[llvm] [NVPTX] Fix lit test issues from masked load/store implementation (PR #169535)

[Aiden Grossman via llvm-commits]

boomanaiden154 wrote:

> Great. From a logistical perspective, can we merge this as soon as the checks are finished and avoid reverting my original change?

You should be able to merge now. AArch64 is currently non-blocking and premerge won't catch any NVPTX issues anyways since we don't have CUDA installed.

https://github.com/llvm/llvm-project/pull/169535