* WIP

* All but two unit tests passing

* Explanatory comment

* Cleanup

* A bit of cleanup stemming from review comments

* Rewrite to use LoopNestBuilder

* Remove unnecessary check from CompiledKernel

* Removed some dead-ish code I missed

* Switch to camelCase in lowerer.cpp

* Fix assignment of resIndexHandles that was triggering an assert

* Add in some safety checks

* dyn_cast -> cast

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.

* - 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

* Fix incorrect callback lookup

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

* Clean up

* revert getCallDecl change

* style

* Support MLIR lowering of Relu

* Use EDSC comparison

* style-apply

* Use .inc file for Conversion classes list

* Disable i32 Relu for plaidml

Improvements and changes in dialect conversion APIs. Type conversion
and op rewriting are now a bit more independent mechanisms. Dialect
conversion now requires a target object with the legal transformations
to be applied. Legality checks to detect illegal operations after
dialect conversion have been strengthened significantly.

MLIR LLVM commit:
commit 82d5084cf28bca7a0eebec283cdeae118dfa1d11 (HEAD)
Author: Jacques Pienaar <jpienaar@google.com>
Date:   Wed Jun 26 11:12:40 2019 -0700

    Split out TranslateClParser and add new parse method that reuses
    SourceMgr.

MLIR commit:
commit 82d5084cf28bca7a0eebec283cdeae118dfa1d11 (HEAD)
Author: Jacques Pienaar <jpienaar@google.com>
Date:   Wed Jun 26 11:12:40 2019 -0700

    Split out TranslateClParser and add new parse method that reuses
    SourceMgr.

Add support for any rank ArgMin. Hacky workaround for IndexType tensors by doing type-conversion after operation

Unused variables.

It refactor MLIR dump code in MLIRCompiler and adds missing dumps for
some steps of the lowering. It also removes some redundant dump for
affine dialect.

Element type was not lowered.

It extends dot definition to be able to deal with operands that are not
square tensors. It also fixes a bug in the lowerer related to that.

* [MLIR] Replace MatmulBiasOp with DotOp

We disable CPUFusion if MLIR is enabled to avoid CPU specific ops to be
introduced for now.

* Upgrade MLIR. Several code fixes based on API changes

* Fixes due to DialectConv API changes

* style-apply

* PR fixes

* Move dialect and types to mlir namespace

* PR fixes and some cleanup

* Merge fix

* Enable module verification. Fix FakeInput result type to memref

* style-apply

* Initial td file. Cmake changes

* Move all ops to .td file.

* Added few more opcodes to show-case. Fixed PR feedback

* Remove NG_ prefix of opcode records. Some fixes

* Added some doc

*  Adding back NG prefix

* Bug fix in MLIR gen

* Use NGRAPH export macros instead of CPU

* Move code to ngmlir namespace

* Move MLIR to src/contrib/mlir

* Create MLIR as cmake external project. Clone and build via ngraph cmake

* Moved code and enabled compilation. Need to clone and build MLIR/LLVM during cmake step, so find_package can work

* clone and build llvm/mlir during configuration. Compiles now. Needs more testing

* Force DEX only if MLIR is ON

* Remove extra cmake file. Style-apply

* Remove redundant files in cpu/mlir

* Update CODEOWNERS. Check for ninja and fail if not found

* Fixing post merge issues

* Implements a simple memory manager that just does malloc for now. Pointers are freed during cleanup.
* Enable JIT call-back to memory manager to allocate temps.
* Memory manager pointer is passed to the JIT'ed code upon invocation. That makes the code re-entrant from different threads in case the code is shared among identical sub-graphs that are executed in parallel.

The following test should work now:
NGRAPH_MLIR_DUMP_ALL=1 NGRAPH_MLIR=1 test/unit-test '--gtest_filter=CPU.dot2d'

* Link MLIR static libs to cpu backend

* Use LLVMConfig.cmake

* Initial commit. Link fails with undefined reference to typeinfo for mlir::Dialect

* Added AddOp

* initial compiler class

* Initialize module/function, and map tensors to arguments

* Code compiles. Moved MLIR building to correct DEX handler

* NGDialect code-gen working

* Use vector instead of sets for i/o tensors. Use functor in executor

* Misc fixes

* style-apply

* WIP: Adding support for dialect lowering.

* WIP: Lowered to affine. Crash on constant ops have side effects in Constant Folding

* Fixed missing whole package linkage.

* Removed fake instruction and update func type

*  Enable lowering to LLVM dialect and IR

* Made loop nest builder handle any rank

* Fixes per PR feedback. Major ones:
- Removed ngdialect namespace
- renamed dialect classes to start with NG prefixwq:w

* Add unreachable assert

* Add reading of LLVM options from an env var MLIR_LLVM_OPTIONS (#5)