[llvm] Revert "[LoopVectorize] Add support for reverse loops in isDereferenceableAndAlignedInLoop (#96752)" (PR #123057)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Jan 15 05:56:27 PST 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-analysis
Author: David Sherwood (david-arm)
<details>
<summary>Changes</summary>
This reverts commit bfedf6460c2cad6e6f966b457d8d27084579dcd8.
---
Patch is 39.30 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/123057.diff
5 Files Affected:
- (modified) llvm/include/llvm/Analysis/LoopAccessAnalysis.h (-19)
- (modified) llvm/lib/Analysis/Loads.cpp (+52-57)
- (modified) llvm/lib/Analysis/LoopAccessAnalysis.cpp (+35-26)
- (modified) llvm/test/Transforms/LoopVectorize/X86/load-deref-pred.ll (+167-71)
- (modified) llvm/test/Transforms/LoopVectorize/load-deref-pred-align.ll (+12-15)
``````````diff
diff --git a/llvm/include/llvm/Analysis/LoopAccessAnalysis.h b/llvm/include/llvm/Analysis/LoopAccessAnalysis.h
index 6fc6ca14d08895..31374a128856c7 100644
--- a/llvm/include/llvm/Analysis/LoopAccessAnalysis.h
+++ b/llvm/include/llvm/Analysis/LoopAccessAnalysis.h
@@ -853,25 +853,6 @@ bool sortPtrAccesses(ArrayRef<Value *> VL, Type *ElemTy, const DataLayout &DL,
bool isConsecutiveAccess(Value *A, Value *B, const DataLayout &DL,
ScalarEvolution &SE, bool CheckType = true);
-/// Calculate Start and End points of memory access.
-/// Let's assume A is the first access and B is a memory access on N-th loop
-/// iteration. Then B is calculated as:
-/// B = A + Step*N .
-/// Step value may be positive or negative.
-/// N is a calculated back-edge taken count:
-/// N = (TripCount > 0) ? RoundDown(TripCount -1 , VF) : 0
-/// Start and End points are calculated in the following way:
-/// Start = UMIN(A, B) ; End = UMAX(A, B) + SizeOfElt,
-/// where SizeOfElt is the size of single memory access in bytes.
-///
-/// There is no conflict when the intervals are disjoint:
-/// NoConflict = (P2.Start >= P1.End) || (P1.Start >= P2.End)
-std::pair<const SCEV *, const SCEV *> getStartAndEndForAccess(
- const Loop *Lp, const SCEV *PtrExpr, Type *AccessTy, const SCEV *MaxBECount,
- ScalarEvolution *SE,
- DenseMap<std::pair<const SCEV *, Type *>,
- std::pair<const SCEV *, const SCEV *>> *PointerBounds);
-
class LoopAccessInfoManager {
/// The cache.
DenseMap<Loop *, std::unique_ptr<LoopAccessInfo>> LoopAccessInfoMap;
diff --git a/llvm/lib/Analysis/Loads.cpp b/llvm/lib/Analysis/Loads.cpp
index cc6760292c2ff1..7bbd469bd035d3 100644
--- a/llvm/lib/Analysis/Loads.cpp
+++ b/llvm/lib/Analysis/Loads.cpp
@@ -13,7 +13,6 @@
#include "llvm/Analysis/Loads.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/AssumeBundleQueries.h"
-#include "llvm/Analysis/LoopAccessAnalysis.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/MemoryBuiltins.h"
#include "llvm/Analysis/MemoryLocation.h"
@@ -276,88 +275,84 @@ static bool AreEquivalentAddressValues(const Value *A, const Value *B) {
bool llvm::isDereferenceableAndAlignedInLoop(
LoadInst *LI, Loop *L, ScalarEvolution &SE, DominatorTree &DT,
AssumptionCache *AC, SmallVectorImpl<const SCEVPredicate *> *Predicates) {
- const Align Alignment = LI->getAlign();
auto &DL = LI->getDataLayout();
Value *Ptr = LI->getPointerOperand();
+
APInt EltSize(DL.getIndexTypeSizeInBits(Ptr->getType()),
DL.getTypeStoreSize(LI->getType()).getFixedValue());
+ const Align Alignment = LI->getAlign();
+
+ Instruction *HeaderFirstNonPHI = L->getHeader()->getFirstNonPHI();
// If given a uniform (i.e. non-varying) address, see if we can prove the
// access is safe within the loop w/o needing predication.
