[llvm] [X86] Refine X86::isOffsetSuitableForCodeModel() (PR #75641)

[Reid Kleckner via llvm-commits]

https://github.com/rnk requested changes to this pull request.

I think it would be better to first make sure we use RIP-relative addressing in all the cases we want, and then to make focus on whether the symbolic displacements are correctly combined or not. Ultimately, we want this function to return true as often as possible in as many code models as possible.

https://github.com/llvm/llvm-project/pull/75641