Skip to content

Commit 3d7a16b

Browse files
RKSimonRKSimon
committed
[VectorCombine] foldShuffleOfShuffles - fold "shuffle (shuffle x, undef), (shuffle y, undef)" -> "shuffle x, y"
Another step towards cleaning up shuffles that have been split, often across bitcasts between SSE intrinsic. Strip shuffles entirely if we fold to an identity shuffle.
1 parent bf1ad1d commit 3d7a16b

File tree

5 files changed

+94
-32
lines changed

5 files changed

+94
-32
lines changed

‎llvm/lib/Transforms/Vectorize/VectorCombine.cpp

Lines changed: 82 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -113,6 +113,7 @@ class VectorCombine {
113113
bool scalarizeLoadExtract(Instruction &I);
114114
bool foldShuffleOfBinops(Instruction &I);
115115
bool foldShuffleOfCastops(Instruction &I);
116+
bool foldShuffleOfShuffles(Instruction &I);
116117
bool foldShuffleFromReductions(Instruction &I);
117118
bool foldTruncFromReductions(Instruction &I);
118119
bool foldSelectShuffle(Instruction &I, bool FromReduction = false);
@@ -1547,6 +1548,86 @@ bool VectorCombine::foldShuffleOfCastops(Instruction &I) {
15471548
return true;
15481549
}
15491550

1551+
/// Try to convert "shuffle (shuffle x, undef), (shuffle y, undef)"
1552+
/// into "shuffle x, y".
1553+
bool VectorCombine::foldShuffleOfShuffles(Instruction &I) {
1554+
Value *V0, *V1, *U0, *U1;
1555+
ArrayRef<int> OuterMask, InnerMask0, InnerMask1;
1556+
if (!match(&I, m_Shuffle(m_OneUse(m_Shuffle(m_Value(V0), m_Value(U0),
1557+
m_Mask(InnerMask0))),
1558+
m_OneUse(m_Shuffle(m_Value(V1), m_Value(U1),
1559+
m_Mask(InnerMask1))),
1560+
m_Mask(OuterMask))) ||
1561+
!isa<UndefValue>(U0) || !isa<UndefValue>(U1))
1562+
return false;
1563+
1564+
auto *ShuffleDstTy = dyn_cast<FixedVectorType>(I.getType());
1565+
auto *ShuffleSrcTy = dyn_cast<FixedVectorType>(V0->getType());
1566+
auto *ShuffleImmTy = dyn_cast<FixedVectorType>(I.getOperand(0)->getType());
1567+
if (!ShuffleDstTy || !ShuffleSrcTy || !ShuffleImmTy ||
1568+
V0->getType() != V1->getType())
1569+
return false;
1570+
1571+
unsigned NumSrcElts = ShuffleSrcTy->getNumElements();
1572+
unsigned NumImmElts = ShuffleImmTy->getNumElements();
1573+
1574+
// Bail if either inner masks reference a RHS undef arg.
1575+
if ((!isa<PoisonValue>(U0) &&
1576+
any_of(InnerMask0, [&](int M) { return M >= (int)NumSrcElts; })) ||
1577+
(!isa<PoisonValue>(U1) &&
1578+
any_of(InnerMask1, [&](int M) { return M >= (int)NumSrcElts; })))
1579+
return false;
1580+
1581+
// Merge shuffles - replace index to the RHS poison arg with PoisonMaskElem,
1582+
SmallVector<int, 16> NewMask(OuterMask.begin(), OuterMask.end());
1583+
for (int &M : NewMask) {
1584+
if (0 <= M && M < (int)NumImmElts) {
1585+
M = (InnerMask0[M] >= (int)NumSrcElts) ? PoisonMaskElem : InnerMask0[M];
1586+
} else if (M >= (int)NumImmElts) {
1587+
if (InnerMask1[M - NumImmElts] >= (int)NumSrcElts)
1588+
M = PoisonMaskElem;
1589+
else
1590+
M = InnerMask1[M - NumImmElts] + (V0 == V1 ? 0 : NumSrcElts);
1591+
}
1592+
}
1593+
1594+
// Have we folded to an Identity shuffle?
1595+
if (ShuffleVectorInst::isIdentityMask(NewMask, NumSrcElts)) {
1596+
replaceValue(I, *V0);
1597+
return true;
1598+
}
1599+
1600+
// Try to merge the shuffles if the new shuffle is not costly.
1601+
TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
1602+
1603+
InstructionCost OldCost =
1604+
TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, ShuffleSrcTy,
1605+
InnerMask0, CostKind) +
1606+
TTI.getShuffleCost(TargetTransformInfo::SK_PermuteSingleSrc, ShuffleSrcTy,
1607+
InnerMask1, CostKind) +
1608+
TTI.getShuffleCost(TargetTransformInfo::SK_PermuteTwoSrc, ShuffleImmTy,
1609+
OuterMask, CostKind, 0, nullptr, std::nullopt, &I);
1610+
1611+
InstructionCost NewCost = TTI.getShuffleCost(
1612+
TargetTransformInfo::SK_PermuteTwoSrc, ShuffleSrcTy, NewMask, CostKind);
1613+
1614+
LLVM_DEBUG(dbgs() << "Found a shuffle feeding two shuffles: " << I
1615+
<< "\n OldCost: " << OldCost << " vs NewCost: " << NewCost
1616+
<< "\n");
1617+
if (NewCost > OldCost)
1618+
return false;
1619+
1620+
// Clear unused sources to poison.
1621+
if (none_of(NewMask, [&](int M) { return 0 <= M && M < (int)NumSrcElts; }))
1622+
V0 = PoisonValue::get(ShuffleSrcTy);
1623+
if (none_of(NewMask, [&](int M) { return (int)NumSrcElts <= M; }))
1624+
V1 = PoisonValue::get(ShuffleSrcTy);
1625+
1626+
Value *Shuf = Builder.CreateShuffleVector(V0, V1, NewMask);
1627+
replaceValue(I, *Shuf);
1628+
return true;
1629+
}
1630+
15501631
/// Given a commutative reduction, the order of the input lanes does not alter
15511632
/// the results. We can use this to remove certain shuffles feeding the
15521633
/// reduction, removing the need to shuffle at all.
@@ -2102,6 +2183,7 @@ bool VectorCombine::run() {
21022183
case Instruction::ShuffleVector:
21032184
MadeChange |= foldShuffleOfBinops(I);
21042185
MadeChange |= foldShuffleOfCastops(I);
2186+
MadeChange |= foldShuffleOfShuffles(I);
21052187
MadeChange |= foldSelectShuffle(I);
21062188
break;
21072189
case Instruction::BitCast:

‎llvm/test/Transforms/VectorCombine/AArch64/select-shuffle.ll

Lines changed: 5 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,5 @@
11
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2-
; RUN: opt -passes='vector-combine' -S %s | FileCheck %s
2+
; RUN: opt -passes=vector-combine -S %s | FileCheck %s
33

44
target triple = "aarch64"
55

@@ -741,37 +741,29 @@ define i32 @full_reorder(ptr nocapture noundef readonly %pix1, i32 noundef %i_pi
741741
; CHECK-NEXT: [[TMP10:%.*]] = load <4 x i8>, ptr [[ARRAYIDX3_2]], align 1
742742
; CHECK-NEXT: [[TMP11:%.*]] = load <4 x i8>, ptr [[ARRAYIDX5_2]], align 1
743743
; CHECK-NEXT: [[TMP12:%.*]] = load <4 x i8>, ptr [[ADD_PTR_2]], align 1
744-
; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <4 x i8> [[TMP12]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
745-
; CHECK-NEXT: [[TMP14:%.*]] = shufflevector <4 x i8> [[TMP8]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
746-
; CHECK-NEXT: [[TMP15:%.*]] = shufflevector <16 x i8> [[TMP13]], <16 x i8> [[TMP14]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
744+
; CHECK-NEXT: [[TMP15:%.*]] = shufflevector <4 x i8> [[TMP12]], <4 x i8> [[TMP8]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
747745
; CHECK-NEXT: [[TMP16:%.*]] = shufflevector <4 x i8> [[TMP4]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
748746
; CHECK-NEXT: [[TMP17:%.*]] = shufflevector <16 x i8> [[TMP15]], <16 x i8> [[TMP16]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
749747
; CHECK-NEXT: [[TMP18:%.*]] = shufflevector <4 x i8> [[TMP0]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
750748
; CHECK-NEXT: [[TMP19:%.*]] = shufflevector <16 x i8> [[TMP17]], <16 x i8> [[TMP18]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
751749
; CHECK-NEXT: [[TMP20:%.*]] = zext <16 x i8> [[TMP19]] to <16 x i32>
752750
; CHECK-NEXT: [[TMP21:%.*]] = load <4 x i8>, ptr [[ADD_PTR64_2]], align 1
753-
; CHECK-NEXT: [[TMP22:%.*]] = shufflevector <4 x i8> [[TMP21]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
754-
; CHECK-NEXT: [[TMP23:%.*]] = shufflevector <4 x i8> [[TMP9]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
755-
; CHECK-NEXT: [[TMP24:%.*]] = shufflevector <16 x i8> [[TMP22]], <16 x i8> [[TMP23]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
751+
; CHECK-NEXT: [[TMP24:%.*]] = shufflevector <4 x i8> [[TMP21]], <4 x i8> [[TMP9]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
756752
; CHECK-NEXT: [[TMP25:%.*]] = shufflevector <4 x i8> [[TMP5]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
757753
; CHECK-NEXT: [[TMP26:%.*]] = shufflevector <16 x i8> [[TMP24]], <16 x i8> [[TMP25]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
758754
; CHECK-NEXT: [[TMP27:%.*]] = shufflevector <4 x i8> [[TMP1]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
759755
; CHECK-NEXT: [[TMP28:%.*]] = shufflevector <16 x i8> [[TMP26]], <16 x i8> [[TMP27]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
760756
; CHECK-NEXT: [[TMP29:%.*]] = zext <16 x i8> [[TMP28]] to <16 x i32>
761757
; CHECK-NEXT: [[TMP30:%.*]] = sub nsw <16 x i32> [[TMP20]], [[TMP29]]
762758
