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

[matejam]

Summary

Remove undef values from the end of the vector operand in image and
buffer store instructions.

Continuation of: 88421ea973916e Trim zero components from buffer and image stores.

[arsenm]

llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp

388–389

You can skip the whole computeKnownFPClass call in this case

llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-simplify-image-buffer-stores.ll

93

Should use poison in place of undef

[matejam]

Instead of call to computeKnownFPClass, use findScalarElement.

[matejam]

Rebase.

[foad]

llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp

387–388

I don't think there's any need for this. Almost everything is an Instruction or a Constant, and anything that is not will just break out of the loop on the first iteration anyway.

391

Should use isNullValue because we don't want to match -0.0.

[matejam]

Remove unnecessary if, use isNullValue instead of isZeroValue.
Thanks @foad.

[foad]

llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp

396

Should really put braces around the "else" part if there are braces around the "if" part.

llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp

 // Trim all zero components from the end of the vector \p UseV and return
 // an appropriate bitset with known elements.
 static APInt trimTrailingZerosInVector(InstCombiner &IC, Value *UseV,
                                        Instruction *I) {
   auto *VTy = cast<FixedVectorType>(UseV->getType());
   unsigned VWidth = VTy->getNumElements();
   APInt DemandedElts = APInt::getAllOnes(VWidth);
 
   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 (!Elt)
       break;
+
+    if (auto *ConstElt = dyn_cast<Constant>(Elt)) {
+      if (!ConstElt->isNullValue() && !isa<UndefValue>(Elt))
+        break;
+    } else
+      break;
+
     DemandedElts.clearBit(i);
   }
+
   return DemandedElts;
 }
 
 static Value *simplifyAMDGCNMemoryIntrinsicDemanded(InstCombiner &IC,
                                                     IntrinsicInst &II,
                                                     APInt DemandedElts,
                                                     int DMaskIdx = -1,
                                                     bool IsLoad = true);

llvm/test/Transforms/InstCombine/AMDGPU/amdgcn-simplify-image-buffer-stores.ll

;
   %newvdata1 = insertelement <4 x float> undef, float 0.0, i32 0
   %newvdata2 = insertelement <4 x float> %newvdata1, float 0.0, i32 1
   %newvdata3 = insertelement <4 x float> %newvdata2, float 0.0, i32 2
   %newvdata4 = insertelement <4 x float> %newvdata3, float %vdata1, i32 3
   call void @llvm.amdgcn.struct.tbuffer.store.v4f32(<4 x float> %newvdata4, <4 x i32> %a, i32 %b, i32 0, i32 42, i32 0, i32 15)
   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
 declare void @llvm.amdgcn.struct.tbuffer.store.v4f32(<4 x float>, <4 x i32>, i32, i32, i32, i32, i32) #0
 declare void @llvm.amdgcn.raw.tbuffer.store.v4f32(<4 x float>, <4 x i32>, i32, i32, i32, i32) #0
 declare void @llvm.amdgcn.tbuffer.store.v4f32(<4 x float>, <4 x i32>, i32, i32, i32, i32, i32, i32, i1, i1) #0
 declare void @llvm.amdgcn.image.store.1d.v4f32.i32(<4 x float>, i32, i32, <8 x i32>, i32, i32) #0
 declare void @llvm.amdgcn.image.store.2d.v4f32.i32(<4 x float>, i32, i32, i32, <8 x i32>, i32, i32) #0