* add top_k operation

* modify topk definition

* Update intelgpu_backend.cpp

* Update intelgpu_op_custom_kernels.cpp

* add topk to graph visualization

* enable index_element_type in graph visualization

* minor changes

* Serializer supports new pad api

* Addressed reviewer comment

* Fixed compiler warning

* Enable previously CI-failing test

* Disable max_3d_to_scalar_int32, still failing on CI

* Remove tests

* [Standalone] Decouple MKLDNN primitive build from code generation

This patch introduces a new pass, MKLDNNPrimitiveBuildPass, which
iterates over all the ops assigned to MKLDNN and builds their
corresponding primitives. Primitive indexes are stored in MKLDNNEmitter
and can easily be retrieved with the get_primitive_index(node)
interface. This decouples the creation of primitives from codegen and
fixes the problem of MKLDNN primitives being created twice
(CommonFunctionCollection pass and codegen).

Current assertions only allow the creation of a single primitive per
node but it should be simple to remove this when needed. Using a pass
might not be the best approach here but I found it convenient for the
current needs and it should be straightforward to convert into a utility,
if needed.

These changes caused a conflict with recently introduced
'build_quantized_inner_product*'. These new build methods will be ported
in a follow up patch to new build approach.

* Remove unrelated comment

* Remove TensorView code

* Set m_node_primitive_map from MKLDNNPrimitiveBuildPass

* Move node->primitive map from mkldnn pass to external function

* Fix struct/class inconsistency en fw declaration

* Add Quantized MatMul OP

* Rename qmatmul -> quantized_matmul

* Use auto everywhere

* More doc review

* Doc to v0.18

* Delete build.md

This reverts commit 6c8284a3.

* add Min/Max support to zero dim tensor elimination

* fix infinity

* [ONNX] Unit test models for QLinearConv

* Temp ref conv

* ref conv temp

* add qlinearconv2d test

* adding conv3d test case

* ignore tests on GPU

* Dynamic scale

* add builder based solution for zero point and conv

* Revert "Dynamic scale"

This reverts commit be8e57bdf7013967e5575164a0402dcc6d16b8ed.

* Revert "Merge remote-tracking branch 'origin/nishant_ref_conv_u8u8' into mkarzyns/qlinear_conv_uts"

This reverts commit dea29a18c474b644b5b531f0e59f21d21bd56bf2, reversing
changes made to b12fc13c5852efa4c335377164d9b7e5d9227a8a.

* style

* [WIP] First commit

* Incremental code changes

* Incremental code changes

* Further mods

* Improve src + add more tests

* Another test added

* clang

* Added NGRAPH_DEBUG statements

* Incorporate Xiaoyu and Scott's comment

* Incorporate Adam's comments in tests

* Incorporate Adam's comments

* Add Jayaram's comments

* Add support for QDotInteger

* Address Feedback

* Support int32 as output of QuantizedDot without requantization similar to QuantizedConvolution

* Initial support for specification of fused ops and type inference

* Added FusedOpDecomposition pass and execution test cases

* Serializer support

* style fix

* Add FusedOpDecomposition to GPU and IGPU backends

* Addressed PR feedback

* Fix comment

* Addressed PR feedback

* add windows support for event trace

* add needed windows.h header

* fix linking error

* Enable multi-threaded builds for windows

* [ONNX] Unit test models for DequantizeLinear

* add unit tests and suppot axis

* ignore unit tests on GPU

* fix build warning

* [ONNX] Unit tests for QuantizeLinear

* Fix some syntax issues

* get tests passing

* adding prototxt files with correct names

* disable new tests in the GPU

* cleanup

* style

* Add some unit tests for provenance with node replacement

* One more test

* add provenace propagation

* Use cpu kernel for constant folding.

* Add default empty map.

* Fix a bug.

* Add new files.

* Address PR feedback.

* Check constant folding map before checking type for unary and binary ops.

* Address PR feedback.

* Address PR feedback.

* Use all_close_f.

Add relu unit test.

Make changes for sqrt and pad.

* Fix a bug.

* Change OneHot to accept only non-real types

* Update docstring

* Update Python test

* Add is_integral to element::Type

* Update docs

* Change is_integral to false for boolean

* Revert "Change is_integral to false for boolean"

This reverts commit 099ff378ae7fcbd1d9346665812f6b95e4886186.

* Revert "Add is_integral to element::Type"

This reverts commit 58fdf76fecaefdad10431f9a894523f326f3adca.

* Change is_integral so it is, by definition, !is_real

TopK: return the lower index for equivalent values

* add unit test for create_tensor passing in buffer to output tensor

* add comments

* disabled gpu create tensor unit test

Update template for release notes and CHANGELOG with latest few summaries from ngraph releases repo (#2696)

* Update Release Notes for Github summaries, incorporate feedback from doc review, fix typos, etc

* Fix links

* Rename RCs on template

* PR reviews

* Remove trailing spaces and add clarification on Python API example

* Fix typo

* [ONNX] Use NullNode for the optional output in MaxPool

* Style check

* add test file

* add new float_util files

* Add unit tests for bfloat. Fix bfloat rounding code since it was incorrectly rounding.

* add more tests

* cleanup

* change trucate to be endian agnostic

* prep work for constexpr ctors

* ready for constexpr

* constexpr ctor for bfloat16

* more bfloating

* write constexpr isnan since it is not constexpr on Macos

* remove cast operator

* add benchmark test and cleanup

* use aligned buffers for benchmark tests

* fix numbers printed in benchmark

* remove union and use cast operator

* all tests passing

* cleanup

* [ONNX] Add verification of split attribute in split op

* Style

* Update split.cpp

* Add verification of split length

IntelGPU backend: Select operation fix (clDNN doesn't support int32 data type); Switched to clDNN 14.0 drop (#2702)

* [Warnings] Silence more warnings

Unused variables and missing switch case.

* Move static global var to the host_tensor translation unit

- Remove non-portable 'unused' attribute.

* Fix style

ConvertLayout is a MKLDNN specific op and the MKLDNN kernel flag was not
set. This led to assume that this operation was not supported by
MKLDNN.