* Improve the readability of imported doxygenclass api doc style
from being so small (WIP).

Also fixed the not op to not import abs doxygenclass.

* Make sure no large files get added to git

* Other doxygen files to exclude from git

* WIP phase two styling code to save and test

* Cleanup for PR

Fix bugs and enable dot in GPU

Add nvrtc support for CUDA

Merge branch 'tfl/gpu_add_nvrtc' of github.com:NervanaSystems/private-ngraph-cpp into tfl/gpu_add_nvrtc

* add TF serialzed graphs

* Replace using aliases with actual classes

Elementwise Add mkldnn support in cpu emitter

- support layout assignment pass to mkldnn add, to insert reorder if the input1 and input0 format are different

Merge branch 'tfl/gpu_add_nvrtc' of github.com:NervanaSystems/private-ngraph-cpp into tfl/gpu_add_nvrtc

- select mkldnn add only if the input_tensor size is >= 1

* initial support for hybrid transformer

* add broadcast_vector_rowwise_reversed for hybrid test

* headerc

* get function placement fix

* conv ref test generator graph node in labmda fuction

* rename map_parameter_to_source_node

* type change map_parameter_to_source_node

* use interpreter for numerical derivative

* better comments

MKLDNN optimized layouts: convolution bprop, avg_pool fprop and bprop

- modified mkldnn add kernel to use mkldnn_utils to build primitive and stream execution

Add simple method to add default values for missing deserialized fields. Provided to ease backwards compatibility. (#547)

* working

* cleanup