[SLP] Fix CommonMask will be transformed into an incorrect mask if createShuffle is called multiple times. #124244
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
createShuffle is called multiple times. We have two types of mask in SLP: a scalar mask and a vector mask. When vectorizing four i32 additions into <4 x i32>, SLP creates a mask of length 4. When vectorizing four <2 x i32> additions into <8 x i32>, SLP also creates a mask of length 4. We refer to the first case as a scalar mask (because the mask element represents a scalar, i32), and the second case as a vector mask (because the mask element represents a vector, <4 x i32>). At some point, we must convert the scalar mask into a vector mask (otherwise, calling TTI cost functions or IRBuilderBase functions may yield incorrect results). Since both ShuffleCostEstimator and ShuffleInstructionBuilder can modify the CommonMask, we have decided to perform the mask transformation only within createShuffle. However, we do not store the transformed result, as createShuffle may be called multiple times.
|
@llvm/pr-subscribers-llvm-transforms @llvm/pr-subscribers-vectorizers Author: Han-Kuan Chen (HanKuanChen) ChangesWe have two types of mask in SLP: a scalar mask and a vector mask. Full diff: https://github.com/llvm/llvm-project/pull/124244.diff 3 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index c98d872fb6467f..8d81fb1427613b 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -9374,8 +9374,15 @@ class BaseShuffleAnalysis {
/// instruction.
template <typename T, typename ShuffleBuilderTy>
static T createShuffle(Value *V1, Value *V2, ArrayRef<int> Mask,
- ShuffleBuilderTy &Builder) {
+ ShuffleBuilderTy &Builder, Type *ScalarTy) {
assert(V1 && "Expected at least one vector value.");
+ unsigned ScalarTyNumElements = getNumElements(ScalarTy);
+ SmallVector<int> NewMask(Mask);
+ if (ScalarTyNumElements != 1) {
+ assert(SLPReVec && "FixedVectorType is not expected.");
+ transformScalarShuffleIndiciesToVector(ScalarTyNumElements, NewMask);
+ Mask = NewMask;
+ }
if (V2)
Builder.resizeToMatch(V1, V2);
int VF = Mask.size();
@@ -9478,7 +9485,6 @@ class BaseShuffleAnalysis {
if (isa<PoisonValue>(V1))
return Builder.createPoison(
cast<VectorType>(V1->getType())->getElementType(), Mask.size());
- SmallVector<int> NewMask(Mask);
bool IsIdentity = peekThroughShuffles(V1, NewMask, /*SinglePermute=*/true);
assert(V1 && "Expected non-null value after looking through shuffles.");
@@ -10643,17 +10649,12 @@ class BoUpSLP::ShuffleCostEstimator : public BaseShuffleAnalysis {
V2 = getAllOnesValue(*R.DL, getWidenedType(ScalarTy, CommonVF));
}
}
- if (auto *VecTy = dyn_cast<FixedVectorType>(ScalarTy)) {
- assert(SLPReVec && "FixedVectorType is not expected.");
- transformScalarShuffleIndiciesToVector(VecTy->getNumElements(),
- CommonMask);
- }
InVectors.front() =
Constant::getNullValue(getWidenedType(ScalarTy, CommonMask.size()));
if (InVectors.size() == 2)
InVectors.pop_back();
return ExtraCost + BaseShuffleAnalysis::createShuffle<InstructionCost>(
- V1, V2, CommonMask, Builder);
+ V1, V2, CommonMask, Builder, ScalarTy);
}
public:
@@ -14198,8 +14199,8 @@ class BoUpSLP::ShuffleInstructionBuilder final : public BaseShuffleAnalysis {
assert(V1 && "Expected at least one vector value.");
ShuffleIRBuilder ShuffleBuilder(Builder, R.GatherShuffleExtractSeq,
R.CSEBlocks, *R.DL);
- return BaseShuffleAnalysis::createShuffle<Value *>(V1, V2, Mask,
- ShuffleBuilder);
