* Migrate #2864

* Remove extra changes

* Fix data type for conv builder

* Added support for other data types

Add build option NGRAPH_ENABLE_CPU_CONV_AUTO to enable/disable mkldnn convolution_auto for CPU backend. (#2868)

mklml on linux has no soname set. This will cause linking against the full path for mklml when using prebuilt mkldnn and mklml. (#2851)

* Klocwork issues.

* Revert to dynamic_pointer_cast

* - add support to algebriac simplification to check dynamic shapes against the Pass Properties

* - added PassProperty to RecurrentMatcher
- added checks to check for dynamic function state  and PassProperty in GraphRewrite before applying graph optimization
- optimize number of calls to f->is_dynamic() in AlgebraicSimplification

* - make changes to Algebriac Simplicfication graph pass to accept PassProperty

* - test case for AlgebraicSimplification pass properties
- set the Pass Property in the pass ctor

* Address PR comments

* fix bug in pass manager

* Addressed PR comments

* Update for v 0.19

* Add rnotes from github releases page

* Add section on graph inspection tools, add link to HE transformer, and fix link to ngraph bridge for Tensorflow

* Fix broken links

* Remove MANYLINUX from list of default flags

* Revise some for new script loc

* Add stub for ngc_util python script

* Might as well link to open source file

* original file from Vasanth Tovinkere

* updated to handle more attributes.

* Update new script loc and details

* Revise intro

* Clarify large graphs generated from script best read on gephi or cytoscape only

* Edit oddly-worded sentence for clarity

* Delete ngraph_json_to_graphml.py

* Delete ngc_util.py

* Add nGraph logo, update color scheme, disallow preprocessing of frontends' markdown UI

* Finalize doc changes for release v0.19

* PR feedback on nGraph-friendly tools

* Basic float16

* Basic fp16 support

* typo

* Create mkldnn primitives at first iteration for CODEGEN.

 OPs: add, lstm, and rnn.

*  OPs: batchnorm.

*  OPs: concat and lrn.

Remove dead code.

* Skip in place concat, relu, reshape, and slice when building node_primitive_string_deps_index map.

* Change NGRAPH_ASSERT to NGRAPH_CHECK.

* Ops: Qconv

* Ops: Convs

* Address PR Feedback.

* Dynamic scale support for qconvs

* updating to amy's recent change

* GroupConv and Cleaning dead code

* Address PR Feedback.

* Remove unused variable.

* Fix a bug.

* Fix style error.

* Change CPU backend to use PassConfig instead of backdoor for codegen

* move is_codegen decl to output dex only flag

* revert change

* Remove parameter, result, and node vector files and class, use vector

* Fix gcc issues

* style

* Remove python references

* update python files

* fix error

* fix build error

* Support constant folding for reshape and broadcast bf16 data

* - Addressed PR feedback
- Adding extra paranthesis for icc

* Generalize types in general convolution

* type-o

* rounding

* Do prod wide

* templatize conv in cpu/kernel & add u8u8 support for Qconv

* Remove cast function

* Avoid compiler warning

* Merge problem

* Implement thread safe calls.

Create MKLDNN primitives on first iteratin for quantized_concat.

* Add buffer_data to CPURuntimeContext.

* Fix bugs.

* Modify unit test.

* Swap vectors of mkldnn primitive pointers for CODEGEN.

* Fix a bug.

* Address PR feedback.

* Rename variables.

* Update Gather, GatherND, and DeconvolutionBias.

* Fix style error.

Disable cpu thread_safe_calls test on Windows.

* - Added support for complex ops used in object detection models
  - PriorBox, PriorBoxClustered, Proposal, RegionYolo, ReorgYolo

* Added  unit test file

* Fix documentation errors

* IntelGPU backend: GroupConvolution implementation

* IntelGPU backend: GroupConvolution with clDNN support added

* use better ctor in cldnn

* IntelGPU backend: Use core fusion pass

* Add BatchFusion pass

* Refactor distributed to be isolated to a few files.

* Fix type-o

* style

* review comments

* type-o

* typo

* Return name

* Add ShapeRelevance pass

* Typo

* Change warmup to run full, normal iterations

* style

* Moving GroupConvolution to ngraph fused opset

* style fix

* remove unused function

* IntelGPU backend: Add GroupConvolution operation into main switch

* Refactor get_default_axis_vector to use std:: functions

* Move get_default_axis_vector to ngraph::op

* Move reorder_axes to ngraph::op::util

* Move reshape helper to ngraph::op::util

* Move DepthToSpace to fused ops

* Add DepthToSpace docstrings

* Move SpaceToDepth to fused ops

* Remove redundant ngraph::op::util::get_default_axis_vector function

* Add ops to serializer

* Change block_size to size_t

* Add fused ops tests

* Add type prop tests

* Add ops to list of ops unsupported on iGPU

* Disable tests in iGPU manifest

* - add safety checks for mkldnn-assignement for quantized inner product

* - add asserts for unsupported data types in builder & emitter code of Quantized Dot

* Add fused version of Elu op

* Refactor ONNX importer prelu function to use fused op

* Style check

* Add docstrings

* Move make_constant_node to op/util

* Use make_constant_node helper

* Remove unneeded std:: prefixes

* Remove make_constant_node function, use builder::make_constant

* Remove redundant includes

* Add Elu to serializer

* Add Elu tests

* Add Elu tests to type prop

* Add Elu to list of ops unsupported on iGPU

* Add Elu to list of ops unsupported on iGPU

* Disable tests in iGPU manifest

* rework PerformanceCounter class

* Update intel gpu to new PerformanceCounter and enable performance on interpreter

* Add new HybridTensor class. Remove Tensor parent.

* upate more backends

* fix GPU tensor

* update intel gpu backend