Upgrade 3rdparty/openmp to release_90 version

[leezu]

Description

Fixes #10856

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

• 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 https://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html
• To the best of my knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

Changes

• Upgrade 3rdparty/openmp to release_90 version

Comments

See #10856 for a discussion on the symptom. It seems it's due to a bug in llvm openmp fixed in the meantime.

The previously failing assertion is still part of the release_90 version. It is at https://github.com/llvm-mirror/openmp/blob/release_90/runtime/src/kmp_runtime.cpp#L6616
But the assertion does not fail anymore (based on my test with cmake -DUSE_CUDA=0 -DUSE_MKLDNN=0 -DCMAKE_EXPORT_COMPILE_COMMANDS=1 -DBUILD_CYTHON_MODULES=1 -DUSE_OPENCV=1 -DCMAKE_BUILD_TYPE=Debug .., which leads to assertion failure without this PR: #10856 (comment))

You can compare the lines in blame mode and conclude they are the same

https://github.com/llvm-mirror/openmp/blame/37c72127e90360a020f351f18d9cccfc30e5145a/runtime/src/kmp_runtime.cpp#L6481
https://github.com/llvm-mirror/openmp/blame/release_90/runtime/src/kmp_runtime.cpp#L6616

[cjolivier01]

Im fine with this after we can resolve that there’s no bug in mxnet which is causing the assert which led to this PR.
Also would be interesting to note if the assert is still there in this version and if not, why was it removed (ie git blame PR)?

[leezu]

Please see the last line of the PR description above. Given that the assert is still there and it is not failing anymore I think we can establish that there is no bug in MXNet. Thus I can't follow your reasoning that there's a bug in MXNet.
I think it's more likely that there was a bug in the 2 year old openmp version.

Then please rescind your veto or give some evidence that there is a bug. Currently I consider this to be a veto without technical justification.

Thanks!

[leezu]

Force push to retrigger CI. Failed with Github API error for 5/11 jobs (other 6 passed):

Pull request #17012 opened

Connecting to https://api.github.com using anirudh2290/****** (anirudh2290/****** (Anirudh Subramanians personal access token to checkout the public MXNet repository. https://github.com/settings/tokens/324757639))

ERROR: Error while retrieving pull request 17012 merge hash : org.kohsuke.github.HttpException: Server returned HTTP response code: -1, message: 'null' for URL: https://api.github.com/repos/apache/incubator-mxnet/commits/83a782e27e6a6b1f9d093a374ef042ba1994ea79

Finished: FAILURE

CC @marcoabreu

[leezu]

Chris rescinded his veto: https://lists.apache.org/thread.html/fbb9c65977afcea064245f7d72bc15a1638fa6d5dcd42c50dba3c701%40%3Cdev.mxnet.apache.org%3E

Thus going ahead and merging this.

[cjolivier01]

i think referring to asking for more information before approving as a “veto” is a bit over-dramatic

…

On Mon, Dec 9, 2019 at 12:19 AM Leonard Lausen ***@***.***> wrote: Chris rescinded his veto: https://lists.apache.org/thread.html/fbb9c65977afcea064245f7d72bc15a1638fa6d5dcd42c50dba3c701%40%3Cdev.mxnet.apache.org%3E Thus going ahead and merging this. — You are receiving this hunk because your review was requested. Reply to this email directly, view it on GitHub <#17012>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACVWZ7JGFBQT5DY3MC7FIDTQXX5PNANCNFSM4JXZLT2A> .

[leezu]

Sorry, that wasn't my intention. I just wanted to reference your approval on the mailinglist.