PawelGlomski-Intel opened a new pull request #19735: URL: https://github.com/apache/incubator-mxnet/pull/19735
## Description ## Right now, oneDNN may choose a blocked format for the output of concat, in cases that may require additional padding (thus more memory). In (de)convolution we make sure to only select a primitive_desc which has the expected (from tensor size) memory size requirements, so the same has to be done in this case. Unlike convolution's, concat's primitive_desc does not support iteration over multiple implementations, so here I manually impose a plain format when oneDNN would choose a blocked one with padding. Resolves #19586 ## Checklist ## ### Essentials ### - [x] PR's title starts with a category (e.g. [BUGFIX], [MODEL], [TUTORIAL], [FEATURE], [DOC], etc) - [ ] Changes are complete (i.e. I finished coding on this PR) - [ ] All changes have test coverage - [ ] Code is well-documented ### Changes ### - [x] Impose a plain format for concat's output when mkldnn uses padding - [x] Fix and add missing comments related to memory planning and padded formats - [ ] Add/Change tests that cover this case ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
