Skip to content

[Caffe2] Add MIOPEN recurrent operator#10840

Closed
rohithkrn wants to merge 7 commits intopytorch:masterfrom
rohithkrn:miopen_rnn_integration
Closed

[Caffe2] Add MIOPEN recurrent operator#10840
rohithkrn wants to merge 7 commits intopytorch:masterfrom
rohithkrn:miopen_rnn_integration

Conversation

@rohithkrn
Copy link
Contributor

The goal of this PR is to enable miopen engine(for hip devices) for recurrent operator and also enable corresponding unit test.
@bddppq @petrex

@bddppq bddppq self-requested a review August 26, 2018 06:07
Copy link
Contributor

@bddppq bddppq left a comment

Choose a reason for hiding this comment

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

Overall looks good. Thanks!

template <typename T>
TensorDescriptors<T>::TensorDescriptors(
size_t n,
std::vector<int>& dim,

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.

This comment was marked as off-topic.


W = np.random.rand(total_sz).astype(np.float32)
self.ws.create_blob("WEIGHT").feed(W, device_option=hu.gpu_do)
self.ws.create_blob("WEIGHT").feed(W, device_option=hu.hip_do if workspace.has_hip_support

This comment was marked as off-topic.

@bddppq
Copy link
Contributor

bddppq commented Aug 26, 2018

@rohithkrn @petrex Looks like miopen RNN doesn't support dropout?

@petrex
Copy link
Contributor

petrex commented Aug 26, 2018 via email

rnnDesc_,
seqLength,
xDesc_->descs(),
InputData(INPUT), //.template data<T>(),./install.sh -icd

This comment was marked as off-topic.

This comment was marked as off-topic.

@bddppq
Copy link
Contributor

bddppq commented Aug 27, 2018

@petrex Is there any plan to add support to dropout in MIOpen?

for input_idx in input_idxs:
self.assertGradientChecks(
hu.gpu_do, op, inputs, input_idx, [0],
device_option, inputs, input_idx, [0],

This comment was marked as off-topic.

Copy link
Contributor

@facebook-github-bot facebook-github-bot left a comment

Choose a reason for hiding this comment

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

bddppq has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

@bddppq bddppq self-requested a review August 27, 2018 20:19
PenghuiCheng pushed a commit to PenghuiCheng/pytorch that referenced this pull request Sep 11, 2018
Summary:
The goal of this PR is to enable miopen engine(for hip devices) for recurrent operator and also enable corresponding unit test.
bddppq petrex
Pull Request resolved: pytorch#10840

Differential Revision: D9518980

Pulled By: bddppq

fbshipit-source-id: 214661e79a47c5dc6b712ef0fba986bd99db051f
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants