don't copy weight gradients in rnn #12600
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
soumith
approved these changes
Oct 12, 2018
Contributor
|
test failures are unrelated |
Contributor
facebook-github-bot
left a comment
facebook-github-bot
left a comment
There was a problem hiding this comment.
soumith is landing this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
Member
|
@pytorchbot retest this please |
zdevito
pushed a commit
to zdevito/ATen
that referenced
this pull request
Oct 12, 2018
Summary: This PR gets rid of unnecessary copy of weight gradients in cudnn rnn. Also removes unnecessary check for input size when deciding whether to use persistent rnn, and adds doc string explaining when persistent rnn can be used. cc ezyang Pull Request resolved: pytorch/pytorch#12600 Differential Revision: D10359981 Pulled By: soumith fbshipit-source-id: 0fce11b527d543fabf21e6e9213fb2879853d7fb
apaszke
reviewed
Oct 14, 2018
| if (copy) { | ||
| param_to.copy_(param_from.view_as(param_to)); | ||
| } else { | ||
| param_from.resize_as_(param_to); |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
| 3) input data has dtype ``torch.float16`` | ||
| 4) V100 GPU is used, | ||
| 5) input data is not in ``PackedSequence`` format | ||
| persistent algorithm can be selected to improve performance. |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
facebook-github-bot
pushed a commit
that referenced
this pull request
Aug 18, 2020
Summary: Should close #36428. The cudnn RNN API expects weights to occupy a flat buffer in memory with a particular layout. This PR implements a "speed of light" fix: [`_cudnn_rnn_cast_reflatten`](https://github.com/pytorch/pytorch/pull/42385/files#diff-9ef93b6a4fb5a06a37c562b83737ac6aR327) (the autocast wrapper assigned to `_cudnn_rnn`) copies weights to the right slices of a flat FP16 buffer with a single read/write per weight (as opposed to casting them to FP16 individually then reflattening the individual FP16 weights, which would require 2 read/writes per weight). It isn't pretty but IMO it doesn't make rnn bindings much more tortuous than they already are. The [test](https://github.com/pytorch/pytorch/pull/42385/files#diff-e68a7bc6ba14f212e5e7eb3727394b40R2683) tries a forward under autocast and a backward for the full cross product of RNN options and input/weight/hidden dtypes. As for all FP16list autocast tests, forward output and backward grads are checked against a control where inputs (including RNN module weights in this case) are precasted to FP16 on the python side. Not sure who to ask for review, tagging ezyang and ngimel because Ed wrote this file (almost 2 years ago) and Natalia did the most recent major [surgery](#12600). Side quests discovered: - Should we update [persistent RNN heuristics](https://github.com/pytorch/pytorch/blob/dbdd28207c5cf6c4a35ceb1de0811c4812e8882c/aten/src/ATen/native/cudnn/RNN.cpp#L584) to include compute capability 8.0? Could be another PR but seems easy enough to include. - Many (maybe all?!) the raw cudnn API calls in [RNN.cpp](https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/cudnn/RNN.cpp) are deprecated in cudnn 8. I don't mind taking the AI to update them since my mental cache is full of rnn stuff, but that would be a substantial separate PR. Pull Request resolved: #42385 Reviewed By: zhangguanheng66 Differential Revision: D23077782 Pulled By: ezyang fbshipit-source-id: a2afb1bdab33ba0442879a703df13dc87f03ec2e
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
6 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR gets rid of unnecessary copy of weight gradients in cudnn rnn. Also removes unnecessary check for input size when deciding whether to use persistent rnn, and adds doc string explaining when persistent rnn can be used. cc @ezyang