Tensor cores used only for fp16 in interleaved multihead attention

[blchu]

Description

Fixed issue where fp32 inputs use tensor cores for the interleaved multihead attention operators, resulting in lower precision calculations and potential reduction in accuracy.

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

• Set interleaved multihead attention GEMM default to not use tensor cores, and only use if input data type is fp16
• No longer checks for tensor input shape divisibility by 8

Comments

[mxnet-bot]

Hey @blchu , 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-cpu, unix-cpu, edge, centos-gpu, miscellaneous, windows-gpu, clang, sanity, unix-gpu, website, 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.

[leezu]

Is this tested on cuda architectures < 5?

[blchu]

I've tested on cuda architecture < 5 (specifically, k80), there's no issue running the operator

[blchu]

@mxnet-bot run ci [all]

[ptrendx]

@ChaiBapchya is something happening to CI right now? bot did not trigger the CI. Thanks!

[leezu]

@mxnet-bot run ci [all]

[leezu]

@blchu can you rebase on master and force push?

[ptrendx]

@mxnet-bot run ci [unix-cpu, unix-gpu]

[mxnet-bot]

Jenkins CI successfully triggered : [unix-gpu, unix-cpu]

[ptrendx]

@mxnet-bot run ci [unix-cpu]

Download of cifar failed in the Perl test.

[mxnet-bot]

Jenkins CI successfully triggered : [unix-cpu]