* "fix accuracy kernel bug"

* "relauch ci"

It caused by a bug of std::call_once described in https://stackoverflow.com/questions/41717579/stdcall-once-hangs-on-second-call-after-callable-threw-on-first-call. It is likely caused by a deeper bug of pthread_once, which is discussed in https://patchwork.ozlabs.org/patch/482350/

CompareOp can run on CPU even other operators are running on GPU, since
opeatations like comparing control flags should be performed only on CPU

* remove unused code

* fix cmake file

* fix build error

* init cudnn batch norm op

* rename batch_norm_cudnn_op.cc batch_norm_op.cu

* correct name style

* add ExtractNCWHD, simplify code

* fix ExtractNCWHD

* use CUDNN_ENFORCE instead of PADDLE_ENFORCE

* "add nccl enforce"

* Dev

* Update comment

* Add nccl test

* Follow comments

* initial matmul operator

Similar to np.matmul, but also has transpose_X and transpose_Y flags,
and only supports tensors from rank 1 to 3 inclusive.

For GPU, uses cublas?gemmStridedBatched. For CPU, uses
cblas_?gemm_batch if available via MKL; otherwise a simple serial
implementation that loops over the batch dimension is employed for now.

* add cudnn_conv_op

* WIP

* update

* update

* fix grad check

* use platform::memory

* add support group for cudnn

* update

* follow comments

* fix onlycpu build

* update cuda define

* follow comments

* follow comments

* merge with updates

* fix compile error

* follow comments

* follow comments

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'`
```

* fix nv_library

* fix symbol in gpu_info.h

* If stat is an integer, bool value will implicit cast to int before
  pass to PADDLE_ENFORCE

* Since GPU is an async device by default. We should sync computation
  when Python invoke `run`. So Python can get the correct computation
  result

Using DeviceContext, not Place to get stream

* Also use `const DeviceContext&` all the time, to prevent `const_cast`

Fix #4169
Fix #3468
Fix #3475