Static memory allocation for cached_op

[piiswrong]

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

[larroy]

where is this used? why expose this detail?

[ThomasDelteil]

@piiswrong I have mixed results with this PR. Did you benchmark your code?
I am running this notebook: https://github.com/ilkarman/DeepLearningFrameworks/blob/master/notebooks/Gluon_MultiGPU.ipynb

It is DenseNet121 running on multilabel classification task on multigpu. Batches are fixed shapes (N, 3, 224, 224)

Here are my results:

• baseline:
  static_shape=False, static_alloc=False -> 8min28

• Supposedly most optimized, is slowest by far
  static_shape=True, static_alloc=True -> 9min32

• only static_alloc, ~speed slightly faster but within normal variations
  static_shape=False, static_alloc=True -> 8min25

• only static_shape, based on docs this shouldn't be possible because:

            Must also
            set static_alloc to True. Change of input shapes is still allowed
            but slower. 

No warning or error was thrown. ~speed slightly faster and outside of normal variations.
static_shape=True, static_alloc=False -> 8min23

@marcoabreu