[llvm] 330cb03 - [LoadStoreVectorizer] Check for guaranteed-to-transfer (PR52950)
Nikita Popov via llvm-commits
llvm-commits at lists.llvm.org
Mon Jan 3 01:57:03 PST 2022
Author: Nikita Popov
Date: 2022-01-03T10:55:47+01:00
New Revision: 330cb0326911ca7090be56c1641ba86f26b6c3c8
URL: https://github.com/llvm/llvm-project/commit/330cb0326911ca7090be56c1641ba86f26b6c3c8
DIFF: https://github.com/llvm/llvm-project/commit/330cb0326911ca7090be56c1641ba86f26b6c3c8.diff
LOG: [LoadStoreVectorizer] Check for guaranteed-to-transfer (PR52950)
Rather than checking for nounwind in particular, make sure the
instruction is guaranteed to transfer execution, which will also
handle non-willreturn calls correctly.
Fixes https://github.com/llvm/llvm-project/issues/52950.
Added:
Modified:
llvm/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp
llvm/test/Transforms/LoadStoreVectorizer/NVPTX/merge-across-side-effects.ll
llvm/test/Transforms/LoadStoreVectorizer/int_sideeffect.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp b/llvm/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp
index 5a4a2f0924f68..d2e0d1d474b0e 100644
--- a/llvm/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp
@@ -698,8 +698,9 @@ Vectorizer::getVectorizablePrefix(ArrayRef<Instruction *> Chain) {
ChainInstrs.push_back(&I);
continue;
}
- if (I.mayThrow()) {
- LLVM_DEBUG(dbgs() << "LSV: Found may-throw operation: " << I << '\n');
+ if (!isGuaranteedToTransferExecutionToSuccessor(&I)) {
+ LLVM_DEBUG(dbgs() << "LSV: Found instruction may not transfer execution: "
+ << I << '\n');
break;
}
if (I.mayReadOrWriteMemory())
diff --git a/llvm/test/Transforms/LoadStoreVectorizer/NVPTX/merge-across-side-effects.ll b/llvm/test/Transforms/LoadStoreVectorizer/NVPTX/merge-across-side-effects.ll
index 72c13b4d12e52..73623dade42f9 100644
--- a/llvm/test/Transforms/LoadStoreVectorizer/NVPTX/merge-across-side-effects.ll
+++ b/llvm/test/Transforms/LoadStoreVectorizer/NVPTX/merge-across-side-effects.ll
@@ -200,10 +200,10 @@ define void @store_fn_readnone(i32* %p) #0 {
}
-attributes #0 = { nounwind }
-attributes #1 = { nounwind writeonly }
-attributes #2 = { nounwind readonly }
+attributes #0 = { nounwind willreturn }
+attributes #1 = { nounwind willreturn writeonly }
+attributes #2 = { nounwind readonly willreturn }
attributes #3 = { writeonly }
attributes #4 = { readonly }
; readnone implies nounwind, so no need to test separately
-attributes #5 = { nounwind readnone }
+attributes #5 = { nounwind willreturn readnone }
diff --git a/llvm/test/Transforms/LoadStoreVectorizer/int_sideeffect.ll b/llvm/test/Transforms/LoadStoreVectorizer/int_sideeffect.ll
index 23108e308f816..bd47d66ddaaa4 100644
--- a/llvm/test/Transforms/LoadStoreVectorizer/int_sideeffect.ll
+++ b/llvm/test/Transforms/LoadStoreVectorizer/int_sideeffect.ll
@@ -44,8 +44,8 @@ define void @test_sideeffect(float* %p) {
declare void @foo()
-define void @test_inaccessiblememonly(float* %p) {
-; CHECK-LABEL: @test_inaccessiblememonly(
+define void @test_inaccessiblememonly_nounwind_willreturn(float* %p) {
+; CHECK-LABEL: @test_inaccessiblememonly_nounwind_willreturn(
; CHECK-NEXT: [[P0:%.*]] = getelementptr float, float* [[P:%.*]], i64 0
; CHECK-NEXT: [[TMP1:%.*]] = bitcast float* [[P0]] to <4 x float>*
; CHECK-NEXT: [[TMP2:%.*]] = load <4 x float>, <4 x float>* [[TMP1]], align 16
@@ -62,6 +62,41 @@ define void @test_inaccessiblememonly(float* %p) {
; CHECK-NEXT: [[TMP7:%.*]] = bitcast float* [[P0]] to <4 x float>*
