[llvm] [SLP][REVEC] Make shufflevector can be vectorized with ReorderIndices and ReuseShuffleIndices. (PR #114965)
via llvm-commits
llvm-commits at lists.llvm.org
Tue Nov 5 02:32:36 PST 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-vectorizers
Author: Han-Kuan Chen (HanKuanChen)
<details>
<summary>Changes</summary>
---
Full diff: https://github.com/llvm/llvm-project/pull/114965.diff
2 Files Affected:
- (modified) llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp (+8-8)
- (modified) llvm/test/Transforms/SLPVectorizer/revec-shufflevector.ll (+37)
``````````diff
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 427b8bd0e75ab0..b14465ecc1dc91 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -15374,9 +15374,6 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E, bool PostponedPHIs) {
case Instruction::ShuffleVector: {
Value *V;
if (SLPReVec && !E->isAltShuffle()) {
- assert(E->ReuseShuffleIndices.empty() &&
- "Not support ReuseShuffleIndices yet.");
- assert(E->ReorderIndices.empty() && "Not support ReorderIndices yet.");
setInsertPointAfterBundle(E);
Value *Src = vectorizeOperand(E, 0, PostponedPHIs);
if (E->VectorizedValue) {
@@ -15394,6 +15391,9 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E, bool PostponedPHIs) {
[&SVSrc](int Mask) { return SVSrc->getShuffleMask()[Mask]; });
V = Builder.CreateShuffleVector(SVSrc->getOperand(0), NewMask);
propagateIRFlags(V, E->Scalars, VL0);
+ if (auto *I = dyn_cast<Instruction>(V))
+ V = propagateMetadata(I, E->Scalars);
+ V = FinalShuffle(V, E);
} else {
assert(E->isAltShuffle() &&
((Instruction::isBinaryOp(E->getOpcode()) &&
@@ -15524,11 +15524,11 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E, bool PostponedPHIs) {
transformScalarShuffleIndiciesToVector(VecTy->getNumElements(), Mask);
}
V = Builder.CreateShuffleVector(V0, V1, Mask);
- }
- if (auto *I = dyn_cast<Instruction>(V)) {
- V = propagateMetadata(I, E->Scalars);
- GatherShuffleExtractSeq.insert(I);
- CSEBlocks.insert(I->getParent());
+ if (auto *I = dyn_cast<Instruction>(V)) {
+ V = propagateMetadata(I, E->Scalars);
+ GatherShuffleExtractSeq.insert(I);
+ CSEBlocks.insert(I->getParent());
+ }
}
E->VectorizedValue = V;
diff --git a/llvm/test/Transforms/SLPVectorizer/revec-shufflevector.ll b/llvm/test/Transforms/SLPVectorizer/revec-shufflevector.ll
index 1fc0b0306d1194..39ba9d52d11e9b 100644
--- a/llvm/test/Transforms/SLPVectorizer/revec-shufflevector.ll
+++ b/llvm/test/Transforms/SLPVectorizer/revec-shufflevector.ll
@@ -82,3 +82,40 @@ entry:
store <4 x i32> %4, ptr %8, align 4
ret void
}
+
+define void @test4(ptr %in) {
+; CHECK-LABEL: @test4(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br label [[LABEL0:%.*]]
+; CHECK: label0:
+; CHECK-NEXT: br label [[LABEL2:%.*]]
+; CHECK: label1:
+; CHECK-NEXT: br label [[LABEL2]]
+; CHECK: label2:
+; CHECK-NEXT: br label [[LABEL0]]
+;
+entry:
+ br label %label0
+
+label0:
+ %0 = load <8 x float>, ptr %in, align 4
+ %1 = shufflevector <8 x float> %0, <8 x float> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+ %2 = shufflevector <8 x float> %0, <8 x float> zeroinitializer, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+ %3 = call <4 x float> @llvm.fma.v4f32(<4 x float> %1, <4 x float> zeroinitializer, <4 x float> zeroinitializer)
+ %4 = call <4 x float> @llvm.fma.v4f32(<4 x float> %2, <4 x float> zeroinitializer, <4 x float> zeroinitializer)
+ %5 = call <4 x float> @llvm.fma.v4f32(<4 x float> %1, <4 x float> zeroinitializer, <4 x float> zeroinitializer)
+ %6 = call <4 x float> @llvm.fma.v4f32(<4 x float> %2, <4 x float> zeroinitializer, <4 x float> zeroinitializer)
+ br label %label2
+
+label1:
+ br label %label2
+
+label2:
+ %7 = phi <4 x float> [ %3, %label0 ], [ zeroinitializer, %label1 ]
+ %8 = phi <4 x float> [ %4, %label0 ], [ zeroinitializer, %label1 ]
+ %9 = phi <4 x float> [ %5, %label0 ], [ zeroinitializer, %label1 ]
+ %10 = phi <4 x float> [ %6, %label0 ], [ zeroinitializer, %label1 ]
+ br label %label0
+}
+
+declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>)
``````````
</details>
https://github.com/llvm/llvm-project/pull/114965
More information about the llvm-commits
mailing list