[MLIR] Initial PoC: NG dialect, dialect code-gen, dialect lowering to affine, no JIT yet

* 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)

src/ngraph/runtime/cpu/CMakeLists.txt

@@ -16,6 +16,13 @@
 
 include(FindOpenMP)
 
+set(MLIR_SRC
+    mlir/dialect/dialect.cpp
+    mlir/dialect/type.cpp
+    mlir/dialect/ops.cpp
+    mlir/compiler.cpp
+    mlir/lowerer.cpp
+)
 set(SRC
     cpu_backend.cpp
     cpu_builder.cpp
@@ -123,6 +130,7 @@ set(SRC
     pass/cpu_rnn_fusion.cpp
     pass/cpu_workspace_insertion.cpp
     ngraph_version.cpp
+    ${MLIR_SRC}
 )
 
 if (NOT NGRAPH_DEX_ONLY)
@@ -197,11 +205,20 @@ if (NGRAPH_CPU_ENABLE)
     endif()
     target_include_directories(cpu_backend SYSTEM PUBLIC libmkldnn)
 
+    if (NOT APPLE AND NOT MSVS)
+        # CPU backend uses third-party libraries like Eigen that might be linked in and
+        # exported by other DSOs as well. In the absence of versioning, this could lead to the
+        # CPU backend picking up the wrong version or even multiple versions of the
+        # third-party library. -Bsymbolic-functions tells the linker to prefer the internal
+        # version inside cpu_backend over what is available through the global symbol table
+        set_property(TARGET cpu_backend APPEND PROPERTY LINK_FLAGS "-Wl,-Bsymbolic-functions -Wl,--exclude-libs=ALL")
+    endif()
+
     # Link LLVM and MLIR
     find_package(LLVM REQUIRED CONFIG)
     message(STATUS "Found LLVM ${LLVM_PACKAGE_VERSION}")
     message(STATUS "Using LLVMConfig.cmake in: ${LLVM_DIR}")
-    
+    message(STATUS "LLVM RTTI is ${LLVM_ENABLE_RTTI}")
     add_definitions(${LLVM_DEFINITIONS})
 
     target_include_directories(cpu_backend PRIVATE ${LLVM_INCLUDE_DIRS})
@@ -214,7 +231,7 @@ if (NGRAPH_CPU_ENABLE)
 
     # Link MLIR libs
     target_link_libraries(
-      cpu_backend PRIVATE 
+      cpu_backend PUBLIC 
       MLIRAnalysis
       MLIREDSC
       MLIRExecutionEngine
@@ -225,24 +242,36 @@ if (NGRAPH_CPU_ENABLE)
       MLIRTargetLLVMIR
       MLIRTransforms
       MLIRSupport
-      MLIRAffineOps
-      MLIRStandardOps
     )
+    # some libs need whole archive linkage because of Globals static initialization
+    # TODO: move this helper somewhere else. 
+    function(whole_archive_link target)
+      if("${CMAKE_SYSTEM_NAME}" STREQUAL "Darwin")
+        set(link_flags "-Llib -Wl,-all_load ")
+        FOREACH(LIB ${ARGN})
+          string(CONCAT link_flags ${link_flags} "${LIB}")
+        ENDFOREACH(LIB)
+      else()
+        set(link_flags "-Llib -Wl,--whole-archive,")
+        FOREACH(LIB ${ARGN})
+          string(CONCAT link_flags ${link_flags} "${LIB},")
+        ENDFOREACH(LIB)
+        string(CONCAT link_flags ${link_flags} "--no-whole-archive")
+      endif()
+      message(STATUS "MLIR Ops link flag: ${link_flags}" )
 
+      set_target_properties(${target} PROPERTIES LINK_FLAGS ${link_flags})
+    endfunction(whole_archive_link)
+
+    whole_archive_link(cpu_backend
+        ${LLVM_BUILD_LIBRARY_DIR}/libMLIRAffineOps.a
+        ${LLVM_BUILD_LIBRARY_DIR}/libMLIRStandardOps.a
+    )
     # Link LLVM libs
     target_link_libraries(
       cpu_backend PRIVATE 
       ${llvm_libs}
     )
 
-    if (NOT APPLE AND NOT MSVS)
-        # CPU backend uses third-party libraries like Eigen that might be linked in and
-        # exported by other DSOs as well. In the absence of versioning, this could lead to the
-        # CPU backend picking up the wrong version or even multiple versions of the
-        # third-party library. -Bsymbolic-functions tells the linker to prefer the internal
-        # version inside cpu_backend over what is available through the global symbol table
-        set_property(TARGET cpu_backend APPEND PROPERTY LINK_FLAGS "-Wl,-Bsymbolic-functions -Wl,--exclude-libs=ALL")
-    endif()
-
     install(TARGETS cpu_backend DESTINATION ${NGRAPH_INSTALL_LIB})
 endif()

src/ngraph/runtime/cpu/cpu_external_function.cpp

@@ -191,6 +191,7 @@
 #include "ngraph/runtime/cpu/pass/cpu_rnn_fusion.hpp"
 #include "ngraph/runtime/cpu/pass/cpu_workspace_insertion.hpp"
 #include "ngraph/runtime/cpu/pass/halide_subgraph_extraction.hpp"
+#include "ngraph/runtime/cpu/mlir/compiler.hpp"
 
 using namespace std;
 using namespace ngraph;
@@ -1392,6 +1393,12 @@ void runtime::cpu::CPU_ExternalFunction::build(ngraph::pass::PassConfig& pass_co
     // After processing inputs, outputs, constants, and intermediates, set the buffer size.
     m_buffer_size = buffer_index;
 
+    if (std::getenv("NGRAPH_MLIR") != nullptr)
+    {
+        // Initialize MLIR compiler
+        MLIRCompiler::init_mlir();
+    }
+
     for (shared_ptr<Node> node : m_function->get_ordered_ops())
     {
         if (node->is_parameter() || node->is_constant())

src/ngraph/runtime/cpu/mlir/compiler.hpp

+//*****************************************************************************
+// Copyright 2017-2019 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//*****************************************************************************
+#pragma once
+
+#include "mlir/IR/Attributes.h"
+#include "mlir/IR/Builders.h"
+#include "mlir/IR/Location.h"
+#include "mlir/IR/MLIRContext.h"
+#include "mlir/IR/Module.h"
+#include "mlir/IR/StandardTypes.h"
+#include "mlir/IR/Types.h"
+#include "mlir/StandardOps/Ops.h"
+
+#include "llvm/ADT/STLExtras.h"
+
+#include "ngraph/descriptor/tensor.hpp"
+#include "ngraph/node.hpp"
+
+namespace ngraph
+{
+    namespace runtime
+    {
+        namespace cpu
+        {
+            class MLIRCompiler
+            {
+            public:
+                using TensorList = std::vector<descriptor::Tensor*>;
+                using TypeList = llvm::SmallVector<mlir::Type, 4>;
+
+                MLIRCompiler(const std::vector<const Node*>& sub_graph)
+                    : m_sub_graph(sub_graph.begin(), sub_graph.end())
+                {
+                }
+                static void init_mlir();
+                // compiles and runs a subgraph in MLIR
+                void compile();
+
+            private:
+                struct TensorInfo
+                {
+                    mlir::Value* m_value; /* mlir value this tensor maps to */
+                    // More info here ?
+                };
+
+            private:
+                void build_module();
+                void lower_dialect();
+                void lower_to_llvm();
+                void build_tensors_list();
+                mlir::Type get_mlir_type(const descriptor::Tensor* tensor);
+                mlir::Type get_mlir_type(const element::Type& type);
+                TensorInfo get_tensor_value(descriptor::Tensor* tensor);
+                void update_tensor_value(descriptor::Tensor* tensor, mlir::Value* value);
+                void build_ng_dialect();
+
+                template <typename OP>
+                static mlir::Value* create_op(MLIRCompiler& compiler, const ngraph::Node* ng_node)
+                {
+                    throw std::runtime_error("Unimplemented op '" + ng_node->description() +
+                                             "' in MLIR Compiler");
+                }
+
+                template <typename BinOp>
+                mlir::Value* create_binary_op(const ngraph::Node* ng_node);
+                void create_return();
+
+            private:
+                mlir::MLIRContext m_context;
+                std::unique_ptr<mlir::Module> m_module;
+                std::unique_ptr<mlir::FuncBuilder> m_builder;
+
+                using TensorToInfo = std::pair<descriptor::Tensor*, TensorInfo>;
+                using TensorToInfoMap = std::unordered_map<descriptor::Tensor*, TensorInfo>;
+                using MLIRCompOpFunction =
+                    std::function<mlir::Value*(MLIRCompiler& compiler, const ngraph::Node*)>;
+                using MLIRCompOpMap = std::unordered_map<std::type_index, MLIRCompOpFunction>;
+
+                llvm::SmallVector<const Node*, 4> m_sub_graph;
+                // Maps tensor to the value it represents in the IR
+                // use for MLIR dialect gen
+                TensorToInfoMap m_tensor_to_value_map;
+                // List of input and output tensors in the graph
+                TensorList m_ip_tensors, m_op_tensors;
+                static const MLIRCompOpMap op_dispatcher;
+            };
+        }
+    }
+}

