implementation.h

This header exposes the core C++ shape inference implementation, including context objects consumed by schema inferencers and helper APIs such as InferShapes and InferFunctionOutputTypes.

namespace ONNX_LIGHT_NAMESPACE

namespace shape_inference

Typedefs

using ModelLocalFunctionsMap = std::unordered_map<std::string, const FunctionProto*>

using DataValueMap = std::unordered_map<std::string, TensorShapeProto>

Functions

std::string GetValueCaseString(const TypeProto &type)

void checkShapesAndTypes(const TypeProto_Sequence &inferredType, const TypeProto_Sequence &existingType)

void checkShapesAndTypes(const TypeProto &inferred_type, const TypeProto &existing_type)

template<typename TensorTypeProto>
void GenerateSymbolicShape(TensorTypeProto *inferred_type, SymbolTable &symbol_table)

void MaterializeSymbolicShape(TypeProto *inferred_type, SymbolTable &symbol_table)

void mergeShapesAndTypes(const TypeProto_Tensor &inferred_type, TypeProto_Tensor *existing_type)

void mergeShapesAndTypes(const TypeProto_SparseTensor &inferred_type, TypeProto_SparseTensor *existing_type)

void mergeShapesAndTypes(const TypeProto_Sequence &inferredType, TypeProto_Tensor *existingType)

void mergeShapesAndTypes(const TypeProto &inferred_type, TypeProto *existing_type)

void InferShapes(GraphProto *g, const std::unordered_map<std::string, int> &opset_imports, const ISchemaRegistry *schema_registry = OpSchemaRegistry::Instance(), const ShapeInferenceOptions &options = ShapeInferenceOptions(), const ModelLocalFunctionsMap &in_model_functions = {})

ModelLocalFunctionsMap is a map of function id -> model local function proto All the ONNX helper utilities expect the function id == <function_proto.domain>:<function_proto.name>

void InferShapes(ModelProto &m, const ISchemaRegistry *schema_registry = OpSchemaRegistry::Instance(), const ShapeInferenceOptions &options = ShapeInferenceOptions(), DataValueMap *generated_shape_data_by_name = nullptr)

void InferShapes(const std::string &model_path, const std::string &save_path = "", const ISchemaRegistry *schema_registry = OpSchemaRegistry::Instance(), const ShapeInferenceOptions &options = ShapeInferenceOptions(), DataValueMap *generated_shape_data_by_name = nullptr)

void InferShapeForFunctionNode(const FunctionProto &func, const ISchemaRegistry *schema_registry, InferenceContext &ctx, const ShapeInferenceOptions &options = ShapeInferenceOptions(), const ModelLocalFunctionsMap &model_local_functions_map = {}, SymbolTable *symbol_table = nullptr, DataValueMap *generated_shape_data_by_name = nullptr)

ModelLocalFunctionsMap is a map of function id -> model local function proto All the ONNX helper utilities expect the function id == <function_proto.domain>:<function_proto.name>

void InferShapeForFunctionNode(const FunctionProto &func_proto, const std::unordered_map<std::string, int> &func_opset_imports, const ISchemaRegistry *schema_registry, InferenceContext &ctx, const ShapeInferenceOptions &options = ShapeInferenceOptions(), const ModelLocalFunctionsMap &model_local_functions_map = {}, SymbolTable *symbol_table = nullptr, DataValueMap *generated_shape_data_by_name = nullptr)

ModelLocalFunctionsMap is a map of function id -> model local function proto All the ONNX helper utilities expect the function id == <function_proto.domain>:<function_proto.name>

std::vector<TypeProto> InferFunctionOutputTypes(const FunctionProto &function_proto, const std::vector<TypeProto> &input_types, const std::vector<AttributeProto> &attributes)

Apply type-and-shape-inference based checks to a Function body. Returns the inferred types of the outputs of the function. Inference depends on the types of the inputs of the function as well as the attribute values supplied. A TypeProto with value_case() == TypeProto::ValueCase::VALUE_NOT_SET is used for missing optional parameters.

std::string GetErrorWithNodeInfo(const NodeProto &n, const std::runtime_error &err)

void TraverseGraphsToAddExistingSymbols(const GraphProto &g, SymbolTable &symbol_table)

struct DataPropagationContextImpl : public ONNX_LIGHT_NAMESPACE::DataPropagationContext

#include <implementation.h>

Public Functions

inline DataPropagationContextImpl(NodeProto &n, const std::unordered_map<std::string, TypeProto*> &valueTypesByName, const std::unordered_map<std::string, const TensorProto*> &inputDataByName, DataValueMap &generatedShapeData)

inline virtual const AttributeProto *getAttribute(const std::string &name) const override

inline virtual size_t getNumInputs() const override

inline virtual const TypeProto *getInputType(size_t index) const override

inline virtual size_t getNumOutputs() const override

inline virtual const TypeProto *getOutputType(size_t index) const override

template<typename INTEGER>
inline void vectorToTensorShapeProto(const std::vector<INTEGER> &input_vals, TensorShapeProto &converted_tsp) const

inline virtual const TensorShapeProto *getInputData(size_t index) override

inline virtual void addOutputData(size_t index, TensorShapeProto &&tsp) override

Public Members

std::vector<const TensorProto*> allInputData_

std::unordered_map<size_t, std::string> inputIndexToNameMap_

std::unordered_map<size_t, std::string> outputIndexToNameMap_

std::vector<const TypeProto*> allInputTypes_

std::vector<TypeProto> allOutputTypes_

DataValueMap &generatedShapeData_

std::unordered_map<std::string, const AttributeProto*> attributesByName_

struct GraphInferenceContext

#include <implementation.h>

Public Functions

inline GraphInferenceContext(const std::unordered_map<std::string, TypeProto*> &outer_scope_value_types_by_name_in, std::unordered_map<std::string, int> opset_imports_in, SymbolTable *symbol_table_in = nullptr, const ModelLocalFunctionsMap &model_local_functions_in = {}, const ISchemaRegistry *schema_registry_in = OpSchemaRegistry::Instance(), DataValueMap *generated_shape_data_by_name_in = nullptr, const int ir_version_in = IR_VERSION)

Public Members

const std::unordered_map<std::string, TypeProto*> *outer_scope_value_types_by_name

const std::unordered_map<std::string, int> opset_imports

SymbolTable *symbol_table

const ModelLocalFunctionsMap &model_local_functions

const ISchemaRegistry *schema_registry

DataValueMap *generated_shape_data_by_name

const int ir_version

class GraphInferencerImpl : public ONNX_LIGHT_NAMESPACE::GraphInferencer

#include <implementation.h>

Public Functions

inline GraphInferencerImpl(GraphProto &g, GraphInferenceContext &context)

inline GraphInferencerImpl(GraphProto &g, GraphInferenceContext &context, const ShapeInferenceOptions &options)

virtual std::vector<const TypeProto*> doInferencing(const std::vector<const TypeProto*> &input_types, const std::vector<const TensorProto*> &input_data) override

Private Members

GraphProto *g_

GraphInferenceContext *context_

ShapeInferenceOptions options_

struct InferenceContextImpl : public ONNX_LIGHT_NAMESPACE::InferenceContext

#include <implementation.h>

Public Functions

inline InferenceContextImpl(NodeProto &n, const std::unordered_map<std::string, TypeProto*> &valueTypesByName, const std::unordered_map<std::string, const TensorProto*> &inputDataByName, const std::unordered_map<std::string, const SparseTensorProto*> &inputSparseDataByName, const ShapeInferenceOptions &options, DataValueMap *generatedShapeData = nullptr, GraphInferenceContext *graphInferenceContext = nullptr)

inline virtual const AttributeProto *getAttribute(const std::string &name) const override

inline virtual size_t getNumInputs() const override

inline virtual const TypeProto *getInputType(size_t index) const override

inline virtual const TensorProto *getInputData(size_t index) const override

inline virtual const TensorShapeProto *getSymbolicInput(size_t index) const override

inline virtual const SparseTensorProto *getInputSparseData(size_t index) const override

inline virtual size_t getNumOutputs() const override

inline virtual TypeProto *getOutputType(size_t index) override

inline virtual GraphInferencer *getGraphAttributeInferencer(const std::string &attr_name) override

inline virtual std::string getDisplayName() const override

Public Members

std::vector<const TensorProto*> allInputData_

std::vector<const SparseTensorProto*> allInputSparseData_

std::vector<const TensorShapeProto*> allShapeInputData_

std::unordered_map<std::string, const AttributeProto*> attributesByName_

std::unordered_map<std::string, GraphProto*> graphProtoAttributesByName_

std::vector<const TypeProto*> allInputTypes_

std::vector<TypeProto> allOutputTypes_

GraphInferenceContext *graphInferenceContext_

mutable std::unordered_map<std::string, std::unique_ptr<GraphInferencer>> graphAttributeInferencers_

ShapeInferenceOptions options_

NodeProto *node_

class SymbolTableImpl : public ONNX_LIGHT_NAMESPACE::SymbolTable

#include <implementation.h>

Public Functions

SymbolTableImpl() = default

inline virtual void addFromGraph(const GraphProto &g) override

inline virtual std::string createNew(const std::string &symbol_prefix) override

Private Functions

template<typename TensorTypeProto>
inline void AddExistingSymbolicDims(const TensorTypeProto &tensorType)

inline void AddExistingSymbolicDims(const TypeProto &typeProto)

inline void AddExistingSymbolicDims(const utils::RepeatedProtoField<ValueInfoProto> &protos)

Private Members

unsigned int index_ = {0}

std::unordered_set<std::string> existing_symbols