if (L->isLoopInvariant(Ptr))
- return isDereferenceableAndAlignedPointer(
- Ptr, Alignment, EltSize, DL, L->getHeader()->getFirstNonPHI(), AC, &DT);
-
- const SCEV *PtrScev = SE.getSCEV(Ptr);
- auto *AddRec = dyn_cast<SCEVAddRecExpr>(PtrScev);
+ return isDereferenceableAndAlignedPointer(Ptr, Alignment, EltSize, DL,
+ HeaderFirstNonPHI, AC, &DT);
- // Check to see if we have a repeating access pattern and it's possible
- // to prove all accesses are well aligned.
+ // Otherwise, check to see if we have a repeating access pattern where we can
+ // prove that all accesses are well aligned and dereferenceable.
+ auto *AddRec = dyn_cast<SCEVAddRecExpr>(SE.getSCEV(Ptr));
if (!AddRec || AddRec->getLoop() != L || !AddRec->isAffine())
return false;
-
auto* Step = dyn_cast<SCEVConstant>(AddRec->getStepRecurrence(SE));
if (!Step)
return false;
- // For the moment, restrict ourselves to the case where the access size is a
- // multiple of the requested alignment and the base is aligned.
- // TODO: generalize if a case found which warrants
- if (EltSize.urem(Alignment.value()) != 0)
+ auto TC = SE.getSmallConstantMaxTripCount(L, Predicates);
+ if (!TC)
return false;
// TODO: Handle overlapping accesses.
- if (EltSize.ugt(Step->getAPInt().abs()))
- return false;
-
- const SCEV *MaxBECount =
- SE.getPredicatedConstantMaxBackedgeTakenCount(L, *Predicates);
- if (isa<SCEVCouldNotCompute>(MaxBECount))
- return false;
-
- const auto &[AccessStart, AccessEnd] = getStartAndEndForAccess(
- L, PtrScev, LI->getType(), MaxBECount, &SE, nullptr);
- if (isa<SCEVCouldNotCompute>(AccessStart) ||
- isa<SCEVCouldNotCompute>(AccessEnd))
+ // We should be computing AccessSize as (TC - 1) * Step + EltSize.
+ if (EltSize.sgt(Step->getAPInt()))
return false;
- // Try to get the access size.
- const SCEV *PtrDiff = SE.getMinusSCEV(AccessEnd, AccessStart);
- APInt MaxPtrDiff = SE.getUnsignedRangeMax(PtrDiff);
+ // Compute the total access size for access patterns with unit stride and
+ // patterns with gaps. For patterns with unit stride, Step and EltSize are the
+ // same.
+ // For patterns with gaps (i.e. non unit stride), we are
+ // accessing EltSize bytes at every Step.
+ APInt AccessSize = TC * Step->getAPInt();
+ assert(SE.isLoopInvariant(AddRec->getStart(), L) &&
+ "implied by addrec definition");
Value *Base = nullptr;
- APInt AccessSize;
- if (const SCEVUnknown *NewBase = dyn_cast<SCEVUnknown>(AccessStart)) {
- Base = NewBase->getValue();
- AccessSize = MaxPtrDiff;
- } else if (auto *MinAdd = dyn_cast<SCEVAddExpr>(AccessStart)) {
- if (MinAdd->getNumOperands() != 2)
- return false;
-
- const auto *Offset = dyn_cast<SCEVConstant>(MinAdd->getOperand(0));
- const auto *NewBase = dyn_cast<SCEVUnknown>(MinAdd->getOperand(1));
- if (!Offset || !NewBase)
- return false;
-
- // The following code below assumes the offset is unsigned, but GEP
- // offsets are treated as signed so we can end up with a signed value
- // here too. For example, suppose the initial PHI value is (i8 255),
- // the offset will be treated as (i8 -1) and sign-extended to (i64 -1).
