[llvm] AMDGPU: Replace unused export inputs with poison instead of undef (PR #131286)
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Tue Mar 18 03:27:32 PDT 2025
https://github.com/arsenm updated https://github.com/llvm/llvm-project/pull/131286
>From 7893fd4cc7d310d7ff290ed966e69cabf8099404 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Fri, 14 Mar 2025 14:02:19 +0700
Subject: [PATCH] AMDGPU: Replace unused export inputs with poison instead of
undef
---
.../AMDGPU/AMDGPUInstCombineIntrinsic.cpp | 4 +-
.../InstCombine/AMDGPU/amdgcn-intrinsics.ll | 42 +++++++++----------
2 files changed, 23 insertions(+), 23 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
index 0bbded6a3422a..be1274e181ce2 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
@@ -806,8 +806,8 @@ GCNTTIImpl::instCombineIntrinsic(InstCombiner &IC, IntrinsicInst &II) const {
if ((!IsCompr && (EnBits & (1 << I)) == 0) ||
(IsCompr && ((EnBits & (0x3 << (2 * I))) == 0))) {
Value *Src = II.getArgOperand(I + 2);
- if (!isa<UndefValue>(Src)) {
- IC.replaceOperand(II, I + 2, UndefValue::get(Src->getType()));
+ if (!isa<PoisonValue>(Src)) {
+ IC.replaceOperand(II, I + 2, PoisonValue::get(Src->getType()));
Changed = true;
}
}
diff --git a/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll b/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll
index 124ea57be35e9..9a42fa723dc48 100644
--- a/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll
+++ b/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll
@@ -1776,18 +1776,18 @@ declare void @llvm.amdgcn.exp.f32(i32 immarg, i32 immarg, float, float, float, f
define void @exp_disabled_inputs_to_undef(float %x, float %y, float %z, float %w) {
; enable src0..src3 constants
; CHECK-LABEL: @exp_disabled_inputs_to_undef(
-; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float 1.000000e+00, float undef, float undef, float undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float undef, float 2.000000e+00, float undef, float undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 4, float undef, float undef, float 5.000000e-01, float undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 8, float undef, float undef, float undef, float 4.000000e+00, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float [[X:%.*]], float undef, float undef, float undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float undef, float [[Y:%.*]], float undef, float undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 4, float undef, float undef, float [[Z:%.*]], float undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 8, float undef, float undef, float undef, float [[W:%.*]], i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 0, float undef, float undef, float undef, float undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 3, float 1.000000e+00, float 2.000000e+00, float undef, float undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 5, float 1.000000e+00, float undef, float 5.000000e-01, float undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 9, float 1.000000e+00, float undef, float undef, float 4.000000e+00, i1 false, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float 1.000000e+00, float poison, float poison, float poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float poison, float 2.000000e+00, float poison, float poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 4, float poison, float poison, float 5.000000e-01, float poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 8, float poison, float poison, float poison, float 4.000000e+00, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float [[X:%.*]], float poison, float poison, float poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float poison, float [[Y:%.*]], float poison, float poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 4, float poison, float poison, float [[Z:%.*]], float poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 8, float poison, float poison, float poison, float [[W:%.*]], i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 0, float poison, float poison, float poison, float poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 3, float 1.000000e+00, float 2.000000e+00, float poison, float poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 5, float 1.000000e+00, float poison, float 5.000000e-01, float poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 9, float 1.000000e+00, float poison, float poison, float 4.000000e+00, i1 false, i1 false)
; CHECK-NEXT: call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float 1.000000e+00, float 2.000000e+00, float 5.000000e-01, float 4.000000e+00, i1 false, i1 false)
; CHECK-NEXT: ret void
;
@@ -1824,15 +1824,15 @@ declare void @llvm.amdgcn.exp.compr.v2f16(i32 immarg, i32 immarg, <2 x half>, <2
define void @exp_compr_disabled_inputs_to_undef(<2 x half> %xy, <2 x half> %zw) {
; CHECK-LABEL: @exp_compr_disabled_inputs_to_undef(
-; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 0, <2 x half> undef, <2 x half> undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 1, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 2, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 3, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 0, <2 x half> undef, <2 x half> undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 1, <2 x half> [[XY:%.*]], <2 x half> undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 2, <2 x half> [[XY]], <2 x half> undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 3, <2 x half> [[XY]], <2 x half> undef, i1 true, i1 false)
-; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 12, <2 x half> undef, <2 x half> [[ZW:%.*]], i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 0, <2 x half> poison, <2 x half> poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 1, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 2, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 3, <2 x half> <half 0xH3C00, half 0xH4000>, <2 x half> poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 0, <2 x half> poison, <2 x half> poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 1, <2 x half> [[XY:%.*]], <2 x half> poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 2, <2 x half> [[XY]], <2 x half> poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 3, <2 x half> [[XY]], <2 x half> poison, i1 true, i1 false)
+; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 12, <2 x half> poison, <2 x half> [[ZW:%.*]], i1 true, i1 false)
; CHECK-NEXT: call void @llvm.amdgcn.exp.compr.v2f16(i32 0, i32 15, <2 x half> [[XY]], <2 x half> [[ZW]], i1 true, i1 false)
; CHECK-NEXT: ret void
;
More information about the llvm-commits
mailing list