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

[James Newling]

newling wrote:

That's a good example of 'getting stuck' @Groverkss. I agree that there are arguments in all directions (no canonicalize vs canonicalize to transpose vs canonicalize to shape_cast). A more advanced canonicalization with multiple 'bubbling' stages (try to get order `shape_cast < broadcast < transpose` then try to get `transpose < shape_cast < broadcast` etc) would result in some nice folding opportunities. 

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