[Mlir-commits] [mlir] [MLIR][Tensor] Canonicalize fully covering slice insertions into tensors with unit prefixes (PR #92912)

[llvmlistbot at llvm.org]

MaheshRavishankar wrote:

> > . For example, bufferization would not be able to handle the expand_shape properly (without essentially doing this analysis in reverse, essentially undoing this canonicalization).
> 
> This does not seem like an argument against canonicalization to me, unless you'd advocate for the opposite canonicalization?

I think we have the same discussion many times. Canonicalizations should be narrow. Changing one op to a completely different op does not seem like a canonicalization to me.

https://github.com/llvm/llvm-project/pull/92912