A Tensor that's backed by a texture object. More...

#include <TexTensor.hpp>

Inheritance diagram for sp::TexTensor< T, R, Options >:

Public Types
using	Base = MemTensorLike< TexTensor< T, R, Options >, T, R, Options >

using	PtrType = typename Base::PtrType

using	ValueType = sp::remove_cvref_t< sp::remove_addrspace_t< T > >

using	Type = T

using	RealType = sp::real_type_t< ValueType >

using	Opts = Options

template<typename NewT >
using	InValueType = TexTensor< NewT, R, Options >

Public Types inherited from sp::MemTensorLike< TexTensor< T, R, sp::DefaultTexTensorOptions >, T, R, sp::DefaultTexTensorOptions >
using	ThisType = MemTensorLike< TexTensor< T, R, sp::DefaultTexTensorOptions >, T, TensorRank, sp::DefaultTexTensorOptions >

using	ValueType = sp::remove_cvref_t< sp::remove_addrspace_t< T > >

using	Type = T

using	Opts = sp::DefaultTexTensorOptions

using	Base = TensorLike< TexTensor< T, R, sp::DefaultTexTensorOptions >, TensorRank >

using	InValueType = MemTensorLike< TexTensor< T, R, sp::DefaultTexTensorOptions >, NewT, TensorRank, sp::DefaultTexTensorOptions >

using	PtrType = T *
	The type of the data pointer. More...

Public Member Functions
__host__	TexTensor (PtrType data, const Vec< int, Rank > &sizes, const Vec< int, Rank > &strides, sp::Vec< cudaTextureAddressMode, Rank > addrMode, ValueType borderColour)

__host__	TexTensor (PtrType data, const Vec< int, Rank > &sizes, const Vec< int, Rank > &strides, cudaTextureAddressMode addrMode, ValueType borderColour=ValueType{0})

__host__	TexTensor (PtrType data, const Vec< int, Rank > &sizes, const Vec< int, Rank > &strides, sp::Vec< cudaTextureAddressMode, Rank > addrMode)

template<int L, int DimRank>
Vec< ValueType, L >	handleReadTexel (Vec< ValueType, MaxVectorSize > texel, const Vec< int, DimRank > &pos) const

template<int DimRank>
bool	boundsCheckImpl (const Vec< int, DimRank > &) const

template<int L, CacheMode Mode, int DimRank>
Vec< ValueType, L >	vectorOffsetReadImpl (const Vec< int, DimRank > &pos, const Vec< int, DimRank > &offset) const

template<int L, CacheMode Mode, int DimRank>
Vec< ValueType, L >	vectorReadImpl (const Vec< int, DimRank > &pos) const

Public Member Functions inherited from sp::MemTensorLike< TexTensor< T, R, sp::DefaultTexTensorOptions >, T, R, sp::DefaultTexTensorOptions >
auto	getSizes () const

void	setSizes (sp::Vec< int, Rank > sz)

auto &	getSize (int n)

const auto &	getSize (int n) const

auto	getStrides () const

void	setStrides (sp::Vec< int, getNumStrides()> sz)

auto &	getStride (int n)

const auto &	getStride (int n) const

auto &	getTotalSize ()

const auto &	getTotalSize () const

auto &	sizeQuantisation ()

const auto &	sizeQuantisation () const

void	setDefaultSizeQuantisation ()

sp::Vec< int, Rank >	getDimsWithPadding ()

void	validateFields ()

void	computeStrides ()
	Compute strides in the default way. More...

int	totalSizeImpl () const
	Get the total memory occupied by the `Tensor`, in terms of elements. More...

sp::Vec< int, Rank >	stridesImpl () const
	Default implementation of the stride-getter. More...

sp::Vec< int, Rank >	strides () const

int	stride (int i) const

	MemTensorLike (PtrType ptr, const Vec< int, Rank > &sizes)
	Create a `Tensor`. More...

