[clang] [clang] teach OpenCL to set the param addrspace before calling CG EmitParmDecl (PR #184264)
Jameson Nash via cfe-commits
cfe-commits at lists.llvm.org
Thu Apr 30 08:52:20 PDT 2026
https://github.com/vtjnash updated https://github.com/llvm/llvm-project/pull/184264
>From ee1746edfd20006c32a307687323fd63569fb56e Mon Sep 17 00:00:00 2001
From: Jameson Nash <vtjnash at gmail.com>
Date: Fri, 27 Feb 2026 18:13:28 +0000
Subject: [PATCH 1/3] [clang] teach OpenCL to set the param addrspace before
calling CG EmitParmDecl
Unfortunately, this ends up in an awkward place between Sema and CG,
since a few CG phases create implicit parameters (e.g. for `this`) which
also need to be deduced into the correct address space. This is intended
to be clearly extensible for other targets that also need this.
Changes the constructor for ImplicitParamDecl to have only one code
path, and the object ctor itself is protected (like all the other
VarDecl subtypes). The memory is later cleaned up by the bump allocator,
and since the stack is basically also a bump allocator, is is typically
equally fast. (Reverts 550d13aebbcc03806811a3aa08e674b9b84e08e8.)
---
clang/include/clang/AST/Decl.h | 25 ++-
clang/lib/AST/Decl.cpp | 38 +++-
clang/lib/AST/DeclObjC.cpp | 5 +-
clang/lib/CodeGen/CGBlocks.cpp | 57 ++---
clang/lib/CodeGen/CGCoroutine.cpp | 16 +-
clang/lib/CodeGen/CGDecl.cpp | 13 +-
clang/lib/CodeGen/CGDeclCXX.cpp | 12 +-
clang/lib/CodeGen/CGExpr.cpp | 36 ++--
clang/lib/CodeGen/CGOpenMPRuntime.cpp | 199 ++++++++++--------
clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp | 18 +-
clang/lib/CodeGen/MicrosoftCXXABI.cpp | 18 +-
clang/lib/Sema/SemaDecl.cpp | 46 ++--
clang/lib/Sema/SemaTemplateInstantiate.cpp | 2 +
.../amdgpu-abi-struct-arg-byref.cl | 112 +++++-----
14 files changed, 324 insertions(+), 273 deletions(-)
diff --git a/clang/include/clang/AST/Decl.h b/clang/include/clang/AST/Decl.h
index 3d1d298479d90..01b9ba45be9ad 100644
--- a/clang/include/clang/AST/Decl.h
+++ b/clang/include/clang/AST/Decl.h
@@ -1728,6 +1728,10 @@ class VarDecl : public DeclaratorDecl, public Redeclarable<VarDecl> {
/// This can only be called for declarations where hasInit() is true.
CharUnits getFlexibleArrayInitChars(const ASTContext &Ctx) const;
+ /// Apply a deduced address space, if one isn't already set.
+ void assignAddressSpace(const ASTContext &Ctxt, LangAS AS);
+ void deduceParmAddressSpace(const ASTContext &Ctxt);
+
// Implement isa/cast/dyncast/etc.
static bool classof(const Decl *D) { return classofKind(D->getKind()); }
static bool classofKind(Kind K) { return K >= firstVar && K <= lastVar; }
@@ -1762,16 +1766,7 @@ enum class ImplicitParamKind {
class ImplicitParamDecl : public VarDecl {
void anchor() override;
-public:
- /// Create implicit parameter.
- static ImplicitParamDecl *Create(ASTContext &C, DeclContext *DC,
- SourceLocation IdLoc, IdentifierInfo *Id,
- QualType T, ImplicitParamKind ParamKind);
- static ImplicitParamDecl *Create(ASTContext &C, QualType T,
- ImplicitParamKind ParamKind);
-
- static ImplicitParamDecl *CreateDeserialized(ASTContext &C, GlobalDeclID ID);
-
+protected:
ImplicitParamDecl(ASTContext &C, DeclContext *DC, SourceLocation IdLoc,
const IdentifierInfo *Id, QualType Type,
ImplicitParamKind ParamKind)
@@ -1789,6 +1784,16 @@ class ImplicitParamDecl : public VarDecl {
setImplicit();
}
+public:
+ /// Create implicit parameter.
+ static ImplicitParamDecl *Create(ASTContext &C, DeclContext *DC,
+ SourceLocation IdLoc,
+ const IdentifierInfo *Id, QualType T,
+ ImplicitParamKind ParamKind);
+ static ImplicitParamDecl *Create(ASTContext &C, QualType T,
+ ImplicitParamKind ParamKind);
+
+ static ImplicitParamDecl *CreateDeserialized(ASTContext &C, GlobalDeclID ID);
/// Returns the implicit parameter kind.
ImplicitParamKind getParameterKind() const {
return static_cast<ImplicitParamKind>(NonParmVarDeclBits.ImplicitParamKind);
diff --git a/clang/lib/AST/Decl.cpp b/clang/lib/AST/Decl.cpp
index a7722aa0be3a4..535adcb2ae109 100644
--- a/clang/lib/AST/Decl.cpp
+++ b/clang/lib/AST/Decl.cpp
@@ -2918,6 +2918,33 @@ VarDecl::setInstantiationOfStaticDataMember(VarDecl *VD,
getASTContext().setInstantiatedFromStaticDataMember(this, VD, TSK);
}
+void VarDecl::assignAddressSpace(const ASTContext &Ctxt, LangAS AS) {
+ QualType Type = getType();
+ if (Type.hasAddressSpace())
+ return;
+ if (Type->isDependentType())
+ return;
+ if (Type->isSamplerT() || Type->isVoidType())
+ return;
+ assert(isa<ParmVarDecl>(this) || isa<ImplicitParamDecl>(this)
+ ? !Type->isArrayType()
+ : !isa<DecayedType>(Type));
+ Type = Ctxt.getAddrSpaceQualType(Type, AS);
+ // Apply any qualifiers (including address space) from the array type to
+ // the element type. This implements C99 6.7.3p8: "If the specification of
+ // an array type includes any type qualifiers, the element type is so
+ // qualified, not the array type."
+ if (Type->isArrayType())
+ Type = QualType(Ctxt.getAsArrayType(Type), 0);
+ setType(Type);
+}
+
+void VarDecl::deduceParmAddressSpace(const ASTContext &Ctxt) {
+ assert(isa<ParmVarDecl>(this) || isa<ImplicitParamDecl>(this));
+ if (Ctxt.getLangOpts().OpenCL)
+ assignAddressSpace(Ctxt, LangAS::opencl_private);
+}
+
//===----------------------------------------------------------------------===//
// ParmVarDecl Implementation
//===----------------------------------------------------------------------===//
@@ -5569,14 +5596,19 @@ void ImplicitParamDecl::anchor() {}
ImplicitParamDecl *ImplicitParamDecl::Create(ASTContext &C, DeclContext *DC,
SourceLocation IdLoc,
- IdentifierInfo *Id, QualType Type,
+ const IdentifierInfo *Id,
+ QualType Type,
ImplicitParamKind ParamKind) {
- return new (C, DC) ImplicitParamDecl(C, DC, IdLoc, Id, Type, ParamKind);
+ auto *Parm = new (C, DC) ImplicitParamDecl(C, DC, IdLoc, Id, Type, ParamKind);
+ Parm->deduceParmAddressSpace(C);
+ return Parm;
}
ImplicitParamDecl *ImplicitParamDecl::Create(ASTContext &C, QualType Type,
ImplicitParamKind ParamKind) {
- return new (C, nullptr) ImplicitParamDecl(C, Type, ParamKind);
+ auto *Parm = new (C, nullptr) ImplicitParamDecl(C, Type, ParamKind);
+ Parm->deduceParmAddressSpace(C);
+ return Parm;
}
ImplicitParamDecl *ImplicitParamDecl::CreateDeserialized(ASTContext &C,
diff --git a/clang/lib/AST/DeclObjC.cpp b/clang/lib/AST/DeclObjC.cpp
index a66eb72981084..5f80bf7aa29d8 100644
--- a/clang/lib/AST/DeclObjC.cpp
+++ b/clang/lib/AST/DeclObjC.cpp
@@ -1200,9 +1200,10 @@ void ObjCMethodDecl::createImplicitParams(ASTContext &Context,
if (selfIsPseudoStrong)
Self->setARCPseudoStrong(true);
- setCmdDecl(ImplicitParamDecl::Create(
+ auto *CmdDecl = ImplicitParamDecl::Create(
Context, this, SourceLocation(), &Context.Idents.get("_cmd"),
- Context.getObjCSelType(), ImplicitParamKind::ObjCCmd));
+ Context.getObjCSelType(), ImplicitParamKind::ObjCCmd);
+ setCmdDecl(CmdDecl);
}
ObjCInterfaceDecl *ObjCMethodDecl::getClassInterface() {
diff --git a/clang/lib/CodeGen/CGBlocks.cpp b/clang/lib/CodeGen/CGBlocks.cpp
index 20a595e58e4c2..1142bf2ecf325 100644
--- a/clang/lib/CodeGen/CGBlocks.cpp
+++ b/clang/lib/CodeGen/CGBlocks.cpp
@@ -1028,8 +1028,8 @@ llvm::Value *CodeGenFunction::EmitBlockLiteral(const CGBlockInfo &blockInfo) {
} else {
// Fake up a new variable so that EmitScalarInit doesn't think
// we're referring to the variable in its own initializer.
- ImplicitParamDecl BlockFieldPseudoVar(getContext(), type,
- ImplicitParamKind::Other);
+ auto *BlockFieldPseudoVar = ImplicitParamDecl::Create(
+ getContext(), type, ImplicitParamKind::Other);
// We use one of these or the other depending on whether the
// reference is nested.
@@ -1042,7 +1042,7 @@ llvm::Value *CodeGenFunction::EmitBlockLiteral(const CGBlockInfo &blockInfo) {
// FIXME: Pass a specific location for the expr init so that the store is
// attributed to a reasonable location - otherwise it may be attributed to
// locations of subexpressions in the initialization.
- EmitExprAsInit(&l2r, &BlockFieldPseudoVar,
+ EmitExprAsInit(&l2r, BlockFieldPseudoVar,
MakeAddrLValue(blockField, type, AlignmentSource::Decl),
/*captured by init*/ false);
}
@@ -1490,10 +1490,10 @@ llvm::Function *CodeGenFunction::GenerateBlockFunction(
const IdentifierInfo *II = &CGM.getContext().Idents.get(".block_descriptor");
- ImplicitParamDecl SelfDecl(getContext(), const_cast<BlockDecl *>(blockDecl),
- SourceLocation(), II, selfTy,
- ImplicitParamKind::ObjCSelf);
- args.push_back(&SelfDecl);
+ auto *SelfDecl = ImplicitParamDecl::Create(
+ getContext(), const_cast<BlockDecl *>(blockDecl), SourceLocation(), II,
+ selfTy, ImplicitParamKind::ObjCSelf);
+ args.push_back(SelfDecl);
// Now add the rest of the parameters.
args.append(blockDecl->param_begin(), blockDecl->param_end());
@@ -1935,10 +1935,12 @@ CodeGenFunction::GenerateCopyHelperFunction(const CGBlockInfo &blockInfo) {
QualType ReturnTy = C.VoidTy;
FunctionArgList args;
- ImplicitParamDecl DstDecl(C, C.VoidPtrTy, ImplicitParamKind::Other);
- args.push_back(&DstDecl);
- ImplicitParamDecl SrcDecl(C, C.VoidPtrTy, ImplicitParamKind::Other);
- args.push_back(&SrcDecl);
+ auto *DstDecl =
+ ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
+ args.push_back(DstDecl);
+ auto *SrcDecl =
+ ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
+ args.push_back(SrcDecl);
const CGFunctionInfo &FI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, args);
@@ -1962,11 +1964,11 @@ CodeGenFunction::GenerateCopyHelperFunction(const CGBlockInfo &blockInfo) {
StartFunction(GlobalDecl(), ReturnTy, Fn, FI, args);
auto AL = ApplyDebugLocation::CreateArtificial(*this);
- Address src = GetAddrOfLocalVar(&SrcDecl);
+ Address src = GetAddrOfLocalVar(SrcDecl);
src = Address(Builder.CreateLoad(src), blockInfo.StructureType,
blockInfo.BlockAlign);
- Address dst = GetAddrOfLocalVar(&DstDecl);
+ Address dst = GetAddrOfLocalVar(DstDecl);
dst = Address(Builder.CreateLoad(dst), blockInfo.StructureType,
blockInfo.BlockAlign);
@@ -2132,8 +2134,9 @@ CodeGenFunction::GenerateDestroyHelperFunction(const CGBlockInfo &blockInfo) {
QualType ReturnTy = C.VoidTy;
FunctionArgList args;
- ImplicitParamDecl SrcDecl(C, C.VoidPtrTy, ImplicitParamKind::Other);
- args.push_back(&SrcDecl);
+ auto *SrcDecl =
+ ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
+ args.push_back(SrcDecl);
const CGFunctionInfo &FI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, args);
@@ -2158,7 +2161,7 @@ CodeGenFunction::GenerateDestroyHelperFunction(const CGBlockInfo &blockInfo) {
auto AL = ApplyDebugLocation::CreateArtificial(*this);
- Address src = GetAddrOfLocalVar(&SrcDecl);
+ Address src = GetAddrOfLocalVar(SrcDecl);
src = Address(Builder.CreateLoad(src), blockInfo.StructureType,
blockInfo.BlockAlign);
@@ -2401,11 +2404,13 @@ generateByrefCopyHelper(CodeGenFunction &CGF, const BlockByrefInfo &byrefInfo,
QualType ReturnTy = Context.VoidTy;
FunctionArgList args;
- ImplicitParamDecl Dst(Context, Context.VoidPtrTy, ImplicitParamKind::Other);
- args.push_back(&Dst);
+ auto *Dst = ImplicitParamDecl::Create(Context, Context.VoidPtrTy,
+ ImplicitParamKind::Other);
+ args.push_back(Dst);
- ImplicitParamDecl Src(Context, Context.VoidPtrTy, ImplicitParamKind::Other);
- args.push_back(&Src);
+ auto *Src = ImplicitParamDecl::Create(Context, Context.VoidPtrTy,
+ ImplicitParamKind::Other);
+ args.push_back(Src);
const CGFunctionInfo &FI =
CGF.CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, args);
@@ -2430,14 +2435,14 @@ generateByrefCopyHelper(CodeGenFunction &CGF, const BlockByrefInfo &byrefInfo,
if (generator.needsCopy()) {
// dst->x
- Address destField = CGF.GetAddrOfLocalVar(&Dst);
+ Address destField = CGF.GetAddrOfLocalVar(Dst);
destField = Address(CGF.Builder.CreateLoad(destField), byrefInfo.Type,
byrefInfo.ByrefAlignment);
destField =
CGF.emitBlockByrefAddress(destField, byrefInfo, false, "dest-object");
// src->x
- Address srcField = CGF.GetAddrOfLocalVar(&Src);
+ Address srcField = CGF.GetAddrOfLocalVar(Src);
srcField = Address(CGF.Builder.CreateLoad(srcField), byrefInfo.Type,
byrefInfo.ByrefAlignment);
srcField =
@@ -2468,9 +2473,9 @@ generateByrefDisposeHelper(CodeGenFunction &CGF,
QualType R = Context.VoidTy;
FunctionArgList args;
- ImplicitParamDecl Src(CGF.getContext(), Context.VoidPtrTy,
- ImplicitParamKind::Other);
- args.push_back(&Src);
+ auto *Src = ImplicitParamDecl::Create(CGF.getContext(), Context.VoidPtrTy,
+ ImplicitParamKind::Other);
+ args.push_back(Src);
const CGFunctionInfo &FI =
CGF.CGM.getTypes().arrangeBuiltinFunctionDeclaration(R, args);
@@ -2494,7 +2499,7 @@ generateByrefDisposeHelper(CodeGenFunction &CGF,
auto AL = ApplyDebugLocation::CreateArtificial(CGF);
if (generator.needsDispose()) {
- Address addr = CGF.GetAddrOfLocalVar(&Src);
+ Address addr = CGF.GetAddrOfLocalVar(Src);
addr = Address(CGF.Builder.CreateLoad(addr), byrefInfo.Type,
byrefInfo.ByrefAlignment);
addr = CGF.emitBlockByrefAddress(addr, byrefInfo, false, "object");
diff --git a/clang/lib/CodeGen/CGCoroutine.cpp b/clang/lib/CodeGen/CGCoroutine.cpp
index 2972f015aa893..7f7de687399ab 100644
--- a/clang/lib/CodeGen/CGCoroutine.cpp
+++ b/clang/lib/CodeGen/CGCoroutine.cpp
@@ -425,12 +425,14 @@ CodeGenFunction::generateAwaitSuspendWrapper(Twine const &CoroName,
FunctionArgList args;
- ImplicitParamDecl AwaiterDecl(C, C.VoidPtrTy, ImplicitParamKind::Other);
- ImplicitParamDecl FrameDecl(C, C.VoidPtrTy, ImplicitParamKind::Other);
+ auto *AwaiterDecl =
+ ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
+ auto *FrameDecl =
+ ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
QualType ReturnTy = S.getSuspendExpr()->getType();
- args.push_back(&AwaiterDecl);
- args.push_back(&FrameDecl);
+ args.push_back(AwaiterDecl);
+ args.push_back(FrameDecl);
const CGFunctionInfo &FI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, args);
@@ -452,12 +454,12 @@ CodeGenFunction::generateAwaitSuspendWrapper(Twine const &CoroName,
StartFunction(GlobalDecl(), ReturnTy, Fn, FI, args);
// FIXME: add TBAA metadata to the loads
- llvm::Value *AwaiterPtr = Builder.CreateLoad(GetAddrOfLocalVar(&AwaiterDecl));
+ llvm::Value *AwaiterPtr = Builder.CreateLoad(GetAddrOfLocalVar(AwaiterDecl));
auto AwaiterLValue =
- MakeNaturalAlignAddrLValue(AwaiterPtr, AwaiterDecl.getType());
+ MakeNaturalAlignAddrLValue(AwaiterPtr, AwaiterDecl->getType());
CurAwaitSuspendWrapper.FramePtr =
- Builder.CreateLoad(GetAddrOfLocalVar(&FrameDecl));
+ Builder.CreateLoad(GetAddrOfLocalVar(FrameDecl));
auto AwaiterBinder = CodeGenFunction::OpaqueValueMappingData::bind(
*this, S.getOpaqueValue(), AwaiterLValue);
diff --git a/clang/lib/CodeGen/CGDecl.cpp b/clang/lib/CodeGen/CGDecl.cpp
index a7479809d219e..22079e0fb3a38 100644
--- a/clang/lib/CodeGen/CGDecl.cpp
+++ b/clang/lib/CodeGen/CGDecl.cpp
@@ -2714,9 +2714,6 @@ void CodeGenFunction::EmitParmDecl(const VarDecl &D, ParamValue Arg,
if (Arg.isIndirect()) {
DeclPtr = Arg.getIndirectAddress();
DeclPtr = DeclPtr.withElementType(ConvertTypeForMem(Ty));
- // Indirect argument is in alloca address space, which may be different
- // from the default address space.
