[AMDGPU] Visit all PHIs in each call to optimizeLiveType #147522
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
Make the Visited set a local variable, otherwise we can reject a PHI (those that do not have a zeroinitializer constant) but mark it as visited, and the rest of the function thinks the PHI is ok when it isn't. This is a bit crude but it's the only fix that consistently worked in my testing. Fixes SWDEV-541767
|
@llvm/pr-subscribers-backend-amdgpu Author: Pierre van Houtryve (Pierre-vh) ChangesMake the Visited set a local variable, otherwise we can reject a PHI Fixes SWDEV-541767 Full diff: https://github.com/llvm/llvm-project/pull/147522.diff 2 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULateCodeGenPrepare.cpp b/llvm/lib/Target/AMDGPU/AMDGPULateCodeGenPrepare.cpp
index df976cf3f7fdb..523c66c72273c 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULateCodeGenPrepare.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULateCodeGenPrepare.cpp
@@ -79,8 +79,6 @@ class LiveRegOptimizer {
/// The scalar type to convert to
Type *const ConvertToScalar;
- /// The set of visited Instructions
- SmallPtrSet<Instruction *, 4> Visited;
/// Map of Value -> Converted Value
ValueToValueMap ValMap;
/// Map of containing conversions from Optimal Type -> Original Type per BB.
@@ -288,6 +286,7 @@ bool LiveRegOptimizer::optimizeLiveType(
SmallPtrSet<PHINode *, 4> PhiNodes;
SmallPtrSet<Instruction *, 4> Defs;
SmallPtrSet<Instruction *, 4> Uses;
+ SmallPtrSet<Instruction *, 4> Visited;
Worklist.push_back(cast<Instruction>(I));
while (!Worklist.empty()) {
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-late-codegenprepare-crash-splat.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-late-codegenprepare-crash-splat.ll
new file mode 100644
index 0000000000000..837b5e0ab5833
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-late-codegenprepare-crash-splat.ll
@@ -0,0 +1,94 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes="amdgpu-late-codegenprepare,verify" %s | FileCheck %s
+
+; This crashed because the PHI with a splat was rejected, but then we marked the PHI
+; as visited and tried to convert one of its user afterwards.
+
+define amdgpu_kernel void @widget(i1 %arg, <4 x i8> %arg1, i64 %arg2) {
+; CHECK-LABEL: define amdgpu_kernel void @widget(
+; CHECK-SAME: i1 [[ARG:%.*]], <4 x i8> [[ARG1:%.*]], i64 [[ARG2:%.*]]) {
+; CHECK-NEXT: [[BB:.*]]:
+; CHECK-NEXT: [[WIDGET_KERNARG_SEGMENT:%.*]] = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
+; CHECK-NEXT: [[ARG_KERNARG_OFFSET_ALIGN_DOWN:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[WIDGET_KERNARG_SEGMENT]], i64 36
+; CHECK-NEXT: [[TMP0:%.*]] = load i32, ptr addrspace(4) [[ARG_KERNARG_OFFSET_ALIGN_DOWN]], align 4
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i32 [[TMP0]] to i1
+; CHECK-NEXT: [[ARG1_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[WIDGET_KERNARG_SEGMENT]], i64 40
+; CHECK-NEXT: [[ARG1_LOAD:%.*]] = load <4 x i8>, ptr addrspace(4) [[ARG1_KERNARG_OFFSET]], align 8
+; CHECK-NEXT: [[ARG2_KERNARG_OFFSET:%.*]] = getelementptr inbounds i8, ptr addrspace(4) [[WIDGET_KERNARG_SEGMENT]], i64 44
+; CHECK-NEXT: [[ARG2_LOAD:%.*]] = load i64, ptr addrspace(4) [[ARG2_KERNARG_OFFSET]], align 4
+; CHECK-NEXT: br label %[[BB_3:.*]]
+; CHECK: [[BB_3]]:
+; CHECK-NEXT: [[PHI:%.*]] = phi ptr addrspace(1) [ null, %[[BB]] ], [ [[GETELEMENTPTR:%.*]], %[[BB_14:.*]] ]
+; CHECK-NEXT: [[PHI4:%.*]] = phi <4 x i8> [ splat (i8 1), %[[BB]] ], [ [[PHI15:%.*]], %[[BB_14]] ]
+; CHECK-NEXT: br i1 [[TMP1]], label %[[BB_6_PREHEADER:.*]], label %[[BB_5:.*]]
+; CHECK: [[BB_5]]:
+; CHECK-NEXT: br label %[[BB_14]]
+; CHECK: [[BB_6_PREHEADER]]:
