TaoLv commented on issue #12265: update mkldnn to 0.16 URL: https://github.com/apache/incubator-mxnet/pull/12265#issuecomment-414519491 Hi Alex, thanks for bring this up. So far as I know, mkl-dnn introduced in padding format for better performance recently and it would require more memory for computation, compared with the logical size of tensor. In the existing integration code, we are re-using those memory from input/output NDArray for computation and they are allocated according to the logical size during memory planning phase. So I don't think we can simply update mkl-dnn to 0.16 at this time. I have talked with @zheng-da and @eric-haibin-lin about that and we are also trying to figure out a work around before we can change the behavior of memory planning for mkldnn backend. @pengzhao-intel
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
