Description

Functions for storing Vec's to memory.

See also: Masked Memory

Functions
template<typename T , size_t SIMD_WIDTH>
static void	simd::store (T *const p, const Vec< T, SIMD_WIDTH > &a)
	Stores a Vec to aligned memory.

template<typename T , size_t SIMD_WIDTH>
static void	simd::store (T *const p, const Vec< T, SIMD_WIDTH > &outVec, size_t numOutVecs)
	Stores a single Vec multiple times to aligned memory.

template<typename T , size_t SIMD_WIDTH>
static void	simd::store (T *const p, const Vec< T, SIMD_WIDTH > outVecs[], size_t numOutVecs)
	Stores multiple successive Vec's to aligned memory.

template<typename T , size_t SIMD_WIDTH>
static void	simd::storeu (T *const p, const Vec< T, SIMD_WIDTH > &a)
	Stores a Vec to unaligned memory.

template<typename T , size_t SIMD_WIDTH>
static void	simd::storeu (T *const p, const Vec< T, SIMD_WIDTH > &outVec, size_t numOutVecs)
	Stores a single Vec multiple times to unaligned memory.

template<typename T , size_t SIMD_WIDTH>
static void	simd::storeu (T *const p, const Vec< T, SIMD_WIDTH > outVecs[], size_t numOutVecs)
	Stores multiple successive Vec's to unaligned memory.

template<typename T , size_t SIMD_WIDTH>
static void	simd::stream_store (T *const p, const Vec< T, SIMD_WIDTH > &a)
	Stores a Vec to aligned memory using a non-temporal memory hint.

Function Documentation

◆ store() [1/3]

template<typename T , size_t SIMD_WIDTH>

static void simd::store	(	T *const	p,
		const Vec< T, SIMD_WIDTH > &	a )

inlinestatic

Stores a Vec to aligned memory.

The memory location must be aligned to the SIMD_WIDTH.

Parameters

[out]	p	pointer to the aligned memory location to store to
	a	Vec to store

◆ store() [2/3]

template<typename T , size_t SIMD_WIDTH>

static void simd::store	(	T *const	p,
		const Vec< T, SIMD_WIDTH > &	outVec,
		size_t	numOutVecs )

inlinestatic

Stores a single Vec multiple times to aligned memory.

The memory location must be aligned to the SIMD_WIDTH.

Parameters

[out]	p	pointer to the aligned memory location to store to
[in]	outVec	Vec to store
	numOutVecs	number of times to store the Vec

◆ store() [3/3]

template<typename T , size_t SIMD_WIDTH>

static void simd::store	(	T *const	p,
		const Vec< T, SIMD_WIDTH >	outVecs[],
		size_t	numOutVecs )

inlinestatic

Stores multiple successive Vec's to aligned memory.

The memory location must be aligned to the SIMD_WIDTH.

Parameters

[out]	p	pointer to the aligned memory location to store to
[in]	outVecs	array of Vec's to store
	numOutVecs	number of Vec's to store

◆ storeu() [1/3]

template<typename T , size_t SIMD_WIDTH>

static void simd::storeu	(	T *const	p,
		const Vec< T, SIMD_WIDTH > &	a )

inlinestatic

Stores a Vec to unaligned memory.

In contrast to store(), the memory location does not need to be aligned to any boundary.

Parameters

[out]	p	pointer to the memory location to store to
	a	Vec to store

◆ storeu() [2/3]

template<typename T , size_t SIMD_WIDTH>

static void simd::storeu	(	T *const	p,
		const Vec< T, SIMD_WIDTH > &	outVec,
		size_t	numOutVecs )

inlinestatic

Stores a single Vec multiple times to unaligned memory.

In contrast to store(T *const p, const Vec<T, SIMD_WIDTH> &outVec, size_t numOutVecs), the memory location does not need to be aligned to any boundary.

Parameters

[out]	p	pointer to the memory location to store to
[in]	outVec	Vec to store
	numOutVecs	number of times to store the Vec

◆ storeu() [3/3]

template<typename T , size_t SIMD_WIDTH>

static void simd::storeu	(	T *const	p,
		const Vec< T, SIMD_WIDTH >	outVecs[],
		size_t	numOutVecs )

inlinestatic

Stores multiple successive Vec's to unaligned memory.

In contrast to store(T *const, const Vec<T, SIMD_WIDTH>[], size_t), the memory location does not need to be aligned to any boundary.

Parameters

[out]	p	pointer to the memory location to store to
[in]	outVecs	array of Vec's to store
	numOutVecs	number of Vec's to store

◆ stream_store()

template<typename T , size_t SIMD_WIDTH>

static void simd::stream_store	(	T *const	p,
		const Vec< T, SIMD_WIDTH > &	a )

inlinestatic

Stores a Vec to aligned memory using a non-temporal memory hint.

This function uses the _mm*_stream_* intrinsics on Intel and regular store intrinsics on NEON. A call to sfence() may be required in order for other threads/processors to see the stored values. This function may improve performance on some architectures compared to store().

The memory location must be aligned to the SIMD_WIDTH.

Parameters

[out]	p	pointer to the aligned memory location to store to
	a	Vec to store