
.isNegative());
4591
4592 case LibFunc_ilogb:
4593 return .isNaN() &&
.isZero() &&
.isInfinity();
4594
4595 case LibFunc_expl:
4596 case LibFunc_exp:
4597 case LibFunc_expf:
4598
4599 if (OpC->getType()->isDoubleTy())
4601 if (OpC->getType()->isFloatTy())
4603 break;
4604
4605 case LibFunc_exp2l:
4606 case LibFunc_exp2:
4607 case LibFunc_exp2f:
4608
4609 if (OpC->getType()->isDoubleTy())
4611 if (OpC->getType()->isFloatTy())
4613 break;
4614
4615 case LibFunc_sinl:
4616 case LibFunc_sin:
4617 case LibFunc_sinf:
4618 case LibFunc_cosl:
4619 case LibFunc_cos:
4620 case LibFunc_cosf:
4621 return .isInfinity();
4622
4623 case LibFunc_tanl:
4624 case LibFunc_tan:
4625 case LibFunc_tanf: {
4626
4627
4628 Type *Ty = OpC->getType();
4629 if (Ty->isDoubleTy() || Ty->isFloatTy() || Ty->isHalfTy())
4630 return ConstantFoldFP(tan, OpC->getValueAPF(), Ty) != nullptr;
4631 break;
4632 }
4633
4634 case LibFunc_atan:
4635 case LibFunc_atanf:
4636 case LibFunc_atanl:
4637
4638 return true;
4639
4640 case LibFunc_asinl:
4641 case LibFunc_asin:
4642 case LibFunc_asinf:
4643 case LibFunc_acosl:
4644 case LibFunc_acos:
4645 case LibFunc_acosf:
4648
4649 case LibFunc_sinh:
4650 case LibFunc_cosh:
4651 case LibFunc_sinhf:
4652 case LibFunc_coshf:
4653 case LibFunc_sinhl:
4654 case LibFunc_coshl:
4655
4656 if (OpC->getType()->isDoubleTy())
4658 if (OpC->getType()->isFloatTy())
4660 break;
4661
4662 case LibFunc_sqrtl:
4663 case LibFunc_sqrt:
4664 case LibFunc_sqrtf:
4665 return Op.isNaN() || Op.isZero() || .isNegative();
4666
4667
4668
4669 default:
4670 break;
4671 }
4672 }
4673 }
4674
4675 if (Call->arg_size() == 2) {
4678 if (Op0C && Op1C) {
4681
4682 switch (Func) {
4683 case LibFunc_powl:
4684 case LibFunc_pow:
4685 case LibFunc_powf: {
4686
4687
4689 if (Ty->isDoubleTy() || Ty->isFloatTy() || Ty->isHalfTy()) {
4690 if (Ty == Op1C->getType())
4691 return ConstantFoldBinaryFP(pow, Op0, Op1, Ty) != nullptr;
4692 }
4693 break;
4694 }
4695
4696 case LibFunc_fmodl:
4697 case LibFunc_fmod:
4698 case LibFunc_fmodf:
4699 case LibFunc_remainderl:
4700 case LibFunc_remainder:
4701 case LibFunc_remainderf:
4702 return Op0.isNaN() || Op1.isNaN() ||
4704
4705 case LibFunc_atan2:
4706 case LibFunc_atan2f:
4707 case LibFunc_atan2l:
4708
4709
4710
4711
4713
4714 default:
4715 break;
4716 }
4717 }
4718 }
4719
4720 return false;
4721}
4722
4726 switch (CastOp) {
4727 case Instruction::BitCast:
4728
4730 case Instruction::Trunc: {
4732 if (Flags) {
4733
4734 Flags->NUW = true;
4735
4736 auto *SExtC =
4738 Flags->NSW = ZExtC == SExtC;
4739 }
4740 return ZExtC;
4741 }
4742 case Instruction::SExt:
4743 case Instruction::ZExt: {
4746
4747 if (!CastInvC || CastInvC != C)
4748 return nullptr;
4749 if (Flags && CastOp == Instruction::ZExt) {
4750 auto *SExtInvC =
4752
4753 Flags->NNeg = CastInvC == SExtInvC;
4754 }
4755 return InvC;
4756 }
4757 default:
4758 return nullptr;
4759 }
4760}
4761
4767
4773
4774void TargetFolder::anchor() {}
assert(UImm &&(UImm !=~static_cast< T >(0)) &&"Invalid immediate!")
This file declares a class to represent arbitrary precision floating point values and provide a varie...
This file implements a class to represent arbitrary precision integral constant values and operations...
This file implements the APSInt class, which is a simple class that represents an arbitrary sized int...
MachineBasicBlock MachineBasicBlock::iterator DebugLoc DL
Function Alias Analysis Results
static GCRegistry::Add< ErlangGC > A("erlang", "erlang-compatible garbage collector")
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
static GCRegistry::Add< OcamlGC > B("ocaml", "ocaml 3.10-compatible GC")
static Constant * FoldBitCast(Constant *V, Type *DestTy)
static ConstantFP * flushDenormalConstant(Type *Ty, const APFloat &APF, DenormalMode::DenormalModeKind Mode)
Definition ConstantFolding.cpp:1331
Constant * getConstantAtOffset(Constant *Base, APInt Offset, const DataLayout &DL)
If this Offset points exactly to the start of an aggregate element, return that element,...
Definition ConstantFolding.cpp:688
static cl::opt< bool > DisableFPCallFolding("disable-fp-call-folding", cl::desc("Disable constant-folding of FP intrinsics and libcalls."), cl::init(false), cl::Hidden)
static ConstantFP * flushDenormalConstantFP(ConstantFP *CFP, const Instruction *Inst, bool IsOutput)
Definition ConstantFolding.cpp:1360
static DenormalMode getInstrDenormalMode(const Instruction *CtxI, Type *Ty)
Return the denormal mode that can be assumed when executing a floating point operation at CtxI.
Definition ConstantFolding.cpp:1354
This file contains the declarations for the subclasses of Constant, which represent the different fla...
This file defines the DenseMap class.
amode Optimize addressing mode
const AbstractManglingParser< Derived, Alloc >::OperatorInfo AbstractManglingParser< Derived, Alloc >::Ops[]
static bool InRange(int64_t Value, unsigned short Shift, int LBound, int HBound)
This file contains the definitions of the enumerations and flags associated with NVVM Intrinsics,...
const SmallVectorImpl< MachineOperand > & Cond
static cl::opt< RegAllocEvictionAdvisorAnalysisLegacy::AdvisorMode > Mode("regalloc-enable-advisor", cl::Hidden, cl::init(RegAllocEvictionAdvisorAnalysisLegacy::AdvisorMode::Default), cl::desc("Enable regalloc advisor mode"), cl::values(clEnumValN(RegAllocEvictionAdvisorAnalysisLegacy::AdvisorMode::Default, "default", "Default"), clEnumValN(RegAllocEvictionAdvisorAnalysisLegacy::AdvisorMode::Release, "release", "precompiled"), clEnumValN(RegAllocEvictionAdvisorAnalysisLegacy::AdvisorMode::Development, "development", "for training")))
This file defines the SmallVector class.
static TableGen::Emitter::OptClass< SkeletonEmitter > X("gen-skeleton-class", "Generate example skeleton class")
static SymbolRef::Type getType(const Symbol *Sym)
The Input class is used to parse a yaml document into in-memory structs and vectors.
static constexpr roundingMode rmTowardZero
llvm::RoundingMode roundingMode
IEEE-754R 4.3: Rounding-direction attributes.
static const fltSemantics & IEEEdouble()
static constexpr roundingMode rmTowardNegative
static constexpr roundingMode rmNearestTiesToEven
static constexpr roundingMode rmTowardPositive
static const fltSemantics & IEEEhalf()
static constexpr roundingMode rmNearestTiesToAway
opStatus
IEEE-754R 7: Default exception handling.
static APFloat getQNaN(const fltSemantics &Sem, bool Negative=false, const APInt *payload=nullptr)
Factory for QNaN values.
opStatus divide(const APFloat &RHS, roundingMode RM)
void copySign(const APFloat &RHS)
LLVM_ABI opStatus convert(const fltSemantics &ToSemantics, roundingMode RM, bool *losesInfo)
opStatus subtract(const APFloat &RHS, roundingMode RM)
LLVM_ABI double convertToDouble() const
Converts this APFloat to host double value.
bool isPosInfinity() const
opStatus add(const APFloat &RHS, roundingMode RM)
const fltSemantics & getSemantics() const
static APFloat getOne(const fltSemantics &Sem, bool Negative=false)
Factory for Positive and Negative One.
opStatus multiply(const APFloat &RHS, roundingMode RM)
LLVM_ABI float convertToFloat() const
Converts this APFloat to host float value.
opStatus fusedMultiplyAdd(const APFloat &Multiplicand, const APFloat &Addend, roundingMode RM)
APInt bitcastToAPInt() const
opStatus convertToInteger(MutableArrayRef< integerPart > Input, unsigned int Width, bool IsSigned, roundingMode RM, bool *IsExact) const
opStatus mod(const APFloat &RHS)
bool isNegInfinity() const
opStatus roundToIntegral(roundingMode RM)
static APFloat getZero(const fltSemantics &Sem, bool Negative=false)
Factory for Positive and Negative Zero.
Class for arbitrary precision integers.
LLVM_ABI APInt umul_ov(const APInt &RHS, bool &Overflow) const
LLVM_ABI APInt usub_sat(const APInt &RHS) const
bool isMinSignedValue() const
Determine if this is the smallest signed value.
uint64_t getZExtValue() const
Get zero extended value.
LLVM_ABI uint64_t extractBitsAsZExtValue(unsigned numBits, unsigned bitPosition) const
LLVM_ABI APInt zextOrTrunc(unsigned width) const
Zero extend or truncate to width.
LLVM_ABI APInt trunc(unsigned width) const
Truncate to new width.
APInt abs() const
Get the absolute value.
LLVM_ABI APInt sadd_sat(const APInt &RHS) const
bool sgt(const APInt &RHS) const
Signed greater than comparison.
LLVM_ABI APInt usub_ov(const APInt &RHS, bool &Overflow) const
bool ugt(const APInt &RHS) const
Unsigned greater than comparison.
bool isZero() const
Determine if this value is zero, i.e. all bits are clear.
LLVM_ABI APInt urem(const APInt &RHS) const
Unsigned remainder operation.
unsigned getBitWidth() const
Return the number of bits in the APInt.
bool ult(const APInt &RHS) const
Unsigned less than comparison.
static APInt getSignedMaxValue(unsigned numBits)
Gets maximum signed value of APInt for a specific bit width.
LLVM_ABI APInt sadd_ov(const APInt &RHS, bool &Overflow) const
LLVM_ABI APInt uadd_ov(const APInt &RHS, bool &Overflow) const
unsigned countr_zero() const
Count the number of trailing zero bits.
unsigned countl_zero() const
The APInt version of std::countl_zero.
static APInt getSignedMinValue(unsigned numBits)
Gets minimum signed value of APInt for a specific bit width.
LLVM_ABI APInt sextOrTrunc(unsigned width) const
Sign extend or truncate to width.
LLVM_ABI APInt uadd_sat(const APInt &RHS) const
APInt ashr(unsigned ShiftAmt) const
Arithmetic right-shift function.
LLVM_ABI APInt smul_ov(const APInt &RHS, bool &Overflow) const
LLVM_ABI APInt sext(unsigned width) const
Sign extend to a new width.
APInt shl(unsigned shiftAmt) const
Left-shift function.
bool slt(const APInt &RHS) const
Signed less than comparison.
LLVM_ABI APInt extractBits(unsigned numBits, unsigned bitPosition) const
Return an APInt with the extracted bits [bitPosition,bitPosition+numBits).
LLVM_ABI APInt ssub_ov(const APInt &RHS, bool &Overflow) const
bool isOne() const
Determine if this is a value of 1.
APInt lshr(unsigned shiftAmt) const
Logical right-shift function.
LLVM_ABI APInt ssub_sat(const APInt &RHS) const
An arbitrary precision integer that knows its signedness.
This class represents an incoming formal argument to a Function.
ArrayRef - Represent a constant reference to an array (0 or more elements consecutively in memory),...
size_t size() const
size - Get the array size.
Base class for all callable instructions (InvokeInst and CallInst) Holds everything related to callin...
static LLVM_ABI Instruction::CastOps getCastOpcode(const Value *Val, bool SrcIsSigned, Type *Ty, bool DstIsSigned)
Returns the opcode necessary to cast Val into Ty using usual casting rules.
static LLVM_ABI unsigned isEliminableCastPair(Instruction::CastOps firstOpcode, Instruction::CastOps secondOpcode, Type *SrcTy, Type *MidTy, Type *DstTy, const DataLayout *DL)
Determine how a pair of casts can be eliminated, if they can be at all.
static LLVM_ABI bool castIsValid(Instruction::CastOps op, Type *SrcTy, Type *DstTy)
This method can be used to determine if a cast from SrcTy to DstTy using Opcode op is valid or not.
Predicate
This enumeration lists the possible predicates for CmpInst subclasses.
Predicate getSwappedPredicate() const
For example, EQ->EQ, SLE->SGE, ULT->UGT, OEQ->OEQ, ULE->UGE, OLT->OGT, etc.
static bool isFPPredicate(Predicate P)
static Constant * get(LLVMContext &Context, ArrayRef< ElementTy > Elts)
get() constructor - Return a constant with array type with an element count and element type matching...
static LLVM_ABI Constant * getIntToPtr(Constant *C, Type *Ty, bool OnlyIfReduced=false)
static LLVM_ABI Constant * getExtractElement(Constant *Vec, Constant *Idx, Type *OnlyIfReducedTy=nullptr)
static LLVM_ABI bool isDesirableCastOp(unsigned Opcode)
Whether creating a constant expression for this cast is desirable.
static LLVM_ABI Constant * getCast(unsigned ops, Constant *C, Type *Ty, bool OnlyIfReduced=false)
Convenience function for getting a Cast operation.
static LLVM_ABI Constant * getSub(Constant *C1, Constant *C2, bool HasNUW=false, bool HasNSW=false)
static LLVM_ABI Constant * getInsertElement(Constant *Vec, Constant *Elt, Constant *Idx, Type *OnlyIfReducedTy=nullptr)
static LLVM_ABI Constant * getShuffleVector(Constant *V1, Constant *V2, ArrayRef< int > Mask, Type *OnlyIfReducedTy=nullptr)
static bool isSupportedGetElementPtr(const Type *SrcElemTy)
Whether creating a constant expression for this getelementptr type is supported.
static LLVM_ABI Constant * get(unsigned Opcode, Constant *C1, Constant *C2, unsigned Flags=0, Type *OnlyIfReducedTy=nullptr)
get - Return a binary or shift operator constant expression, folding if possible.
static LLVM_ABI bool isDesirableBinOp(unsigned Opcode)
Whether creating a constant expression for this binary operator is desirable.
static Constant * getGetElementPtr(Type *Ty, Constant *C, ArrayRef< Constant * > IdxList, GEPNoWrapFlags NW=GEPNoWrapFlags::none(), std::optional< ConstantRange > InRange=std::nullopt, Type *OnlyIfReducedTy=nullptr)
Getelementptr form.