; CHECK-NEXT: store <4 x float> [[TMP6]], <4 x float>* [[TMP7]], align 16
; CHECK-NEXT: ret void
+;
+ %p0 = getelementptr float, float* %p, i64 0
+ %p1 = getelementptr float, float* %p, i64 1
+ %p2 = getelementptr float, float* %p, i64 2
+ %p3 = getelementptr float, float* %p, i64 3
+ %l0 = load float, float* %p0, align 16
+ %l1 = load float, float* %p1
+ %l2 = load float, float* %p2
+ call void @foo() inaccessiblememonly nounwind willreturn
+ %l3 = load float, float* %p3
+ store float %l0, float* %p0, align 16
+ call void @foo() inaccessiblememonly nounwind willreturn
+ store float %l1, float* %p1
+ store float %l2, float* %p2
+ store float %l3, float* %p3
+ ret void
+}
+
+define void @test_inaccessiblememonly_not_willreturn(float* %p) {
+; CHECK-LABEL: @test_inaccessiblememonly_not_willreturn(
+; CHECK-NEXT: [[P0:%.*]] = getelementptr float, float* [[P:%.*]], i64 0
+; CHECK-NEXT: [[P1:%.*]] = getelementptr float, float* [[P]], i64 1
+; CHECK-NEXT: [[P2:%.*]] = getelementptr float, float* [[P]], i64 2
+; CHECK-NEXT: [[P3:%.*]] = getelementptr float, float* [[P]], i64 3
+; CHECK-NEXT: [[L0:%.*]] = load float, float* [[P0]], align 16
+; CHECK-NEXT: [[L1:%.*]] = load float, float* [[P1]], align 4
+; CHECK-NEXT: [[L2:%.*]] = load float, float* [[P2]], align 4
+; CHECK-NEXT: call void @foo() #[[ATTR2:[0-9]+]]
+; CHECK-NEXT: [[L3:%.*]] = load float, float* [[P3]], align 4
+; CHECK-NEXT: store float [[L0]], float* [[P0]], align 16
+; CHECK-NEXT: call void @foo() #[[ATTR2]]
+; CHECK-NEXT: store float [[L1]], float* [[P1]], align 4
+; CHECK-NEXT: store float [[L2]], float* [[P2]], align 4
+; CHECK-NEXT: store float [[L3]], float* [[P3]], align 4
+; CHECK-NEXT: ret void
;
%p0 = getelementptr float, float* %p, i64 0
%p1 = getelementptr float, float* %p, i64 1
@@ -79,3 +114,38 @@ define void @test_inaccessiblememonly(float* %p) {
store float %l3, float* %p3
ret void
}
+
+define void @test_inaccessiblememonly_not_nounwind(float* %p) {
+; CHECK-LABEL: @test_inaccessiblememonly_not_nounwind(
+; CHECK-NEXT: [[P0:%.*]] = getelementptr float, float* [[P:%.*]], i64 0
+; CHECK-NEXT: [[P1:%.*]] = getelementptr float, float* [[P]], i64 1
+; CHECK-NEXT: [[P2:%.*]] = getelementptr float, float* [[P]], i64 2
+; CHECK-NEXT: [[P3:%.*]] = getelementptr float, float* [[P]], i64 3
+; CHECK-NEXT: [[L0:%.*]] = load float, float* [[P0]], align 16
+; CHECK-NEXT: [[L1:%.*]] = load float, float* [[P1]], align 4
+; CHECK-NEXT: [[L2:%.*]] = load float, float* [[P2]], align 4
+; CHECK-NEXT: call void @foo() #[[ATTR3:[0-9]+]]
+; CHECK-NEXT: [[L3:%.*]] = load float, float* [[P3]], align 4
+; CHECK-NEXT: store float [[L0]], float* [[P0]], align 16
+; CHECK-NEXT: call void @foo() #[[ATTR3]]
+; CHECK-NEXT: store float [[L1]], float* [[P1]], align 4
+; CHECK-NEXT: store float [[L2]], float* [[P2]], align 4
+; CHECK-NEXT: store float [[L3]], float* [[P3]], align 4
+; CHECK-NEXT: ret void
+;
+ %p0 = getelementptr float, float* %p, i64 0
+ %p1 = getelementptr float, float* %p, i64 1
+ %p2 = getelementptr float, float* %p, i64 2
+ %p3 = getelementptr float, float* %p, i64 3
+ %l0 = load float, float* %p0, align 16
+ %l1 = load float, float* %p1
+ %l2 = load float, float* %p2
+ call void @foo() inaccessiblememonly willreturn
+ %l3 = load float, float* %p3
+ store float %l0, float* %p0, align 16
+ call void @foo() inaccessiblememonly willreturn
+ store float %l1, float* %p1
+ store float %l2, float* %p2
+ store float %l3, float* %p3
+ ret void
+}
More information about the llvm-commits
mailing list