-
Notifications
You must be signed in to change notification settings - Fork 10.8k
/
InstCombineVectorOps.cpp
1491 lines (1354 loc) · 58.9 KB
/
InstCombineVectorOps.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
//===- InstCombineVectorOps.cpp -------------------------------------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements instcombine for ExtractElement, InsertElement and
// ShuffleVector.
//
//===----------------------------------------------------------------------===//
#include "InstCombineInternal.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/VectorUtils.h"
#include "llvm/IR/PatternMatch.h"
using namespace llvm;
using namespace PatternMatch;
#define DEBUG_TYPE "instcombine"
/// Return true if the value is cheaper to scalarize than it is to leave as a
/// vector operation. isConstant indicates whether we're extracting one known
/// element. If false we're extracting a variable index.
static bool cheapToScalarize(Value *V, bool isConstant) {
if (Constant *C = dyn_cast<Constant>(V)) {
if (isConstant) return true;
// If all elts are the same, we can extract it and use any of the values.
if (Constant *Op0 = C->getAggregateElement(0U)) {
for (unsigned i = 1, e = V->getType()->getVectorNumElements(); i != e;
++i)
if (C->getAggregateElement(i) != Op0)
return false;
return true;
}
}
Instruction *I = dyn_cast<Instruction>(V);
if (!I) return false;
// Insert element gets simplified to the inserted element or is deleted if
// this is constant idx extract element and its a constant idx insertelt.
if (I->getOpcode() == Instruction::InsertElement && isConstant &&
isa<ConstantInt>(I->getOperand(2)))
return true;
if (I->getOpcode() == Instruction::Load && I->hasOneUse())
return true;
if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I))
if (BO->hasOneUse() &&
(cheapToScalarize(BO->getOperand(0), isConstant) ||
cheapToScalarize(BO->getOperand(1), isConstant)))
return true;
if (CmpInst *CI = dyn_cast<CmpInst>(I))
if (CI->hasOneUse() &&
(cheapToScalarize(CI->getOperand(0), isConstant) ||
cheapToScalarize(CI->getOperand(1), isConstant)))
return true;
return false;
}
// If we have a PHI node with a vector type that is only used to feed
// itself and be an operand of extractelement at a constant location,
// try to replace the PHI of the vector type with a PHI of a scalar type.
Instruction *InstCombiner::scalarizePHI(ExtractElementInst &EI, PHINode *PN) {
SmallVector<Instruction *, 2> Extracts;
// The users we want the PHI to have are:
// 1) The EI ExtractElement (we already know this)
// 2) Possibly more ExtractElements with the same index.
// 3) Another operand, which will feed back into the PHI.
Instruction *PHIUser = nullptr;
for (auto U : PN->users()) {
if (ExtractElementInst *EU = dyn_cast<ExtractElementInst>(U)) {
if (EI.getIndexOperand() == EU->getIndexOperand())
Extracts.push_back(EU);
else
return nullptr;
} else if (!PHIUser) {
PHIUser = cast<Instruction>(U);
} else {
return nullptr;
}
}
if (!PHIUser)
return nullptr;
// Verify that this PHI user has one use, which is the PHI itself,
// and that it is a binary operation which is cheap to scalarize.
// otherwise return NULL.
if (!PHIUser->hasOneUse() || !(PHIUser->user_back() == PN) ||
!(isa<BinaryOperator>(PHIUser)) || !cheapToScalarize(PHIUser, true))
return nullptr;
// Create a scalar PHI node that will replace the vector PHI node
// just before the current PHI node.
PHINode *scalarPHI = cast<PHINode>(InsertNewInstWith(
PHINode::Create(EI.getType(), PN->getNumIncomingValues(), ""), *PN));
// Scalarize each PHI operand.
for (unsigned i = 0; i < PN->getNumIncomingValues(); i++) {
Value *PHIInVal = PN->getIncomingValue(i);
BasicBlock *inBB = PN->getIncomingBlock(i);
Value *Elt = EI.getIndexOperand();
// If the operand is the PHI induction variable:
if (PHIInVal == PHIUser) {
// Scalarize the binary operation. Its first operand is the
// scalar PHI, and the second operand is extracted from the other
// vector operand.
BinaryOperator *B0 = cast<BinaryOperator>(PHIUser);
unsigned opId = (B0->getOperand(0) == PN) ? 1 : 0;
Value *Op = InsertNewInstWith(
ExtractElementInst::Create(B0->getOperand(opId), Elt,
B0->getOperand(opId)->getName() + ".Elt"),
*B0);
Value *newPHIUser = InsertNewInstWith(
BinaryOperator::CreateWithCopiedFlags(B0->getOpcode(),
scalarPHI, Op, B0), *B0);
scalarPHI->addIncoming(newPHIUser, inBB);
} else {
// Scalarize PHI input:
Instruction *newEI = ExtractElementInst::Create(PHIInVal, Elt, "");
// Insert the new instruction into the predecessor basic block.
Instruction *pos = dyn_cast<Instruction>(PHIInVal);
BasicBlock::iterator InsertPos;
if (pos && !isa<PHINode>(pos)) {
InsertPos = ++pos->getIterator();
} else {
InsertPos = inBB->getFirstInsertionPt();
}
InsertNewInstWith(newEI, *InsertPos);
scalarPHI->addIncoming(newEI, inBB);
}
}
for (auto E : Extracts)
replaceInstUsesWith(*E, scalarPHI);
return &EI;
}
Instruction *InstCombiner::visitExtractElementInst(ExtractElementInst &EI) {
if (Value *V = SimplifyExtractElementInst(
EI.getVectorOperand(), EI.getIndexOperand(), DL, &TLI, &DT, &AC))
return replaceInstUsesWith(EI, V);
// If vector val is constant with all elements the same, replace EI with
// that element. We handle a known element # below.
if (Constant *C = dyn_cast<Constant>(EI.getOperand(0)))
if (cheapToScalarize(C, false))
return replaceInstUsesWith(EI, C->getAggregateElement(0U));
// If extracting a specified index from the vector, see if we can recursively
// find a previously computed scalar that was inserted into the vector.
if (ConstantInt *IdxC = dyn_cast<ConstantInt>(EI.getOperand(1))) {
unsigned IndexVal = IdxC->getZExtValue();
unsigned VectorWidth = EI.getVectorOperandType()->getNumElements();
// InstSimplify handles cases where the index is invalid.
assert(IndexVal < VectorWidth);
// This instruction only demands the single element from the input vector.
// If the input vector has a single use, simplify it based on this use
// property.
if (EI.getOperand(0)->hasOneUse() && VectorWidth != 1) {
APInt UndefElts(VectorWidth, 0);
APInt DemandedMask(VectorWidth, 0);
DemandedMask.setBit(IndexVal);
if (Value *V = SimplifyDemandedVectorElts(EI.getOperand(0), DemandedMask,
UndefElts)) {
EI.setOperand(0, V);
return &EI;
}
}
// If this extractelement is directly using a bitcast from a vector of
// the same number of elements, see if we can find the source element from
// it. In this case, we will end up needing to bitcast the scalars.
if (BitCastInst *BCI = dyn_cast<BitCastInst>(EI.getOperand(0))) {
if (VectorType *VT = dyn_cast<VectorType>(BCI->getOperand(0)->getType()))
if (VT->getNumElements() == VectorWidth)
if (Value *Elt = findScalarElement(BCI->getOperand(0), IndexVal))
return new BitCastInst(Elt, EI.getType());
}
// If there's a vector PHI feeding a scalar use through this extractelement
// instruction, try to scalarize the PHI.
if (PHINode *PN = dyn_cast<PHINode>(EI.getOperand(0))) {
Instruction *scalarPHI = scalarizePHI(EI, PN);
if (scalarPHI)
return scalarPHI;
}
}
if (Instruction *I = dyn_cast<Instruction>(EI.getOperand(0))) {
// Push extractelement into predecessor operation if legal and
// profitable to do so.
if (BinaryOperator *BO = dyn_cast<BinaryOperator>(I)) {
if (I->hasOneUse() &&
cheapToScalarize(BO, isa<ConstantInt>(EI.getOperand(1)))) {
Value *newEI0 =
Builder->CreateExtractElement(BO->getOperand(0), EI.getOperand(1),
EI.getName()+".lhs");
Value *newEI1 =
Builder->CreateExtractElement(BO->getOperand(1), EI.getOperand(1),
EI.getName()+".rhs");
return BinaryOperator::CreateWithCopiedFlags(BO->getOpcode(),
newEI0, newEI1, BO);
}
} else if (InsertElementInst *IE = dyn_cast<InsertElementInst>(I)) {
// Extracting the inserted element?
if (IE->getOperand(2) == EI.getOperand(1))
return replaceInstUsesWith(EI, IE->getOperand(1));
// If the inserted and extracted elements are constants, they must not
// be the same value, extract from the pre-inserted value instead.
if (isa<Constant>(IE->getOperand(2)) && isa<Constant>(EI.getOperand(1))) {
Worklist.AddValue(EI.getOperand(0));
EI.setOperand(0, IE->getOperand(0));
return &EI;
}
} else if (ShuffleVectorInst *SVI = dyn_cast<ShuffleVectorInst>(I)) {
// If this is extracting an element from a shufflevector, figure out where
// it came from and extract from the appropriate input element instead.
if (ConstantInt *Elt = dyn_cast<ConstantInt>(EI.getOperand(1))) {
int SrcIdx = SVI->getMaskValue(Elt->getZExtValue());
Value *Src;
unsigned LHSWidth =
SVI->getOperand(0)->getType()->getVectorNumElements();
if (SrcIdx < 0)
return replaceInstUsesWith(EI, UndefValue::get(EI.getType()));
if (SrcIdx < (int)LHSWidth)
Src = SVI->getOperand(0);
else {
SrcIdx -= LHSWidth;
Src = SVI->getOperand(1);
}
Type *Int32Ty = Type::getInt32Ty(EI.getContext());
return ExtractElementInst::Create(Src,
ConstantInt::get(Int32Ty,
SrcIdx, false));
}
} else if (CastInst *CI = dyn_cast<CastInst>(I)) {
// Canonicalize extractelement(cast) -> cast(extractelement).
// Bitcasts can change the number of vector elements, and they cost
// nothing.
if (CI->hasOneUse() && (CI->getOpcode() != Instruction::BitCast)) {
Value *EE = Builder->CreateExtractElement(CI->getOperand(0),
EI.getIndexOperand());
Worklist.AddValue(EE);
return CastInst::Create(CI->getOpcode(), EE, EI.getType());
}
} else if (SelectInst *SI = dyn_cast<SelectInst>(I)) {
if (SI->hasOneUse()) {
// TODO: For a select on vectors, it might be useful to do this if it
// has multiple extractelement uses. For vector select, that seems to
// fight the vectorizer.
// If we are extracting an element from a vector select or a select on
// vectors, create a select on the scalars extracted from the vector
// arguments.
Value *TrueVal = SI->getTrueValue();
Value *FalseVal = SI->getFalseValue();
Value *Cond = SI->getCondition();
if (Cond->getType()->isVectorTy()) {
Cond = Builder->CreateExtractElement(Cond,
EI.getIndexOperand(),
Cond->getName() + ".elt");
}
Value *V1Elem
= Builder->CreateExtractElement(TrueVal,
EI.getIndexOperand(),
TrueVal->getName() + ".elt");
Value *V2Elem
= Builder->CreateExtractElement(FalseVal,
EI.getIndexOperand(),
FalseVal->getName() + ".elt");
return SelectInst::Create(Cond,
V1Elem,
V2Elem,
SI->getName() + ".elt");
}
}
}
return nullptr;
}
/// If V is a shuffle of values that ONLY returns elements from either LHS or
/// RHS, return the shuffle mask and true. Otherwise, return false.
static bool collectSingleShuffleElements(Value *V, Value *LHS, Value *RHS,
SmallVectorImpl<Constant*> &Mask) {
assert(LHS->getType() == RHS->getType() &&
"Invalid CollectSingleShuffleElements");
unsigned NumElts = V->getType()->getVectorNumElements();
if (isa<UndefValue>(V)) {
Mask.assign(NumElts, UndefValue::get(Type::getInt32Ty(V->getContext())));
return true;
}
if (V == LHS) {
for (unsigned i = 0; i != NumElts; ++i)
Mask.push_back(ConstantInt::get(Type::getInt32Ty(V->getContext()), i));
return true;
}
if (V == RHS) {
for (unsigned i = 0; i != NumElts; ++i)
Mask.push_back(ConstantInt::get(Type::getInt32Ty(V->getContext()),
i+NumElts));
return true;
}
if (InsertElementInst *IEI = dyn_cast<InsertElementInst>(V)) {
// If this is an insert of an extract from some other vector, include it.
Value *VecOp = IEI->getOperand(0);
Value *ScalarOp = IEI->getOperand(1);
Value *IdxOp = IEI->getOperand(2);
if (!isa<ConstantInt>(IdxOp))
return false;
unsigned InsertedIdx = cast<ConstantInt>(IdxOp)->getZExtValue();
if (isa<UndefValue>(ScalarOp)) { // inserting undef into vector.
// We can handle this if the vector we are inserting into is
// transitively ok.
if (collectSingleShuffleElements(VecOp, LHS, RHS, Mask)) {
// If so, update the mask to reflect the inserted undef.
Mask[InsertedIdx] = UndefValue::get(Type::getInt32Ty(V->getContext()));
return true;
}
} else if (ExtractElementInst *EI = dyn_cast<ExtractElementInst>(ScalarOp)){
if (isa<ConstantInt>(EI->getOperand(1))) {
unsigned ExtractedIdx =
cast<ConstantInt>(EI->getOperand(1))->getZExtValue();
unsigned NumLHSElts = LHS->getType()->getVectorNumElements();
// This must be extracting from either LHS or RHS.
if (EI->getOperand(0) == LHS || EI->getOperand(0) == RHS) {
// We can handle this if the vector we are inserting into is
// transitively ok.
if (collectSingleShuffleElements(VecOp, LHS, RHS, Mask)) {
// If so, update the mask to reflect the inserted value.
if (EI->getOperand(0) == LHS) {
Mask[InsertedIdx % NumElts] =
ConstantInt::get(Type::getInt32Ty(V->getContext()),
ExtractedIdx);
} else {
assert(EI->getOperand(0) == RHS);
Mask[InsertedIdx % NumElts] =
ConstantInt::get(Type::getInt32Ty(V->getContext()),
ExtractedIdx + NumLHSElts);
}
return true;
}
}
}
}
}
return false;
}
/// If we have insertion into a vector that is wider than the vector that we
/// are extracting from, try to widen the source vector to allow a single
/// shufflevector to replace one or more insert/extract pairs.
static void replaceExtractElements(InsertElementInst *InsElt,
ExtractElementInst *ExtElt,
InstCombiner &IC) {
VectorType *InsVecType = InsElt->getType();
VectorType *ExtVecType = ExtElt->getVectorOperandType();
unsigned NumInsElts = InsVecType->getVectorNumElements();
unsigned NumExtElts = ExtVecType->getVectorNumElements();
// The inserted-to vector must be wider than the extracted-from vector.
if (InsVecType->getElementType() != ExtVecType->getElementType() ||
NumExtElts >= NumInsElts)
return;
// Create a shuffle mask to widen the extended-from vector using undefined
// values. The mask selects all of the values of the original vector followed
// by as many undefined values as needed to create a vector of the same length
// as the inserted-to vector.
