IntelĀ® Machine Learning Scaling Library  2018
A library providing an efficient implementation of communication patterns used in deep learning.
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros
Public Member Functions | List of all members
MLSL::OperationRegInfo Class Reference

A class to hold Operation registration information. More...

#include <mlsl.hpp>

Public Member Functions

void SetName (const char *name)
 
size_t AddInput (size_t featureMapCount, size_t featureMapSize, DataType dataType)
 
size_t AddOutput (size_t featureMapCount, size_t featureMapSize, DataType dataType)
 
size_t AddParameterSet (size_t kernelCount, size_t kernelSize, DataType dataType, bool distributedUpdate=false, CompressionType compressType=CT_NONE)
 
void Validate (Distribution *dist=NULL)
 

Detailed Description

A class to hold Operation registration information.

Holds the information about learnable parameters and activation shapes and is used for creation of an Operation object. All the input/output activation shapes and parameter shapes (if any) should be added before calling Session::AddOperation()

Member Function Documentation

size_t MLSL::OperationRegInfo::AddInput ( size_t  featureMapCount,
size_t  featureMapSize,
DataType  dataType 
)

Adds an input activation shape to the operation.

Parameters
featureMapCountthe number of feature maps
featureMapSizethe size of feature maps in MLSL::DataType elements
dataTypethe data type of feature map elements
Returns
The index of the input activation.
size_t MLSL::OperationRegInfo::AddOutput ( size_t  featureMapCount,
size_t  featureMapSize,
DataType  dataType 
)

Adds an output activation shape to the operation.

Parameters
featureMapCountthe number of feature maps
featureMapSizethe size of feature maps in MLSL::DataType elements
dataTypethe data type of feature map's elements
Returns
The index of the output activation.
size_t MLSL::OperationRegInfo::AddParameterSet ( size_t  kernelCount,
size_t  kernelSize,
DataType  dataType,
bool  distributedUpdate = false,
CompressionType  compressType = CT_NONE 
)

Adds a parameter set shape to the operation.

Parameters
kernelCountthe number of kernels
kernelSizethe size of a single kernel in MLSL::DataType elements
dataTypethe data type of kernel's elements
distributedUpdatetells whether to use the distributed update of parameters (ReduceScatter() + AllGather() instead of AllReduce())
compressTypethe type of compression (Environment::SetQuantizationParams() must be called before AddParameterSet() with quantization enabled)
Returns
The index of the parameter set.
void MLSL::OperationRegInfo::SetName ( const char *  name)

Sets the operation name (for debugging purposes)

Parameters
namethe operation name
void MLSL::OperationRegInfo::Validate ( Distribution dist = NULL)

Validates the activations/parameter sets information.

Parameters
distthe distribution (currently not used)

The documentation for this class was generated from the following file: