Mila::Dnn::TensorPtr< T, IsHostAccessible > Class Template Reference

Base tensor pointer class that wraps a raw pointer with memory-type safety. More...

Public Member Functions
	TensorPtr (T *ptr) noexcept
	Constructs a new TensorPtr.
T *	get () const noexcept
	Gets the raw pointer (explicit conversion)
	operator T* () const
	Explicit conversion to raw pointer for CUDA kernels.
T &	operator* () const
	Host-only: Dereference operator.
TensorPtr	operator+ (size_t offset) const
	Host-only: Pointer addition.
T &	operator[] (size_t index) const
	Host-only: Subscript operator.

Private Attributes
T *	ptr_

Detailed Description

template<typename T, bool IsHostAccessible>
class Mila::Dnn::TensorPtr< T, IsHostAccessible >

Base tensor pointer class that wraps a raw pointer with memory-type safety.

Template Parameters

TPrecision	Element type
IsHostAccessible	Whether the pointer points to host-accessible memory

Constructor & Destructor Documentation

TensorPtr()

template<typename T , bool IsHostAccessible>

Mila::Dnn::TensorPtr< T, IsHostAccessible >::TensorPtr ( T *  ptr )

inlineexplicitnoexcept

Constructs a new TensorPtr.

Parameters

ptr	Raw pointer to wrap

Member Function Documentation

get()

template<typename T , bool IsHostAccessible>

T * Mila::Dnn::TensorPtr< T, IsHostAccessible >::get ( ) const

inlinenoexcept

Gets the raw pointer (explicit conversion)

Returns

TPrecision* Raw pointer

Here is the caller graph for this function:

operator T*()

template<typename T , bool IsHostAccessible>

Mila::Dnn::TensorPtr< T, IsHostAccessible >::operator T* ( ) const

inline

Explicit conversion to raw pointer for CUDA kernels.

Returns

Raw pointer for use in CUDA kernels

operator*()

template<typename T , bool IsHostAccessible>

T & Mila::Dnn::TensorPtr< T, IsHostAccessible >::operator* ( ) const

inline

Host-only: Dereference operator.

Returns

Reference to the element

Exceptions

std::runtime_error

if memory is not host-accessible

operator+()

template<typename T , bool IsHostAccessible>

TensorPtr Mila::Dnn::TensorPtr< T, IsHostAccessible >::operator+ ( size_t  offset ) const

inline

Host-only: Pointer addition.

Parameters

offset

Offset to add

Returns

TensorPtr at the offset position

Exceptions

std::runtime_error

if memory is not host-accessible

operator[]()

template<typename T , bool IsHostAccessible>

T & Mila::Dnn::TensorPtr< T, IsHostAccessible >::operator[] ( size_t  index ) const

inline

Host-only: Subscript operator.

Parameters

index

Element index

Returns

Reference to the element

Exceptions

std::runtime_error

if memory is not host-accessible

Member Data Documentation

ptr_

template<typename T , bool IsHostAccessible>

T* Mila::Dnn::TensorPtr< T, IsHostAccessible >::ptr_

private

---

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

• /home/runner/work/Mila/Mila/Mila/Src/Dnn/Tensor/TensorPtr.ixx