src/ngraph/runtime/cpu/mlir/dialect/dialect.cpp

+//*****************************************************************************
+// Copyright 2017-2019 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//*****************************************************************************
+
+#include "dialect.hpp"
+#include "ops.hpp"
+#include "type.hpp"
+namespace ngraph
+{
+    using namespace runtime::cpu;
+
+    /// Register a dialect and its types
+    /// Usage:
+    /// mlir::registerDialect<ngraph::runtime::cpu::ngdialect::Dialect>();
+    NGDialect::NGDialect(mlir::MLIRContext* ctx)
+        : mlir::Dialect("ng", ctx)
+    {
+        addTypes<NGTensorType>();
+        addOperations<NG_AddOp>();
+        addOperations<NG_ReturnOp>();
+        addOperations<NG_FakeOutput>();
+    }
+
+    void NGDialect::printType(mlir::Type type, raw_ostream& os) const
+    {
+        auto arrayTy = type.dyn_cast<NGTensorType>();
+        if (!arrayTy)
+        {
+            NGRAPH_ASSERT(0) << "Incorrect type to print?";
+        }
+        os << "tensor";
+        if (!arrayTy.getShape().empty())
+        {
+            os << "<";
+            mlir::interleaveComma(arrayTy.getShape(), os);
+            os << ">";
+        }
+    }
+}

src/ngraph/runtime/cpu/mlir/dialect/dialect.hpp

+//*****************************************************************************
+// Copyright 2017-2019 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//*****************************************************************************
+#pragma once
+
+#include "mlir/IR/Dialect.h"
+#include "mlir/IR/Function.h"
+#include "mlir/IR/OpDefinition.h"
+#include "mlir/IR/OpImplementation.h"
+#include "mlir/IR/StandardTypes.h"
+#include "mlir/IR/TypeSupport.h"
+#include "mlir/IR/Types.h"
+#include "ngraph/assertion.hpp"
+
+namespace ngraph
+{
+    namespace runtime
+    {
+        namespace cpu
+        {
+            class NGDialect : public mlir::Dialect
+            {
+            public:
+                explicit NGDialect(mlir::MLIRContext* ctx);
+                mlir::Type parseType(llvm::StringRef tyData, mlir::Location loc) const override
+                {
+                    NGRAPH_ASSERT(0) << "Unsupported type parsing.";
+                    return mlir::Type();
+                }
+                void printType(mlir::Type type, llvm::raw_ostream& os) const override;
+            };
+        }
+    }
+}

