Skip to content
This repository was archived by the owner on Nov 17, 2023. It is now read-only.

Fix cuDNN RNN dtype_with_fallback_ bug#16671

Merged
ptrendx merged 1 commit intoapache:masterfrom
stu1130:fix_cudnn_rnn_dtype
Oct 30, 2019
Merged

Fix cuDNN RNN dtype_with_fallback_ bug#16671
ptrendx merged 1 commit intoapache:masterfrom
stu1130:fix_cudnn_rnn_dtype

Conversation

@stu1130
Copy link
Contributor

@stu1130 stu1130 commented Oct 29, 2019 •
edited
Loading

Description

#16670

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

  • The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant JIRA issue created (except PRs with tiny changes)
  • Changes are complete (i.e. I finished coding on this PR)
  • All changes have test coverage:
  • Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
  • Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore)
  • Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL)
  • Code is well-documented:
  • For user-facing API changes, API doc string has been updated.
  • For new C++ functions in header files, their functionalities and arguments are documented.
  • For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable
  • Check the API doc at https://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html
  • 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

@DickJC123

@stu1130 stu1130 mentioned this pull request Oct 29, 2019
Closed
DickJC123
DickJC123 reviewed Oct 29, 2019
View reviewed changes
Copy link
Contributor

@DickJC123 DickJC123 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please replace CUDNN_VERSION with cudnnGetVersion(). Since we only warn users if the linked-against version mismatches the compiled-against version, I think we need to consider that case here. In most cases, the user would just not receive the additional cuDNN functionality. However for the issue this PR is addressing, a user that compiled against cuDNN v7.4, but then ran with cuDNN v7.5 would subtly have their RNNs converted from pseudo-fp16 to true-fp16. Use of cudnnGetVersion() corrects this and delivers consistent pseudo-fp16 behavior. Sorry I didn't think of this when creating the original issue.

@stu1130 stu1130 force-pushed the fix_cudnn_rnn_dtype branch from 32c713f to 2b95500 Compare October 29, 2019 22:55
@DickJC123
Copy link
Contributor

DickJC123 commented Oct 30, 2019

I've selectively retriggered the CI, as I don't think the recent CI failures are related to this PR. The Clojure CPU Integration error was a bit strange, so I've filed it as issue #16674 for someone to look at.

@stu1130
Copy link
Contributor Author

stu1130 commented Oct 30, 2019

Thanks @DickJC123

@DickJC123 DickJC123 requested a review from ptrendx October 30, 2019 23:25
@ptrendx ptrendx merged commit 77e8f51 into apache:master Oct 30, 2019
yajiedesign pushed a commit to yajiedesign/mxnet that referenced this pull request Nov 6, 2019
@stu1130 @yajiedesign
fix cuDNN RNN dtype_with_fallback_ bug (apache#16671)
9ddea08
@samskalicky samskalicky mentioned this pull request Nov 11, 2019
Open
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Reviewers

2 more reviewers

@DickJC123 DickJC123 DickJC123 approved these changes

@ptrendx ptrendx ptrendx approved these changes

Reviewers whose approvals may not affect merge requirements

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@stu1130 @DickJC123 @ptrendx