You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@mxnet.apache.org by GitBox <gi...@apache.org> on 2018/11/30 03:10:39 UTC

[GitHub] pengzhao-intel commented on issue #13409: [MXNET-1234] Fix shape inference problems in Activation backward

pengzhao-intel commented on issue #13409: [MXNET-1234] Fix shape inference problems in Activation backward
URL: https://github.com/apache/incubator-mxnet/pull/13409#issuecomment-443076037
 
 
   @larroy Thanks for the fix :) The current solution can work and resolve the problem. 
   
   However,  two things I want to highlight:
   * Divergence 
     this kind of workaround will lead divergent codes which are hard to understand and maintain later
   * Performance degradation
     As @TaoLv mentioned, definitely, the performance will drop in CUXXMKL build compared with pure MKL build. And it's really hard to debug for end-user too.
   
   So, I think we can work together to make
   * An testcase/example to reproduce the issue. 
      We still need to futher analysis and see if there're other solutions (maybe inside MKLDNN OP).
   * A plan for the executor improvements
      Agree @larroy's comments, this would be a big refactor or chanllange so it's better to have a plan earily. 
   
   Feel free to correct me :)

----------------------------------------------------------------
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:
users@infra.apache.org


With regards,
Apache Git Services