[v1.4.x] mkldnn s8 conv API change for 1.4.x branch

[ZhennanQin]

Description

This is extracted from #13697, to make API change for 1.4.x branch.
@KellenSunderland @pengzhao-intel

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

[stu1130]

@mxnet-label-bot add [pr-awaiting-review]
Thanks for your contribution @ZhennanQin

[pengzhao-intel]

@KellenSunderland could you help take a review this PR? The deadline of 1.4 is very near :(

[pengzhao-intel]

@TaoLv @xinyu-intel please help take a review too :)

[srochel]

Please review and merge asap, this is the last PR blocking 1.4.0.rc1. Please keep an eye on builds and CI.

[szha]

@zheng-da @reminisce @eric-haibin-lin ping for review.
@srochel this change concerns what API to expose in 1.4.0 so please allow some time to get it right.

[KellenSunderland]

Having a more detailed look now ...

[KellenSunderland]

LGTM.

[pengzhao-intel]

LGTM

[TaoLv]

The changes look good to me. Merging now to catch the code freezing of 1.4.0rc1. Please make sure the same changes will also happen on master branch, so we would not have API discrepancy between minor versions.

[szha]

The concern I have on this API design is the fact that types that are exposed in the public API come from mkldnn, which evolves independently from mxnet.

[ZhennanQin]

@szha , A better design is isolating public API with internal API, which is, create a new ndarray wrapper class that only exposes public API. But this probably can't be done within 1.4.x. Consider there're many mkldnn APIs already existing in ndarray.h, e.g.

https://github.com/apache/incubator-mxnet/blob/3165262f1f4c94aaf9bb9a333f32d4c94fddf293/include/mxnet/ndarray.h#L722-L745

so I think it's fine for this time.

[szha]

Yes, IMO those shouldn't have happened either. We'd have to address those in 2.0.

An alternative to extending the NDArray class is to pass around the metadata using nnvm within the abstraction of subgraph.

[szha]

#9686 (comment)