[Mlir-commits] [mlir] [mlir][vector] Don't fold in_bounds for negative constant indices (PR #219681)

[Alessandro Potenza]

alepot55 wrote:

That covers it. Naming `ShapedType::kDynamic` and `INT64_MIN` is better than what I asked for, because it makes the reason greppable rather than something a reader has to reconstruct from the guard fifteen lines up.

@banach-space, to close the loop on your condition: everything I raised is either addressed or answered. The `kDynamic` comment is in, and the two suggestions I made about sharing the guards with #215340 and matching its `maxStart < 0` bail were declined with better reasons than the suggestions had, which I agreed with above. Nothing of mine is outstanding.


https://github.com/llvm/llvm-project/pull/219681