https://github.com/arsenm created
https://github.com/llvm/llvm-project/pull/131288
None
>From b9480c7cf5ec5c227352feb77fac9baf5e11462b Mon Sep 17 00:00:00 2001
From: Matt Arsenault <matthew.arsena...@amd.com>
Date: Fri, 14 Mar 2025 14:12:08 +0700
Subject: [PATCH] AMDGPU: Replace unused permlane inputs with poison instead of
undef
---
llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp | 4 ++--
.../Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll | 8 ++++----
2 files changed, 6 insertions(+), 6 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
b/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
index 26c48c19ebfd8..7cd97e95b0189 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
@@ -1115,7 +1115,7 @@ GCNTTIImpl::instCombineIntrinsic(InstCombiner &IC,
IntrinsicInst &II) const {
case Intrinsic::amdgcn_permlanex16_var: {
// Discard vdst_in if it's not going to be read.
Value *VDstIn = II.getArgOperand(0);
- if (isa<UndefValue>(VDstIn))
+ if (isa<PoisonValue>(VDstIn))
break;
// FetchInvalid operand idx.
@@ -1134,7 +1134,7 @@ GCNTTIImpl::instCombineIntrinsic(InstCombiner &IC,
IntrinsicInst &II) const {
if (!FetchInvalid->getZExtValue() && !BoundCtrl->getZExtValue())
break;
- return IC.replaceOperand(II, 0, UndefValue::get(VDstIn->getType()));
+ return IC.replaceOperand(II, 0, PoisonValue::get(VDstIn->getType()));
}
case Intrinsic::amdgcn_permlane64:
case Intrinsic::amdgcn_readfirstlane:
diff --git a/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll
b/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll
index 3e30b2a5b0911..99e6fc98588ae 100644
--- a/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll
+++ b/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll
@@ -3272,7 +3272,7 @@ define amdgpu_kernel void @permlane16(ptr addrspace(1)
%out, i32 %src0, i32 %src
define amdgpu_kernel void @permlane16_bound_ctrl(ptr addrspace(1) %out, i32
%src0, i32 %src1, i32 %src2) {
; CHECK-LABEL: @permlane16_bound_ctrl(
-; CHECK-NEXT: [[RES:%.*]] = call i32 @llvm.amdgcn.permlane16.i32(i32 undef,
i32 [[SRC0:%.*]], i32 [[SRC1:%.*]], i32 [[SRC2:%.*]], i1 false, i1 true)
+; CHECK-NEXT: [[RES:%.*]] = call i32 @llvm.amdgcn.permlane16.i32(i32
poison, i32 [[SRC0:%.*]], i32 [[SRC1:%.*]], i32 [[SRC2:%.*]], i1 false, i1 true)
; CHECK-NEXT: store i32 [[RES]], ptr addrspace(1) [[OUT:%.*]], align 4
; CHECK-NEXT: ret void
;
@@ -3283,7 +3283,7 @@ define amdgpu_kernel void @permlane16_bound_ctrl(ptr
addrspace(1) %out, i32 %src
define amdgpu_kernel void @permlane16_fetch_invalid_bound_ctrl(ptr
addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; CHECK-LABEL: @permlane16_fetch_invalid_bound_ctrl(
-; CHECK-NEXT: [[RES:%.*]] = call i32 @llvm.amdgcn.permlane16.i32(i32 undef,
i32 [[SRC0:%.*]], i32 [[SRC1:%.*]], i32 [[SRC2:%.*]], i1 true, i1 true)
+; CHECK-NEXT: [[RES:%.*]] = call i32 @llvm.amdgcn.permlane16.i32(i32
poison, i32 [[SRC0:%.*]], i32 [[SRC1:%.*]], i32 [[SRC2:%.*]], i1 true, i1 true)
; CHECK-NEXT: store i32 [[RES]], ptr addrspace(1) [[OUT:%.*]], align 4
; CHECK-NEXT: ret void
;
@@ -3311,7 +3311,7 @@ define amdgpu_kernel void @permlanex16(ptr addrspace(1)
%out, i32 %src0, i32 %sr
define amdgpu_kernel void @permlanex16_bound_ctrl(ptr addrspace(1) %out, i32
%src0, i32 %src1, i32 %src2) {
; CHECK-LABEL: @permlanex16_bound_ctrl(
-; CHECK-NEXT: [[RES:%.*]] = call i32 @llvm.amdgcn.permlanex16.i32(i32
undef, i32 [[SRC0:%.*]], i32 [[SRC1:%.*]], i32 [[SRC2:%.*]], i1 false, i1 true)
+; CHECK-NEXT: [[RES:%.*]] = call i32 @llvm.amdgcn.permlanex16.i32(i32
poison, i32 [[SRC0:%.*]], i32 [[SRC1:%.*]], i32 [[SRC2:%.*]], i1 false, i1 true)
; CHECK-NEXT: store i32 [[RES]], ptr addrspace(1) [[OUT:%.*]], align 4
; CHECK-NEXT: ret void
;
@@ -3322,7 +3322,7 @@ define amdgpu_kernel void @permlanex16_bound_ctrl(ptr
addrspace(1) %out, i32 %sr
define amdgpu_kernel void @permlanex16_fetch_invalid_bound_ctrl(ptr
addrspace(1) %out, i32 %src0, i32 %src1, i32 %src2) {
; CHECK-LABEL: @permlanex16_fetch_invalid_bound_ctrl(
-; CHECK-NEXT: [[RES:%.*]] = call i32 @llvm.amdgcn.permlanex16.i32(i32
undef, i32 [[SRC0:%.*]], i32 [[SRC1:%.*]], i32 [[SRC2:%.*]], i1 true, i1 true)
+; CHECK-NEXT: [[RES:%.*]] = call i32 @llvm.amdgcn.permlanex16.i32(i32
poison, i32 [[SRC0:%.*]], i32 [[SRC1:%.*]], i32 [[SRC2:%.*]], i1 true, i1 true)
; CHECK-NEXT: store i32 [[RES]], ptr addrspace(1) [[OUT:%.*]], align 4
; CHECK-NEXT: ret void
;
_______________________________________________
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