SmallVector<Constant *, 16> ExtendMask;
IntegerType *IntType = Type::getInt32Ty(InsElt->getContext());
for (unsigned i = 0; i < NumExtElts; ++i)
ExtendMask.push_back(ConstantInt::get(IntType, i));
for (unsigned i = NumExtElts; i < NumInsElts; ++i)
ExtendMask.push_back(UndefValue::get(IntType));
Value *ExtVecOp = ExtElt->getVectorOperand();
auto *ExtVecOpInst = dyn_cast<Instruction>(ExtVecOp);
BasicBlock *InsertionBlock = (ExtVecOpInst && !isa<PHINode>(ExtVecOpInst))
? ExtVecOpInst->getParent()
: ExtElt->getParent();
// TODO: This restriction matches the basic block check below when creating
// new extractelement instructions. If that limitation is removed, this one
// could also be removed. But for now, we just bail out to ensure that we
// will replace the extractelement instruction that is feeding our
// insertelement instruction. This allows the insertelement to then be
// replaced by a shufflevector. If the insertelement is not replaced, we can
// induce infinite looping because there's an optimization for extractelement
// that will delete our widening shuffle. This would trigger another attempt
// here to create that shuffle, and we spin forever.
if (InsertionBlock != InsElt->getParent())
return;
// TODO: This restriction matches the check in visitInsertElementInst() and
// prevents an infinite loop caused by not turning the extract/insert pair
// into a shuffle. We really should not need either check, but we're lacking
// folds for shufflevectors because we're afraid to generate shuffle masks
// that the backend can't handle.
if (InsElt->hasOneUse() && isa<InsertElementInst>(InsElt->user_back()))
return;
auto *WideVec = new ShuffleVectorInst(ExtVecOp, UndefValue::get(ExtVecType),
ConstantVector::get(ExtendMask));
// Insert the new shuffle after the vector operand of the extract is defined
// (as long as it's not a PHI) or at the start of the basic block of the
// extract, so any subsequent extracts in the same basic block can use it.
// TODO: Insert before the earliest ExtractElementInst that is replaced.
if (ExtVecOpInst && !isa<PHINode>(ExtVecOpInst))
WideVec->insertAfter(ExtVecOpInst);
else
IC.InsertNewInstWith(WideVec, *ExtElt->getParent()->getFirstInsertionPt());
// Replace extracts from the original narrow vector with extracts from the new
// wide vector.
for (User *U : ExtVecOp->users()) {
ExtractElementInst *OldExt = dyn_cast<ExtractElementInst>(U);
if (!OldExt || OldExt->getParent() != WideVec->getParent())
continue;
auto *NewExt = ExtractElementInst::Create(WideVec, OldExt->getOperand(1));
NewExt->insertAfter(WideVec);
IC.replaceInstUsesWith(*OldExt, NewExt);
}
}
/// We are building a shuffle to create V, which is a sequence of insertelement,
/// extractelement pairs. If PermittedRHS is set, then we must either use it or
/// not rely on the second vector source. Return a std::pair containing the
/// left and right vectors of the proposed shuffle (or 0), and set the Mask
/// parameter as required.
///
/// Note: we intentionally don't try to fold earlier shuffles since they have
/// often been chosen carefully to be efficiently implementable on the target.
typedef std::pair<Value *, Value *> ShuffleOps;
static ShuffleOps collectShuffleElements(Value *V,
SmallVectorImpl<Constant *> &Mask,
Value *PermittedRHS,
InstCombiner &IC) {
assert(V->getType()->isVectorTy() && "Invalid shuffle!");
unsigned NumElts = V->getType()->getVectorNumElements();
if (isa<UndefValue>(V)) {
Mask.assign(NumElts, UndefValue::get(Type::getInt32Ty(V->getContext())));
return std::make_pair(
PermittedRHS ? UndefValue::get(PermittedRHS->getType()) : V, nullptr);
}
if (isa<ConstantAggregateZero>(V)) {
Mask.assign(NumElts, ConstantInt::get(Type::getInt32Ty(V->getContext()),0));
return std::make_pair(V, nullptr);
}
if (InsertElementInst *IEI = dyn_cast<InsertElementInst>(V)) {
// If this is an insert of an extract from some other vector, include it.
Value *VecOp = IEI->getOperand(0);
Value *ScalarOp = IEI->getOperand(1);
Value *IdxOp = IEI->getOperand(2);
if (ExtractElementInst *EI = dyn_cast<ExtractElementInst>(ScalarOp)) {
if (isa<ConstantInt>(EI->getOperand(1)) && isa<ConstantInt>(IdxOp)) {
unsigned ExtractedIdx =
cast<ConstantInt>(EI->getOperand(1))->getZExtValue();
unsigned InsertedIdx = cast<ConstantInt>(IdxOp)->getZExtValue();
// Either the extracted from or inserted into vector must be RHSVec,
// otherwise we'd end up with a shuffle of three inputs.
if (EI->getOperand(0) == PermittedRHS || PermittedRHS == nullptr) {
Value *RHS = EI->getOperand(0);
ShuffleOps LR = collectShuffleElements(VecOp, Mask, RHS, IC);
assert(LR.second == nullptr || LR.second == RHS);
if (LR.first->getType() != RHS->getType()) {
// Although we are giving up for now, see if we can create extracts
// that match the inserts for another round of combining.
