[Mlir-commits] [mlir] [MLIR] Folding unpack and pack sequence in data layout propagation from padded domain (PR #138332)

[Han-Chung Wang]

https://github.com/hanhanW commented:

> The idea of this patch is to optimize the propagation by never creating the unpack + pack in cases where the padding value does not matter the the op that is being propagated through (for example, most non-reduction ops). If we do not do this, then the pack + unpack canonicalization patterns will not have enough information to know whether the folding is possible, and we will end up with pack + unpack pairs in the final IR.

Thanks for the details! IIUC, this is a special case that the pack op is generated because of the unpack op propagation, so they can be folded away.

LGTM, just one bug and few nits, please take a look.

https://github.com/llvm/llvm-project/pull/138332