[llvm-branch-commits] [llvm] da33d40 - [SLP] Don't try to vectorize pair with insertelement
Tom Stellard via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Tue Mar 1 14:31:15 PST 2022
Author: Anton Afanasyev
Date: 2022-03-01T14:29:22-08:00
New Revision: da33d400682a8cf93062fe61a9f0b6ec1d60c8ad
URL: https://github.com/llvm/llvm-project/commit/da33d400682a8cf93062fe61a9f0b6ec1d60c8ad
DIFF: https://github.com/llvm/llvm-project/commit/da33d400682a8cf93062fe61a9f0b6ec1d60c8ad.diff
LOG: [SLP] Don't try to vectorize pair with insertelement
Particularly this breaks vectorization of insertelements where some of
intermediate (i.e. not last) insertelements are used externally.
Fixes PR52275
Fixes #51617
Reviewed by: ABataev
Differential Revision: https://reviews.llvm.org/D119679
(cherry picked from commit b7574b0)
Added:
Modified:
llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
llvm/test/Transforms/SLPVectorizer/X86/pr52275.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 9eafd94efea28..644372483edde 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -3849,13 +3849,7 @@ void BoUpSLP::buildTree_rec(ArrayRef<Value *> VL, unsigned Depth,
ValueSet SourceVectors;
for (Value *V : VL) {
SourceVectors.insert(cast<Instruction>(V)->getOperand(0));
- if (getInsertIndex(V) == None) {
- LLVM_DEBUG(dbgs() << "SLP: Gather of insertelement vectors with "
- "non-constant or undef index.\n");
- newTreeEntry(VL, None /*not vectorized*/, S, UserTreeIdx);
- BS.cancelScheduling(VL, VL0);
- return;
- }
+ assert(getInsertIndex(V) != None && "Non-constant or undef index?");
}
if (count_if(VL, [&SourceVectors](Value *V) {
@@ -8343,6 +8337,8 @@ void SLPVectorizerPass::collectSeedInstructions(BasicBlock *BB) {
bool SLPVectorizerPass::tryToVectorizePair(Value *A, Value *B, BoUpSLP &R) {
if (!A || !B)
return false;
+ if (isa<InsertElementInst>(A) || isa<InsertElementInst>(B))
+ return false;
Value *VL[] = {A, B};
return tryToVectorizeList(VL, R);
}
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/pr52275.ll b/llvm/test/Transforms/SLPVectorizer/X86/pr52275.ll
index 6794553ffd6d0..9205ef0b375d1 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/pr52275.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/pr52275.ll
@@ -1,16 +1,7 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -slp-vectorizer -S -mtriple=x86_64-- | FileCheck %s
-define <4 x i8> @pr52275(<4 x i8> %v, i8* %x) {
-; CHECK-LABEL: @pr52275(
