[clang] [llvm] [NVPTX] Add clang builtin for `__nvvm_reflect` intrinsic (PR #81277)

[Joseph Huber via llvm-commits]

================
@@ -1624,8 +1624,9 @@ def int_nvvm_compiler_error :
 def int_nvvm_compiler_warn :
     Intrinsic<[], [llvm_anyptr_ty], [], "llvm.nvvm.compiler.warn">;
 
-def int_nvvm_reflect :
-  Intrinsic<[llvm_i32_ty], [llvm_anyptr_ty], [IntrNoMem], "llvm.nvvm.reflect">;
+def int_nvvm_reflect : 
+  Intrinsic<[llvm_i32_ty], [llvm_ptr_ty], [IntrNoMem], "llvm.nvvm.reflect">,
+  ClangBuiltin<"__nvvm_reflect">;
----------------
jhuber6 wrote:

It seems this broke the NVPTX tests as well. I'll try to check w/ OpenCL, but I know next to nothing about the OpenCL NVPTX support, I was surprised when I leanred that `nxptx64-nvidia-` can be anything other than CUDA.

https://github.com/llvm/llvm-project/pull/81277