static LLVM_ABI Constant * getBitCast(Constant *C, Type *Ty, bool OnlyIfReduced=false)
static LLVM_ABI Constant * getTrunc(Constant *C, Type *Ty, bool OnlyIfReduced=false)
ConstantFP - Floating Point Values [float, double].
const APFloat & getValueAPF() const
static LLVM_ABI Constant * getZero(Type *Ty, bool Negative=false)
This is the shared class of boolean and integer constants.
static LLVM_ABI ConstantInt * getTrue(LLVMContext &Context)
static ConstantInt * getSigned(IntegerType *Ty, int64_t V)
Return a ConstantInt with the specified value for the specified type.
static LLVM_ABI ConstantInt * getFalse(LLVMContext &Context)
int64_t getSExtValue() const
Return the constant as a 64-bit integer value after it has been sign extended as appropriate for the ...
static LLVM_ABI ConstantInt * getBool(LLVMContext &Context, bool V)
static LLVM_ABI Constant * get(StructType *T, ArrayRef< Constant * > V)
static LLVM_ABI Constant * getSplat(ElementCount EC, Constant *Elt)
Return a ConstantVector with the specified constant in each element.
static LLVM_ABI Constant * get(ArrayRef< Constant * > V)
This is an important base class in LLVM.
LLVM_ABI Constant * getSplatValue(bool AllowPoison=false) const
If all elements of the vector constant have the same value, return that value.
static LLVM_ABI Constant * getAllOnesValue(Type *Ty)
static LLVM_ABI Constant * getNullValue(Type *Ty)
Constructor to create a '0' constant of arbitrary type.
LLVM_ABI Constant * getAggregateElement(unsigned Elt) const
For aggregates (struct/array/vector) return the constant that corresponds to the specified element if...
LLVM_ABI bool isZeroValue() const
Return true if the value is negative zero or null value.
LLVM_ABI bool isNullValue() const
Return true if this is the value that would be returned by getNullValue.
Constrained floating point compare intrinsics.
This is the common base class for constrained floating point intrinsics.
LLVM_ABI std::optional< fp::ExceptionBehavior > getExceptionBehavior() const
LLVM_ABI std::optional< RoundingMode > getRoundingMode() const
Wrapper for a function that represents a value that functionally represents the original function.
A parsed version of the target data layout string in and methods for querying it.
iterator find(const_arg_type_t< KeyT > Val)
std::pair< iterator, bool > insert(const std::pair< KeyT, ValueT > &KV)
static LLVM_ABI bool compare(const APFloat &LHS, const APFloat &RHS, FCmpInst::Predicate Pred)
Return result of LHS Pred RHS comparison.
This provides a helper for copying FMF from an instruction or setting specified flags.
Class to represent fixed width SIMD vectors.
unsigned getNumElements() const
static LLVM_ABI FixedVectorType * get(Type *ElementType, unsigned NumElts)
DenormalMode getDenormalMode(const fltSemantics &FPType) const
Returns the denormal handling type for the default rounding mode of the function.
Represents flags for the getelementptr instruction/expression.
static GEPNoWrapFlags inBounds()
GEPNoWrapFlags withoutNoUnsignedSignedWrap() const
static GEPNoWrapFlags noUnsignedWrap()
bool hasNoUnsignedSignedWrap() const
static LLVM_ABI Type * getIndexedType(Type *Ty, ArrayRef< Value * > IdxList)
Returns the result type of a getelementptr with the given source element type and indexes.
PointerType * getType() const
Global values are always pointers.
LLVM_ABI const DataLayout & getDataLayout() const
Get the data layout of the module this global belongs to.
const Constant * getInitializer() const
getInitializer - Return the initializer for this global variable.
bool isConstant() const
If the value is a global constant, its value is immutable throughout the runtime execution of the pro...
bool hasDefinitiveInitializer() const
hasDefinitiveInitializer - Whether the global variable has an initializer, and any other instances of...
static LLVM_ABI bool compare(const APInt &LHS, const APInt &RHS, ICmpInst::Predicate Pred)
Return result of LHS Pred RHS comparison.
Predicate getSignedPredicate() const
For example, EQ->EQ, SLE->SLE, UGT->SGT, etc.
bool isEquality() const
Return true if this predicate is either EQ or NE.
LLVM_ABI const Function * getFunction() const
Return the function this instruction belongs to.
static LLVM_ABI IntegerType * get(LLVMContext &C, unsigned NumBits)
This static method is the primary way of constructing an IntegerType.
This is an important class for using LLVM in a threaded context.
static APInt getSaturationPoint(Intrinsic::ID ID, unsigned numBits)
Min/max intrinsics are monotonic, they operate on a fixed-bitwidth values, so there is a certain thre...
static ICmpInst::Predicate getPredicate(Intrinsic::ID ID)
Returns the comparison predicate underlying the intrinsic.
static LLVM_ABI PoisonValue * get(Type *T)
Static factory methods - Return an 'poison' object of the specified type.
