[clang] [llvm] [AMDGPU] Implement AsyncMark Stages (PR #220442)

[Sameer Sahasrabuddhe via cfe-commits]

ssahasra wrote:

@vporpo wrote:

> the exact rules on what gets marked are not super obvious. Like for example it is not clear what is being marked if there are other instructions in between, or if the asyncmark is at the top of the block and etc.

Can you elaborate? The rules of asyncmarks are abundantly clearly about program order. 
https://llvm.org/docs/AMDGPUAsyncOperations.html#completion-of-asyncmarks

An asyncmark orders every operation that is program-ordered before it. 

https://github.com/llvm/llvm-project/pull/220442