[clang] [llvm] [AMDGPU] Introduce asyncmark/wait intrinsics (PR #173259)

[Krzysztof Drewniak via cfe-commits]

https://github.com/krzysz00 commented:

Missing test: a combination of async DMA ops and regular `load` operations from global on gfx9

Missing test: a pipelined loop example wheer we show that you don't get conservative waitcnt insertion when using the async intrinsics

https://github.com/llvm/llvm-project/pull/173259