- auto AllocaAS = CGM.getASTAllocaAddressSpace();
auto *V = DeclPtr.emitRawPointer(*this);
AllocaPtr = RawAddress(V, DeclPtr.getElementType(), DeclPtr.getAlignment());
@@ -2732,13 +2729,9 @@ void CodeGenFunction::EmitParmDecl(const VarDecl &D, ParamValue Arg,
EmitStoreOfScalar(V, AllocaPtr, /* Volatile */ false, PtrTy);
}
- auto SrcLangAS = getLangOpts().OpenCL ? LangAS::opencl_private : AllocaAS;
- auto DestLangAS =
- getLangOpts().OpenCL ? LangAS::opencl_private : LangAS::Default;
- if (SrcLangAS != DestLangAS) {
- assert(getContext().getTargetAddressSpace(SrcLangAS) ==
- CGM.getDataLayout().getAllocaAddrSpace());
- auto DestAS = getContext().getTargetAddressSpace(DestLangAS);
+ LangAS DestLangAS = Ty.getAddressSpace();
+ unsigned DestAS = getContext().getTargetAddressSpace(DestLangAS);
+ if (DeclPtr.getAddressSpace() != DestAS) {
auto *T = llvm::PointerType::get(getLLVMContext(), DestAS);
DeclPtr = DeclPtr.withPointer(performAddrSpaceCast(V, T),
DeclPtr.isKnownNonNull());
diff --git a/clang/lib/CodeGen/CGDeclCXX.cpp b/clang/lib/CodeGen/CGDeclCXX.cpp
index d47c50bf26243..1798d42b69d91 100644
--- a/clang/lib/CodeGen/CGDeclCXX.cpp
+++ b/clang/lib/CodeGen/CGDeclCXX.cpp
@@ -298,9 +298,9 @@ llvm::Function *CodeGenFunction::createTLSAtExitStub(
CodeGenFunction CGF(CGM);
FunctionArgList Args;
- ImplicitParamDecl IPD(CGM.getContext(), CGM.getContext().IntTy,
- ImplicitParamKind::Other);
- Args.push_back(&IPD);
+ auto *IPD = ImplicitParamDecl::Create(
+ CGM.getContext(), CGM.getContext().IntTy, ImplicitParamKind::Other);
+ Args.push_back(IPD);
QualType ResTy = CGM.getContext().IntTy;
CGF.StartFunction(GlobalDecl(&D, DynamicInitKind::AtExit), ResTy, DtorStub,
@@ -1197,9 +1197,9 @@ llvm::Function *CodeGenFunction::generateDestroyHelper(
Address addr, QualType type, Destroyer *destroyer,
bool useEHCleanupForArray, const VarDecl *VD) {
FunctionArgList args;
- ImplicitParamDecl Dst(getContext(), getContext().VoidPtrTy,
- ImplicitParamKind::Other);
- args.push_back(&Dst);
+ auto *Dst = ImplicitParamDecl::Create(getContext(), getContext().VoidPtrTy,
+ ImplicitParamKind::Other);
+ args.push_back(Dst);
const CGFunctionInfo &FI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(getContext().VoidTy, args);
diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
index 5275115626f1e..c2dbc375ec7f6 100644
--- a/clang/lib/CodeGen/CGExpr.cpp
+++ b/clang/lib/CodeGen/CGExpr.cpp
@@ -4368,13 +4368,15 @@ void CodeGenFunction::EmitCfiCheckStub() {
QualType QInt64Ty = C.getIntTypeForBitwidth(64, false);
FunctionArgList FnArgs;
- ImplicitParamDecl ArgCallsiteTypeId(C, QInt64Ty, ImplicitParamKind::Other);
- ImplicitParamDecl ArgAddr(C, C.VoidPtrTy, ImplicitParamKind::Other);
- ImplicitParamDecl ArgCFICheckFailData(C, C.VoidPtrTy,
- ImplicitParamKind::Other);
- FnArgs.push_back(&ArgCallsiteTypeId);
- FnArgs.push_back(&ArgAddr);
- FnArgs.push_back(&ArgCFICheckFailData);
+ auto *ArgCallsiteTypeId =
+ ImplicitParamDecl::Create(C, QInt64Ty, ImplicitParamKind::Other);
+ auto *ArgAddr =
+ ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
+ auto *ArgCFICheckFailData =
+ ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
+ FnArgs.push_back(ArgCallsiteTypeId);
+ FnArgs.push_back(ArgAddr);
+ FnArgs.push_back(ArgCFICheckFailData);
const CGFunctionInfo &FI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, FnArgs);
@@ -4414,12 +4416,12 @@ void CodeGenFunction::EmitCfiCheckFail() {
SanitizerKind::SO_CFIICall},
CheckHandler);
FunctionArgList Args;
- ImplicitParamDecl ArgData(getContext(), getContext().VoidPtrTy,
- ImplicitParamKind::Other);
- ImplicitParamDecl ArgAddr(getContext(), getContext().VoidPtrTy,
- ImplicitParamKind::Other);
- Args.push_back(&ArgData);
- Args.push_back(&ArgAddr);
+ auto *ArgData = ImplicitParamDecl::Create(
+ getContext(), getContext().VoidPtrTy, ImplicitParamKind::Other);
+ auto *ArgAddr = ImplicitParamDecl::Create(
+ getContext(), getContext().VoidPtrTy, ImplicitParamKind::Other);
+ Args.push_back(ArgData);
+ Args.push_back(ArgAddr);
const CGFunctionInfo &FI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(getContext().VoidTy, Args);
@@ -4443,11 +4445,11 @@ void CodeGenFunction::EmitCfiCheckFail() {
SanOpts = CGM.getLangOpts().Sanitize;
llvm::Value *Data =
- EmitLoadOfScalar(GetAddrOfLocalVar(&ArgData), /*Volatile=*/false,
- CGM.getContext().VoidPtrTy, ArgData.getLocation());
+ EmitLoadOfScalar(GetAddrOfLocalVar(ArgData), /*Volatile=*/false,
+ CGM.getContext().VoidPtrTy, ArgData->getLocation());
llvm::Value *Addr =
- EmitLoadOfScalar(GetAddrOfLocalVar(&ArgAddr), /*Volatile=*/false,
- CGM.getContext().VoidPtrTy, ArgAddr.getLocation());
+ EmitLoadOfScalar(GetAddrOfLocalVar(ArgAddr), /*Volatile=*/false,
+ CGM.getContext().VoidPtrTy, ArgAddr->getLocation());
// Data == nullptr means the calling module has trap behaviour for this check.
llvm::Value *DataIsNotNullPtr =
diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index d5371e43f5930..7caa71d2b9b99 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -1086,12 +1086,14 @@ emitCombinerOrInitializer(CodeGenModule &CGM, QualType Ty,
ASTContext &C = CGM.getContext();
QualType PtrTy = C.getPointerType(Ty).withRestrict();
FunctionArgList Args;
- ImplicitParamDecl OmpOutParm(C, /*DC=*/nullptr, Out->getLocation(),
- /*Id=*/nullptr, PtrTy, ImplicitParamKind::Other);
- ImplicitParamDecl OmpInParm(C, /*DC=*/nullptr, In->getLocation(),
- /*Id=*/nullptr, PtrTy, ImplicitParamKind::Other);
- Args.push_back(&OmpOutParm);
- Args.push_back(&OmpInParm);
+ auto *OmpOutParm = ImplicitParamDecl::Create(
+ C, /*DC=*/nullptr, Out->getLocation(),
+ /*Id=*/nullptr, PtrTy, ImplicitParamKind::Other);
+ auto *OmpInParm = ImplicitParamDecl::Create(
+ C, /*DC=*/nullptr, In->getLocation(),
+ /*Id=*/nullptr, PtrTy, ImplicitParamKind::Other);
+ Args.push_back(OmpOutParm);
+ Args.push_back(OmpInParm);
const CGFunctionInfo &FnInfo =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
llvm::FunctionType *FnTy = CGM.getTypes().GetFunctionType(FnInfo);
@@ -1113,11 +1115,11 @@ emitCombinerOrInitializer(CodeGenModule &CGM, QualType Ty,
CGF.StartFunction(GlobalDecl(), C.VoidTy, Fn, FnInfo, Args, In->getLocation(),
Out->getLocation());
CodeGenFunction::OMPPrivateScope Scope(CGF);
- Address AddrIn = CGF.GetAddrOfLocalVar(&OmpInParm);
+ Address AddrIn = CGF.GetAddrOfLocalVar(OmpInParm);
Scope.addPrivate(
In, CGF.EmitLoadOfPointerLValue(AddrIn, PtrTy->castAs<PointerType>())
.getAddress());
- Address AddrOut = CGF.GetAddrOfLocalVar(&OmpOutParm);
+ Address AddrOut = CGF.GetAddrOfLocalVar(OmpOutParm);
Scope.addPrivate(
Out, CGF.EmitLoadOfPointerLValue(AddrOut, PtrTy->castAs<PointerType>())
.getAddress());
@@ -1656,10 +1658,10 @@ llvm::Function *CGOpenMPRuntime::emitThreadPrivateVarDefinition(
// threadprivate copy of the variable VD
CodeGenFunction CtorCGF(CGM);
FunctionArgList Args;
- ImplicitParamDecl Dst(CGM.getContext(), /*DC=*/nullptr, Loc,
- /*Id=*/nullptr, CGM.getContext().VoidPtrTy,
- ImplicitParamKind::Other);
- Args.push_back(&Dst);
+ auto *Dst = ImplicitParamDecl::Create(
+ CGM.getContext(), /*DC=*/nullptr, Loc,
+ /*Id=*/nullptr, CGM.getContext().VoidPtrTy, ImplicitParamKind::Other);
+ Args.push_back(Dst);
const auto &FI = CGM.getTypes().arrangeBuiltinFunctionDeclaration(
CGM.getContext().VoidPtrTy, Args);
@@ -1670,15 +1672,15 @@ llvm::Function *CGOpenMPRuntime::emitThreadPrivateVarDefinition(
CtorCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidPtrTy, Fn, FI,
Args, Loc, Loc);
llvm::Value *ArgVal = CtorCGF.EmitLoadOfScalar(
- CtorCGF.GetAddrOfLocalVar(&Dst), /*Volatile=*/false,
- CGM.getContext().VoidPtrTy, Dst.getLocation());
+ CtorCGF.GetAddrOfLocalVar(Dst), /*Volatile=*/false,
+ CGM.getContext().VoidPtrTy, Dst->getLocation());
Address Arg(ArgVal, CtorCGF.ConvertTypeForMem(ASTTy),
VDAddr.getAlignment());
CtorCGF.EmitAnyExprToMem(Init, Arg, Init->getType().getQualifiers(),
/*IsInitializer=*/true);
ArgVal = CtorCGF.EmitLoadOfScalar(
- CtorCGF.GetAddrOfLocalVar(&Dst), /*Volatile=*/false,
- CGM.getContext().VoidPtrTy, Dst.getLocation());
+ CtorCGF.GetAddrOfLocalVar(Dst), /*Volatile=*/false,
+ CGM.getContext().VoidPtrTy, Dst->getLocation());
CtorCGF.Builder.CreateStore(ArgVal, CtorCGF.ReturnValue);
CtorCGF.FinishFunction();
Ctor = Fn;
@@ -1688,10 +1690,10 @@ llvm::Function *CGOpenMPRuntime::emitThreadPrivateVarDefinition(
// of the variable VD
CodeGenFunction DtorCGF(CGM);
FunctionArgList Args;
- ImplicitParamDecl Dst(CGM.getContext(), /*DC=*/nullptr, Loc,
- /*Id=*/nullptr, CGM.getContext().VoidPtrTy,
- ImplicitParamKind::Other);
- Args.push_back(&Dst);
+ auto *Dst = ImplicitParamDecl::Create(
+ CGM.getContext(), /*DC=*/nullptr, Loc,
+ /*Id=*/nullptr, CGM.getContext().VoidPtrTy, ImplicitParamKind::Other);
+ Args.push_back(Dst);
const auto &FI = CGM.getTypes().arrangeBuiltinFunctionDeclaration(
CGM.getContext().VoidTy, Args);
@@ -1705,8 +1707,8 @@ llvm::Function *CGOpenMPRuntime::emitThreadPrivateVarDefinition(
// Create a scope with an artificial location for the body of this function.
auto AL = ApplyDebugLocation::CreateArtificial(DtorCGF);
llvm::Value *ArgVal = DtorCGF.EmitLoadOfScalar(
- DtorCGF.GetAddrOfLocalVar(&Dst),
- /*Volatile=*/false, CGM.getContext().VoidPtrTy, Dst.getLocation());
+ DtorCGF.GetAddrOfLocalVar(Dst),
+ /*Volatile=*/false, CGM.getContext().VoidPtrTy, Dst->getLocation());
DtorCGF.emitDestroy(
Address(ArgVal, DtorCGF.Int8Ty, VDAddr.getAlignment()), ASTTy,
DtorCGF.getDestroyer(ASTTy.isDestructedType()),
@@ -2260,12 +2262,15 @@ static llvm::Value *emitCopyprivateCopyFunction(
ASTContext &C = CGM.getContext();
// void copy_func(void *LHSArg, void *RHSArg);
FunctionArgList Args;
- ImplicitParamDecl LHSArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.VoidPtrTy,
- ImplicitParamKind::Other);
- ImplicitParamDecl RHSArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.VoidPtrTy,
- ImplicitParamKind::Other);
- Args.push_back(&LHSArg);
- Args.push_back(&RHSArg);
+
+ auto *LHSArg =
+ ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+ C.VoidPtrTy, ImplicitParamKind::Other);
+ auto *RHSArg =
+ ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+ C.VoidPtrTy, ImplicitParamKind::Other);
+ Args.push_back(LHSArg);
+ Args.push_back(RHSArg);
const auto &CGFI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
std::string Name =
@@ -2282,11 +2287,11 @@ static llvm::Value *emitCopyprivateCopyFunction(
// Dest = (void*[n])(LHSArg);
// Src = (void*[n])(RHSArg);
Address LHS(CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
- CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(&LHSArg)),
+ CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(LHSArg)),
CGF.Builder.getPtrTy(0)),
ArgsElemType, CGF.getPointerAlign());
Address RHS(CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
- CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(&RHSArg)),
+ CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(RHSArg)),
CGF.Builder.getPtrTy(0)),
ArgsElemType, CGF.getPointerAlign());
// *(Type0*)Dst[0] = *(Type0*)Src[0];
@@ -3135,13 +3140,14 @@ emitProxyTaskFunction(CodeGenModule &CGM, SourceLocation Loc,
llvm::Value *TaskPrivatesMap) {
ASTContext &C = CGM.getContext();
FunctionArgList Args;
- ImplicitParamDecl GtidArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, KmpInt32Ty,
- ImplicitParamKind::Other);
- ImplicitParamDecl TaskTypeArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
- KmpTaskTWithPrivatesPtrQTy.withRestrict(),
- ImplicitParamKind::Other);
- Args.push_back(&GtidArg);
- Args.push_back(&TaskTypeArg);
+ auto *GtidArg =
+ ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+ KmpInt32Ty, ImplicitParamKind::Other);
+ auto *TaskTypeArg = ImplicitParamDecl::Create(
+ C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+ KmpTaskTWithPrivatesPtrQTy.withRestrict(), ImplicitParamKind::Other);
+ Args.push_back(GtidArg);
+ Args.push_back(TaskTypeArg);
const auto &TaskEntryFnInfo =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(KmpInt32Ty, Args);
llvm::FunctionType *TaskEntryTy =
@@ -3163,9 +3169,9 @@ emitProxyTaskFunction(CodeGenModule &CGM, SourceLocation Loc,
// tt->task_data.lb, tt->task_data.ub, tt->task_data.st, tt->task_data.liter,
// tt->task_data.shareds);
llvm::Value *GtidParam = CGF.EmitLoadOfScalar(
- CGF.GetAddrOfLocalVar(&GtidArg), /*Volatile=*/false, KmpInt32Ty, Loc);
+ CGF.GetAddrOfLocalVar(GtidArg), /*Volatile=*/false, KmpInt32Ty, Loc);
LValue TDBase = CGF.EmitLoadOfPointerLValue(
- CGF.GetAddrOfLocalVar(&TaskTypeArg),
+ CGF.GetAddrOfLocalVar(TaskTypeArg),
KmpTaskTWithPrivatesPtrQTy->castAs<PointerType>());
const auto *KmpTaskTWithPrivatesQTyRD =
KmpTaskTWithPrivatesQTy->castAsRecordDecl();
@@ -3239,13 +3245,14 @@ static llvm::Value *emitDestructorsFunction(CodeGenModule &CGM,
QualType KmpTaskTWithPrivatesQTy) {
ASTContext &C = CGM.getContext();
FunctionArgList Args;
- ImplicitParamDecl GtidArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, KmpInt32Ty,
- ImplicitParamKind::Other);
- ImplicitParamDecl TaskTypeArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
- KmpTaskTWithPrivatesPtrQTy.withRestrict(),
- ImplicitParamKind::Other);
- Args.push_back(&GtidArg);
- Args.push_back(&TaskTypeArg);
+ auto *GtidArg =
+ ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+ KmpInt32Ty, ImplicitParamKind::Other);
+ auto *TaskTypeArg = ImplicitParamDecl::Create(
+ C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+ KmpTaskTWithPrivatesPtrQTy.withRestrict(), ImplicitParamKind::Other);
+ Args.push_back(GtidArg);
+ Args.push_back(TaskTypeArg);
const auto &DestructorFnInfo =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(KmpInt32Ty, Args);
llvm::FunctionType *DestructorFnTy =
@@ -3265,7 +3272,7 @@ static llvm::Value *emitDestructorsFunction(CodeGenModule &CGM,
Args, Loc, Loc);
LValue Base = CGF.EmitLoadOfPointerLValue(
- CGF.GetAddrOfLocalVar(&TaskTypeArg),
+ CGF.GetAddrOfLocalVar(TaskTypeArg),
KmpTaskTWithPrivatesPtrQTy->castAs<PointerType>());
const auto *KmpTaskTWithPrivatesQTyRD =
KmpTaskTWithPrivatesQTy->castAsRecordDecl();
@@ -3298,11 +3305,11 @@ emitTaskPrivateMappingFunction(CodeGenModule &CGM, SourceLocation Loc,
ArrayRef<PrivateDataTy> Privates) {
ASTContext &C = CGM.getContext();
FunctionArgList Args;
- ImplicitParamDecl TaskPrivatesArg(
+ auto *TaskPrivatesArg = ImplicitParamDecl::Create(
C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
C.getPointerType(PrivatesQTy).withConst().withRestrict(),
ImplicitParamKind::Other);
- Args.push_back(&TaskPrivatesArg);
+ Args.push_back(TaskPrivatesArg);
llvm::DenseMap<CanonicalDeclPtr<const VarDecl>, unsigned> PrivateVarsPos;
unsigned Counter = 1;
for (const Expr *E : Data.PrivateVars) {
@@ -3376,8 +3383,8 @@ emitTaskPrivateMappingFunction(CodeGenModule &CGM, SourceLocation Loc,
// *privi = &.privates.privi;
LValue Base = CGF.EmitLoadOfPointerLValue(
- CGF.GetAddrOfLocalVar(&TaskPrivatesArg),
- TaskPrivatesArg.getType()->castAs<PointerType>());
+ CGF.GetAddrOfLocalVar(TaskPrivatesArg),
+ TaskPrivatesArg->getType()->castAs<PointerType>());
const auto *PrivatesQTyRD = PrivatesQTy->castAsRecordDecl();
Counter = 0;
for (const FieldDecl *Field : PrivatesQTyRD->fields()) {
@@ -3551,17 +3558,18 @@ emitTaskDupFunction(CodeGenModule &CGM, SourceLocation Loc,
ArrayRef<PrivateDataTy> Privates, bool WithLastIter) {
ASTContext &C = CGM.getContext();
FunctionArgList Args;
- ImplicitParamDecl DstArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
- KmpTaskTWithPrivatesPtrQTy,
- ImplicitParamKind::Other);
- ImplicitParamDecl SrcArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
- KmpTaskTWithPrivatesPtrQTy,
- ImplicitParamKind::Other);
- ImplicitParamDecl LastprivArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.IntTy,
+ auto *DstArg = ImplicitParamDecl::Create(
+ C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, KmpTaskTWithPrivatesPtrQTy,
+ ImplicitParamKind::Other);
+ auto *SrcArg = ImplicitParamDecl::Create(
+ C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, KmpTaskTWithPrivatesPtrQTy,
+ ImplicitParamKind::Other);
+ auto *LastprivArg =
+ ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.IntTy,
ImplicitParamKind::Other);
- Args.push_back(&DstArg);
- Args.push_back(&SrcArg);
- Args.push_back(&LastprivArg);
+ Args.push_back(DstArg);
+ Args.push_back(SrcArg);
+ Args.push_back(LastprivArg);
const auto &TaskDupFnInfo =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
llvm::FunctionType *TaskDupTy = CGM.getTypes().GetFunctionType(TaskDupFnInfo);
@@ -3577,7 +3585,7 @@ emitTaskDupFunction(CodeGenModule &CGM, SourceLocation Loc,
Loc);
LValue TDBase = CGF.EmitLoadOfPointerLValue(
- CGF.GetAddrOfLocalVar(&DstArg),
+ CGF.GetAddrOfLocalVar(DstArg),
KmpTaskTWithPrivatesPtrQTy->castAs<PointerType>());
// task_dst->liter = lastpriv;
if (WithLastIter) {
@@ -3586,7 +3594,7 @@ emitTaskDupFunction(CodeGenModule &CGM, SourceLocation Loc,
TDBase, *KmpTaskTWithPrivatesQTyRD->field_begin());
LValue LILVal = CGF.EmitLValueForField(Base, *LIFI);
llvm::Value *Lastpriv = CGF.EmitLoadOfScalar(
- CGF.GetAddrOfLocalVar(&LastprivArg), /*Volatile=*/false, C.IntTy, Loc);
+ CGF.GetAddrOfLocalVar(LastprivArg), /*Volatile=*/false, C.IntTy, Loc);
CGF.EmitStoreOfScalar(Lastpriv, LILVal);
}
@@ -3595,7 +3603,7 @@ emitTaskDupFunction(CodeGenModule &CGM, SourceLocation Loc,
Address KmpTaskSharedsPtr = Address::invalid();
if (!Data.FirstprivateVars.empty()) {
LValue TDBase = CGF.EmitLoadOfPointerLValue(
- CGF.GetAddrOfLocalVar(&SrcArg),
+ CGF.GetAddrOfLocalVar(SrcArg),
KmpTaskTWithPrivatesPtrQTy->castAs<PointerType>());
LValue Base = CGF.EmitLValueForField(
TDBase, *KmpTaskTWithPrivatesQTyRD->field_begin());
@@ -4972,12 +4980,14 @@ llvm::Function *CGOpenMPRuntime::emitReductionFunction(
// void reduction_func(void *LHSArg, void *RHSArg);
FunctionArgList Args;
- ImplicitParamDecl LHSArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.VoidPtrTy,
- ImplicitParamKind::Other);
- ImplicitParamDecl RHSArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.VoidPtrTy,
- ImplicitParamKind::Other);
- Args.push_back(&LHSArg);
- Args.push_back(&RHSArg);
+ auto *LHSArg =
+ ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+ C.VoidPtrTy, ImplicitParamKind::Other);
+ auto *RHSArg =
+ ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+ C.VoidPtrTy, ImplicitParamKind::Other);
+ Args.push_back(LHSArg);
+ Args.push_back(RHSArg);
const auto &CGFI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
std::string Name = getReductionFuncName(ReducerName);
@@ -4994,11 +5004,11 @@ llvm::Function *CGOpenMPRuntime::emitReductionFunction(
// Dst = (void*[n])(LHSArg);
// Src = (void*[n])(RHSArg);
Address LHS(CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
- CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(&LHSArg)),
+ CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(LHSArg)),
CGF.Builder.getPtrTy(0)),
ArgsElemType, CGF.getPointerAlign());
Address RHS(CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
- CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(&RHSArg)),
+ CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(RHSArg)),
CGF.Builder.getPtrTy(0)),
ArgsElemType, CGF.getPointerAlign());
@@ -5687,12 +5697,14 @@ static llvm::Value *emitReduceInitFunction(CodeGenModule &CGM,
QualType VoidPtrTy = C.VoidPtrTy;
VoidPtrTy.addRestrict();
FunctionArgList Args;
- ImplicitParamDecl Param(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, VoidPtrTy,
- ImplicitParamKind::Other);
- ImplicitParamDecl ParamOrig(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, VoidPtrTy,
- ImplicitParamKind::Other);
- Args.emplace_back(&Param);
- Args.emplace_back(&ParamOrig);
+ auto *Param =
+ ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+ VoidPtrTy, ImplicitParamKind::Other);
+ auto *ParamOrig =
+ ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+ VoidPtrTy, ImplicitParamKind::Other);
+ Args.emplace_back(Param);
+ Args.emplace_back(ParamOrig);
const auto &FnInfo =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
llvm::FunctionType *FnTy = CGM.getTypes().GetFunctionType(FnInfo);
@@ -5707,7 +5719,7 @@ static llvm::Value *emitReduceInitFunction(CodeGenModule &CGM,
CGF.StartFunction(GlobalDecl(), C.VoidTy, Fn, FnInfo, Args, Loc, Loc);
QualType PrivateType = RCG.getPrivateType(N);
Address PrivateAddr = CGF.EmitLoadOfPointer(
- CGF.GetAddrOfLocalVar(&Param).withElementType(CGF.Builder.getPtrTy(0)),
+ CGF.GetAddrOfLocalVar(Param).withElementType(CGF.Builder.getPtrTy(0)),
C.getPointerType(PrivateType)->castAs<PointerType>());
llvm::Value *Size = nullptr;
// If the size of the reduction item is non-constant, load it from global
@@ -5725,7 +5737,7 @@ static llvm::Value *emitReduceInitFunction(CodeGenModule &CGM,
// pointer to the address of the original reduction item (reuired by reduction
// initializer)
if (RCG.usesReductionInitializer(N)) {
- Address SharedAddr = CGF.GetAddrOfLocalVar(&ParamOrig);
+ Address SharedAddr = CGF.GetAddrOfLocalVar(ParamOrig);
OrigAddr = CGF.EmitLoadOfPointer(
SharedAddr,
CGM.getContext().VoidPtrTy.castAs<PointerType>()->getTypePtr());
@@ -5759,12 +5771,14 @@ static llvm::Value *emitReduceCombFunction(CodeGenModule &CGM,
const auto *LHSVD = cast<VarDecl>(cast<DeclRefExpr>(LHS)->getDecl());
const auto *RHSVD = cast<VarDecl>(cast<DeclRefExpr>(RHS)->getDecl());
FunctionArgList Args;
- ImplicitParamDecl ParamInOut(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
- C.VoidPtrTy, ImplicitParamKind::Other);
- ImplicitParamDecl ParamIn(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.VoidPtrTy,
- ImplicitParamKind::Other);
- Args.emplace_back(&ParamInOut);
- Args.emplace_back(&ParamIn);
+ auto *ParamInOut =
+ ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+ C.VoidPtrTy, ImplicitParamKind::Other);
+ auto *ParamIn =
+ ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+ C.VoidPtrTy, ImplicitParamKind::Other);
+ Args.emplace_back(ParamInOut);
+ Args.emplace_back(ParamIn);
const auto &FnInfo =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
llvm::FunctionType *FnTy = CGM.getTypes().GetFunctionType(FnInfo);
@@ -5796,14 +5810,14 @@ static llvm::Value *emitReduceCombFunction(CodeGenModule &CGM,
LHSVD,
// Pull out the pointer to the variable.
