[CUDA] Set LLVM calling convention for CUDA kernel

Some targets need special LLVM calling convention for CUDA kernel.
This patch does that through a TargetCodeGenInfo hook.

It only affects amdgcn target.

Patch by Greg Rodgers.
Revised and lit tests added by Yaxun Liu.

Differential Revision: https://reviews.llvm.org/D45223

llvm-svn: 330447
diff --git a/clang/lib/CodeGen/TargetInfo.cpp b/clang/lib/CodeGen/TargetInfo.cpp
index 5e842fa..99e4b0d 100644
--- a/clang/lib/CodeGen/TargetInfo.cpp
+++ b/clang/lib/CodeGen/TargetInfo.cpp
@@ -7637,6 +7637,7 @@
                             llvm::Function *BlockInvokeFunc,
                             llvm::Value *BlockLiteral) const override;
   bool shouldEmitStaticExternCAliases() const override;
+  void setCUDAKernelCallingConvention(llvm::Function *F) const override;
 };
 }
 
@@ -7772,6 +7773,11 @@
   return false;
 }
 
+void AMDGPUTargetCodeGenInfo::setCUDAKernelCallingConvention(
+    llvm::Function *F) const {
+  F->setCallingConv(llvm::CallingConv::AMDGPU_KERNEL);
+}
+
 //===----------------------------------------------------------------------===//
 // SPARC v8 ABI Implementation.
 // Based on the SPARC Compliance Definition version 2.4.1.