Added a test in test_NetworkCompare to verify cudnn and exconv, fixed a bug in cudnn_conv in dealing with groups

The instance weight and coeff are only mulipled to the gradient with respect to input.

* error when paddle has a wrong version number.

* Because it is a cuda source file, and we need to support c++ 03 in
  cuda.

* Different Type compare.
* ostream << should pass a const object.
* remove always true checks.

* Because it is a cuda source file, and we need to support c++ 03 in
  cuda.

* Different Type compare.
* ostream << should pass a const object.
* remove always true checks.

* Mount local Paddle instead of git clone from remote.
* Use official chinese ubuntu source instead of 163 mirror.

* unittest need to be added. But fix the bugs first.

Change-Id: Ic80845c892c96c37a0df0ddc433fe1aeaa5a9d1c

out = w * input
where w is a parameter of size 1

Change-Id: Ife682d62323ceb1a20cbbf6269421b20a862d888