src/ngraph/runtime/cpu/mlir/dialect/ops.cpp

+//*****************************************************************************
+// Copyright 2017-2019 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//*****************************************************************************
+#include "ops.hpp"
+#include "assertion.hpp"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/Regex.h"
+#include "llvm/Support/raw_ostream.h"
+#include "type.hpp"
+
+using llvm::ArrayRef;
+using llvm::raw_ostream;
+using llvm::raw_string_ostream;
+using llvm::SmallVector;
+using llvm::StringRef;
+using llvm::Twine;
+
+namespace ngraph
+{
+    namespace runtime
+    {
+        namespace cpu
+        {
+            template <typename T>
+            static mlir::LogicalResult verifyBinOperands(T* op)
+            {
+                if (!op->getOperand(0)->getType().template isa<NGTensorType>())
+                {
+                    std::string msg;
+                    raw_string_ostream os(msg);
+                    os << "expects a Tensor type for LHS, got " << op->getOperand(0)->getType();
+                    return op->emitOpError(os.str());
+                }
+                if (!op->getOperand(1)->getType().template isa<NGTensorType>())
+                {
+                    std::string msg;
+                    raw_string_ostream os(msg);
+                    os << "expects a Tensor type for RHS, got " << op->getOperand(0)->getType();
+                    return op->emitOpError(os.str());
+                }
+                return mlir::success();
+            }
+
+            template <typename T>
+            static mlir::LogicalResult verifySingleOperand(T* op)
+            {
+                if (!op->getOperand()->getType().template isa<NGTensorType>())
+                {
+                    std::string msg;
+                    raw_string_ostream os(msg);
+                    os << "expects a Tensor Type for its argument, got "
+                       << op->getOperand()->getType();
+                    return op->emitOpError(os.str());
+                }
+                return mlir::success();
+            }
+        }
+    }
+
+    void runtime::cpu::NG_FakeOutput::build(mlir::Builder* builder,
+                                            mlir::OperationState* state,
+                                            mlir::Type resultType)
+    {
+        state->types.push_back(std::move(resultType));
+    }
+
+    mlir::LogicalResult runtime::cpu::NG_FakeOutput::verify()
+    {
+        // TODO: Verify returned tensor types must match function return type.
+        return mlir::success();
+    }
+
+    void runtime::cpu::NG_AddOp::build(mlir::Builder* builder,
+                                       mlir::OperationState* state,
+                                       mlir::Value* lhs,
+                                       mlir::Value* rhs)
+    {
+        state->types.push_back(lhs->getType());
+        state->operands.push_back(lhs);
+        state->operands.push_back(rhs);
+    }
+
+    mlir::LogicalResult runtime::cpu::NG_AddOp::verify()
+    {
+        // TODO: verify matching elt types
+        verifyBinOperands(this);
+        return mlir::success();
+    }
+
+    void runtime::cpu::NG_ReturnOp::build(mlir::Builder* builder,
+                                          mlir::OperationState* state,
+                                          std::vector<mlir::Value*> value_list)
+    {
+        for (auto value : value_list)
+        {
+            if (value)
+                state->operands.push_back(value);
+        }
+    }
+
+    mlir::LogicalResult runtime::cpu::NG_ReturnOp::verify()
+    {
+        // TODO: Verify returned tensor types must match function return type.
+        return mlir::success();
+    }
+}

src/ngraph/runtime/cpu/mlir/dialect/ops.hpp

+//*****************************************************************************
+// Copyright 2017-2019 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//*****************************************************************************
+#pragma once
+
+#include <cstdarg>
+#include "mlir/IR/Builders.h"
+#include "mlir/IR/OpDefinition.h"
+#include "mlir/IR/OperationSupport.h"
+#include "mlir/IR/StandardTypes.h"
+#include "mlir/Support/STLExtras.h"
+
+namespace ngraph
+{
+    namespace runtime
+    {
+        namespace cpu
+        {
+            // Fake instructions
+            class NG_FakeOutput : public mlir::Op<NG_FakeOutput,
+                                                  mlir::OpTrait::NOperands<0>::Impl,
+                                                  mlir::OpTrait::OneResult,
+                                                  mlir::OpTrait::HasNoSideEffect>
+            {
+            public:
+                static llvm::StringRef getOperationName() { return "ng.fake.output"; }
+                mlir::LogicalResult verify();
+                static void
+                    build(mlir::Builder* builder, mlir::OperationState* state, mlir::Type type);
+                /// Inherit constructor.
+                using Op::Op;
+            };
+
+            // Binary instructions
+            class NG_AddOp : public mlir::Op<NG_AddOp,
+                                             mlir::OpTrait::NOperands<2>::Impl,
+                                             mlir::OpTrait::OneResult,
+                                             mlir::OpTrait::HasNoSideEffect>
+            {
+            public:
+                static llvm::StringRef getOperationName() { return "ng.add"; }
+                /// custom verification
+                mlir::LogicalResult verify();
+                static void build(mlir::Builder* builder,
+                                  mlir::OperationState* state,
+                                  mlir::Value* lhs,
+                                  mlir::Value* rhs);
+
+                /// Convenience accessor for LHS of the expression.
+                mlir::Value* getLHS() { return getOperand(0); }
+                /// Convenience accessor for RHS of the expression.
+                mlir::Value* getRHS() { return getOperand(1); }
+                /// Inherit constructor.
+                using Op::Op;
+            };
+
+            /// Return operations terminate blocks (and functions as well). They take a
+            /// single argument and the type must match the function return type.
+            class NG_ReturnOp : public mlir::Op<NG_ReturnOp,
+                                                mlir::OpTrait::VariadicOperands,
+                                                mlir::OpTrait::ZeroResult,
+                                                mlir::OpTrait::IsTerminator>
+            {
+            public:
+                static llvm::StringRef getOperationName() { return "ng.return"; }
+                /// Operations can add custom verification beyond the traits they define.
+                mlir::LogicalResult verify();
+
+                /// Interface to mlir::Builder::create<PrintOp>(...)
+                /// This method populate the `state` that MLIR use to create operations.
+                /// The `toy.return` operation accepts an optional single array as an argument
+                /// and does not have any returned value.
+                static void build(mlir::Builder* builder,
+                                  mlir::OperationState* state,
+                                  std::vector<mlir::Value*> value_list);
+
+                /// Return true if there is a returned value.
+                bool hasOperand() { return 0 != getNumOperands(); }
+                /// Helper to return the optional operand. Caller must check if the operand
+                /// is present before calling this.
+                mlir::Value* getOperand() { return getOperation()->getOperand(0); }
+                mlir::Value* getOperand(unsigned i) { return getOperation()->getOperand(i); }
+                /// Inherit constructor.
+                using Op::Op;
+            };
+        }
+    }
+}
\ No newline at end of file

