fix memory corruption in gluoncv ssd training

[arcadiaphy]

Description

Fix #13710

In bipartite match, the temporary workspace size is miscalculated and not enough to hold the index tensor.

Since it is related to sizeof(DType) and sizeof(int32_t), the bug is only triggered on some platforms.

Checklist

Essentials

Please feel free to remove inapplicable items for your PR.

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

[zhreshold]

LGTM, thanks @arcadiaphy

[sandeep-krishnamurthy]

Sorry for my naive understanding. But, I do not understand why +1? And this mean, we will have one extra size in the workspace always?

[arcadiaphy]

Yes, we will have extra size sometimes when we need to hold int32_t data in DType workspace, and the calculation assures that only the smallest needed space is allocated.

if workspace = (N * sizeof(int32_t) - 1) / sizeof(DType) + 1,
then workspace * sizeof(DType) >= N * sizeof(int32_t) - 1 - (sizeof(DType) - 1) + sizeof(DType)
                                = N * sizeof(int32_t)

Actually, L426 has already calculated correctly.

[wkcn]

Is it better to create two workspaces for DType and int32_t respectively?

[zhreshold]

workspace allocation is limited to one time only, so total space has to be calculated aforehead

[wkcn]

@zhreshold Thanks!
I found there is the definition Tensor<xpu, 1, char>.

[Roshrini]

@zhreshold Is this PR good to go?