[clang] [HLSL] Support vector swizzles on scalars (PR #67700)
Chris B via cfe-commits
cfe-commits at lists.llvm.org
Fri Oct 27 10:47:27 PDT 2023
https://github.com/llvm-beanz updated https://github.com/llvm/llvm-project/pull/67700
>From 2fa183da3991c0bc4da6163609331d198f4a37af Mon Sep 17 00:00:00 2001
From: Chris Bieneman <chris.bieneman at me.com>
Date: Tue, 26 Sep 2023 19:48:18 -0500
Subject: [PATCH 1/4] [HLSL] Support vector swizzles on scalars
HLSL supports vector swizzles on scalars by implicitly converting the
scalar to a single-element vector. This syntax is a convienent way to
initialize vectors based on filling a scalar value.
There are two parts of this change. The first part in the Lexer splits
numeric constant tokens when a `.x` or `.r` suffix is encountered. This
splitting is a bit hacky but allows the numeric constant to be parsed
separately from the vector element expression. There is an ambiguity
here with the `r` suffix used by fixed point types, however fixed point
types aren't supported in HLSL so this should not cause any exposable
problems (a separate issue has been filed to track validating language
options for HLSL: #67689).
The second part of this change is in Sema::LookupMemberExpr. For HLSL,
if the base type is a scalar, we implicit cast the scalar to a
one-element vector then call back to perform the vector lookup.
Fixes #56658 and #67511
---
clang/lib/Lex/Lexer.cpp | 4 +
clang/lib/Lex/LiteralSupport.cpp | 6 +-
clang/lib/Sema/SemaExprMember.cpp | 10 +++
.../BuiltinVector/ScalarSwizzleErrors.hlsl | 14 ++++
.../Types/BuiltinVector/ScalarSwizzles.hlsl | 78 +++++++++++++++++++
5 files changed, 111 insertions(+), 1 deletion(-)
create mode 100644 clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl
create mode 100644 clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzles.hlsl
diff --git a/clang/lib/Lex/Lexer.cpp b/clang/lib/Lex/Lexer.cpp
index 37c3e4175d4736e..65588191cfd8fdb 100644
--- a/clang/lib/Lex/Lexer.cpp
+++ b/clang/lib/Lex/Lexer.cpp
@@ -1950,6 +1950,10 @@ bool Lexer::LexNumericConstant(Token &Result, const char *CurPtr) {
while (isPreprocessingNumberBody(C)) {
CurPtr = ConsumeChar(CurPtr, Size, Result);
PrevCh = C;
+ if (LangOpts.HLSL && C == '.' && (*CurPtr == 'x' || *CurPtr == 'r')) {
+ CurPtr--;
+ break;
+ }
C = getCharAndSize(CurPtr, Size);
}
diff --git a/clang/lib/Lex/LiteralSupport.cpp b/clang/lib/Lex/LiteralSupport.cpp
index 2de307883b97ce7..0a78638f680511d 100644
--- a/clang/lib/Lex/LiteralSupport.cpp
+++ b/clang/lib/Lex/LiteralSupport.cpp
@@ -930,7 +930,11 @@ NumericLiteralParser::NumericLiteralParser(StringRef TokSpelling,
// and FP constants (specifically, the 'pp-number' regex), and assumes that
// the byte at "*end" is both valid and not part of the regex. Because of
// this, it doesn't have to check for 'overscan' in various places.
- if (isPreprocessingNumberBody(*ThisTokEnd)) {
+ // Note: For HLSL, the end token is allowed to be '.' which would be in the
+ // 'pp-number' regex. This is required to support vector swizzles on numeric
+ // constants (i.e. 1.xx or 1.5f.rrr).
+ if (isPreprocessingNumberBody(*ThisTokEnd) &&
+ !(LangOpts.HLSL && *ThisTokEnd == '.')) {
Diags.Report(TokLoc, diag::err_lexing_numeric);
hadError = true;
return;
diff --git a/clang/lib/Sema/SemaExprMember.cpp b/clang/lib/Sema/SemaExprMember.cpp
index fe92215ae46776f..2de5579df27153d 100644
--- a/clang/lib/Sema/SemaExprMember.cpp
+++ b/clang/lib/Sema/SemaExprMember.cpp
@@ -1687,6 +1687,16 @@ static ExprResult LookupMemberExpr(Sema &S, LookupResult &R,
ObjCImpDecl, HasTemplateArgs, TemplateKWLoc);
}
+ // HLSL supports implicit conversion of scalar types to single element vector
+ // rvalues in member expressions.
+ if (S.getLangOpts().HLSL && BaseType->isScalarType()) {
+ QualType VectorTy = S.Context.getExtVectorType(BaseType, 1);
+ BaseExpr = S.ImpCastExprToType(BaseExpr.get(), VectorTy, CK_VectorSplat,
+ BaseExpr.get()->getValueKind());
+ return LookupMemberExpr(S, R, BaseExpr, IsArrow, OpLoc, SS,
+ ObjCImpDecl, HasTemplateArgs, TemplateKWLoc);
+ }
+
S.Diag(OpLoc, diag::err_typecheck_member_reference_struct_union)
<< BaseType << BaseExpr.get()->getSourceRange() << MemberLoc;
diff --git a/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl b/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl
new file mode 100644
index 000000000000000..56c6ab537261769
--- /dev/null
+++ b/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl
@@ -0,0 +1,14 @@
+// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-library -x hlsl -finclude-default-header -verify %s
+
+
+int2 ToTwoInts(int V) {
+ return V.xy; // expected-error{{vector component access exceeds type 'int __attribute__((ext_vector_type(1)))' (vector of 1 'int' value)}}
+}
+
+float2 ToTwoFloats(float V) {
+ return V.rg; // expected-error{{vector component access exceeds type 'float __attribute__((ext_vector_type(1)))' (vector of 1 'float' value)}}
+}
+
+int4 SomeNonsense(int V) {
+ return V.poop; // expected-error{{illegal vector component name 'p'}}
+}
diff --git a/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzles.hlsl b/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzles.hlsl
new file mode 100644
index 000000000000000..b25d73499fdba0f
--- /dev/null
+++ b/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzles.hlsl
@@ -0,0 +1,78 @@
+// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-library -x hlsl \
+// RUN: -finclude-default-header -ast-dump %s | FileCheck %s
+
+
+// CHECK: ExtVectorElementExpr {{.*}} 'int __attribute__((ext_vector_type(2)))' xx
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'int __attribute__((ext_vector_type(1)))' lvalue <VectorSplat>
+// CHECK-NEXT: DeclRefExpr {{.*}} 'int' lvalue ParmVar {{.*}} 'V' 'int'
+
+int2 ToTwoInts(int V){
+ return V.xx;
+}
+
+// CHECK: ExtVectorElementExpr {{.*}} 'float __attribute__((ext_vector_type(4)))' rrrr
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'float __attribute__((ext_vector_type(1)))' lvalue <VectorSplat>
+// CHECK-NEXT: DeclRefExpr {{.*}} 'float' lvalue ParmVar {{.*}} 'V' 'float'
+
+
+float4 ToThreeFloats(float V){
+ return V.rrrr;
+}
+
+// CHECK: ExtVectorElementExpr {{.*}} 'int __attribute__((ext_vector_type(2)))' xx
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'int __attribute__((ext_vector_type(1)))' <VectorSplat>
+// CHECK-NEXT: IntegerLiteral {{.*}} 'int' 1
+
+int2 FillOne(){
+ return 1.xx;
+}
+
+
+// CHECK: ExtVectorElementExpr {{.*}} 'unsigned int __attribute__((ext_vector_type(3)))' xxx
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'unsigned int __attribute__((ext_vector_type(1)))' <VectorSplat>
+// CHECK-NEXT: IntegerLiteral {{.*}} 'unsigned int' 1
+
+uint3 FillOneUnsigned(){
+ return 1u.xxx;
+}
+
+// CHECK: ExtVectorElementExpr {{.*}} 'unsigned long __attribute__((ext_vector_type(4)))' xxxx
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'unsigned long __attribute__((ext_vector_type(1)))' <VectorSplat>
+// CHECK-NEXT: IntegerLiteral {{.*}} 'unsigned long' 1
+
+vector<uint64_t,4> FillOneUnsignedLong(){
+ return 1ul.xxxx;
+}
+
+// CHECK: ExtVectorElementExpr {{.*}} 'double __attribute__((ext_vector_type(2)))' rr
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'double __attribute__((ext_vector_type(1)))' <VectorSplat>
+// CHECK-NEXT: FloatingLiteral {{.*}} 'double' 2.500000e+00
+
+double2 FillTwoPointFive(){
+ return 2.5.rr;
+}
+
+// CHECK: ExtVectorElementExpr {{.*}} 'double __attribute__((ext_vector_type(3)))' rrr
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'double __attribute__((ext_vector_type(1)))' <VectorSplat>
+// CHECK-NEXT: FloatingLiteral {{.*}} 'double' 5.000000e-01
+
+double3 FillOneHalf(){
+ return .5.rrr;
+}
+
+// CHECK: ExtVectorElementExpr {{.*}} 'float __attribute__((ext_vector_type(4)))' rrrr
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'float __attribute__((ext_vector_type(1)))' <VectorSplat>
+// CHECK-NEXT: FloatingLiteral {{.*}} 'float' 2.500000e+00
+
+float4 FillTwoPointFiveFloat(){
+ return 2.5f.rrrr;
+}
+
+// CHECK: InitListExpr {{.*}} 'vector<float, 1>':'float __attribute__((ext_vector_type(1)))'
+// CHECK-NEXT: ExtVectorElementExpr {{.*}} 'float' r
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'float __attribute__((ext_vector_type(1)))' <VectorSplat>
+// CHECK-NEXT: FloatingLiteral {{.*}} 'float' 5.000000e-01
+
+vector<float, 1> FillOneHalfFloat(){
+ return .5f.r;
+}
>From d43bbc555e15721d8f4805b8df62951f67551c3b Mon Sep 17 00:00:00 2001
From: Chris Bieneman <chris.bieneman at me.com>
Date: Mon, 23 Oct 2023 16:10:28 -0500
Subject: [PATCH 2/4] Updates bassed on review feedback.
../clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl
---
clang/lib/Lex/Lexer.cpp | 2 +-
.../BuiltinVector/ScalarSwizzleErrors.hlsl | 19 ++++++++++++++++++-
2 files changed, 19 insertions(+), 2 deletions(-)
diff --git a/clang/lib/Lex/Lexer.cpp b/clang/lib/Lex/Lexer.cpp
index 65588191cfd8fdb..c501d01f79fd033 100644
--- a/clang/lib/Lex/Lexer.cpp
+++ b/clang/lib/Lex/Lexer.cpp
@@ -1951,7 +1951,7 @@ bool Lexer::LexNumericConstant(Token &Result, const char *CurPtr) {
CurPtr = ConsumeChar(CurPtr, Size, Result);
PrevCh = C;
if (LangOpts.HLSL && C == '.' && (*CurPtr == 'x' || *CurPtr == 'r')) {
- CurPtr--;
+ CurPtr -= Size;
break;
}
C = getCharAndSize(CurPtr, Size);
diff --git a/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl b/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl
index 56c6ab537261769..49111691fba209c 100644
--- a/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl
+++ b/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl
@@ -1,6 +1,5 @@
// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-library -x hlsl -finclude-default-header -verify %s
-
int2 ToTwoInts(int V) {
return V.xy; // expected-error{{vector component access exceeds type 'int __attribute__((ext_vector_type(1)))' (vector of 1 'int' value)}}
}
@@ -12,3 +11,21 @@ float2 ToTwoFloats(float V) {
int4 SomeNonsense(int V) {
return V.poop; // expected-error{{illegal vector component name 'p'}}
}
+
+float2 WhatIsHappening(float V) {
+ return V.; // expected-error{{expected unqualified-id}}
+}
+
+// These cases produce no error.
+
+float2 HowManyFloats(float V) {
+ return V.rr.rr;
+}
+
+int64_t4 HooBoy() {
+ return 4l.xxxx;
+}
+
+float3 err() {
+ return 1..rrr;
+}
>From abc17120e38f49fa40fd573cbe11374f9324b80d Mon Sep 17 00:00:00 2001
From: Chris Bieneman <chris.bieneman at me.com>
Date: Tue, 24 Oct 2023 12:19:35 -0500
Subject: [PATCH 3/4] [NFC] Fix clang-format violation
---
clang/lib/Sema/SemaExprMember.cpp | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/clang/lib/Sema/SemaExprMember.cpp b/clang/lib/Sema/SemaExprMember.cpp
index 2de5579df27153d..96db63739f9762a 100644
--- a/clang/lib/Sema/SemaExprMember.cpp
+++ b/clang/lib/Sema/SemaExprMember.cpp
@@ -1693,8 +1693,8 @@ static ExprResult LookupMemberExpr(Sema &S, LookupResult &R,
QualType VectorTy = S.Context.getExtVectorType(BaseType, 1);
BaseExpr = S.ImpCastExprToType(BaseExpr.get(), VectorTy, CK_VectorSplat,
BaseExpr.get()->getValueKind());
- return LookupMemberExpr(S, R, BaseExpr, IsArrow, OpLoc, SS,
- ObjCImpDecl, HasTemplateArgs, TemplateKWLoc);
+ return LookupMemberExpr(S, R, BaseExpr, IsArrow, OpLoc, SS, ObjCImpDecl,
+ HasTemplateArgs, TemplateKWLoc);
}
S.Diag(OpLoc, diag::err_typecheck_member_reference_struct_union)
>From cf357c63a4d230bb530b16a6ecdd6bf0c271b6af Mon Sep 17 00:00:00 2001
From: Chris Bieneman <chris.bieneman at me.com>
Date: Fri, 27 Oct 2023 12:45:08 -0500
Subject: [PATCH 4/4] Codegen support
This change adds CodeGen support for writing to scalar l-values through
ext vector component l-value expressions. This is something only HLSL is
crazy enough to support.
../clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl
../clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl
../clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzles.hlsl
---
clang/lib/CodeGen/CGExpr.cpp | 33 +++-
.../CodeGenHLSL/builtins/ScalarSwizzles.hlsl | 174 ++++++++++++++++++
.../BuiltinVector/ScalarSwizzleErrors.hlsl | 2 +-
.../Types/BuiltinVector/ScalarSwizzles.hlsl | 58 +++++-
4 files changed, 261 insertions(+), 6 deletions(-)
create mode 100644 clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl
diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
index 3123e278985f210..bd008596ed0b319 100644
--- a/clang/lib/CodeGen/CGExpr.cpp
+++ b/clang/lib/CodeGen/CGExpr.cpp
@@ -2035,6 +2035,14 @@ RValue CodeGenFunction::EmitLoadOfExtVectorElementLValue(LValue LV) {
llvm::Value *Vec = Builder.CreateLoad(LV.getExtVectorAddress(),
LV.isVolatileQualified());
+ // HLSL allows treating scalars as one-element vectors. Converting the scalar
+ // IR value to a vector here allows the rest of codegen to behave as normal.
+ if (getLangOpts().HLSL && !Vec->getType()->isVectorTy()) {
+ llvm::Type *DstTy = llvm::FixedVectorType::get(Vec->getType(), 1);
+ llvm::Value *Zero = llvm::Constant::getNullValue(CGM.Int64Ty);
+ Vec = Builder.CreateInsertElement(DstTy, Vec, Zero, "cast.splat");
+ }
+
const llvm::Constant *Elts = LV.getExtVectorElts();
// If the result of the expression is a non-vector type, we must be extracting
@@ -2304,10 +2312,20 @@ void CodeGenFunction::EmitStoreThroughBitfieldLValue(RValue Src, LValue Dst,
void CodeGenFunction::EmitStoreThroughExtVectorComponentLValue(RValue Src,
LValue Dst) {
+ // HLSL allows storing to scalar values through ExtVector component LValues.
+ // To support this we need to handle the case where the destination address is
+ // a scalar.
+ Address DstAddr = Dst.getExtVectorAddress();
+ if (!DstAddr.getElementType()->isVectorTy()) {
+ assert(!Dst.getType()->isVectorType() &&
+ "this should only occur for non-vector l-values");
+ Builder.CreateStore(Src.getScalarVal(), DstAddr, Dst.isVolatileQualified());
+ return;
+ }
+
// This access turns into a read/modify/write of the vector. Load the input
// value now.
- llvm::Value *Vec = Builder.CreateLoad(Dst.getExtVectorAddress(),
- Dst.isVolatileQualified());
+ llvm::Value *Vec = Builder.CreateLoad(DstAddr, Dst.isVolatileQualified());
const llvm::Constant *Elts = Dst.getExtVectorElts();
llvm::Value *SrcVal = Src.getScalarVal();
@@ -2355,7 +2373,8 @@ void CodeGenFunction::EmitStoreThroughExtVectorComponentLValue(RValue Src,
llvm_unreachable("unexpected shorten vector length");
}
} else {
- // If the Src is a scalar (not a vector) it must be updating one element.
+ // If the Src is a scalar (not a vector), and the target is a vector it must
+ // be updating one element.
unsigned InIdx = getAccessedFieldNo(0, Elts);
llvm::Value *Elt = llvm::ConstantInt::get(SizeTy, InIdx);
Vec = Builder.CreateInsertElement(Vec, SrcVal, Elt);
@@ -4734,7 +4753,6 @@ LValue CodeGenFunction::EmitCastLValue(const CastExpr *E) {
case CK_IntegralToPointer:
case CK_PointerToIntegral:
case CK_PointerToBoolean:
- case CK_VectorSplat:
case CK_IntegralCast:
case CK_BooleanToSignedIntegral:
case CK_IntegralToBoolean:
@@ -4899,6 +4917,13 @@ LValue CodeGenFunction::EmitCastLValue(const CastExpr *E) {
}
case CK_ZeroToOCLOpaqueType:
llvm_unreachable("NULL to OpenCL opaque type lvalue cast is not valid");
+
+ case CK_VectorSplat: {
+ // LValue results of vector splats are only supported in HLSL.
+ if (!getLangOpts().HLSL)
+ return EmitUnsupportedLValue(E, "unexpected cast lvalue");
+ return EmitLValue(E->getSubExpr());
+ }
}
llvm_unreachable("Unhandled lvalue cast kind?");
diff --git a/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl b/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl
new file mode 100644
index 000000000000000..2e169ef60c70c4b
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/ScalarSwizzles.hlsl
@@ -0,0 +1,174 @@
+// RUN: %clang_cc1 -std=hlsl2021 -finclude-default-header -x hlsl -triple \
+// RUN: dxil-pc-shadermodel6.3-library %s -emit-llvm -disable-llvm-passes \
+// RUN: -o - | FileCheck %s
+
+// CHECK-LABEL: ToTwoInts
+// CHECK: [[splat:%.*]] = insertelement <1 x i32> poison, i32 {{.*}}, i64 0
+// CHECK: [[vec2:%.*]] = shufflevector <1 x i32> [[splat]], <1 x i32> poison, <2 x i32> zeroinitializer
+// CHECK: ret <2 x i32> [[vec2]]
+int2 ToTwoInts(int V){
+ return V.xx;
+}
+
+// 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
+// ret <4 x float> [[vec4]]
+float4 ToFourFloats(float V){
+ return V.rrrr;
+}
+
+// CHECK-LABEL: FillOne
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x i32>, align 4
+// CHECK: store <1 x i32> <i32 1>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec1:%.*]] = load <1 x i32>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec2:%.*]] = shufflevector <1 x i32> [[vec1]], <1 x i32> poison, <2 x i32> zeroinitializer
+// CHECK: ret <2 x i32> [[vec2]]
+int2 FillOne(){
+ return 1.xx;
+}
+
+// CHECK-LABEL: FillOneUnsigned
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x i32>, align 4
+// CHECK: store <1 x i32> <i32 1>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec1:%.*]] = load <1 x i32>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec3:%.*]] = shufflevector <1 x i32> [[vec1]], <1 x i32> poison, <3 x i32> zeroinitializer
+// CHECK: ret <3 x i32> [[vec3]]
+uint3 FillOneUnsigned(){
+ return 1u.xxx;
+}
+
+// CHECK-LABEL: FillOneUnsignedLong
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x i64>, align 8
+// CHECK: store <1 x i64> <i64 1>, ptr [[vec1Ptr]], align 8
+// CHECK: [[vec1:%.*]] = load <1 x i64>, ptr [[vec1Ptr]], align 8
+// CHECK: [[vec4:%.*]] = shufflevector <1 x i64> [[vec1]], <1 x i64> poison, <4 x i32> zeroinitializer
+// CHECK: ret <4 x i64> [[vec4]]
+vector<uint64_t,4> FillOneUnsignedLong(){
+ return 1ul.xxxx;
+}
+
+// CHECK-LABEL: FillTwoPointFive
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x double>, align 8
+// CHECK: store <1 x double> <double 2.500000e+00>, ptr [[vec1Ptr]], align 8
+// CHECK: [[vec1:%.*]] = load <1 x double>, ptr [[vec1Ptr]], align 8
+// CHECK: [[vec2:%.*]] = shufflevector <1 x double> [[vec1]], <1 x double> poison, <2 x i32> zeroinitializer
+// CHECK: ret <2 x double> [[vec2]]
+double2 FillTwoPointFive(){
+ return 2.5.rr;
+}
+
+// CHECK-LABEL: FillOneHalf
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x double>, align 8
+// CHECK: store <1 x double> <double 5.000000e-01>, ptr [[vec1Ptr]], align 8
+// CHECK: [[vec1:%.*]] = load <1 x double>, ptr [[vec1Ptr]], align 8
+// CHECK: [[vec3:%.*]] = shufflevector <1 x double> [[vec1]], <1 x double> poison, <3 x i32> zeroinitializer
+// CHECK: ret <3 x double> [[vec3]]
+double3 FillOneHalf(){
+ return .5.rrr;
+}
+
+// CHECK-LABEL: FillTwoPointFiveFloat
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x float>, align 4
+// CHECK: store <1 x float> <float 2.500000e+00>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec1:%.*]] = load <1 x float>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec4:%.*]] = shufflevector <1 x float> [[vec1]], <1 x float> poison, <4 x i32> zeroinitializer
+// CHECK: ret <4 x float> [[vec4]]
+float4 FillTwoPointFiveFloat(){
+ return 2.5f.rrrr;
+}
+
+// The initial codegen for this case is correct but a bit odd. The IR optimizer
+// cleans this up very nicely.
+
+// CHECK-LABEL: FillOneHalfFloat
+// CHECK: [[vec1Ptr:%.*]] = alloca <1 x float>, align 4
+// CHECK: store <1 x float> <float 5.000000e-01>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec1:%.*]] = load <1 x float>, ptr [[vec1Ptr]], align 4
+// CHECK: [[vec1Ret:%.*]] = shufflevector <1 x float> [[vec1]], <1 x float> undef, <1 x i32> zeroinitializer
+// CHECK: ret <1 x float> [[vec1Ret]]
+vector<float, 1> FillOneHalfFloat(){
+ return .5f.r;
+}
+
+// The initial codegen for this case is correct but a bit odd. The IR optimizer
+// cleans this up very nicely.
+
+// CHECK-LABEL: HowManyFloats
+// CHECK: [[VAddr:%.*]] = alloca float, align 4
+// CHECK: [[vec2Ptr:%.*]] = alloca <2 x float>, align 8
+// CHECK: [[VVal:%.*]] = load float, ptr [[VAddr]], align 4
+// CHECK: [[splat:%.*]] = insertelement <1 x float> poison, float [[VVal]], i64 0
+// CHECK: [[vec2:%.*]] = shufflevector <1 x float> [[splat]], <1 x float> poison, <2 x i32> zeroinitializer
+// CHECK: store <2 x float> [[vec2]], ptr [[vec2Ptr]], align 8
+// CHECK: [[vec2:%.*]] = load <2 x float>, ptr [[vec2Ptr]], align 8
+// CHECK: [[vec2Res:%.*]] = shufflevector <2 x float> [[vec2]], <2 x float> poison, <2 x i32> zeroinitializer
+// CHECK: ret <2 x float> [[vec2Res]]
+float2 HowManyFloats(float V) {
+ return V.rr.rr;
+}
+
+// This codegen is gnarly because `1.` is a double, so this creates double
+// vectors that need to be truncated down to floats. The optimizer cleans this
+// up nicely too.
