[llvm] SystemZ: Handle gr128 to fp128 copies in copyPhysReg (PR #90861)

[Ulrich Weigand via llvm-commits]

uweigand wrote:

Hi @arsenm, coming back to this I don't see how this is actually needed (any more).  There are no types that can live in both FP128 and GR128, so nothing should ever generate such COPYs.  I guess you ran into this with an early version of the f128 atomic load/store bitcast patch, but the current version splits into two 64-bit copies very early and never trigger this code path either.

Unless I'm missing something here, I'd prefer to revert this PR again rather than keep this as dead code.


https://github.com/llvm/llvm-project/pull/90861