[WIP] Fix the unit test for depthwise convolution

[TaoLv]

Description

Should fix #14052 .

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

• The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant JIRA issue created (except PRs with tiny changes)
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage:
• Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
• Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore)
• Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL)
• Code is well-documented:
• For user-facing API changes, API doc string has been updated.
• For new C++ functions in header files, their functionalities and arguments are documented.
• For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable
• Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html
• To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

Changes

• Feature1, tests, (and when applicable, API doc)
• Feature2, tests, (and when applicable, API doc)

Comments

• If this change is a backward incompatible change, why must this change be made.
• Interesting edge cases to note here

[karan6181]

@mxnet-label-bot add [Test, Python, Flaky, pr-work-in-progress]

[TaoLv]

Looks like the issue is still there and flaky. New MKL-DNN doesn't mitigate the problem. @pengzhao-intel @szha

[ZhennanQin]

Looked into the CI result, test_depthwise_convolution can pass now. The failures seem don't relate to test_depthwise_convolution. Please try to retrigger CI.

[TaoLv]

CI was triggered 3 times in this PR and test_depthwise_convolution failed twice in them:
http://jenkins.mxnet-ci.amazon-ml.com/blue/organizations/jenkins/mxnet-validation%2Fwindows-gpu/detail/PR-14452/1/pipeline#step-178-log-2982
http://jenkins.mxnet-ci.amazon-ml.com/blue/organizations/jenkins/mxnet-validation%2Fwindows-gpu/detail/PR-14452/3/pipeline/#step-156-log-2968

As the the second failure happened on python3: GPU Win, I suspect that the issue was not caused by MKL-DNN. Besides seems it's not the same issue descried in #14052 @mseth10 .

[mseth10]

This seems like a new failure and needs to be investigated, it was not there when I raised the issue #14052 .
Are we done with MKLDNN fix? What about the other CI failures?

[TaoLv]

@DickJC123 Could you take a look at the CI failure on Windows GPU? This PR tries to re-enable the test for depthwise convolution.
http://jenkins.mxnet-ci.amazon-ml.com/blue/organizations/jenkins/mxnet-validation%2Fwindows-gpu/detail/PR-14452/4/pipeline#step-178-log-2966
http://jenkins.mxnet-ci.amazon-ml.com/blue/organizations/jenkins/mxnet-validation%2Fwindows-gpu/detail/PR-14452/4/pipeline/110#step-179-log-3000

[pengzhao-intel]

Maybe @ptrendx can help too :)

[piyushghai]

@TaoLv Were you able to figure out the test failures here ?

[karan6181]

@TaoLv Could you please provide an update on this PR? It's being idle for sometime now. Thanks!

[pengzhao-intel]

@TaoLv any plan on this? Still on 0.20 or 1.0?

[roywei]

I see the issue is closed, please close if not required