[llvm][sroa] Disable support for `invariant.group` #151743

tommymcm · 2025-08-01T18:06:21Z

SROA pass does not perform aggregate load/store rewriting on a pointer whose source is a launder.invariant.group.

This causes failed assertion in AllocaSlices.
void (anonymous namespace)::AllocaSlices::SliceBuilder::visitStoreInst(StoreInst &):
 Assertion `(!SI.isSimple() || ValOp->getType()->isSingleValueType()) &&
 "All simple FCA stores should have been pre-split"' failed.

Disables support for {launder,strip}.invariant.group intrinsics in SROA.

Updates SROA test for invariant.group support.

github-actions · 2025-08-01T18:06:38Z

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

llvmbot · 2025-08-01T18:07:10Z

@llvm/pr-subscribers-llvm-transforms

Author: Tommy MᶜMichen (tommymcm)

Changes

Resolves #151574

> SROA pass does not perform aggregate load/store rewriting on a pointer whose source is a launder.invariant.group.
>
> This causes failed assertion in AllocaSlices.
>
> > void (anonymous namespace)::AllocaSlices::SliceBuilder::visitStoreInst(StoreInst &): > Assertion `(!SI.isSimple() || ValOp->getType()->isSingleValueType()) && > "All simple FCA stores should have been pre-split"' failed. >

Disables support for {launder,strip}.invariant.group intrinsics in SROA.

Updates SROA test for invariant.group support.

Full diff: https://github.com/llvm/llvm-project/pull/151743.diff

2 Files Affected:

(modified) llvm/lib/Transforms/Scalar/SROA.cpp (+2-13)
(modified) llvm/test/Transforms/SROA/invariant-group.ll (+27-7)

diff --git a/llvm/lib/Transforms/Scalar/SROA.cpp b/llvm/lib/Transforms/Scalar/SROA.cpp
index 23256cf2acbd2..03d9f32aa3f03 100644
--- a/llvm/lib/Transforms/Scalar/SROA.cpp
+++ b/llvm/lib/Transforms/Scalar/SROA.cpp
@@ -1247,8 +1247,7 @@ class AllocaSlices::SliceBuilder : public PtrUseVisitor<SliceBuilder> {
            "Map index doesn't point back to a slice with this user.");
   }
 
-  // Disable SRoA for any intrinsics except for lifetime invariants and
-  // invariant group.
+  // Disable SRoA for any intrinsics except for lifetime invariants.
   // FIXME: What about debug intrinsics? This matches old behavior, but
   // doesn't make sense.
   void visitIntrinsicInst(IntrinsicInst &II) {
@@ -1268,12 +1267,6 @@ class AllocaSlices::SliceBuilder : public PtrUseVisitor<SliceBuilder> {
       return;
     }
 
-    if (II.isLaunderOrStripInvariantGroup()) {
-      insertUse(II, Offset, AllocSize, true);
-      enqueueUsers(II);
-      return;
-    }
-
     Base::visitIntrinsicInst(II);
   }
 
@@ -3607,8 +3600,7 @@ class AllocaSliceRewriter : public InstVisitor<AllocaSliceRewriter, bool> {
   }
 
   bool visitIntrinsicInst(IntrinsicInst &II) {
-    assert((II.isLifetimeStartOrEnd() || II.isLaunderOrStripInvariantGroup() ||
-            II.isDroppable()) &&
+    assert((II.isLifetimeStartOrEnd() || II.isDroppable()) &&
            "Unexpected intrinsic!");
     LLVM_DEBUG(dbgs() << "    original: " << II << "\n");
 
@@ -3622,9 +3614,6 @@ class AllocaSliceRewriter : public InstVisitor<AllocaSliceRewriter, bool> {
       return true;
     }
 
