[GitHub] [incubator-mxnet] DickJC123 opened a new pull request #14310: Cudnn conv dgrad algo filtering

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

DickJC123 opened a new pull request #14310: Cudnn conv dgrad algo filtering
URL: https://github.com/apache/incubator-mxnet/pull/14310
 
 
   ## Description ##
   I've learned that for cudnn versions in the range [7.3.1,7.5.0), the Convolution dgrad algorithm 3 (CUDNN_CONVOLUTION_BWD_DATA_ALGO_FFT_TILING) may produce incorrect results for some strided convolutions.  This is not something that would generally appear in the current CI, which builds against cuda9.1 and cudnn7.1(so before the problem was introduced).
   
   I've created a test to expose the problem, as well as a fix.  I've noticed that there is a CI build for tensorrt that uses cuda10.0 and cudnn7.4, which should exhibit the problem.  Thus, the plan is:
   
   - Do an initial PR submission that includes the problem-exposing test, and with a temporary addition of that test to the tensorrt CI test suite.
   - If hopefully this demonstrates the problem, then push the fix (which involves filtering-out algo 3 from the results of cudnnFind when necessary).
   - When the fix is shown to work, remove the explicit running of the test from the tensorrt CI, allowing the normal running of the test as part of test_operator_gpu.py to catch any return of the problem.
   
   ## Checklist ##
   ### Essentials ###
   Please feel free to remove inapplicable items for your PR.
   - [ ] The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant [JIRA issue](https://issues.apache.org/jira/projects/MXNET/issues) created (except PRs with tiny changes)
   - [ ] Changes are complete (i.e. I finished coding on this PR)
   - [X] All changes have test coverage:
   - Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
   - Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore)
   - Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL)
   - [X] Code is well-documented: 
   - For user-facing API changes, API doc string has been updated. 
   - For new C++ functions in header files, their functionalities and arguments are documented. 
   - For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable
   - Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html
   - [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 ###
   - [ ] Feature1, tests, (and when applicable, API doc)
   - [ ] Feature2, tests, (and when applicable, API doc)
   
   ## Comments ##
   - If this change is a backward incompatible change, why must this change be made.
   - Interesting edge cases to note here
   

----------------------------------------------------------------
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