CGF.EmitLoadOfPointer(
- CGF.GetAddrOfLocalVar(&ParamInOut)
+ CGF.GetAddrOfLocalVar(ParamInOut)
.withElementType(CGF.Builder.getPtrTy(0)),
C.getPointerType(LHSVD->getType())->castAs<PointerType>()));
PrivateScope.addPrivate(
RHSVD,
// Pull out the pointer to the variable.
CGF.EmitLoadOfPointer(
- CGF.GetAddrOfLocalVar(&ParamIn).withElementType(
+ CGF.GetAddrOfLocalVar(ParamIn).withElementType(
CGF.Builder.getPtrTy(0)),
C.getPointerType(RHSVD->getType())->castAs<PointerType>()));
PrivateScope.Privatize();
@@ -5832,9 +5846,10 @@ static llvm::Value *emitReduceFiniFunction(CodeGenModule &CGM,
return nullptr;
ASTContext &C = CGM.getContext();
FunctionArgList Args;
- ImplicitParamDecl Param(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.VoidPtrTy,
- ImplicitParamKind::Other);
- Args.emplace_back(&Param);
+ auto *Param =
+ ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+ C.VoidPtrTy, ImplicitParamKind::Other);
+ Args.emplace_back(Param);
const auto &FnInfo =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
llvm::FunctionType *FnTy = CGM.getTypes().GetFunctionType(FnInfo);
@@ -5848,7 +5863,7 @@ static llvm::Value *emitReduceFiniFunction(CodeGenModule &CGM,
CodeGenFunction CGF(CGM);
CGF.StartFunction(GlobalDecl(), C.VoidTy, Fn, FnInfo, Args, Loc, Loc);
Address PrivateAddr = CGF.EmitLoadOfPointer(
- CGF.GetAddrOfLocalVar(&Param), C.VoidPtrTy.castAs<PointerType>());
+ CGF.GetAddrOfLocalVar(Param), C.VoidPtrTy.castAs<PointerType>());
llvm::Value *Size = nullptr;
// If the size of the reduction item is non-constant, load it from global
// threadprivate variable.
diff --git a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
index 943c2ac9f8491..53088b4422fcf 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
@@ -1894,14 +1894,14 @@ llvm::Function *CGOpenMPRuntimeGPU::createParallelDataSharingWrapper(
Ctx.getIntTypeForBitwidth(/*DestWidth=*/16, /*Signed=*/false);
QualType Int32QTy =
Ctx.getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/false);
- ImplicitParamDecl ParallelLevelArg(Ctx, /*DC=*/nullptr, D.getBeginLoc(),
- /*Id=*/nullptr, Int16QTy,
- ImplicitParamKind::Other);
- ImplicitParamDecl WrapperArg(Ctx, /*DC=*/nullptr, D.getBeginLoc(),
- /*Id=*/nullptr, Int32QTy,
- ImplicitParamKind::Other);
- WrapperArgs.emplace_back(&ParallelLevelArg);
- WrapperArgs.emplace_back(&WrapperArg);
+ auto *ParallelLevelArg = ImplicitParamDecl::Create(
+ Ctx, /*DC=*/nullptr, D.getBeginLoc(),
+ /*Id=*/nullptr, Int16QTy, ImplicitParamKind::Other);
+ auto *WrapperArg = ImplicitParamDecl::Create(
+ Ctx, /*DC=*/nullptr, D.getBeginLoc(),
+ /*Id=*/nullptr, Int32QTy, ImplicitParamKind::Other);
+ WrapperArgs.emplace_back(ParallelLevelArg);
+ WrapperArgs.emplace_back(WrapperArg);
const CGFunctionInfo &CGFI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(Ctx.VoidTy, WrapperArgs);
@@ -1935,7 +1935,7 @@ llvm::Function *CGOpenMPRuntimeGPU::createParallelDataSharingWrapper(
// Get the array of arguments.
SmallVector<llvm::Value *, 8> Args;
- Args.emplace_back(CGF.GetAddrOfLocalVar(&WrapperArg).emitRawPointer(CGF));
+ Args.emplace_back(CGF.GetAddrOfLocalVar(WrapperArg).emitRawPointer(CGF));
Args.emplace_back(ZeroAddr.emitRawPointer(CGF));
CGBuilderTy &Bld = CGF.Builder;
diff --git a/clang/lib/CodeGen/MicrosoftCXXABI.cpp b/clang/lib/CodeGen/MicrosoftCXXABI.cpp
index 2588efb6f15fd..9b444206e8a3d 100644
--- a/clang/lib/CodeGen/MicrosoftCXXABI.cpp
+++ b/clang/lib/CodeGen/MicrosoftCXXABI.cpp
@@ -4163,25 +4163,25 @@ MicrosoftCXXABI::getAddrOfCXXCtorClosure(const CXXConstructorDecl *CD,
// Following the 'this' pointer is a reference to the source object that we
// are copying from.
- ImplicitParamDecl SrcParam(
+ auto *SrcParam = ImplicitParamDecl::Create(
getContext(), /*DC=*/nullptr, SourceLocation(),
&getContext().Idents.get("src"),
getContext().getLValueReferenceType(RecordTy,
/*SpelledAsLValue=*/true),
ImplicitParamKind::Other);
if (IsCopy)
- FunctionArgs.push_back(&SrcParam);
+ FunctionArgs.push_back(SrcParam);
// Constructors for classes which utilize virtual bases have an additional
// parameter which indicates whether or not it is being delegated to by a more
// derived constructor.
- ImplicitParamDecl IsMostDerived(getContext(), /*DC=*/nullptr,
- SourceLocation(),
- &getContext().Idents.get("is_most_derived"),
- getContext().IntTy, ImplicitParamKind::Other);
+ auto *IsMostDerived =
+ ImplicitParamDecl::Create(getContext(), /*DC=*/nullptr, SourceLocation(),
+ &getContext().Idents.get("is_most_derived"),
+ getContext().IntTy, ImplicitParamKind::Other);
// Only add the parameter to the list if the class has virtual bases.
if (RD->getNumVBases() > 0)
- FunctionArgs.push_back(&IsMostDerived);
+ FunctionArgs.push_back(IsMostDerived);
// Start defining the function.
auto NL = ApplyDebugLocation::CreateEmpty(CGF);
@@ -4193,7 +4193,7 @@ MicrosoftCXXABI::getAddrOfCXXCtorClosure(const CXXConstructorDecl *CD,
llvm::Value *This = getThisValue(CGF);
llvm::Value *SrcVal =
- IsCopy ? CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(&SrcParam), "src")
+ IsCopy ? CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(SrcParam), "src")
: nullptr;
CallArgList Args;
@@ -4203,7 +4203,7 @@ MicrosoftCXXABI::getAddrOfCXXCtorClosure(const CXXConstructorDecl *CD,
// Push the src ptr.
if (SrcVal)
- Args.add(RValue::get(SrcVal), SrcParam.getType());
+ Args.add(RValue::get(SrcVal), SrcParam->getType());
// Add the rest of the default arguments.
SmallVector<const Stmt *, 4> ArgVec;
diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
index 22e85e213c5db..eb5b6d65b4d58 100644
--- a/clang/lib/Sema/SemaDecl.cpp
+++ b/clang/lib/Sema/SemaDecl.cpp
@@ -7075,13 +7075,6 @@ static void SetNestedNameSpecifier(Sema &S, DeclaratorDecl *DD, Declarator &D) {
}
void Sema::deduceOpenCLAddressSpace(VarDecl *Var) {
- QualType Type = Var->getType();
- if (Type.hasAddressSpace())
- return;
- if (Type->isDependentType())
- return;
- if (Type->isSamplerT() || Type->isVoidType())
- return;
LangAS ImplAS = LangAS::opencl_private;
// OpenCL C v3.0 s6.7.8 - For OpenCL C 2.0 or with the
// __opencl_c_program_scope_global_variables feature, the address space
@@ -7090,27 +7083,7 @@ void Sema::deduceOpenCLAddressSpace(VarDecl *Var) {
if (getOpenCLOptions().areProgramScopeVariablesSupported(getLangOpts()) &&
Var->hasGlobalStorage())
ImplAS = LangAS::opencl_global;
- // If the original type from a decayed type is an array type and that array
- // type has no address space yet, deduce it now.
- if (auto DT = dyn_cast<DecayedType>(Type)) {
- auto OrigTy = DT->getOriginalType();
- if (!OrigTy.hasAddressSpace() && OrigTy->isArrayType()) {
- // Add the address space to the original array type and then propagate
- // that to the element type through `getAsArrayType`.
- OrigTy = Context.getAddrSpaceQualType(OrigTy, ImplAS);
- OrigTy = QualType(Context.getAsArrayType(OrigTy), 0);
- // Re-generate the decayed type.
- Type = Context.getDecayedType(OrigTy);
- }
- }
- Type = Context.getAddrSpaceQualType(Type, ImplAS);
- // Apply any qualifiers (including address space) from the array type to
- // the element type. This implements C99 6.7.3p8: "If the specification of
- // an array type includes any type qualifiers, the element type is so
- // qualified, not the array type."
- if (Type->isArrayType())
- Type = QualType(Context.getAsArrayType(Type), 0);
- Var->setType(Type);
+ Var->assignAddressSpace(Context, ImplAS);
}
static void checkWeakAttr(Sema &S, NamedDecl &ND) {
@@ -15814,8 +15787,7 @@ Decl *Sema::ActOnParamDeclarator(Scope *S, Declarator &D,
Diag(New->getLocation(), diag::err_block_on_nonlocal);
}
- if (getLangOpts().OpenCL)
- deduceOpenCLAddressSpace(New);
+ New->deduceParmAddressSpace(Context);
return New;
}
@@ -15906,6 +15878,20 @@ ParmVarDecl *Sema::CheckParameter(DeclContext *DC, SourceLocation StartLoc,
T = Context.getLifetimeQualifiedType(T, lifetime);
}
+ if (getLangOpts().OpenCL) {
+ assert(!isa<DecayedType>(T));
+ if (T->isArrayType() && !T.hasAddressSpace()) {
+ QualType ET = Context.getAsArrayType(T)->getElementType();
+ if (!ET.hasAddressSpace()) {
+ // Add the private address space to the contents of the pointer when a
+ // pointer parameter is declared as an array and not declared.
+ LangAS ImplAS = LangAS::opencl_private;
+ T = Context.getAddrSpaceQualType(T, ImplAS);
+ T = QualType(Context.getAsArrayType(T), 0);
+ }
+ }
+ }
+
ParmVarDecl *New = ParmVarDecl::Create(Context, DC, StartLoc, NameLoc, Name,
Context.getAdjustedParameterType(T),
TSInfo, SC, nullptr);
diff --git a/clang/lib/Sema/SemaTemplateInstantiate.cpp b/clang/lib/Sema/SemaTemplateInstantiate.cpp
index 59a3b01e963ca..f9dcd8b24b783 100644
--- a/clang/lib/Sema/SemaTemplateInstantiate.cpp
+++ b/clang/lib/Sema/SemaTemplateInstantiate.cpp
@@ -3234,6 +3234,8 @@ Sema::SubstParmVarDecl(ParmVarDecl *OldParm,
InstantiateAttrs(TemplateArgs, OldParm, NewParm);
+ NewParm->deduceParmAddressSpace(Context);
+
return NewParm;
}
diff --git a/clang/test/CodeGenOpenCL/amdgpu-abi-struct-arg-byref.cl b/clang/test/CodeGenOpenCL/amdgpu-abi-struct-arg-byref.cl
index 1cf1fcf1c9f9f..bef560c6f119f 100644
--- a/clang/test/CodeGenOpenCL/amdgpu-abi-struct-arg-byref.cl
+++ b/clang/test/CodeGenOpenCL/amdgpu-abi-struct-arg-byref.cl
@@ -122,6 +122,7 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[IN1:%.*]] = addrspacecast ptr addrspace(5) [[IN]] to ptr
// AMDGCN-NEXT: [[COERCE_DIVE:%.*]] = getelementptr inbounds nuw [[STRUCT_MAT3X3]], ptr [[IN1]], i32 0, i32 0
// AMDGCN-NEXT: store [9 x i32] [[IN_COERCE]], ptr [[COERCE_DIVE]], align 4
+// AMDGCN-NEXT: [[IN1_ASCAST:%.*]] = addrspacecast ptr [[IN1]] to ptr addrspace(5)
// AMDGCN-NEXT: [[TMP0:%.*]] = load [[STRUCT_MAT4X4]], ptr addrspace(5) [[RETVAL]], align 4
// AMDGCN-NEXT: ret [[STRUCT_MAT4X4]] [[TMP0]]
//
@@ -137,12 +138,12 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: store ptr addrspace(1) [[OUT]], ptr [[OUT_ADDR_ASCAST]], align 8
// AMDGCN-NEXT: [[TMP0:%.*]] = load ptr addrspace(1), ptr [[IN_ADDR_ASCAST]], align 8
// AMDGCN-NEXT: [[TMP1:%.*]] = load ptr addrspace(1), ptr [[OUT_ADDR_ASCAST]], align 8
-// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_ker(ptr addrspace(1) noundef align 4 [[TMP0]], ptr addrspace(1) noundef align 4 [[TMP1]]) #[[ATTR4:[0-9]+]]
+// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_ker(ptr addrspace(1) noundef align 4 [[TMP0]], ptr addrspace(1) noundef align 4 [[TMP1]]) #[[ATTR3:[0-9]+]]
// AMDGCN-NEXT: ret void
//
//
// AMDGCN-LABEL: define dso_local void @__clang_ocl_kern_imp_ker(
-// AMDGCN-SAME: ptr addrspace(1) noundef align 4 [[IN:%.*]], ptr addrspace(1) noundef align 4 [[OUT:%.*]]) #[[ATTR2:[0-9]+]] !kernel_arg_addr_space [[META4]] !kernel_arg_access_qual [[META5]] !kernel_arg_type [[META6]] !kernel_arg_base_type [[META6]] !kernel_arg_type_qual [[META7]] {
+// AMDGCN-SAME: ptr addrspace(1) noundef align 4 [[IN:%.*]], ptr addrspace(1) noundef align 4 [[OUT:%.*]]) #[[ATTR1]] !kernel_arg_addr_space [[META4]] !kernel_arg_access_qual [[META5]] !kernel_arg_type [[META6]] !kernel_arg_base_type [[META6]] !kernel_arg_type_qual [[META7]] {
// AMDGCN-NEXT: [[ENTRY:.*:]]
// AMDGCN-NEXT: [[IN_ADDR:%.*]] = alloca ptr addrspace(1), align 8, addrspace(5)
// AMDGCN-NEXT: [[OUT_ADDR:%.*]] = alloca ptr addrspace(1), align 8, addrspace(5)
@@ -157,7 +158,7 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds [[STRUCT_MAT3X3:%.*]], ptr addrspace(1) [[TMP1]], i64 1
// AMDGCN-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw [[STRUCT_MAT3X3]], ptr addrspace(1) [[ARRAYIDX1]], i32 0, i32 0
// AMDGCN-NEXT: [[TMP3:%.*]] = load [9 x i32], ptr addrspace(1) [[TMP2]], align 4
-// AMDGCN-NEXT: [[CALL:%.*]] = call [[STRUCT_MAT4X4]] @[[FOO:[a-zA-Z0-9_$\"\\.-]*[a-zA-Z_$\"\\.-][a-zA-Z0-9_$\"\\.-]*]]([9 x i32] [[TMP3]]) #[[ATTR4]]
+// AMDGCN-NEXT: [[CALL:%.*]] = call [[STRUCT_MAT4X4]] @[[FOO:[a-zA-Z0-9_$\"\\.-]*[a-zA-Z_$\"\\.-][a-zA-Z0-9_$\"\\.-]*]]([9 x i32] [[TMP3]]) #[[ATTR3]]
// AMDGCN-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT_MAT4X4]], ptr addrspace(5) [[TMP]], i32 0, i32 0
// AMDGCN-NEXT: [[TMP5:%.*]] = extractvalue [[STRUCT_MAT4X4]] [[CALL]], 0
// AMDGCN-NEXT: store [16 x i32] [[TMP5]], ptr addrspace(5) [[TMP4]], align 4
@@ -171,6 +172,7 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[COERCE:%.*]] = alloca [[STRUCT_MAT32X32]], align 4, addrspace(5)
// AMDGCN-NEXT: [[IN:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN-NEXT: call void @llvm.memcpy.p0.p5.i64(ptr align 4 [[IN]], ptr addrspace(5) align 4 [[TMP0]], i64 4096, i1 false)
+// AMDGCN-NEXT: [[IN_ASCAST:%.*]] = addrspacecast ptr [[IN]] to ptr addrspace(5)
// AMDGCN-NEXT: ret void
//
//
@@ -185,12 +187,12 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: store ptr addrspace(1) [[OUT]], ptr [[OUT_ADDR_ASCAST]], align 8
// AMDGCN-NEXT: [[TMP0:%.*]] = load ptr addrspace(1), ptr [[IN_ADDR_ASCAST]], align 8
// AMDGCN-NEXT: [[TMP1:%.*]] = load ptr addrspace(1), ptr [[OUT_ADDR_ASCAST]], align 8
-// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_ker_large(ptr addrspace(1) noundef align 4 [[TMP0]], ptr addrspace(1) noundef align 4 [[TMP1]]) #[[ATTR4]]
+// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_ker_large(ptr addrspace(1) noundef align 4 [[TMP0]], ptr addrspace(1) noundef align 4 [[TMP1]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
// AMDGCN-LABEL: define dso_local void @__clang_ocl_kern_imp_ker_large(
-// AMDGCN-SAME: ptr addrspace(1) noundef align 4 [[IN:%.*]], ptr addrspace(1) noundef align 4 [[OUT:%.*]]) #[[ATTR2]] !kernel_arg_addr_space [[META4]] !kernel_arg_access_qual [[META5]] !kernel_arg_type [[META8]] !kernel_arg_base_type [[META8]] !kernel_arg_type_qual [[META7]] {
+// AMDGCN-SAME: ptr addrspace(1) noundef align 4 [[IN:%.*]], ptr addrspace(1) noundef align 4 [[OUT:%.*]]) #[[ATTR1]] !kernel_arg_addr_space [[META4]] !kernel_arg_access_qual [[META5]] !kernel_arg_type [[META8]] !kernel_arg_base_type [[META8]] !kernel_arg_type_qual [[META7]] {
// AMDGCN-NEXT: [[ENTRY:.*:]]
// AMDGCN-NEXT: [[IN_ADDR:%.*]] = alloca ptr addrspace(1), align 8, addrspace(5)
// AMDGCN-NEXT: [[OUT_ADDR:%.*]] = alloca ptr addrspace(1), align 8, addrspace(5)
@@ -205,7 +207,7 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[TMP1:%.*]] = load ptr addrspace(1), ptr [[IN_ADDR_ASCAST]], align 8
// AMDGCN-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds [[STRUCT_MAT32X32]], ptr addrspace(1) [[TMP1]], i64 1
// AMDGCN-NEXT: call void @llvm.memcpy.p5.p1.i64(ptr addrspace(5) align 4 [[BYVAL_TEMP]], ptr addrspace(1) align 4 [[ARRAYIDX1]], i64 4096, i1 false)
-// AMDGCN-NEXT: call void @foo_large(ptr addrspace(5) dead_on_unwind writable sret([[STRUCT_MAT64X64]]) align 4 [[TMP]], ptr addrspace(5) noundef byref([[STRUCT_MAT32X32]]) align 4 [[BYVAL_TEMP]]) #[[ATTR4]]
+// AMDGCN-NEXT: call void @foo_large(ptr addrspace(5) dead_on_unwind writable sret([[STRUCT_MAT64X64]]) align 4 [[TMP]], ptr addrspace(5) noundef byref([[STRUCT_MAT32X32]]) align 4 [[BYVAL_TEMP]]) #[[ATTR3]]
// AMDGCN-NEXT: call void @llvm.memcpy.p1.p5.i64(ptr addrspace(1) align 4 [[ARRAYIDX]], ptr addrspace(5) align 4 [[TMP]], i64 16384, i1 false)
// AMDGCN-NEXT: ret void
//
@@ -219,10 +221,11 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[DOTCOMPOUNDLITERAL_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCOMPOUNDLITERAL]] to ptr
// AMDGCN-NEXT: [[COERCE_DIVE:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr [[U1]], i32 0, i32 0
// AMDGCN-NEXT: store <2 x i32> [[U_COERCE]], ptr [[COERCE_DIVE]], align 8
+// AMDGCN-NEXT: [[U1_ASCAST:%.*]] = addrspacecast ptr [[U1]] to ptr addrspace(5)
// AMDGCN-NEXT: store <2 x i32> zeroinitializer, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
// AMDGCN-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
-// AMDGCN-NEXT: [[X:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr [[U1]], i32 0, i32 0
-// AMDGCN-NEXT: store <2 x i32> [[TMP0]], ptr [[X]], align 8
+// AMDGCN-NEXT: [[X:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr addrspace(5) [[U1_ASCAST]], i32 0, i32 0
+// AMDGCN-NEXT: store <2 x i32> [[TMP0]], ptr addrspace(5) [[X]], align 8
// AMDGCN-NEXT: ret void
//
//
@@ -234,11 +237,12 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[U:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN-NEXT: [[DOTCOMPOUNDLITERAL_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCOMPOUNDLITERAL]] to ptr
// AMDGCN-NEXT: call void @llvm.memcpy.p0.p5.i64(ptr align 8 [[U]], ptr addrspace(5) align 8 [[TMP0]], i64 800, i1 false)
+// AMDGCN-NEXT: [[U_ASCAST:%.*]] = addrspacecast ptr [[U]] to ptr addrspace(5)
// AMDGCN-NEXT: store <2 x i32> zeroinitializer, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
// AMDGCN-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
-// AMDGCN-NEXT: [[X:%.*]] = getelementptr inbounds nuw [[STRUCT_LARGESTRUCTONEMEMBER]], ptr [[U]], i32 0, i32 0
-// AMDGCN-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [100 x <2 x i32>], ptr [[X]], i64 0, i64 0
-// AMDGCN-NEXT: store <2 x i32> [[TMP1]], ptr [[ARRAYIDX]], align 8
+// AMDGCN-NEXT: [[X:%.*]] = getelementptr inbounds nuw [[STRUCT_LARGESTRUCTONEMEMBER]], ptr addrspace(5) [[U_ASCAST]], i32 0, i32 0
+// AMDGCN-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [100 x <2 x i32>], ptr addrspace(5) [[X]], i64 0, i64 0
+// AMDGCN-NEXT: store <2 x i32> [[TMP1]], ptr addrspace(5) [[ARRAYIDX]], align 8
// AMDGCN-NEXT: ret void
//
//
@@ -247,23 +251,23 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[ENTRY:.*:]]
// AMDGCN-NEXT: [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_LARGESTRUCTONEMEMBER:%.*]], align 8, addrspace(5)
// AMDGCN-NEXT: call void @llvm.memcpy.p5.p1.i64(ptr addrspace(5) align 8 [[BYVAL_TEMP]], ptr addrspace(1) align 8 @g_s, i64 800, i1 false)
-// AMDGCN-NEXT: call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[BYVAL_TEMP]]) #[[ATTR4]]
+// AMDGCN-NEXT: call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[BYVAL_TEMP]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
// AMDGCN-LABEL: define dso_local amdgpu_kernel void @test_indirect_arg_local(
// AMDGCN-SAME: ) #[[ATTR1]] !kernel_arg_addr_space [[META9:![0-9]+]] !kernel_arg_access_qual [[META9]] !kernel_arg_type [[META9]] !kernel_arg_base_type [[META9]] !kernel_arg_type_qual [[META9]] {
// AMDGCN-NEXT: [[ENTRY:.*:]]
-// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_test_indirect_arg_local() #[[ATTR4]]
+// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_test_indirect_arg_local() #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
// AMDGCN-LABEL: define dso_local void @__clang_ocl_kern_imp_test_indirect_arg_local(
-// AMDGCN-SAME: ) #[[ATTR2]] !kernel_arg_addr_space [[META9]] !kernel_arg_access_qual [[META9]] !kernel_arg_type [[META9]] !kernel_arg_base_type [[META9]] !kernel_arg_type_qual [[META9]] {
+// AMDGCN-SAME: ) #[[ATTR1]] !kernel_arg_addr_space [[META9]] !kernel_arg_access_qual [[META9]] !kernel_arg_type [[META9]] !kernel_arg_base_type [[META9]] !kernel_arg_type_qual [[META9]] {
// AMDGCN-NEXT: [[ENTRY:.*:]]
// AMDGCN-NEXT: [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_LARGESTRUCTONEMEMBER:%.*]], align 8, addrspace(5)
// AMDGCN-NEXT: call void @llvm.memcpy.p5.p3.i64(ptr addrspace(5) align 8 [[BYVAL_TEMP]], ptr addrspace(3) align 8 @test_indirect_arg_local.l_s, i64 800, i1 false)
-// AMDGCN-NEXT: call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[BYVAL_TEMP]]) #[[ATTR4]]
+// AMDGCN-NEXT: call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[BYVAL_TEMP]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
@@ -271,7 +275,7 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-SAME: ) #[[ATTR0]] {
// AMDGCN-NEXT: [[ENTRY:.*:]]
// AMDGCN-NEXT: [[P_S:%.*]] = alloca [[STRUCT_LARGESTRUCTONEMEMBER:%.*]], align 8, addrspace(5)
-// AMDGCN-NEXT: call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[P_S]]) #[[ATTR4]]
+// AMDGCN-NEXT: call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[P_S]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
@@ -282,22 +286,24 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[U1:%.*]] = addrspacecast ptr addrspace(5) [[U]] to ptr
// AMDGCN-NEXT: [[COERCE_DIVE:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr [[U1]], i32 0, i32 0
// AMDGCN-NEXT: store <2 x i32> [[U_COERCE]], ptr [[COERCE_DIVE]], align 8
-// AMDGCN-NEXT: [[COERCE_DIVE2:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr [[U1]], i32 0, i32 0
-// AMDGCN-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[COERCE_DIVE2]], align 8
-// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_KernelOneMember(<2 x i32> [[TMP0]]) #[[ATTR4]]
+// AMDGCN-NEXT: [[U1_ASCAST:%.*]] = addrspacecast ptr [[U1]] to ptr addrspace(5)
+// AMDGCN-NEXT: [[COERCE_DIVE2:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr addrspace(5) [[U1_ASCAST]], i32 0, i32 0
+// AMDGCN-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr addrspace(5) [[COERCE_DIVE2]], align 8
+// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_KernelOneMember(<2 x i32> [[TMP0]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
// AMDGCN-LABEL: define dso_local void @__clang_ocl_kern_imp_KernelOneMember(
-// AMDGCN-SAME: <2 x i32> [[U_COERCE:%.*]]) #[[ATTR2]] !kernel_arg_addr_space [[META10]] !kernel_arg_access_qual [[META11]] !kernel_arg_type [[META12]] !kernel_arg_base_type [[META12]] !kernel_arg_type_qual [[META13]] {
+// AMDGCN-SAME: <2 x i32> [[U_COERCE:%.*]]) #[[ATTR1]] !kernel_arg_addr_space [[META10]] !kernel_arg_access_qual [[META11]] !kernel_arg_type [[META12]] !kernel_arg_base_type [[META12]] !kernel_arg_type_qual [[META13]] {
// AMDGCN-NEXT: [[ENTRY:.*:]]
// AMDGCN-NEXT: [[U:%.*]] = alloca [[STRUCT_STRUCTONEMEMBER:%.*]], align 8, addrspace(5)
// AMDGCN-NEXT: [[U1:%.*]] = addrspacecast ptr addrspace(5) [[U]] to ptr
// AMDGCN-NEXT: [[COERCE_DIVE:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr [[U1]], i32 0, i32 0
// AMDGCN-NEXT: store <2 x i32> [[U_COERCE]], ptr [[COERCE_DIVE]], align 8
-// AMDGCN-NEXT: [[COERCE_DIVE2:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr [[U1]], i32 0, i32 0
-// AMDGCN-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[COERCE_DIVE2]], align 8
-// AMDGCN-NEXT: call void @FuncOneMember(<2 x i32> [[TMP0]]) #[[ATTR4]]
+// AMDGCN-NEXT: [[U1_ASCAST:%.*]] = addrspacecast ptr [[U1]] to ptr addrspace(5)
+// AMDGCN-NEXT: [[COERCE_DIVE2:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr addrspace(5) [[U1_ASCAST]], i32 0, i32 0
+// AMDGCN-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr addrspace(5) [[COERCE_DIVE2]], align 8
+// AMDGCN-NEXT: call void @FuncOneMember(<2 x i32> [[TMP0]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
@@ -308,12 +314,12 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[U_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[U_ADDR]] to ptr
// AMDGCN-NEXT: store ptr addrspace(1) [[U]], ptr [[U_ADDR_ASCAST]], align 8
// AMDGCN-NEXT: [[TMP0:%.*]] = load ptr addrspace(1), ptr [[U_ADDR_ASCAST]], align 8
-// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_KernelOneMemberSpir(ptr addrspace(1) noundef align 8 [[TMP0]]) #[[ATTR4]]
+// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_KernelOneMemberSpir(ptr addrspace(1) noundef align 8 [[TMP0]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
// AMDGCN-LABEL: define dso_local void @__clang_ocl_kern_imp_KernelOneMemberSpir(
-// AMDGCN-SAME: ptr addrspace(1) noundef align 8 [[U:%.*]]) #[[ATTR2]] !kernel_arg_addr_space [[META14]] !kernel_arg_access_qual [[META11]] !kernel_arg_type [[META15]] !kernel_arg_base_type [[META15]] !kernel_arg_type_qual [[META13]] {
+// AMDGCN-SAME: ptr addrspace(1) noundef align 8 [[U:%.*]]) #[[ATTR1]] !kernel_arg_addr_space [[META14]] !kernel_arg_access_qual [[META11]] !kernel_arg_type [[META15]] !kernel_arg_base_type [[META15]] !kernel_arg_type_qual [[META13]] {
// AMDGCN-NEXT: [[ENTRY:.*:]]
// AMDGCN-NEXT: [[U_ADDR:%.*]] = alloca ptr addrspace(1), align 8, addrspace(5)
// AMDGCN-NEXT: [[U_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[U_ADDR]] to ptr
@@ -321,7 +327,7 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[TMP0:%.*]] = load ptr addrspace(1), ptr [[U_ADDR_ASCAST]], align 8
// AMDGCN-NEXT: [[COERCE_DIVE:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER:%.*]], ptr addrspace(1) [[TMP0]], i32 0, i32 0
// AMDGCN-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr addrspace(1) [[COERCE_DIVE]], align 8
-// AMDGCN-NEXT: call void @FuncOneMember(<2 x i32> [[TMP1]]) #[[ATTR4]]
+// AMDGCN-NEXT: call void @FuncOneMember(<2 x i32> [[TMP1]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
@@ -332,19 +338,18 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[U:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN-NEXT: call void @llvm.memcpy.p0.p4.i64(ptr align 8 [[U]], ptr addrspace(4) align 8 [[TMP0]], i64 800, i1 false)
// AMDGCN-NEXT: [[U_ASCAST:%.*]] = addrspacecast ptr [[U]] to ptr addrspace(5)
-// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_KernelLargeOneMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[U_ASCAST]]) #[[ATTR4]]
+// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_KernelLargeOneMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[U_ASCAST]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
// AMDGCN-LABEL: define dso_local void @__clang_ocl_kern_imp_KernelLargeOneMember(
-// AMDGCN-SAME: ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER:%.*]]) align 8 [[TMP0:%.*]]) #[[ATTR2]] !kernel_arg_addr_space [[META10]] !kernel_arg_access_qual [[META11]] !kernel_arg_type [[META16]] !kernel_arg_base_type [[META16]] !kernel_arg_type_qual [[META13]] {
+// AMDGCN-SAME: ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER:%.*]]) align 8 [[TMP0:%.*]]) #[[ATTR1]] !kernel_arg_addr_space [[META10]] !kernel_arg_access_qual [[META11]] !kernel_arg_type [[META16]] !kernel_arg_base_type [[META16]] !kernel_arg_type_qual [[META13]] {
// AMDGCN-NEXT: [[ENTRY:.*:]]
// AMDGCN-NEXT: [[COERCE:%.*]] = alloca [[STRUCT_LARGESTRUCTONEMEMBER]], align 8, addrspace(5)
-// AMDGCN-NEXT: [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_LARGESTRUCTONEMEMBER]], align 8, addrspace(5)