+ return BaseShuffleAnalysis::createShuffle<Value *>(
+ V1, V2, Mask, ShuffleBuilder, ScalarTy);
}
/// Cast value \p V to the vector type with the same number of elements, but
@@ -14518,14 +14519,6 @@ class BoUpSLP::ShuffleInstructionBuilder final : public BaseShuffleAnalysis {
ArrayRef<int> SubVectorsMask, unsigned VF = 0,
function_ref<void(Value *&, SmallVectorImpl<int> &)> Action = {}) {
IsFinalized = true;
- unsigned ScalarTyNumElements = getNumElements(ScalarTy);
- SmallVector<int> NewExtMask(ExtMask);
- if (ScalarTyNumElements != 1) {
- assert(SLPReVec && "FixedVectorType is not expected.");
- transformScalarShuffleIndiciesToVector(ScalarTyNumElements, CommonMask);
- transformScalarShuffleIndiciesToVector(ScalarTyNumElements, NewExtMask);
- ExtMask = NewExtMask;
- }
if (Action) {
Value *Vec = InVectors.front();
if (InVectors.size() == 2) {
@@ -14566,17 +14559,15 @@ class BoUpSLP::ShuffleInstructionBuilder final : public BaseShuffleAnalysis {
return !isKnownNonNegative(
V, SimplifyQuery(*R.DL));
}));
- unsigned InsertionIndex = Idx * ScalarTyNumElements;
+ unsigned InsertionIndex = Idx * getNumElements(ScalarTy);
Vec = createInsertVector(
Builder, Vec, V, InsertionIndex,
std::bind(&ShuffleInstructionBuilder::createShuffle, this, _1, _2,
_3));
if (!CommonMask.empty()) {
- std::iota(
- std::next(CommonMask.begin(), InsertionIndex),
- std::next(CommonMask.begin(),
- (Idx + E->getVectorFactor()) * ScalarTyNumElements),
- InsertionIndex);
+ std::iota(std::next(CommonMask.begin(), Idx),
+ std::next(CommonMask.begin(), Idx + E->getVectorFactor()),
+ Idx);
}
}
return Vec;
diff --git a/llvm/test/Transforms/SLPVectorizer/RISCV/revec.ll b/llvm/test/Transforms/SLPVectorizer/RISCV/revec.ll
index 61ff4f5766d309..dd7a21198ac1f5 100644
--- a/llvm/test/Transforms/SLPVectorizer/RISCV/revec.ll
+++ b/llvm/test/Transforms/SLPVectorizer/RISCV/revec.ll
@@ -231,3 +231,40 @@ define ptr @test4() {
%28 = tail call float @llvm.sqrt.f32(float %26)
ret ptr null
}
+
+define i32 @test5() {
+; CHECK-LABEL: @test5(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[TMP0:%.*]] = call <4 x double> @llvm.vector.insert.v4f64.v2f64(<4 x double> poison, <2 x double> zeroinitializer, i64 0)
+; CHECK-NEXT: [[TMP1:%.*]] = call <4 x double> @llvm.vector.insert.v4f64.v2f64(<4 x double> [[TMP0]], <2 x double> zeroinitializer, i64 2)
+; CHECK-NEXT: [[TMP2:%.*]] = fdiv <4 x double> [[TMP1]], [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = call <8 x double> @llvm.vector.insert.v8f64.v2f64(<8 x double> poison, <2 x double> zeroinitializer, i64 0)
+; CHECK-NEXT: [[TMP4:%.*]] = call <8 x double> @llvm.vector.insert.v8f64.v2f64(<8 x double> [[TMP3]], <2 x double> zeroinitializer, i64 2)
+; CHECK-NEXT: [[TMP5:%.*]] = call <8 x double> @llvm.vector.insert.v8f64.v2f64(<8 x double> [[TMP4]], <2 x double> zeroinitializer, i64 4)
+; CHECK-NEXT: [[TMP6:%.*]] = call <8 x double> @llvm.vector.insert.v8f64.v2f64(<8 x double> [[TMP5]], <2 x double> zeroinitializer, i64 6)
+; CHECK-NEXT: [[TMP7:%.*]] = call <8 x double> @llvm.vector.insert.v8f64.v2f64(<8 x double> poison, <2 x double> zeroinitializer, i64 2)
+; CHECK-NEXT: [[TMP8:%.*]] = call <8 x double> @llvm.vector.insert.v8f64.v2f64(<8 x double> [[TMP7]], <2 x double> zeroinitializer, i64 6)
+; CHECK-NEXT: [[TMP9:%.*]] = call <8 x double> @llvm.vector.insert.v8f64.v4f64(<8 x double> poison, <4 x double> [[TMP2]], i64 0)