	MemTensorLike (PtrType ptr, const Vec< int, Rank > &sizes, const Vec< int, Rank > &strides)
	Create a `Tensor` with specified strides. More...

int	computeOffsetImpl (const Vec< int, Rank > &pos) const
	Calculate the offset from the start of the buffer that a particular element can be found at. More...

int	computeOffset (const Vec< int, Rank > &pos) const

const PtrType	ptr (const Vec< int, Rank > &pos) const
	Get a pointer to a particular element. More...

PtrType	ptr (const Vec< int, Rank > &pos)

const PtrType	ptr () const
	Get a pointer to the start of the underlying buffer. More...

PtrType	ptr ()

auto	begin () const

auto	end () const

auto	getViewImpl (const sp::Vec< int, Rank > &start, const sp::Vec< int, Rank > &size)
	Create a view into this Tensor using identical striding. More...

auto &	operator[] (int idx)
	Handy 1D subscript operator. More...

const auto &	operator[] (int idx) const

void	setSizeQuantisation (int x)
	Set the size quantisation for the bounds checks. More...

Public Member Functions inherited from sp::TensorLike< Subclass, TensorRank >
auto	dims () const

bool	boundsCheck (const Vec< int, Rank > &pos) const

template<int L, CacheMode Mode = CacheMode::DEFAULT>
auto	vectorRead (const Vec< int, Rank > &pos) const

template<int L, CacheMode Mode = CacheMode::DEFAULT>
auto	vectorOffsetRead (const Vec< int, Rank > &base, const Vec< int, Rank > &offset) const

template<int L, CacheMode Mode = CacheMode::DEFAULT>
auto	maskedVectorRead (const Vec< int, Rank > &pos) const

template<int L, CacheMode Mode = CacheMode::DEFAULT, typename T >
void	vectorWrite (const Vec< int, Rank > &pos, const Vec< T, L > &values)

template<int L, CacheMode Mode = CacheMode::DEFAULT, typename T >
void	maskedVectorWrite (const Vec< int, Rank > &pos, const Vec< T, L > &values)

int	sizeQuantisation () const

int	totalSize () const
	Get the total memory occupied by the `Tensor`, in terms of elements. More...

auto	getView (const sp::Vec< int, Rank > &start, const sp::Vec< int, Rank > &size)
	Get an object that represents (and aliases) a portion of this object. More...

int	dim (int d) const
	Behaviour common to all TensorLikes ///. More...

template<CacheMode Mode = CacheMode::DEFAULT>
auto	read (const Vec< int, Rank > &pos) const
	Get the element at a given position. More...

template<CacheMode Mode = CacheMode::DEFAULT, typename T >
void	write (const Vec< int, Rank > &pos, const T &value)
	Set a single element. More...

template<int Dummy = 0>
int	size () const

template<int L>
void	boundsCheckAccess (Vec< int, Rank > pos) const
	Bounds-check an L-element vector read at `pos`. More...

Static Public Attributes
constexpr static int	Rank = R

constexpr static int	MaxVectorSize = sp::getMaxGMemVectorSize<ValueType>()

constexpr static int	ElementsPerTexel = 16 / sizeof(ValueType)

constexpr static int	ChannelsPerElement = 4 / ElementsPerTexel

Static Public Attributes inherited from sp::MemTensorLike< TexTensor< T, R, sp::DefaultTexTensorOptions >, T, R, sp::DefaultTexTensorOptions >
static constexpr int	Rank

static constexpr bool	AllowVectorMemoryOps
	An overrideable flag to indicate if a concrete `MemTensorLike` subclass supports vector memory operations. More...

static constexpr int	NumStrides
	The number of strides to explicitly store. More...

static constexpr int	ImplicitStrides
	The number of strides that are not stored. More...

constexpr static bool	HasTotalSize

Static Public Attributes inherited from sp::TensorLike< Subclass, TensorRank >
constexpr static int	Rank = TensorRank

Additional Inherited Members
Static Public Member Functions inherited from sp::MemTensorLike< TexTensor< T, R, sp::DefaultTexTensorOptions >, T, R, sp::DefaultTexTensorOptions >
static constexpr int	calculateNumStrides ()
	Compute the default value for `NumStrides`. More...

