[Mlir-commits] [mlir] [mlir][vector] Canonicalize/fold 'order preserving' transposes (PR #135841)

[Andrzej Warzyński]

https://github.com/banach-space commented:

Thanks, makes sense!

Quite possibly there’s no need to be so strict with "scalable" vectors (thanks for checking!) — let me experiment and confirm, probably tomorrow.

Btw:
* "not of size 1" vs "non-one dimensions" vs "non-one indices"

We tend to say "non-unit dims" 😉 I hope I'm not being overly pedantic by suggesting using that wording in this PR as well. (Also, ESL here — so feel free to suggest something else if you prefer!)

https://github.com/llvm/llvm-project/pull/135841