[InstCombine]PtrReplacer: Correctly handle select with unavailable operands #148829
Conversation
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
…erands The testcase I added previously failed because a SelectInst with invalid operands was created (one side `addrspace(4)`, the other `addrspace(5)`). PointerReplacer needs to dig deeper if the true and/or false instructions of the select are not available. Fixes SWDEV-542957
Contributor
Author
This stack of pull requests is managed by Graphite. Learn more about stacking. |
llvmbot
added
backend:AMDGPU
llvm:instcombine
Member
|
@llvm/pr-subscribers-llvm-transforms Author: Pierre van Houtryve (Pierre-vh) ChangesThe testcase I added previously failed because a SelectInst with invalid operands was created (one side Fixes SWDEV-542957 Full diff: https://github.com/llvm/llvm-project/pull/148829.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp b/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
index e791bc4b56e52..2cc1bc9fa4a67 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
@@ -277,6 +277,15 @@ bool PointerReplacer::collectUsers() {
Worklist.emplace_back(I);
};
+ auto TryPushInstOperand = [&](Instruction *InstOp) {
+ if (!UsersToReplace.contains(InstOp)) {
+ if (!ValuesToRevisit.insert(InstOp))
+ return false;
+ Worklist.emplace_back(InstOp);
+ }
+ return true;
+ };
+
PushUsersToWorklist(&Root);
while (!Worklist.empty()) {
Instruction *Inst = Worklist.pop_back_val();
@@ -309,12 +318,8 @@ bool PointerReplacer::collectUsers() {
// incoming values.
Worklist.emplace_back(PHI);
for (unsigned Idx = 0; Idx < PHI->getNumIncomingValues(); ++Idx) {
- auto *IncomingValue = cast<Instruction>(PHI->getIncomingValue(Idx));
- if (UsersToReplace.contains(IncomingValue))
- continue;
- if (!ValuesToRevisit.insert(IncomingValue))
+ if (!TryPushInstOperand(cast<Instruction>(PHI->getIncomingValue(Idx))))
return false;
- Worklist.emplace_back(IncomingValue);
}
} else if (auto *SI = dyn_cast<SelectInst>(Inst)) {
auto *TrueInst = dyn_cast<Instruction>(SI->getTrueValue());
@@ -322,8 +327,17 @@ bool PointerReplacer::collectUsers() {
if (!TrueInst || !FalseInst)
return false;
- UsersToReplace.insert(SI);
- PushUsersToWorklist(SI);
+ if (isAvailable(TrueInst) && isAvailable(FalseInst)) {
+ UsersToReplace.insert(SI);
+ PushUsersToWorklist(SI);
+ continue;
+ }
+
+ // Push select back onto the stack, followed by unavailable true/false
+ // value.
+ Worklist.emplace_back(SI);
+ if (!TryPushInstOperand(TrueInst) || !TryPushInstOperand(FalseInst))
+ return false;
} else if (auto *GEP = dyn_cast<GetElementPtrInst>(Inst)) {
UsersToReplace.insert(GEP);
PushUsersToWorklist(GEP);
diff --git a/llvm/test/Transforms/InstCombine/AMDGPU/ptr-replacer-select-addrspacecast-crash.ll b/llvm/test/Transforms/InstCombine/AMDGPU/ptr-replacer-select-addrspacecast-crash.ll
new file mode 100644
index 0000000000000..921a72364e6f9
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/AMDGPU/ptr-replacer-select-addrspacecast-crash.ll
@@ -0,0 +1,20 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -mtriple=amdgcn-amd-amdhsa -passes=instcombine %s -S -o - | FileCheck %s
+
+; Crashed in IC PtrReplacer because an invalid select was generated with addrspace(4) and addrspace(5)
+; operands.