src/ngraph/runtime/cpu/mlir/dialect/type.cpp

+//*****************************************************************************
+// Copyright 2017-2019 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "type.hpp"
+
+#include "llvm/ADT/iterator_range.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/Regex.h"
+#include "llvm/Support/raw_ostream.h"
+#include "mlir/IR/Builders.h"
+#include "mlir/IR/StandardTypes.h"
+#include "mlir/Support/STLExtras.h"
+#include "ngraph/assertion.hpp"
+
+using llvm::ArrayRef;
+using llvm::raw_ostream;
+using llvm::raw_string_ostream;
+using llvm::SmallVector;
+using llvm::StringRef;
+using llvm::Twine;
+
+namespace ngraph
+{
+    using namespace runtime::cpu;
+    /// Creates TensorType objects. They all point to the same storage if
+    /// element type and shape are the same.
+    NGTensorType NGTensorType::get(mlir::MLIRContext* context, EltType eltType, Shape shape)
+    {
+        return Base::get(context, NGTypeKind::TENSOR_TYPE_ID, eltType, shape);
+    }
+
+    mlir::MemRefType NGTensorType::toMemref()
+    {
+        auto memRefType =
+            mlir::MemRefType::get(getShape(), getElementType(), {/* no map used */}, 0);
+        return memRefType;
+    }
+}

