freetensor::Array Class Reference

#include <array.h>

Public Member Functions
	~Array ()
	Array (Array &&)
	Array (const Array &)=delete
Array &	operator= (Array &&)
Array &	operator= (const Array &)=delete
size_t	size () const
size_t	nElem () const
const std::vector< size_t > &	shape () const
DataType	dtype () const
bool	dontDropBorrow () const
void	setDontDropBorrow (bool flag)
bool	moved () const
void	setMoved (bool flag)
void *	rawSharedTo (const Ref< Device > &device)
void *	rawMovedTo (const Ref< Device > &device)
void *	rawInitTo (const Ref< Device > &device)
void *	rawTemporarilyCopiedTo (const Ref< Device > &device)
void	makePrivateCopy ()

Static Public Member Functions
static Array	moveFromRaw (void *ptr, const std::vector< size_t > &shape, DataType dtype, const Ref< Device > &device)
static Array	borrowFromRaw (void *ptr, const std::vector< size_t > &shape, DataType dtype, const Ref< Device > &device, bool dontDropBorrow, bool moved)

Detailed Description

Data stored on a Device or shared by multiple Devices

An Array manages one or more ArrayCopys, locating on different devices. An ArrayCopy can be a memory buffer managed by itself, or can be borrowed from a user object.

When an Array is required for read-only, an ArrayCopy will be copied to a specific device, if there isn't one on the device.

When an Array is requried for write-only, an ArrayCopy will be allocated without initialization on a specific device, if there isn't one, and copies on other devices will be dropped. Writing to a lending user object at a different device is not supported.

When an Array is requried for read-write, an ArrayCopy will be copied to a specific device, if there isn't one, and copies on other devices will be dropped. Writing to a lending user object at a different device is not supported.

Constructor & Destructor Documentation

~Array()

freetensor::Array::~Array

(

)

Array() [1/2]

freetensor::Array::Array

(

Array &&

other

)

Array() [2/2]

freetensor::Array::Array ( const Array &  )

delete

Member Function Documentation

borrowFromRaw()

Array freetensor::Array::borrowFromRaw ( void *  ptr, const std::vector< size_t > &  shape, DataType  dtype, const Ref< Device > &  device, bool  dontDropBorrow, bool  moved  )

static

Borrow from raw pointer.

Please make sure the owner keeps alive. Use keep_alive when exposing with PyBind11

dontDropBorrow()

bool freetensor::Array::dontDropBorrow ( ) const

inline

dtype()

DataType freetensor::Array::dtype ( ) const

inline

makePrivateCopy()

void freetensor::Array::makePrivateCopy

(

)

Somethings we can't keep track of user objects, so we need to ensure we don't borrow from user data

moved()

bool freetensor::Array::moved ( ) const

inline

moveFromRaw()

Array freetensor::Array::moveFromRaw ( void *  ptr, const std::vector< size_t > &  shape, DataType  dtype, const Ref< Device > &  device  )

static

Move from raw pointer. Use with cautious

nElem()

size_t freetensor::Array::nElem ( ) const

inline

operator=() [1/2]

Array & freetensor::Array::operator=

(

Array &&

other

)

operator=() [2/2]

Array & freetensor::Array::operator= ( const Array &  )

delete

rawInitTo()

void * freetensor::Array::rawInitTo

(

const Ref< Device > &

device

)

rawMovedTo()

void * freetensor::Array::rawMovedTo

(

const Ref< Device > &

device

)

rawSharedTo()

void * freetensor::Array::rawSharedTo

(

const Ref< Device > &

device

)

rawTemporarilyCopiedTo()

void * freetensor::Array::rawTemporarilyCopiedTo

(

const Ref< Device > &

device

)

setDontDropBorrow()

void freetensor::Array::setDontDropBorrow ( bool  flag )

inline

setMoved()

void freetensor::Array::setMoved ( bool  flag )

inline

shape()

const std::vector< size_t > & freetensor::Array::shape ( ) const

inline

size()

size_t freetensor::Array::size ( ) const

inline

---

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

• include/driver/array.h
• src/driver/array.cc