[clang] [HLSL] Make memory representation of boolean vectors in HLSL, vectors of i32. Add support for boolean swizzling. (PR #123977)

Sarah Spall via cfe-commits cfe-commits at lists.llvm.org
Wed Feb 12 19:27:16 PST 2025


https://github.com/spall updated https://github.com/llvm/llvm-project/pull/123977

>From 50f8b16bf83d70cc37f2ac7e70c9195a4c4cda02 Mon Sep 17 00:00:00 2001
From: Sarah Spall <spall at planetbauer.com>
Date: Thu, 16 Jan 2025 01:32:25 +0000
Subject: [PATCH 01/12] wip

---
 clang/lib/CodeGen/CGExpr.cpp       | 19 +++++++++++++++++++
 clang/lib/CodeGen/CodeGenTypes.cpp |  6 ++++++
 2 files changed, 25 insertions(+)

diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
index 054f8d1eadb8c..be8effb4dd40e 100644
--- a/clang/lib/CodeGen/CGExpr.cpp
+++ b/clang/lib/CodeGen/CGExpr.cpp
@@ -1986,6 +1986,10 @@ llvm::Value *CodeGenFunction::EmitLoadOfScalar(Address Addr, bool Volatile,
   if (const auto *ClangVecTy = Ty->getAs<VectorType>()) {
     // Boolean vectors use `iN` as storage type.
     if (ClangVecTy->isExtVectorBoolType()) {
+      if (getLangOpts().HLSL) {
+	llvm::Value *Value = Builder.CreateLoad(Addr, Volatile, "load_boolvec");
+	return EmitFromMemory(Value, Ty);
+      }
       llvm::Type *ValTy = ConvertType(Ty);
       unsigned ValNumElems =
           cast<llvm::FixedVectorType>(ValTy)->getNumElements();
@@ -2064,6 +2068,9 @@ llvm::Value *CodeGenFunction::EmitToMemory(llvm::Value *Value, QualType Ty) {
 
   if (Ty->isExtVectorBoolType()) {
     llvm::Type *StoreTy = convertTypeForLoadStore(Ty, Value->getType());
+    if (getLangOpts().HLSL)
+      return Builder.CreateZExt(Value, StoreTy);
+
     // Expand to the memory bit width.
     unsigned MemNumElems = StoreTy->getPrimitiveSizeInBits();
     // <N x i1> --> <P x i1>.
@@ -2081,6 +2088,9 @@ llvm::Value *CodeGenFunction::EmitToMemory(llvm::Value *Value, QualType Ty) {
 llvm::Value *CodeGenFunction::EmitFromMemory(llvm::Value *Value, QualType Ty) {
   if (Ty->isExtVectorBoolType()) {
     const auto *RawIntTy = Value->getType();
+    if (getLangOpts().HLSL)
+      return Builder.CreateTrunc(Value, ConvertType(Ty), "loadedv");
+	
     // Bitcast iP --> <P x i1>.
     auto *PaddedVecTy = llvm::FixedVectorType::get(
         Builder.getInt1Ty(), RawIntTy->getPrimitiveSizeInBits());
@@ -2407,6 +2417,11 @@ void CodeGenFunction::EmitStoreThroughLValue(RValue Src, LValue Dst,
       // Read/modify/write the vector, inserting the new element.
       llvm::Value *Vec = Builder.CreateLoad(Dst.getVectorAddress(),
                                             Dst.isVolatileQualified());
+      llvm::Type *OldVecTy = Vec->getType();
+      if (getLangOpts().HLSL && Dst.getType()->isExtVectorBoolType())
+        
+	Vec = Builder.CreateTrunc(Vec, ConvertType(Dst.getType()), "truncboolv");
+      
       auto *IRStoreTy = dyn_cast<llvm::IntegerType>(Vec->getType());
       if (IRStoreTy) {
         auto *IRVecTy = llvm::FixedVectorType::get(
@@ -2420,6 +2435,10 @@ void CodeGenFunction::EmitStoreThroughLValue(RValue Src, LValue Dst,
         // <N x i1> --> <iN>.
         Vec = Builder.CreateBitCast(Vec, IRStoreTy);
       }
+
+      if (getLangOpts().HLSL && Dst.getType()->isExtVectorBoolType())
+	Vec = Builder.CreateZExt(Vec, OldVecTy);
+      
       Builder.CreateStore(Vec, Dst.getVectorAddress(),
                           Dst.isVolatileQualified());
       return;
diff --git a/clang/lib/CodeGen/CodeGenTypes.cpp b/clang/lib/CodeGen/CodeGenTypes.cpp
index 09191a4901f49..e7f280e3823a3 100644
--- a/clang/lib/CodeGen/CodeGenTypes.cpp
+++ b/clang/lib/CodeGen/CodeGenTypes.cpp
@@ -112,6 +112,12 @@ llvm::Type *CodeGenTypes::ConvertTypeForMem(QualType T) {
   // Check for the boolean vector case.
   if (T->isExtVectorBoolType()) {
     auto *FixedVT = cast<llvm::FixedVectorType>(R);
+    
+    if (Context.getLangOpts().HLSL) {
+      llvm::Type *IRElemTy = ConvertTypeForMem(Context.BoolTy);
+      return llvm::FixedVectorType::get(IRElemTy, FixedVT->getNumElements());
+    }
+    
     // Pad to at least one byte.
     uint64_t BytePadded = std::max<uint64_t>(FixedVT->getNumElements(), 8);
     return llvm::IntegerType::get(FixedVT->getContext(), BytePadded);

>From b231f4e47ed263bf798455d8d303383caf1be635 Mon Sep 17 00:00:00 2001
From: Sarah Spall <spall at planetbauer.com>
Date: Thu, 16 Jan 2025 19:30:58 +0000
Subject: [PATCH 02/12] bool vector swizzle is allowed in hlsl

---
 clang/lib/Sema/SemaExprMember.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/clang/lib/Sema/SemaExprMember.cpp b/clang/lib/Sema/SemaExprMember.cpp
index d130e8b86bc56..dcf7feab752b5 100644
--- a/clang/lib/Sema/SemaExprMember.cpp
+++ b/clang/lib/Sema/SemaExprMember.cpp
@@ -1697,7 +1697,7 @@ static ExprResult LookupMemberExpr(Sema &S, LookupResult &R,
         QualType(), false);
   }
 
-  if (BaseType->isExtVectorBoolType()) {
+  if (BaseType->isExtVectorBoolType() && !S.Context.getLangOpts().HLSL) {
     // We disallow element access for ext_vector_type bool.  There is no way to
     // materialize a reference to a vector element as a pointer (each element is
     // one bit in the vector).

>From dd7e4595f47fb02fab4450e69a5ab07fd21f67d3 Mon Sep 17 00:00:00 2001
From: Sarah Spall <spall at planetbauer.com>
Date: Tue, 21 Jan 2025 16:50:33 +0000
Subject: [PATCH 03/12] swizzle tests for boolean vectors

---
 .../CodeGenHLSL/builtins/ScalarSwizzles.hlsl  | 79 ++++++++++++++++++-
 1 file changed, 77 insertions(+), 2 deletions(-)

diff --git a/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl b/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl
index 97711c9ee25a1..8e246959df94f 100644
--- a/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl
@@ -11,13 +11,23 @@ int2 ToTwoInts(int V){
 }
 
 // CHECK-LABEL: ToFourFloats
-// [[splat:%.*]] = insertelement <1 x float> poison, float {{.*}}, i64 0
-// [[vec4:%.*]] = shufflevector <1 x float> [[splat]], <1 x float> poison, <4 x i32> zeroinitializer
+// CHECK: [[splat:%.*]] = insertelement <1 x float> poison, float {{.*}}, i64 0
+// CHECK: [[vec4:%.*]] = shufflevector <1 x float> [[splat]], <1 x float> poison, <4 x i32> zeroinitializer
 // ret <4 x float> [[vec4]]
 float4 ToFourFloats(float V){
   return V.rrrr;
 }
 
+// CHECK-LABEL: ToFourBools
+// CHECK: {{%.*}} = zext i1 {{.*}} to i32
+// CHECK: [[splat:%.*]] = insertelement <1 x i32> poison, i32 {{.*}}, i64 0
+// CHECK-NEXT: [[vec4:%.*]] = shufflevector <1 x i32> [[splat]], <1 x i32> poison, <4 x i32> zeroinitializer
+// CHECK-NEXT: [[vec2Ret:%.*]] = trunc <4 x i32> [[vec4]] to <4 x i1>
+// CHECK-NEXT: ret <4 x i1> [[vec2Ret]]
+bool4 ToFourBools(bool V) {
+  return V.rrrr;
+}
+
 // CHECK-LABEL: FillOne
 // CHECK: [[vec1Ptr:%.*]] = alloca <1 x i32>, align 4
 // CHECK: store <1 x i32> splat (i32 1), ptr [[vec1Ptr]], align 4
@@ -93,6 +103,17 @@ vector<float, 1> FillOneHalfFloat(){
   return .5f.r;
 }
 
+// CHECK-LABEL: FillTrue
+// CHECK: [[Tmp:%.*]] = alloca <1 x i32>, align 1
+// CHECK-NEXT: store <1 x i1> splat (i1 true), ptr [[Tmp]], align 1
+// CHECK-NEXT: [[Vec1:%.*]] = load <1 x i32>, ptr [[Tmp]], align 1
+// CHECK-NEXT: [[Vec2:%.*]] = shufflevector <1 x i32> [[Vec1]], <1 x i32> poison, <2 x i32> zeroinitializer
+// CHECK-NEXT: [[Vec2Ret:%.*]] = trunc <2 x i32> [[Vec2]] to <2 x i1>
+// CHECK-NEXT: ret <2 x i1> [[Vec2Ret]]
+bool2 FillTrue() {
+  return true.xx;
+}
+
 // The initial codegen for this case is correct but a bit odd. The IR optimizer
 // cleans this up very nicely.
 
@@ -110,6 +131,24 @@ float2 HowManyFloats(float V) {
   return V.rr.rr;
 }
 
+// CHECK-LABEL: HowManyBools
+// CHECK: [[VAddr:%.*]] = alloca i32, align 4
+// CHECK-NEXT: [[Vec2Ptr:%.*]] = alloca <2 x i32>, align 1
+// CHECK-NEXT: [[Tmp:%.*]] = zext i1 {{.*}} to i32
+// CHECK-NEXT: store i32 [[Tmp]], ptr [[VAddr]], align 4
+// CHECK-NEXT: [[VVal:%.*]] = load i32, ptr [[VAddr]], align 4
+// CHECK-NEXT: [[Splat:%.*]] = insertelement <1 x i32> poison, i32 [[VVal]], i64 0
+// CHECK-NEXT: [[Vec2:%.*]] = shufflevector <1 x i32> [[Splat]], <1 x i32> poison, <2 x i32> zeroinitializer
+// CHECK-NEXT: [[Trunc:%.*]] = trunc <2 x i32> [[Vec2]] to <2 x i1>
+// CHECK-NEXT: store <2 x i1> [[Trunc]], ptr [[Vec2Ptr]], align 1
+// CHECK-NEXT: [[V2:%.*]] = load <2 x i32>, ptr [[Vec2Ptr]], align 1
+// CHECK-NEXT: [[V3:%.*]] = shufflevector <2 x i32> [[V2]], <2 x i32> poison, <2 x i32> zeroinitializer
+// CHECK-NEXT: [[LV1:%.*]] = trunc <2 x i32> [[V3]] to <2 x i1>
+// CHECK-NEXT: ret <2 x i1> [[LV1]]
+bool2 HowManyBools(bool V) {
+  return V.rr.rr;
+}
+
 // This codegen is gnarly because `1.l` is a double, so this creates double
 // vectors that need to be truncated down to floats. The optimizer cleans this
 // up nicely too.
@@ -166,3 +205,39 @@ int AssignInt(int V){
   X.x = V.x + V.x;
   return X;
 }
+
+// CHECK-LABEL: AssignBool
+// CHECK: [[VAddr:%.*]] = alloca i32, align 4
+// CHECK-NEXT: [[XAddr:%.*]] = alloca i32, align 4
+// CHECK-NEXT: [[Zext:%.*]] = zext i1 %V to i32
+// CHECK-NEXT: store i32 [[Zext]], ptr [[VAddr]], align 4
+// CHECK-NEXT: [[X:%.*]] = load i32, ptr [[VAddr]], align 4
+// CHECK-NEXT: [[Splat:%.*]] = insertelement <1 x i32> poison, i32 [[X]], i64 0
+// CHECK-NEXT: [[Y:%.*]] = extractelement <1 x i32> [[Splat]], i32 0
+// CHECK-NEXT: [[Z:%.*]] = trunc i32 [[Y]] to i1
+// CHECK-NEXT: [[A:%.*]] = zext i1 [[Z]] to i32
+// CHECK-NEXT: store i32 [[A]], ptr [[XAddr]], align 4
+// CHECK-NEXT: [[B:%.*]] = load i32, ptr [[VAddr]], align 4
+// CHECK-NEXT: [[Splat2:%.*]] = insertelement <1 x i32> poison, i32 [[B]], i64 0
+// CHECK-NEXT: [[C:%.*]] = extractelement <1 x i32> [[Splat2]], i32 0
+// CHECK-NEXT: [[D:%.*]] = trunc i32 [[C]] to i1
+// CHECK-NEXT: br i1 [[D]], label %lor.end, label %lor.rhs
+
+// CHECK: lor.rhs:
+// CHECK-NEXT: [[E:%.*]] = load i32, ptr [[VAddr]], align 4
+// CHECK-NEXT: [[Splat3:%.*]] = insertelement <1 x i32> poison, i32 [[E]], i64 0
+// CHECK-NEXT: [[F:%.*]] = extractelement <1 x i32> [[Splat3]], i32 0
+// CHECK-NEXT: [[G:%.*]] = trunc i32 [[F]] to i1
+// CHECK-NEXT: br label %lor.end
+
+// CHECK: lor.end:
+// CHECK-NEXT: [[H:%.*]] = phi i1 [ true, %entry ], [ [[G]], %lor.rhs ]
+// CHECK-NEXT: store i1 [[H]], ptr [[XAddr]], align 4
+// CHECK-NEXT: [[I:%.*]] = load i32, ptr [[XAddr]], align 4
+// CHECK-NEXT: [[LoadV:%.*]] = trunc i32 [[I]] to i1
+// CHECK-NEXT: ret i1 [[LoadV]]
+bool AssignBool(bool V) {
+  bool X = V.x;
+  X.x = V.x || V.x;
+  return X;
+}

>From 6734eedc466fc6202f7eb84484864d63dd0bb39d Mon Sep 17 00:00:00 2001
From: Sarah Spall <spall at planetbauer.com>
Date: Tue, 21 Jan 2025 16:53:10 +0000
Subject: [PATCH 04/12] vector swizzle boolean vectors

---
 clang/lib/CodeGen/CGExpr.cpp | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
index be8effb4dd40e..331da27764df1 100644
--- a/clang/lib/CodeGen/CGExpr.cpp
+++ b/clang/lib/CodeGen/CGExpr.cpp
@@ -2353,7 +2353,13 @@ RValue CodeGenFunction::EmitLoadOfExtVectorElementLValue(LValue LV) {
   if (!ExprVT) {
     unsigned InIdx = getAccessedFieldNo(0, Elts);
     llvm::Value *Elt = llvm::ConstantInt::get(SizeTy, InIdx);
-    return RValue::get(Builder.CreateExtractElement(Vec, Elt));
+
+    llvm::Value *Element = Builder.CreateExtractElement(Vec, Elt);
+
+    if (getLangOpts().HLSL && LV.getType()->isBooleanType())
+      Element = Builder.CreateTrunc(Element, ConvertType(LV.getType()));
+
+    return RValue::get(Element);
   }
 
   // Always use shuffle vector to try to retain the original program structure
@@ -2364,6 +2370,10 @@ RValue CodeGenFunction::EmitLoadOfExtVectorElementLValue(LValue LV) {
     Mask.push_back(getAccessedFieldNo(i, Elts));
 
   Vec = Builder.CreateShuffleVector(Vec, Mask);
+
+  if (getLangOpts().HLSL && LV.getType()->isExtVectorBoolType())
+    Vec = EmitFromMemory(Vec, LV.getType());
+
   return RValue::get(Vec);
 }
 

>From d9a4777cc755bdd9950e50f3f2bcf0a6f844a4a1 Mon Sep 17 00:00:00 2001
From: Sarah Spall <spall at planetbauer.com>
Date: Tue, 21 Jan 2025 18:17:07 +0000
Subject: [PATCH 05/12] fix test broken by changes

---
 .../standard_conversion_sequences.hlsl             | 14 ++++++--------
 1 file changed, 6 insertions(+), 8 deletions(-)

diff --git a/clang/test/CodeGenHLSL/BasicFeatures/standard_conversion_sequences.hlsl b/clang/test/CodeGenHLSL/BasicFeatures/standard_conversion_sequences.hlsl
index 1665a0260ab05..6770efefe94fe 100644
--- a/clang/test/CodeGenHLSL/BasicFeatures/standard_conversion_sequences.hlsl
+++ b/clang/test/CodeGenHLSL/BasicFeatures/standard_conversion_sequences.hlsl
@@ -91,13 +91,12 @@ void l4_to_i2() {
 
 // CHECK-LABEL: i2_to_b2
 // CHECK: [[l2:%.*]] = alloca <2 x i32>
-// CHECK: [[b2:%.*]] = alloca i8
+// CHECK: [[b2:%.*]] = alloca <2 x i32>
 // CHECK: store <2 x i32> splat (i32 8), ptr [[i2]]
 // CHECK: [[veci2:%.*]] = load <2 x i32>, ptr [[i2]]
 // CHECK: [[vecb2:%.*]] = icmp ne <2 x i32> [[veci2]], zeroinitializer
-// CHECK: [[vecb8:%.*]] = shufflevector <2 x i1> [[vecb2]], <2 x i1> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
-// CHECK: [[i8:%.*]] = bitcast <8 x i1> [[vecb8]] to i8
-// CHECK: store i8 [[i8]], ptr [[b2]]
+// CHECK: [[vecb8:%.*]] = zext <2 x i1> [[vecb2]] to <2 x i32>
+// CHECK: store <2 x i32> [[vecb8]], ptr [[b2]]
 void i2_to_b2() {
   vector<int, 2> i2 = 8;
   vector<bool, 2> b2 = i2;
@@ -105,14 +104,13 @@ void i2_to_b2() {
 
 // CHECK-LABEL: d4_to_b2
 // CHECK: [[d4:%.*]] = alloca <4 x double>
-// CHECK: [[b2:%.*]] = alloca i8
+// CHECK: [[b2:%.*]] = alloca <2 x i32>
 // CHECK: store <4 x double> splat (double 9.000000e+00), ptr [[d4]]
 // CHECK: [[vecd4:%.*]] = load <4 x double>, ptr [[d4]]
 // CHECK: [[vecb4:%.*]] = fcmp reassoc nnan ninf nsz arcp afn une <4 x double> [[vecd4]], zeroinitializer
 // CHECK: [[vecd2:%.*]] = shufflevector <4 x i1> [[vecb4]], <4 x i1> poison, <2 x i32> <i32 0, i32 1>
-// CHECK: [[vecb8:%.*]] = shufflevector <2 x i1> [[vecd2]], <2 x i1> poison, <8 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
-// CHECK: [[i8:%.*]] = bitcast <8 x i1> [[vecb8]] to i8
-// CHECK: store i8 [[i8]], ptr [[b2]]
+// CHECK: [[vecb8:%.*]] = zext <2 x i1> [[vecd2]] to <2 x i32>
+// CHECK: store <2 x i32> [[vecb8]], ptr [[b2]]
 void d4_to_b2() {
   vector<double,4> d4 = 9.0;
   vector<bool, 2> b2 = d4;

>From e0638d10baadcf7793ed307a73e36f9159877c21 Mon Sep 17 00:00:00 2001
From: Sarah Spall <spall at planetbauer.com>
Date: Tue, 21 Jan 2025 18:18:59 +0000
Subject: [PATCH 06/12] actually make clang format happy

---
 clang/lib/CodeGen/CGExpr.cpp       | 17 +++++++++--------
 clang/lib/CodeGen/CodeGenTypes.cpp |  4 ++--
 2 files changed, 11 insertions(+), 10 deletions(-)

diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
index 331da27764df1..906876f1d878b 100644
--- a/clang/lib/CodeGen/CGExpr.cpp
+++ b/clang/lib/CodeGen/CGExpr.cpp
@@ -1987,8 +1987,8 @@ llvm::Value *CodeGenFunction::EmitLoadOfScalar(Address Addr, bool Volatile,
     // Boolean vectors use `iN` as storage type.
     if (ClangVecTy->isExtVectorBoolType()) {
       if (getLangOpts().HLSL) {
-	llvm::Value *Value = Builder.CreateLoad(Addr, Volatile, "load_boolvec");
-	return EmitFromMemory(Value, Ty);
+        llvm::Value *Value = Builder.CreateLoad(Addr, Volatile, "load_boolvec");
+        return EmitFromMemory(Value, Ty);
       }
       llvm::Type *ValTy = ConvertType(Ty);
       unsigned ValNumElems =
@@ -2090,7 +2090,7 @@ llvm::Value *CodeGenFunction::EmitFromMemory(llvm::Value *Value, QualType Ty) {
     const auto *RawIntTy = Value->getType();
     if (getLangOpts().HLSL)
       return Builder.CreateTrunc(Value, ConvertType(Ty), "loadedv");
-	
+
     // Bitcast iP --> <P x i1>.
     auto *PaddedVecTy = llvm::FixedVectorType::get(
         Builder.getInt1Ty(), RawIntTy->getPrimitiveSizeInBits());
@@ -2429,9 +2429,10 @@ void CodeGenFunction::EmitStoreThroughLValue(RValue Src, LValue Dst,
                                             Dst.isVolatileQualified());
       llvm::Type *OldVecTy = Vec->getType();
       if (getLangOpts().HLSL && Dst.getType()->isExtVectorBoolType())
-        
-	Vec = Builder.CreateTrunc(Vec, ConvertType(Dst.getType()), "truncboolv");
-      
+
+        Vec =
+            Builder.CreateTrunc(Vec, ConvertType(Dst.getType()), "truncboolv");
+
       auto *IRStoreTy = dyn_cast<llvm::IntegerType>(Vec->getType());
       if (IRStoreTy) {
         auto *IRVecTy = llvm::FixedVectorType::get(
@@ -2447,8 +2448,8 @@ void CodeGenFunction::EmitStoreThroughLValue(RValue Src, LValue Dst,
       }
 
       if (getLangOpts().HLSL && Dst.getType()->isExtVectorBoolType())
-	Vec = Builder.CreateZExt(Vec, OldVecTy);
-      
+        Vec = Builder.CreateZExt(Vec, OldVecTy);
+
       Builder.CreateStore(Vec, Dst.getVectorAddress(),
                           Dst.isVolatileQualified());
       return;
diff --git a/clang/lib/CodeGen/CodeGenTypes.cpp b/clang/lib/CodeGen/CodeGenTypes.cpp
index e7f280e3823a3..778161c6a5818 100644
--- a/clang/lib/CodeGen/CodeGenTypes.cpp
+++ b/clang/lib/CodeGen/CodeGenTypes.cpp
@@ -112,12 +112,12 @@ llvm::Type *CodeGenTypes::ConvertTypeForMem(QualType T) {
   // Check for the boolean vector case.
   if (T->isExtVectorBoolType()) {
     auto *FixedVT = cast<llvm::FixedVectorType>(R);
-    
+
     if (Context.getLangOpts().HLSL) {
       llvm::Type *IRElemTy = ConvertTypeForMem(Context.BoolTy);
       return llvm::FixedVectorType::get(IRElemTy, FixedVT->getNumElements());
     }
-    
+
     // Pad to at least one byte.
     uint64_t BytePadded = std::max<uint64_t>(FixedVT->getNumElements(), 8);
     return llvm::IntegerType::get(FixedVT->getContext(), BytePadded);

>From 2e534a50c7059080b295a31d6cd8b4ac8072a688 Mon Sep 17 00:00:00 2001
From: Sarah Spall <spall at planetbauer.com>
Date: Wed, 22 Jan 2025 18:05:17 +0000
Subject: [PATCH 07/12] new test

---
 clang/test/CodeGenHLSL/BoolVector.hlsl | 32 ++++++++++++++++++++++++++
 1 file changed, 32 insertions(+)
 create mode 100644 clang/test/CodeGenHLSL/BoolVector.hlsl

diff --git a/clang/test/CodeGenHLSL/BoolVector.hlsl b/clang/test/CodeGenHLSL/BoolVector.hlsl
new file mode 100644
index 0000000000000..16aa6c0e56cee
--- /dev/null
+++ b/clang/test/CodeGenHLSL/BoolVector.hlsl
@@ -0,0 +1,32 @@
+// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.3-library -emit-llvm -disable-llvm-passes -o - %s | FileCheck %s
+
+// CHECK-LABEL: fn
+// CHECK: [[B:%.*]] = alloca <2 x i32>, align 1
+// CHECK-NEXT: store <2 x i32> splat (i32 1), ptr [[B]], align 1
+// CHECK-NEXT: [[BoolVec:%.*]] = load <2 x i32>, ptr [[B]], align 1
+// CHECK-NEXT: [[L:%.*]] = trunc <2 x i32> [[BoolVec:%.*]] to <2 x i1>
+// CHECK-NEXT: [[VecExt:%.*]] = extractelement <2 x i1> [[L]], i32 0
+// CHECK-NEXT: ret i1 [[VecExt]]
+bool fn() {
+  bool2 B = {true,true};
+  return B[0];
+}
+
+// CHECK-LABEL: fn2
+// CHECK: [[VAddr:%.*]] = alloca i32, align 4
+// CHECK-NEXT: [[A:%.*]] = alloca <2 x i32>, align 1
+// CHECK-NEXT: [[StoreV:%.*]] = zext i1 {{.*}} to i32
+// CHECK-NEXT: store i32 [[StoreV]], ptr [[VAddr]], align 4
+// CHECK-NEXT: [[L:%.*]] = load i32, ptr [[VAddr]], align 4
+// CHECK-NEXT: [[LoadV:%.*]] = trunc i32 [[L]] to i1
+// CHECK-NEXT: [[Vec:%.*]] = insertelement <2 x i1> poison, i1 [[LoadV]], i32 0
+// CHECK-NEXT: [[Vec1:%.*]] = insertelement <2 x i1> [[Vec]], i1 true, i32 1
+// CHECK-NEXT: [[Z:%.*]] = zext <2 x i1> [[Vec1]] to <2 x i32>
+// CHECK-NEXT: store <2 x i32> [[Z]], ptr [[A]], align 1
+// CHECK-NEXT: [[LoadBV:%.*]] = load <2 x i32>, ptr [[A]], align 1
+// CHECK-NEXT: [[LoadV2:%.*]] = trunc <2 x i32> [[LoadBV]] to <2 x i1>
+// CHECK-NEXT: ret <2 x i1> [[LoadV2]]
+bool2 fn2(bool V) {
+  bool2 A = {V,true};
+  return A;
+}

>From 793541d14300cdcedcebf4b18bd6e7373e1a2e70 Mon Sep 17 00:00:00 2001
From: Sarah Spall <spall at planetbauer.com>
Date: Fri, 31 Jan 2025 20:32:49 +0000
Subject: [PATCH 08/12] test with struct that contains bool vector. make sure
 bool vector in struct is a vector of i32. make sure constant struct has
 vector of i32

make clang format happy
---
 clang/lib/AST/ASTContext.cpp           |  5 ++--
 clang/lib/CodeGen/CGExprConstant.cpp   |  9 ++++++
 clang/test/CodeGenHLSL/BoolVector.hlsl | 38 ++++++++++++++++++++------
 3 files changed, 41 insertions(+), 11 deletions(-)

diff --git a/clang/lib/AST/ASTContext.cpp b/clang/lib/AST/ASTContext.cpp
index 155dbcfcaeed3..7aea590189766 100644
--- a/clang/lib/AST/ASTContext.cpp
+++ b/clang/lib/AST/ASTContext.cpp
@@ -2016,8 +2016,9 @@ TypeInfo ASTContext::getTypeInfoImpl(const Type *T) const {
   case Type::Vector: {
     const auto *VT = cast<VectorType>(T);
     TypeInfo EltInfo = getTypeInfo(VT->getElementType());
-    Width = VT->isExtVectorBoolType() ? VT->getNumElements()
-                                      : EltInfo.Width * VT->getNumElements();
+    Width = (VT->isExtVectorBoolType() && !getLangOpts().HLSL)
+                ? VT->getNumElements()
+                : EltInfo.Width * VT->getNumElements();
     // Enforce at least byte size and alignment.
     Width = std::max<unsigned>(8, Width);
     Align = std::max<unsigned>(8, Width);
diff --git a/clang/lib/CodeGen/CGExprConstant.cpp b/clang/lib/CodeGen/CGExprConstant.cpp
index 655fc3dc954c8..afdc8c4a6cb98 100644
--- a/clang/lib/CodeGen/CGExprConstant.cpp
+++ b/clang/lib/CodeGen/CGExprConstant.cpp
@@ -1984,6 +1984,15 @@ llvm::Constant *ConstantEmitter::emitForMemory(CodeGenModule &CGM,
     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);
diff --git a/clang/test/CodeGenHLSL/BoolVector.hlsl b/clang/test/CodeGenHLSL/BoolVector.hlsl
index 16aa6c0e56cee..329f3c7e51b7a 100644
--- a/clang/test/CodeGenHLSL/BoolVector.hlsl
+++ b/clang/test/CodeGenHLSL/BoolVector.hlsl
@@ -1,20 +1,27 @@
 // RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.3-library -emit-llvm -disable-llvm-passes -o - %s | FileCheck %s
 
-// CHECK-LABEL: fn
-// CHECK: [[B:%.*]] = alloca <2 x i32>, align 1
-// CHECK-NEXT: store <2 x i32> splat (i32 1), ptr [[B]], align 1
-// CHECK-NEXT: [[BoolVec:%.*]] = load <2 x i32>, ptr [[B]], align 1
+// CHECK: %struct.S = type { <2 x i32>, float }
+// CHECK: [[ConstS:@.*]] = private unnamed_addr constant %struct.S { <2 x i32> splat (i32 1), float 1.000000e+00 }, align 8
+struct S {
+    bool2 bv;
+    float f;
+};
+
+// CHECK-LABEL: define noundef i1 {{.*}}fn1{{.*}}
+// CHECK: [[B:%.*]] = alloca <2 x i32>, align 8
+// CHECK-NEXT: store <2 x i32> splat (i32 1), ptr [[B]], align 8
+// CHECK-NEXT: [[BoolVec:%.*]] = load <2 x i32>, ptr [[B]], align 8
 // CHECK-NEXT: [[L:%.*]] = trunc <2 x i32> [[BoolVec:%.*]] to <2 x i1>
 // CHECK-NEXT: [[VecExt:%.*]] = extractelement <2 x i1> [[L]], i32 0
 // CHECK-NEXT: ret i1 [[VecExt]]
-bool fn() {
+bool fn1() {
   bool2 B = {true,true};
   return B[0];
 }
 
-// CHECK-LABEL: fn2
+// CHECK-LABEL: define noundef <2 x i1> {{.*}}fn2{{.*}}
 // CHECK: [[VAddr:%.*]] = alloca i32, align 4
-// CHECK-NEXT: [[A:%.*]] = alloca <2 x i32>, align 1
+// CHECK-NEXT: [[A:%.*]] = alloca <2 x i32>, align 8
 // CHECK-NEXT: [[StoreV:%.*]] = zext i1 {{.*}} to i32
 // CHECK-NEXT: store i32 [[StoreV]], ptr [[VAddr]], align 4
 // CHECK-NEXT: [[L:%.*]] = load i32, ptr [[VAddr]], align 4
@@ -22,11 +29,24 @@ bool fn() {
 // CHECK-NEXT: [[Vec:%.*]] = insertelement <2 x i1> poison, i1 [[LoadV]], i32 0
 // CHECK-NEXT: [[Vec1:%.*]] = insertelement <2 x i1> [[Vec]], i1 true, i32 1
 // CHECK-NEXT: [[Z:%.*]] = zext <2 x i1> [[Vec1]] to <2 x i32>
-// CHECK-NEXT: store <2 x i32> [[Z]], ptr [[A]], align 1
-// CHECK-NEXT: [[LoadBV:%.*]] = load <2 x i32>, ptr [[A]], align 1
+// CHECK-NEXT: store <2 x i32> [[Z]], ptr [[A]], align 8
+// CHECK-NEXT: [[LoadBV:%.*]] = load <2 x i32>, ptr [[A]], align 8
 // CHECK-NEXT: [[LoadV2:%.*]] = trunc <2 x i32> [[LoadBV]] to <2 x i1>
 // CHECK-NEXT: ret <2 x i1> [[LoadV2]]
 bool2 fn2(bool V) {
   bool2 A = {V,true};
   return A;
 }
+
+// CHECK-LABEL: define noundef i1 {{.*}}fn3{{.*}}
+// CHECK: [[s:%.*]] = alloca %struct.S, align 8
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 8 [[s]], ptr align 8 [[ConstS]], i32 16, i1 false)
+// CHECK-NEXT: [[BV:%.*]] = getelementptr inbounds nuw %struct.S, ptr [[s]], i32 0, i32 0
+// CHECK-NEXT: [[LBV:%.*]] = load <2 x i32>, ptr [[BV]], align 8
+// CHECK-NEXT: [[LV:%.*]] = trunc <2 x i32> [[LBV]] to <2 x i1>
+// CHECK-NEXT: [[VX:%.*]] = extractelement <2 x i1> [[LV]], i32 0
+// CHECK-NEXT: ret i1 [[VX]]
+bool fn3() {
+  S s = {{true,true}, 1.0};
+  return s.bv[0];
+}

>From 1fe0951164a4fbc50b99ee35dd4f37c59a6224db Mon Sep 17 00:00:00 2001
From: Sarah Spall <spall at planetbauer.com>
Date: Fri, 31 Jan 2025 21:18:10 +0000
Subject: [PATCH 09/12] update test

---
 clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl b/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl
index 8e246959df94f..9c64ceb3089da 100644
--- a/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl
@@ -104,9 +104,9 @@ vector<float, 1> FillOneHalfFloat(){
 }
 
