search

tensorflow / mlir-hlo

398 stars 70 forks source link

Add canonicalization for mhlo.dynamic_conv to mhlo.convolution #54

Closed rsuderman closed 1 year ago

rsuderman commented 1 year ago

The case where the padding value is constant for mhlo.convolution can be lowered to mhlo.convolution directly by copying the pad values in.

burmako commented 1 year ago

Hi Rob! Thank you for your contribution and LGTM. We're not yet at a stage where contributions to MHLO can be conveniently done through GitHub (e.g. this repository is read-only), so may I ask you to send this patch as a CL through Critique or as a PR through tensorflow/tensorflow?