Change `PADDLE_ONLY_CPU` to `PADDLE_WITH_GPU`

By shell command

```bash
sed -i 's#ifdef PADDLE_ONLY_CPU#ifndef PADDLE_WITH_GPU#g' `find ./paddle/ -name '*.h' -o -name '*.cc' -o -name '*.cpp' -o -name '*.c' -o -name '*.cu'`
sed -i 's#ifndef PADDLE_ONLY_CPU#ifdef PADDLE_WITH_GPU#g' `find ./paddle/ -name '*.h' -o -name '*.cc' -o -name '*.cpp' -o -name '*.c' -o -name '*.cu'`
```

Bug fix

Add helper function in GradOpDescMakerBase. Make it easier to use.

add generic add operator

gather, scatter with gpu support, passed python test

* Changing learning rate from attribute to input(float)
* Removing obsolete code

Feature/simplify sum op

Fix MacOS compile error

The private data `tensor_shared_` is not used.

remove rnn infershape

macro header