replaceExtractElements(IEI, EI, IC);
// We tried our best, but we can't find anything compatible with RHS
// further up the chain. Return a trivial shuffle.
for (unsigned i = 0; i < NumElts; ++i)
Mask[i] = ConstantInt::get(Type::getInt32Ty(V->getContext()), i);
return std::make_pair(V, nullptr);
}
unsigned NumLHSElts = RHS->getType()->getVectorNumElements();
Mask[InsertedIdx % NumElts] =
ConstantInt::get(Type::getInt32Ty(V->getContext()),
NumLHSElts+ExtractedIdx);
return std::make_pair(LR.first, RHS);
}
if (VecOp == PermittedRHS) {
// We've gone as far as we can: anything on the other side of the
// extractelement will already have been converted into a shuffle.
unsigned NumLHSElts =
EI->getOperand(0)->getType()->getVectorNumElements();
for (unsigned i = 0; i != NumElts; ++i)
Mask.push_back(ConstantInt::get(
Type::getInt32Ty(V->getContext()),
i == InsertedIdx ? ExtractedIdx : NumLHSElts + i));
return std::make_pair(EI->getOperand(0), PermittedRHS);
}
// If this insertelement is a chain that comes from exactly these two
// vectors, return the vector and the effective shuffle.
if (EI->getOperand(0)->getType() == PermittedRHS->getType() &&
collectSingleShuffleElements(IEI, EI->getOperand(0), PermittedRHS,
Mask))
return std::make_pair(EI->getOperand(0), PermittedRHS);
}
}
}
// Otherwise, we can't do anything fancy. Return an identity vector.
for (unsigned i = 0; i != NumElts; ++i)
Mask.push_back(ConstantInt::get(Type::getInt32Ty(V->getContext()), i));
return std::make_pair(V, nullptr);
}
/// Try to find redundant insertvalue instructions, like the following ones:
/// %0 = insertvalue { i8, i32 } undef, i8 %x, 0
/// %1 = insertvalue { i8, i32 } %0, i8 %y, 0
/// Here the second instruction inserts values at the same indices, as the
/// first one, making the first one redundant.
/// It should be transformed to:
/// %0 = insertvalue { i8, i32 } undef, i8 %y, 0
Instruction *InstCombiner::visitInsertValueInst(InsertValueInst &I) {
bool IsRedundant = false;
ArrayRef<unsigned int> FirstIndices = I.getIndices();
// If there is a chain of insertvalue instructions (each of them except the
// last one has only one use and it's another insertvalue insn from this
// chain), check if any of the 'children' uses the same indices as the first
// instruction. In this case, the first one is redundant.
Value *V = &I;
unsigned Depth = 0;
while (V->hasOneUse() && Depth < 10) {
User *U = V->user_back();
auto UserInsInst = dyn_cast<InsertValueInst>(U);
if (!UserInsInst || U->getOperand(0) != V)
break;
if (UserInsInst->getIndices() == FirstIndices) {
IsRedundant = true;
break;
}
V = UserInsInst;
Depth++;
}
if (IsRedundant)
return replaceInstUsesWith(I, I.getOperand(0));
return nullptr;
}
static bool isShuffleEquivalentToSelect(ShuffleVectorInst &Shuf) {
int MaskSize = Shuf.getMask()->getType()->getVectorNumElements();
int VecSize = Shuf.getOperand(0)->getType()->getVectorNumElements();
// A vector select does not change the size of the operands.
if (MaskSize != VecSize)
return false;
// Each mask element must be undefined or choose a vector element from one of
// the source operands without crossing vector lanes.
for (int i = 0; i != MaskSize; ++i) {
int Elt = Shuf.getMaskValue(i);
if (Elt != -1 && Elt != i && Elt != i + VecSize)
return false;
}
return true;
}
// Turn a chain of inserts that splats a value into a canonical insert + shuffle
// splat. That is:
// insertelt(insertelt(insertelt(insertelt X, %k, 0), %k, 1), %k, 2) ... ->
// shufflevector(insertelt(X, %k, 0), undef, zero)
static Instruction *foldInsSequenceIntoBroadcast(InsertElementInst &InsElt) {
// We are interested in the last insert in a chain. So, if this insert
// has a single user, and that user is an insert, bail.
if (InsElt.hasOneUse() && isa<InsertElementInst>(InsElt.user_back()))
return nullptr;
VectorType *VT = cast<VectorType>(InsElt.getType());
int NumElements = VT->getNumElements();
// Do not try to do this for a one-element vector, since that's a nop,
// and will cause an inf-loop.
if (NumElements == 1)
return nullptr;
Value *SplatVal = InsElt.getOperand(1);
InsertElementInst *CurrIE = &InsElt;
SmallVector<bool, 16> ElementPresent(NumElements, false);
// Walk the chain backwards, keeping track of which indices we inserted into,
// until we hit something that isn't an insert of the splatted value.
while (CurrIE) {
ConstantInt *Idx = dyn_cast<ConstantInt>(CurrIE->getOperand(2));
if (!Idx || CurrIE->getOperand(1) != SplatVal)
return nullptr;
// Check none of the intermediate steps have any additional uses.
if ((CurrIE != &InsElt) && !CurrIE->hasOneUse())
return nullptr;
ElementPresent[Idx->getZExtValue()] = true;
CurrIE = dyn_cast<InsertElementInst>(CurrIE->getOperand(0));
}
// Make sure we've seen an insert into every element.
if (llvm::any_of(ElementPresent, [](bool Present) { return !Present; }))
return nullptr;
// All right, create the insert + shuffle.
