[llvm] [WIP][AMDGPU] combine uniform AMDGPU lane Intrinsics (PR #116953)

[Sameer Sahasrabuddhe via llvm-commits]

ssahasra wrote:

> No, it could be anything, like a simple `add`. The transitive users may be convergent, but I don't think it really matters. I think there would need to be some kind of convergent operation inserted here to maintain the use position

This still sounds overly conservative. If a speculatable use depends on the value being uniform, then it has crossthread semantics and needs to be marked convergent. Still not seeing a reason to always maintain the position of the use.

https://github.com/llvm/llvm-project/pull/116953