[CUDA][HIP] Set kernel calling convention before arrange function
Currently clang set kernel calling convention for CUDA/HIP after
arranging function, which causes incorrect kernel function type since
it depends on calling convention.
This patch moves setting kernel convention before arranging
function.
Differential Revision: https://reviews.llvm.org/D47733
llvm-svn: 334457
diff --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp
index 2d8339e..064b5fc 100644
--- a/clang/lib/CodeGen/CGCall.cpp
+++ b/clang/lib/CodeGen/CGCall.cpp
@@ -255,6 +255,16 @@
FTP->getCanonicalTypeUnqualified().getAs<FunctionProtoType>(), MD);
}
+/// Set calling convention for CUDA/HIP kernel.
+static void setCUDAKernelCallingConvention(CanQualType &FTy, CodeGenModule &CGM,
+ const FunctionDecl *FD) {
+ if (FD->hasAttr<CUDAGlobalAttr>()) {
+ const FunctionType *FT = FTy->getAs<FunctionType>();
+ CGM.getTargetCodeGenInfo().setCUDAKernelCallingConvention(FT);
+ FTy = FT->getCanonicalTypeUnqualified();
+ }
+}
+
/// Arrange the argument and result information for a declaration or
/// definition of the given C++ non-static member function. The
/// member function must be an ordinary function, i.e. not a
@@ -264,7 +274,9 @@
assert(!isa<CXXConstructorDecl>(MD) && "wrong method for constructors!");
assert(!isa<CXXDestructorDecl>(MD) && "wrong method for destructors!");
- CanQual<FunctionProtoType> prototype = GetFormalType(MD);
+ CanQualType FT = GetFormalType(MD).getAs<Type>();
+ setCUDAKernelCallingConvention(FT, CGM, MD);
+ auto prototype = FT.getAs<FunctionProtoType>();
if (MD->isInstance()) {
// The abstract case is perfectly fine.
@@ -424,6 +436,7 @@
CanQualType FTy = FD->getType()->getCanonicalTypeUnqualified();
assert(isa<FunctionType>(FTy));
+ setCUDAKernelCallingConvention(FTy, CGM, FD);
// When declaring a function without a prototype, always use a
// non-variadic type.