src/ngraph/runtime/cpu/mlir/dialect/type.hpp

+//*****************************************************************************
+// Copyright 2017-2019 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//*****************************************************************************
+#pragma once
+
+#include "mlir/IR/Dialect.h"
+#include "mlir/IR/Function.h"
+#include "mlir/IR/OpDefinition.h"
+#include "mlir/IR/OpImplementation.h"
+#include "mlir/IR/StandardTypes.h"
+#include "mlir/IR/TypeSupport.h"
+#include "mlir/IR/Types.h"
+
+namespace ngraph
+{
+    namespace runtime
+    {
+        namespace cpu
+        {
+            using llvm::raw_ostream;
+
+            enum NGTypeKind
+            {
+                // The enum starts at the range reserved for this dialect.
+                // These values are pre-defined in MLIR lib and not configurable from here.
+                NG_TYPE = mlir::Type::Kind::FIRST_PRIVATE_EXPERIMENTAL_0_TYPE,
+                TENSOR_TYPE_ID
+            };
+
+            using EltType = mlir::Type;
+            // TODO: Can we use ngraph::shape here (given the hashing requirements)
+            using Shape = llvm::ArrayRef<int64_t>;
+
+            /// Tensor Type storage. There is a unique instance per type attributes.
+            /// Tensor Type is combination of the element type and shape. Each different
+            /// shape is a unique type.
+            struct NGTensorTypeStorage : public mlir::TypeStorage
+            {
+                // Tensor key is its type and shape.
+                // This is called when the user requests a specific tensor type
+                using KeyTy = std::tuple<EltType, Shape>;
+
+                static unsigned hashKey(const KeyTy& key)
+                {
+                    return llvm::hash_combine(std::get<0>(key), std::get<1>(key));
+                }
+
+                bool operator==(const KeyTy& key) const
+                {
+                    return key == KeyTy(getElementType(), getShape());
+                }
+
+                static NGTensorTypeStorage* construct(mlir::TypeStorageAllocator& allocator,
+                                                      const KeyTy& key)
+                {
+                    // Deep copy the type shape over to MLIR context
+                    EltType eltType = std::get<0>(key);
+                    Shape shape = allocator.copyInto(std::get<1>(key));
+                    auto* storage = allocator.allocate<NGTensorTypeStorage>();
+                    return new (storage) NGTensorTypeStorage(eltType, shape);
+                }
+
+                Shape getShape() const { return m_shape; }
+                EltType getElementType() const { return m_eltType; }
+            private:
+                NGTensorTypeStorage(EltType eltType, Shape shape)
+                    : m_eltType(eltType)
+                    , m_shape(shape)
+                {
+                }
+
+            private:
+                EltType m_eltType;
+                Shape m_shape;
+            };
+
+            class NGTensorType
+                : public mlir::Type::TypeBase<NGTensorType, mlir::Type, NGTensorTypeStorage>
+            {
+            public:
+                using Base::Base;
+                EltType getElementType() const { return getImpl()->getElementType(); }
+                Shape getShape() const { return getImpl()->getShape(); }
+                int getRank() { return getShape().size(); }
+                /// convert to memref native MLIR type. Used for lowering.
+                mlir::MemRefType toMemref();
+                /// create a unique tensor type based on element type and shape.
+                static NGTensorType get(mlir::MLIRContext* context, EltType eltType, Shape shape);
+                /// for llvm RTTI
+                static bool kindof(unsigned kind) { return kind == NGTypeKind::TENSOR_TYPE_ID; }
+            };
+        }
+    }
+}