Class to represent scalable SIMD vectors.
void push_back(const T &Elt)
pointer data()
Return a pointer to the vector's buffer, even if empty().
This is a 'vector' (really, a variable-sized array), optimized for the case when the array is small.
StringRef - Represent a constant reference to a string, i.e.
Used to lazily calculate structure layout information for a target machine, based on the DataLayout s...
LLVM_ABI unsigned getElementContainingOffset(uint64_t FixedOffset) const
Given a valid byte offset into the structure, returns the structure index that contains it.
TypeSize getElementOffset(unsigned Idx) const
Class to represent struct types.
unsigned getNumElements() const
Random access to the elements.
Provides information about what library functions are available for the current target.
bool has(LibFunc F) const
Tests whether a library function is available.
bool getLibFunc(StringRef funcName, LibFunc &F) const
Searches for a particular function name.
The instances of the Type class are immutable: once they are created, they are never changed.
static LLVM_ABI IntegerType * getInt64Ty(LLVMContext &C)
bool isVectorTy() const
True if this is an instance of VectorType.
static LLVM_ABI IntegerType * getInt32Ty(LLVMContext &C)
bool isPointerTy() const
True if this is an instance of PointerType.
LLVM_ABI unsigned getPointerAddressSpace() const
Get the address space of this pointer or pointer vector type.
@ HalfTyID
16-bit floating point type
@ FloatTyID
32-bit floating point type
@ DoubleTyID
64-bit floating point type
static LLVM_ABI IntegerType * getInt8Ty(LLVMContext &C)
Type * getScalarType() const
If this is a vector type, return the element type, otherwise return 'this'.
LLVM_ABI TypeSize getPrimitiveSizeInBits() const LLVM_READONLY
Return the basic size of this type if it is a primitive type.
static LLVM_ABI IntegerType * getInt16Ty(LLVMContext &C)
bool isSized(SmallPtrSetImpl< Type * > *Visited=nullptr) const
Return true if it makes sense to take the size of this type.
LLVMContext & getContext() const
Return the LLVMContext in which this type was uniqued.
LLVM_ABI unsigned getScalarSizeInBits() const LLVM_READONLY
If this is a vector type, return the getPrimitiveSizeInBits value for the element type.
static LLVM_ABI IntegerType * getInt1Ty(LLVMContext &C)
bool isFloatingPointTy() const
Return true if this is one of the floating-point types.
bool isPtrOrPtrVectorTy() const
Return true if this is a pointer type or a vector of pointer types.
bool isX86_AMXTy() const
Return true if this is X86 AMX.
bool isIntegerTy() const
True if this is an instance of IntegerType.
static LLVM_ABI IntegerType * getIntNTy(LLVMContext &C, unsigned N)
Type * getContainedType(unsigned i) const
This method is used to implement the type iterator (defined at the end of the file).
LLVM_ABI const fltSemantics & getFltSemantics() const
static LLVM_ABI UndefValue * get(Type *T)
Static factory methods - Return an 'undef' object of the specified type.
A Use represents the edge between a Value definition and its users.
LLVM Value Representation.
Type * getType() const
All values are typed, get the type of this value.
LLVM_ABI const Value * stripAndAccumulateConstantOffsets(const DataLayout &DL, APInt &Offset, bool AllowNonInbounds, bool AllowInvariantGroup=false, function_ref< bool(Value &Value, APInt &Offset)> ExternalAnalysis=nullptr, bool LookThroughIntToPtr=false) const
Accumulate the constant offset this value has compared to a base pointer.
LLVM_ABI LLVMContext & getContext() const
All values hold a context through their type.
LLVM_ABI uint64_t getPointerDereferenceableBytes(const DataLayout &DL, bool &CanBeNull, bool &CanBeFreed) const
Returns the number of bytes known to be dereferenceable for the pointer value.
Base class of all SIMD vector types.
ElementCount getElementCount() const
Return an ElementCount instance to represent the (possibly scalable) number of elements in the vector...
Type * getElementType() const
constexpr ScalarTy getFixedValue() const
constexpr bool isScalable() const
Returns whether the quantity is scaled by a runtime quantity (vscale).
constexpr bool isFixed() const
Returns true if the quantity is not scaled by vscale.
constexpr LeafTy divideCoefficientBy(ScalarTy RHS) const
We do not provide the '/' operator here because division for polynomial types does not work in the sa...
static constexpr bool isKnownGE(const FixedOrScalableQuantity &LHS, const FixedOrScalableQuantity &RHS)
const ParentTy * getParent() const
#define llvm_unreachable(msg)
Marks that the current location is not supposed to be reachable.
const APInt & smin(const APInt &A, const APInt &B)
Determine the smaller of two APInts considered to be signed.
const APInt & smax(const APInt &A, const APInt &B)
Determine the larger of two APInts considered to be signed.
const APInt & umin(const APInt &A, const APInt &B)
Determine the smaller of two APInts considered to be unsigned.
const APInt & umax(const APInt &A, const APInt &B)
Determine the larger of two APInts considered to be unsigned.
constexpr std::underlying_type_t< E > Mask()
Get a bitmask with 1s in all places up to the high-order bit of E's largest value.
unsigned ID
LLVM IR allows to use arbitrary numbers as calling convention identifiers.