+
+define amdgpu_kernel void @eggs(ptr addrspace(4) byref([12 x i8]) align 16 %arg) {
+; CHECK-LABEL: define amdgpu_kernel void @eggs(
+; CHECK-SAME: ptr addrspace(4) byref([12 x i8]) align 16 [[ARG:%.*]]) {
+; CHECK-NEXT: [[BB:.*:]]
+; CHECK-NEXT: ret void
+;
+bb:
+ %alloca = alloca i32, i32 0, align 8, addrspace(5)
+ %alloca1 = alloca [12 x i8], align 16, addrspace(5)
+ call void @llvm.memcpy.p5.p4.i64(ptr addrspace(5) %alloca1, ptr addrspace(4) %arg, i64 0, i1 false)
+ %select = select i1 false, ptr addrspace(5) %alloca1, ptr addrspace(5) %alloca
+ call void @llvm.memcpy.p0.p5.i64(ptr null, ptr addrspace(5) %select, i64 0, i1 false)
+ ret void
+}
|
Member
|
@llvm/pr-subscribers-backend-amdgpu Author: Pierre van Houtryve (Pierre-vh) ChangesThe testcase I added previously failed because a SelectInst with invalid operands was created (one side Fixes SWDEV-542957 Full diff: https://github.com/llvm/llvm-project/pull/148829.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp b/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
index e791bc4b56e52..2cc1bc9fa4a67 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
@@ -277,6 +277,15 @@ bool PointerReplacer::collectUsers() {
Worklist.emplace_back(I);
};
+ auto TryPushInstOperand = [&](Instruction *InstOp) {
+ if (!UsersToReplace.contains(InstOp)) {
+ if (!ValuesToRevisit.insert(InstOp))
+ return false;
+ Worklist.emplace_back(InstOp);
+ }
+ return true;
+ };
+
PushUsersToWorklist(&Root);
while (!Worklist.empty()) {
Instruction *Inst = Worklist.pop_back_val();
@@ -309,12 +318,8 @@ bool PointerReplacer::collectUsers() {
// incoming values.
Worklist.emplace_back(PHI);
for (unsigned Idx = 0; Idx < PHI->getNumIncomingValues(); ++Idx) {
- auto *IncomingValue = cast<Instruction>(PHI->getIncomingValue(Idx));
- if (UsersToReplace.contains(IncomingValue))
- continue;
- if (!ValuesToRevisit.insert(IncomingValue))
+ if (!TryPushInstOperand(cast<Instruction>(PHI->getIncomingValue(Idx))))
return false;
- Worklist.emplace_back(IncomingValue);
}
} else if (auto *SI = dyn_cast<SelectInst>(Inst)) {
auto *TrueInst = dyn_cast<Instruction>(SI->getTrueValue());
@@ -322,8 +327,17 @@ bool PointerReplacer::collectUsers() {
if (!TrueInst || !FalseInst)
return false;
- UsersToReplace.insert(SI);
- PushUsersToWorklist(SI);
+ if (isAvailable(TrueInst) && isAvailable(FalseInst)) {
+ UsersToReplace.insert(SI);
+ PushUsersToWorklist(SI);
+ continue;
+ }
+
+ // Push select back onto the stack, followed by unavailable true/false
+ // value.
+ Worklist.emplace_back(SI);
+ if (!TryPushInstOperand(TrueInst) || !TryPushInstOperand(FalseInst))
+ return false;
} else if (auto *GEP = dyn_cast<GetElementPtrInst>(Inst)) {
UsersToReplace.insert(GEP);
PushUsersToWorklist(GEP);
diff --git a/llvm/test/Transforms/InstCombine/AMDGPU/ptr-replacer-select-addrspacecast-crash.ll b/llvm/test/Transforms/InstCombine/AMDGPU/ptr-replacer-select-addrspacecast-crash.ll
new file mode 100644
index 0000000000000..921a72364e6f9
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/AMDGPU/ptr-replacer-select-addrspacecast-crash.ll
@@ -0,0 +1,20 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -mtriple=amdgcn-amd-amdhsa -passes=instcombine %s -S -o - | FileCheck %s
+
+; Crashed in IC PtrReplacer because an invalid select was generated with addrspace(4) and addrspace(5)
+; operands.
