avir::float8 Class Reference

SIMD packed 8-float type. More...

#include <avir_float8_avx.h>

Public Member Functions
	float8 (const __m256 s)
	float8 (const double s)
	float8 (const float s)
	float8 (const float8 &s)
	float8 (const int s)
float	hadd () const
	Returns horizontal sum of elements.
	operator float () const
float8	operator* (const float8 &s) const
float8 &	operator*= (const float8 &s)
float8	operator+ (const float8 &s) const
float8 &	operator+= (const float8 &s)
float8	operator- (const float8 &s) const
float8 &	operator-= (const float8 &s)
float8	operator/ (const float8 &s) const
float8 &	operator/= (const float8 &s)
float8 &	operator= (const __m256 s)
float8 &	operator= (const float s)
float8 &	operator= (const float8 &s)
void	store (float *const p) const
	Stores this value to the specified memory location.
void	storeu (float *const p) const
	Stores this value to the specified memory location.
void	storeu (float *p, int lim) const
	Stores "lim" lower elements of this value to the specified memory location.

Static Public Member Functions
static void	addu (float *const p, const float8 &v)
	Performs in-place addition of a value located in memory, and the specified value.
static void	addu (float *const p, const float8 &v, const int lim)
	Performs in-place addition of a value located in memory, and the specified value. Limited to the specfied number of elements.
static float8	load (const float *const p)
	Returns float8 value loaded from the specified memory location.
static float8	loadu (const float *const p)
	Returns float8 value loaded from the specified memory location.
static float8	loadu (const float *const p, const int lim)
	Returns float8 value loaded from the specified memory location, with elements beyond "lim" set to 0.

Public Attributes
__m256	value
	Packed value of 8 floats.

Detailed Description

SIMD packed 8-float type.

This class implements a packed 8-float type that can be used to perform parallel computation using SIMD instructions on AVX-enabled processors. This class can be used as the "fptype" argument of the avir::fpclass_def or avir::fpclass_def_dil class.

Member Function Documentation

addu() [1/2]

static void avir::float8::addu ( float *const p, const float8 & v )

static

Performs in-place addition of a value located in memory, and the specified value.

Parameters

p	Pointer to value where addition happens. May be unaligned.
v	Value to add.

addu() [2/2]

static void avir::float8::addu ( float *const p, const float8 & v, const int lim )

static

Performs in-place addition of a value located in memory, and the specified value. Limited to the specfied number of elements.

Parameters

p	Pointer to value where addition happens. May be unaligned.
v	Value to add.
lim	The element number limit, greater than 0.

load()

static float8 avir::float8::load ( const float *const p )

static

Returns float8 value loaded from the specified memory location.

Parameters

p	Pointer to memory from where the value should be loaded, should be 32-byte aligned.

Returns

Loaded value.

loadu() [1/2]

static float8 avir::float8::loadu ( const float *const p )

static

Returns float8 value loaded from the specified memory location.

Parameters

p	Pointer to memory from where the value should be loaded, may have any alignment.

Returns

Loaded value.

loadu() [2/2]

static float8 avir::float8::loadu ( const float *const p, const int lim )

static

Returns float8 value loaded from the specified memory location, with elements beyond "lim" set to 0.

Parameters

p	Pointer to memory from where the value should be loaded, may have any alignment.
lim	The maximum number of elements to load, greater than 0.

Returns

Loaded value.

store()

void avir::float8::store

(

float *const

p

)

const

Stores this value to the specified memory location.

Parameters

[out]

p

Output memory location, should be 32-byte aligned.

storeu() [1/2]

void avir::float8::storeu

(

float *const

p

)

const

Stores this value to the specified memory location.

Parameters

[out]

p

Output memory location, may have any alignment.

storeu() [2/2]

void avir::float8::storeu	(	float *	p,
		int	lim ) const

Stores "lim" lower elements of this value to the specified memory location.

Parameters

[out]	p	Output memory location, may have any alignment.
	lim	The number of lower elements to store, greater than 0.