; CHECK-NEXT: [[TMP31:%.*]] = load <4 x i8>, ptr [[ARRAYIDX3_3]], align 1
763-
; CHECK-NEXT: [[TMP32:%.*]] = shufflevector <4 x i8> [[TMP31]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
764-
; CHECK-NEXT: [[TMP33:%.*]] = shufflevector <4 x i8> [[TMP10]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
765-
; CHECK-NEXT: [[TMP34:%.*]] = shufflevector <16 x i8> [[TMP32]], <16 x i8> [[TMP33]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
759+
; CHECK-NEXT: [[TMP34:%.*]] = shufflevector <4 x i8> [[TMP31]], <4 x i8> [[TMP10]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
766760
; CHECK-NEXT: [[TMP35:%.*]] = shufflevector <4 x i8> [[TMP6]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
767761
; CHECK-NEXT: [[TMP36:%.*]] = shufflevector <16 x i8> [[TMP34]], <16 x i8> [[TMP35]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
768762
; CHECK-NEXT: [[TMP37:%.*]] = shufflevector <4 x i8> [[TMP2]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
769763
; CHECK-NEXT: [[TMP38:%.*]] = shufflevector <16 x i8> [[TMP36]], <16 x i8> [[TMP37]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
770764
; CHECK-NEXT: [[TMP39:%.*]] = zext <16 x i8> [[TMP38]] to <16 x i32>
771765
; CHECK-NEXT: [[TMP40:%.*]] = load <4 x i8>, ptr [[ARRAYIDX5_3]], align 1
772-
; CHECK-NEXT: [[TMP41:%.*]] = shufflevector <4 x i8> [[TMP40]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
773-
; CHECK-NEXT: [[TMP42:%.*]] = shufflevector <4 x i8> [[TMP11]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
774-
; CHECK-NEXT: [[TMP43:%.*]] = shufflevector <16 x i8> [[TMP41]], <16 x i8> [[TMP42]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
766+
; CHECK-NEXT: [[TMP43:%.*]] = shufflevector <4 x i8> [[TMP40]], <4 x i8> [[TMP11]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
775767
; CHECK-NEXT: [[TMP44:%.*]] = shufflevector <4 x i8> [[TMP7]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
776768
; CHECK-NEXT: [[TMP45:%.*]] = shufflevector <16 x i8> [[TMP43]], <16 x i8> [[TMP44]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
777769
; CHECK-NEXT: [[TMP46:%.*]] = shufflevector <4 x i8> [[TMP3]], <4 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>

‎llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll

Lines changed: 2 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -1,14 +1,11 @@
11
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2-
; RUN: opt -passes='vector-combine' -S %s | FileCheck %s
2+
; RUN: opt -passes=vector-combine -S %s | FileCheck %s
33

44
target triple = "aarch64"
55

66
define <8 x i8> @trivial(<8 x i8> %a) {
77
; CHECK-LABEL: @trivial(
8-
; CHECK-NEXT: [[AB:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
9-
; CHECK-NEXT: [[AT:%.*]] = shufflevector <8 x i8> [[A]], <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>
10-
; CHECK-NEXT: [[R:%.*]] = shufflevector <4 x i8> [[AT]], <4 x i8> [[AB]], <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
11-
; CHECK-NEXT: ret <8 x i8> [[R]]
8+
; CHECK-NEXT: ret <8 x i8> [[R:%.*]]
129
;
1310
%ab = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
1411
%at = shufflevector <8 x i8> %a, <8 x i8> poison, <4 x i32> <i32 7, i32 6, i32 5, i32 4>

