add unittest of retry_allocator

* set the right shape of selected_rows
test=develop

* enhance check

* fix activation_op

* remove cast

* use ShareDimInfo replace SetDim and ShareLod

* use ShareDimAndLod
test=develop

* follow comment

test=develop

* check whether the input has lod
test=develop

* Split ShareDimAndLod

test=develop

* checkout clip.py
test=develop

fix lod tensor

test=develop

* Fix bug in reduce_op caused by PR #13534
* Fix output shape and enhance unit test.
test=develop

revert with_fast_math to ON

test=develop

optimize pyreader

* "avoid copy back to cpu"

* "add infinity support"

* "fix ci"

* "add cpu macro"

* rerun ci; test=develop

* "fix api"
test=develop

* test=develop

* test=develop

* test=develop

* test=develop

* test=develop

add a fake reader for speed test

clean unused code and small optimize

* fix cuda9 docker build test=develop

* update test=develop

* update test

disable EIGEN_FAST_MATH and use_fast_math

test=develop

fix compile error on latest Mac OS 10.14

test=develop

Add nodes for Dyrnn

test=develop

test=develop

clarify the fraction_of_gpu_memory flag

fix blas include failed problem

test=develop

test=develop

 add resnet50 inference unit-test

* "fix operators cmake"

* "rerun ci"
test=develop

Fixed missing set_attr

Enable MKL-DNN in Naive Executor and Analysis Predictor