 // CHECK-LABEL: FillTrue
-// CHECK: [[Tmp:%.*]] = alloca <1 x i32>, align 1
-// CHECK-NEXT: store <1 x i1> splat (i1 true), ptr [[Tmp]], align 1
-// CHECK-NEXT: [[Vec1:%.*]] = load <1 x i32>, ptr [[Tmp]], align 1
+// CHECK: [[Tmp:%.*]] = alloca <1 x i32>, align 4
+// CHECK-NEXT: store <1 x i1> splat (i1 true), ptr [[Tmp]], align 4
+// CHECK-NEXT: [[Vec1:%.*]] = load <1 x i32>, ptr [[Tmp]], align 4
 // CHECK-NEXT: [[Vec2:%.*]] = shufflevector <1 x i32> [[Vec1]], <1 x i32> poison, <2 x i32> zeroinitializer
 // CHECK-NEXT: [[Vec2Ret:%.*]] = trunc <2 x i32> [[Vec2]] to <2 x i1>
 // CHECK-NEXT: ret <2 x i1> [[Vec2Ret]]
@@ -133,15 +133,15 @@ float2 HowManyFloats(float V) {
 
 // CHECK-LABEL: HowManyBools
 // CHECK: [[VAddr:%.*]] = alloca i32, align 4
-// CHECK-NEXT: [[Vec2Ptr:%.*]] = alloca <2 x i32>, align 1
+// CHECK-NEXT: [[Vec2Ptr:%.*]] = alloca <2 x i32>, align 8
 // CHECK-NEXT: [[Tmp:%.*]] = zext i1 {{.*}} to i32
 // CHECK-NEXT: store i32 [[Tmp]], ptr [[VAddr]], align 4
 // CHECK-NEXT: [[VVal:%.*]] = load i32, ptr [[VAddr]], align 4
 // CHECK-NEXT: [[Splat:%.*]] = insertelement <1 x i32> poison, i32 [[VVal]], i64 0
 // CHECK-NEXT: [[Vec2:%.*]] = shufflevector <1 x i32> [[Splat]], <1 x i32> poison, <2 x i32> zeroinitializer
 // CHECK-NEXT: [[Trunc:%.*]] = trunc <2 x i32> [[Vec2]] to <2 x i1>
-// CHECK-NEXT: store <2 x i1> [[Trunc]], ptr [[Vec2Ptr]], align 1
-// CHECK-NEXT: [[V2:%.*]] = load <2 x i32>, ptr [[Vec2Ptr]], align 1
+// CHECK-NEXT: store <2 x i1> [[Trunc]], ptr [[Vec2Ptr]], align 8
+// CHECK-NEXT: [[V2:%.*]] = load <2 x i32>, ptr [[Vec2Ptr]], align 8
 // CHECK-NEXT: [[V3:%.*]] = shufflevector <2 x i32> [[V2]], <2 x i32> poison, <2 x i32> zeroinitializer
 // CHECK-NEXT: [[LV1:%.*]] = trunc <2 x i32> [[V3]] to <2 x i1>
 // CHECK-NEXT: ret <2 x i1> [[LV1]]

>From 505b17ba61867969bf3df3b47ed7db8a3f0c5d39 Mon Sep 17 00:00:00 2001
From: Sarah Spall <spall at planetbauer.com>
Date: Fri, 31 Jan 2025 23:37:55 +0000
Subject: [PATCH 10/12] test bool vector in array

---
 clang/test/CodeGenHLSL/BoolVector.hlsl | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/clang/test/CodeGenHLSL/BoolVector.hlsl b/clang/test/CodeGenHLSL/BoolVector.hlsl
index 329f3c7e51b7a..ac2638963f70d 100644
--- a/clang/test/CodeGenHLSL/BoolVector.hlsl
+++ b/clang/test/CodeGenHLSL/BoolVector.hlsl
@@ -2,6 +2,8 @@
 
 // CHECK: %struct.S = type { <2 x i32>, float }
 // CHECK: [[ConstS:@.*]] = private unnamed_addr constant %struct.S { <2 x i32> splat (i32 1), float 1.000000e+00 }, align 8
+// CHECK: [[ConstArr:.*]] = private unnamed_addr constant [2 x <2 x i32>] [<2 x i32> splat (i32 1), <2 x i32> zeroinitializer], align 8
+
 struct S {
     bool2 bv;
     float f;
@@ -50,3 +52,16 @@ bool fn3() {
   S s = {{true,true}, 1.0};
   return s.bv[0];
 }
+
+// CHECK-LABEL: define noundef i1 {{.*}}fn4{{.*}}
+// CHECK: [[Arr:%.*]] = alloca [2 x <2 x i32>], align 8
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 8 [[Arr]], ptr align 8 [[ConstArr]], i32 16, i1 false)
+// CHECK-NEXT: [[Idx:%.*]] = getelementptr inbounds [2 x <2 x i32>], ptr [[Arr]], i32 0, i32 0
+// CHECK-NEXT: [[L:%.*]] = load <2 x i32>, ptr [[Idx]], align 8
+// CHECK-NEXT: [[LV:%.*]] = trunc <2 x i32> [[L]] to <2 x i1>
+// CHECK-NEXT: [[VX:%.*]] = extractelement <2 x i1> [[LV]], i32 1
+// CHECK-NEXT: ret i1 [[VX]]
+bool fn4() {
+  bool2 Arr[2] = {{true,true}, {false,false}};
+  return Arr[0][1];
+}

>From 0e6da2bc0ee55edb76710fc87e6cb2ac19433e80 Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahspall at microsoft.com>
Date: Tue, 11 Feb 2025 19:35:40 -0800
Subject: [PATCH 11/12] create isPackedVectorBoolType which checks if the type
 is an extboolvector and not hlsl

---
 clang/include/clang/AST/Type.h                  | 3 +++
 clang/lib/AST/ASTContext.cpp                    | 2 +-
 clang/lib/AST/ByteCode/InterpBuiltinBitCast.cpp | 5 +++--
 clang/lib/AST/ExprConstant.cpp                  | 9 +++++----
 clang/lib/AST/Type.cpp                          | 6 ++++++
 clang/lib/CodeGen/CGDebugInfo.cpp               | 2 +-
 clang/lib/CodeGen/CGExpr.cpp                    | 9 +++------
 clang/lib/CodeGen/CodeGenTypes.cpp              | 2 +-
 clang/lib/Sema/SemaExprMember.cpp               | 2 +-
 9 files changed, 24 insertions(+), 16 deletions(-)

