[SLP] Fix assertion "Deleting out-of-tree value" for struct-typed vectorized operands#199679
Merged
Conversation
Created using spr 1.3.7
alexey-bataev
deleted the
users/alexey-bataev/spr/slp-fix-assertion-deleting-out-of-tree-value-for-struct-typed-vectorized-operands
branch
|
@llvm/pr-subscribers-llvm-transforms Author: Alexey Bataev (alexey-bataev) ChangesWhen an external-use scalar had a struct-typed vectorized operand, OperandIsScalar
Full diff: https://github.com/llvm/llvm-project/pull/199679.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index bdd5311ab55ad..caee72eb0547a 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -19708,6 +19708,8 @@ InstructionCost BoUpSLP::getTreeCost(InstructionCost TreeCost,
return !EE->hasOneUse() || !MustGather.contains(EE);
return true;
}
+ if (isa<StructType>(V->getType()))
+ return false;
return ValueToExtUses->contains(V);
};
bool CanBeUsedAsScalar = all_of(Inst->operands(), OperandIsScalar);
diff --git a/llvm/test/Transforms/SLPVectorizer/AArch64/slp-extval-struct-operand-no-scalar-reuse.ll b/llvm/test/Transforms/SLPVectorizer/AArch64/slp-extval-struct-operand-no-scalar-reuse.ll
new file mode 100644
index 0000000000000..2f84803994136
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/AArch64/slp-extval-struct-operand-no-scalar-reuse.ll
@@ -0,0 +1,38 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S --passes=slp-vectorizer -slp-threshold=-10 -mtriple=arm64-apple-macosx12.0.0 < %s | FileCheck %s
+
+
+define [2 x i64] @test() {
+; CHECK-LABEL: define [2 x i64] @test() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[TMP0:%.*]] = call <2 x i32> @llvm.sadd.sat.v2i32(<2 x i32> zeroinitializer, <2 x i32> zeroinitializer)
+; CHECK-NEXT: [[TMP1:%.*]] = call { <2 x i32>, <2 x i1> } @llvm.ssub.with.overflow.v2i32(<2 x i32> zeroinitializer, <2 x i32> [[TMP0]])
+; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <2 x i32>, <2 x i1> } [[TMP1]], 0
+; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[TMP2]], i32 0
+; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <2 x i32>, <2 x i1> } [[TMP1]], 1
+; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[TMP4]], i32 0
+; CHECK-NEXT: [[RETVAL_SROA_5_8_INSERT_SHIFT_I52_I:%.*]] = select i1 [[TMP5]], i64 0, i64 0
+; CHECK-NEXT: [[TMP6:%.*]] = zext <2 x i1> [[TMP4]] to <2 x i64>
+; CHECK-NEXT: store <2 x i64> [[TMP6]], ptr getelementptr inbounds nuw (i8, ptr null, i64 104), align 8
+; CHECK-NEXT: ret [2 x i64] zeroinitializer
+;
+entry:
+ %retval.0.i.i.i.i.i14.i.i.i = tail call i32 @llvm.sadd.sat.i32(i32 0, i32 0)
+ %0 = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 0, i32 %retval.0.i.i.i.i.i14.i.i.i)
+ %1 = extractvalue { i32, i1 } %0, 0
+ %2 = extractvalue { i32, i1 } %0, 1
+ %3 = zext i1 %2 to i64
+ store i64 %3, ptr getelementptr inbounds nuw (i8, ptr null, i64 104), align 8
+ %retval.sroa.5.8.insert.shift.i52.i = select i1 %2, i64 0, i64 0
+ %retval.0.i.i.i.i.i13.i.i.i = tail call i32 @llvm.sadd.sat.i32(i32 0, i32 0)
+ %4 = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 0, i32 %retval.0.i.i.i.i.i13.i.i.i)
+ %5 = extractvalue { i32, i1 } %4, 1
+ %6 = zext i1 %5 to i64
+ store i64 %6, ptr getelementptr inbounds nuw (i8, ptr null, i64 112), align 8
+ ret [2 x i64] zeroinitializer
+}
+
+declare { i32, i1 } @llvm.ssub.with.overflow.i32(i32, i32)
+
+declare i32 @llvm.sadd.sat.i32(i32, i32)
+
|
|
@llvm/pr-subscribers-vectorizers Author: Alexey Bataev (alexey-bataev) ChangesWhen an external-use scalar had a struct-typed vectorized operand, OperandIsScalar
Full diff: https://github.com/llvm/llvm-project/pull/199679.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index bdd5311ab55ad..caee72eb0547a 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -19708,6 +19708,8 @@ InstructionCost BoUpSLP::getTreeCost(InstructionCost TreeCost,
return !EE->hasOneUse() || !MustGather.contains(EE);
return true;
}
+ if (isa<StructType>(V->getType()))
+ return false;
return ValueToExtUses->contains(V);
};
bool CanBeUsedAsScalar = all_of(Inst->operands(), OperandIsScalar);
diff --git a/llvm/test/Transforms/SLPVectorizer/AArch64/slp-extval-struct-operand-no-scalar-reuse.ll b/llvm/test/Transforms/SLPVectorizer/AArch64/slp-extval-struct-operand-no-scalar-reuse.ll
new file mode 100644
index 0000000000000..2f84803994136
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/AArch64/slp-extval-struct-operand-no-scalar-reuse.ll
@@ -0,0 +1,38 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S --passes=slp-vectorizer -slp-threshold=-10 -mtriple=arm64-apple-macosx12.0.0 < %s | FileCheck %s
+
+
+define [2 x i64] @test() {
+; CHECK-LABEL: define [2 x i64] @test() {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[TMP0:%.*]] = call <2 x i32> @llvm.sadd.sat.v2i32(<2 x i32> zeroinitializer, <2 x i32> zeroinitializer)
+; CHECK-NEXT: [[TMP1:%.*]] = call { <2 x i32>, <2 x i1> } @llvm.ssub.with.overflow.v2i32(<2 x i32> zeroinitializer, <2 x i32> [[TMP0]])
+; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <2 x i32>, <2 x i1> } [[TMP1]], 0
+; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[TMP2]], i32 0
+; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <2 x i32>, <2 x i1> } [[TMP1]], 1
+; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[TMP4]], i32 0
+; CHECK-NEXT: [[RETVAL_SROA_5_8_INSERT_SHIFT_I52_I:%.*]] = select i1 [[TMP5]], i64 0, i64 0
+; CHECK-NEXT: [[TMP6:%.*]] = zext <2 x i1> [[TMP4]] to <2 x i64>
+; CHECK-NEXT: store <2 x i64> [[TMP6]], ptr getelementptr inbounds nuw (i8, ptr null, i64 104), align 8
+; CHECK-NEXT: ret [2 x i64] zeroinitializer
+;
+entry:
+ %retval.0.i.i.i.i.i14.i.i.i = tail call i32 @llvm.sadd.sat.i32(i32 0, i32 0)
+ %0 = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 0, i32 %retval.0.i.i.i.i.i14.i.i.i)
+ %1 = extractvalue { i32, i1 } %0, 0
+ %2 = extractvalue { i32, i1 } %0, 1
+ %3 = zext i1 %2 to i64
+ store i64 %3, ptr getelementptr inbounds nuw (i8, ptr null, i64 104), align 8
+ %retval.sroa.5.8.insert.shift.i52.i = select i1 %2, i64 0, i64 0
+ %retval.0.i.i.i.i.i13.i.i.i = tail call i32 @llvm.sadd.sat.i32(i32 0, i32 0)
+ %4 = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 0, i32 %retval.0.i.i.i.i.i13.i.i.i)
+ %5 = extractvalue { i32, i1 } %4, 1
+ %6 = zext i1 %5 to i64
+ store i64 %6, ptr getelementptr inbounds nuw (i8, ptr null, i64 112), align 8
+ ret [2 x i64] zeroinitializer
+}
+
+declare { i32, i1 } @llvm.ssub.with.overflow.i32(i32, i32)
+
+declare i32 @llvm.sadd.sat.i32(i32, i32)
+
|
llvm-upstreamsync Bot
pushed a commit
to qualcomm/cpullvm-toolchain
that referenced
this pull request
May 26, 2026
…t-typed vectorized operands When an external-use scalar had a struct-typed vectorized operand, OperandIsScalar incorrectly returned true (because the struct scalar was in ValueToExtUses), causing the scalar to be cloned into ExternalUsesAsOriginalScalar. The clone used the struct scalar directly, but struct-typed vectorized scalars are deleted after vectorization - their external-use handling erases the extractvalue user rather than replacing the struct value via replaceAllUsesWith. Reviewers: Pull Request: llvm/llvm-project#199679
llvm-sync Bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
May 26, 2026
…t-typed vectorized operands When an external-use scalar had a struct-typed vectorized operand, OperandIsScalar incorrectly returned true (because the struct scalar was in ValueToExtUses), causing the scalar to be cloned into ExternalUsesAsOriginalScalar. The clone used the struct scalar directly, but struct-typed vectorized scalars are deleted after vectorization - their external-use handling erases the extractvalue user rather than replacing the struct value via replaceAllUsesWith. Reviewers: Pull Request: llvm/llvm-project#199679
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
When an external-use scalar had a struct-typed vectorized operand, OperandIsScalar
incorrectly returned true (because the struct scalar was in ValueToExtUses), causing
the scalar to be cloned into ExternalUsesAsOriginalScalar. The clone used the struct
scalar directly, but struct-typed vectorized scalars are deleted after vectorization
struct value via replaceAllUsesWith.