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

[Shilei Tian via cfe-commits]

shiltian wrote:

> This "convention" that intrinsics should correspond to instructions is severely outdated. We have moved beyond that, where we use intrinsics to represent semantics rather than instructions.

> I am strongly opposed to any use of that legacy convention on that.

> I strongly agree too. We should do what's best for the compiler and UX, not follow some arbitrary convention from the past.

Not specific to this `.async` case, but regarding the general intrinsic convention, I don't agree. LLVM intrinsics are not intended to provide a nice library-like interface, and the backend is not the right place to do the heavy lifting. There should be only a very limited number of exceptions. Otherwise, instruction selection would become unmanageable. We have not moved in that direction so far, and I don't think we ever should.

As for this specific `.async` case, I don't have a particularly strong opinion from the perspective of intrinsic naming conventions.

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