test=develop

test=develop

* Initialize the elementwise plugin.

* Implement the basic CUDA kernel of elementwise plugin.
test=develop

* refine GraphCheck
test=develop

* fix ci fail
test=develop

This reverts commit be506703.

* Modify some infer-shape in compile-time.

* Optimize layer_norm operator with AVX intrinsic functions

* Revert the wrong modifications

* Implement the jit kernel for layer_norm operator

* Add math headfile to fix the compile issue (test=develop)

* Add math headfile to fix the compile issue (test=develop)

* Fixed the intrinsic headfile issue (test=develop)

* Fix the conflicts (test=develop)

* Revert for CUDA compiler (test=develop)

* Fixed the cuda depency (test=develop)

* Fix the marco issues (test=develop)

test=develop

* Convolution fusion operator.
* Clean code
test=develop

test=develop

* fix dist deps test=develop

* update test=develop

* update test=develop

* update test=develop

* update test=develop

test=develop

test=develop

test=develop

test=develop

test=develop

test=develop

test=develop

the parameters will load from CPUPlace, that will keep copying data
between CPU and GPU places.

test=develop

* wip simplify operator framework

* wip

* wip

* done test=develop

* clean test=develop

* fix test=develop

* fix deps test=develop

* fix cpu build test=develop

* fix tensorrt build test=develop

* fix tests test=develop

* fix test=develop

* fix cpu build test=develop

test=develop

* fix space_to_depth_op unicode problem

* test=develop

test=develop

- Added profiling to softmax functors

- MKL based softmax inference op

- Fix to softmax compuation via MKL

- cleaning

- Cosmetic fixes to softmax MKL

- Fix to ON_INFER lack of propagation

* implements reachability check between identity node and non-identity argument to elementwise_add
* implements handling identity node as x and as y argument to elementwise_add