+
+// CHECK-LABEL: AllRighty
+// CHECK: [[XTmp:%.*]] = alloca <1 x double>, align 8
+// CHECK: [[YTmp:%.*]] = alloca <1 x double>, align 8
+// CHECK: [[ZTmp:%.*]] = alloca <1 x double>, align 8
+
+// CHECK: store <1 x double> <double 1.000000e+00>, ptr [[XTmp]], align 8
+// CHECK: [[XVec:%.*]] = load <1 x double>, ptr [[XTmp]], align 8
+// CHECK: [[XVec3:%.*]] = shufflevector <1 x double> [[XVec]], <1 x double> poison, <3 x i32> zeroinitializer
+// CHECK: [[XVal:%.*]] = extractelement <3 x double> [[XVec3]], i32 0
+// CHECK: [[XValF:%.*]] = fptrunc double [[XVal]] to float
+// CHECK: [[Vec3F1:%.*]] = insertelement <3 x float> undef, float [[XValF]], i32 0
+
+// CHECK: store <1 x double> <double 1.000000e+00>, ptr [[YTmp]], align 8
+// CHECK: [[YVec:%.*]] = load <1 x double>, ptr [[YTmp]], align 8
+// CHECK: [[YVec3:%.*]] = shufflevector <1 x double> [[YVec]], <1 x double> poison, <3 x i32> zeroinitializer
+// CHECK: [[YVal:%.*]] = extractelement <3 x double> [[YVec3]], i32 1
+// CHECK: [[YValF:%.*]] = fptrunc double [[YVal]] to float
+// CHECK: [[Vec3F2:%.*]] = insertelement <3 x float> [[Vec3F1]], float [[YValF]], i32 1
+
+// CHECK: store <1 x double> <double 1.000000e+00>, ptr [[ZTmp]], align 8
+// CHECK: [[ZVec:%.*]] = load <1 x double>, ptr [[ZTmp]], align 8
+// CHECK: [[ZVec3:%.*]] = shufflevector <1 x double> [[ZVec]], <1 x double> poison, <3 x i32> zeroinitializer
+// CHECK: [[ZVal:%.*]] = extractelement <3 x double> [[ZVec3]], i32 2
+// CHECK: [[ZValF:%.*]] = fptrunc double [[ZVal]] to float
+// CHECK: [[Vec3F3:%.*]] = insertelement <3 x float> [[Vec3F2]], float [[ZValF]], i32 2
+
+// ret <3 x float> [[Vec3F3]]
+float3 AllRighty() {
+ return 1..rrr;
+}
+
+// CHECK-LABEL: AssignInt
+// CHECK: [[VAddr:%.*]] = alloca i32, align 4
+// CHECK: [[XAddr:%.*]] = alloca i32, align 4
+
+// Load V into a vector, then extract V out and store it to X.
+// CHECK: [[V:%.*]] = load i32, ptr [[VAddr]], align 4
+// CHECK: [[Splat:%.*]] = insertelement <1 x i32> poison, i32 [[V]], i64 0
+// CHECK: [[VExtVal:%.*]] = extractelement <1 x i32> [[Splat]], i32 0
+// CHECK: store i32 [[VExtVal]], ptr [[XAddr]], align 4
+
+// Load V into two separate vectors, then add the extracted X components.
+// CHECK: [[V:%.*]] = load i32, ptr [[VAddr]], align 4
+// CHECK: [[Splat:%.*]] = insertelement <1 x i32> poison, i32 [[V]], i64 0
+// CHECK: [[LHS:%.*]] = extractelement <1 x i32> [[Splat]], i32 0
+
+// CHECK: [[V:%.*]] = load i32, ptr [[VAddr]], align 4
+// CHECK: [[Splat:%.*]] = insertelement <1 x i32> poison, i32 [[V]], i64 0
+// CHECK: [[RHS:%.*]] = extractelement <1 x i32> [[Splat]], i32 0
+
+// CHECK: [[Sum:%.*]] = add nsw i32 [[LHS]], [[RHS]]
+// CHECK: store i32 [[Sum]], ptr [[XAddr]], align 4
+// CHECK: [[X:%.*]] = load i32, ptr [[XAddr]], align 4
+// CHECK: ret i32 [[X]]
+
+int AssignInt(int V){
+ int X = V.x;
+ X.x = V.x + V.x;
+ return X;
+}
diff --git a/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl b/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl
index 49111691fba209c..170be6a13c36600 100644
--- a/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl
+++ b/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzleErrors.hlsl
@@ -26,6 +26,6 @@ int64_t4 HooBoy() {
return 4l.xxxx;
}
-float3 err() {
+float3 AllRighty() {
return 1..rrr;
}
diff --git a/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzles.hlsl b/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzles.hlsl
index b25d73499fdba0f..57d55dcc9de0b7f 100644
--- a/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzles.hlsl
+++ b/clang/test/SemaHLSL/Types/BuiltinVector/ScalarSwizzles.hlsl
@@ -15,7 +15,7 @@ int2 ToTwoInts(int V){
// CHECK-NEXT: DeclRefExpr {{.*}} 'float' lvalue ParmVar {{.*}} 'V' 'float'
-float4 ToThreeFloats(float V){
+float4 ToFourFloats(float V){
return V.rrrr;
}
@@ -68,6 +68,10 @@ float4 FillTwoPointFiveFloat(){
return 2.5f.rrrr;
}
+// Because a signle-element accessor returns the element type rather than a
+// truncated vector, this AST formulation has an initialization list to
+// initialze the returned vector.
