* Move split utility functions into core builder.

* Move activation functions to nGraph core.

* RNN cell base class.

* LSTM cell fused operator.

* Update LSTM ONNX operator to use LSTMCell fused op.

* Use Constant::create instead of make_constant.

* Remove ngraph:: prefixes and include standard headers.

* Store member shared_ptrs as object.

* Formatting.

* Run validation at the end of constructor.

* Add more doc to ActivationFunction.

* Run FusedOpDecomposition pass two times in interpreter backend.

* Remove unnecesary class member.

* Add node validation.

* Disambiguate constructors.

* Add type property test.

* Formatting and add comment with equations.

* Update IGPU backend with LSTMCell fused op.

* Fix: clip activation function input.

* Unit tests.

* Workaround for nested fused op: run FusedOpDecomposition twice.

* Fix compilation on CentOS and on GPU.

* PR feedback.

* Fix CentOS bugs.

* Address review comments.

Remove stored inputs as class members. Use node inputs directly in
decomposition.

* Fix errors.

* Review feedback: don't use decompose_op while generating Function in UTs.

* Fix merge artifacts.

* Move RNNCellBase to op/util directory.

* Fix typo for avg_pool setter method.

* Set default values for optional inputs.

* Fix typo in comment.

Fix corner case where op::Constant ctor is called with a zero-element shape and a vector of one string (#3082)

* Adding GroupConvTranspose fused operator.

* Add missing header and remove commented code.

* Remove unused variable.

* Add a few more convieniece constructors.

* Add more type prop UTs.

* Remove unused post validation functions.

* Style apply.

* Fix conversion of vector to CoordinateDiff

* Add GroupConvolutionTranspose to intel gpu backend.

* Add documentation.

* Use default (python-like) divide.

* Added missing attribute to Result serialization

* Fix default layout attribute

* use helper routine for optional attrs

* Initial implementation of the Shrink op

* Multiply the values by the correct masks

* Basic test case for Shrink with floats

* Shrink test on integers

* Code formatting

* Shrink documentation and typo fix

* Rephrase the Shrink docs

* Out of <memory> ;)

* Add DynElimination for Broadcast

* Change silent bailouts for invalid shape/ETs to NGRAPH_CHECKs

* Add execution tests for dynamic reduction ops; fix validation logic

* Add dynamic_GPU.all to manifest

* Be explicit about the dynamic_GPU prefix in manifest

* Initial implementation

* Added test case

* Bug fix; Dropout with 2 outputs, WIP

* Fixed in unit-testl; WIP for model

* Nothing is working

* Revert "Nothing is working"

This reverts commit d3ff09bb7a0d0519ab70ac85f2e7f30721afea96.

* Fixed unit-test; fusion with 2 outputs

* Fix style check, file permissions

* Changed input arg to Node

* Fix order of declaration

* Improved performance

* some cleanup

* Fixed CI error

* Fixed review comments

* Fix CI error

* Remove unused variable

* Fix other CI errors

* Changed type

* Fix style check

* Add codegen code for Dropout

* addressed PR feedback; will add codegen support later

* Cleanup; change variable name

* Support for use_seed

* Add setter for use_seed

* Add setter for use_seed

* Fix CI error

* Make use_seed as arg

* Fix CI error

* Fix CI error

* Simple DynElimination test (not passing yet)

* Implement DynElimination for DynSlice; simple test passing, but more needed

* Add test generator for DynSlice

* Add more tests (one not passing)

* Rename update_reference.sh to update_convolution_reference.sh, to (hopefully) reduce confusion

* Comment edits

* Fix a couple more bugs, add a bunch of unit tests

* A few more tests for the negative-stride slicing issue that's worrying me.

* Refactor dyn slice test to improve compile time

* Update comment

* Further test refactoring: generate separate tests rather than one big one

* More element type coverage because hey why not

* Add more tests, per review comments

* Copy friendly name when copying node

* add unit test

* style

* wip

* hybrid as a static backend and not part of ngraph

* only for linux

* fix link problem

* style

* remove hybrid

* fix compile error

* Change reduction operations to 2-input dynamic variants with convenience constructors for cases where reduction AxisSet is known at op construction time

* Modify rest of arithmetic and logical reduction ops to 2-input dynamic variants. Some fixes to existing passes to keep constant reduction axes inputs intact

* add new All tests to GPU manifest

* Backward comptability for GenerateMask nbench

* PR feedback - use get_or_default

* Added support for ceil mode in AvgPool

* Added ceil mode to MaxPool

* remove extra semicolon

* Add more constructor variants to support pybind which seems to have issues with multiple optional arguments

* More constructor variants for AvgPool

* More constructor variants for MaxPool

* Style fix

* Avoid constructor delegation

* Revert "Avoid constructor delegation"

This reverts commit 8efd59127bc9a16bae93b3c6b67dbcccfa95648f.

* Unit tests for ConvInteger

* Add ONNX ConvInteger op

* Add QuantizedConvInteger builder

* Add unit tests

* Exclude tests on nVidia GPU backend

* Fix merge artifact

* Add const-correctness and allow RVO

* Change behavior of elementwise divide for integral type to match Python.

* Fix CPU codegen.

* Temp fix: Disable failing UT for IntelGPU

* Divide: Add constructor option to specify rounding mode for Integral types.

* Update serializer to support legacy Divide dump.

* Restore modified UT.

* Rename PlaidML_Executable::save -> save_as_format

* Repair regression in PlaidML tensor impl

This was caused by the recent removal of the offset parameter for tensor read/write operations -- we missed a
spot where read/write were being called for synchronization purposes.

* Disable a few more PlaidML tests pending triage

* Skip elision of reshape->reshape

It turns out this doesn't work, because the downstream reshape's input_order axis vector is incorrect if the
upstream reshape is removed.

* Add element type to PlaidML tensor debug output

* Use nGraph booleans for PlaidML boolean data

We'd previously been using i8; that's been deprecated for boolean data now that we have an explicit boolean
element type.

* Set PlaidML convolution output shapes correctly

We weren't transposing the output shape; we were computing the right data, but the incorrect shape metadata
causes validation to fail.

* Add a PlaidML implicit broadcast op

Better nGraph shape validation was tripping up PlaidML's use of a reshape to replace explicit broadcasts with
implicit NumPy-style broadcasts (since the reshape's output shape would be incorrect for the downstream
elementwise operation).  Adding this implicit broadcast operation lets PlaidML tell nGraph something useful
about the shapes, making validation pass (when it's otherwise correct).

* GenerateMask correction
Add an attribute that controls if the seed should be set on each use
Convert to new virtual method for description implementatin

* Support for switching to dynamic attributes.

* GenerateMask changes in CPU backend (#3042)

* Add CPU builder and kernel for new GenerateMask API

* Remove dead code

* Fix unit-test, PR feedback, file permissions

* Disable new test for non-supporting backends

* Fix CI error

* Codegen support

* Style check

* Fix CI error

* Unit tests for MatMulInteger

* Add ONNX MatMulInteger op

* Add QuantizedLinearMatmulInteger builder

* Additional unit test

* Exclude tests on nVidia GPU backend

* Add 4D test case

* Enable >2D MatMulInteger

* Refactoring to MatMulFactory - step 1

* Refactoring to MatMulFactory - step 2

* Remove `using namespace ngraph` to make `Node` unambiguous.

* Disable quantized ops tests on GPU backend

* Remove unused `includes`

* Remove redundant dynamic_pointer_cast

* Remove redundant `move`

* Add const correctness

* Code review comments

* Style apply

* Add documentation

* Use more complex shapes in tests