[Mlir-commits] [mlir] [mlir][vector] Fold `in_bounds` for transfers with loop-derived indices (PR #215340)

[Federico Bruzzone]

FedericoBruzzone wrote:

Thanks @dhairyashilRG! 
It makes perfect sense to me, given the absence of an upper bound for scalable vectors.

> With fixed-length SVE codegen my expectation is it'd look like the AVX-512 rows, not the NEON ones, native predication means the win is repeated mask setup, not avoiding per-lane scalarization. Happy to run that if it'd be useful to your study.

You are very kind, but I don't need it. Thanks a lot! 🙏
Unfortunately, I don't have a machine where I can run `sve` tests (only `ssve`). While I believe the patch brings benefits, I would hope that the baseline, when specifying `-msve-vector-bits=<N>`, also performs better than plain SVE.
Is this what you expect too?


https://github.com/llvm/llvm-project/pull/215340