[llvm] [CodeGen] Use 128bits for LaneBitmask. (PR #111157)

[Sander de Smalen via llvm-commits]

sdesmalen-arm wrote:

I've found the right way to model this now, avoiding the need for >64 lanes, so will close this PR. Thanks for the feedback!

https://github.com/llvm/llvm-project/pull/111157