[llvm] [AMDGPU] Allow hoisting of V_READFIRSTLANE_B32 for uniform operand (PR #178312)

[Sameer Sahasrabuddhe via llvm-commits]

ssahasra wrote:

> > In that respect, I think the new approach is safer.
> 
> I think I do object this 'safer'. This is safe, so far I have not seen a counter example.

And absence of proof has never been considered proof of absence. 

> So, finally we have only that here: the question if it is non-convergent if the operand is uniform. The rest is landed.

No. The distinction between convergent and non-convergent has never been about uniform operands. It is wrong to say that something _became_ non-convergent because its input is uniform. 

https://github.com/llvm/llvm-project/pull/178312