- if (Offset->getAPInt().isNegative())
- return false;
+ if (auto *StartS = dyn_cast<SCEVUnknown>(AddRec->getStart())) {
+ Base = StartS->getValue();
+ } else if (auto *StartS = dyn_cast<SCEVAddExpr>(AddRec->getStart())) {
+ // Handle (NewBase + offset) as start value.
+ const auto *Offset = dyn_cast<SCEVConstant>(StartS->getOperand(0));
+ const auto *NewBase = dyn_cast<SCEVUnknown>(StartS->getOperand(1));
+ if (StartS->getNumOperands() == 2 && Offset && NewBase) {
+ // The following code below assumes the offset is unsigned, but GEP
+ // offsets are treated as signed so we can end up with a signed value
+ // here too. For example, suppose the initial PHI value is (i8 255),
+ // the offset will be treated as (i8 -1) and sign-extended to (i64 -1).
+ if (Offset->getAPInt().isNegative())
+ return false;
- // For the moment, restrict ourselves to the case where the offset is a
- // multiple of the requested alignment and the base is aligned.
- // TODO: generalize if a case found which warrants
- if (Offset->getAPInt().urem(Alignment.value()) != 0)
- return false;
+ // For the moment, restrict ourselves to the case where the offset is a
+ // multiple of the requested alignment and the base is aligned.
+ // TODO: generalize if a case found which warrants
+ if (Offset->getAPInt().urem(Alignment.value()) != 0)
+ return false;
+ Base = NewBase->getValue();
+ bool Overflow = false;
+ AccessSize = AccessSize.uadd_ov(Offset->getAPInt(), Overflow);
+ if (Overflow)
+ return false;
+ }
+ }
- AccessSize = MaxPtrDiff + Offset->getAPInt();
- Base = NewBase->getValue();
- } else
+ if (!Base)
return false;
- Instruction *HeaderFirstNonPHI = L->getHeader()->getFirstNonPHI();
+ // For the moment, restrict ourselves to the case where the access size is a
+ // multiple of the requested alignment and the base is aligned.
+ // TODO: generalize if a case found which warrants
+ if (EltSize.urem(Alignment.value()) != 0)
+ return false;
return isDereferenceableAndAlignedPointer(Base, Alignment, AccessSize, DL,
HeaderFirstNonPHI, AC, &DT);
}
diff --git a/llvm/lib/Analysis/LoopAccessAnalysis.cpp b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
index 11e0a221fc8878..2a68979add666d 100644
--- a/llvm/lib/Analysis/LoopAccessAnalysis.cpp
+++ b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
@@ -190,20 +190,31 @@ RuntimeCheckingPtrGroup::RuntimeCheckingPtrGroup(
Members.push_back(Index);
}
-std::pair<const SCEV *, const SCEV *> llvm::getStartAndEndForAccess(
- const Loop *Lp, const SCEV *PtrExpr, Type *AccessTy, const SCEV *MaxBECount,
- ScalarEvolution *SE,
+/// Calculate Start and End points of memory access.
+/// Let's assume A is the first access and B is a memory access on N-th loop
+/// iteration. Then B is calculated as:
+/// B = A + Step*N .
+/// Step value may be positive or negative.
+/// N is a calculated back-edge taken count:
+/// N = (TripCount > 0) ? RoundDown(TripCount -1 , VF) : 0
+/// Start and End points are calculated in the following way:
+/// Start = UMIN(A, B) ; End = UMAX(A, B) + SizeOfElt,
+/// where SizeOfElt is the size of single memory access in bytes.
