[llvm] [AMDGPU] ISel for @llvm.amdgcn.cs.chain intrinsic (PR #68186)

[via llvm-commits]

https://github.com/ruiling approved this pull request.

I think the intrinsic here is quite special which makes us to lower it in a different way from ordinary intrinsic. I think it works quite like `invoke`, which acts as kind of indirect call with additional custom operations being capture in extra operands and with custom calling convention. So I think it sounds reasonable to lower it in a similar way as `invoke`. The change LGTM overall. Please wait some time to see if there are further concerns.

https://github.com/llvm/llvm-project/pull/68186