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

[Ryan Mitchell 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">;
----------------
RyanRio wrote:

Yeah exactly, gfx1250 has the global_load_async_to_lds intrinsics (asynccnt)

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