Instruction *InsertFirst = InsertElementInst::Create(
UndefValue::get(VT), SplatVal,
ConstantInt::get(Type::getInt32Ty(InsElt.getContext()), 0), "", &InsElt);
Constant *ZeroMask = ConstantAggregateZero::get(
VectorType::get(Type::getInt32Ty(InsElt.getContext()), NumElements));
return new ShuffleVectorInst(InsertFirst, UndefValue::get(VT), ZeroMask);
}
/// If we have an insertelement instruction feeding into another insertelement
/// and the 2nd is inserting a constant into the vector, canonicalize that
/// constant insertion before the insertion of a variable:
///
/// insertelement (insertelement X, Y, IdxC1), ScalarC, IdxC2 -->
/// insertelement (insertelement X, ScalarC, IdxC2), Y, IdxC1
///
/// This has the potential of eliminating the 2nd insertelement instruction
/// via constant folding of the scalar constant into a vector constant.
static Instruction *hoistInsEltConst(InsertElementInst &InsElt2,
InstCombiner::BuilderTy &Builder) {
auto *InsElt1 = dyn_cast<InsertElementInst>(InsElt2.getOperand(0));
if (!InsElt1 || !InsElt1->hasOneUse())
return nullptr;
Value *X, *Y;
Constant *ScalarC;
ConstantInt *IdxC1, *IdxC2;
if (match(InsElt1->getOperand(0), m_Value(X)) &&
match(InsElt1->getOperand(1), m_Value(Y)) && !isa<Constant>(Y) &&
match(InsElt1->getOperand(2), m_ConstantInt(IdxC1)) &&
match(InsElt2.getOperand(1), m_Constant(ScalarC)) &&
match(InsElt2.getOperand(2), m_ConstantInt(IdxC2)) && IdxC1 != IdxC2) {
Value *NewInsElt1 = Builder.CreateInsertElement(X, ScalarC, IdxC2);
return InsertElementInst::Create(NewInsElt1, Y, IdxC1);
}
return nullptr;
}
/// insertelt (shufflevector X, CVec, Mask|insertelt X, C1, CIndex1), C, CIndex
/// --> shufflevector X, CVec', Mask'
static Instruction *foldConstantInsEltIntoShuffle(InsertElementInst &InsElt) {
auto *Inst = dyn_cast<Instruction>(InsElt.getOperand(0));
// Bail out if the parent has more than one use. In that case, we'd be
// replacing the insertelt with a shuffle, and that's not a clear win.
if (!Inst || !Inst->hasOneUse())
return nullptr;
if (auto *Shuf = dyn_cast<ShuffleVectorInst>(InsElt.getOperand(0))) {
// The shuffle must have a constant vector operand. The insertelt must have
// a constant scalar being inserted at a constant position in the vector.
Constant *ShufConstVec, *InsEltScalar;
uint64_t InsEltIndex;
if (!match(Shuf->getOperand(1), m_Constant(ShufConstVec)) ||
!match(InsElt.getOperand(1), m_Constant(InsEltScalar)) ||
!match(InsElt.getOperand(2), m_ConstantInt(InsEltIndex)))
return nullptr;
// Adding an element to an arbitrary shuffle could be expensive, but a
// shuffle that selects elements from vectors without crossing lanes is
// assumed cheap.
// If we're just adding a constant into that shuffle, it will still be
// cheap.
if (!isShuffleEquivalentToSelect(*Shuf))
return nullptr;
// From the above 'select' check, we know that the mask has the same number
// of elements as the vector input operands. We also know that each constant
// input element is used in its lane and can not be used more than once by
// the shuffle. Therefore, replace the constant in the shuffle's constant
// vector with the insertelt constant. Replace the constant in the shuffle's
// mask vector with the insertelt index plus the length of the vector
// (because the constant vector operand of a shuffle is always the 2nd
// operand).
Constant *Mask = Shuf->getMask();
unsigned NumElts = Mask->getType()->getVectorNumElements();
SmallVector<Constant *, 16> NewShufElts(NumElts);
SmallVector<Constant *, 16> NewMaskElts(NumElts);
for (unsigned I = 0; I != NumElts; ++I) {
if (I == InsEltIndex) {
NewShufElts[I] = InsEltScalar;
Type *Int32Ty = Type::getInt32Ty(Shuf->getContext());
NewMaskElts[I] = ConstantInt::get(Int32Ty, InsEltIndex + NumElts);
} else {
// Copy over the existing values.
