[Clang][OpenCL][AMDGPU] Allow a kernel to call another kernel (#115821)

This feature is currently not supported in the compiler.
To facilitate this we emit a stub version of each kernel
function body with different name mangling scheme, and
replaces the respective kernel call-sites appropriately.
    
Fixes https://github.com/llvm/llvm-project/issues/60313
    
D120566 was an earlier attempt made to upstream a solution
for this issue.

---------

Co-authored-by: anikelal <anikelal@amd.com>

1 files changed, 11 insertions, 0 deletions

diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
index 8f9cf96..0154799 100644
--- a/clang/lib/CodeGen/CodeGenModule.cpp
+++ b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -1903,6 +1903,9 @@ static std::string getMangledNameImpl(CodeGenModule &CGM, GlobalDecl GD,
     } else if (FD && FD->hasAttr<CUDAGlobalAttr>() &&
                GD.getKernelReferenceKind() == KernelReferenceKind::Stub) {
       Out << "__device_stub__" << II->getName();
+    } else if (FD && FD->hasAttr<OpenCLKernelAttr>() &&
+               GD.getKernelReferenceKind() == KernelReferenceKind::Stub) {
+      Out << "__clang_ocl_kern_imp_" << II->getName();
     } else {
       Out << II->getName();
     }
@@ -3890,6 +3893,9 @@ void CodeGenModule::EmitGlobal(GlobalDecl GD) {
 
   // Ignore declarations, they will be emitted on their first use.
   if (const auto *FD = dyn_cast<FunctionDecl>(Global)) {
+    if (FD->hasAttr<OpenCLKernelAttr>() && FD->doesThisDeclarationHaveABody())
+      addDeferredDeclToEmit(GlobalDecl(FD, KernelReferenceKind::Stub));
+
     // Update deferred annotations with the latest declaration if the function
     // function was already used or defined.
     if (FD->hasAttr<AnnotateAttr>()) {
@@ -4857,6 +4863,11 @@ CodeGenModule::GetAddrOfFunction(GlobalDecl GD, llvm::Type *Ty, bool ForVTable,
   if (!Ty) {
     const auto *FD = cast<FunctionDecl>(GD.getDecl());
     Ty = getTypes().ConvertType(FD->getType());
+    if (FD->hasAttr<OpenCLKernelAttr>() &&
+        GD.getKernelReferenceKind() == KernelReferenceKind::Stub) {
+      const CGFunctionInfo &FI = getTypes().arrangeGlobalDeclaration(GD);
+      Ty = getTypes().GetFunctionType(FI);
+    }
   }
 
   // Devirtualized destructor calls may come through here instead of via