address pr comments cont

Author: spall

clang/lib/CodeGen/CGExpr.cpp

@@ -2429,7 +2429,7 @@ void CodeGenFunction::EmitStoreThroughLValue(RValue Src, LValue Dst,
       llvm::Value *SrcVal = Src.getScalarVal();
 
       if (SrcVal->getType()->getPrimitiveSizeInBits() <
-                                     VecTy->getScalarSizeInBits())
+          VecTy->getScalarSizeInBits())
         SrcVal = Builder.CreateZExt(SrcVal, VecTy->getScalarType());
 
       auto *IRStoreTy = dyn_cast<llvm::IntegerType>(Vec->getType());
@@ -2655,8 +2655,7 @@ void CodeGenFunction::EmitStoreThroughExtVectorComponentLValue(RValue Src,
 
   if (const VectorType *VTy = Dst.getType()->getAs<VectorType>()) {
     unsigned NumSrcElts = VTy->getNumElements();
-    unsigned NumDstElts =
-        cast<llvm::FixedVectorType>(VecTy)->getNumElements();
+    unsigned NumDstElts = cast<llvm::FixedVectorType>(VecTy)->getNumElements();
     if (NumDstElts == NumSrcElts) {
       // Use shuffle vector is the src and destination are the same number of
       // elements and restore the vector mask since it is on the side it will be
@@ -2666,7 +2665,8 @@ void CodeGenFunction::EmitStoreThroughExtVectorComponentLValue(RValue Src,
         Mask[getAccessedFieldNo(i, Elts)] = i;
 
       llvm::Value *SrcVal = Src.getScalarVal();
-      if (VecTy->getScalarSizeInBits() > SrcVal->getType()->getScalarSizeInBits())
+      if (VecTy->getScalarSizeInBits() >
+          SrcVal->getType()->getScalarSizeInBits())
         SrcVal = Builder.CreateZExt(SrcVal, VecTy);
 
       Vec = Builder.CreateShuffleVector(SrcVal, Mask);
@@ -2679,7 +2679,8 @@ void CodeGenFunction::EmitStoreThroughExtVectorComponentLValue(RValue Src,
       for (unsigned i = 0; i != NumSrcElts; ++i)
         ExtMask.push_back(i);
       ExtMask.resize(NumDstElts, -1);
-      llvm::Value *ExtSrcVal = Builder.CreateShuffleVector(Src.getScalarVal(), ExtMask);
+      llvm::Value *ExtSrcVal =
+          Builder.CreateShuffleVector(Src.getScalarVal(), ExtMask);
       // build identity
       SmallVector<int, 4> Mask;
       for (unsigned i = 0; i != NumDstElts; ++i)

clang/lib/CodeGen/CGExprConstant.cpp

@@ -1978,23 +1978,17 @@ llvm::Constant *ConstantEmitter::emitForMemory(CodeGenModule &CGM,
   }
 
   // Zero-extend bool.
-  if (C->getType()->isIntegerTy(1) && !destType->isBitIntType()) {
+  // In HLSL bool vectors are stored in memory as a vector of i32
+  if ((C->getType()->isIntegerTy(1) && !destType->isBitIntType()) ||
+      (destType->isExtVectorBoolType() &&
+       !destType->isPackedVectorBoolType(CGM.getContext()))) {
     llvm::Type *boolTy = CGM.getTypes().ConvertTypeForMem(destType);
     llvm::Constant *Res = llvm::ConstantFoldCastOperand(
         llvm::Instruction::ZExt, C, boolTy, CGM.getDataLayout());
     assert(Res && "Constant folding must succeed");
     return Res;
   }
 
-  // In HLSL bool vectors are stored in memory as a vector of i32
-  if (destType->isExtVectorBoolType() && CGM.getContext().getLangOpts().HLSL) {
-    llvm::Type *boolVecTy = CGM.getTypes().ConvertTypeForMem(destType);
-    llvm::Constant *Res = llvm::ConstantFoldCastOperand(
-        llvm::Instruction::ZExt, C, boolVecTy, CGM.getDataLayout());
-    assert(Res && "Constant folding must succeed");
-    return Res;
-  }
-
   if (destType->isBitIntType()) {
     ConstantAggregateBuilder Builder(CGM);
     llvm::Type *LoadStoreTy = CGM.getTypes().convertTypeForLoadStore(destType);

clang/test/CodeGenHLSL/BoolVector.hlsl

@@ -108,4 +108,4 @@ void fn6() {
 void fn7() {
   bool2 Arr[2] = {{true,true}, {false,false}};
   Arr[0][1] = false;
-}
+}

clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl

@@ -301,4 +301,4 @@ void AssignBool3(bool2 V) {
 bool2 AccessBools() {
   bool4 X = true.xxxx;
   return X.zw;
-}
+}