[v1.6.x] Backport [MKL-DNN] Integrate Conv3d and Pool3d/1d (#17884) and Fix Sanity pipeline in 1.6.x

[ChaiBapchya]

• Integrate MKl-DNN conv3d and pool3d/1d

• fix UT & address comments

• clean code

• rebase against latest master

[mxnet-bot]

Hey @ChaiBapchya , 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: [sanity, centos-gpu, miscellaneous, windows-cpu, edge, website, clang, windows-gpu, unix-gpu, unix-cpu, centos-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.

[ChaiBapchya]

@mxnet-bot run ci [sanity]

[mxnet-bot]

Jenkins CI successfully triggered : [sanity]

[TaoLv]

@wuxun-zhang please help to review ~

[wuxun-zhang]

There are still some incompatible changes derived from master branch. Please modify them according to the comments.

[ChaiBapchya]

@wuxun-zhang Thanks for pointing it out. Addressed your comments.

[ChaiBapchya]

test_quantization_mkldnn.test_quantized_conv failed with

[2020-05-04T03:07:42.847Z] ----------------------------------------------------------------------

[2020-05-04T03:07:42.847Z] Traceback (most recent call last):

[2020-05-04T03:07:42.847Z]   File "/usr/local/lib/python3.5/dist-packages/nose/case.py", line 198, in runTest

[2020-05-04T03:07:42.847Z]     self.test(*self.arg)

[2020-05-04T03:07:42.847Z]   File "/usr/local/lib/python3.5/dist-packages/nose/util.py", line 620, in newfunc

[2020-05-04T03:07:42.847Z]     return func(*arg, **kw)

[2020-05-04T03:07:42.847Z]   File "/work/mxnet/tests/python/mkl/../unittest/common.py", line 177, in test_new

[2020-05-04T03:07:42.847Z]     orig_test(*args, **kwargs)

[2020-05-04T03:07:42.847Z]   File "/work/mxnet/tests/python/mkl/../quantization/test_quantization.py", line 283, in test_quantized_conv

[2020-05-04T03:07:42.847Z]     check_quantized_conv((1, 3, 4, 28, 28), (1, 3, 3), 128, (1, 1, 1), (1, 1, 1), (2, 2, 2), False, qdtype)

[2020-05-04T03:07:42.847Z]   File "/work/mxnet/tests/python/mkl/../quantization/test_quantization.py", line 217, in check_quantized_conv

[2020-05-04T03:07:42.847Z]     arg_shapes, _, _ = conv.infer_shape(data=data_shape)

[2020-05-04T03:07:42.847Z]   File "/work/mxnet/python/mxnet/symbol/symbol.py", line 1103, in infer_shape

[2020-05-04T03:07:42.847Z]     res = self._infer_shape_impl(False, *args, **kwargs)

[2020-05-04T03:07:42.847Z]   File "/work/mxnet/python/mxnet/symbol/symbol.py", line 1267, in _infer_shape_impl

[2020-05-04T03:07:42.847Z]     ctypes.byref(complete)))

[2020-05-04T03:07:42.847Z]   File "/work/mxnet/python/mxnet/base.py", line 255, in check_call

[2020-05-04T03:07:42.847Z]     raise MXNetError(py_str(_LIB.MXGetLastError()))

[2020-05-04T03:07:42.847Z] mxnet.base.MXNetError: Error in operator conv: [03:07:30] src/operator/nn/convolution.cc:242: Check failed: param_.dilate.Size() == 1U (8 vs. 1) : Dilate is not supported in 3d convolution

Cherrypicked #17491 which removed the conv3d dilate restriction.
That should resolve unix-cpu pipeline errors.

[ChaiBapchya]

@wuxun-zhang One last failure
unix-gpu pipeline
Tests / Cpp: MKLDNN+GPU

pooling-inl.h:313: Check failed: outputs.size() == GetNumOutputs(param) (1 vs. 2) :  

I was able to repro locally. Not sure why output size is mismatch for pooling.

[wuxun-zhang]

@ChaiBapchya To fix the unix-gpu issue, you can refer to this link, like what I did when backported this PR to v1.x branch before.

[ChaiBapchya]

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

[mxnet-bot]

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

[ChaiBapchya]

@mxnet-bot run ci [unix-cpu]

[mxnet-bot]

Jenkins CI successfully triggered : [unix-cpu]

[wuxun-zhang]

LGTM

[ChaiBapchya]

@mxnet-label-bot add [pr-awaiting-merge]