[GitHub] [incubator-mxnet] MoisesHer opened a new pull request #17002: Multi_sum_sq review, AtomicAdd removal

[GitBox <gi...@apache.org>]

MoisesHer opened a new pull request #17002: Multi_sum_sq review, AtomicAdd removal
URL: https://github.com/apache/incubator-mxnet/pull/17002
 
 
   ## Description ##
   Modified multi_sum_sq operator to avoid nondeterministic behavior, which was potentially caused by AtomicAdd operation on GPU kernel.
     
   ## Checklist ##
   ### Essentials ###
   Please feel free to remove inapplicable items for your PR.
   
   - [x] Changes are complete (i.e. I finished coding on this PR)
   - [x] All changes have test coverage: an specific test for multi_sum_sq operator has been included
   - [x] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change
   
   ### Changes ###
   - [x] Modified GPU implementation of multi_sum_sq operator. The operator is launching several CUDA blocks per tensor. In the previous version, an atomic operation was used to reduce the sumSQ coming from different tensors. In this new version, we instead use a temporal storage space to write the reduction from each block, and we launch a second kernel to reduce those.
   
   

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


With regards,
Apache Git Services