Base class for all compute operations in the Mila neural network framework. More...
Public Member Functions | |
| OperationBase (OperationType operation_type, std::shared_ptr< DeviceContext > context) | |
| Constructs an OperationBase object with a specific device context and compute precision. | |
| virtual | ~OperationBase ()=default |
| Virtual destructor for the OperationBase class. | |
| std::shared_ptr< DeviceContext > | getDeviceContext () const |
| Gets the device context associated with this operation. | |
| DeviceType | getDeviceType () const |
| Gets the device type for this operation. | |
| virtual std::string | getName () const =0 |
| Gets the name of the operation. | |
| OperationType | getOperationType () const |
| Gets the operation type enumeration value. | |
Private Attributes | |
| std::shared_ptr< DeviceContext > | device_context_ |
| The device context for execution. | |
| OperationType | operation_type_ |
| The operation type identifier. | |
Detailed Description
requires ValidTensorTypes<TInput1, TInput2> && ValidFloatTensorType<TOutput>
class Mila::Dnn::Compute::OperationBase< TDeviceType, TInput1, TInput2, TOutput >
Base class for all compute operations in the Mila neural network framework.
This abstract base class defines the common interface for all operations that can be performed in the neural network computation graph, regardless of the device type (CPU, CUDA, etc). Specific operations inherit from this class and implement their specialized behavior while adhering to a consistent interface.
- Template Parameters
-
TInput1 The data type of the first input tensor elements. Must satisfy ValidTensorType constraint. TInput2 The data type of the second input tensor elements, defaults to TInput1. Must satisfy ValidTensorType constraint. TOutput The data type of the output tensor elements, defaults to TInput1. Must satisfy ValidFloatTensorType constraint. TDeviceType The target device type for the operation, defaults to DeviceType::Cuda.
Constructor & Destructor Documentation
◆ OperationBase()
|
inline |
Constructs an OperationBase object with a specific device context and compute precision.
Initializes the operation with the specified operation type and device context, using the template parameter-specified compute precision.
- Parameters
-
operation_type The type of the operation (from OperationType enum). context The device context to use for this operation. Must not be null. precision_policy The compute precision policy to use for this operation.
- Exceptions
-
std::invalid_argument May throw if context is null (implementation dependent).
◆ ~OperationBase()
|
virtualdefault |
Virtual destructor for the OperationBase class.
Ensures proper cleanup of derived class resources when destroyed through a base class pointer. Default implementation is sufficient for this base class.
Member Function Documentation
◆ getDeviceContext()
|
inline |
Gets the device context associated with this operation.
The device context contains information about the execution environment, including the device, streams, and memory resources. This context is used for all device interactions performed by this operation.
- Returns
- std::shared_ptr<DeviceContext> The device context for this operation.
◆ getDeviceType()
|
inline |
Gets the device type for this operation.
This is a convenience method that retrieves the device type from the associated device context. It delegates to the device context's device to determine the actual hardware target.
- Returns
- DeviceType The type of device (CPU, CUDA, etc.) for this operation.
◆ getName()
|
pure virtual |
Gets the name of the operation.
This pure virtual function must be implemented by derived classes to return a unique identifier string for the specific operation type. The name should be descriptive and consistent across framework components.
- Returns
- std::string A unique name identifying this operation type.
Implemented in Mila::Dnn::Compute::CpuCrossEntropyOp, Mila::Dnn::Compute::CpuEncoderOp, Mila::Dnn::Compute::CpuGeluOp, Mila::Dnn::Compute::CpuLayerNormOp, Mila::Dnn::Compute::CpuLinearOp, Mila::Dnn::Compute::CpuMultiHeadAttentionOp, Mila::Dnn::Compute::CpuResidualOp, Mila::Dnn::Compute::CpuSoftmaxOp, Mila::Dnn::Compute::CudaEncoderOp< TInput, TOutput >, Mila::Dnn::Compute::CudaGeluOp< TDataType >, Mila::Dnn::Compute::CudaLayerNormOp< TInput, TOutput >, Mila::Dnn::Compute::CudaLinearOp< TInput, TOutput >, Mila::Dnn::Compute::CudaMultiHeadAttentionOp< TInput, TOutput >, Mila::Dnn::Compute::CudaResidualOp< TInput, TOutput >, Mila::Dnn::Compute::CudaSoftmaxOp< TInput, TOutput >, Mila::Dnn::Compute::FusedSoftmaxCrossEntropyOp< TPrecision >, and Mila::Dnn::Compute::CudaMatMulBiasGeluOp< TInput, TOutput >.
◆ getOperationType()
|
inline |
Gets the operation type enumeration value.
Returns the operation type that was specified during construction. This identifies the category of neural network operation being performed.
- Returns
- OperationType The enumeration value identifying this operation's category.
Member Data Documentation
◆ device_context_
|
private |
The device context for execution.
◆ operation_type_
|
private |
The operation type identifier.
The documentation for this class was generated from the following file:
- /home/runner/work/Mila/Mila/Mila/Src/Dnn/Compute/Operations/OperationBase.ixx