// AMDGCN-NEXT: [[U:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN-NEXT: call void @llvm.memcpy.p0.p5.i64(ptr align 8 [[U]], ptr addrspace(5) align 8 [[TMP0]], i64 800, i1 false)
-// AMDGCN-NEXT: call void @llvm.memcpy.p5.p0.i64(ptr addrspace(5) align 8 [[BYVAL_TEMP]], ptr align 8 [[U]], i64 800, i1 false)
-// AMDGCN-NEXT: call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[BYVAL_TEMP]]) #[[ATTR4]]
+// AMDGCN-NEXT: [[U_ASCAST:%.*]] = addrspacecast ptr [[U]] to ptr addrspace(5)
+// AMDGCN-NEXT: call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[U_ASCAST]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
@@ -359,10 +364,11 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: store <2 x i32> [[U_COERCE0]], ptr [[TMP0]], align 8
// AMDGCN-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U1]], i32 0, i32 1
// AMDGCN-NEXT: store <2 x i32> [[U_COERCE1]], ptr [[TMP1]], align 8
+// AMDGCN-NEXT: [[U1_ASCAST:%.*]] = addrspacecast ptr [[U1]] to ptr addrspace(5)
// AMDGCN-NEXT: store <2 x i32> zeroinitializer, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
// AMDGCN-NEXT: [[TMP2:%.*]] = load <2 x i32>, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
-// AMDGCN-NEXT: [[Y:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U1]], i32 0, i32 1
-// AMDGCN-NEXT: store <2 x i32> [[TMP2]], ptr [[Y]], align 8
+// AMDGCN-NEXT: [[Y:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U1_ASCAST]], i32 0, i32 1
+// AMDGCN-NEXT: store <2 x i32> [[TMP2]], ptr addrspace(5) [[Y]], align 8
// AMDGCN-NEXT: ret void
//
//
@@ -374,11 +380,12 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[U:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN-NEXT: [[DOTCOMPOUNDLITERAL_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCOMPOUNDLITERAL]] to ptr
// AMDGCN-NEXT: call void @llvm.memcpy.p0.p5.i64(ptr align 8 [[U]], ptr addrspace(5) align 8 [[TMP0]], i64 480, i1 false)
+// AMDGCN-NEXT: [[U_ASCAST:%.*]] = addrspacecast ptr [[U]] to ptr addrspace(5)
// AMDGCN-NEXT: store <2 x i32> zeroinitializer, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
// AMDGCN-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
-// AMDGCN-NEXT: [[Y:%.*]] = getelementptr inbounds nuw [[STRUCT_LARGESTRUCTTWOMEMBER]], ptr [[U]], i32 0, i32 1
-// AMDGCN-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [20 x <2 x i32>], ptr [[Y]], i64 0, i64 0
-// AMDGCN-NEXT: store <2 x i32> [[TMP1]], ptr [[ARRAYIDX]], align 8
+// AMDGCN-NEXT: [[Y:%.*]] = getelementptr inbounds nuw [[STRUCT_LARGESTRUCTTWOMEMBER]], ptr addrspace(5) [[U_ASCAST]], i32 0, i32 1
+// AMDGCN-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [20 x <2 x i32>], ptr addrspace(5) [[Y]], i64 0, i64 0
+// AMDGCN-NEXT: store <2 x i32> [[TMP1]], ptr addrspace(5) [[ARRAYIDX]], align 8
// AMDGCN-NEXT: ret void
//
//
@@ -388,16 +395,17 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[COERCE:%.*]] = alloca [[STRUCT_STRUCTTWOMEMBER]], align 8, addrspace(5)
// AMDGCN-NEXT: [[U:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN-NEXT: call void @llvm.memcpy.p0.p4.i64(ptr align 8 [[U]], ptr addrspace(4) align 8 [[TMP0]], i64 16, i1 false)
-// AMDGCN-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U]], i32 0, i32 0
-// AMDGCN-NEXT: [[TMP2:%.*]] = load <2 x i32>, ptr [[TMP1]], align 8
-// AMDGCN-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U]], i32 0, i32 1
-// AMDGCN-NEXT: [[TMP4:%.*]] = load <2 x i32>, ptr [[TMP3]], align 8
-// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_KernelTwoMember(<2 x i32> [[TMP2]], <2 x i32> [[TMP4]]) #[[ATTR4]]
+// AMDGCN-NEXT: [[U_ASCAST:%.*]] = addrspacecast ptr [[U]] to ptr addrspace(5)
+// AMDGCN-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U_ASCAST]], i32 0, i32 0
+// AMDGCN-NEXT: [[TMP2:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP1]], align 8
+// AMDGCN-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U_ASCAST]], i32 0, i32 1
+// AMDGCN-NEXT: [[TMP4:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP3]], align 8
+// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_KernelTwoMember(<2 x i32> [[TMP2]], <2 x i32> [[TMP4]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
// AMDGCN-LABEL: define dso_local void @__clang_ocl_kern_imp_KernelTwoMember(
-// AMDGCN-SAME: <2 x i32> [[U_COERCE0:%.*]], <2 x i32> [[U_COERCE1:%.*]]) #[[ATTR2]] !kernel_arg_addr_space [[META10]] !kernel_arg_access_qual [[META11]] !kernel_arg_type [[META17]] !kernel_arg_base_type [[META17]] !kernel_arg_type_qual [[META13]] {
+// AMDGCN-SAME: <2 x i32> [[U_COERCE0:%.*]], <2 x i32> [[U_COERCE1:%.*]]) #[[ATTR1]] !kernel_arg_addr_space [[META10]] !kernel_arg_access_qual [[META11]] !kernel_arg_type [[META17]] !kernel_arg_base_type [[META17]] !kernel_arg_type_qual [[META13]] {
// AMDGCN-NEXT: [[ENTRY:.*:]]
// AMDGCN-NEXT: [[U:%.*]] = alloca [[STRUCT_STRUCTTWOMEMBER:%.*]], align 8, addrspace(5)
// AMDGCN-NEXT: [[U1:%.*]] = addrspacecast ptr addrspace(5) [[U]] to ptr
@@ -405,11 +413,12 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: store <2 x i32> [[U_COERCE0]], ptr [[TMP0]], align 8
// AMDGCN-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U1]], i32 0, i32 1
// AMDGCN-NEXT: store <2 x i32> [[U_COERCE1]], ptr [[TMP1]], align 8
-// AMDGCN-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U1]], i32 0, i32 0
-// AMDGCN-NEXT: [[TMP3:%.*]] = load <2 x i32>, ptr [[TMP2]], align 8
-// AMDGCN-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U1]], i32 0, i32 1
-// AMDGCN-NEXT: [[TMP5:%.*]] = load <2 x i32>, ptr [[TMP4]], align 8
-// AMDGCN-NEXT: call void @FuncTwoMember(<2 x i32> [[TMP3]], <2 x i32> [[TMP5]]) #[[ATTR4]]
+// AMDGCN-NEXT: [[U1_ASCAST:%.*]] = addrspacecast ptr [[U1]] to ptr addrspace(5)
+// AMDGCN-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U1_ASCAST]], i32 0, i32 0
+// AMDGCN-NEXT: [[TMP3:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP2]], align 8
+// AMDGCN-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U1_ASCAST]], i32 0, i32 1
+// AMDGCN-NEXT: [[TMP5:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP4]], align 8
+// AMDGCN-NEXT: call void @FuncTwoMember(<2 x i32> [[TMP3]], <2 x i32> [[TMP5]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
@@ -420,19 +429,18 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN-NEXT: [[U:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN-NEXT: call void @llvm.memcpy.p0.p4.i64(ptr align 8 [[U]], ptr addrspace(4) align 8 [[TMP0]], i64 480, i1 false)
// AMDGCN-NEXT: [[U_ASCAST:%.*]] = addrspacecast ptr [[U]] to ptr addrspace(5)
-// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_KernelLargeTwoMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTTWOMEMBER]]) align 8 [[U_ASCAST]]) #[[ATTR4]]
+// AMDGCN-NEXT: call void @__clang_ocl_kern_imp_KernelLargeTwoMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTTWOMEMBER]]) align 8 [[U_ASCAST]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//
// AMDGCN-LABEL: define dso_local void @__clang_ocl_kern_imp_KernelLargeTwoMember(
-// AMDGCN-SAME: ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTTWOMEMBER:%.*]]) align 8 [[TMP0:%.*]]) #[[ATTR2]] !kernel_arg_addr_space [[META10]] !kernel_arg_access_qual [[META11]] !kernel_arg_type [[META18]] !kernel_arg_base_type [[META18]] !kernel_arg_type_qual [[META13]] {
+// AMDGCN-SAME: ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTTWOMEMBER:%.*]]) align 8 [[TMP0:%.*]]) #[[ATTR1]] !kernel_arg_addr_space [[META10]] !kernel_arg_access_qual [[META11]] !kernel_arg_type [[META18]] !kernel_arg_base_type [[META18]] !kernel_arg_type_qual [[META13]] {
// AMDGCN-NEXT: [[ENTRY:.*:]]
// AMDGCN-NEXT: [[COERCE:%.*]] = alloca [[STRUCT_LARGESTRUCTTWOMEMBER]], align 8, addrspace(5)
-// AMDGCN-NEXT: [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_LARGESTRUCTTWOMEMBER]], align 8, addrspace(5)
// AMDGCN-NEXT: [[U:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN-NEXT: call void @llvm.memcpy.p0.p5.i64(ptr align 8 [[U]], ptr addrspace(5) align 8 [[TMP0]], i64 480, i1 false)
-// AMDGCN-NEXT: call void @llvm.memcpy.p5.p0.i64(ptr addrspace(5) align 8 [[BYVAL_TEMP]], ptr align 8 [[U]], i64 480, i1 false)
-// AMDGCN-NEXT: call void @FuncLargeTwoMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTTWOMEMBER]]) align 8 [[BYVAL_TEMP]]) #[[ATTR4]]
+// AMDGCN-NEXT: [[U_ASCAST:%.*]] = addrspacecast ptr [[U]] to ptr addrspace(5)
+// AMDGCN-NEXT: call void @FuncLargeTwoMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTTWOMEMBER]]) align 8 [[U_ASCAST]]) #[[ATTR3]]
// AMDGCN-NEXT: ret void
//
//.
>From 027d956764598b9ceb1e70d06722aac786fe9e24 Mon Sep 17 00:00:00 2001
From: Jameson Nash <vtjnash at gmail.com>
Date: Thu, 5 Mar 2026 23:42:34 +0000
Subject: [PATCH 2/3] api cleanup, per review
---
clang/lib/CodeGen/CGBlocks.cpp | 29 ++++---------------------
clang/lib/CodeGen/CGCall.h | 4 +++-
clang/lib/CodeGen/CGCoroutine.cpp | 6 +-----
clang/lib/CodeGen/CGDeclCXX.cpp | 6 ++----
clang/lib/CodeGen/CGExpr.cpp | 9 ++------
clang/lib/CodeGen/CGOpenMPRuntime.cpp | 31 +++++++--------------------
6 files changed, 20 insertions(+), 65 deletions(-)
diff --git a/clang/lib/CodeGen/CGBlocks.cpp b/clang/lib/CodeGen/CGBlocks.cpp
index 1142bf2ecf325..1ce22df11e6a7 100644
--- a/clang/lib/CodeGen/CGBlocks.cpp
+++ b/clang/lib/CodeGen/CGBlocks.cpp
@@ -1934,14 +1934,12 @@ CodeGenFunction::GenerateCopyHelperFunction(const CGBlockInfo &blockInfo) {
QualType ReturnTy = C.VoidTy;
- FunctionArgList args;
auto *DstDecl =
ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
- args.push_back(DstDecl);
auto *SrcDecl =
ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
- args.push_back(SrcDecl);
+ FunctionArgList args{DstDecl, SrcDecl};
const CGFunctionInfo &FI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, args);
@@ -1955,10 +1953,6 @@ CodeGenFunction::GenerateCopyHelperFunction(const CGBlockInfo &blockInfo) {
if (CGM.supportsCOMDAT())
Fn->setComdat(CGM.getModule().getOrInsertComdat(FuncName));
- SmallVector<QualType, 2> ArgTys;
- ArgTys.push_back(C.VoidPtrTy);
- ArgTys.push_back(C.VoidPtrTy);
-
setBlockHelperAttributesVisibility(blockInfo.CapturesNonExternalType, Fn, FI,
CGM);
StartFunction(GlobalDecl(), ReturnTy, Fn, FI, args);
@@ -2133,11 +2127,10 @@ CodeGenFunction::GenerateDestroyHelperFunction(const CGBlockInfo &blockInfo) {
QualType ReturnTy = C.VoidTy;
- FunctionArgList args;
auto *SrcDecl =
ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
- args.push_back(SrcDecl);
+ FunctionArgList args{SrcDecl};
const CGFunctionInfo &FI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, args);
@@ -2151,9 +2144,6 @@ CodeGenFunction::GenerateDestroyHelperFunction(const CGBlockInfo &blockInfo) {
if (CGM.supportsCOMDAT())
Fn->setComdat(CGM.getModule().getOrInsertComdat(FuncName));
- SmallVector<QualType, 1> ArgTys;
- ArgTys.push_back(C.VoidPtrTy);
-
setBlockHelperAttributesVisibility(blockInfo.CapturesNonExternalType, Fn, FI,
CGM);
StartFunction(GlobalDecl(), ReturnTy, Fn, FI, args);
@@ -2403,15 +2393,12 @@ generateByrefCopyHelper(CodeGenFunction &CGF, const BlockByrefInfo &byrefInfo,
QualType ReturnTy = Context.VoidTy;
- FunctionArgList args;
auto *Dst = ImplicitParamDecl::Create(Context, Context.VoidPtrTy,
ImplicitParamKind::Other);
- args.push_back(Dst);
-
auto *Src = ImplicitParamDecl::Create(Context, Context.VoidPtrTy,
ImplicitParamKind::Other);
- args.push_back(Src);
+ FunctionArgList args{Dst, Src};
const CGFunctionInfo &FI =
CGF.CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, args);
@@ -2423,10 +2410,6 @@ generateByrefCopyHelper(CodeGenFunction &CGF, const BlockByrefInfo &byrefInfo,
llvm::Function::Create(LTy, llvm::GlobalValue::InternalLinkage,
"__Block_byref_object_copy_", &CGF.CGM.getModule());
- SmallVector<QualType, 2> ArgTys;
- ArgTys.push_back(Context.VoidPtrTy);
- ArgTys.push_back(Context.VoidPtrTy);
-
CGF.CGM.SetInternalFunctionAttributes(GlobalDecl(), Fn, FI);
CGF.StartFunction(GlobalDecl(), ReturnTy, Fn, FI, args);
@@ -2472,11 +2455,10 @@ generateByrefDisposeHelper(CodeGenFunction &CGF,
ASTContext &Context = CGF.getContext();
QualType R = Context.VoidTy;
- FunctionArgList args;
auto *Src = ImplicitParamDecl::Create(CGF.getContext(), Context.VoidPtrTy,
ImplicitParamKind::Other);
- args.push_back(Src);
+ FunctionArgList args{Src};
const CGFunctionInfo &FI =
CGF.CGM.getTypes().arrangeBuiltinFunctionDeclaration(R, args);
@@ -2489,9 +2471,6 @@ generateByrefDisposeHelper(CodeGenFunction &CGF,
"__Block_byref_object_dispose_",
&CGF.CGM.getModule());
- SmallVector<QualType, 1> ArgTys;
- ArgTys.push_back(Context.VoidPtrTy);
-
CGF.CGM.SetInternalFunctionAttributes(GlobalDecl(), Fn, FI);
CGF.StartFunction(GlobalDecl(), R, Fn, FI, args);
diff --git a/clang/lib/CodeGen/CGCall.h b/clang/lib/CodeGen/CGCall.h
index 145992652934f..fdf17fb530788 100644
--- a/clang/lib/CodeGen/CGCall.h
+++ b/clang/lib/CodeGen/CGCall.h
@@ -372,7 +372,9 @@ class CallArgList : public SmallVector<CallArg, 8> {
/// FunctionArgList - Type for representing both the decl and type
/// of parameters to a function. The decl must be either a
/// ParmVarDecl or ImplicitParamDecl.
