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

[Sameer Sahasrabuddhe via cfe-commits]

ssahasra wrote:

> @ssahasra Does the stuff I wrote above match your understanding of what we're going for, and will it help update the documentation?

@krzysz00, sorry I didn't realize you might be waiting for feedback. Yes, what you wrote matches our intentions. I have the updated text up for review now. There was another issue with my text, because "program order" creates the wrong picture.

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