[clang] [llvm] [AMDGPU] Add async variants of tensor load/store LDS intrinsics. (PR #200775)

[Shilei Tian via cfe-commits]

================
@@ -959,6 +959,8 @@ def __builtin_amdgcn_ds_atomic_barrier_arrive_rtn_b64 : AMDGPUBuiltin<"long int(
 
 def __builtin_amdgcn_tensor_load_to_lds : AMDGPUBuiltin<"void(_ExtVector<4, unsigned int>, _ExtVector<8, int>, _ExtVector<4, int>, _ExtVector<4, int>, _ExtVector<8, int>, _Constant int)", [Const], "gfx1250-insts">;
 def __builtin_amdgcn_tensor_store_from_lds : AMDGPUBuiltin<"void(_ExtVector<4, unsigned int>, _ExtVector<8, int>, _ExtVector<4, int>, _ExtVector<4, int>, _ExtVector<8, int>, _Constant int)", [Const], "gfx1250-insts">;
+def __builtin_amdgcn_tensor_load_async_to_lds : AMDGPUBuiltin<"void(_ExtVector<4, unsigned int>, _ExtVector<8, int>, _ExtVector<4, int>, _ExtVector<4, int>, _ExtVector<8, int>, _Constant int)", [Const], "gfx1250-insts">;
----------------
shiltian wrote:

add documentation pls https://llvm.org/docs/AMDGPU/DeveloperGuideline.html#document-new-builtins

also, I'd prefer to use feature check instead of `gfx1250-ints`.

https://github.com/llvm/llvm-project/pull/200775