+
// CHECK: InitListExpr {{.*}} 'vector<float, 1>':'float __attribute__((ext_vector_type(1)))'
// CHECK-NEXT: ExtVectorElementExpr {{.*}} 'float' r
// CHECK-NEXT: ImplicitCastExpr {{.*}} 'float __attribute__((ext_vector_type(1)))' <VectorSplat>
@@ -76,3 +80,55 @@ float4 FillTwoPointFiveFloat(){
vector<float, 1> FillOneHalfFloat(){
return .5f.r;
}
+
+// CHECK: ExtVectorElementExpr {{.*}} 'float __attribute__((ext_vector_type(2)))' rr
+// CHECK-NEXT: ExtVectorElementExpr {{.*}} 'float __attribute__((ext_vector_type(2)))' rr
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'float __attribute__((ext_vector_type(1)))' lvalue <VectorSplat>
+// CHECK-NEXT: DeclRefExpr {{.*}} 'float' lvalue ParmVar {{.*}} 'V' 'float'
+
+float2 HowManyFloats(float V) {
+ return V.rr.rr;
+}
+
+// CHECK: ExtVectorElementExpr {{.*}} 'long __attribute__((ext_vector_type(4)))' xxxx
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'long __attribute__((ext_vector_type(1)))' <VectorSplat>
+// CHECK-NEXT: IntegerLiteral {{.*}} 'long' 4
+
+int64_t4 HooBoy() {
+ return 4l.xxxx;
+}
+
+// This one gets a pretty wierd AST because in addition to the vector splat it
+// is a double->float conversion, which results in generating an initializtion
+// list with float truncation casts.
+
+
+// CHECK: InitListExpr {{.*}} 'float3':'float __attribute__((ext_vector_type(3)))'
+
+// Vector element 0:
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'float':'float' <FloatingCast>
+// CHECK-NEXT: ArraySubscriptExpr {{.*}} 'double'
+// CHECK-NEXT: ExtVectorElementExpr {{.*}} 'double __attribute__((ext_vector_type(3)))' rrr
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'double __attribute__((ext_vector_type(1)))' <VectorSplat>
+// CHECK-NEXT: FloatingLiteral {{.*}} 'double' 1.000000e+00
+// CHECK-NEXT: IntegerLiteral {{.*}} 'int' 0
+
+// Vector element 1:
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'float':'float' <FloatingCast>
+// CHECK-NEXT: ArraySubscriptExpr {{.*}} 'double'
+// CHECK-NEXT: ExtVectorElementExpr {{.*}} 'double __attribute__((ext_vector_type(3)))' rrr
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'double __attribute__((ext_vector_type(1)))' <VectorSplat>
+// CHECK-NEXT: FloatingLiteral {{.*}} 'double' 1.000000e+00
+// CHECK-NEXT: IntegerLiteral {{.*}} 'int' 1
+
+// Vector element 2:
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'float':'float' <FloatingCast>
+// CHECK-NEXT: ArraySubscriptExpr {{.*}} 'double'
+// CHECK-NEXT: ExtVectorElementExpr {{.*}} 'double __attribute__((ext_vector_type(3)))' rrr
+// CHECK-NEXT: ImplicitCastExpr {{.*}} 'double __attribute__((ext_vector_type(1)))' <VectorSplat>
+// CHECK-NEXT: FloatingLiteral {{.*}} 'double' 1.000000e+00
+// CHECK-NEXT: IntegerLiteral {{.*}} 'int' 2
+
+float3 AllRighty() {
+ return 1..rrr;
+}
More information about the cfe-commits
mailing list