[Mlir-commits] [mlir] [vector][mlir] Canonicalize to shape_cast where possible (PR #140583)

[Mehdi Amini]

joker-eph wrote:

>  @ftynse 's talk on “Normal Forms for MLIR” from last week (*) offers a promising direction,

I talked to @ftynse after the talk, and it seems that canonicalization is orthogonal to this concept of normal form, and I really don't see how canonicalization as a concept would go away.

> However, in the absence of a clear technical criterion (“energy function”) for preferring one form over another, much of this comes down to preference.

This is not a rebuttal of canonicalization: we can try to apply principles to pick one or another form, but if at the end of everything two forms are completely equivalent, picking one is just a convention (e.g. "constants are moved to the right of commutative operations").


https://github.com/llvm/llvm-project/pull/140583