[llvm] [WIP][AMDGPU] Enable "amdgpu-uniform-intrinsic-combine" pass in pipeline. (PR #128687)

[Pankaj Dwivedi via llvm-commits]

PankajDwivedi-25 wrote:

> Why do we need this in the codegen pipeline as well as the optimization pipeline?

Right, at this point, we don't have any concrete proof for that. Its effectiveness is entirely dependent on experimentation. If it doesn’t prove beneficial, the pass will be disabled by default in the backend pipeline, but can still be optionally enabled via a command-line flag if needed later.

https://github.com/llvm/llvm-project/pull/128687