Mourad Gouicem

Thank you @maayaneh for the proposal. Could you clarify which API you want discussed? From the branch you link to, it is simply using a preprocessor variable (so compile time...

@maayaneh , what is the expectations for conv with post-ops ? We currently allow conversions to f32 between the main operations (e.g. convolution) and post-ops (e.g relu). This can cause...

THank you for the PR @TejaX-Alaghari . Could you share on which platform(s) you validated this change?

> Another option I thought of was using a binary add post_op, as the documentation mentioned that these are applied after output scaling. But that doesn't seem to support dimensions...

Thanks for the clarification. Ideally, oneDNN primitive cache should remove the need to pre-generate all possible shapes or the need to constraint your shapes. When the shapes are not known...

Also it seems that the performance gap is not that huge between all those shapes (when using fixed shapes). When measuring matmul for the shapes you shared on a 16...

> This looks good to me, but I'm not entirely sure what's expected. Should this in fact generate an error in oneDNN, for example? This seems good to me, and...

> Hi @mgouicem , thanks for your comment! The three algorithms 'logistic/soft_relu/elu' show these inaccuracies, and all the test log in attachment. [LastTest.log.zip](https://github.com/oneapi-src/oneDNN/files/9618412/LastTest.log.zip). Thanks for the details. > ACL calculates...

Maybe I am missing something, but I don't see the changes mentioned in your comments (removing amx_tileconfigure headers, printf, ...). Did you push those changes?

One last thing: please fix formatting using [clang-format](https://dev.azure.com/mkldnn/dnnl/_build/results?buildId=1550&view=logs&j=0bf7d65c-1444-5302-7319-c6947569d8d5&t=85a4fd38-810c-57ff-5840-0f66a7da7b77).