Also modify existing kernel so it works within the builder framework

and remove broadcast, which will be replaced with an Eigen implementation

This allows op builders to be self-contained changesets

* update version and add glossary defs

* clean up graph rewrite code blocks

* PR feedback

* add better details to LSTM def

* RNN def generalized

* adding fancy formulas to RNN def glossary entry

* Address API breaking change in PR 1164

* all of the documentation re default install path needed updated with pr 1164

* Assert manual compilation process to build ngraph_dist locally as a sensible default

* IntelGPUBackend: const, div, maxpool and max operations

* IntelGPUBackend: negative, abs, relu, sqrt, tanh and substract operations

* Update intelgpu_backend.cpp

* make pool test check backends other than CPU

* more unit test cleanup

* onnx: add 'constant' operator
Signed-off-by: Artur Wojcik <artur.wojcik@intel.com>

* onnx: getting attribute value by name
Signed-off-by: Artur Wojcik <artur.wojcik@intel.com>

* onnx: fix code style
Signed-off-by: Artur Wojcik <artur.wojcik@intel.com>

* onnx: fix clang compilation warnings
Signed-off-by: Artur Wojcik <artur.wojcik@intel.com>

* onnx: incorporate review comments
Signed-off-by: Artur Wojcik <artur.wojcik@intel.com>

* cpu loop kernel fusion pass

*  remove extra code

* bounded relu test

* address scotts feedback

* IntelGPUBackend: BatchNorm 5x1 operation

* Update intelgpu_op_batchnorm.cpp

* PR1244 Comments are adressed

* Fix a segfault in the strided conv optimization

* Only bail if all *live* users are not Convolution

* Draft of updates for JIRA tasks WIP

* meta to correct spelling

* Modify TBB graph nodes creation and deletion
* Add a graph* member to CPURuntimeContext.
* Create nodes the first time a function is called, all the following calls only exectue the computation.
* Delete nodes when cleanup_runtime_context is called.

* Add TBB global_control* and task_scheduler_init* members to CPURuntimeContext.

* Remove one comment.
Do not write two TBB header files and one #define to generated c++ source code.

* Move TBB header file and #define before other header files in generated c++ source code.

* Move one comment to the top in generated c++ source code.

* inplace results

* fix parameter propagation

* fix python tests

* change GPU to use cfe pass

* update per review comments

* CPU Direct Execution: Implement ConvertLayout and refactor

* CPU Direct Execution: Implement Convolution

* 1) Adds computation reuse to direct execution
2) Add avg_pool, broadcast and convolution_bias to direct execution
3) Moved some computation reuse utility functions to graph_utils

* Use lists instead of vectors to avoid reallocation overheads

* - Added convolution variants to direct execution
- Removed ConvolutionBiasRelu, use ConvolutionBias instead
- Reduced code duplication by moving functionality to mkldnn_emitter
  from cpu_emitter

* Style fix

* Moved mkldnn build_convolution to a templated method

* Style fix

* refactored mkldnn conv bprop builders

* Style fix

move long building tests to the be the first tests built with the hope of reducing build time. (#1229)

* using async gpu timers

* remove sync for cuda calls, add async gpu stopwatch, add count to timing-detail

* add debug sync

* make timer static

* move timer to runtime context

* Draft of updates for JIRA tasks WIP

* correct typo

* more cleanup

* more cleanup

* Refactored GPU backend state into BackendContext and moved it to the highest level GPU_Backend.
Some bugs have appeared in so doing. Needs investigation.

* extra *block_size

* change grid_size to threads

* Bug fix in softmax cache parameters.

* Additional bug fix for maxpool1d cache parameters.

* Bug fix in softmax cache parameters.

* Additional bug fix for maxpool1d cache parameters.

* Remove temporary print statements.

* Use nthreads in primitive hash.

* Switched from using stack references for cudnn and cublas handles to heap pointers held only the c-struct GPURuntimeContext but managed by the GPU_Backend.

* Refactored the use of GPURuntimeContext* ctx throughout the emitters.

* Use std::prev instead of operator-- for memory iteratory capture

* bug fix from abaf1d7

* Backend/API: Implementation of the call method for IntelGPU

* intel_gpu_style_fix_1199

* Copy memory from clDNN to Tensor

* Code style fix in 1199.2

* get_subgraph_outputs

* simplify the condition