[NumPy] enable large tensor in np

[szha]

Description

(Brief description on what this PR is about)

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

• The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant JIRA issue created (except PRs with tiny changes)
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage:
• Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
• Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore)
• Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL)
• Code is well-documented:
• For user-facing API changes, API doc string has been updated.
• For new C++ functions in header files, their functionalities and arguments are documented.
• For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable
• Check the API doc at https://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$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

• 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

[mxnet-bot]

Hey @szha , Thanks for submitting the PR
All tests are already queued to run once. If tests fail, you can trigger one or more tests again with the following commands:

• To trigger all jobs: @mxnet-bot run ci [all]
• To trigger specific jobs: @mxnet-bot run ci [job1, job2]

---

CI supported jobs: [centos-gpu, windows-gpu, miscellaneous, clang, unix-cpu, website, unix-gpu, centos-cpu, sanity, edge, windows-cpu]

---

Note:
Only following 3 categories can trigger CI :PR Author, MXNet Committer, Jenkins Admin.
All CI tests must pass before the PR can be merged.

[szha]

@TaoLv @PatricZhao large tensor build for MKL builds seems to be failing. See the CI checks.

[TaoLv]

@szha, do you mean the errors below? As I mentioned in #18645 (comment), we need also enable MKL_USE_ILP64 in cmake line when LTS is enabled. More information, see https://software.intel.com/content/www/us/en/develop/documentation/mkl-macos-developer-guide/top/linking-your-application-with-the-intel-math-kernel-library/linking-in-detail/linking-with-interface-libraries/using-the-ilp64-interface-vs-lp64-interface.html. I'm afraid that similar problem may also exist in mshadow code: https://github.com/apache/incubator-mxnet/blob/master/3rdparty/mshadow/mshadow/dot_engine-inl.h#L317.

[2020-07-03T21:00:50.972Z] /work/mxnet/src/operator/contrib/transformer.cc:143:31: error: narrowing conversion of 'm' from 'mxnet::index_t {aka long int}' to 'int' inside { } [-Werror=narrowing]
[2020-07-03T21:00:50.972Z]    MKL_INT p_m[GROUP_SIZE] = {m};
[2020-07-03T21:00:50.972Z]                                ^
[2020-07-03T21:00:50.972Z] /work/mxnet/src/operator/contrib/transformer.cc:144:31: error: narrowing conversion of 'n' from 'mxnet::index_t {aka long int}' to 'int' inside { } [-Werror=narrowing]
[2020-07-03T21:00:50.972Z]    MKL_INT p_n[GROUP_SIZE] = {n};
[2020-07-03T21:00:50.972Z]                                ^
[2020-07-03T21:00:50.972Z] /work/mxnet/src/operator/contrib/transformer.cc:145:31: error: narrowing conversion of 'k' from 'mxnet::index_t {aka long int}' to 'int' inside { } [-Werror=narrowing]
[2020-07-03T21:00:50.972Z]    MKL_INT p_k[GROUP_SIZE] = {k};
[2020-07-03T21:00:50.972Z]                                ^
[2020-07-03T21:00:50.972Z] /work/mxnet/src/operator/contrib/transformer.cc:146:35: error: narrowing conversion of 'lda' from 'mxnet::index_t {aka long int}' to 'int' inside { } [-Werror=narrowing]
[2020-07-03T21:00:50.972Z]    MKL_INT p_lda[GROUP_SIZE] = {lda};
[2020-07-03T21:00:50.972Z]                                    ^
[2020-07-03T21:00:50.972Z] /work/mxnet/src/operator/contrib/transformer.cc:147:35: error: narrowing conversion of 'ldb' from 'mxnet::index_t {aka long int}' to 'int' inside { } [-Werror=narrowing]
[2020-07-03T21:00:50.972Z]    MKL_INT p_ldb[GROUP_SIZE] = {ldb};
[2020-07-03T21:00:50.972Z]                                    ^
[2020-07-03T21:00:50.972Z] /work/mxnet/src/operator/contrib/transformer.cc:148:35: error: narrowing conversion of 'ldc' from 'mxnet::index_t {aka long int}' to 'int' inside { } [-Werror=narrowing]
[2020-07-03T21:00:50.972Z]    MKL_INT p_ldc[GROUP_SIZE] = {ldc};

[szha]

@TaoLv thanks. I will wait for that PR to be resolved.

[sandeep-krishnamurthy]

@access2rohit - Please help review. Thanks.

[szha]

@sandeep-krishnamurthy the CI is currently stuck on build issue that @TaoLv pointed out. @access2rohit 's PR on the build fix is needed to address the issue but it seems that there has been no progress in #18645. What's the plan?

[sandeep-krishnamurthy]

As discussed in this issue #17331 (comment) as a first step we are focussed on OpenBLAS updates only as it is the primary BLAS engine we ship with PyPi. MKL updates is next.

[access2rohit]

@szha
Can we move the file: tests/python/unittest/test_np_large_array.py to nightly folder instead? Keeping them in unittest will significantly slowdown the CI and may result in timeouts as we are allocating tensors with over 4.3 Billion elements. Even if CI doesn't timeout right now once all numpy ops are added it will definitely timeout.

[szha]

@sandeep-krishnamurthy I'm not asking about MKL as feature request. I'm saying that the current build is broken and it needs fix.

[access2rohit]

Overall functionality seems fine. Can you run all the tests in the file tests/python/unittest/test_np_large_array.py using pytest and paste the output in the comments section so we can be sure that nightly CD pipeline doesn't break.

[szha]

@access2rohit will do. We still need to fix the CI for enabling large tensor first.

[access2rohit]

@sandeep-krishnamurthy I'm not asking about MKL as feature request. I'm saying that the current build is broken and it needs fix.

Basically I will disable large tensor build with MKL in CI when enabling Large Tensor by default and that should fix it(Since we are currently focusing on making Large Tensor work with openBLAS). For now you can remove your change to make USE_INT64_TENSOR_SIZE=ON as default from CMakeLists.txt and your PR should pass CI. Sounds good ?

[access2rohit]

LGTM! BTW codecov is failing, does it block us from merging the PR ?

[szha]

@access2rohit thanks for the review. no codecov isn't enforced yet.