NewShufElts[I] = ShufConstVec->getAggregateElement(I);
NewMaskElts[I] = Mask->getAggregateElement(I);
}
}
// Create new operands for a shuffle that includes the constant of the
// original insertelt. The old shuffle will be dead now.
return new ShuffleVectorInst(Shuf->getOperand(0),
ConstantVector::get(NewShufElts),
ConstantVector::get(NewMaskElts));
} else if (auto *IEI = dyn_cast<InsertElementInst>(Inst)) {
// Transform sequences of insertelements ops with constant data/indexes into
// a single shuffle op.
unsigned NumElts = InsElt.getType()->getNumElements();
uint64_t InsertIdx[2];
Constant *Val[2];
if (!match(InsElt.getOperand(2), m_ConstantInt(InsertIdx[0])) ||
!match(InsElt.getOperand(1), m_Constant(Val[0])) ||
!match(IEI->getOperand(2), m_ConstantInt(InsertIdx[1])) ||
!match(IEI->getOperand(1), m_Constant(Val[1])))
return nullptr;
SmallVector<Constant *, 16> Values(NumElts);
SmallVector<Constant *, 16> Mask(NumElts);
auto ValI = std::begin(Val);
// Generate new constant vector and mask.
// We have 2 values/masks from the insertelements instructions. Insert them
// into new value/mask vectors.
for (uint64_t I : InsertIdx) {
if (!Values[I]) {
assert(!Mask[I]);
Values[I] = *ValI;
Mask[I] = ConstantInt::get(Type::getInt32Ty(InsElt.getContext()),
NumElts + I);
}
++ValI;
}
// Remaining values are filled with 'undef' values.
for (unsigned I = 0; I < NumElts; ++I) {
if (!Values[I]) {
assert(!Mask[I]);
Values[I] = UndefValue::get(InsElt.getType()->getElementType());
Mask[I] = ConstantInt::get(Type::getInt32Ty(InsElt.getContext()), I);
}
}
// Create new operands for a shuffle that includes the constant of the
// original insertelt.
return new ShuffleVectorInst(IEI->getOperand(0),
ConstantVector::get(Values),
ConstantVector::get(Mask));
}
return nullptr;
}
Instruction *InstCombiner::visitInsertElementInst(InsertElementInst &IE) {
Value *VecOp = IE.getOperand(0);
Value *ScalarOp = IE.getOperand(1);
Value *IdxOp = IE.getOperand(2);
// Inserting an undef or into an undefined place, remove this.
if (isa<UndefValue>(ScalarOp) || isa<UndefValue>(IdxOp))
replaceInstUsesWith(IE, VecOp);
// If the inserted element was extracted from some other vector, and if the
// indexes are constant, try to turn this into a shufflevector operation.
if (ExtractElementInst *EI = dyn_cast<ExtractElementInst>(ScalarOp)) {
if (isa<ConstantInt>(EI->getOperand(1)) && isa<ConstantInt>(IdxOp)) {
unsigned NumInsertVectorElts = IE.getType()->getNumElements();
unsigned NumExtractVectorElts =
EI->getOperand(0)->getType()->getVectorNumElements();
unsigned ExtractedIdx =
cast<ConstantInt>(EI->getOperand(1))->getZExtValue();
unsigned InsertedIdx = cast<ConstantInt>(IdxOp)->getZExtValue();
if (ExtractedIdx >= NumExtractVectorElts) // Out of range extract.
return replaceInstUsesWith(IE, VecOp);
if (InsertedIdx >= NumInsertVectorElts) // Out of range insert.
return replaceInstUsesWith(IE, UndefValue::get(IE.getType()));
// If we are extracting a value from a vector, then inserting it right
// back into the same place, just use the input vector.
if (EI->getOperand(0) == VecOp && ExtractedIdx == InsertedIdx)
return replaceInstUsesWith(IE, VecOp);
// If this insertelement isn't used by some other insertelement, turn it
// (and any insertelements it points to), into one big shuffle.
if (!IE.hasOneUse() || !isa<InsertElementInst>(IE.user_back())) {
SmallVector<Constant*, 16> Mask;
ShuffleOps LR = collectShuffleElements(&IE, Mask, nullptr, *this);
// The proposed shuffle may be trivial, in which case we shouldn't
// perform the combine.
if (LR.first != &IE && LR.second != &IE) {
// We now have a shuffle of LHS, RHS, Mask.
if (LR.second == nullptr)
LR.second = UndefValue::get(LR.first->getType());
return new ShuffleVectorInst(LR.first, LR.second,
ConstantVector::get(Mask));
}
}
}
}
unsigned VWidth = VecOp->getType()->getVectorNumElements();
APInt UndefElts(VWidth, 0);
APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
if (Value *V = SimplifyDemandedVectorElts(&IE, AllOnesEltMask, UndefElts)) {
if (V != &IE)
return replaceInstUsesWith(IE, V);
return &IE;
}
if (Instruction *Shuf = foldConstantInsEltIntoShuffle(IE))
return Shuf;
if (Instruction *NewInsElt = hoistInsEltConst(IE, *Builder))
return NewInsElt;
// Turn a sequence of inserts that broadcasts a scalar into a single
// insert + shufflevector.
if (Instruction *Broadcast = foldInsSequenceIntoBroadcast(IE))
return Broadcast;
return nullptr;
}
/// Return true if we can evaluate the specified expression tree if the vector
/// elements were shuffled in a different order.
static bool CanEvaluateShuffled(Value *V, ArrayRef<int> Mask,
unsigned Depth = 5) {
// We can always reorder the elements of a constant.
if (isa<Constant>(V))
return true;
// We won't reorder vector arguments. No IPO here.
