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

[Pierre van Houtryve via cfe-commits]

Pierre-vh wrote:

> The change introduces new builtins and intrinsics, implements them end-to-end, and documents their semantics. It's a complete self-contained patch. 

At a glance it looks like asyncmark/asyncwait could have been a separate patch.

> I don't see what value is added by splitting this into smaller patches, just for the sake of smallness.

It's easier to review, even if everything lands all at once. e.g. in this patch, there's ~2k lines, and in my first read through I missed the entirety of the `InsertWaitCnt` changes because GitHub hid it due to size, and it was sandwiched between two larger files.

In any case, I don't want to side track the review too much. It's fine for this review, this is a big patch, but it's no so big that splitting it is a blocker for review in my book. Just keep it in mind next time and try to split things up a bit more when there is an opportunity to do so. It really helps with reviews :smile: 

> What's missing from the overview presented by the patch description?

The new builtins like `__builtin_amdgcn_raw_ptr_buffer_load_async_lds`

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