Aliasing prevents vectorization and other optimizations in LLVM (dot2d,
for example). I found that LLVM dialect contemplates the possibility of
previous dialects adding llvm.noalias attribute to function arguments so
that the NoAlias attribute is generated in LLVM-IR. This should be good for
now although we should discuss how to model this in Standard dialect and co.

* [WIP] Fix unreachable warnings in constant_folding.cpp

* Introduce NGRAPH_UNREACHABLE macro and silent unreachable warnings

* Fix typo and add missing NGRAPH_UNREACHABLE

* Some comments for replace_node

* Update src/ngraph/graph_util.hpp

@csullivan done figured it out
Co-Authored-By: Chris Sullivan <chris.sullivan@intel.com>

* Update graph_util.hpp

* - templatize computing binary elementwise
- added lowering support for Add, Sub, Multiply, Divide

* - Added Support for Greater and less Op

* -Add support for Minimum and Maximum

* use edsc::intrinsics::select instead of terenary operator

* Addressed PR comments

* - return after the conditional check

* Add instructions for building the NGRAPH_PLAIDML backend, ref on latest ToC

* Fix typo

* Reword paragraph to reference ngraph-plaidml build instruction page

* PR review suggestions added

* Fix incorrect callback lookup

* Use std.AllocOp(malloc) for static tensor allocations

* Clean up

* revert getCallDecl change

* style

Update jenkins-trigger.groovy

* Re-enable -Wswitch and -Wswitch-enum

* Collapse identical switch cases

* Fix (I think) compilation errors

* Partially fix missing cases in GPU xformer, so clang will show me the rest of them...

* One last enum value

* Fixes pprovenance issue for replace_node() when new nodes are added by the replacement.

* Updates unit test comment and adds one more.

* Fix shrink operator for uint* types

* Add a comment for negative_lambd=0 for uint*

* Helper function converting ONNX auto_pad into nGraph PadType.

Separate auto_pad support from explicitly provided paddings.

* Add support for more PadType values for GroupConvolutionTranspose.

* Pass auto_pad attribute value to nGraph operator.

* Helper class for generating ONNX pooling operators.

* Pass auto pad type to nGraph Convolution operator.

* Use pooling factory.

* Helper function calculating pads taking into account auto_pad attribute.

* Fix attribute type in UT ONNX models.

* Take auto_pad attribute value into account.

* Rename helper function and update doc.

* Retain old API for GroupConvoloutionTranspose

* Remove PadType::INVALID and use other approach to validate auto_pad
value.

* Style apply.

* Fix clang err on documentation style.

* Refactor get_auto_pad method.

* Fix segfaults on CentOS 7.

When using const reference the node member m_attributes had invalid
data.

* Fix used operator for reversing input sequences in LSTM.

* Fix backward pass for bidirectional LSTM.

* UT for LSTM with sequence_lens shorter than input sequence size.

* Skip LSTM UT using ReverseSequence since it is not supported yet on
PlaidML.

* Maintain control_deps in replace_node
Add helpers for mainting control_deps in other ways
Fix compiler warning about shadowing a local

* Clone control dependencies

* Export symbol for Windows

* Callers of copy_with_new_args

* Need copy

* Need shared pointer

* GOE needed

* goe

* goe

* Need pointer

* Review comments

Fix a mismatch after changing get_inputs API.

* Start splitting type_prop.cpp (just a few for now, to get feedback on general pattern)

* Split out some more type_prop stuff

* A bunch more

* Move one test out of type_prop into build_graph

* Split Reverse from ReverseSequence (oops), and fix a typo I noticed in dyn_reshape.cpp

* fix EOF newline

* Style.

* Add newline at eof.

Wait until performance counter data is set to release function for CODEGEN.

*  Add Gelu OP.

* CPU backend: Enable Erf UTs

* Add UT.

* Address review comment.

First cut at async interface for backend tensor allocation and execution

Handle negative padding

* Stabilize node sorting and fix some bugs.

* Review comments

* Fix broken tests

* Implement traverse nodes with pointers

* Let sort gather nodes for get_ordered_ops

* Use stacks for stacks

* Keep control deps ordered

* Optimize subgraph sort

* Add unordered map over function ops

* Don't recheck children

* Use vectors in stacks, avoid std::list::size()

* use bash method for style on Linux and Apple because it is faster

* cleanup

* address PR review comment

* make scripts executable

Update unit_test.manifest