diff --git a/clang/include/clang/AST/Type.h b/clang/include/clang/AST/Type.h
index 3457d524c63aa..12a50357cc1e6 100644
--- a/clang/include/clang/AST/Type.h
+++ b/clang/include/clang/AST/Type.h
@@ -2567,6 +2567,9 @@ class alignas(TypeAlignment) Type : public ExtQualsTypeCommonBase {
   bool isVectorType() const;                    // GCC vector type.
   bool isExtVectorType() const;                 // Extended vector type.
   bool isExtVectorBoolType() const;             // Extended vector type with bool element.
+  // Extended vector type with bool element that is packed. HLSL doesn't pack
+  // its bool vectors.
+  bool isPackedVectorBoolType(const ASTContext &ctx) const;
   bool isSubscriptableVectorType() const;
   bool isMatrixType() const;                    // Matrix type.
   bool isConstantMatrixType() const;            // Constant matrix type.
diff --git a/clang/lib/AST/ASTContext.cpp b/clang/lib/AST/ASTContext.cpp
index 7aea590189766..9abf86e8d6acb 100644
--- a/clang/lib/AST/ASTContext.cpp
+++ b/clang/lib/AST/ASTContext.cpp
@@ -2016,7 +2016,7 @@ TypeInfo ASTContext::getTypeInfoImpl(const Type *T) const {
   case Type::Vector: {
     const auto *VT = cast<VectorType>(T);
     TypeInfo EltInfo = getTypeInfo(VT->getElementType());
-    Width = (VT->isExtVectorBoolType() && !getLangOpts().HLSL)
+    Width = VT->isPackedVectorBoolType(*this)
                 ? VT->getNumElements()
                 : EltInfo.Width * VT->getNumElements();
     // Enforce at least byte size and alignment.
diff --git a/clang/lib/AST/ByteCode/InterpBuiltinBitCast.cpp b/clang/lib/AST/ByteCode/InterpBuiltinBitCast.cpp
index f4c54551a9a60..6b8860c09167c 100644
--- a/clang/lib/AST/ByteCode/InterpBuiltinBitCast.cpp
+++ b/clang/lib/AST/ByteCode/InterpBuiltinBitCast.cpp
@@ -94,7 +94,8 @@ static bool enumerateData(const Pointer &P, const Context &Ctx, Bits Offset,
     Bits ElemSize = Bits(Ctx.getASTContext().getTypeSize(ElemType));
     PrimType ElemT = *Ctx.classify(ElemType);
     // Special case, since the bools here are packed.
-    bool PackedBools = FieldDesc->getType()->isExtVectorBoolType();
+    bool PackedBools =
+        FieldDesc->getType()->isPackedVectorBoolType(Ctx.getASTContext());
     unsigned NumElems = FieldDesc->getNumElems();
     bool Ok = true;
     for (unsigned I = P.getIndex(); I != NumElems; ++I) {
@@ -227,7 +228,7 @@ static bool CheckBitcastType(InterpState &S, CodePtr OpPC, QualType T,
     QualType EltTy = VT->getElementType();
     unsigned NElts = VT->getNumElements();
     unsigned EltSize =
-        VT->isExtVectorBoolType() ? 1 : ASTCtx.getTypeSize(EltTy);
+        VT->isPackedVectorBoolType(ASTCtx) ? 1 : ASTCtx.getTypeSize(EltTy);
 
     if ((NElts * EltSize) % ASTCtx.getCharWidth() != 0) {
       // The vector's size in bits is not a multiple of the target's byte size,
diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
index 2e680d1569f60..0eaee68d35f47 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -7353,7 +7353,7 @@ class APValueToBufferConverter {
     QualType EltTy = VTy->getElementType();
     unsigned NElts = VTy->getNumElements();
 
-    if (VTy->isExtVectorBoolType()) {
+    if (VTy->isPackedVectorBoolType(Info.Ctx)) {
       // Special handling for OpenCL bool vectors:
       // Since these vectors are stored as packed bits, but we can't write
       // individual bits to the BitCastBuffer, we'll buffer all of the elements
@@ -7616,11 +7616,11 @@ class BufferToAPValueConverter {
     QualType EltTy = VTy->getElementType();
     unsigned NElts = VTy->getNumElements();
     unsigned EltSize =
-        VTy->isExtVectorBoolType() ? 1 : Info.Ctx.getTypeSize(EltTy);
+        VTy->isPackedVectorBoolType(Info.Ctx) ? 1 : Info.Ctx.getTypeSize(EltTy);
 
     SmallVector<APValue, 4> Elts;
     Elts.reserve(NElts);
-    if (VTy->isExtVectorBoolType()) {
+    if (VTy->isPackedVectorBoolType(Info.Ctx)) {
       // Special handling for OpenCL bool vectors:
       // Since these vectors are stored as packed bits, but we can't read
       // individual bits from the BitCastBuffer, we'll buffer all of the
@@ -7749,7 +7749,8 @@ static bool checkBitCastConstexprEligibilityType(SourceLocation Loc,
   if (const auto *VTy = Ty->getAs<VectorType>()) {
     QualType EltTy = VTy->getElementType();
     unsigned NElts = VTy->getNumElements();
-    unsigned EltSize = VTy->isExtVectorBoolType() ? 1 : Ctx.getTypeSize(EltTy);
+    unsigned EltSize =
+        VTy->isPackedVectorBoolType(Ctx) ? 1 : Ctx.getTypeSize(EltTy);
 
     if ((NElts * EltSize) % Ctx.getCharWidth() != 0) {
       // The vector's size in bits is not a multiple of the target's byte size,
diff --git a/clang/lib/AST/Type.cpp b/clang/lib/AST/Type.cpp
index caa0ac858a1be..7d7177d802a23 100644
--- a/clang/lib/AST/Type.cpp
+++ b/clang/lib/AST/Type.cpp
@@ -410,6 +410,12 @@ VectorType::VectorType(TypeClass tc, QualType vecType, unsigned nElements,
   VectorTypeBits.NumElements = nElements;
 }
 
+bool Type::isPackedVectorBoolType(const ASTContext &ctx) const {
+  if (ctx.getLangOpts().HLSL)
+    return false;
+  return isExtVectorBoolType();
+}
+
 BitIntType::BitIntType(bool IsUnsigned, unsigned NumBits)
     : Type(BitInt, QualType{}, TypeDependence::None), IsUnsigned(IsUnsigned),
       NumBits(NumBits) {}
diff --git a/clang/lib/CodeGen/CGDebugInfo.cpp b/clang/lib/CodeGen/CGDebugInfo.cpp
index 6cbcaf0384410..6bdc924a6559c 100644
--- a/clang/lib/CodeGen/CGDebugInfo.cpp
+++ b/clang/lib/CodeGen/CGDebugInfo.cpp
@@ -3292,7 +3292,7 @@ llvm::DIType *CGDebugInfo::CreateTypeDefinition(const ObjCInterfaceType *Ty,
 
 llvm::DIType *CGDebugInfo::CreateType(const VectorType *Ty,
                                       llvm::DIFile *Unit) {
-  if (Ty->isExtVectorBoolType()) {
+  if (Ty->isPackedVectorBoolType(CGM.getContext())) {
     // Boolean ext_vector_type(N) are special because their real element type
     // (bits of bit size) is not their Clang element type (_Bool of size byte).
     // For now, we pretend the boolean vector were actually a vector of bytes
diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
index 906876f1d878b..ed17757e33c66 100644
--- a/clang/lib/CodeGen/CGExpr.cpp
+++ b/clang/lib/CodeGen/CGExpr.cpp
@@ -1985,11 +1985,7 @@ llvm::Value *CodeGenFunction::EmitLoadOfScalar(Address Addr, bool Volatile,
 