src/ngraph/runtime/cpu/mlir/lowerer.cpp

+//*****************************************************************************
+// Copyright 2017-2019 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//*****************************************************************************
+
+#include "lowerer.hpp"
+#include <map>
+#include "llvm/ADT/DenseSet.h"
+#include "mlir/EDSC/Builders.h"
+#include "mlir/EDSC/Helpers.h"
+#include "mlir/EDSC/Intrinsics.h"
+#include "mlir/IR/MLIRContext.h"
+#include "mlir/IR/StandardTypes.h"
+#include "mlir/Transforms/DialectConversion.h"
+#include "ngraph/assertion.hpp"
+#include "ngraph/runtime/cpu/mlir/dialect/ops.hpp"
+#include "ngraph/runtime/cpu/mlir/dialect/type.hpp"
+
+using namespace ngraph::runtime::cpu;
+// anonymous namespace
+// no need to expose any of the following outside of this file
+namespace
+{
+    using namespace mlir;
+    using namespace mlir::edsc;
+    using namespace ngraph::runtime::cpu;
+
+    class DialectLoweringPass;
+#include "op_lowerers.inc"
+
+    /// Use Dialect Converson Framework
+    class DialectLowerer : public DialectConversion
+    {
+    public:
+        DialectLowerer(DialectLoweringPass& pass)
+            : DialectConversion()
+            , m_pass(pass)
+        {
+        }
+
+        Type convertType(Type t) override;
+
+    protected:
+        // Initialize the list of converters.
+        llvm::DenseSet<DialectOpConversion*> initConverters(MLIRContext* context) override
+        {
+            return ConversionListBuilder<NG_AddOpConversion, NG_ReturnOpConversion>::build(
+                &allocator, context, m_pass);
+        }
+
+    private:
+        DialectLoweringPass& m_pass;
+        llvm::BumpPtrAllocator allocator;
+    };
+
+    /// Dialect Lowering Pass to affine ops
+    class DialectLoweringPass : public ModulePass<DialectLoweringPass>
+    {
+    public:
+        DialectLoweringPass()
+            : m_dialectLowerer(*this)
+        {
+        }
+        void runOnModule() override;
+        std::map<Value*, unsigned>& getOutputValueMap() { return m_outputValueMap; };
+        SmallVector<Value*, 4> buildOutputDefs(Operation* op, FuncBuilder& rewriter);
+
+    private:
+        void findOutputValues();
+        void fixOutputs();
+
+    private:
+        DialectLowerer m_dialectLowerer;
+        // maps output ng dialect values to args pos
+        std::map<Value*, unsigned> m_outputValueMap;
+        // list of results values to add to func signature
+        SmallVector<Value*, 4> m_loweredOutputValues;
+    };
+
+    Type DialectLowerer::convertType(Type t)
+    {
+        if (auto tensor = t.cast<NGTensorType>())
+        {
+            return tensor.toMemref();
+        }
+        return t;
+    }
+
+    void DialectLoweringPass::runOnModule()
+    {
+        // capture output values by looking for the Return and grabbing the values
+        // the order of the returned values matches the order of the lowered func signature for
+        // results. This is used to find the arg_id that a defined value maps to if it is an output
+        findOutputValues();
+
+        if (failed(m_dialectLowerer.convert(&getModule())))
+        {
+            getModule().getContext()->emitError(mlir::UnknownLoc::get(getModule().getContext()),
+                                                "Error lowering dialect\n");
+            signalPassFailure();
+        }
+        if (std::getenv("NGRAPH_MLIR_DUMP_ALL") != nullptr)
+        {
+            getModule().dump();
+        }
+        fixOutputs();
+        if (std::getenv("NGRAPH_MLIR_DUMP_ALL") != nullptr)
+        {
+            getModule().dump();
+        }
+    }
+
+    void DialectLoweringPass::findOutputValues()
+    {
+        auto f = getModule().getNamedFunction("main");
+        SmallVector<Value*, 4> outputList;
+        unsigned outputCount = 0;
+
+        // we find out output values by looking at returned values
+        // any return should return all outputs of the subgraph
+        f->walk<NG_ReturnOp>([this, &outputCount](NG_ReturnOp ret) {
+            for (unsigned i = 0; i < ret.getNumOperands(); i++)
+            {
+                this->m_outputValueMap.insert(std::pair<Value*, unsigned>(ret.getOperand(i), i));
+            }
+            NGRAPH_ASSERT(outputCount == 0 || outputCount == ret.getNumOperands())
+                << "Inconsistent returns in function";
+            outputCount = ret.getNumOperands();
+        });
+        // will be populated with lowered output values later
+        m_loweredOutputValues.resize(outputCount, nullptr);
+    }
+
+    // NGDialect converters
+    // ADD
+    SmallVector<Value*, 4> NG_AddOpConversion::rewrite(Operation* op,
+                                                       ArrayRef<Value*> operands,
+                                                       FuncBuilder& rewriter) const
+    {
+        auto add = op->cast<NG_AddOp>();
+        auto loc = add.getLoc();
+        Value *origResult, *newResult;
+
+        auto result = m_pass.buildOutputDefs(op, rewriter)[0];
+        NGRAPH_ASSERT(result->getType().isa<MemRefType>());
+        // NOte that builder's current function is still the original function body.
+        // use getBlock to get the new block instead.
+
+        // get new operands
+        Value* lhs = operands[0];
+        Value* rhs = operands[1];
+
+        ScopedContext scope(rewriter, loc);
+        // Views
+        MemRefView vRes(result), vLHS(lhs), vRHS(rhs);
+        // Index Values
+        IndexedValue iRes(result), iLHS(lhs), iRHS(rhs);
+        // Bounds Index Handles
+        auto lbs = vLHS.getLbs();
+        auto ubs = vLHS.getUbs();
+        // Loop induction vars
+        auto ivs = IndexHandle::makeIndexHandles(vLHS.rank());
+        auto pivs = IndexHandle::makeIndexHandlePointers(ivs);
+        // Steps
+        auto steps = vLHS.getSteps();
+
+        LoopNestBuilder(pivs, lbs, ubs, steps)({// single stmt body
+                                                iRes(ivs) = iLHS(ivs) + iRHS(ivs)});
+        // return result memref
+        return {result};
+    }
+
+    SmallVector<Value*, 4> NG_ReturnOpConversion::rewrite(Operation* op,
+                                                          ArrayRef<Value*> operands,
+                                                          FuncBuilder& rewriter) const
+    {
+        rewriter.create<ReturnOp>(op->getLoc());
+        return {};
+    }
+
+    SmallVector<Value*, 4> DialectLoweringPass::buildOutputDefs(Operation* op,
+                                                                FuncBuilder& rewriter)
+    {
+        auto& outputMap = getOutputValueMap();
+        SmallVector<Value*, 4> newResults;
+        for (auto origResult : op->getResults())
+        {
+            auto it = outputMap.find(origResult);
+            // create output def if this operation produces any sub-graph outputs
+            if (it != outputMap.end())
+            {
+                unsigned argId = (*it).second;
+                auto newResult = rewriter
+                                     .create<NG_FakeOutput>(
+                                         op->getLoc(),
+                                         m_dialectLowerer.convertType(
+                                             origResult->getType()) /* convert to lowered type */
+                                         )
+                                     .getResult();
+                newResults.push_back(newResult);
+                m_loweredOutputValues[argId] = newResult;
+            }
+        }
+        return newResults;
+    }
+
+    void DialectLoweringPass::fixOutputs()
+    {
+        auto context = getModule().getContext();
+        auto f = getModule().getNamedFunction("main");
+        mlir::Block* entryBlock = &*(f->begin());
+        auto oldFuncType = f->getType();
+        ArrayRef<mlir::Type> ipArgs = oldFuncType.getInputs();
+        ArrayRef<mlir::Type> opArgs = oldFuncType.getResults();
+        SmallVector<mlir::Type, 4> allArgs;
+
+        // Move all args as inputs in new type
+        for (auto type : ipArgs)
+        {
+            allArgs.push_back(type);
+        }
+        for (auto type : opArgs)
+        {
+            allArgs.push_back(type);
+            // add new value for result
+            entryBlock->addArgument(type);
+        }
+        // update type
+        auto newFuncType = mlir::FunctionType::get(allArgs, {}, context);
+        f->setType(newFuncType);
+
+        // RAUW fake outputs with result values
+        unsigned i = 0;
+        for (auto value : m_loweredOutputValues)
+        {
+            auto op = value->getDefiningOp();
+            NGRAPH_ASSERT(op->isa<NG_FakeOutput>()) << "output value not defined by fake output?";
+            value->replaceAllUsesWith(entryBlock->getArgument(oldFuncType.getNumInputs() + i));
+            op->erase();
+            i++;
+        }
+    }
+}
+
+namespace ngraph
+{
+    namespace runtime
+    {
+        namespace cpu
+        {
+            Pass* createDialectLoweringPass() { return new DialectLoweringPass(); }
+        }
+    }
+}