+///
+/// There is no conflict when the intervals are disjoint:
+/// NoConflict = (P2.Start >= P1.End) || (P1.Start >= P2.End)
+static std::pair<const SCEV *, const SCEV *> getStartAndEndForAccess(
+ const Loop *Lp, const SCEV *PtrExpr, Type *AccessTy,
+ PredicatedScalarEvolution &PSE,
DenseMap<std::pair<const SCEV *, Type *>,
- std::pair<const SCEV *, const SCEV *>> *PointerBounds) {
- std::pair<const SCEV *, const SCEV *> *PtrBoundsPair;
- if (PointerBounds) {
- auto [Iter, Ins] = PointerBounds->insert(
- {{PtrExpr, AccessTy},
- {SE->getCouldNotCompute(), SE->getCouldNotCompute()}});
- if (!Ins)
- return Iter->second;
- PtrBoundsPair = &Iter->second;
- }
+ std::pair<const SCEV *, const SCEV *>> &PointerBounds) {
+ ScalarEvolution *SE = PSE.getSE();
+
+ auto [Iter, Ins] = PointerBounds.insert(
+ {{PtrExpr, AccessTy},
+ {SE->getCouldNotCompute(), SE->getCouldNotCompute()}});
+ if (!Ins)
+ return Iter->second;
const SCEV *ScStart;
const SCEV *ScEnd;
@@ -211,8 +222,10 @@ std::pair<const SCEV *, const SCEV *> llvm::getStartAndEndForAccess(
if (SE->isLoopInvariant(PtrExpr, Lp)) {
ScStart = ScEnd = PtrExpr;
} else if (auto *AR = dyn_cast<SCEVAddRecExpr>(PtrExpr)) {
+ const SCEV *Ex = PSE.getSymbolicMaxBackedgeTakenCount();
+
ScStart = AR->getStart();
- ScEnd = AR->evaluateAtIteration(MaxBECount, *SE);
+ ScEnd = AR->evaluateAtIteration(Ex, *SE);
const SCEV *Step = AR->getStepRecurrence(*SE);
// For expressions with negative step, the upper bound is ScStart and the
@@ -231,7 +244,7 @@ std::pair<const SCEV *, const SCEV *> llvm::getStartAndEndForAccess(
return {SE->getCouldNotCompute(), SE->getCouldNotCompute()};
assert(SE->isLoopInvariant(ScStart, Lp) && "ScStart needs to be invariant");
- assert(SE->isLoopInvariant(ScEnd, Lp) && "ScEnd needs to be invariant");
+ assert(SE->isLoopInvariant(ScEnd, Lp)&& "ScEnd needs to be invariant");
// Add the size of the pointed element to ScEnd.
auto &DL = Lp->getHeader()->getDataLayout();
@@ -239,10 +252,8 @@ std::pair<const SCEV *, const SCEV *> llvm::getStartAndEndForAccess(
const SCEV *EltSizeSCEV = SE->getStoreSizeOfExpr(IdxTy, AccessTy);
ScEnd = SE->getAddExpr(ScEnd, EltSizeSCEV);
- std::pair<const SCEV *, const SCEV *> Res = {ScStart, ScEnd};
- if (PointerBounds)
- *PtrBoundsPair = Res;
- return Res;
+ Iter->second = {ScStart, ScEnd};
+ return Iter->second;
}
/// Calculate Start and End points of memory access using
@@ -252,9 +263,8 @@ void RuntimePointerChecking::insert(Loop *Lp, Value *Ptr, const SCEV *PtrExpr,
unsigned DepSetId, unsigned ASId,
PredicatedScalarEvolution &PSE,
bool NeedsFreeze) {
- const SCEV *MaxBECount = PSE.getSymbolicMaxBackedgeTakenCount();
const auto &[ScStart, ScEnd] = getStartAndEndForAccess(
- Lp, PtrExpr, AccessTy, MaxBECount, PSE.getSE(), &DC.getPointerBounds());
+ Lp, PtrExpr, AccessTy, PSE, DC.getPointerBounds());
assert(!isa<SCEVCouldNotCompute>(ScStart) &&
!isa<SCEVCouldNotCompute>(ScEnd) &&
"must be able to compute both start and end expressions");
@@ -1928,11 +1938,10 @@ MemoryDepChecker::getDependenceDistanceStrideAndSize(
// required for correctness.