-; CHECK-NEXT: [[G1:%.*]] = getelementptr inbounds i8, i8* [[X:%.*]], i64 1
-; CHECK-NEXT: [[TMP1:%.*]] = bitcast i8* [[X]] to <2 x i8>*
-; CHECK-NEXT: [[TMP2:%.*]] = load <2 x i8>, <2 x i8>* [[TMP1]], align 4
-; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <2 x i8> [[TMP2]], <2 x i8> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
-; CHECK-NEXT: [[V11:%.*]] = shufflevector <4 x i8> [[V:%.*]], <4 x i8> [[TMP3]], <4 x i32> <i32 4, i32 5, i32 2, i32 3>
-; CHECK-NEXT: [[V2:%.*]] = add <4 x i8> [[V11]], [[V11]]
-; CHECK-NEXT: ret <4 x i8> [[V2]]
-;
+define <4 x i8> @test(<4 x i8> %v, i8* %x) {
%x0 = load i8, i8* %x, align 4
%g1 = getelementptr inbounds i8, i8* %x, i64 1
%x1 = load i8, i8* %g1, align 4
@@ -19,3 +10,139 @@ define <4 x i8> @pr52275(<4 x i8> %v, i8* %x) {
%v2 = add <4 x i8> %v0, %v1
ret <4 x i8> %v2
}
+
+define <2 x i8> @test2(<2 x i8> %t6, i32* %t1) {
+; CHECK-LABEL: @test2(
+; CHECK-NEXT: [[T3:%.*]] = load i32, i32* [[T1:%.*]], align 4
+; CHECK-NEXT: [[T4:%.*]] = getelementptr inbounds i32, i32* [[T1]], i64 1
+; CHECK-NEXT: [[T5:%.*]] = load i32, i32* [[T4]], align 4
+; CHECK-NEXT: [[T7:%.*]] = trunc i32 [[T3]] to i8
+; CHECK-NEXT: [[T8:%.*]] = insertelement <2 x i8> [[T6:%.*]], i8 [[T7]], i64 0
+; CHECK-NEXT: [[T9:%.*]] = trunc i32 [[T5]] to i8
+; CHECK-NEXT: [[T10:%.*]] = insertelement <2 x i8> [[T8]], i8 [[T9]], i64 1
+; CHECK-NEXT: [[T11:%.*]] = add <2 x i8> [[T10]], [[T8]]
+; CHECK-NEXT: ret <2 x i8> [[T11]]
+;
+; FORCE_SLP-LABEL: @test2(
+; FORCE_SLP-NEXT: [[T3:%.*]] = load i32, i32* [[T1:%.*]], align 4
+; FORCE_SLP-NEXT: [[T4:%.*]] = getelementptr inbounds i32, i32* [[T1]], i64 1
+; FORCE_SLP-NEXT: [[T5:%.*]] = load i32, i32* [[T4]], align 4
+; FORCE_SLP-NEXT: [[T7:%.*]] = trunc i32 [[T3]] to i8
+; FORCE_SLP-NEXT: [[T8:%.*]] = insertelement <2 x i8> [[T6:%.*]], i8 [[T7]], i64 0
+; FORCE_SLP-NEXT: [[T9:%.*]] = trunc i32 [[T5]] to i8
+; FORCE_SLP-NEXT: [[T10:%.*]] = insertelement <2 x i8> [[T8]], i8 [[T9]], i64 1
+; FORCE_SLP-NEXT: [[T11:%.*]] = add <2 x i8> [[T10]], [[T8]]
+; FORCE_SLP-NEXT: ret <2 x i8> [[T11]]
+;
+ %t3 = load i32, i32* %t1, align 4
+ %t4 = getelementptr inbounds i32, i32* %t1, i64 1
+ %t5 = load i32, i32* %t4, align 4
+ %t7 = trunc i32 %t3 to i8
+ %t8 = insertelement <2 x i8> %t6, i8 %t7, i64 0
+ %t9 = trunc i32 %t5 to i8
+ %t10 = insertelement <2 x i8> %t8, i8 %t9, i64 1
+ %t11 = add <2 x i8> %t10, %t8
+ ret <2 x i8> %t11
+}
+
