https://github.com/s-perron updated
https://github.com/llvm/llvm-project/pull/156075
>From aaecbc5efd38190e8aedd9b3e3c95b2ede791a09 Mon Sep 17 00:00:00 2001
From: Steven Perron <stevenper...@google.com>
Date: Fri, 29 Aug 2025 13:40:55 -0400
Subject: [PATCH] [HLSL] Add copy assignment and construtor to resource types
The wrapper used to hold the handle for resource type has just the
default copy constructor and assignment operator. This causes clang to
insert memcpys when it does an assignment of a resource type. The
memcpy then cause optimizations to fail when the memcpy is turned into a
load and store of an i64.
To fix this, we should define copying of a resource type by adding the
operator= and copy constructor.
This exposed an issue in the hanlding of OpaqueValueExpr. If there were
OpaqueValueExpr in the sub tree of an OpaqueValueExpr, there were not
getting evaluated correctly, where causing an assert. This fixs that,
creating an exception for HLSLOutArgExpr nodes. Their OpaqueValueExpr
need to be handled different, and are already handled correctly.
Partially fixes #154669
---
clang/lib/CodeGen/CGHLSLRuntime.cpp | 20 ++-
clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp | 69 ++++++++++
clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h | 2 +
clang/lib/Sema/HLSLExternalSemaSource.cpp | 2 +
.../CodeGenHLSL/builtins/hlsl_resource_t.hlsl | 12 +-
.../debug/rwbuffer_debug_info.hlsl | 4 +-
.../implicit-norecurse-attrib.hlsl | 2 +-
.../resources/res-array-local-multi-dim.hlsl | 114 +++++++++-------
.../resources/res-array-local1.hlsl | 128 +++++++++---------
.../resources/res-array-local3.hlsl | 124 ++++++++---------
10 files changed, 291 insertions(+), 186 deletions(-)
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.cpp
b/clang/lib/CodeGen/CGHLSLRuntime.cpp
index f32d01ae78658..99a654aec29e5 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.cpp
+++ b/clang/lib/CodeGen/CGHLSLRuntime.cpp
@@ -762,11 +762,27 @@ llvm::Instruction
*CGHLSLRuntime::getConvergenceToken(BasicBlock &BB) {
class OpaqueValueVisitor : public RecursiveASTVisitor<OpaqueValueVisitor> {
public:
- llvm::SmallPtrSet<OpaqueValueExpr *, 8> OVEs;
+ llvm::SmallVector<OpaqueValueExpr *, 8> OVEs;
+ llvm::SmallPtrSet<OpaqueValueExpr *, 8> Visited;
OpaqueValueVisitor() {}
+ bool VisitHLSLOutArgExpr(HLSLOutArgExpr *) {
+ // These need to be bound in CodeGenFunction::EmitHLSLOutArgLValues
+ // or CodeGenFunction::EmitHLSLOutArgExpr. If they are part of this
+ // traversal, the temporary containing the copy out will not have
+ // been created yet.
+ return false;
+ }
+
bool VisitOpaqueValueExpr(OpaqueValueExpr *E) {
- OVEs.insert(E);
+ // Traverse the source expression first.
+ if (E->getSourceExpr())
+ TraverseStmt(E->getSourceExpr());
+
+ // Then add this OVE if we haven't seen it before.
+ if (Visited.insert(E).second)
+ OVEs.push_back(E);
+
return true;
}
};
diff --git a/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp
b/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp
index 806800cb7b213..6c39fe504d3f3 100644
--- a/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp
+++ b/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp
@@ -676,6 +676,75 @@
BuiltinTypeDeclBuilder::addHandleConstructorFromImplicitBinding() {
.finalize();
}
+BuiltinTypeDeclBuilder &BuiltinTypeDeclBuilder::addCopyConstructor() {
+ if (Record->isCompleteDefinition())
+ return *this;
+
+ ASTContext &AST = SemaRef.getASTContext();
+ QualType RecordType = AST.getCanonicalTagType(Record);
+ QualType ConstRecordType = RecordType.withConst();
+ QualType ConstRecordRefType = AST.getLValueReferenceType(ConstRecordType);
+
+ using PH = BuiltinTypeMethodBuilder::PlaceHolder;
+
+ BuiltinTypeMethodBuilder Builder(*this, "", AST.VoidTy, false, true);
+ Builder.addParam("other", ConstRecordRefType);
+ Builder.ensureCompleteDecl();
+
+ ParmVarDecl *OtherParam = Builder.Method->getParamDecl(0);
+
+ Expr *OtherDeclRef = DeclRefExpr::Create(
+ AST, NestedNameSpecifierLoc(), SourceLocation(), OtherParam, false,
+ DeclarationNameInfo(OtherParam->getDeclName(), SourceLocation()),
+ ConstRecordType, VK_LValue);
+
+ FieldDecl *HandleField = getResourceHandleField();
+ Expr *OtherHandleMemberExpr = MemberExpr::CreateImplicit(
+ AST, OtherDeclRef, false, HandleField, HandleField->getType(), VK_LValue,
+ OK_Ordinary);
+
+ return Builder.assign(PH::Handle, OtherHandleMemberExpr).finalize();
+}
+
+BuiltinTypeDeclBuilder &BuiltinTypeDeclBuilder::addCopyAssignmentOperator() {
+ if (Record->isCompleteDefinition())
+ return *this;
+
+ ASTContext &AST = SemaRef.getASTContext();
+ QualType RecordType = AST.getCanonicalTagType(Record);
+ QualType ConstRecordType = RecordType.withConst();
+ QualType ConstRecordRefType = AST.getLValueReferenceType(ConstRecordType);
+ QualType RecordRefType = AST.getLValueReferenceType(RecordType);
+
+ using PH = BuiltinTypeMethodBuilder::PlaceHolder;
+
+ DeclarationName Name = AST.DeclarationNames.getCXXOperatorName(OO_Equal);
+ BuiltinTypeMethodBuilder Builder(*this, Name, RecordRefType, false, false);
+ Builder.addParam("other", ConstRecordRefType);
+ Builder.ensureCompleteDecl();
+
+ ParmVarDecl *OtherParam = Builder.Method->getParamDecl(0);
+ Expr *OtherDeclRef = DeclRefExpr::Create(
+ AST, NestedNameSpecifierLoc(), SourceLocation(), OtherParam, false,
+ DeclarationNameInfo(OtherParam->getDeclName(), SourceLocation()),
+ ConstRecordType, VK_LValue);
+
+ FieldDecl *HandleField = getResourceHandleField();
+ Expr *OtherHandleMemberExpr = MemberExpr::CreateImplicit(
+ AST, OtherDeclRef, false, HandleField, HandleField->getType(), VK_LValue,
+ OK_Ordinary);
+
+ Builder.assign(PH::Handle, OtherHandleMemberExpr);
+
+ // return *this;
+ CXXThisExpr *This = CXXThisExpr::Create(
+ AST, SourceLocation(), Builder.Method->getFunctionObjectParameterType(),
+ true);
+ Builder.StmtsList.push_back(This);
+
+ return Builder.finalize();
+}
+
BuiltinTypeDeclBuilder &BuiltinTypeDeclBuilder::addArraySubscriptOperators() {
ASTContext &AST = Record->getASTContext();
DeclarationName Subscript =
diff --git a/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h
b/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h
index 098b72692bd3a..4c4c2083a8440 100644
--- a/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h
+++ b/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h
@@ -80,6 +80,8 @@ class BuiltinTypeDeclBuilder {
BuiltinTypeDeclBuilder &addDefaultHandleConstructor();
BuiltinTypeDeclBuilder &addHandleConstructorFromBinding();
BuiltinTypeDeclBuilder &addHandleConstructorFromImplicitBinding();
+ BuiltinTypeDeclBuilder &addCopyConstructor();
+ BuiltinTypeDeclBuilder &addCopyAssignmentOperator();
// Builtin types methods
BuiltinTypeDeclBuilder &addLoadMethods();
diff --git a/clang/lib/Sema/HLSLExternalSemaSource.cpp
b/clang/lib/Sema/HLSLExternalSemaSource.cpp
index 726581d131623..8c893c0c30baf 100644
--- a/clang/lib/Sema/HLSLExternalSemaSource.cpp
+++ b/clang/lib/Sema/HLSLExternalSemaSource.cpp
@@ -132,6 +132,8 @@ static BuiltinTypeDeclBuilder setupBufferType(CXXRecordDecl
*Decl, Sema &S,
return BuiltinTypeDeclBuilder(S, Decl)
.addHandleMember(RC, IsROV, RawBuffer)
.addDefaultHandleConstructor()
+ .addCopyConstructor()
+ .addCopyAssignmentOperator()
.addHandleConstructorFromBinding()
.addHandleConstructorFromImplicitBinding();
}
diff --git a/clang/test/CodeGenHLSL/builtins/hlsl_resource_t.hlsl
b/clang/test/CodeGenHLSL/builtins/hlsl_resource_t.hlsl
index 24114b11c7602..1f64037e1f9b4 100644
--- a/clang/test/CodeGenHLSL/builtins/hlsl_resource_t.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/hlsl_resource_t.hlsl
@@ -21,9 +21,9 @@ void fb(handle_float_t a) {
handle_float_t b = a;
}
-// CHECK: define hidden void @_Z2fcN4hlsl8RWBufferIDv4_fEE(ptr noundef
byval(%"class.hlsl::RWBuffer") align 4 %a)
-// CHECK: call void @_Z4foo2N4hlsl8RWBufferIDv4_fEE(ptr noundef
byval(%"class.hlsl::RWBuffer") align 4 %agg.tmp)
-// CHECK: declare hidden void @_Z4foo2N4hlsl8RWBufferIDv4_fEE(ptr noundef
byval(%"class.hlsl::RWBuffer") align 4)
+// CHECK: define hidden void @_Z2fcN4hlsl8RWBufferIDv4_fEE(ptr dead_on_return
noundef %a)
+// CHECK: call void @_Z4foo2N4hlsl8RWBufferIDv4_fEE(ptr dead_on_return noundef
%{{.*}})
+// CHECK: declare hidden void @_Z4foo2N4hlsl8RWBufferIDv4_fEE(ptr
dead_on_return noundef)
void foo2(RWBuffer<float4> buf);
void fc(RWBuffer<float4> a) {
@@ -39,9 +39,9 @@ struct MyStruct {
int2 i;
};
-// CHECK: define hidden void @_Z2feN4hlsl16StructuredBufferI8MyStructEE(ptr
noundef byval(%"class.hlsl::StructuredBuffer") align 4 %a)
-// CHECK: call void @_Z4foo3N4hlsl16StructuredBufferI8MyStructEE(ptr noundef
byval(%"class.hlsl::StructuredBuffer") align 4 %agg.tmp)
-// CHECK: declare hidden void @_Z4foo3N4hlsl16StructuredBufferI8MyStructEE(ptr
noundef byval(%"class.hlsl::StructuredBuffer") align 4)
+// CHECK: define hidden void @_Z2feN4hlsl16StructuredBufferI8MyStructEE(ptr
dead_on_return noundef %a)
+// CHECK: call void @_Z4foo3N4hlsl16StructuredBufferI8MyStructEE(ptr
dead_on_return noundef %{{.*}})
+// CHECK: declare hidden void @_Z4foo3N4hlsl16StructuredBufferI8MyStructEE(ptr
dead_on_return noundef)
void foo3(StructuredBuffer<MyStruct> buf);
void fe(StructuredBuffer<MyStruct> a) {
diff --git a/clang/test/CodeGenHLSL/debug/rwbuffer_debug_info.hlsl
b/clang/test/CodeGenHLSL/debug/rwbuffer_debug_info.hlsl
index db0388e41eae9..76ea87fd534c1 100644
--- a/clang/test/CodeGenHLSL/debug/rwbuffer_debug_info.hlsl
+++ b/clang/test/CodeGenHLSL/debug/rwbuffer_debug_info.hlsl
@@ -1,11 +1,11 @@
// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-compute -x hlsl -emit-llvm
-disable-llvm-passes -o - -hlsl-entry main %s -debug-info-kind=standalone
-dwarf-version=4 | FileCheck %s
-// CHECK: [[DWTag:![0-9]+]] = distinct !DICompositeType(tag:
DW_TAG_class_type, name: "RWBuffer<float>",
+// CHECK: [[DWTag:![0-9]+]] = distinct !DICompositeType(tag:
DW_TAG_class_type, name: "RWBuffer<float>",
+// CHECK: [[thisType:![0-9]+]] = !DIDerivedType(tag: DW_TAG_reference_type,
baseType: [[DWTag]], size: 32)
// CHECK: [[RWBuffer:![0-9]+]] = distinct !DISubprogram(name: "RWBuffer",
// CHECK-SAME: scope: [[DWTag]]
// CHECK: [[FirstThis:![0-9]+]] = !DILocalVariable(name: "this", arg: 1,
scope: [[RWBuffer]], type: [[thisType:![0-9]+]]
-// CHECK: [[thisType]] = !DIDerivedType(tag: DW_TAG_reference_type, baseType:
[[DWTag]], size: 32)
RWBuffer<float> Out : register(u7, space4);
[numthreads(8,1,1)]
diff --git a/clang/test/CodeGenHLSL/implicit-norecurse-attrib.hlsl
b/clang/test/CodeGenHLSL/implicit-norecurse-attrib.hlsl
index 60238cbf8eff5..0c96b73ac6cb1 100644
--- a/clang/test/CodeGenHLSL/implicit-norecurse-attrib.hlsl
+++ b/clang/test/CodeGenHLSL/implicit-norecurse-attrib.hlsl
@@ -31,7 +31,7 @@ uint Find(Node SortedTree[MAX], uint key) {
}
// CHECK: Function Attrs:{{.*}}norecurse
-// CHECK: define noundef i1 @_Z8InitTreeA100_4NodeN4hlsl8RWBufferIDv4_jEEj(ptr
noundef byval([100 x %struct.Node]) align 1 %tree, ptr noundef
byval(%"class.hlsl::RWBuffer") align 4 %encodedTree, i32 noundef %maxDepth)
[[Attr:\#[0-9]+]]
+// CHECK: define noundef i1 @_Z8InitTreeA100_4NodeN4hlsl8RWBufferIDv4_jEEj(ptr
noundef byval([100 x %struct.Node]) align 1 %tree, ptr dead_on_return noundef
%encodedTree, i32 noundef %maxDepth) [[Attr:\#[0-9]+]]
// CHECK: ret i1
// Initialize tree with given buffer
// Imagine the inout works
diff --git a/clang/test/CodeGenHLSL/resources/res-array-local-multi-dim.hlsl
b/clang/test/CodeGenHLSL/resources/res-array-local-multi-dim.hlsl
index d803882fd2f72..d9bd944ba6297 100644
--- a/clang/test/CodeGenHLSL/resources/res-array-local-multi-dim.hlsl
+++ b/clang/test/CodeGenHLSL/resources/res-array-local-multi-dim.hlsl
@@ -1,49 +1,65 @@
-// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-compute
-finclude-default-header \
-// RUN: -emit-llvm -disable-llvm-passes -o - %s | FileCheck %s
-
-// This test verifies handling of multi-dimensional local arrays of resources
-// when used as a function argument and local variable.
-
-// CHECK: @_ZL1A = internal global %"class.hlsl::RWBuffer" poison, align 4
-// CHECK: @_ZL1B = internal global %"class.hlsl::RWBuffer" poison, align 4
-
-RWBuffer<float> A : register(u10);
-RWBuffer<float> B : register(u20);
-RWStructuredBuffer<float> Out;
-
-// NOTE: _ZN4hlsl8RWBufferIfEixEj is the subscript operator for
RWBuffer<float> and
-// _ZN4hlsl18RWStructuredBufferIfEixEj is the subscript operator for
RWStructuredBuffer<float>
-
-// CHECK: define {{.*}} float @_Z3fooA2_A2_N4hlsl8RWBufferIfEE(ptr noundef
byval([2 x [2 x %"class.hlsl::RWBuffer"]]) align 4 %Arr)
-// CHECK-NEXT: entry:
-float foo(RWBuffer<float> Arr[2][2]) {
-// CHECK-NEXT: %[[Arr_1_Ptr:.*]] = getelementptr inbounds [2 x [2 x
%"class.hlsl::RWBuffer"]], ptr %Arr, i32 0, i32 1
-// CHECK-NEXT: %[[Arr_1_1_Ptr:.*]] = getelementptr inbounds [2 x
%"class.hlsl::RWBuffer"], ptr %[[Arr_1_Ptr]], i32 0, i32 1
-// CHECK-NEXT: %[[BufPtr:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIfEixEj(ptr
{{.*}} %[[Arr_1_1_Ptr]], i32 noundef 0)
-// CHECK-NEXT: %[[Value:.*]] = load float, ptr %[[BufPtr]], align 4
-// CHECK-NEXT: ret float %[[Value]]
- return Arr[1][1][0];
-}
-
-// CHECK: define internal void @_Z4mainv()
-// CHECK-NEXT: entry:
-[numthreads(4,1,1)]
-void main() {
-// CHECK-NEXT: %L = alloca [2 x [2 x %"class.hlsl::RWBuffer"]], align 4
-// CHECK-NEXT: %[[Tmp:.*]] = alloca [2 x [2 x %"class.hlsl::RWBuffer"]], align
4
-// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %L, ptr align 4
@_ZL1A, i32 4, i1 false)
-// CHECK-NEXT: %[[Ptr1:.*]] = getelementptr inbounds %"class.hlsl::RWBuffer",
ptr %L, i32 1
-// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Ptr1]], ptr
align 4 @_ZL1B, i32 4, i1 false)
-// CHECK-NEXT: %[[Ptr2:.*]] = getelementptr inbounds [2 x
%"class.hlsl::RWBuffer"], ptr %L, i32 1
-// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Ptr2]], ptr
align 4 @_ZL1A, i32 4, i1 false)
-// CHECK-NEXT: %[[Ptr3:.*]] = getelementptr inbounds %"class.hlsl::RWBuffer",
ptr %[[Ptr2]], i32 1
-// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Ptr3]], ptr
align 4 @_ZL1B, i32 4, i1 false)
- RWBuffer<float> L[2][2] = { { A, B }, { A, B } };
-
-// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Tmp]], ptr
align 4 %L, i32 16, i1 false)
-// CHECK-NEXT: %[[ReturnedValue:.*]] = call {{.*}}float
@_Z3fooA2_A2_N4hlsl8RWBufferIfEE(ptr noundef byval([2 x [2 x
%"class.hlsl::RWBuffer"]]) align 4 %[[Tmp]])
-// CHECK-NEXT: %[[OutBufPtr:.*]] = call {{.*}} ptr
@_ZN4hlsl18RWStructuredBufferIfEixEj(ptr {{.*}} @_ZL3Out, i32 noundef 0)
-// CHECK-NEXT: store float %[[ReturnedValue]], ptr %[[OutBufPtr]], align 4
-// CHECK-NEXT: ret void
- Out[0] = foo(L);
-}
+// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-compute
-finclude-default-header \
+// RUN: -emit-llvm -disable-llvm-passes -o - %s | FileCheck %s
+
+// This test verifies handling of multi-dimensional local arrays of resources
+// when used as a function argument and local variable.
+
+// CHECK: @_ZL1A = internal global %"class.hlsl::RWBuffer" poison, align 4
+// CHECK: @_ZL1B = internal global %"class.hlsl::RWBuffer" poison, align 4
+
+RWBuffer<float> A : register(u10);
+RWBuffer<float> B : register(u20);
+RWStructuredBuffer<float> Out;
+
+// NOTE: _ZN4hlsl8RWBufferIfEixEj is the subscript operator for
RWBuffer<float> and
+// _ZN4hlsl18RWStructuredBufferIfEixEj is the subscript operator for
RWStructuredBuffer<float>
+
+// CHECK: define {{.*}} float @_Z3fooA2_A2_N4hlsl8RWBufferIfEE(ptr noundef
byval([2 x [2 x %"class.hlsl::RWBuffer"]]) align 4 %Arr)
+// CHECK-NEXT: entry:
+float foo(RWBuffer<float> Arr[2][2]) {
+// CHECK-NEXT: %[[Arr_1_Ptr:.*]] = getelementptr inbounds [2 x [2 x
%"class.hlsl::RWBuffer"]], ptr %Arr, i32 0, i32 1
+// CHECK-NEXT: %[[Arr_1_1_Ptr:.*]] = getelementptr inbounds [2 x
%"class.hlsl::RWBuffer"], ptr %[[Arr_1_Ptr]], i32 0, i32 1
+// CHECK-NEXT: %[[BufPtr:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIfEixEj(ptr
{{.*}} %[[Arr_1_1_Ptr]], i32 noundef 0)
+// CHECK-NEXT: %[[Value:.*]] = load float, ptr %[[BufPtr]], align 4
+// CHECK-NEXT: ret float %[[Value]]
+ return Arr[1][1][0];
+}
+
+// CHECK: define internal void @_Z4mainv()
+// CHECK-NEXT: entry:
+[numthreads(4,1,1)]
+void main() {
+// CHECK: %L = alloca [2 x [2 x %"class.hlsl::RWBuffer"]], align 4
+// CHECK: %[[ref_tmp:.*]] = alloca %"class.hlsl::RWBuffer", align 4
+// CHECK: %[[ref_tmp1:.*]] = alloca %"class.hlsl::RWBuffer", align 4
+// CHECK: %[[ref_tmp2:.*]] = alloca %"class.hlsl::RWBuffer", align 4
+// CHECK: %[[ref_tmp3:.*]] = alloca %"class.hlsl::RWBuffer", align 4
+// CHECK: %[[ref_tmp4:.*]] = alloca %"class.hlsl::RWBuffer", align 4
+// CHECK: %[[ref_tmp5:.*]] = alloca %"class.hlsl::RWBuffer", align 4
+// CHECK: %[[ref_tmp6:.*]] = alloca %"class.hlsl::RWBuffer", align 4
+// CHECK: %[[ref_tmp7:.*]] = alloca %"class.hlsl::RWBuffer", align 4
+// CHECK: %[[agg_tmp:.*]] = alloca [2 x [2 x %"class.hlsl::RWBuffer"]], align 4
+// CHECK: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr noundef nonnull align 4
dereferenceable(4) %[[ref_tmp]], ptr noundef nonnull align 4 dereferenceable(4)
@_ZL1A)
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr noundef nonnull
align 4 dereferenceable(4) %[[ref_tmp1]], ptr noundef nonnull align 4
dereferenceable(4) %[[ref_tmp]])
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr noundef nonnull
align 4 dereferenceable(4) %[[ref_tmp2]], ptr noundef nonnull align 4
dereferenceable(4) @_ZL1B)
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr noundef nonnull
align 4 dereferenceable(4) %[[ref_tmp3]], ptr noundef nonnull align 4
dereferenceable(4) %[[ref_tmp2]])
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr noundef nonnull
align 4 dereferenceable(4) %[[ref_tmp4]], ptr noundef nonnull align 4
dereferenceable(4) @_ZL1A)
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr noundef nonnull
align 4 dereferenceable(4) %[[ref_tmp5]], ptr noundef nonnull align 4
dereferenceable(4) %[[ref_tmp4]])
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr noundef nonnull
align 4 dereferenceable(4) %[[ref_tmp6]], ptr noundef nonnull align 4
dereferenceable(4) @_ZL1B)
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr noundef nonnull
align 4 dereferenceable(4) %[[ref_tmp7]], ptr noundef nonnull align 4
dereferenceable(4) %[[ref_tmp6]])
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr noundef nonnull
align 4 dereferenceable(4) %L, ptr noundef nonnull align 4 dereferenceable(4)
%[[ref_tmp1]])
+// CHECK-NEXT: %[[arrayinit_element:.*]] = getelementptr inbounds
%"class.hlsl::RWBuffer", ptr %L, i32 1
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr noundef nonnull
align 4 dereferenceable(4) %[[arrayinit_element]], ptr noundef nonnull align 4
dereferenceable(4) %[[ref_tmp3]])
+// CHECK-NEXT: %[[arrayinit_element8:.*]] = getelementptr inbounds [2 x
%"class.hlsl::RWBuffer"], ptr %L, i32 1
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr noundef nonnull
align 4 dereferenceable(4) %[[arrayinit_element8]], ptr noundef nonnull align 4
dereferenceable(4) %[[ref_tmp5]])
+// CHECK-NEXT: %[[arrayinit_element9:.*]] = getelementptr inbounds
%"class.hlsl::RWBuffer", ptr %[[arrayinit_element8]], i32 1
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr noundef nonnull
align 4 dereferenceable(4) %[[arrayinit_element9]], ptr noundef nonnull align 4
dereferenceable(4) %[[ref_tmp7]])
+ RWBuffer<float> L[2][2] = { { A, B }, { A, B } };
+
+// CHECK: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[agg_tmp]], ptr align
4 %L, i32 16, i1 false)
+// CHECK-NEXT: %[[ReturnedValue:.*]] = call {{.*}}float
@_Z3fooA2_A2_N4hlsl8RWBufferIfEE(ptr noundef byval([2 x [2 x
%"class.hlsl::RWBuffer"]]) align 4 %[[agg_tmp]])
+// CHECK-NEXT: %[[OutBufPtr:.*]] = call {{.*}} ptr
@_ZN4hlsl18RWStructuredBufferIfEixEj(ptr {{.*}} @_ZL3Out, i32 noundef 0)
+// CHECK-NEXT: store float %[[ReturnedValue]], ptr %[[OutBufPtr]], align 4
+// CHECK-NEXT: ret void
+ Out[0] = foo(L);
+}
diff --git a/clang/test/CodeGenHLSL/resources/res-array-local1.hlsl
b/clang/test/CodeGenHLSL/resources/res-array-local1.hlsl
index c0d508b1395c3..9e31f4f150c52 100644
--- a/clang/test/CodeGenHLSL/resources/res-array-local1.hlsl
+++ b/clang/test/CodeGenHLSL/resources/res-array-local1.hlsl
@@ -1,64 +1,64 @@
-// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-compute
-finclude-default-header \
-// RUN: -emit-llvm -disable-llvm-passes -o - %s | FileCheck %s
-
-// This test verifies local arrays of resources in HLSL.
-
-// CHECK: @_ZL1A = internal global %"class.hlsl::RWBuffer" poison, align 4
-// CHECK: @_ZL1B = internal global %"class.hlsl::RWBuffer" poison, align 4
-// CHECK: @_ZL1C = internal global %"class.hlsl::RWBuffer" poison, align 4
-
-RWBuffer<float> A : register(u1);
-RWBuffer<float> B : register(u2);
-RWBuffer<float> C : register(u3);
-RWStructuredBuffer<float> Out : register(u0);
-
-// CHECK: define internal void @_Z4mainv()
-// CHECK-NEXT: entry:
-[numthreads(4,1,1)]
-void main() {
-// CHECK-NEXT: %First = alloca [3 x %"class.hlsl::RWBuffer"], align 4
-// CHECK-NEXT: %Second = alloca [4 x %"class.hlsl::RWBuffer"], align 4
- RWBuffer<float> First[3] = { A, B, C };
- RWBuffer<float> Second[4];
-
-// Verify initialization of First array from an initialization list
-// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %First, ptr align
4 @_ZL1A, i32 4, i1 false)
-// CHECK-NEXT: %[[Ptr1:.*]] = getelementptr inbounds %"class.hlsl::RWBuffer",
ptr %First, i32 1
-// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Ptr1]], ptr
align 4 @_ZL1B, i32 4, i1 false)
-// CHECK-NEXT: %[[Ptr2:.*]] = getelementptr inbounds %"class.hlsl::RWBuffer",
ptr %First, i32 2
-// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Ptr2]], ptr
align 4 @_ZL1C, i32 4, i1 false)
-
-// Verify default initialization of Second array, which means there is a loop
iterating
-// over the array elements and calling the default constructor for each
-// CHECK-NEXT: %[[ArrayBeginPtr:.*]] = getelementptr inbounds [4 x
%"class.hlsl::RWBuffer"], ptr %Second, i32 0, i32 0
-// CHECK-NEXT: %[[ArrayEndPtr:.*]] = getelementptr inbounds
%"class.hlsl::RWBuffer", ptr %[[ArrayBeginPtr]], i32 4
-// CHECK-NEXT: br label %[[ArrayInitLoop:.*]]
-// CHECK: [[ArrayInitLoop]]:
-// CHECK-NEXT: %[[ArrayCurPtr:.*]] = phi ptr [ %[[ArrayBeginPtr]], %entry ], [
%[[ArrayNextPtr:.*]], %[[ArrayInitLoop]] ]
-// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1Ev(ptr {{.*}} %[[ArrayCurPtr]])
-// CHECK-NEXT: %[[ArrayNextPtr]] = getelementptr inbounds
%"class.hlsl::RWBuffer", ptr %[[ArrayCurPtr]], i32 1
-// CHECK-NEXT: %[[ArrayInitDone:.*]] = icmp eq ptr %[[ArrayNextPtr]],
%[[ArrayEndPtr]]
-// CHECK-NEXT: br i1 %[[ArrayInitDone]], label %[[AfterArrayInit:.*]], label
%[[ArrayInitLoop]]
-// CHECK: [[AfterArrayInit]]:
-
-// Initialize First[2] with C
-// CHECK: %[[Ptr3:.*]] = getelementptr inbounds [4 x %"class.hlsl::RWBuffer"],
ptr %Second, i32 0, i32 2
-// CHECK: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Ptr3]], ptr align 4
@_ZL1C, i32 4, i1 false)
- Second[2] = C;
-
- // NOTE: _ZN4hlsl8RWBufferIfEixEj is the subscript operator for
RWBuffer<float>
-
-// get First[1][0] value
-// CHECK: %[[First_1_Ptr:.*]] = getelementptr inbounds [3 x
%"class.hlsl::RWBuffer"], ptr %First, i32 0, i32 1
-// CHECK: %[[BufPtr1:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIfEixEj(ptr
{{.*}} %[[First_1_Ptr]], i32 noundef 0)
-// CHECK: %[[Value1:.*]] = load float, ptr %[[BufPtr1]], align 4
-
-// get Second[2][0] value
-// CHECK: %[[Second_2_Ptr:.*]] = getelementptr inbounds [4 x
%"class.hlsl::RWBuffer"], ptr %Second, i32 0, i32 2
-// CHECK: %[[BufPtr2:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIfEixEj(ptr
{{.*}} %[[Second_2_Ptr]], i32 noundef 0)
-// CHECK: %[[Value2:.*]] = load float, ptr %[[BufPtr2]], align 4
-
-// add them
-// CHECK: %{{.*}} = fadd {{.*}} float %[[Value1]], %[[Value2]]
- Out[0] = First[1][0] + Second[2][0];
-}
+// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-compute
-finclude-default-header \
+// RUN: -emit-llvm -disable-llvm-passes -o - %s | FileCheck %s
+
+// This test verifies local arrays of resources in HLSL.
+
+// CHECK: @_ZL1A = internal global %"class.hlsl::RWBuffer" poison, align 4
+// CHECK: @_ZL1B = internal global %"class.hlsl::RWBuffer" poison, align 4
+// CHECK: @_ZL1C = internal global %"class.hlsl::RWBuffer" poison, align 4
+
+RWBuffer<float> A : register(u1);
+RWBuffer<float> B : register(u2);
+RWBuffer<float> C : register(u3);
+RWStructuredBuffer<float> Out : register(u0);
+
+// CHECK: define internal void @_Z4mainv()
+// CHECK-NEXT: entry:
+[numthreads(4,1,1)]
+void main() {
+// CHECK-NEXT: %First = alloca [3 x %"class.hlsl::RWBuffer"], align 4
+// CHECK-NEXT: %Second = alloca [4 x %"class.hlsl::RWBuffer"], align 4
+ RWBuffer<float> First[3] = { A, B, C };
+ RWBuffer<float> Second[4];
+
+// Verify initialization of First array from an initialization list
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr {{.*}} %First, ptr
{{.*}} @_ZL1A)
+// CHECK-NEXT: %[[Ptr1:.*]] = getelementptr inbounds %"class.hlsl::RWBuffer",
ptr %First, i32 1
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr {{.*}} %[[Ptr1]],
ptr {{.*}} @_ZL1B)
+// CHECK-NEXT: %[[Ptr2:.*]] = getelementptr inbounds %"class.hlsl::RWBuffer",
ptr %First, i32 2
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1ERKS1_(ptr {{.*}} %[[Ptr2]],
ptr {{.*}} @_ZL1C)
+
+// Verify default initialization of Second array, which means there is a loop
iterating
+// over the array elements and calling the default constructor for each
+// CHECK-NEXT: %[[ArrayBeginPtr:.*]] = getelementptr inbounds [4 x
%"class.hlsl::RWBuffer"], ptr %Second, i32 0, i32 0
+// CHECK-NEXT: %[[ArrayEndPtr:.*]] = getelementptr inbounds
%"class.hlsl::RWBuffer", ptr %[[ArrayBeginPtr]], i32 4
+// CHECK-NEXT: br label %[[ArrayInitLoop:.*]]
+// CHECK: [[ArrayInitLoop]]:
+// CHECK-NEXT: %[[ArrayCurPtr:.*]] = phi ptr [ %[[ArrayBeginPtr]], %entry ], [
%[[ArrayNextPtr:.*]], %[[ArrayInitLoop]] ]
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIfEC1Ev(ptr {{.*}} %[[ArrayCurPtr]])
+// CHECK-NEXT: %[[ArrayNextPtr]] = getelementptr inbounds
%"class.hlsl::RWBuffer", ptr %[[ArrayCurPtr]], i32 1
+// CHECK-NEXT: %[[ArrayInitDone:.*]] = icmp eq ptr %[[ArrayNextPtr]],
%[[ArrayEndPtr]]
+// CHECK-NEXT: br i1 %[[ArrayInitDone]], label %[[AfterArrayInit:.*]], label
%[[ArrayInitLoop]]
+// CHECK: [[AfterArrayInit]]:
+
+// Initialize First[2] with C
+// CHECK: %[[Ptr3:.*]] = getelementptr inbounds [4 x %"class.hlsl::RWBuffer"],
ptr %Second, i32 0, i32 2
+// CHECK: call {{.*}} @_ZN4hlsl8RWBufferIfEaSERKS1_(ptr {{.*}} %[[Ptr3]], ptr
{{.*}} @_ZL1C)
+ Second[2] = C;
+
+ // NOTE: _ZN4hlsl8RWBufferIfEixEj is the subscript operator for
RWBuffer<float>
+
+// get First[1][0] value
+// CHECK: %[[First_1_Ptr:.*]] = getelementptr inbounds [3 x
%"class.hlsl::RWBuffer"], ptr %First, i32 0, i32 1
+// CHECK: %[[BufPtr1:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIfEixEj(ptr
{{.*}} %[[First_1_Ptr]], i32 noundef 0)
+// CHECK: %[[Value1:.*]] = load float, ptr %[[BufPtr1]], align 4
+
+// get Second[2][0] value
+// CHECK: %[[Second_2_Ptr:.*]] = getelementptr inbounds [4 x
%"class.hlsl::RWBuffer"], ptr %Second, i32 0, i32 2
+// CHECK: %[[BufPtr2:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIfEixEj(ptr
{{.*}} %[[Second_2_Ptr]], i32 noundef 0)
+// CHECK: %[[Value2:.*]] = load float, ptr %[[BufPtr2]], align 4
+
+// add them
+// CHECK: %{{.*}} = fadd {{.*}} float %[[Value1]], %[[Value2]]
+ Out[0] = First[1][0] + Second[2][0];
+}
diff --git a/clang/test/CodeGenHLSL/resources/res-array-local3.hlsl
b/clang/test/CodeGenHLSL/resources/res-array-local3.hlsl
index e5bcdc651254f..21ca3f4a98f99 100644
--- a/clang/test/CodeGenHLSL/resources/res-array-local3.hlsl
+++ b/clang/test/CodeGenHLSL/resources/res-array-local3.hlsl
@@ -1,62 +1,62 @@
-// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-compute
-finclude-default-header \
-// RUN: -emit-llvm -disable-llvm-passes -o - %s | FileCheck %s
-
-// This test verifies handling of local arrays of resources when used
-// as a function argument that is modified inside the function.
-
-// CHECK: @_ZL1X = internal global %"class.hlsl::RWBuffer" poison, align 4
-// CHECK: @_ZL1Y = internal global %"class.hlsl::RWBuffer" poison, align 4
-
-RWBuffer<int> X : register(u0);
-RWBuffer<int> Y : register(u1);
-
-// CHECK: define {{.*}} @_Z6SomeFnA2_N4hlsl8RWBufferIiEEji(
-// CHECK-SAME: ptr noundef byval([2 x %"class.hlsl::RWBuffer"]) align 4 %B,
i32 noundef %Idx, i32 noundef %Val0)
-// CHECK-NEXT: entry:
-// CHECK-NEXT: %[[Idx_addr:.*]] = alloca i32, align 4
-// CHECK-NEXT: %[[Val0_addr:.*]] = alloca i32, align 4
-// CHECK-NEXT: store i32 %Idx, ptr %[[Idx_addr]], align 4
-// CHECK-NEXT: store i32 %Val0, ptr %[[Val0_addr]], align 4
-void SomeFn(RWBuffer<int> B[2], uint Idx, int Val0) {
-
-// CHECK-NEXT: %[[B_0_Ptr:.*]] = getelementptr inbounds [2 x
%"class.hlsl::RWBuffer"], ptr %B, i32 0, i32 0
-// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[B_0_Ptr]], ptr
align 4 @_ZL1Y, i32 4, i1 false)
- B[0] = Y;
-
-// NOTE: _ZN4hlsl8RWBufferIiEixEj is the subscript operator for RWBuffer<int>
-
-// CHECK-NEXT: %[[Val0:.*]] = load i32, ptr %[[Val0_addr]], align 4
-// CHECK-NEXT: %[[B_0_Ptr:.*]] = getelementptr inbounds [2 x
%"class.hlsl::RWBuffer"], ptr %B, i32 0, i32 0
-// CHECK-NEXT: %[[Idx:.*]] = load i32, ptr %[[Idx_addr]], align 4
-// CHECK-NEXT: %[[BufPtr:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIiEixEj(ptr
{{.*}} %[[B_0_Ptr]], i32 noundef %[[Idx]])
-// CHECK-NEXT: store i32 %[[Val0]], ptr %[[BufPtr]], align 4
- B[0][Idx] = Val0;
-}
-
-// CHECK: define {{.*}} void @_Z4mainj(i32 noundef %GI)
-// CHECK-NEXT: entry:
-// CHECK-NEXT: %[[GI_addr:.*]] = alloca i32, align 4
-[numthreads(4,1,1)]
-void main(uint GI : SV_GroupIndex) {
-// CHECK-NEXT: %A = alloca [2 x %"class.hlsl::RWBuffer"], align 4
-// CHECK-NEXT: %[[Tmp:.*]] = alloca [2 x %"class.hlsl::RWBuffer"], align 4
-// CHECK-NEXT: store i32 %GI, ptr %GI.addr, align 4
-
-// Initialization of array A with resources X and Y
-// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %A, ptr align 4
@_ZL1X, i32 4, i1 false)
-// CHECK-NEXT: %[[A_1_Ptr:.*]] = getelementptr inbounds
%"class.hlsl::RWBuffer", ptr %A, i32 1
-// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[A_1_Ptr]], ptr
align 4 @_ZL1Y, i32 4, i1 false)
- RWBuffer<int> A[2] = {X, Y};
-
-// Verify that SomeFn is called with a local copy of the array A
-// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Tmp]], ptr
align 4 %A, i32 8, i1 false)
-// CHECK-NEXT: %[[GI:.*]] = load i32, ptr %[[GI_addr]], align 4
-// CHECK-NEXT: call void @_Z6SomeFnA2_N4hlsl8RWBufferIiEEji(ptr noundef
byval([2 x %"class.hlsl::RWBuffer"]) align 4 %[[Tmp]], i32 noundef %[[GI]], i32
noundef 1)
- SomeFn(A, GI, 1);
-
-// CHECK-NEXT: %[[A_0_Ptr:.*]] = getelementptr inbounds [2 x
%"class.hlsl::RWBuffer"], ptr %A, i32 0, i32 0
-// CHECK-NEXT: %[[GI:.*]] = load i32, ptr %[[GI_addr]], align 4
-// CHECK-NEXT: %[[BufPtr:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIiEixEj(ptr
{{.*}} %[[A_0_Ptr]], i32 noundef %[[GI]])
-// CHECK-NEXT: store i32 2, ptr %[[BufPtr]], align 4
- A[0][GI] = 2;
-}
+// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.6-compute
-finclude-default-header \
+// RUN: -emit-llvm -disable-llvm-passes -o - %s | FileCheck %s
+
+// This test verifies handling of local arrays of resources when used
+// as a function argument that is modified inside the function.
