* add op inout check macro, test=develop

* fix enforce_test, test=develop

* Delete is_test from activation operators
test=develop

* Revent unneeded changes
test=develop

As nccl comm is not created by CUDADeviceContext, it should be destroyed by the creator as the best practice of RAII.

* refine output of profiler for child event 

initialize global nccl context in dygraph
test=develop

* refine the profiler print test=develop

* add recorded cuda memory apis, fix typo, test=develop

* add more ut, test=develop

* follow comments, test=develop

* fix py35 incompatible issues, test=develop

* add profiler_help.h to refine the code test=develop

* add framework overhead ratio, test=develop

* print GpuMemcpy overhead, test=develop

* add support for the driver api callback and fix the profiler name show bug

* fix the profile print error test=develop

* add python flag to control profile level test=develop

* reproduce match error, test=develop, test=document_fix

* fix mismatch error, test=develop, test=document_fix

Compile without nccl deps. [1/2]
Co-authored-by: N石晓伟 <39303645+Shixiaowei02@users.noreply.github.com>

* optimize interpolate op, test=develop

cmake选项中添加了WITH_NCCL，显示指定是否编译NCCL的部分代码，WITH_NCCL默认打开，但如果WITH_GPU为OFF，则关闭WITH_NCCL

添加了PADDLE_WITH_NCCL定义

单机单卡能够关闭NCCL编译，多卡的话需要默认打开NCCL，如果关闭NCCL，则只能使用单卡
Co-authored-by: N石晓伟 <39303645+Shixiaowei02@users.noreply.github.com>

* Enable quantize to reorder to nchw as well

* Correct FC MKL-DNN input dim requirements to accept 3D

* Improve DNNL FC format, error and 3D input handling

test=develop

* Improve error checking in FC

test=develop

* Improve PADDLE_ENFORCE messages in fc-related files

* Remove data layout attribute from obligatory pass args

test=develop

* Fix message in fc_mkldnn_pass to be logically correct

test=develop

* fix the bug of profile update test=develop

* Polish the PADDLE_ENFORCE in fusion_group pass related codes.
test=develop

* Correct the unittest because of the change relu_grad's formula.
test=develop

* add support for nested profiling event and printing in different level

stack's wait cost a lot of cpu time, use cuda kernel to do memory copy
will reduce cpu time.
Signed-off-by: Nzhaoyuchen <zhaoyuchen01@baidu.com>