-
Notifications
You must be signed in to change notification settings - Fork 22.1k
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
Make CUDNN an alias of MIOPEN for HIP ops #12278
Conversation
@petrex @ashishfarmer There are some tests failing for the MIOPEN conv ops, could you help taking a look? error message, detailed jenkins log |
@farmer, Ashish<mailto:Ashish.Farmer@amd.com> has a PR for group conv? Maybe we need to land that one first?
From: bddppq <notifications@github.com>
Reply-To: pytorch/pytorch <reply@reply.github.com>
Date: Wednesday, October 3, 2018 at 10:57 AM
To: pytorch/pytorch <pytorch@noreply.github.com>
Cc: "Yeh, Peter" <Peter.Yeh@amd.com>, Mention <mention@noreply.github.com>
Subject: Re: [pytorch/pytorch] Make CUDNN an alias of MIOPEN for HIP ops (#12278)
@petrex<https://github.com/petrex> @ashishfarmer<https://github.com/ashishfarmer> There are some tests failing for the MIOPEN conv ops, could you help taking a look? error message<https://gist.github.com/bddppq/9d8db91f2d77ce58e4668cdea1b32739>, detailed jenkins log<https://ci.pytorch.org/jenkins/job/caffe2-builds/job/py2-clang7-rocmdeb-ubuntu16.04-test/33/consoleFull>
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub<#12278 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AAqt_63zlfYfh7EFk_wWTN5G0oi9bpWcks5uhPpwgaJpZM4XFTA0>.
|
This is a regression that was caused by my earlier check in for group conv optimizations. The changes in PR#12273 will fix this issue. Could we land that one first please |
@ashishfarmer @petrex ah I see. landing #12273 |
9ed9f5f
to
dc4efa1
Compare
@petrex @ashishfarmer there is a new error in test_convolution_layout https://gist.github.com/bddppq/b40b12637acb9b2554412525d1e78094 |
I am guessing { name: "order" s: "NHWC" }
@farmer, Ashish<mailto:Ashish.Farmer@amd.com> Can you pls take a look ? thx
From: bddppq <notifications@github.com>
Reply-To: pytorch/pytorch <reply@reply.github.com>
Date: Wednesday, October 3, 2018 at 3:50 PM
To: pytorch/pytorch <pytorch@noreply.github.com>
Cc: "Yeh, Peter" <Peter.Yeh@amd.com>, Mention <mention@noreply.github.com>
Subject: Re: [pytorch/pytorch] Make CUDNN an alias of MIOPEN for HIP ops (#12278)
@petrex<https://github.com/petrex> @ashishfarmer<https://github.com/ashishfarmer> there is a new error in test_convolution_layout https://gist.github.com/bddppq/b40b12637acb9b2554412525d1e78094
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub<#12278 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AAqt_2MFry3swnEKm_HfOKLQds73i0wKks5uhT8hgaJpZM4XFTA0>.
|
@bddppq, It is trying to run NHWC using MIOpen. Currently we don't support NHWC layout in MIOpen.
similar to the checks in spatial_bn_op_test.py |
@ashishfarmer Thanks for looking! Could you take this PR over to fix all the failed miopen tests? |
Sure thing. Will check and enable appropriate tests for miopen |
@pytorchbot retest this please |
engine_list.append('CUDNN') | ||
if _run_in_hip(gc, dc): | ||
if order == 'NCHW': | ||
engine_list.append('CUDNN') |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
@bddppq, triggered tests after the commits: https://ci.pytorch.org/jenkins/job/caffe2-builds/job/py2-clang7-rocmdeb-ubuntu16.04-trigger-test/1015/ |
failing test : operator_test.rnn_cell_test.RNNCellTest.test_lstm_with_dot_attention_different_dim (from pytest) IIRC , miopen does not support attention in lstm at this moment. Shall we skip this test? |
This test doesn't use miopen implementation of LSTM. |
Triggered tests: https://ci.pytorch.org/jenkins/job/caffe2-builds/job/py2-clang7-rocmdeb-ubuntu16.04-trigger-test/2115/ @bddppq, the PR is ready for review |
@@ -19,6 +19,9 @@ | |||
import unittest | |||
import os | |||
|
|||
def _run_in_hip(gc, dc): |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
def test_spatialbn_test_mode_1d( | ||
self, size, input_channels, batch_size, seed, order, epsilon, | ||
inplace, engine, gc, dc): | ||
# Currently MIOPEN SpatialBN only supports 2D | ||
if _run_in_hip(gc, dc) and engine != "": |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
def test_spatialbn_train_mode_gradient_check_1d( | ||
self, size, input_channels, batch_size, seed, order, epsilon, | ||
momentum, engine, gc, dc): | ||
# Currently MIOPEN SpatialBN only supports 2D | ||
if _run_in_hip(gc, dc) and engine != "": | ||
return |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
def test_spatialbn_test_mode_3d( | ||
self, size, input_channels, batch_size, seed, order, epsilon, | ||
inplace, engine, gc, dc): | ||
# Currently MIOPEN SpatialBN only supports 2D | ||
if _run_in_hip(gc, dc) and engine != "": | ||
return |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
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.
bddppq has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.
This is mostly for reusing all the cudnn test cases in our python operator_tests.