Repository navigation
[SLP]Fix external uses of the pointers of the jumbled copyable loads - #240127
Merged
alexey-bataev merged 1 commit intoNov 8, 2026
Conversation
Created using spr 1.3.7
alexey-bataev
deleted the
users/alexey-bataev/spr/slpfix-external-uses-of-the-pointers-of-the-jumbled-copyable-loads
branch
November 8, 2026 12:39
|
@llvm/pr-subscribers-llvm-vectorizers Author: Alexey Bataev (alexey-bataev) ChangesThe wide load of the copyable loads uses the pointer of the first load, Fixes #233297 (comment) Assisted-by: Cursor Full diff: https://github.com/llvm/llvm-project/pull/240127.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 920205b1f9ce1..f3efbff76b778 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -8070,6 +8070,9 @@ Instruction *BoUpSLP::getRootEntryInstruction(const TreeEntry &Entry) const {
Entry.State == TreeEntry::StridedVectorize && Entry.isReverse() &&
!isWidenedStridedLoad(&Entry))
return dyn_cast<Instruction>(Entry.Scalars[Entry.ReorderIndices.front()]);
+ // The wide load of the copyable loads uses the pointer of the first load.
+ if (Entry.hasState() && Entry.getOpcode() == Instruction::Load)
+ return Entry.getFirstLoad();
return Entry.hasState() ? Entry.getMainOp()
: dyn_cast<Instruction>(Entry.Scalars.front());
}
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/loads-from-single-predecessor.ll b/llvm/test/Transforms/SLPVectorizer/X86/loads-from-single-predecessor.ll
index f0e8526fc81d6..2ea4b8ef5c534 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/loads-from-single-predecessor.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/loads-from-single-predecessor.ll
@@ -567,6 +567,64 @@ exit:
ret i32 0
}
+; The pointers of the jumbled loads are vectorized, the first load is not the
+; main op of the node.
+define i64 @jumbled_vectorized_ptrs(ptr %b, ptr noalias %dst, i1 %c) {
+; CHECK-LABEL: define i64 @jumbled_vectorized_ptrs(
+; CHECK-SAME: ptr [[B:%.*]], ptr noalias [[DST:%.*]], i1 [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x ptr> poison, ptr [[B]], i64 0
+; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x ptr> [[TMP0]], <4 x ptr> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, <4 x ptr> [[TMP1]], <4 x i64> <i64 16, i64 8, i64 24, i64 32>
+; CHECK-NEXT: [[G0:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 8
+; CHECK-NEXT: [[G2:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 24
+; CHECK-NEXT: [[L2:%.*]] = load i64, ptr [[G2]], align 8
+; CHECK-NEXT: br i1 [[C]], label %[[BB:.*]], label %[[EXIT:.*]]
+; CHECK: [[BB]]:
+; CHECK-NEXT: [[TMP3:%.*]] = ptrtoint <4 x ptr> [[TMP2]] to <4 x i64>
+; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i64>, ptr [[G0]], align 8
+; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <4 x i64> [[TMP4]], <4 x i64> poison, <4 x i32> <i32 1, i32 0, i32 2, i32 3>
+; CHECK-NEXT: [[TMP6:%.*]] = add <4 x i64> [[TMP5]], [[TMP3]]
+; CHECK-NEXT: store <4 x i64> [[TMP6]], ptr [[DST]], align 8
+; CHECK-NEXT: br label %[[EXIT]]
+; CHECK: [[EXIT]]:
+; CHECK-NEXT: [[R:%.*]] = phi i64 [ [[L2]], %[[ENTRY]] ], [ 0, %[[BB]] ]
+; CHECK-NEXT: ret i64 [[R]]
+;
+entry:
+ %g0 = getelementptr inbounds i8, ptr %b, i64 8
+ %g1 = getelementptr inbounds i8, ptr %b, i64 16
+ %g2 = getelementptr inbounds i8, ptr %b, i64 24
+ %g3 = getelementptr inbounds i8, ptr %b, i64 32
+ %l2 = load i64, ptr %g2, align 8
+ br i1 %c, label %bb, label %exit
+
+bb:
+ %l1 = load i64, ptr %g1, align 8
+ %l0 = load i64, ptr %g0, align 8
+ %l3 = load i64, ptr %g3, align 8
+ %p1 = ptrtoint ptr %g1 to i64
+ %p0 = ptrtoint ptr %g0 to i64
+ %p2 = ptrtoint ptr %g2 to i64
+ %p3 = ptrtoint ptr %g3 to i64
+ %s1 = add i64 %l1, %p1
+ %s0 = add i64 %l0, %p0
+ %s2 = add i64 %l2, %p2
+ %s3 = add i64 %l3, %p3
+ %d1 = getelementptr inbounds i64, ptr %dst, i64 1
+ %d2 = getelementptr inbounds i64, ptr %dst, i64 2
+ %d3 = getelementptr inbounds i64, ptr %dst, i64 3
+ store i64 %s1, ptr %dst, align 8
+ store i64 %s0, ptr %d1, align 8
+ store i64 %s2, ptr %d2, align 8
+ store i64 %s3, ptr %d3, align 8
+ br label %exit
+
+exit:
+ %r = phi i64 [ %l2, %entry ], [ 0, %bb ]
+ ret i64 %r
+}
+
!0 = !{!1, !1, i64 0}
!1 = !{!"int", !2, i64 0}
!2 = !{!"omnipotent char", !4, i64 0}
|
|
@llvm/pr-subscribers-llvm-transforms Author: Alexey Bataev (alexey-bataev) ChangesThe wide load of the copyable loads uses the pointer of the first load, Fixes #233297 (comment) Assisted-by: Cursor Full diff: https://github.com/llvm/llvm-project/pull/240127.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 920205b1f9ce1..f3efbff76b778 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -8070,6 +8070,9 @@ Instruction *BoUpSLP::getRootEntryInstruction(const TreeEntry &Entry) const {
Entry.State == TreeEntry::StridedVectorize && Entry.isReverse() &&
!isWidenedStridedLoad(&Entry))
return dyn_cast<Instruction>(Entry.Scalars[Entry.ReorderIndices.front()]);
+ // The wide load of the copyable loads uses the pointer of the first load.
