fix control_flow ops in outs

Imperative

fix bug of trt pool2d converter

update README to 1.2

* add more log for dist test for ci test=develop

* increase deadline test=develop

test=develop

When data flow from a MKLDNN OP kernel to a non-MKLDNN OP kernel,
data layout transform (via MKLDNN reorder) will occur even when
those two OP kernels share same layout. Add code to remove this
unnecessary reorder.

test=develop

Add cuda cudnn version check

Fix numpy bug on Ubuntu16 and Ubuntu18 which will cause segmentfault

test=develop

test=develop

test=develop

Revert "Merge pull request #14798 from PaddlePaddle/revert-14786-revert-14782-revert-14398-imperative"

This reverts commit b1d3a1c8, reversing
changes made to f1fb64b1.

* Choose to turn on use_mkldnn attribute v1

* Fix mkldnn_op empty bug

* format change test=develop

* fix ci test=develop

* fix ci test and add test in dam test=develop

* add example to dam compare test test=develop

* review changes test=develop

Revert "Revert "Revert "Imperative"""

Implement the fusion of convolution 3D and bias for mkldnn

Fix eigen deallocate bug

* enable test_gradient_clip in mac
test=develop

* refine test_gradient_clip
test=develop

* fix io doc
test=develop

* refine data_feeder doc
test=develop

test=develop

Revert "Revert "Imperative""

test=develop

test=develop

Implement conv3d with mkldnn library

Async executor bugfix: Tensor changed to LoDTensor

support loading from memory

test=develop

* add nccl2 dist tests test=develop

* fix dist_base test=develop

* fix tests test=develop

* fix test on mac test=develop

Revert "Imperative"

test=develop

test=develop

test=develop

test=develop