+
+// CHECK: @_ZL1X = internal global %"class.hlsl::RWBuffer" poison, align 4
+// CHECK: @_ZL1Y = internal global %"class.hlsl::RWBuffer" poison, align 4
+
+RWBuffer<int> X : register(u0);
+RWBuffer<int> Y : register(u1);
+
+// CHECK: define {{.*}} @_Z6SomeFnA2_N4hlsl8RWBufferIiEEji(
+// CHECK-SAME: ptr noundef byval([2 x %"class.hlsl::RWBuffer"]) align 4 %B,
i32 noundef %Idx, i32 noundef %Val0)
+// CHECK-NEXT: entry:
+// CHECK-NEXT: %[[Idx_addr:.*]] = alloca i32, align 4
+// CHECK-NEXT: %[[Val0_addr:.*]] = alloca i32, align 4
+// CHECK-NEXT: store i32 %Idx, ptr %[[Idx_addr]], align 4
+// CHECK-NEXT: store i32 %Val0, ptr %[[Val0_addr]], align 4
+void SomeFn(RWBuffer<int> B[2], uint Idx, int Val0) {
+
+// CHECK-NEXT: %[[B_0_Ptr:.*]] = getelementptr inbounds [2 x
%"class.hlsl::RWBuffer"], ptr %B, i32 0, i32 0
+// CHECK-NEXT: call {{.*}} @_ZN4hlsl8RWBufferIiEaSERKS1_(ptr {{.*}}
%[[B_0_Ptr]], ptr {{.*}} @_ZL1Y)
+ B[0] = Y;
+
+// NOTE: _ZN4hlsl8RWBufferIiEixEj is the subscript operator for RWBuffer<int>
+
+// CHECK-NEXT: %[[Val0:.*]] = load i32, ptr %[[Val0_addr]], align 4
+// CHECK-NEXT: %[[B_0_Ptr:.*]] = getelementptr inbounds [2 x
%"class.hlsl::RWBuffer"], ptr %B, i32 0, i32 0
+// CHECK-NEXT: %[[Idx:.*]] = load i32, ptr %[[Idx_addr]], align 4
+// CHECK-NEXT: %[[BufPtr:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIiEixEj(ptr
{{.*}} %[[B_0_Ptr]], i32 noundef %[[Idx]])
+// CHECK-NEXT: store i32 %[[Val0]], ptr %[[BufPtr]], align 4
+ B[0][Idx] = Val0;
+}
+
+// CHECK: define {{.*}} void @_Z4mainj(i32 noundef %GI)
+// CHECK-NEXT: entry:
+// CHECK-NEXT: %[[GI_addr:.*]] = alloca i32, align 4
+[numthreads(4,1,1)]
+void main(uint GI : SV_GroupIndex) {
+// CHECK-NEXT: %A = alloca [2 x %"class.hlsl::RWBuffer"], align 4
+// CHECK-NEXT: %[[Tmp:.*]] = alloca [2 x %"class.hlsl::RWBuffer"], align 4
+// CHECK-NEXT: store i32 %GI, ptr %GI.addr, align 4
+
+// Initialization of array A with resources X and Y
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIiEC1ERKS1_(ptr {{.*}} %A, ptr
{{.*}} @_ZL1X)
+// CHECK-NEXT: %[[A_1_Ptr:.*]] = getelementptr inbounds
%"class.hlsl::RWBuffer", ptr %A, i32 1
+// CHECK-NEXT: call void @_ZN4hlsl8RWBufferIiEC1ERKS1_(ptr {{.*}}
%[[A_1_Ptr]], ptr {{.*}} @_ZL1Y)
+ RWBuffer<int> A[2] = {X, Y};
+
+// Verify that SomeFn is called with a local copy of the array A
+// CHECK-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 %[[Tmp]], ptr
align 4 %A, i32 8, i1 false)
+// CHECK-NEXT: %[[GI:.*]] = load i32, ptr %[[GI_addr]], align 4
+// CHECK-NEXT: call void @_Z6SomeFnA2_N4hlsl8RWBufferIiEEji(ptr noundef
byval([2 x %"class.hlsl::RWBuffer"]) align 4 %[[Tmp]], i32 noundef %[[GI]], i32
noundef 1)
+ SomeFn(A, GI, 1);
+
+// CHECK-NEXT: %[[A_0_Ptr:.*]] = getelementptr inbounds [2 x
%"class.hlsl::RWBuffer"], ptr %A, i32 0, i32 0
+// CHECK-NEXT: %[[GI:.*]] = load i32, ptr %[[GI_addr]], align 4
+// CHECK-NEXT: %[[BufPtr:.*]] = call {{.*}} ptr @_ZN4hlsl8RWBufferIiEixEj(ptr
{{.*}} %[[A_0_Ptr]], i32 noundef %[[GI]])
+// CHECK-NEXT: store i32 2, ptr %[[BufPtr]], align 4
+ A[0][GI] = 2;
+}
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