‎llvm/test/Transforms/VectorCombine/X86/pr67803.ll

Lines changed: 1 addition & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -6,10 +6,7 @@ define <4 x i64> @PR67803(<8 x i32> %x, <8 x i32> %y, <8 x float> %a, <8 x float
66
; CHECK-LABEL: @PR67803(
77
; CHECK-NEXT: entry:
88
; CHECK-NEXT: [[CMP:%.*]] = icmp sgt <8 x i32> [[X:%.*]], [[Y:%.*]]
9-
; CHECK-NEXT: [[CMP_LO:%.*]] = shufflevector <8 x i1> [[CMP]], <8 x i1> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
10-
; CHECK-NEXT: [[CMP_HI:%.*]] = shufflevector <8 x i1> [[CMP]], <8 x i1> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
11-
; CHECK-NEXT: [[TMP0:%.*]] = shufflevector <4 x i1> [[CMP_LO]], <4 x i1> [[CMP_HI]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
12-
; CHECK-NEXT: [[TMP1:%.*]] = sext <8 x i1> [[TMP0]] to <8 x i32>
9+
; CHECK-NEXT: [[TMP1:%.*]] = sext <8 x i1> [[CMP]] to <8 x i32>
1310
; CHECK-NEXT: [[CONCAT:%.*]] = bitcast <8 x i32> [[TMP1]] to <4 x i64>
1411
; CHECK-NEXT: [[MASK:%.*]] = bitcast <4 x i64> [[CONCAT]] to <8 x float>
1512
; CHECK-NEXT: [[SEL:%.*]] = tail call noundef <8 x float> @llvm.x86.avx.blendv.ps.256(<8 x float> [[A:%.*]], <8 x float> [[B:%.*]], <8 x float> [[MASK]])

‎llvm/test/Transforms/VectorCombine/X86/shuffle-of-shuffles.ll

Lines changed: 4 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -2,15 +2,12 @@
22
; RUN: opt < %s -passes=vector-combine -S -mtriple=x86_64-- -mattr=sse2 | FileCheck %s
33
; RUN: opt < %s -passes=vector-combine -S -mtriple=x86_64-- -mattr=avx2 | FileCheck %s
44

5-
; TODO: fold to identity
5+
; fold to identity
66

77
define <8 x i32> @concat_extract_subvectors(<8 x i32> %x) {
88
; CHECK-LABEL: define <8 x i32> @concat_extract_subvectors(
99
; CHECK-SAME: <8 x i32> [[X:%.*]]) #[[ATTR0:[0-9]+]] {
10-
; CHECK-NEXT: [[LO:%.*]] = shufflevector <8 x i32> [[X]], <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
11-
; CHECK-NEXT: [[HI:%.*]] = shufflevector <8 x i32> [[X]], <8 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
12-
; CHECK-NEXT: [[CONCAT:%.*]] = shufflevector <4 x i32> [[LO]], <4 x i32> [[HI]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
13-
; CHECK-NEXT: ret <8 x i32> [[CONCAT]]
10+
; CHECK-NEXT: ret <8 x i32> [[X]]
1411
;
1512
%lo = shufflevector <8 x i32> %x, <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
1613
%hi = shufflevector <8 x i32> %x, <8 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
@@ -34,15 +31,12 @@ define <8 x i32> @concat_extract_subvectors_undef(<8 x i32> %x) {
3431
ret <8 x i32> %concat
3532
}
3633

37-
; negative test - shuffle contains poision
34+
; shuffle contains poison
3835

3936
define <8 x i32> @concat_extract_subvectors_poison(<8 x i32> %x) {
4037
; CHECK-LABEL: define <8 x i32> @concat_extract_subvectors_poison(
4138
; CHECK-SAME: <8 x i32> [[X:%.*]]) #[[ATTR0]] {
42-
; CHECK-NEXT: [[LO:%.*]] = shufflevector <8 x i32> [[X]], <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 8>
43-
; CHECK-NEXT: [[HI:%.*]] = shufflevector <8 x i32> [[X]], <8 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 8>
44-
; CHECK-NEXT: [[CONCAT:%.*]] = shufflevector <4 x i32> [[LO]], <4 x i32> [[HI]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
45-
; CHECK-NEXT: ret <8 x i32> [[CONCAT]]
39+
; CHECK-NEXT: ret <8 x i32> [[X]]
4640
;
4741
%lo = shufflevector <8 x i32> %x, <8 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 8>
4842
%hi = shufflevector <8 x i32> %x, <8 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 8>

0 commit comments

Comments
 (0)