fix dist unit test

Rewrite eager deletion

* Add psroi_pool operator.

test=develop

fix clang

* add l2 unit test
test=develop

* code refine
test=develop

* fix nccl dist test acc test=develop

* fix test=develop

[MKL-DNN] Concat Layer

test=develop

test=develop

* Fix the gelu backward to avoid nan

test=develop

* Remove unnecessary calls

test=develop

an sample of recording benchmark to file for trt

pserver should crash early whe has problem

fix visualizer

Follow ups on nGraph PR comments

test=develop

test=develop

Add gperftools supports for PE

Fix Eigen macro when using GPU

test=develop

test=develop

test=develop

test=develop

fix the wrong format

test=develop

test=develop

The macro should be defined by compiler rather than by source.

test=develop

split selected rows op should always init output selected rows

test=develop

test=develop

add Opdesc's HasProtoAttr

test=develop

* random failed. rerun ci. test=develop

* windows failed. rerun ci. test=develop

add bpr_loss operator

test=develop

fix control_flow ops in outs

test=develop

Imperative