TorchDNNV2CUDA Class Reference

A class that implements a configurable deep neural network using PyTorch with CUDA support, inheriting from MLFunction. More...

#include <functions.h>

Inheritance diagram for TorchDNNV2CUDA:

Public Member Functions
	TorchDNNV2CUDA (std::vector< velox::dl::KernelType > kernelTypes, std::vector< float * > weights, std::vector< int > dimensions)
	Constructor that initializes the neural network with kernel types, weights, and dimensions.
void	apply (const SelectivityVector &rows, std::vector< VectorPtr > &args, const TypePtr &type, exec::EvalCtx &context, VectorPtr &output) const override
	Applies the neural network to the input array using CUDA.
float *	getTensor () const override
	Returns the tensor associated with this function.
const std::vector< float * > &	getWeights () const
	Returns the weights of the neural network.
const std::vector< float * > &	getBias () const
	Returns the biases of the neural network.
std::string	getFuncName ()
	Returns the name of the function.
std::vector< velox::dl::KernelType >	getKernelTypes () const
	Returns the kernel types used in the neural network.
CostEstimate	getCost (std::vector< int > inputDims)
	Estimates the computational cost of applying the neural network.
Public Member Functions inherited from MLFunction
virtual	~MLFunction ()=default
	Virtual destructor.
virtual std::vector< int >	getDims ()
	Returns the dimensions of the function.
virtual int	getNumDims ()
	Returns the number of dimensions of the function.

Static Public Member Functions
static std::vector< std::shared_ptr< exec::FunctionSignature > >	signatures ()
	Returns the function signatures supported by this class.
static std::string	getName ()
	Static method to return the name of the function.

Additional Inherited Members
Protected Member Functions inherited from MLFunction
double	getWeightedCost (std::string name, float cost)
	Calculates the weighted cost of the function.
std::vector< double >	getCoefficientVector (std::string name)
	Retrieves the cost coefficients for the function.
Protected Attributes inherited from MLFunction
std::vector< int >	dims
	Dimensions of the function.

Detailed Description

A class that implements a configurable deep neural network using PyTorch with CUDA support, inheriting from MLFunction.

This class provides functionality to apply a neural network with configurable layers (e.g., MatMul, ReLU, Softmax, etc.) on a CUDA device.

Constructor & Destructor Documentation

TorchDNNV2CUDA()

TorchDNNV2CUDA::TorchDNNV2CUDA ( std::vector< velox::dl::KernelType > kernelTypes, std::vector< float * > weights, std::vector< int > dimensions )

inline

Constructor that initializes the neural network with kernel types, weights, and dimensions.

Parameters

kernelTypes	A vector of KernelType specifying the types of layers in the network.
weights	A vector of pointers to weight matrices for each layer.
dimensions	A vector containing the dimensions of the neural network layers.

Member Function Documentation

apply()

void TorchDNNV2CUDA::apply ( const SelectivityVector & rows, std::vector< VectorPtr > & args, const TypePtr & type, exec::EvalCtx & context, VectorPtr & output ) const

inlineoverride

Applies the neural network to the input array using CUDA.

This method processes the input array, applies the neural network on a CUDA device, and stores the result in the output vector.

Parameters

rows	A SelectivityVector specifying the rows to process.
args	A vector of input arguments (e.g., the input array).
type	The type of the output vector.
context	The execution context.
output	The output vector where the result will be stored.

getBias()

const std::vector< float * > & TorchDNNV2CUDA::getBias ( ) const

inline

Returns the biases of the neural network.

Returns

A vector of pointers to bias vectors.

getCost()

CostEstimate TorchDNNV2CUDA::getCost ( std::vector< int > inputDims )

inlinevirtual

Estimates the computational cost of applying the neural network.

Parameters

inputDims

A vector containing the dimensions of the input.

Returns

A CostEstimate object representing the estimated cost.

Reimplemented from MLFunction.

getFuncName()

std::string TorchDNNV2CUDA::getFuncName ( )

inlinevirtual

Returns the name of the function.

Returns

The name of the function as a string.

Reimplemented from MLFunction.

getKernelTypes()

std::vector< velox::dl::KernelType > TorchDNNV2CUDA::getKernelTypes ( ) const

inline

Returns the kernel types used in the neural network.

Returns

A vector of KernelType specifying the types of layers.

getName()

static std::string TorchDNNV2CUDA::getName ( )

inlinestatic

Static method to return the name of the function.

Returns

The name of the function as a string ("complexTorchNN_GPU").

getTensor()

float * TorchDNNV2CUDA::getTensor ( ) const

inlineoverridevirtual

Returns the tensor associated with this function.

Returns

A pointer to an empty float array (no weights for TorchDNNV2CUDA).

Implements MLFunction.

getWeights()

const std::vector< float * > & TorchDNNV2CUDA::getWeights ( ) const

inline

Returns the weights of the neural network.

Returns

A vector of pointers to weight matrices.

signatures()

static std::vector< std::shared_ptr< exec::FunctionSignature > > TorchDNNV2CUDA::signatures ( )

inlinestatic

Returns the function signatures supported by this class.

Returns

A vector of shared pointers to FunctionSignature objects.

---

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

• functions.h