+; CHECK-NEXT: br label %[[BB_6:.*]]
+; CHECK: [[BB_6]]:
+; CHECK-NEXT: [[PHI7:%.*]] = phi <4 x i8> [ [[PHI13:%.*]], %[[BB_12:.*]] ], [ [[PHI4]], %[[BB_6_PREHEADER]] ]
+; CHECK-NEXT: br i1 [[TMP1]], label %[[BB_8:.*]], label %[[BB_12]]
+; CHECK: [[BB_8]]:
+; CHECK-NEXT: br i1 [[TMP1]], label %[[BB_10:.*]], label %[[BB_9:.*]]
+; CHECK: [[BB_9]]:
+; CHECK-NEXT: br label %[[BB_10]]
+; CHECK: [[BB_10]]:
+; CHECK-NEXT: [[PHI11:%.*]] = phi <4 x i8> [ [[PHI7]], %[[BB_9]] ], [ zeroinitializer, %[[BB_8]] ]
+; CHECK-NEXT: [[EXTRACTELEMENT:%.*]] = extractelement <4 x i8> [[PHI11]], i64 0
+; CHECK-NEXT: store i8 [[EXTRACTELEMENT]], ptr addrspace(1) [[PHI]], align 1
+; CHECK-NEXT: br label %[[BB_12]]
+; CHECK: [[BB_12]]:
+; CHECK-NEXT: [[PHI13]] = phi <4 x i8> [ zeroinitializer, %[[BB_10]] ], [ [[PHI7]], %[[BB_6]] ]
+; CHECK-NEXT: br i1 [[TMP1]], label %[[BB_6]], label %[[BB_14]]
+; CHECK: [[BB_14]]:
+; CHECK-NEXT: [[PHI15]] = phi <4 x i8> [ [[ARG1_LOAD]], %[[BB_5]] ], [ zeroinitializer, %[[BB_12]] ]
+; CHECK-NEXT: [[GETELEMENTPTR]] = getelementptr i8, ptr addrspace(1) [[PHI]], i64 [[ARG2_LOAD]]
+; CHECK-NEXT: br label %[[BB_3]]
+;
+bb:
+ %widget.kernarg.segment = call nonnull align 16 dereferenceable(272) ptr addrspace(4) @llvm.amdgcn.kernarg.segment.ptr()
+ %arg.kernarg.offset.align.down = getelementptr inbounds i8, ptr addrspace(4) %widget.kernarg.segment, i64 36
+ %0 = load i32, ptr addrspace(4) %arg.kernarg.offset.align.down, align 4
+ %1 = trunc i32 %0 to i1
+ %arg1.kernarg.offset = getelementptr inbounds i8, ptr addrspace(4) %widget.kernarg.segment, i64 40
+ %arg1.load = load <4 x i8>, ptr addrspace(4) %arg1.kernarg.offset, align 8
+ %arg2.kernarg.offset = getelementptr inbounds i8, ptr addrspace(4) %widget.kernarg.segment, i64 44
+ %arg2.load = load i64, ptr addrspace(4) %arg2.kernarg.offset, align 4
+ br label %bb.3
+
+bb.3: ; preds = %bb.14, %bb
+ %phi = phi ptr addrspace(1) [ null, %bb ], [ %getelementptr, %bb.14 ]
+ %phi4 = phi <4 x i8> [ splat (i8 1), %bb ], [ %phi15, %bb.14 ]
+ br i1 %1, label %bb.6.preheader, label %bb.5
+
+bb.5: ; preds = %bb.3
+ br label %bb.14
+
+bb.6.preheader: ; preds = %bb.3
+ br label %bb.6
+
+bb.6: ; preds = %bb.6.preheader, %bb.12
+ %phi7 = phi <4 x i8> [ %phi13, %bb.12 ], [ %phi4, %bb.6.preheader ]
+ br i1 %1, label %bb.8, label %bb.12
+
+bb.8: ; preds = %bb.6
+ br i1 %1, label %bb.10, label %bb.9
+
+bb.9: ; preds = %bb.8
+ br label %bb.10
+
+bb.10: ; preds = %bb.9, %bb.8
+ %phi11 = phi <4 x i8> [ %phi7, %bb.9 ], [ zeroinitializer, %bb.8 ]
+ %extractelement = extractelement <4 x i8> %phi11, i64 0
+ store i8 %extractelement, ptr addrspace(1) %phi, align 1
+ br label %bb.12
+
+bb.12: ; preds = %bb.10, %bb.6
+ %phi13 = phi <4 x i8> [ zeroinitializer, %bb.10 ], [ %phi7, %bb.6 ]
+ br i1 %1, label %bb.6, label %bb.14
+
+bb.14: ; preds = %bb.5, %bb.12
+ %phi15 = phi <4 x i8> [ %arg1.load, %bb.5 ], [ zeroinitializer, %bb.12 ]
+ %getelementptr = getelementptr i8, ptr addrspace(1) %phi, i64 %arg2.load
+ br label %bb.3
+}
|
shiltian
reviewed
Jul 8, 2025
llvm/test/CodeGen/AMDGPU/amdgpu-late-codegenprepare-crash-splat.ll
Outdated
Show resolved
Hide resolved
shiltian
reviewed
Jul 8, 2025
arsenm
reviewed
Jul 8, 2025
llvm/test/CodeGen/AMDGPU/amdgpu-late-codegenprepare-crash-splat.ll
Outdated
Show resolved
Hide resolved
shiltian
approved these changes
Jul 9, 2025
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.
Make the Visited set a local variable, otherwise we can reject a PHI
(those that do not have a zeroinitializer constant) but mark it as visited,
and the rest of the function thinks the PHI is ok when it isn't.
This is a bit crude but it's the only fix that consistently worked in my testing.
Fixes SWDEV-541767