+
+define amdgpu_kernel void @eggs(ptr addrspace(4) byref([12 x i8]) align 16 %arg) {
+; CHECK-LABEL: define amdgpu_kernel void @eggs(
+; CHECK-SAME: ptr addrspace(4) byref([12 x i8]) align 16 [[ARG:%.*]]) {
+; CHECK-NEXT: [[BB:.*:]]
+; CHECK-NEXT: ret void
+;
+bb:
+ %alloca = alloca i32, i32 0, align 8, addrspace(5)
+ %alloca1 = alloca [12 x i8], align 16, addrspace(5)
+ call void @llvm.memcpy.p5.p4.i64(ptr addrspace(5) %alloca1, ptr addrspace(4) %arg, i64 0, i1 false)
+ %select = select i1 false, ptr addrspace(5) %alloca1, ptr addrspace(5) %alloca
+ call void @llvm.memcpy.p0.p5.i64(ptr null, ptr addrspace(5) %select, i64 0, i1 false)
+ ret void
+}
|
arsenm
reviewed
Jul 15, 2025
llvm/test/Transforms/InstCombine/AMDGPU/ptr-replacer-select-addrspacecast-crash.ll
Outdated
Show resolved
Hide resolved
llvm/test/Transforms/InstCombine/AMDGPU/ptr-replacer-select-addrspacecast-crash.ll
Outdated
Show resolved
Hide resolved
shiltian
reviewed
Jul 15, 2025
| if (UsersToReplace.contains(IncomingValue)) | ||
| continue; | ||
| if (!ValuesToRevisit.insert(IncomingValue)) | ||
| if (!TryPushInstOperand(cast<Instruction>(PHI->getIncomingValue(Idx)))) |
Contributor
There was a problem hiding this comment.
This is a NFC and is already being tested in AMDGPU/ptr-replace-alloca.ll
gandhi56
reviewed
Jul 15, 2025
| ; Crashed in IC PtrReplacer because an invalid select was generated with addrspace(4) and addrspace(5) | ||
| ; operands. | ||
|
|
||
| define amdgpu_kernel void @eggs(ptr addrspace(4) byref([12 x i8]) align 16 %arg) { |
Contributor
There was a problem hiding this comment.
More descriptive function name, please.
Contributor
There was a problem hiding this comment.
eggs and milk, I actually love it. :-D
Contributor
There was a problem hiding this comment.
Sure, I don't mind as long as other tests are named bagel, croissant, etc. Consistent breakfast is key for a healthy compiler ;)
Contributor
There was a problem hiding this comment.
no I was just kidding :-D
Contributor
Author
There was a problem hiding this comment.
In my defense, the first one came from -passes=metarenamer (which I always use on reduced test cases), but the second one is just me continuing the trend 😄
I'll update the test names
gandhi56
reviewed
Jul 15, 2025
| ret void | ||
| } | ||
|
|
||
| define amdgpu_kernel void @milk(ptr addrspace(4) byref([12 x i8]) align 16 %arg) { |
gandhi56
reviewed
Jul 15, 2025
Contributor
There was a problem hiding this comment.
These tests could be merged into AMDGPU/ptr-replace-alloca.ll
gandhi56
approved these changes
Jul 15, 2025
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/162/builds/26944 Here is the relevant piece of the build log for the reference |
searlmc1
pushed a commit
to ROCm/llvm-project
that referenced
this pull request
Jul 17, 2025
…erands (llvm#3119) The testcase I added previously failed because a SelectInst with invalid operands was created (one side `addrspace(4)`, the other `addrspace(5)`). PointerReplacer needs to dig deeper if the true and/or false instructions of the select are not available. Fixes SWDEV-542957 llvm#148829
rahulc-gh
pushed a commit
to ROCm/llvm-project
that referenced
this pull request
Jul 25, 2025
…erands The testcase I added previously failed because a SelectInst with invalid operands was created (one side `addrspace(4)`, the other `addrspace(5)`). PointerReplacer needs to dig deeper if the true and/or false instructions of the select are not available. Fixes SWDEV-542957 llvm#148829
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 testcase I added previously failed because a SelectInst with invalid operands was created (one side
addrspace(4), the otheraddrspace(5)).PointerReplacer needs to dig deeper if the true and/or false instructions of the select are not available.
Fixes SWDEV-542957