* fix nccl op unit test

* fix build error

* format code

* refine nccl related unit test

* fix build error

* add setGPUData

* clean up

* follow comments

* rm test_nccl.cu

* follow comment

* rm wait

* compile with nccl2

* add ncclGroup; it is necessary in nccl2

* add back libnccl-dev

* move float16 into fluid

* fix include

* move to platform folder

* "change operator interface"

* "move devicepool to device_context"

* "fix operator test"

* "fix op_registry Run interface"

* "net op passed. Need to fix nccl multi-Context"

* "add nccl group function"

* "add nccl group function"

* "fix gpu count exceed 32 error"

* "fix recurrent op, nccl op"

* "change the other operators interface with Place"

* "fix typo"

* "fix pybind"

* "fix device in python side"

* "fix pybind failed"

* "add init for test"

* "fix CI"

* make enforce a target and dependent on nccl when gpu is enabled

* add some more dependency

* remove unused code

* fix cmake file

* fix build error

* "add nccl enforce"

* Dev

* Update comment

* Add nccl test

* Follow comments

Using DeviceContext, not Place to get stream

* with unit-tests
* Also complete `memcpy`

* add device_context

* add unittest for device_context

* transfer to use function paddle::platform::throw_on_error

* fix cuda build error

* using dynload functions

* follow comments