[llvm] [LSV] Enhance LoadStoreVectorizer to Handle Disjoint Flag in OR Instructions and Restore Vectorization Opportunities (PR #96495)

[Hao Li via llvm-commits]

================

----------------
LiHao217 wrote:

This is my mistake, the issue is not related to the -mtriple=nvptx64-nvidia-cuda option, and it can be removed. The nvptx triple does not break the disjoint inference.

https://github.com/llvm/llvm-project/pull/96495