magmadnn::op::Variable< T > Class Template Reference

#include <variable.h>

Inheritance diagram for magmadnn::op::Variable< T >:

Collaboration diagram for magmadnn::op::Variable< T >:

Public Member Functions
	Variable (std::string name, std::vector< unsigned int > shape, tensor_filler_t< T > filler, memory_t mem_type)
	Variable (std::string name, Tensor< T > *val)
std::string	to_string ()
std::string	get_name ()
Public Member Functions inherited from magmadnn::op::Operation< T >
	Operation ()
	Operation (std::vector< Operation< T > *> inputs, bool needs_grad=true)
virtual std::vector< unsigned int >	get_output_shape () const
virtual unsigned int	get_output_shape (unsigned int idx) const
virtual unsigned int	get_output_size () const
virtual memory_t	get_memory_type () const
virtual Tensor< T > *	eval (bool recompute=true)
virtual void	reset ()
virtual Tensor< T > *	grad (Operation< T > *consumer, Operation< T > *var, Tensor< T > *grad, bool recompute=true)
virtual void	add_consumer (Operation< T > *consumer)
virtual std::vector< Operation< T > * >	get_consumers ()
virtual std::vector< Operation< T > * >	get_inputs ()
virtual Tensor< T > *	get_output_tensor ()
virtual Tensor< T > *	get_grad_tensor (Operation< T > *wrt)

Protected Member Functions
Tensor< T > *	_eval (bool recompute=true)
Tensor< T > *	_grad (Operation< T > *consumer, Operation< T > *var, Tensor< T > *grad)

Protected Attributes
std::string	name
Tensor< T > *	val
bool	delete_tensor
Protected Attributes inherited from magmadnn::op::Operation< T >
std::vector< Operation< T > * >	inputs
std::vector< Operation< T > * >	consumers
std::vector< unsigned int >	output_shape
memory_t	mem_type
std::map< uintptr_t, Tensor< T > * >	_grad_cache
std::string	name = "DefaultOpName"
Tensor< T > *	output_tensor
bool	needs_grad
bool	has_been_computed
bool	has_grad_been_computed

Detailed Description

template<typename T>
class magmadnn::op::Variable< T >

Variable Operation. The most basic operation; it simply wraps around a tensor.

Template Parameters

T

Member Function Documentation

_eval()

template<typename T >

Tensor< T > * magmadnn::op::Variable< T >::_eval ( bool  recompute = true )

protectedvirtual

Sets this->output_tensor to the value of this operation

Returns

Tensor<T>* the evaluated tensor

Implements magmadnn::op::Operation< T >.

Here is the caller graph for this function:

_grad()

template<typename T >

Tensor< T > * magmadnn::op::Variable< T >::_grad ( Operation< T > *  consumer, Operation< T > *  var, Tensor< T > *  grad  )

protectedvirtual

Computes the gradient of this operation wrt the output of consumer.

Parameters

consumer
var
grad
recompute

Returns

Tensor<T>*

Implements magmadnn::op::Operation< T >.

Here is the caller graph for this function:

get_name()

template<typename T>

std::string magmadnn::op::Variable< T >::get_name ( )

inlinevirtual

the name of this operation.

Returns

std::string

Reimplemented from magmadnn::op::Operation< T >.

Here is the call graph for this function:

to_string()

template<typename T>

std::string magmadnn::op::Variable< T >::to_string ( )

inlinevirtual

string form of the given operation. Expands on input.

Returns

std::string

Implements magmadnn::op::Operation< T >.

---

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

• include/compute/variable.h
• src/compute/variable.cpp