src/ngraph/runtime/cpu/mlir/lowerer.hpp

+//*****************************************************************************
+// Copyright 2017-2019 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//*****************************************************************************
+
+#pragma once
+
+#include "mlir/Pass/Pass.h"
+#include "mlir/Support/LLVM.h"
+
+namespace ngraph
+{
+    namespace runtime
+    {
+        namespace cpu
+        {
+            mlir::Pass* createDialectLoweringPass();
+        }
+    }
+}

src/ngraph/runtime/cpu/mlir/op_lowerers.inc

+//*****************************************************************************
+// Copyright 2017-2019 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//*****************************************************************************
+
+
+// Add new dialect ops lowerers to this file
+
+#define DECL_OP_CONV(OP) \
+class OP##Conversion : public mlir::DialectOpConversion \
+{\
+public:\
+    explicit OP##Conversion(mlir::MLIRContext *context, DialectLoweringPass& pass)\
+    : mlir::DialectOpConversion(ngraph::runtime::cpu::OP::getOperationName(), 1, context),\
+      m_pass(pass)\
+    {} \
+    SmallVector<Value *, 4> rewrite(Operation *op, ArrayRef<Value *> operands, FuncBuilder &rewriter) const override; \
+    DialectLoweringPass& m_pass;\
+};
+
+DECL_OP_CONV(NG_AddOp)
+DECL_OP_CONV(NG_ReturnOp)
+
+#undef DECL_OP_CONV