-class FunctionArgList : public SmallVector<const VarDecl *, 16> {};
+class FunctionArgList : public SmallVector<const VarDecl *, 16> {
+ using SmallVector::SmallVector;
+};
/// ReturnValueSlot - Contains the address where the return value of a
/// function can be stored, and whether the address is volatile or not.
diff --git a/clang/lib/CodeGen/CGCoroutine.cpp b/clang/lib/CodeGen/CGCoroutine.cpp
index 7f7de687399ab..fac2047d54d63 100644
--- a/clang/lib/CodeGen/CGCoroutine.cpp
+++ b/clang/lib/CodeGen/CGCoroutine.cpp
@@ -423,17 +423,13 @@ CodeGenFunction::generateAwaitSuspendWrapper(Twine const &CoroName,
ASTContext &C = getContext();
- FunctionArgList args;
-
auto *AwaiterDecl =
ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
auto *FrameDecl =
ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
QualType ReturnTy = S.getSuspendExpr()->getType();
- args.push_back(AwaiterDecl);
- args.push_back(FrameDecl);
-
+ FunctionArgList args{AwaiterDecl, FrameDecl};
const CGFunctionInfo &FI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, args);
diff --git a/clang/lib/CodeGen/CGDeclCXX.cpp b/clang/lib/CodeGen/CGDeclCXX.cpp
index 1798d42b69d91..57c6bad67dab9 100644
--- a/clang/lib/CodeGen/CGDeclCXX.cpp
+++ b/clang/lib/CodeGen/CGDeclCXX.cpp
@@ -297,12 +297,11 @@ llvm::Function *CodeGenFunction::createTLSAtExitStub(
CodeGenFunction CGF(CGM);
- FunctionArgList Args;
auto *IPD = ImplicitParamDecl::Create(
CGM.getContext(), CGM.getContext().IntTy, ImplicitParamKind::Other);
- Args.push_back(IPD);
QualType ResTy = CGM.getContext().IntTy;
+ FunctionArgList Args{IPD};
CGF.StartFunction(GlobalDecl(&D, DynamicInitKind::AtExit), ResTy, DtorStub,
FI, Args, D.getLocation(), D.getInit()->getExprLoc());
@@ -1196,11 +1195,10 @@ void CodeGenFunction::GenerateCXXGlobalCleanUpFunc(
llvm::Function *CodeGenFunction::generateDestroyHelper(
Address addr, QualType type, Destroyer *destroyer,
bool useEHCleanupForArray, const VarDecl *VD) {
- FunctionArgList args;
auto *Dst = ImplicitParamDecl::Create(getContext(), getContext().VoidPtrTy,
ImplicitParamKind::Other);
- args.push_back(Dst);
+ FunctionArgList args{Dst};
const CGFunctionInfo &FI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(getContext().VoidTy, args);
llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
index c2dbc375ec7f6..9107553652688 100644
--- a/clang/lib/CodeGen/CGExpr.cpp
+++ b/clang/lib/CodeGen/CGExpr.cpp
@@ -4367,16 +4367,13 @@ void CodeGenFunction::EmitCfiCheckStub() {
ASTContext &C = getContext();
QualType QInt64Ty = C.getIntTypeForBitwidth(64, false);
- FunctionArgList FnArgs;
auto *ArgCallsiteTypeId =
ImplicitParamDecl::Create(C, QInt64Ty, ImplicitParamKind::Other);
auto *ArgAddr =
ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
auto *ArgCFICheckFailData =
ImplicitParamDecl::Create(C, C.VoidPtrTy, ImplicitParamKind::Other);
- FnArgs.push_back(ArgCallsiteTypeId);
- FnArgs.push_back(ArgAddr);
- FnArgs.push_back(ArgCFICheckFailData);
+ FunctionArgList FnArgs{ArgCallsiteTypeId, ArgAddr, ArgCFICheckFailData};
const CGFunctionInfo &FI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, FnArgs);
@@ -4415,14 +4412,12 @@ void CodeGenFunction::EmitCfiCheckFail() {
SanitizerKind::SO_CFIDerivedCast, SanitizerKind::SO_CFIUnrelatedCast,
SanitizerKind::SO_CFIICall},
CheckHandler);
- FunctionArgList Args;
auto *ArgData = ImplicitParamDecl::Create(
getContext(), getContext().VoidPtrTy, ImplicitParamKind::Other);
auto *ArgAddr = ImplicitParamDecl::Create(
getContext(), getContext().VoidPtrTy, ImplicitParamKind::Other);
- Args.push_back(ArgData);
- Args.push_back(ArgAddr);
+ FunctionArgList Args{ArgData, ArgAddr};
const CGFunctionInfo &FI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(getContext().VoidTy, Args);
diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index 7caa71d2b9b99..a99a257c14a2a 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -1085,15 +1085,13 @@ emitCombinerOrInitializer(CodeGenModule &CGM, QualType Ty,
// void .omp_combiner.(Ty *in, Ty *out);
ASTContext &C = CGM.getContext();
QualType PtrTy = C.getPointerType(Ty).withRestrict();
- FunctionArgList Args;
auto *OmpOutParm = ImplicitParamDecl::Create(
C, /*DC=*/nullptr, Out->getLocation(),
/*Id=*/nullptr, PtrTy, ImplicitParamKind::Other);
auto *OmpInParm = ImplicitParamDecl::Create(
C, /*DC=*/nullptr, In->getLocation(),
/*Id=*/nullptr, PtrTy, ImplicitParamKind::Other);
- Args.push_back(OmpOutParm);
- Args.push_back(OmpInParm);
+ FunctionArgList Args{OmpOutParm, OmpInParm};
const CGFunctionInfo &FnInfo =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
llvm::FunctionType *FnTy = CGM.getTypes().GetFunctionType(FnInfo);
@@ -1657,12 +1655,11 @@ llvm::Function *CGOpenMPRuntime::emitThreadPrivateVarDefinition(
// Generate function that re-emits the declaration's initializer into the
// threadprivate copy of the variable VD
CodeGenFunction CtorCGF(CGM);
- FunctionArgList Args;
auto *Dst = ImplicitParamDecl::Create(
CGM.getContext(), /*DC=*/nullptr, Loc,
/*Id=*/nullptr, CGM.getContext().VoidPtrTy, ImplicitParamKind::Other);
- Args.push_back(Dst);
+ FunctionArgList Args{Dst};
const auto &FI = CGM.getTypes().arrangeBuiltinFunctionDeclaration(
CGM.getContext().VoidPtrTy, Args);
llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
@@ -1689,12 +1686,11 @@ llvm::Function *CGOpenMPRuntime::emitThreadPrivateVarDefinition(
// Generate function that emits destructor call for the threadprivate copy
// of the variable VD
CodeGenFunction DtorCGF(CGM);
- FunctionArgList Args;
auto *Dst = ImplicitParamDecl::Create(
CGM.getContext(), /*DC=*/nullptr, Loc,
/*Id=*/nullptr, CGM.getContext().VoidPtrTy, ImplicitParamKind::Other);
- Args.push_back(Dst);
+ FunctionArgList Args{Dst};
const auto &FI = CGM.getTypes().arrangeBuiltinFunctionDeclaration(
CGM.getContext().VoidTy, Args);
llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
@@ -2261,7 +2257,6 @@ static llvm::Value *emitCopyprivateCopyFunction(
SourceLocation Loc) {
ASTContext &C = CGM.getContext();
// void copy_func(void *LHSArg, void *RHSArg);
- FunctionArgList Args;
auto *LHSArg =
ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
@@ -2269,8 +2264,7 @@ static llvm::Value *emitCopyprivateCopyFunction(
auto *RHSArg =
ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
C.VoidPtrTy, ImplicitParamKind::Other);
- Args.push_back(LHSArg);
- Args.push_back(RHSArg);
+ FunctionArgList Args{LHSArg, RHSArg};
const auto &CGFI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
std::string Name =
@@ -3139,15 +3133,13 @@ emitProxyTaskFunction(CodeGenModule &CGM, SourceLocation Loc,
QualType SharedsPtrTy, llvm::Function *TaskFunction,
llvm::Value *TaskPrivatesMap) {
ASTContext &C = CGM.getContext();
- FunctionArgList Args;
auto *GtidArg =
ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
KmpInt32Ty, ImplicitParamKind::Other);
auto *TaskTypeArg = ImplicitParamDecl::Create(
C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
KmpTaskTWithPrivatesPtrQTy.withRestrict(), ImplicitParamKind::Other);
- Args.push_back(GtidArg);
- Args.push_back(TaskTypeArg);
+ FunctionArgList Args{GtidArg, TaskTypeArg};
const auto &TaskEntryFnInfo =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(KmpInt32Ty, Args);
llvm::FunctionType *TaskEntryTy =
@@ -3244,15 +3236,13 @@ static llvm::Value *emitDestructorsFunction(CodeGenModule &CGM,
QualType KmpTaskTWithPrivatesPtrQTy,
QualType KmpTaskTWithPrivatesQTy) {
ASTContext &C = CGM.getContext();
- FunctionArgList Args;
auto *GtidArg =
ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
KmpInt32Ty, ImplicitParamKind::Other);
auto *TaskTypeArg = ImplicitParamDecl::Create(
C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
KmpTaskTWithPrivatesPtrQTy.withRestrict(), ImplicitParamKind::Other);
- Args.push_back(GtidArg);
- Args.push_back(TaskTypeArg);
+ FunctionArgList Args{GtidArg, TaskTypeArg};
const auto &DestructorFnInfo =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(KmpInt32Ty, Args);
llvm::FunctionType *DestructorFnTy =
@@ -3557,7 +3547,6 @@ emitTaskDupFunction(CodeGenModule &CGM, SourceLocation Loc,
QualType SharedsPtrTy, const OMPTaskDataTy &Data,
ArrayRef<PrivateDataTy> Privates, bool WithLastIter) {
ASTContext &C = CGM.getContext();
- FunctionArgList Args;
auto *DstArg = ImplicitParamDecl::Create(
C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, KmpTaskTWithPrivatesPtrQTy,
ImplicitParamKind::Other);
@@ -3567,9 +3556,7 @@ emitTaskDupFunction(CodeGenModule &CGM, SourceLocation Loc,
auto *LastprivArg =
ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, C.IntTy,
ImplicitParamKind::Other);
- Args.push_back(DstArg);
- Args.push_back(SrcArg);
- Args.push_back(LastprivArg);
+ FunctionArgList Args{DstArg, SrcArg, LastprivArg};
const auto &TaskDupFnInfo =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
llvm::FunctionType *TaskDupTy = CGM.getTypes().GetFunctionType(TaskDupFnInfo);
@@ -4979,15 +4966,13 @@ llvm::Function *CGOpenMPRuntime::emitReductionFunction(
ASTContext &C = CGM.getContext();
// void reduction_func(void *LHSArg, void *RHSArg);
- FunctionArgList Args;
auto *LHSArg =
ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
C.VoidPtrTy, ImplicitParamKind::Other);
auto *RHSArg =
ImplicitParamDecl::Create(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
C.VoidPtrTy, ImplicitParamKind::Other);
- Args.push_back(LHSArg);
- Args.push_back(RHSArg);
+ FunctionArgList Args{LHSArg, RHSArg};
const auto &CGFI =
CGM.getTypes().arrangeBuiltinFunctionDeclaration(C.VoidTy, Args);
std::string Name = getReductionFuncName(ReducerName);
>From e5cfaba7e4d66d083f4c289d4033691d8917b1a7 Mon Sep 17 00:00:00 2001
From: Jameson Nash <vtjnash at gmail.com>
Date: Thu, 30 Apr 2026 15:51:50 +0000
Subject: [PATCH 3/3] rebase test fix
---
.../CodeGenOpenCL/addr-space-struct-arg.cl | 64 +++++++++++--------
1 file changed, 36 insertions(+), 28 deletions(-)
diff --git a/clang/test/CodeGenOpenCL/addr-space-struct-arg.cl b/clang/test/CodeGenOpenCL/addr-space-struct-arg.cl
index 9d333ec5cbc6f..6000ba07978cb 100644
--- a/clang/test/CodeGenOpenCL/addr-space-struct-arg.cl
+++ b/clang/test/CodeGenOpenCL/addr-space-struct-arg.cl
@@ -648,6 +648,7 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN20-NEXT: [[IN1:%.*]] = addrspacecast ptr addrspace(5) [[IN]] to ptr
// AMDGCN20-NEXT: [[COERCE_DIVE:%.*]] = getelementptr inbounds nuw [[STRUCT_MAT3X3]], ptr [[IN1]], i32 0, i32 0
// AMDGCN20-NEXT: store [9 x i32] [[IN_COERCE]], ptr [[COERCE_DIVE]], align 4
+// AMDGCN20-NEXT: [[IN1_ASCAST:%.*]] = addrspacecast ptr [[IN1]] to ptr addrspace(5)
// AMDGCN20-NEXT: [[TMP0:%.*]] = load [[STRUCT_MAT4X4]], ptr addrspace(5) [[RETVAL]], align 4
// AMDGCN20-NEXT: ret [[STRUCT_MAT4X4]] [[TMP0]]
//
@@ -697,6 +698,7 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN20-NEXT: [[COERCE:%.*]] = alloca [[STRUCT_MAT32X32]], align 4, addrspace(5)
// AMDGCN20-NEXT: [[IN:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN20-NEXT: call void @llvm.memcpy.p0.p5.i64(ptr align 4 [[IN]], ptr addrspace(5) align 4 [[TMP0]], i64 4096, i1 false)
+// AMDGCN20-NEXT: [[IN_ASCAST:%.*]] = addrspacecast ptr [[IN]] to ptr addrspace(5)
// AMDGCN20-NEXT: ret void
//
//
@@ -745,10 +747,11 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN20-NEXT: [[DOTCOMPOUNDLITERAL_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCOMPOUNDLITERAL]] to ptr
// AMDGCN20-NEXT: [[COERCE_DIVE:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr [[U1]], i32 0, i32 0
