-
Notifications
You must be signed in to change notification settings - Fork 22.2k
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
Remove support for CUDNN 6 #15851
Remove support for CUDNN 6 #15851
Conversation
.circleci/config.yml
Outdated
environment: | ||
JOB_BASE_NAME: pytorch-linux-xenial-cuda8-cudnn6-py3-build | ||
JOB_BASE_NAME: pytorch-linux-xenial-cuda8-cudnn7-py3-build | ||
DOCKER_IMAGE: "308535385114.dkr.ecr.us-east-1.amazonaws.com/pytorch/pytorch-linux-xenial-cuda8-cudnn6-py3:278" |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Docker image is still cudnn6? Are there cuda8+cudnn7 images already pushed to aws?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
the cuda8+cudnn7 docker image is not pushed
I don't know how cudnn is discovered now, but it probably also has to be changed to require minimum 7.0 version. |
@ngimel Added a check for minimum cudnn version. Will verify if it works tomorrow. |
@ngimel verified that minimum version requirement change in the cmake file works:
|
5f6e2ad
to
2b2cb51
Compare
Prepare for pytorch/pytorch#15851. The docker side change was done in pietern/pytorch-dockerfiles@913d15b, but this is needed to make Jenkins really build these dockers.
2b2cb51
to
e6a0168
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Lovely!
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@ezyang is landing this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
Summary: This PR aims to remove support for cuDNN 6. Differential Revision: D13709595 Pulled By: ezyang fbshipit-source-id: 853624db1cf66b0534d7028654c38c2806fb4107
@syed-ahmed btw if you wanted to also delete
that'd be peachy |
Summary: This PR aims to remove support for cuDNN 6. Differential Revision: D13709595 Pulled By: ezyang fbshipit-source-id: 853624db1cf66b0534d7028654c38c2806fb4107
Sounds good! Peachy it is. I'll follow up on that soon. |
Summary: This PR aims to remove support for cuDNN 6. Differential Revision: D13709595 Pulled By: ezyang fbshipit-source-id: 853624db1cf66b0534d7028654c38c2806fb4107
Summary: This PR updates the logic for using cudnnGet* and cudnnFind*. Current version of cudnn find and get (v7) returns a pair of best algorithm and the convDesc mathType. While we were using the returned algorithm, we didn't update the mathType. As a result, we ended up with a slow choice of algorithm and math type. Without this patch, we are seeing a 10x regression in group convolutions. Changelist: - Changed the template arguments to be `perf_t` instead of `algo_t` to unify cudnnFind and cudnnGet. Both cudnnFind and cudnnGet have the same purpose and hence, it made sense to unify them and get rid of `getAlgorithm`. - Used cudnnGet*_v7 everywhere cudnnGet* was being used. - Removed all cudnn6 paths (This PR depends on #15851) Differential Revision: D13787601 Pulled By: ezyang fbshipit-source-id: 81fe86727673d021306fe1c99c3e528b7c9ad17f
Summary: This PR updates the logic for using cudnnGet* and cudnnFind*. Current version of cudnn find and get (v7) returns a pair of best algorithm and the convDesc mathType. While we were using the returned algorithm, we didn't update the mathType. As a result, we ended up with a slow choice of algorithm and math type. Without this patch, we are seeing a 10x regression in group convolutions. Changelist: - Changed the template arguments to be `perf_t` instead of `algo_t` to unify cudnnFind and cudnnGet. Both cudnnFind and cudnnGet have the same purpose and hence, it made sense to unify them and get rid of `getAlgorithm`. - Used cudnnGet*_v7 everywhere cudnnGet* was being used. - Removed all cudnn6 paths (This PR depends on pytorch/pytorch#15851) Differential Revision: D13787601 Pulled By: ezyang fbshipit-source-id: 81fe86727673d021306fe1c99c3e528b7c9ad17f
Summary: This PR updates the logic for using cudnnGet* and cudnnFind*. Current version of cudnn find and get (v7) returns a pair of best algorithm and the convDesc mathType. While we were using the returned algorithm, we didn't update the mathType. As a result, we ended up with a slow choice of algorithm and math type. Without this patch, we are seeing a 10x regression in group convolutions. Changelist: - Changed the template arguments to be `perf_t` instead of `algo_t` to unify cudnnFind and cudnnGet. Both cudnnFind and cudnnGet have the same purpose and hence, it made sense to unify them and get rid of `getAlgorithm`. - Used cudnnGet*_v7 everywhere cudnnGet* was being used. - Removed all cudnn6 paths (This PR depends on #15851) Differential Revision: D13957944 Pulled By: ezyang fbshipit-source-id: a88c39d80ae37f2d686665622302b62b50fab404
Summary: This PR updates the logic for using cudnnGet* and cudnnFind*. Current version of cudnn find and get (v7) returns a pair of best algorithm and the convDesc mathType. While we were using the returned algorithm, we didn't update the mathType. As a result, we ended up with a slow choice of algorithm and math type. Without this patch, we are seeing a 10x regression in group convolutions. Changelist: - Changed the template arguments to be `perf_t` instead of `algo_t` to unify cudnnFind and cudnnGet. Both cudnnFind and cudnnGet have the same purpose and hence, it made sense to unify them and get rid of `getAlgorithm`. - Used cudnnGet*_v7 everywhere cudnnGet* was being used. - Removed all cudnn6 paths (This PR depends on pytorch/pytorch#15851) Differential Revision: D13957944 Pulled By: ezyang fbshipit-source-id: a88c39d80ae37f2d686665622302b62b50fab404
Remove pointer to nonexistent Note. It is already removed in "Remove support for CUDNN 6 (pytorch#15851)"
Summary: Remove pointer to nonexistent Note. It is already removed in "Remove support for CUDNN 6 (pytorch#15851)" Pull Request resolved: pytorch#19148 Differential Revision: D14891514 Pulled By: soumith fbshipit-source-id: dd33cfefa3a21e18afae5b3992dea085adaabda8
This PR aims to remove support for cuDNN 6.
CC: @soumith @ngimel