if (SE.isLoopInvariant(Src, InnermostLoop) ||
SE.isLoopInvariant(Sink, InnermostLoop)) {
- const SCEV *MaxBECount = PSE.getSymbolicMaxBackedgeTakenCount();
- const auto &[SrcStart_, SrcEnd_] = getStartAndEndForAccess(
- InnermostLoop, Src, ATy, MaxBECount, PSE.getSE(), &PointerBounds);
- const auto &[SinkStart_, SinkEnd_] = getStartAndEndForAccess(
- InnermostLoop, Sink, BTy, MaxBECount, PSE.getSE(), &PointerBounds);
+ const auto &[SrcStart_, SrcEnd_] =
+ getStartAndEndForAccess(InnermostLoop, Src, ATy, PSE, PointerBounds);
+ const auto &[SinkStart_, SinkEnd_] =
+ getStartAndEndForAccess(InnermostLoop, Sink, BTy, PSE, PointerBounds);
if (!isa<SCEVCouldNotCompute>(SrcStart_) &&
!isa<SCEVCouldNotCompute>(SrcEnd_) &&
!isa<SCEVCouldNotCompute>(SinkStart_) &&
diff --git a/llvm/test/Transforms/LoopVectorize/X86/load-deref-pred.ll b/llvm/test/Transforms/LoopVectorize/X86/load-deref-pred.ll
index 3e50ee42866b9b..1433e48690bc60 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/load-deref-pred.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/load-deref-pred.ll
@@ -2920,8 +2920,8 @@ loop_exit:
ret i32 %accum.next
}
-define i32 @test_non_unit_stride_off_by_four_bytes(i64 %len, ptr %test_base) {
-; CHECK-LABEL: @test_non_unit_stride_off_by_four_bytes(
+define i32 @neg_test_non_unit_stride_off_by_four_bytes(i64 %len, ptr %test_base) {
+; CHECK-LABEL: @neg_test_non_unit_stride_off_by_four_bytes(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[ALLOCA:%.*]] = alloca [103 x i32], align 4
; CHECK-NEXT: call void @init(ptr [[ALLOCA]])
@@ -2929,11 +2929,11 @@ define i32 @test_non_unit_stride_off_by_four_bytes(i64 %len, ptr %test_base) {
; CHECK: vector.ph:
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
-; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP112:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP113:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI2:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP114:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP115:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE33:%.*]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP144:%.*]], [[PRED_LOAD_CONTINUE33]] ]
+; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP145:%.*]], [[PRED_LOAD_CONTINUE33]] ]
+; CHECK-NEXT: [[VEC_PHI2:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP146:%.*]], [[PRED_LOAD_CONTINUE33]] ]
+; CHECK-NEXT: [[VEC_PHI3:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP147:%.*]], [[PRED_LOAD_CONTINUE33]] ]
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 2
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[OFFSET_IDX]], 0
; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[OFFSET_IDX]], 2
@@ -2999,74 +2999,170 @@ define i32 @test_non_unit_stride_off_by_four_bytes(i64 %len, ptr %test_base) {
; CHECK-NEXT: [[TMP61:%.*]] = insertelement <4 x i1> [[TMP60]], i1 [[TMP57]], i32 1
; CHECK-NEXT: [[TMP62:%.*]] = insertelement <4 x i1> [[TMP61]], i1 [[TMP58]], i32 2
; CHECK-NEXT: [[TMP63:%.*]] = insertelement <4 x i1> [[TMP62]], i1 [[TMP59]], i32 3
-; CHECK-NEXT: [[TMP64:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP0]]
-; CHECK-NEXT: [[TMP65:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP1]]
-; CHECK-NEXT: [[TMP66:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP2]]
-; CHECK-NEXT: [[TMP67:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP3]]
-; CHECK-NEXT: [[TMP68:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP4]]
-; CHECK-NEXT: [[TMP69:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP5]]
-; CHECK-NEXT: [[TMP70:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP6]]
-; CHECK-NEXT: [[TMP71:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP7]]
-; CHECK-NEXT: [[TMP72:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP8]]
-; CHECK-NEXT: [[TMP73:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP9]]
-; CHECK-NEXT: [[TMP74:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP10]]
-; CHECK-NEXT: [[TMP75:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP11]]
-; CHECK-NEXT: [[TMP76:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP12]]
-; CHECK-NEXT: [[TMP77:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP13]]
-; CHECK-NEXT: [[TMP78:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP14]]
-; CHECK-NEXT: [[TMP79:%.*]] = getelementptr inbounds i32, ptr [[ALLOCA]], i64 [[TMP15]]
-; CHECK-NEXT: [[TMP80:%.*]] = load i32, ptr [[TMP64]], align 4
-; CHECK-NEXT: [[TMP81:%.*]] = load i32, ptr [[TMP65]], align 4
-; CHECK-NEXT: [[TMP82:%.*]] = load i32, ptr [[TMP66]], align 4
-; CHECK-NEXT: [[TMP83:%.*]] = load i32, ptr [[TMP67]], align 4
-; CHECK-NEXT: [[TMP84:%.*]] = insertelement <4 x i32> poison, i32 [[TMP80]], i32 0
-; CHECK-NEXT: [[TMP85:%.*]] = insertelement <4 x i32> [[TMP84]], i32 [[TMP81]], i32 1
-; CHECK-NEXT: [[TMP86:%.*]] = insertelement <4 x i32> [[TMP85]], i32 [[TMP82]], i32 2
-; CHECK-NEXT: [[TMP87:%.*]] = insertelement <4 x i32> [[TMP86]], i32 [[TMP83]], i32 3
-; CHECK-NEXT: [[TMP88:%.*]] = load i32, ptr [[TMP68]], align 4
-; CHECK-NEXT: [[TMP89:%.*]] = load i32, ptr [[TMP69]], align 4
-; CHECK-NEXT: [[TMP90:%.*]] = load i32, ptr [[TMP70]], align 4
-; CHECK-NEXT: [[TMP91:%.*]] = load i32, ptr [[TMP71]], align 4
-; CHECK-NEXT: [[TMP92:%.*]] = insertelement <4 x i32> poison, i32 [[TMP88]], i32 0
-; CHECK-NEXT: [[TMP93:%.*]] = insertelement <4 x i32> [[TMP92]], i32 [[TMP89]], i32 1
-; CHECK-NEXT: [[TMP94:%.*]] = insertelement <4 x i32> [[TMP93]], i32 [[TMP90]], i32 2
-; CHECK-NEXT: [[TMP95:%.*]] = insertelement <4 x i32> [[TMP94]], i32 [[TMP91]], i32 3
-; CHECK-NEXT: [[TMP96:%.*]] = load i32, ptr [[TMP72]], align 4
-; CHECK-NEXT: [[TMP97:%.*]] = load i32, ptr [[TMP73]], align 4
-; CHECK-NEXT: [[TMP98:%.*]] = load i32, ptr [[TMP74]], align 4
-; CHECK-NEXT: [[TMP99:%.*]] = load i32, ptr [[TMP75]], align 4
-; CHECK-NEXT: [[TMP100:%.*]] = insertelement <4 x i32> poison, i32 [[TMP96]], i32 0
-; CHECK-NEXT: [[TMP101:%.*]] = insertelement <4 x i32> [[TMP100]], i32 [[TMP97]], i32 1
-; CHECK-NEXT: [[TMP102:%.*]] = insertelement <4 x i32> [[TMP101]], i32 [[TMP98]], i32 2
-; CHECK-NEXT: [[TMP103:%.*]] = insertelement <4 x i32> [[TMP102]], i32 [[TMP99]], i32 3
-; CHECK-NEXT: [[TMP104:%.*]] = load i32, ptr [[TMP76]], align 4
-; CHECK-NEXT: [[TMP105:%.*]] = load i32, ptr [[TMP77]], align 4
-; CHECK-NEXT: [[TMP106:%.*]] = load i32, ptr [[TMP78]], align 4
-; CHECK-NEXT: [[TMP107:%.*]] = load i32, ptr [[TMP79]], align 4
-; CHECK-NEXT: [[TMP108:%.*]] = insertelement <4 x i32> poison, i32 [[TMP104]], i32 0
-; CHECK-NEXT: [[TMP109:%.*]] = insertelement <4 x i32> [[TMP108]], i32 [[TMP105]], i32 1
-; CHECK-NEXT: [[TMP110:%.*]] = insertelement <4 x i32> [[TMP109]], i32 [[TMP106]], i32 2
-; CHECK-NEXT: [[TMP111:%.*]] = insertelement <4 x i32> [[TMP110]], i32 [[TMP107]], i32 3
-; CHECK-NEXT: ...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/123057
More information about the llvm-commits
mailing list