+; CHECK-NEXT: [[TMP10:%.*]] = shufflevector <8 x double> [[TMP9]], <8 x double> [[TMP8]], <8 x i32> <i32 0, i32 1, i32 10, i32 11, i32 2, i32 3, i32 14, i32 15>
+; CHECK-NEXT: [[TMP11:%.*]] = fadd <8 x double> [[TMP6]], [[TMP10]]
+; CHECK-NEXT: br label [[FOR_END47:%.*]]
+; CHECK: for.end47:
+; CHECK-NEXT: [[TMP12:%.*]] = phi <8 x double> [ [[TMP11]], [[ENTRY:%.*]] ]
+; CHECK-NEXT: ret i32 0
+;
+entry:
+ %div0 = fdiv <2 x double> zeroinitializer, zeroinitializer
+ %div1 = fdiv <2 x double> zeroinitializer, zeroinitializer
+ %add0 = fadd <2 x double> zeroinitializer, %div0
+ %add1 = fadd <2 x double> zeroinitializer, zeroinitializer
+ %add2 = fadd <2 x double> %div1, zeroinitializer
+ %add3 = fadd <2 x double> zeroinitializer, zeroinitializer
+ br label %for.end47
+
+for.end47: ; preds = %entry
+ %add0.lcssa = phi <2 x double> [ %add0, %entry ]
+ %add1.lcssa = phi <2 x double> [ %add1, %entry ]
+ %add2.lcssa = phi <2 x double> [ %add2, %entry ]
+ %add3.lcssa = phi <2 x double> [ %add3, %entry ]
+ ret i32 0
+}
diff --git a/llvm/test/Transforms/SLPVectorizer/revec-shufflevector.ll b/llvm/test/Transforms/SLPVectorizer/revec-shufflevector.ll
index a2673d81068d8d..d6c09bc224a7d1 100644
--- a/llvm/test/Transforms/SLPVectorizer/revec-shufflevector.ll
+++ b/llvm/test/Transforms/SLPVectorizer/revec-shufflevector.ll
@@ -121,3 +121,117 @@ entry:
store <4 x i32> %1, ptr %3, align 4
ret void
}
+
+define void @test6(ptr %in0, ptr %in1, ptr %in2) {
+; CHECK-LABEL: @test6(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[GEP1:%.*]] = getelementptr inbounds nuw i8, ptr [[IN0:%.*]], i64 32
+; CHECK-NEXT: [[LOAD2:%.*]] = load <4 x float>, ptr [[GEP1]], align 16
+; CHECK-NEXT: [[TMP0:%.*]] = load <8 x float>, ptr [[IN0]], align 16
+; CHECK-NEXT: [[TMP1:%.*]] = load <32 x i8>, ptr [[IN1:%.*]], align 1
+; CHECK-NEXT: [[TMP2:%.*]] = uitofp <32 x i8> [[TMP1]] to <32 x float>
+; CHECK-NEXT: [[TMP14:%.*]] = shufflevector <8 x float> [[TMP0]], <8 x float> poison, <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>
+; CHECK-NEXT: [[TMP15:%.*]] = shufflevector <4 x float> [[LOAD2]], <4 x float> 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>
+; CHECK-NEXT: [[TMP16:%.*]] = shufflevector <16 x float> [[TMP14]], <16 x float> [[TMP15]], <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 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <16 x float> [[TMP16]], <16 x float> poison, <32 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 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT: [[TMP4:%.*]] = fmul <32 x float> [[TMP3]], [[TMP2]]
+; CHECK-NEXT: store <32 x float> [[TMP4]], ptr [[IN2:%.*]], align 16
+; CHECK-NEXT: [[GEP10:%.*]] = getelementptr inbounds nuw i8, ptr [[IN1]], i64 32
+; CHECK-NEXT: [[LOAD5:%.*]] = load <16 x i8>, ptr [[GEP10]], align 1
+; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <8 x float> [[TMP0]], <8 x float> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT: [[GEP11:%.*]] = getelementptr inbounds nuw i8, ptr [[IN2]], i64 128
+; CHECK-NEXT: [[TMP6:%.*]] = uitofp <16 x i8> [[LOAD5]] to <16 x float>
+; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <4 x float> [[LOAD2]], <4 x float> 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>
+; CHECK-NEXT: [[TMP8:%.*]] = shufflevector <8 x float> [[TMP0]], <8 x float> 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>
+; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <16 x float> [[TMP7]], <16 x float> [[TMP8]], <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT: [[TMP10:%.*]] = shufflevector <4 x float> [[TMP5]], <4 x float> 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>
+; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <16 x float> [[TMP9]], <16 x float> [[TMP10]], <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 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <16 x float> [[TMP11]], <16 x float> poison, <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 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT: [[TMP13:%.*]] = fmul <16 x float> [[TMP12]], [[TMP6]]
+; CHECK-NEXT: store <16 x float> [[TMP13]], ptr [[GEP11]], align 16
+; CHECK-NEXT: ret void
+;
+entry:
+ %gep0 = getelementptr inbounds i8, ptr %in0, i64 16
+ %gep1 = getelementptr inbounds i8, ptr %in0, i64 32
+ %load0 = load <4 x float>, ptr %in0, align 16
+ %load1 = load <4 x float>, ptr %gep0, align 16
+ %load2 = load <4 x float>, ptr %gep1, align 16
+ %gep2 = getelementptr inbounds i8, ptr %in1, i64 16
+ %load3 = load <16 x i8>, ptr %in1, align 1
+ %load4 = load <16 x i8>, ptr %gep2, align 1
+ %shufflevector0 = shufflevector <16 x i8> %load3, <16 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ %shufflevector1 = shufflevector <16 x i8> %load3, <16 x i8> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+ %shufflevector2 = shufflevector <16 x i8> %load4, <16 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ %shufflevector3 = shufflevector <16 x i8> %load4, <16 x i8> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+ %zext0 = zext <8 x i8> %shufflevector0 to <8 x i16>
+ %zext1 = zext <8 x i8> %shufflevector1 to <8 x i16>
+ %zext2 = zext <8 x i8> %shufflevector2 to <8 x i16>
+ %zext3 = zext <8 x i8> %shufflevector3 to <8 x i16>
+ %shufflevector4 = shufflevector <8 x i16> %zext0, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %shufflevector5 = shufflevector <8 x i16> %zext0, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+ %shufflevector6 = shufflevector <8 x i16> %zext1, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %shufflevector7 = shufflevector <8 x i16> %zext1, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+ %shufflevector8 = shufflevector <8 x i16> %zext2, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %shufflevector9 = shufflevector <8 x i16> %zext2, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+ %shufflevector10 = shufflevector <8 x i16> %zext3, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %shufflevector11 = shufflevector <8 x i16> %zext3, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+ %uitofp0 = uitofp nneg <4 x i16> %shufflevector4 to <4 x float>
+ %uitofp1 = uitofp nneg <4 x i16> %shufflevector5 to <4 x float>
+ %uitofp2 = uitofp nneg <4 x i16> %shufflevector6 to <4 x float>
+ %uitofp3 = uitofp nneg <4 x i16> %shufflevector7 to <4 x float>
+ %uitofp4 = uitofp nneg <4 x i16> %shufflevector8 to <4 x float>
+ %uitofp5 = uitofp nneg <4 x i16> %shufflevector9 to <4 x float>
+ %uitofp6 = uitofp nneg <4 x i16> %shufflevector10 to <4 x float>
+ %uitofp7 = uitofp nneg <4 x i16> %shufflevector11 to <4 x float>
+ %fmul0 = fmul <4 x float> %load0, %uitofp0
+ %fmul1 = fmul <4 x float> %load1, %uitofp1
+ %fmul2 = fmul <4 x float> %load2, %uitofp2
+ %fmul3 = fmul <4 x float> %load0, %uitofp3
+ %fmul4 = fmul <4 x float> %load1, %uitofp4
+ %fmul5 = fmul <4 x float> %load2, %uitofp5
+ %fmul6 = fmul <4 x float> %load0, %uitofp6
+ %fmul7 = fmul <4 x float> %load1, %uitofp7