+ if (Entry.hasState() && Entry.getOpcode() == Instruction::Load)
+ return Entry.getFirstLoad();
return Entry.hasState() ? Entry.getMainOp()
: dyn_cast<Instruction>(Entry.Scalars.front());
}
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/loads-from-single-predecessor.ll b/llvm/test/Transforms/SLPVectorizer/X86/loads-from-single-predecessor.ll
index f0e8526fc81d6..2ea4b8ef5c534 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/loads-from-single-predecessor.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/loads-from-single-predecessor.ll
@@ -567,6 +567,64 @@ exit:
ret i32 0
}
+; The pointers of the jumbled loads are vectorized, the first load is not the
+; main op of the node.
+define i64 @jumbled_vectorized_ptrs(ptr %b, ptr noalias %dst, i1 %c) {
+; CHECK-LABEL: define i64 @jumbled_vectorized_ptrs(
+; CHECK-SAME: ptr [[B:%.*]], ptr noalias [[DST:%.*]], i1 [[C:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x ptr> poison, ptr [[B]], i64 0
+; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x ptr> [[TMP0]], <4 x ptr> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, <4 x ptr> [[TMP1]], <4 x i64> <i64 16, i64 8, i64 24, i64 32>
+; CHECK-NEXT: [[G0:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 8
+; CHECK-NEXT: [[G2:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 24
+; CHECK-NEXT: [[L2:%.*]] = load i64, ptr [[G2]], align 8
+; CHECK-NEXT: br i1 [[C]], label %[[BB:.*]], label %[[EXIT:.*]]
+; CHECK: [[BB]]:
+; CHECK-NEXT: [[TMP3:%.*]] = ptrtoint <4 x ptr> [[TMP2]] to <4 x i64>
+; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i64>, ptr [[G0]], align 8
+; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <4 x i64> [[TMP4]], <4 x i64> poison, <4 x i32> <i32 1, i32 0, i32 2, i32 3>
+; CHECK-NEXT: [[TMP6:%.*]] = add <4 x i64> [[TMP5]], [[TMP3]]
+; CHECK-NEXT: store <4 x i64> [[TMP6]], ptr [[DST]], align 8
+; CHECK-NEXT: br label %[[EXIT]]
+; CHECK: [[EXIT]]:
+; CHECK-NEXT: [[R:%.*]] = phi i64 [ [[L2]], %[[ENTRY]] ], [ 0, %[[BB]] ]
+; CHECK-NEXT: ret i64 [[R]]
+;
+entry:
+ %g0 = getelementptr inbounds i8, ptr %b, i64 8
+ %g1 = getelementptr inbounds i8, ptr %b, i64 16
+ %g2 = getelementptr inbounds i8, ptr %b, i64 24
+ %g3 = getelementptr inbounds i8, ptr %b, i64 32
+ %l2 = load i64, ptr %g2, align 8
+ br i1 %c, label %bb, label %exit
+
+bb:
+ %l1 = load i64, ptr %g1, align 8
+ %l0 = load i64, ptr %g0, align 8
+ %l3 = load i64, ptr %g3, align 8
+ %p1 = ptrtoint ptr %g1 to i64
+ %p0 = ptrtoint ptr %g0 to i64
+ %p2 = ptrtoint ptr %g2 to i64
+ %p3 = ptrtoint ptr %g3 to i64
+ %s1 = add i64 %l1, %p1
+ %s0 = add i64 %l0, %p0
+ %s2 = add i64 %l2, %p2
+ %s3 = add i64 %l3, %p3
+ %d1 = getelementptr inbounds i64, ptr %dst, i64 1
+ %d2 = getelementptr inbounds i64, ptr %dst, i64 2
+ %d3 = getelementptr inbounds i64, ptr %dst, i64 3
+ store i64 %s1, ptr %dst, align 8
+ store i64 %s0, ptr %d1, align 8
+ store i64 %s2, ptr %d2, align 8
+ store i64 %s3, ptr %d3, align 8
+ br label %exit
+
+exit:
+ %r = phi i64 [ %l2, %entry ], [ 0, %bb ]
+ ret i64 %r
+}
+
!0 = !{!1, !1, i64 0}
!1 = !{!"int", !2, i64 0}
!2 = !{!"omnipotent char", !4, i64 0}
|
llvm-upstreamsync Bot
pushed a commit
to qualcomm/cpullvm-toolchain
that referenced
this pull request
Nov 8, 2026
…able loads The wide load of the copyable loads uses the pointer of the first load, which may differ from the main op after reordering. Use it as the root instruction to correctly detect the pointers, which need extracts. Fixes llvm/llvm-project#233297 (comment) Assisted-by: Cursor Reviewers: Pull Request: llvm/llvm-project#240127
llvm-sync Bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Nov 8, 2026
…able loads The wide load of the copyable loads uses the pointer of the first load, which may differ from the main op after reordering. Use it as the root instruction to correctly detect the pointers, which need extracts. Fixes llvm/llvm-project#233297 (comment) Assisted-by: Cursor Reviewers: Pull Request: llvm/llvm-project#240127
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
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.
The wide load of the copyable loads uses the pointer of the first load,
which may differ from the main op after reordering. Use it as the root
instruction to correctly detect the pointers, which need extracts.
Fixes #233297 (comment)
Assisted-by: Cursor