   if (const auto *ClangVecTy = Ty->getAs<VectorType>()) {
     // Boolean vectors use `iN` as storage type.
-    if (ClangVecTy->isExtVectorBoolType()) {
-      if (getLangOpts().HLSL) {
-        llvm::Value *Value = Builder.CreateLoad(Addr, Volatile, "load_boolvec");
-        return EmitFromMemory(Value, Ty);
-      }
+    if (ClangVecTy->isPackedVectorBoolType(getContext())) {
       llvm::Type *ValTy = ConvertType(Ty);
       unsigned ValNumElems =
           cast<llvm::FixedVectorType>(ValTy)->getNumElements();
@@ -2162,7 +2158,8 @@ void CodeGenFunction::EmitStoreOfScalar(llvm::Value *Value, Address Addr,
     if (auto *VecTy = dyn_cast<llvm::FixedVectorType>(SrcTy)) {
       auto *NewVecTy =
           CGM.getABIInfo().getOptimalVectorMemoryType(VecTy, getLangOpts());
-      if (!ClangVecTy->isExtVectorBoolType() && VecTy != NewVecTy) {
+      if (!ClangVecTy->isPackedVectorBoolType(getContext()) &&
+          VecTy != NewVecTy) {
         SmallVector<int, 16> Mask(NewVecTy->getNumElements(), -1);
         std::iota(Mask.begin(), Mask.begin() + VecTy->getNumElements(), 0);
         Value = Builder.CreateShuffleVector(Value, Mask, "extractVec");
diff --git a/clang/lib/CodeGen/CodeGenTypes.cpp b/clang/lib/CodeGen/CodeGenTypes.cpp
index 778161c6a5818..07d2d15fc08f5 100644
--- a/clang/lib/CodeGen/CodeGenTypes.cpp
+++ b/clang/lib/CodeGen/CodeGenTypes.cpp
@@ -654,7 +654,7 @@ llvm::Type *CodeGenTypes::ConvertType(QualType T) {
   case Type::Vector: {
     const auto *VT = cast<VectorType>(Ty);
     // An ext_vector_type of Bool is really a vector of bits.
-    llvm::Type *IRElemTy = VT->isExtVectorBoolType()
+    llvm::Type *IRElemTy = VT->isPackedVectorBoolType(Context)
                                ? llvm::Type::getInt1Ty(getLLVMContext())
                                : ConvertType(VT->getElementType());
     ResultType = llvm::FixedVectorType::get(IRElemTy, VT->getNumElements());
diff --git a/clang/lib/Sema/SemaExprMember.cpp b/clang/lib/Sema/SemaExprMember.cpp
index dcf7feab752b5..1d9efbeb5ccb5 100644
--- a/clang/lib/Sema/SemaExprMember.cpp
+++ b/clang/lib/Sema/SemaExprMember.cpp
@@ -1697,7 +1697,7 @@ static ExprResult LookupMemberExpr(Sema &S, LookupResult &R,
         QualType(), false);
   }
 
-  if (BaseType->isExtVectorBoolType() && !S.Context.getLangOpts().HLSL) {
+  if (BaseType->isPackedVectorBoolType(S.Context)) {
     // We disallow element access for ext_vector_type bool.  There is no way to
     // materialize a reference to a vector element as a pointer (each element is
     // one bit in the vector).

>From fbaf536b943b3630f39cff2e2e6d78b176ee374c Mon Sep 17 00:00:00 2001
From: Sarah Spall <sarahspall at microsoft.com>
Date: Wed, 12 Feb 2025 19:22:42 -0800
Subject: [PATCH 12/12] fix 2 incorrect tests

---
 clang/lib/CodeGen/CGExpr.cpp                        | 8 ++++++--
 clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl | 5 +++--
 2 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
index ed17757e33c66..be5e086d41a79 100644
--- a/clang/lib/CodeGen/CGExpr.cpp
+++ b/clang/lib/CodeGen/CGExpr.cpp
@@ -4718,9 +4718,13 @@ EmitExtVectorElementExpr(const ExtVectorElementExpr *E) {
 
     // Store the vector to memory (because LValue wants an address).
     Address VecMem = CreateMemTemp(E->getBase()->getType());
+    // To be consistent need to zero extend an hlsl boolean vector to store it back
+    // to memory
+    QualType Ty = E->getBase()->getType();
+    if (Ty->isExtVectorBoolType() && getLangOpts().HLSL)
+      Vec = Builder.CreateZExt(Vec, convertTypeForLoadStore(Ty, Vec->getType()));
     Builder.CreateStore(Vec, VecMem);
-    Base = MakeAddrLValue(VecMem, E->getBase()->getType(),
-                          AlignmentSource::Decl);
+    Base = MakeAddrLValue(VecMem, Ty, AlignmentSource::Decl);
   }
 
   QualType type =
diff --git a/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl b/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl
index 9c64ceb3089da..063669448ed82 100644
--- a/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl
@@ -105,7 +105,7 @@ vector<float, 1> FillOneHalfFloat(){
 
 // CHECK-LABEL: FillTrue
 // CHECK: [[Tmp:%.*]] = alloca <1 x i32>, align 4
-// CHECK-NEXT: store <1 x i1> splat (i1 true), ptr [[Tmp]], align 4
+// CHECK-NEXT: store <1 x i32> splat (i32 1), ptr [[Tmp]], align 4
 // CHECK-NEXT: [[Vec1:%.*]] = load <1 x i32>, ptr [[Tmp]], align 4
 // CHECK-NEXT: [[Vec2:%.*]] = shufflevector <1 x i32> [[Vec1]], <1 x i32> poison, <2 x i32> zeroinitializer
 // CHECK-NEXT: [[Vec2Ret:%.*]] = trunc <2 x i32> [[Vec2]] to <2 x i1>
@@ -140,7 +140,8 @@ float2 HowManyFloats(float V) {
 // CHECK-NEXT: [[Splat:%.*]] = insertelement <1 x i32> poison, i32 [[VVal]], i64 0
 // CHECK-NEXT: [[Vec2:%.*]] = shufflevector <1 x i32> [[Splat]], <1 x i32> poison, <2 x i32> zeroinitializer
 // CHECK-NEXT: [[Trunc:%.*]] = trunc <2 x i32> [[Vec2]] to <2 x i1>
-// CHECK-NEXT: store <2 x i1> [[Trunc]], ptr [[Vec2Ptr]], align 8
+// CHECK-NEXT: [[Ext:%.*]] = zext <2 x i1> [[Trunc]] to <2 x i32>
+// CHECK-NEXT: store <2 x i32> [[Ext]], ptr [[Vec2Ptr]], align 8
 // CHECK-NEXT: [[V2:%.*]] = load <2 x i32>, ptr [[Vec2Ptr]], align 8
 // CHECK-NEXT: [[V3:%.*]] = shufflevector <2 x i32> [[V2]], <2 x i32> poison, <2 x i32> zeroinitializer
 // CHECK-NEXT: [[LV1:%.*]] = trunc <2 x i32> [[V3]] to <2 x i1>



More information about the cfe-commits mailing list