Instruction *I = dyn_cast<Instruction>(V);
if (!I) return false;
// Two users may expect different orders of the elements. Don't try it.
if (!I->hasOneUse())
return false;
if (Depth == 0) return false;
switch (I->getOpcode()) {
case Instruction::Add:
case Instruction::FAdd:
case Instruction::Sub:
case Instruction::FSub:
case Instruction::Mul:
case Instruction::FMul:
case Instruction::UDiv:
case Instruction::SDiv:
case Instruction::FDiv:
case Instruction::URem:
case Instruction::SRem:
case Instruction::FRem:
case Instruction::Shl:
case Instruction::LShr:
case Instruction::AShr:
case Instruction::And:
case Instruction::Or:
case Instruction::Xor:
case Instruction::ICmp:
case Instruction::FCmp:
case Instruction::Trunc:
case Instruction::ZExt:
case Instruction::SExt:
case Instruction::FPToUI:
case Instruction::FPToSI:
case Instruction::UIToFP:
case Instruction::SIToFP:
case Instruction::FPTrunc:
case Instruction::FPExt:
case Instruction::GetElementPtr: {
for (Value *Operand : I->operands()) {
if (!CanEvaluateShuffled(Operand, Mask, Depth-1))
return false;
}
return true;
}
case Instruction::InsertElement: {
ConstantInt *CI = dyn_cast<ConstantInt>(I->getOperand(2));
if (!CI) return false;
int ElementNumber = CI->getLimitedValue();
// Verify that 'CI' does not occur twice in Mask. A single 'insertelement'
// can't put an element into multiple indices.
bool SeenOnce = false;
for (int i = 0, e = Mask.size(); i != e; ++i) {
if (Mask[i] == ElementNumber) {
if (SeenOnce)
return false;
SeenOnce = true;
}
}
return CanEvaluateShuffled(I->getOperand(0), Mask, Depth-1);
}
}
return false;
}
/// Rebuild a new instruction just like 'I' but with the new operands given.
/// In the event of type mismatch, the type of the operands is correct.
static Value *buildNew(Instruction *I, ArrayRef<Value*> NewOps) {
// We don't want to use the IRBuilder here because we want the replacement
// instructions to appear next to 'I', not the builder's insertion point.
switch (I->getOpcode()) {
case Instruction::Add:
case Instruction::FAdd:
case Instruction::Sub:
case Instruction::FSub:
case Instruction::Mul:
case Instruction::FMul:
case Instruction::UDiv:
case Instruction::SDiv:
case Instruction::FDiv:
case Instruction::URem:
case Instruction::SRem:
case Instruction::FRem:
case Instruction::Shl:
case Instruction::LShr:
case Instruction::AShr:
case Instruction::And:
case Instruction::Or:
case Instruction::Xor: {
BinaryOperator *BO = cast<BinaryOperator>(I);
assert(NewOps.size() == 2 && "binary operator with #ops != 2");
BinaryOperator *New =
BinaryOperator::Create(cast<BinaryOperator>(I)->getOpcode(),
NewOps[0], NewOps[1], "", BO);
if (isa<OverflowingBinaryOperator>(BO)) {
New->setHasNoUnsignedWrap(BO->hasNoUnsignedWrap());
New->setHasNoSignedWrap(BO->hasNoSignedWrap());
}
if (isa<PossiblyExactOperator>(BO)) {
New->setIsExact(BO->isExact());
}
if (isa<FPMathOperator>(BO))
New->copyFastMathFlags(I);
return New;
}
case Instruction::ICmp:
assert(NewOps.size() == 2 && "icmp with #ops != 2");
return new ICmpInst(I, cast<ICmpInst>(I)->getPredicate(),
NewOps[0], NewOps[1]);
case Instruction::FCmp:
assert(NewOps.size() == 2 && "fcmp with #ops != 2");
return new FCmpInst(I, cast<FCmpInst>(I)->getPredicate(),
NewOps[0], NewOps[1]);
case Instruction::Trunc:
case Instruction::ZExt:
case Instruction::SExt:
case Instruction::FPToUI:
case Instruction::FPToSI:
case Instruction::UIToFP:
case Instruction::SIToFP:
case Instruction::FPTrunc:
case Instruction::FPExt: {
// It's possible that the mask has a different number of elements from
// the original cast. We recompute the destination type to match the mask.
Type *DestTy =
VectorType::get(I->getType()->getScalarType(),
NewOps[0]->getType()->getVectorNumElements());
assert(NewOps.size() == 1 && "cast with #ops != 1");
return CastInst::Create(cast<CastInst>(I)->getOpcode(), NewOps[0], DestTy,
"", I);
}
case Instruction::GetElementPtr: {
Value *Ptr = NewOps[0];
ArrayRef<Value*> Idx = NewOps.slice(1);
GetElementPtrInst *GEP = GetElementPtrInst::Create(
cast<GetElementPtrInst>(I)->getSourceElementType(), Ptr, Idx, "", I);
GEP->setIsInBounds(cast<GetElementPtrInst>(I)->isInBounds());
return GEP;
}
}