Skip to content

Basic cuDNN v3 support (update) #3160

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.

Sign up for GitHub

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

Jump to bottom
Merged
merged 1 commit into from
Oct 16, 2015
Merged

Basic cuDNN v3 support (update) #3160

merged 1 commit into from
Oct 16, 2015

Conversation

shelhamer
Copy link
Member

This is the same as #2737 except for

  • Restoring the pooling layer fallback for max + argmax output when the layer is configured to have two tops. The padding is dropped since it is now supported by cuDNN pooling.
  • Clearing the warnings for unused variables in cuDNNConvolutionLayer::Forward_gpu() now that algo and workspace are determined in Reshape().

Thanks @slayton58 for the integration.

@shelhamer shelhamer mentioned this pull request Oct 6, 2015
Closed
@slayton58
Copy link
Contributor

@shelhamer Regarding the test failure for Groups -- it seems like this->weight_offset_ for the cuDNN routines is getting set incorrectly / not set (either way, it's wrong!) This seems to have been introduced in 9d8206e

Setting it back explicitly to:
this->weight_offset_ = (this->num_output_ / this->group_) * (this->channels_ / this->group_) * kernel_h * kernel_w;

in CuDNNConvolutionLayer::Setup seems to fix the issue, not sure if there's a better way - let me know and I'll update the PR

shelhamer added a commit that referenced this pull request Oct 16, 2015
@shelhamer
Merge pull request #3160 from shelhamer/cudnnV3
321720d 
Basic cuDNN v3 support
@shelhamer shelhamer merged commit 321720d into BVLC:master Oct 16, 2015
@shelhamer shelhamer deleted the cudnnV3 branch October 16, 2015 03:17
@ronghanghu
Copy link
Member

Great 👍

@shelhamer
Copy link
Member Author

cuDNN v3 is not itself backward compatible with v2, so adopting v3 in this PR does deprecate v2. We plan to follow the latest cuDNN version in master but keep compatability as the cuDNN interface itself allows.

@shelhamer shelhamer mentioned this pull request Oct 19, 2015
Merged
@ronghanghu ronghanghu mentioned this pull request Oct 28, 2015
Open
ronghanghu
ronghanghu reviewed Oct 28, 2015
View reviewed changes
src/caffe/layers/cudnn_conv_layer.cu
Copy link
Member

Choose a reason for hiding this comment

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

@shelhamer @slayton58 I am a bit confused here... Why do we need to zero out the diff? It confuses me as parameter gradients should be accumulated.

Copy link
Member Author

Choose a reason for hiding this comment

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

This is definitely a bug -- thanks for the fix in #3254.

@ronghanghu ronghanghu mentioned this pull request Oct 28, 2015
Merged
@lukeyeager lukeyeager mentioned this pull request Nov 12, 2015
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

ready for review speed-up

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@shelhamer @slayton58 @ronghanghu