-    if (II.isLaunderOrStripInvariantGroup())
-      return true;
-
     assert(II.getArgOperand(1) == OldPtr);
     // Lifetime intrinsics are only promotable if they cover the whole alloca.
     // Therefore, we drop lifetime intrinsics which don't cover the whole
diff --git a/llvm/test/Transforms/SROA/invariant-group.ll b/llvm/test/Transforms/SROA/invariant-group.ll
index 1be6f6e2fc32b..147ea3704f6b9 100644
--- a/llvm/test/Transforms/SROA/invariant-group.ll
+++ b/llvm/test/Transforms/SROA/invariant-group.ll
@@ -11,10 +11,17 @@ declare i32 @somevalue()
 
 define void @f() {
 ; CHECK-LABEL: @f(
+; CHECK-NEXT:    [[A:%.*]] = alloca [[T:%.*]], align 8
+; CHECK-NEXT:    [[A1_I8_INV:%.*]] = call ptr @llvm.launder.invariant.group.p0(ptr [[A]])
+; CHECK-NEXT:    [[A2:%.*]] = getelementptr inbounds [[T]], ptr [[A]], i32 0, i32 1
 ; CHECK-NEXT:    [[SV1:%.*]] = call i32 @somevalue()
 ; CHECK-NEXT:    [[SV2:%.*]] = call i32 @somevalue()
-; CHECK-NEXT:    call void @h(i32 [[SV1]])
-; CHECK-NEXT:    call void @h(i32 [[SV2]])
+; CHECK-NEXT:    store i32 [[SV1]], ptr [[A1_I8_INV]], align 4, !invariant.group [[META0:![0-9]+]]
+; CHECK-NEXT:    store i32 [[SV2]], ptr [[A2]], align 4
+; CHECK-NEXT:    [[V1:%.*]] = load i32, ptr [[A1_I8_INV]], align 4, !invariant.group [[META0]]
+; CHECK-NEXT:    [[V2:%.*]] = load i32, ptr [[A2]], align 4
+; CHECK-NEXT:    call void @h(i32 [[V1]])
+; CHECK-NEXT:    call void @h(i32 [[V2]])
 ; CHECK-NEXT:    ret void
 ;
   %a = alloca %t
@@ -44,7 +51,7 @@ define void @g() {
 ; CHECK-NEXT:    [[A2:%.*]] = getelementptr inbounds [[T]], ptr [[A]], i32 0, i32 1
 ; CHECK-NEXT:    [[SV1:%.*]] = call i32 @somevalue()
 ; CHECK-NEXT:    [[SV2:%.*]] = call i32 @somevalue()
-; CHECK-NEXT:    store i32 [[SV1]], ptr [[A1_I8_INV]], align 4, !invariant.group [[META0:![0-9]+]]
+; CHECK-NEXT:    store i32 [[SV1]], ptr [[A1_I8_INV]], align 4, !invariant.group [[META0]]
 ; CHECK-NEXT:    store i32 [[SV2]], ptr [[A2]], align 4
 ; CHECK-NEXT:    [[V1:%.*]] = load i32, ptr [[A1_I8_INV]], align 4, !invariant.group [[META0]]
 ; CHECK-NEXT:    [[V2:%.*]] = load i32, ptr [[A2]], align 4
@@ -81,6 +88,9 @@ define void @g() {
 
 define void @store_and_launder() {
 ; CHECK-LABEL: @store_and_launder(
+; CHECK-NEXT:    [[VALPTR:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    store i32 0, ptr [[VALPTR]], align 4
+; CHECK-NEXT:    [[BARR:%.*]] = call ptr @llvm.launder.invariant.group.p0(ptr [[VALPTR]])
 ; CHECK-NEXT:    ret void
 ;
   %valptr = alloca i32, align 4
@@ -91,7 +101,10 @@ define void @store_and_launder() {
 
 define i32 @launder_and_load() {
 ; CHECK-LABEL: @launder_and_load(
-; CHECK-NEXT:    ret i32 undef
+; CHECK-NEXT:    [[VALPTR:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    [[BARR:%.*]] = call ptr @llvm.launder.invariant.group.p0(ptr [[VALPTR]])
+; CHECK-NEXT:    [[V2:%.*]] = load i32, ptr [[VALPTR]], align 4
+; CHECK-NEXT:    ret i32 [[V2]]
 ;
   %valptr = alloca i32, align 4
   %barr = call ptr @llvm.launder.invariant.group.p0(ptr %valptr)
@@ -101,6 +114,9 @@ define i32 @launder_and_load() {
 