@ C
The default llvm calling convention, compatible with C.
@ CE
Windows NT (Windows on ARM)
initializer< Ty > init(const Ty &Val)
@ ebStrict
This corresponds to "fpexcept.strict".
@ ebIgnore
This corresponds to "fpexcept.ignore".
APFloat::roundingMode GetFMARoundingMode(Intrinsic::ID IntrinsicID)
DenormalMode GetNVVMDenormMode(bool ShouldFTZ)
bool FPToIntegerIntrinsicNaNZero(Intrinsic::ID IntrinsicID)
APFloat::roundingMode GetFDivRoundingMode(Intrinsic::ID IntrinsicID)
bool FPToIntegerIntrinsicResultIsSigned(Intrinsic::ID IntrinsicID)
APFloat::roundingMode GetFPToIntegerRoundingMode(Intrinsic::ID IntrinsicID)
bool RCPShouldFTZ(Intrinsic::ID IntrinsicID)
bool FPToIntegerIntrinsicShouldFTZ(Intrinsic::ID IntrinsicID)
bool FDivShouldFTZ(Intrinsic::ID IntrinsicID)
bool FAddShouldFTZ(Intrinsic::ID IntrinsicID)
bool FMinFMaxIsXorSignAbs(Intrinsic::ID IntrinsicID)
APFloat::roundingMode GetFMulRoundingMode(Intrinsic::ID IntrinsicID)
bool UnaryMathIntrinsicShouldFTZ(Intrinsic::ID IntrinsicID)
bool FMinFMaxShouldFTZ(Intrinsic::ID IntrinsicID)
APFloat::roundingMode GetFAddRoundingMode(Intrinsic::ID IntrinsicID)
bool FMAShouldFTZ(Intrinsic::ID IntrinsicID)
bool FMulShouldFTZ(Intrinsic::ID IntrinsicID)
APFloat::roundingMode GetRCPRoundingMode(Intrinsic::ID IntrinsicID)
bool FMinFMaxPropagatesNaNs(Intrinsic::ID IntrinsicID)
NodeAddr< FuncNode * > Func
LLVM_ABI std::error_code status(const Twine &path, file_status &result, bool follow=true)
Get file status as if by POSIX stat().
This is an optimization pass for GlobalISel generic memory operations.
auto drop_begin(T &&RangeOrContainer, size_t N=1)
Return a range covering RangeOrContainer with the first N elements excluded.
LLVM_ABI Constant * ConstantFoldBinaryIntrinsic(Intrinsic::ID ID, Constant *LHS, Constant *RHS, Type *Ty, Instruction *FMFSource)
Definition ConstantFolding.cpp:4508
bool all_of(R &&range, UnaryPredicate P)
Provide wrappers to std::all_of which take ranges instead of having to pass begin/end explicitly.
LLVM_ABI Constant * ConstantFoldLoadThroughBitcast(Constant *C, Type *DestTy, const DataLayout &DL)
ConstantFoldLoadThroughBitcast - try to cast constant to destination type returning null if unsuccess...
Definition ConstantFolding.cpp:358
static double log2(double V)
LLVM_ABI Constant * ConstantFoldSelectInstruction(Constant *Cond, Constant *V1, Constant *V2)
Attempt to constant fold a select instruction with the specified operands.
LLVM_ABI Constant * ConstantFoldFPInstOperands(unsigned Opcode, Constant *LHS, Constant *RHS, const DataLayout &DL, const Instruction *I, bool AllowNonDeterministic=true)
Attempt to constant fold a floating point binary operation with the specified operands,...
Definition ConstantFolding.cpp:1440
auto enumerate(FirstRange &&First, RestRanges &&...Rest)
Given two or more input ranges, returns a new range whose values are tuples (A, B,...
LLVM_ABI bool canConstantFoldCallTo(const CallBase *Call, const Function *F)
canConstantFoldCallTo - Return true if its even possible to fold a call to the specified function.
Definition ConstantFolding.cpp:1600
unsigned getPointerAddressSpace(const Type *T)
decltype(auto) dyn_cast(const From &Val)
dyn_cast - Return the argument parameter cast to the specified type.
LLVM_ABI Constant * ConstantFoldInstruction(const Instruction *I, const DataLayout &DL, const TargetLibraryInfo *TLI=nullptr)
ConstantFoldInstruction - Try to constant fold the specified instruction.
Definition ConstantFolding.cpp:1135
APFloat abs(APFloat X)
Returns the absolute value of the argument.
LLVM_ABI Constant * ConstantFoldCompareInstruction(CmpInst::Predicate Predicate, Constant *C1, Constant *C2)
LLVM_ABI Constant * ConstantFoldUnaryInstruction(unsigned Opcode, Constant *V)
LLVM_ABI bool IsConstantOffsetFromGlobal(Constant *C, GlobalValue *&GV, APInt &Offset, const DataLayout &DL, DSOLocalEquivalent **DSOEquiv=nullptr)
If this constant is a constant offset from a global, return the global and the constant.
Definition ConstantFolding.cpp:304
LLVM_ABI bool isMathLibCallNoop(const CallBase *Call, const TargetLibraryInfo *TLI)
Check whether the given call has no side-effects.