+define <2 x i8> @test_reorder(<2 x i8> %t6, i32* %t1) {
+; CHECK-LABEL: @test_reorder(
+; CHECK-NEXT: [[T3:%.*]] = load i32, i32* [[T1:%.*]], align 4
+; CHECK-NEXT: [[T4:%.*]] = getelementptr inbounds i32, i32* [[T1]], i64 1
+; CHECK-NEXT: [[T5:%.*]] = load i32, i32* [[T4]], align 4
+; CHECK-NEXT: [[T7:%.*]] = trunc i32 [[T3]] to i8
+; CHECK-NEXT: [[T8:%.*]] = insertelement <2 x i8> [[T6:%.*]], i8 [[T7]], i64 1
+; CHECK-NEXT: [[T9:%.*]] = trunc i32 [[T5]] to i8
+; CHECK-NEXT: [[T10:%.*]] = insertelement <2 x i8> [[T8]], i8 [[T9]], i64 0
+; CHECK-NEXT: [[T11:%.*]] = add <2 x i8> [[T10]], [[T8]]
+; CHECK-NEXT: ret <2 x i8> [[T11]]
+;
+; FORCE_SLP-LABEL: @test_reorder(
+; FORCE_SLP-NEXT: [[T3:%.*]] = load i32, i32* [[T1:%.*]], align 4
+; FORCE_SLP-NEXT: [[T4:%.*]] = getelementptr inbounds i32, i32* [[T1]], i64 1
+; FORCE_SLP-NEXT: [[T5:%.*]] = load i32, i32* [[T4]], align 4
+; FORCE_SLP-NEXT: [[T7:%.*]] = trunc i32 [[T3]] to i8
+; FORCE_SLP-NEXT: [[T8:%.*]] = insertelement <2 x i8> [[T6:%.*]], i8 [[T7]], i64 1
+; FORCE_SLP-NEXT: [[T9:%.*]] = trunc i32 [[T5]] to i8
+; FORCE_SLP-NEXT: [[T10:%.*]] = insertelement <2 x i8> [[T8]], i8 [[T9]], i64 0
+; FORCE_SLP-NEXT: [[T11:%.*]] = add <2 x i8> [[T10]], [[T8]]
+; FORCE_SLP-NEXT: ret <2 x i8> [[T11]]
+;
+ %t3 = load i32, i32* %t1, align 4
+ %t4 = getelementptr inbounds i32, i32* %t1, i64 1
+ %t5 = load i32, i32* %t4, align 4
+ %t7 = trunc i32 %t3 to i8
+ %t8 = insertelement <2 x i8> %t6, i8 %t7, i64 1
+ %t9 = trunc i32 %t5 to i8
+ %t10 = insertelement <2 x i8> %t8, i8 %t9, i64 0
+ %t11 = add <2 x i8> %t10, %t8
+ ret <2 x i8> %t11
+}
+
+define <4 x i8> @test_subvector(<4 x i8> %t6, i32* %t1) {
+; CHECK-LABEL: @test_subvector(
+; CHECK-NEXT: [[T3:%.*]] = load i32, i32* [[T1:%.*]], align 4
+; CHECK-NEXT: [[T4:%.*]] = getelementptr inbounds i32, i32* [[T1]], i64 1
+; CHECK-NEXT: [[T5:%.*]] = load i32, i32* [[T4]], align 4
+; CHECK-NEXT: [[T7:%.*]] = trunc i32 [[T3]] to i8
+; CHECK-NEXT: [[T8:%.*]] = insertelement <4 x i8> [[T6:%.*]], i8 [[T7]], i64 0
+; CHECK-NEXT: [[T9:%.*]] = trunc i32 [[T5]] to i8
+; CHECK-NEXT: [[T10:%.*]] = insertelement <4 x i8> [[T8]], i8 [[T9]], i64 1
+; CHECK-NEXT: [[T11:%.*]] = add <4 x i8> [[T10]], [[T8]]
+; CHECK-NEXT: ret <4 x i8> [[T11]]
+;
+; FORCE_SLP-LABEL: @test_subvector(
+; FORCE_SLP-NEXT: [[T3:%.*]] = load i32, i32* [[T1:%.*]], align 4
+; FORCE_SLP-NEXT: [[T4:%.*]] = getelementptr inbounds i32, i32* [[T1]], i64 1
+; FORCE_SLP-NEXT: [[T5:%.*]] = load i32, i32* [[T4]], align 4