// AMDGCN20-NEXT: store <2 x i32> [[U_COERCE]], ptr [[COERCE_DIVE]], align 8
+// AMDGCN20-NEXT: [[U1_ASCAST:%.*]] = addrspacecast ptr [[U1]] to ptr addrspace(5)
// AMDGCN20-NEXT: store <2 x i32> zeroinitializer, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
// AMDGCN20-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
-// AMDGCN20-NEXT: [[X:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr [[U1]], i32 0, i32 0
-// AMDGCN20-NEXT: store <2 x i32> [[TMP0]], ptr [[X]], align 8
+// AMDGCN20-NEXT: [[X:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr addrspace(5) [[U1_ASCAST]], i32 0, i32 0
+// AMDGCN20-NEXT: store <2 x i32> [[TMP0]], ptr addrspace(5) [[X]], align 8
// AMDGCN20-NEXT: ret void
//
//
@@ -760,11 +763,12 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN20-NEXT: [[U:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN20-NEXT: [[DOTCOMPOUNDLITERAL_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCOMPOUNDLITERAL]] to ptr
// AMDGCN20-NEXT: call void @llvm.memcpy.p0.p5.i64(ptr align 8 [[U]], ptr addrspace(5) align 8 [[TMP0]], i64 800, i1 false)
+// AMDGCN20-NEXT: [[U_ASCAST:%.*]] = addrspacecast ptr [[U]] to ptr addrspace(5)
// AMDGCN20-NEXT: store <2 x i32> zeroinitializer, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
// AMDGCN20-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
-// AMDGCN20-NEXT: [[X:%.*]] = getelementptr inbounds nuw [[STRUCT_LARGESTRUCTONEMEMBER]], ptr [[U]], i32 0, i32 0
-// AMDGCN20-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [100 x <2 x i32>], ptr [[X]], i64 0, i64 0
-// AMDGCN20-NEXT: store <2 x i32> [[TMP1]], ptr [[ARRAYIDX]], align 8
+// AMDGCN20-NEXT: [[X:%.*]] = getelementptr inbounds nuw [[STRUCT_LARGESTRUCTONEMEMBER]], ptr addrspace(5) [[U_ASCAST]], i32 0, i32 0
+// AMDGCN20-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [100 x <2 x i32>], ptr addrspace(5) [[X]], i64 0, i64 0
+// AMDGCN20-NEXT: store <2 x i32> [[TMP1]], ptr addrspace(5) [[ARRAYIDX]], align 8
// AMDGCN20-NEXT: ret void
//
//
@@ -808,8 +812,9 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN20-NEXT: [[U1:%.*]] = addrspacecast ptr addrspace(5) [[U]] to ptr
// AMDGCN20-NEXT: [[COERCE_DIVE:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr [[U1]], i32 0, i32 0
// AMDGCN20-NEXT: store <2 x i32> [[U_COERCE]], ptr [[COERCE_DIVE]], align 8
-// AMDGCN20-NEXT: [[COERCE_DIVE2:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr [[U1]], i32 0, i32 0
-// AMDGCN20-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[COERCE_DIVE2]], align 8
+// AMDGCN20-NEXT: [[U1_ASCAST:%.*]] = addrspacecast ptr [[U1]] to ptr addrspace(5)
+// AMDGCN20-NEXT: [[COERCE_DIVE2:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr addrspace(5) [[U1_ASCAST]], i32 0, i32 0
+// AMDGCN20-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr addrspace(5) [[COERCE_DIVE2]], align 8
// AMDGCN20-NEXT: call void @__clang_ocl_kern_imp_KernelOneMember(<2 x i32> [[TMP0]]) #[[ATTR3]]
// AMDGCN20-NEXT: ret void
//
@@ -821,8 +826,9 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN20-NEXT: [[U1:%.*]] = addrspacecast ptr addrspace(5) [[U]] to ptr
// AMDGCN20-NEXT: [[COERCE_DIVE:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr [[U1]], i32 0, i32 0
// AMDGCN20-NEXT: store <2 x i32> [[U_COERCE]], ptr [[COERCE_DIVE]], align 8
-// AMDGCN20-NEXT: [[COERCE_DIVE2:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr [[U1]], i32 0, i32 0
-// AMDGCN20-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[COERCE_DIVE2]], align 8
+// AMDGCN20-NEXT: [[U1_ASCAST:%.*]] = addrspacecast ptr [[U1]] to ptr addrspace(5)
+// AMDGCN20-NEXT: [[COERCE_DIVE2:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTONEMEMBER]], ptr addrspace(5) [[U1_ASCAST]], i32 0, i32 0
+// AMDGCN20-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr addrspace(5) [[COERCE_DIVE2]], align 8
// AMDGCN20-NEXT: call void @FuncOneMember(<2 x i32> [[TMP0]]) #[[ATTR3]]
// AMDGCN20-NEXT: ret void
//
@@ -866,11 +872,10 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN20-SAME: ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER:%.*]]) align 8 [[TMP0:%.*]]) #[[ATTR1]] !kernel_arg_addr_space [[META9]] !kernel_arg_access_qual [[META10]] !kernel_arg_type [[META15]] !kernel_arg_base_type [[META15]] !kernel_arg_type_qual [[META12]] {
// AMDGCN20-NEXT: [[ENTRY:.*:]]
// AMDGCN20-NEXT: [[COERCE:%.*]] = alloca [[STRUCT_LARGESTRUCTONEMEMBER]], align 8, addrspace(5)
-// AMDGCN20-NEXT: [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_LARGESTRUCTONEMEMBER]], align 8, addrspace(5)
// AMDGCN20-NEXT: [[U:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN20-NEXT: call void @llvm.memcpy.p0.p5.i64(ptr align 8 [[U]], ptr addrspace(5) align 8 [[TMP0]], i64 800, i1 false)
-// AMDGCN20-NEXT: call void @llvm.memcpy.p5.p0.i64(ptr addrspace(5) align 8 [[BYVAL_TEMP]], ptr align 8 [[U]], i64 800, i1 false)
-// AMDGCN20-NEXT: call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[BYVAL_TEMP]]) #[[ATTR3]]
+// AMDGCN20-NEXT: [[U_ASCAST:%.*]] = addrspacecast ptr [[U]] to ptr addrspace(5)
+// AMDGCN20-NEXT: call void @FuncOneLargeMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTONEMEMBER]]) align 8 [[U_ASCAST]]) #[[ATTR3]]
// AMDGCN20-NEXT: ret void
//
//
@@ -885,10 +890,11 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN20-NEXT: store <2 x i32> [[U_COERCE0]], ptr [[TMP0]], align 8
// AMDGCN20-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U1]], i32 0, i32 1
// AMDGCN20-NEXT: store <2 x i32> [[U_COERCE1]], ptr [[TMP1]], align 8
+// AMDGCN20-NEXT: [[U1_ASCAST:%.*]] = addrspacecast ptr [[U1]] to ptr addrspace(5)
// AMDGCN20-NEXT: store <2 x i32> zeroinitializer, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
// AMDGCN20-NEXT: [[TMP2:%.*]] = load <2 x i32>, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
-// AMDGCN20-NEXT: [[Y:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U1]], i32 0, i32 1
-// AMDGCN20-NEXT: store <2 x i32> [[TMP2]], ptr [[Y]], align 8
+// AMDGCN20-NEXT: [[Y:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U1_ASCAST]], i32 0, i32 1
+// AMDGCN20-NEXT: store <2 x i32> [[TMP2]], ptr addrspace(5) [[Y]], align 8
// AMDGCN20-NEXT: ret void
//
//
@@ -900,11 +906,12 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN20-NEXT: [[U:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN20-NEXT: [[DOTCOMPOUNDLITERAL_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[DOTCOMPOUNDLITERAL]] to ptr
// AMDGCN20-NEXT: call void @llvm.memcpy.p0.p5.i64(ptr align 8 [[U]], ptr addrspace(5) align 8 [[TMP0]], i64 480, i1 false)
+// AMDGCN20-NEXT: [[U_ASCAST:%.*]] = addrspacecast ptr [[U]] to ptr addrspace(5)
// AMDGCN20-NEXT: store <2 x i32> zeroinitializer, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
// AMDGCN20-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr [[DOTCOMPOUNDLITERAL_ASCAST]], align 8
-// AMDGCN20-NEXT: [[Y:%.*]] = getelementptr inbounds nuw [[STRUCT_LARGESTRUCTTWOMEMBER]], ptr [[U]], i32 0, i32 1
-// AMDGCN20-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [20 x <2 x i32>], ptr [[Y]], i64 0, i64 0
-// AMDGCN20-NEXT: store <2 x i32> [[TMP1]], ptr [[ARRAYIDX]], align 8
+// AMDGCN20-NEXT: [[Y:%.*]] = getelementptr inbounds nuw [[STRUCT_LARGESTRUCTTWOMEMBER]], ptr addrspace(5) [[U_ASCAST]], i32 0, i32 1
+// AMDGCN20-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [20 x <2 x i32>], ptr addrspace(5) [[Y]], i64 0, i64 0
+// AMDGCN20-NEXT: store <2 x i32> [[TMP1]], ptr addrspace(5) [[ARRAYIDX]], align 8
// AMDGCN20-NEXT: ret void
//
//
@@ -914,10 +921,11 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN20-NEXT: [[COERCE:%.*]] = alloca [[STRUCT_STRUCTTWOMEMBER]], align 8, addrspace(5)
// AMDGCN20-NEXT: [[U:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN20-NEXT: call void @llvm.memcpy.p0.p4.i64(ptr align 8 [[U]], ptr addrspace(4) align 8 [[TMP0]], i64 16, i1 false)
-// AMDGCN20-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U]], i32 0, i32 0
-// AMDGCN20-NEXT: [[TMP2:%.*]] = load <2 x i32>, ptr [[TMP1]], align 8
-// AMDGCN20-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U]], i32 0, i32 1
-// AMDGCN20-NEXT: [[TMP4:%.*]] = load <2 x i32>, ptr [[TMP3]], align 8
+// AMDGCN20-NEXT: [[U_ASCAST:%.*]] = addrspacecast ptr [[U]] to ptr addrspace(5)
+// AMDGCN20-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U_ASCAST]], i32 0, i32 0
+// AMDGCN20-NEXT: [[TMP2:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP1]], align 8
+// AMDGCN20-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U_ASCAST]], i32 0, i32 1
+// AMDGCN20-NEXT: [[TMP4:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP3]], align 8
// AMDGCN20-NEXT: call void @__clang_ocl_kern_imp_KernelTwoMember(<2 x i32> [[TMP2]], <2 x i32> [[TMP4]]) #[[ATTR3]]
// AMDGCN20-NEXT: ret void
//
@@ -931,10 +939,11 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN20-NEXT: store <2 x i32> [[U_COERCE0]], ptr [[TMP0]], align 8
// AMDGCN20-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U1]], i32 0, i32 1
// AMDGCN20-NEXT: store <2 x i32> [[U_COERCE1]], ptr [[TMP1]], align 8
-// AMDGCN20-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U1]], i32 0, i32 0
-// AMDGCN20-NEXT: [[TMP3:%.*]] = load <2 x i32>, ptr [[TMP2]], align 8
-// AMDGCN20-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr [[U1]], i32 0, i32 1
-// AMDGCN20-NEXT: [[TMP5:%.*]] = load <2 x i32>, ptr [[TMP4]], align 8
+// AMDGCN20-NEXT: [[U1_ASCAST:%.*]] = addrspacecast ptr [[U1]] to ptr addrspace(5)
+// AMDGCN20-NEXT: [[TMP2:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U1_ASCAST]], i32 0, i32 0
+// AMDGCN20-NEXT: [[TMP3:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP2]], align 8
+// AMDGCN20-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT_STRUCTTWOMEMBER]], ptr addrspace(5) [[U1_ASCAST]], i32 0, i32 1
+// AMDGCN20-NEXT: [[TMP5:%.*]] = load <2 x i32>, ptr addrspace(5) [[TMP4]], align 8
// AMDGCN20-NEXT: call void @FuncTwoMember(<2 x i32> [[TMP3]], <2 x i32> [[TMP5]]) #[[ATTR3]]
// AMDGCN20-NEXT: ret void
//
@@ -954,11 +963,10 @@ kernel void KernelLargeTwoMember(struct LargeStructTwoMember u) {
// AMDGCN20-SAME: ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTTWOMEMBER:%.*]]) align 8 [[TMP0:%.*]]) #[[ATTR1]] !kernel_arg_addr_space [[META9]] !kernel_arg_access_qual [[META10]] !kernel_arg_type [[META17]] !kernel_arg_base_type [[META17]] !kernel_arg_type_qual [[META12]] {
// AMDGCN20-NEXT: [[ENTRY:.*:]]
// AMDGCN20-NEXT: [[COERCE:%.*]] = alloca [[STRUCT_LARGESTRUCTTWOMEMBER]], align 8, addrspace(5)
-// AMDGCN20-NEXT: [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_LARGESTRUCTTWOMEMBER]], align 8, addrspace(5)
// AMDGCN20-NEXT: [[U:%.*]] = addrspacecast ptr addrspace(5) [[COERCE]] to ptr
// AMDGCN20-NEXT: call void @llvm.memcpy.p0.p5.i64(ptr align 8 [[U]], ptr addrspace(5) align 8 [[TMP0]], i64 480, i1 false)
-// AMDGCN20-NEXT: call void @llvm.memcpy.p5.p0.i64(ptr addrspace(5) align 8 [[BYVAL_TEMP]], ptr align 8 [[U]], i64 480, i1 false)
-// AMDGCN20-NEXT: call void @FuncLargeTwoMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTTWOMEMBER]]) align 8 [[BYVAL_TEMP]]) #[[ATTR3]]
+// AMDGCN20-NEXT: [[U_ASCAST:%.*]] = addrspacecast ptr [[U]] to ptr addrspace(5)
+// AMDGCN20-NEXT: call void @FuncLargeTwoMember(ptr addrspace(5) noundef byref([[STRUCT_LARGESTRUCTTWOMEMBER]]) align 8 [[U_ASCAST]]) #[[ATTR3]]
// AMDGCN20-NEXT: ret void
//
//
More information about the cfe-commits
mailing list