Definition ConstantFolding.cpp:4563
LLVM_ABI Constant * ReadByteArrayFromGlobal(const GlobalVariable *GV, uint64_t Offset)
Definition ConstantFolding.cpp:658
auto dyn_cast_if_present(const Y &Val)
dyn_cast_if_present - Functionally identical to dyn_cast, except that a null (or none in the case ...
LLVM_READONLY APFloat maximum(const APFloat &A, const APFloat &B)
Implements IEEE 754-2019 maximum semantics.
LLVM_ABI Constant * ConstantFoldCompareInstOperands(unsigned Predicate, Constant *LHS, Constant *RHS, const DataLayout &DL, const TargetLibraryInfo *TLI=nullptr, const Instruction *I=nullptr)
Attempt to constant fold a compare instruction (icmp/fcmp) with the specified operands.
Definition ConstantFolding.cpp:1199
int ilogb(const APFloat &Arg)
Returns the exponent of the internal representation of the APFloat.
bool isa_and_nonnull(const Y &Val)
LLVM_ABI Constant * ConstantFoldCall(const CallBase *Call, Function *F, ArrayRef< Constant * > Operands, const TargetLibraryInfo *TLI=nullptr, bool AllowNonDeterministic=true)
ConstantFoldCall - Attempt to constant fold a call to the specified function with the specified argum...
Definition ConstantFolding.cpp:4519
APFloat frexp(const APFloat &X, int &Exp, APFloat::roundingMode RM)
Equivalent of C standard library function.
LLVM_ABI Constant * ConstantFoldExtractValueInstruction(Constant *Agg, ArrayRef< unsigned > Idxs)
Attempt to constant fold an extractvalue instruction with the specified operands and indices.
LLVM_ABI Constant * ConstantFoldConstant(const Constant *C, const DataLayout &DL, const TargetLibraryInfo *TLI=nullptr)
ConstantFoldConstant - Fold the constant using the specified DataLayout.
Definition ConstantFolding.cpp:1184
auto dyn_cast_or_null(const Y &Val)
bool any_of(R &&range, UnaryPredicate P)
Provide wrappers to std::any_of which take ranges instead of having to pass begin/end explicitly.
LLVM_READONLY APFloat maxnum(const APFloat &A, const APFloat &B)
Implements IEEE-754 2008 maxNum semantics.
LLVM_ABI Constant * ConstantFoldLoadFromUniformValue(Constant *C, Type *Ty, const DataLayout &DL)
If C is a uniform value where all bits are the same (either all zero, all ones, all undef or all pois...
Definition ConstantFolding.cpp:773
LLVM_ABI Constant * ConstantFoldUnaryOpOperand(unsigned Opcode, Constant *Op, const DataLayout &DL)
Attempt to constant fold a unary operation with the specified operand.
Definition ConstantFolding.cpp:1311
LLVM_ABI Constant * FlushFPConstant(Constant *Operand, const Instruction *I, bool IsOutput)
Attempt to flush float point constant according to denormal mode set in the instruction's parent func...
Definition ConstantFolding.cpp:1372
LLVM_ABI Constant * getLosslessUnsignedTrunc(Constant *C, Type *DestTy, const DataLayout &DL, PreservedCastFlags *Flags=nullptr)
Definition ConstantFolding.cpp:4762
decltype(auto) get(const PointerIntPair< PointerTy, IntBits, IntType, PtrTraits, Info > &Pair)
LLVM_READONLY APFloat minimumnum(const APFloat &A, const APFloat &B)
Implements IEEE 754-2019 minimumNumber semantics.
FPClassTest
Floating-point class tests, supported by 'is_fpclass' intrinsic.
APFloat scalbn(APFloat X, int Exp, APFloat::roundingMode RM)
Returns: X * 2^Exp for integral exponents.
LLVM_ABI void computeKnownBits(const Value *V, KnownBits &Known, const DataLayout &DL, AssumptionCache *AC=nullptr, const Instruction *CxtI=nullptr, const DominatorTree *DT=nullptr, bool UseInstrInfo=true, unsigned Depth=0)
Determine which bits of V are known to be either zero or one and return them in the KnownZero/KnownOn...
LLVM_ABI bool NullPointerIsDefined(const Function *F, unsigned AS=0)
Check whether null pointer dereferencing is considered undefined behavior for a given function or an ...
LLVM_ABI Constant * getLosslessSignedTrunc(Constant *C, Type *DestTy, const DataLayout &DL, PreservedCastFlags *Flags=nullptr)
Definition ConstantFolding.cpp:4768
LLVM_ABI Constant * ConstantFoldCastOperand(unsigned Opcode, Constant *C, Type *DestTy, const DataLayout &DL)
Attempt to constant fold a cast with the specified operand.
Definition ConstantFolding.cpp:1483
LLVM_ABI Constant * ConstantFoldLoadFromConst(Constant *C, Type *Ty, const APInt &Offset, const DataLayout &DL)
Extract value of C at the given Offset reinterpreted as Ty.
Definition ConstantFolding.cpp:714
bool isa(const From &Val)
isa - Return true if the parameter to the template is an instance of one of the template type argu...