+; FORCE_SLP-NEXT: [[T7:%.*]] = trunc i32 [[T3]] to i8
+; FORCE_SLP-NEXT: [[T8:%.*]] = insertelement <4 x i8> [[T6:%.*]], i8 [[T7]], i64 0
+; FORCE_SLP-NEXT: [[T9:%.*]] = trunc i32 [[T5]] to i8
+; FORCE_SLP-NEXT: [[T10:%.*]] = insertelement <4 x i8> [[T8]], i8 [[T9]], i64 1
+; FORCE_SLP-NEXT: [[T11:%.*]] = add <4 x i8> [[T10]], [[T8]]
+; FORCE_SLP-NEXT: ret <4 x i8> [[T11]]
+;
+ %t3 = load i32, i32* %t1, align 4
+ %t4 = getelementptr inbounds i32, i32* %t1, i64 1
+ %t5 = load i32, i32* %t4, align 4
+ %t7 = trunc i32 %t3 to i8
+ %t8 = insertelement <4 x i8> %t6, i8 %t7, i64 0
+ %t9 = trunc i32 %t5 to i8
+ %t10 = insertelement <4 x i8> %t8, i8 %t9, i64 1
+ %t11 = add <4 x i8> %t10, %t8
+ ret <4 x i8> %t11
+}
+
+define <4 x i8> @test_subvector_reorder(<4 x i8> %t6, i32* %t1) {
+; CHECK-LABEL: @test_subvector_reorder(
+; CHECK-NEXT: [[T3:%.*]] = load i32, i32* [[T1:%.*]], align 4
+; CHECK-NEXT: [[T4:%.*]] = getelementptr inbounds i32, i32* [[T1]], i64 1
+; CHECK-NEXT: [[T5:%.*]] = load i32, i32* [[T4]], align 4
+; CHECK-NEXT: [[T7:%.*]] = trunc i32 [[T3]] to i8
+; CHECK-NEXT: [[T8:%.*]] = insertelement <4 x i8> [[T6:%.*]], i8 [[T7]], i64 3
+; CHECK-NEXT: [[T9:%.*]] = trunc i32 [[T5]] to i8
+; CHECK-NEXT: [[T10:%.*]] = insertelement <4 x i8> [[T8]], i8 [[T9]], i64 2
+; CHECK-NEXT: [[T11:%.*]] = add <4 x i8> [[T10]], [[T8]]
+; CHECK-NEXT: ret <4 x i8> [[T11]]
+;
+; FORCE_SLP-LABEL: @test_subvector_reorder(
+; FORCE_SLP-NEXT: [[T3:%.*]] = load i32, i32* [[T1:%.*]], align 4
+; FORCE_SLP-NEXT: [[T4:%.*]] = getelementptr inbounds i32, i32* [[T1]], i64 1
+; FORCE_SLP-NEXT: [[T5:%.*]] = load i32, i32* [[T4]], align 4
+; FORCE_SLP-NEXT: [[T7:%.*]] = trunc i32 [[T3]] to i8
+; FORCE_SLP-NEXT: [[T8:%.*]] = insertelement <4 x i8> [[T6:%.*]], i8 [[T7]], i64 3
+; FORCE_SLP-NEXT: [[T9:%.*]] = trunc i32 [[T5]] to i8
+; FORCE_SLP-NEXT: [[T10:%.*]] = insertelement <4 x i8> [[T8]], i8 [[T9]], i64 2
+; FORCE_SLP-NEXT: [[T11:%.*]] = add <4 x i8> [[T10]], [[T8]]
+; FORCE_SLP-NEXT: ret <4 x i8> [[T11]]
+;
+ %t3 = load i32, i32* %t1, align 4
+ %t4 = getelementptr inbounds i32, i32* %t1, i64 1
+ %t5 = load i32, i32* %t4, align 4
+ %t7 = trunc i32 %t3 to i8
+ %t8 = insertelement <4 x i8> %t6, i8 %t7, i64 3
+ %t9 = trunc i32 %t5 to i8
+ %t10 = insertelement <4 x i8> %t8, i8 %t9, i64 2
+ %t11 = add <4 x i8> %t10, %t8
+ ret <4 x i8> %t11
+}
More information about the llvm-branch-commits
mailing list