Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

CuDNN improvements #4864

merged 8 commits into from Mar 29, 2018


Copy link

commented Mar 28, 2018

Fixes: #4863


  • Convolution helper: does manual input copy to handle the fact that CuDNN doesn't support SAME mode padding (i.e., asymmetric padding)

General improvements here:

  • Much more informative errors for CudnnConvolutionHelper
  • Adds option (enabled by default, disable-able via config) to fall back to built-in CNN/Subsampling implementation when CuDNN fails
  • Add CuDNN fall-back to IMPLICIT_GEMM mode if forward algorithm cannot be inferred during convolution layer forward pass (logs this once)

Tests are now passing on CuDNN (with 1 exception):

Note that the separable conv failure is noted here: #4862

@AlexDBlack AlexDBlack force-pushed the ab_4863_cudnn branch from 2f45b76 to 04618b2 Mar 29, 2018

AlexDBlack added 2 commits Mar 29, 2018

@AlexDBlack AlexDBlack merged commit de39473 into master Mar 29, 2018

0 of 2 checks passed

codeclimate Code Climate is analyzing this code.
continuous-integration/jenkins/pr-merge This commit is being built

@AlexDBlack AlexDBlack deleted the ab_4863_cudnn branch Mar 29, 2018

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
None yet
1 participant
You can’t perform that action at this time.