 define void @launder_and_ptr_arith() {
 ; CHECK-LABEL: @launder_and_ptr_arith(
+; CHECK-NEXT:    [[VALPTR:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    [[BARR:%.*]] = call ptr @llvm.launder.invariant.group.p0(ptr [[VALPTR]])
+; CHECK-NEXT:    [[A2:%.*]] = getelementptr inbounds i32, ptr [[VALPTR]], i32 0
 ; CHECK-NEXT:    ret void
 ;
   %valptr = alloca i32, align 4
@@ -140,9 +156,13 @@ end:
 
 define void @partial_promotion_of_alloca() {
 ; CHECK-LABEL: @partial_promotion_of_alloca(
-; CHECK-NEXT:    [[STRUCT_PTR_SROA_2:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    store volatile i32 0, ptr [[STRUCT_PTR_SROA_2]], align 4
-; CHECK-NEXT:    [[STRUCT_PTR_SROA_2_0_STRUCT_PTR_SROA_2_4_LOAD_VAL:%.*]] = load volatile i32, ptr [[STRUCT_PTR_SROA_2]], align 4
+; CHECK-NEXT:    [[STRUCT_PTR:%.*]] = alloca [[T:%.*]], align 4
+; CHECK-NEXT:    [[FIELD_PTR:%.*]] = getelementptr inbounds [[T]], ptr [[STRUCT_PTR]], i32 0, i32 0
+; CHECK-NEXT:    store i32 0, ptr [[FIELD_PTR]], align 4
+; CHECK-NEXT:    [[VOLATILE_FIELD_PTR:%.*]] = getelementptr inbounds [[T]], ptr [[STRUCT_PTR]], i32 0, i32 1
+; CHECK-NEXT:    store volatile i32 0, ptr [[VOLATILE_FIELD_PTR]], align 4, !invariant.group [[META0]]
+; CHECK-NEXT:    [[BARR:%.*]] = call ptr @llvm.launder.invariant.group.p0(ptr [[STRUCT_PTR]])
+; CHECK-NEXT:    [[LOAD_VAL:%.*]] = load volatile i32, ptr [[VOLATILE_FIELD_PTR]], align 4, !invariant.group [[META0]]
 ; CHECK-NEXT:    ret void
 ;
   %struct_ptr = alloca %t, align 4

nikic

Can you please also add the tests from #151574 and #107557 to make sure that these issues are not reintroduced later?

…151574

tommymcm · 2025-08-01T20:20:50Z

Can you please also add the tests from #151574 and #107557 to make sure that these issues are not reintroduced later?

@nikic I've added the tests, thanks for pointing them out.

nikic · 2025-08-02T07:51:10Z

llvm/test/Transforms/SROA/invariant-group.ll

+  ret void
+}
+
+define void @launder_in_loop() {


I don't think the fact that it's in a loop is actually relevant?

%t = type { i32, i32 } define void @test_agg_store() { %struct_ptr = alloca %t, align 4 %struct_ptr_fresh = call ptr @llvm.launder.invariant.group.p0(ptr %struct_ptr) store %t zeroinitializer, ptr %struct_ptr_fresh, align 4, !invariant.group !{} ret void }

is enough to reproduce the crash. To avoid everything being optimized away only keeping the loop body is sufficient:

%t = type { i32, i32 } define void @test_agg_store() { %struct_ptr = alloca %t, align 4 %struct_ptr_fresh = call ptr @llvm.launder.invariant.group.p0(ptr %struct_ptr) %struct = call %t @make_t() store %t %struct, ptr %struct_ptr_fresh, align 4, !invariant.group !{} %first_ptr = getelementptr %t, ptr %struct_ptr_fresh, i32 0, i32 0 %first = load i32, ptr %first_ptr, align 4 %second_ptr = getelementptr %t, ptr %struct_ptr_fresh, i32 0, i32 1 %second = load i32, ptr %second_ptr, align 4 ret void } declare %t @make_t()

Good catch, I've updated the test.

…ant.group test

nikic

LGTM

github-actions · 2025-08-05T07:59:24Z

@tommymcm Congratulations on having your first Pull Request (PR) merged into the LLVM Project!

Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR.

Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues.

