[llvm] [AMDGPU] Use poison instead of undef for non-demanded elements (PR #75914)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-amdgpu

Author: Nikita Popov (nikic)

<details>
<summary>Changes</summary>

Return poison instead of undef for non-demanded lanes in the AMDGPU demanded element simplification hook.

I'm putting up a PR for this in case the notion of dead lanes for these dmask target intrinsics is different from LLVM's usual.

---
Full diff: https://github.com/llvm/llvm-project/pull/75914.diff


3 Files Affected:

- (modified) llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp (+2-2) 
- (modified) llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-demanded-vector-elts-inseltpoison.ll (+1-1) 
- (modified) llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-demanded-vector-elts.ll (+1-1) 


``````````diff
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
index ee93d9eb4c0a05..69867bf218efc9 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
@@ -1261,7 +1261,7 @@ static Value *simplifyAMDGCNMemoryIntrinsicDemanded(InstCombiner &IC,
 
   unsigned NewNumElts = DemandedElts.popcount();
   if (!NewNumElts)
-    return UndefValue::get(IIVTy);
+    return PoisonValue::get(IIVTy);
 
   if (NewNumElts >= VWidth && DemandedElts.isMask()) {
     if (DMaskIdx >= 0)
@@ -1299,7 +1299,7 @@ static Value *simplifyAMDGCNMemoryIntrinsicDemanded(InstCombiner &IC,
 
   if (IsLoad) {
     if (NewNumElts == 1) {
-      return IC.Builder.CreateInsertElement(UndefValue::get(IIVTy), NewCall,
+      return IC.Builder.CreateInsertElement(PoisonValue::get(IIVTy), NewCall,
                                             DemandedElts.countr_zero());
     }
 
diff --git a/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-demanded-vector-elts-inseltpoison.ll b/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-demanded-vector-elts-inseltpoison.ll
index 53300e0c9771a9..1567a9cb7b3f92 100644
--- a/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-demanded-vector-elts-inseltpoison.ll
+++ b/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-demanded-vector-elts-inseltpoison.ll
@@ -4792,7 +4792,7 @@ define amdgpu_ps float @extract_elt0_image_sample_2d_v4f32_f32(float %s, float %
 
 define amdgpu_ps float @extract_elt0_dmask_0000_image_sample_3d_v4f32_f32(float %s, float %t, float %r, <8 x i32> inreg %sampler, <4 x i32> inreg %rsrc) #0 {
 ; CHECK-LABEL: @extract_elt0_dmask_0000_image_sample_3d_v4f32_f32(
-; CHECK-NEXT:    ret float undef
+; CHECK-NEXT:    ret float poison
 ;
   %data = call <4 x float> @llvm.amdgcn.image.sample.3d.v4f32.f32(i32 0, float %s, float %t, float %r, <8 x i32> %sampler, <4 x i32> %rsrc, i1 false, i32 0, i32 0)
   %elt0 = extractelement <4 x float> %data, i32 0
diff --git a/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-demanded-vector-elts.ll b/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-demanded-vector-elts.ll
index ce809a5676b82f..3465bd7be3e678 100644
--- a/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-demanded-vector-elts.ll
+++ b/llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-demanded-vector-elts.ll
@@ -4791,7 +4791,7 @@ define amdgpu_ps float @extract_elt0_image_sample_2d_v4f32_f32(float %s, float %
 
 define amdgpu_ps float @extract_elt0_dmask_0000_image_sample_3d_v4f32_f32(float %s, float %t, float %r, <8 x i32> inreg %sampler, <4 x i32> inreg %rsrc) #0 {
 ; CHECK-LABEL: @extract_elt0_dmask_0000_image_sample_3d_v4f32_f32(
-; CHECK-NEXT:    ret float undef
+; CHECK-NEXT:    ret float poison
 ;
   %data = call <4 x float> @llvm.amdgcn.image.sample.3d.v4f32.f32(i32 0, float %s, float %t, float %r, <8 x i32> %sampler, <4 x i32> %rsrc, i1 false, i32 0, i32 0)
   %elt0 = extractelement <4 x float> %data, i32 0

``````````

</details>


https://github.com/llvm/llvm-project/pull/75914