[clang] [llvm] [AMDGPU][GFX1250] Support asyncmark builtin (PR #205697)

[Shilei Tian via cfe-commits]

================
@@ -299,9 +299,9 @@ def __builtin_amdgcn_av_store_b128
 // Async mark builtins.
 //===----------------------------------------------------------------------===//
 
-// FIXME: Not supported on GFX12 yet. Will need a new feature when we do.
-def __builtin_amdgcn_asyncmark : AMDGPUBuiltin<"void()", [], "vmem-to-lds-load-insts">;
-def __builtin_amdgcn_wait_asyncmark : AMDGPUBuiltin<"void(_Constant unsigned short)", [], "vmem-to-lds-load-insts">;
+// Mirrors GCNSubtarget::hasAsyncMark()
+def __builtin_amdgcn_asyncmark : AMDGPUBuiltin<"void()", [], "vmem-to-lds-load-insts|asynccnt">;
+def __builtin_amdgcn_wait_asyncmark : AMDGPUBuiltin<"void(_Constant unsigned short)", [], "vmem-to-lds-load-insts|asynccnt">;
----------------
shiltian wrote:

Do I understand it correctly that `vmem-to-lds-load-insts` is not supported by gfx12, but `asynccnt` is?

https://github.com/llvm/llvm-project/pull/205697