* Convolution fusion operator.
* Clean code
test=develop

* add cudnn ctc loss

* wip add test test=develop

* wip

* wip

* done test=develop

* move include cudnn test=develop

* test test=develop

* fix build test=develop

* fix build test=develop

* fix build on cudnn5 test=develop

* fix cudnn5 build test=develop

* fix cudnn5 build test=develop

* merge develop softmax functor change test=develop

test=develop

* exhaustive search for cuDNN conv.
* Refine code and add unit testing.
* Fix model load in fluid/inference and unit testing in conv2d
* Follow comments.
* Fix compiling test=develop

Fix code to support cpplint syntax check

test=develop

test=develop

This reverts commit ce7d9b07.

* exhaustive search for cuDNN conv.
* Refine code and add unit testing.
* Clean code
* Fix model load in fluid/inference and unit testing in conv2d
* Follow comments.

test=develop

test=develop

Warning only at first when CUDA CC not matched.

test=develop

test=develop

test=develop

* Add affine grid generator.

* fix ffine grid.

* Add unitest.

* Add CPU kernel and fix unitest.

* Fix CPU kernel.

* Refine code.
test=develop

* Fix python api.
test=develop

* Update python api.
test=develop

* Fix comment.
test=develop

* Rename affine_grid_generator to affine_grid and enhence unitest.
test=develop

* Fix unitest.
test=develop

test=develop

* feat(platform): lazy initialization of devicecontext in pool

Use std::async(deferer, []{...}) to lazy initialize DeviceContext in Pool

test=develop

* Add future includes

test=develop

test=develop

test=develop

Make all blob info saved in global device context to be thread based.
Meanwhile save thread id in thread local storage in ParallelDo

Add rpc profiler flags

test=develop