Bugfix: kAddTo support for swap_axes.

[tdomhan]

Description

With this PR the swapaxes operator no longer ignores the req argument. Previously this silently lead to wrong gradients.

Checklist

Essentials

• Passed code style checking (make lint)
• Changes are complete (i.e. I finished coding on this PR)
• To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change

[KellenSunderland]

@eric-haibin-lin Any chance we can get this in the next release?

[KellenSunderland]

@tdomhan Could you amend the commit and add

Co-authored-by: Hetzel <ahhetzel@amazon.com>

at the bottom?

[tdomhan]

yeah of course. I also wouldn't mind to close the PR and update the other PR. Just whatever solution gets the fix into the next release works for me :)

[asmushetzel]

@piiswrong
Somehow I forgot about the pending #9495 , sorry. This here is indeed the suitable fix. I will close 9495.
That definitely should go into next release

[sxjscience]

Like my comments in #9495, we should also revise the test here (https://github.com/apache/incubator-mxnet/blob/master/tests/python/unittest/test_operator.py#L361-L377) to add a test for “grad_req=addto”. You may refer to the test codes in https://github.com/apache/incubator-mxnet/blob/master/tests/python/unittest/test_operator.py#L775-L781

[piiswrong]

I'm merging this because code freeze is today. Let's update tests later.

@marcoabreu Please track this. We should actually test for addto for ALL operators automatically.

[marcoabreu]

Tracked at #9497

@tdomhan would you mind adding a test in a second PR? It would be great if we could test this ASAP.

[tdomhan]

thanks for merging! I will try to create an appropriate test based on pointers from @sxjscience.