How to do this, and the rest of the post-merge process, is covered in detail here.

If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again.

If you don't get any reports, no action is required from you. Your changes are working as expected, well done!

nikic · 2025-08-05T07:59:36Z

/cherry-pick 155359c

llvmbot · 2025-08-05T08:06:26Z

/pull-request #152099

Resolves llvm#151574. > SROA pass does not perform aggregate load/store rewriting on a pointer whose source is a `launder.invariant.group`. > > This causes failed assertion in `AllocaSlices`. > > ``` > void (anonymous namespace)::AllocaSlices::SliceBuilder::visitStoreInst(StoreInst &): > Assertion `(!SI.isSimple() || ValOp->getType()->isSingleValueType()) && > "All simple FCA stores should have been pre-split"' failed. > ``` Disables support for `{launder,strip}.invariant.group` intrinsics in SROA. Updates SROA test for `invariant.group` support. (cherry picked from commit 155359c)

llvmbot added the llvm:transforms label Aug 1, 2025

tommymcm mentioned this pull request Aug 1, 2025

[llvm][sroa] Fixed failing assertion caused by no handling for {launder,strip}.invariant.group in AggLoadStoreRewriter #151574

Closed

nikic reviewed Aug 1, 2025

View reviewed changes

tommymcm added 2 commits August 1, 2025 13:17

[llvm][sroa] Disabled support for invariant.group

5129ea5

[llvm][sroa] Added invariant.group tests from llvm#107557 and llvm#…

02c381f

…151574

tommymcm force-pushed the disable-sroa-invariant-group branch from 30ac2e8 to 02c381f Compare August 1, 2025 20:18

nikic reviewed Aug 2, 2025

View reviewed changes

tommymcm and others added 2 commits August 4, 2025 10:04

[llvm][sroa] Removed unecessary loop from aggregate store with invari…

ee5d72d

…ant.group test

Merge branch 'main' into disable-sroa-invariant-group

a0147cb

nikic approved these changes Aug 4, 2025

View reviewed changes

nikic merged commit 155359c into llvm:main Aug 5, 2025
9 checks passed

nikic added this to the LLVM 21.x Release milestone Aug 5, 2025

github-project-automation bot added this to LLVM Release Status Aug 5, 2025

github-project-automation bot moved this to Needs Triage in LLVM Release Status Aug 5, 2025

llvmbot moved this from Needs Triage to Done in LLVM Release Status Aug 5, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[llvm][sroa] Disable support for `invariant.group` #151743

[llvm][sroa] Disable support for `invariant.group` #151743

Uh oh!

tommymcm commented Aug 1, 2025

Uh oh!

github-actions bot commented Aug 1, 2025

Uh oh!

llvmbot commented Aug 1, 2025

Uh oh!

nikic left a comment

Uh oh!

tommymcm commented Aug 1, 2025

Uh oh!

nikic Aug 2, 2025

Uh oh!

tommymcm Aug 4, 2025

Uh oh!

nikic left a comment

Uh oh!

Uh oh!

github-actions bot commented Aug 5, 2025

Uh oh!

nikic commented Aug 5, 2025

Uh oh!

llvmbot commented Aug 5, 2025

Uh oh!

Uh oh!

[llvm][sroa] Disable support for invariant.group #151743

[llvm][sroa] Disable support for invariant.group #151743

Uh oh!

Conversation

tommymcm commented Aug 1, 2025

Uh oh!

github-actions bot commented Aug 1, 2025

Uh oh!

llvmbot commented Aug 1, 2025

Uh oh!

nikic left a comment

Choose a reason for hiding this comment

Uh oh!

tommymcm commented Aug 1, 2025

Uh oh!

nikic Aug 2, 2025

Choose a reason for hiding this comment

Uh oh!

tommymcm Aug 4, 2025

Choose a reason for hiding this comment

Uh oh!

nikic left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

github-actions bot commented Aug 5, 2025

Uh oh!

nikic commented Aug 5, 2025

Uh oh!

llvmbot commented Aug 5, 2025

Uh oh!

Uh oh!

[llvm][sroa] Disable support for `invariant.group` #151743

[llvm][sroa] Disable support for `invariant.group` #151743