LLVM_ABI bool intrinsicPropagatesPoison(Intrinsic::ID IID)
Return whether this intrinsic propagates poison for all operands.
LLVM_ABI Constant * ConstantFoldBinaryOpOperands(unsigned Opcode, Constant *LHS, Constant *RHS, const DataLayout &DL)
Attempt to constant fold a binary operation with the specified operands.
Definition ConstantFolding.cpp:1318
MutableArrayRef(T &OneElt) -> MutableArrayRef< T >
LLVM_READONLY APFloat minnum(const APFloat &A, const APFloat &B)
Implements IEEE-754 2008 minNum semantics.
@ Sub
Subtraction of integers.
LLVM_ABI bool isVectorIntrinsicWithScalarOpAtArg(Intrinsic::ID ID, unsigned ScalarOpdIdx, const TargetTransformInfo *TTI)
Identifies if the vector form of the intrinsic has a scalar operand.
DWARFExpression::Operation Op
RoundingMode
Rounding mode.
@ NearestTiesToEven
roundTiesToEven.
@ Dynamic
Denotes mode unknown at compile time.
LLVM_ABI bool isGuaranteedNotToBeUndefOrPoison(const Value *V, AssumptionCache *AC=nullptr, const Instruction *CtxI=nullptr, const DominatorTree *DT=nullptr, unsigned Depth=0)
Return true if this function can prove that V does not have undef bits and is never poison.
constexpr unsigned BitWidth
LLVM_ABI Constant * getLosslessInvCast(Constant *C, Type *InvCastTo, unsigned CastOp, const DataLayout &DL, PreservedCastFlags *Flags=nullptr)
Try to cast C to InvC losslessly, satisfying CastOp(InvC) equals C, or CastOp(InvC) is a refined valu...
Definition ConstantFolding.cpp:4723
decltype(auto) cast(const From &Val)
cast - Return the argument parameter cast to the specified type.
bool all_equal(std::initializer_list< T > Values)
Returns true if all Values in the initializer lists are equal or the list.
LLVM_ABI Constant * ConstantFoldCastInstruction(unsigned opcode, Constant *V, Type *DestTy)
LLVM_ABI Constant * ConstantFoldInsertValueInstruction(Constant *Agg, Constant *Val, ArrayRef< unsigned > Idxs)
Attempt to constant fold an insertvalue instruction with the specified operands and indices.
LLVM_ABI Constant * ConstantFoldLoadFromConstPtr(Constant *C, Type *Ty, APInt Offset, const DataLayout &DL)
Return the value that a load from C with offset Offset would produce if it is constant and determinab...
Definition ConstantFolding.cpp:745
LLVM_ABI Constant * ConstantFoldInstOperands(const Instruction *I, ArrayRef< Constant * > Ops, const DataLayout &DL, const TargetLibraryInfo *TLI=nullptr, bool AllowNonDeterministic=true)
ConstantFoldInstOperands - Attempt to constant fold an instruction with the specified operands.
Definition ConstantFolding.cpp:1190
LLVM_READONLY APFloat minimum(const APFloat &A, const APFloat &B)
Implements IEEE 754-2019 minimum semantics.
LLVM_READONLY APFloat maximumnum(const APFloat &A, const APFloat &B)
Implements IEEE 754-2019 maximumNumber semantics.
LLVM_ABI const Value * getUnderlyingObject(const Value *V, unsigned MaxLookup=MaxLookupSearchDepth)
This method strips off any GEP address adjustments, pointer casts or llvm.threadlocal....
LLVM_ABI Constant * ConstantFoldIntegerCast(Constant *C, Type *DestTy, bool IsSigned, const DataLayout &DL)
Constant fold a zext, sext or trunc, depending on IsSigned and whether the DestTy is wider or narrowe...
Definition ConstantFolding.cpp:1584
LLVM_ABI bool isTriviallyVectorizable(Intrinsic::ID ID)
Identify if the intrinsic is trivially vectorizable.
constexpr detail::IsaCheckPredicate< Types... > IsaPred
Function object wrapper for the llvm::isa type check.
LLVM_ABI Constant * ConstantFoldBinaryInstruction(unsigned Opcode, Constant *V1, Constant *V2)
Represent subnormal handling kind for floating point instruction inputs and outputs.
DenormalModeKind Input
Denormal treatment kind for floating point instruction inputs in the default floating-point environme...
DenormalModeKind
Represent handled modes for denormal (aka subnormal) modes in the floating point environment.
@ PreserveSign
The sign of a flushed-to-zero number is preserved in the sign of 0.
@ PositiveZero
Denormals are flushed to positive zero.
@ Dynamic
Denormals have unknown treatment.
@ IEEE
IEEE-754 denormal numbers preserved.
DenormalModeKind Output
Denormal flushing mode for floating point instruction results in the default floating point environme...
static constexpr DenormalMode getDynamic()
static constexpr DenormalMode getIEEE()
Incoming for lane maks phi as machine instruction, incoming register Reg and incoming block Block are...
bool isConstant() const
Returns true if we know the value of all bits.
const APInt & getConstant() const
Returns the value when all bits have a known value.