OperationRegistry.ixx File Reference

Provides a registry for neural network operations in the Mila framework. More...

#include <string>
#include <format>
#include <memory>
#include <optional>
#include <unordered_map>
#include <functional>
#include <stdexcept>
#include <type_traits>
#include <typeindex>
#include <vector>
import Compute.CudaMemoryResource;
import Dnn.TensorTraits;
import Compute.MemoryResource;
import Dnn.ComponentConfig;
import Compute.Precision;
import Compute.OperationType;
import Compute.CpuMemoryResource;
import Compute.OperationBase;
import Compute.DeviceType;
import Compute.UnaryOperation;
import Compute.CpuDevice;
import Compute.BinaryOperation;
import Compute.DeviceContext;
import Compute.CudaDevice;

Classes
struct	Mila::Dnn::Compute::FusedOpMeta
	Metadata for fused operations in the neural network. More...
class	Mila::Dnn::Compute::OperationRegistry
	A registry for operations that can be created based on operation names, type information, and device type. More...
struct	Mila::Dnn::Compute::OperationRegistry::TypeID
	Type ID structure to uniquely identify operations based on input types and device type. More...
struct	Mila::Dnn::Compute::OperationRegistry::TypeIDHash
	Hash function for TypeID to use in unordered_map. More...

Namespaces
namespace	Mila
namespace	Mila::Dnn
namespace	Mila::Dnn::Compute

Detailed Description

Provides a registry for neural network operations in the Mila framework.

This file implements a comprehensive registry system that manages neural network operations across different device types (CPU, CUDA) and with various data types. It enables:

• Registration of operation implementations for specific device types and data types
• Creation of operations based on type information and device contexts
• Support for operation variants with automatic fallback to default implementations
• Registration and discovery of fused operations for performance optimization

The OperationRegistry serves as the central hub for operation management in the compute framework, allowing the system to dynamically select the appropriate implementation based on the current context and requirements.