[llvm] [AMDGPU]: Rewrite mbcnt_lo/mbcnt_hi to work item ID where applicable (PR #160496)

[Krzysztof Drewniak via llvm-commits]

krzysz00 wrote:

> The AMDGPUCodegenPrepare pass is scheduled only for the higher optimization levels and it isn't included in the -O0 pipeline. Isn't this patch relevant for -O0 compilation?

I'd argue that is the sort of "optimization" that `-O0` can not do

https://github.com/llvm/llvm-project/pull/160496