Public Attributes inherited from sp::MemTensorLike< TexTensor< T, R, sp::DefaultTexTensorOptions >, T, R, sp::DefaultTexTensorOptions >
friend	Base

PtrType	dataPtr

sp::Vec< int, Rank+getNumStrides()+HasTotalSize+AssertionsEnabled >	variousIntegers

Protected Member Functions inherited from sp::MemTensorLike< TexTensor< T, R, sp::DefaultTexTensorOptions >, T, R, sp::DefaultTexTensorOptions >
sp::Vec< int, Rank >	dimsImpl () const

int	sizeQuantisationImpl () const

Vec< ValueType, L >	vectorReadImpl (const Vec< int, Rank > &pos) const
	This will yield a vector load instruction if the alignment permits. More...

void	vectorWriteImpl (const Vec< int, Rank > &pos, const Vec< ValueType, L > &values)

Protected Member Functions inherited from sp::TensorLike< Subclass, TensorRank >
template<typename Dummy = void*>
auto	dimsImpl () const
	Get an `sp::Vec<int, X>` representing the dimensions of the object. More...

bool	boundsCheckImpl (const sp::Vec< int, Rank > &) const
	Return true iff the given coordinates are inside the object (Default implementation below) More...

int	sizeQuantisationImpl () const
	The last dimension is rounded up to the next multiple of this value for the purposes of bounds checks. More...

template<int L, CacheMode Mode = CacheMode::DEFAULT>
auto	vectorReadImpl (const Vec< int, Rank > &) const
	Read L elements, adjacent in the last dimension, starting at the given position. More...

template<int L, CacheMode Mode = CacheMode::DEFAULT>
auto	vectorOffsetReadImpl (const Vec< int, Rank > &base, const Vec< int, Rank > &offset) const
	Read L elements, adjacent in the last dimension, from position `base + offset`. More...

template<int L, CacheMode Mode = CacheMode::DEFAULT>
auto	maskedVectorReadImpl (const Vec< int, Rank > &pos) const
	Do a vectorRead that copes with the possibility of part of the vector being out of bounds. More...

template<int L, CacheMode Mode = CacheMode::DEFAULT, typename T >
void	maskedVectorWriteImpl (const Vec< int, Rank > &pos, const Vec< T, L > &values)
	Do a vectorWrite that copes with the possibility of part of the vector being out of bounds. More...

template<int L, CacheMode Mode, typename T >
void	vectorWriteImpl (const Vec< int, Rank > &pos, const Vec< T, L > &values)
	Write L elements, adjacent in the last dimension, starting at the given position. More...

auto	getViewImpl (const sp::Vec< int, Rank > &start, const sp::Vec< int, Rank > &size)
	The default implementation just makes a `TensorView`. Not hugely fast, but always works. More...

Detailed Description

template<typename T, int R, typename Options = sp::DefaultTexTensorOptions>
class sp::TexTensor< T, R, Options >

A Tensor that's backed by a texture object.

Usually useful to combine with TexTensorSpecialiser, which will swap input tensors for TexTensors when the (many) necessary constraints are met.

TexTensors are writeable, but subject to the documented limitations of writes to texture-backed buffers:

The texture and surface memory is cached (see Device Memory Accesses) and within the same kernel call, the cache is not kept coherent with respect to global memory writes and surface memory writes, so any texture fetch or surface read to an address that has been written to via a global write or a surface write in the same kernel call returns undefined data. In other words, a thread can safely read some texture or surface memory location only if this memory location has been updated by a previous kernel call or memory copy, but not if it has been previously updated by the same thread or another thread from the same kernel call.

So: writing is allowed, but it is up to the user to never read an address that has previously been written while in the same kernel. There is nothing to catch this error condition: you'll just get undefined behaviour.

Public Types

Public Member Functions

Static Public Attributes

Additional Inherited Members

Detailed Description