+ %gep3 = getelementptr inbounds i8, ptr %in2, i64 16
+ %gep4 = getelementptr inbounds i8, ptr %in2, i64 32
+ %gep5 = getelementptr inbounds i8, ptr %in2, i64 48
+ %gep6 = getelementptr inbounds i8, ptr %in2, i64 64
+ %gep7 = getelementptr inbounds i8, ptr %in2, i64 80
+ %gep8 = getelementptr inbounds i8, ptr %in2, i64 96
+ %gep9 = getelementptr inbounds i8, ptr %in2, i64 112
+ store <4 x float> %fmul0, ptr %in2, align 16
+ store <4 x float> %fmul1, ptr %gep3, align 16
+ store <4 x float> %fmul2, ptr %gep4, align 16
+ store <4 x float> %fmul3, ptr %gep5, align 16
+ store <4 x float> %fmul4, ptr %gep6, align 16
+ store <4 x float> %fmul5, ptr %gep7, align 16
+ store <4 x float> %fmul6, ptr %gep8, align 16
+ store <4 x float> %fmul7, ptr %gep9, align 16
+ %gep10 = getelementptr inbounds i8, ptr %in1, i64 32
+ %load5 = load <16 x i8>, ptr %gep10, align 1
+ %shufflevector12 = shufflevector <16 x i8> %load5, <16 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ %shufflevector13 = shufflevector <16 x i8> %load5, <16 x i8> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+ %zext4 = zext <8 x i8> %shufflevector12 to <8 x i16>
+ %zext5 = zext <8 x i8> %shufflevector13 to <8 x i16>
+ %shufflevector14 = shufflevector <8 x i16> %zext4, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %shufflevector15 = shufflevector <8 x i16> %zext4, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+ %shufflevector16 = shufflevector <8 x i16> %zext5, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %shufflevector17 = shufflevector <8 x i16> %zext5, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+ %uitofp8 = uitofp nneg <4 x i16> %shufflevector14 to <4 x float>
+ %uitofp9 = uitofp nneg <4 x i16> %shufflevector15 to <4 x float>
+ %uitofp10 = uitofp nneg <4 x i16> %shufflevector16 to <4 x float>
+ %uitofp11 = uitofp nneg <4 x i16> %shufflevector17 to <4 x float>
+ %fmul8 = fmul <4 x float> %load2, %uitofp8
+ %fmul9 = fmul <4 x float> %load0, %uitofp9
+ %fmul10 = fmul <4 x float> %load1, %uitofp10
+ %fmul11 = fmul <4 x float> %load2, %uitofp11
+ %gep11 = getelementptr inbounds i8, ptr %in2, i64 128
+ %gep12 = getelementptr inbounds i8, ptr %in2, i64 144
+ %gep13 = getelementptr inbounds i8, ptr %in2, i64 160
+ %gep14 = getelementptr inbounds i8, ptr %in2, i64 176
+ store <4 x float> %fmul8, ptr %gep11, align 16
+ store <4 x float> %fmul9, ptr %gep12, align 16
+ store <4 x float> %fmul10, ptr %gep13, align 16
+ store <4 x float> %fmul11, ptr %gep14, align 16
+ ret void
+}
|
This was referenced Jan 28, 2025
Closed
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
We have two types of mask in SLP: a scalar mask and a vector mask.
When vectorizing four i32 additions into <4 x i32>, SLP creates a mask of length 4.
When vectorizing four <2 x i32> additions into <8 x i32>, SLP also creates a mask of length 4.
We refer to the first case as a scalar mask (because the mask element represents a scalar, i32), and the second case as a vector mask (because the mask element represents a vector, <4 x i32>).
At some point, we must convert the scalar mask into a vector mask (otherwise, calling TTI cost functions or IRBuilderBase functions may yield incorrect results).
Since both ShuffleCostEstimator and ShuffleInstructionBuilder can modify the CommonMask, we have decided to perform the mask transformation only within createShuffle. However, we do not store the transformed result, as createShuffle may be called multiple times.