[PATCH] D152440: [AMDGPU] Trim trailing undefs from the end of image and buffer store

[Mateja Marjanovic via Phabricator via llvm-commits]

matejam updated this revision to Diff 529915.
matejam added a comment.

Instead of call to computeKnownFPClass, use findScalarElement.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D152440/new/

https://reviews.llvm.org/D152440

Files:
  llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
  llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-simplify-image-buffer-stores.ll


Index: llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-simplify-image-buffer-stores.ll
===================================================================
--- llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-simplify-image-buffer-stores.ll
+++ llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-simplify-image-buffer-stores.ll
@@ -84,6 +84,19 @@
   ret void
 }
 
+define amdgpu_ps void @struct_tbuffer_store_insert_undefs(<4 x i32> inreg %a, float %vdata1, i32 %b) {
+; GCN-LABEL: @struct_tbuffer_store_insert_undefs(
+; GCN-NEXT:    [[TMP1:%.*]] = insertelement <2 x float> <float poison, float 1.000000e+00>, float [[VDATA1:%.*]], i64 0
+; GCN-NEXT:    call void @llvm.amdgcn.struct.tbuffer.store.v2f32(<2 x float> [[TMP1]], <4 x i32> [[A:%.*]], i32 [[B:%.*]], i32 0, i32 42, i32 0, i32 15)
+; GCN-NEXT:    ret void
+;
+  %newvdata1 = insertelement <4 x float> poison, float %vdata1, i32 0
+  %newvdata2 = insertelement <4 x float> %newvdata1, float 1.0, i32 1
+  call void @llvm.amdgcn.struct.tbuffer.store.v4f32(<4 x float> %newvdata2, <4 x i32> %a, i32 %b, i32 0, i32 42, i32 0, i32 15)
+  ret void
+}
+
+
 declare void @llvm.amdgcn.raw.buffer.store.format.v4f32(<4 x float>, <4 x i32>, i32, i32, i32) #2
 declare void @llvm.amdgcn.buffer.store.format.v4f32(<4 x float>, <4 x i32>, i32, i32, i1, i1) #1
 declare void @llvm.amdgcn.struct.buffer.store.format.v4f32(<4 x float>, <4 x i32>, i32, i32, i32, i32) #2
Index: llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
+++ llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp
@@ -384,18 +384,21 @@
   unsigned VWidth = VTy->getNumElements();
   APInt DemandedElts = APInt::getAllOnes(VWidth);
 
+  if (!isa<Constant>(UseV) && !isa<Instruction>(UseV))
+    return DemandedElts;
+
   for (int i = VWidth - 1; i > 0; --i) {
-    APInt DemandOneElt = APInt::getOneBitSet(VWidth, i);
-    KnownFPClass KnownFPClass =
-        computeKnownFPClass(UseV, DemandOneElt, IC.getDataLayout(),
-                            /*InterestedClasses=*/fcAllFlags,
-                            /*Depth=*/0, &IC.getTargetLibraryInfo(),
-                            &IC.getAssumptionCache(), I,
-                            &IC.getDominatorTree());
-    if (KnownFPClass.KnownFPClasses != fcPosZero)
+    auto *Elt = findScalarElement(UseV, i);
+
+    if (auto *ConstElt = dyn_cast<Constant>(Elt)) {
+      if (!ConstElt->isZeroValue() && !isa<UndefValue>(Elt))
+        break;
+    } else
       break;
+
     DemandedElts.clearBit(i);
   }
+
   return DemandedElts;
 }
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D152440.529915.patch
Type: text/x-patch
Size: 2627 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230609/57c7d2d6/attachment.bin>