Description

Operations that are based on floating point arithmetic, but can be applied to arbitrary input and output types.

These functions take an input of arbitrary type, convert it to a floating point type, perform an operation in floating point arithmetic, and convert the result to an output of arbitrary type.

Similar to convert(), if the input and output types are of the same size, both the inputs and outputs consist of one Vec. If the types are of different sizes, the input or output consists of multiple Vec's, such that the number of input elements is equal to the number of output elements.

Typedefs
template<typename Tout , typename Tin >
using	simd::BigEnoughFloat
	Smallest floating point type that is at least as big as the input and output types.

Functions
template<typename Tout , typename Tin , typename Tfloat = BigEnoughFloat<Tout, Tin>, size_t SIMD_WIDTH>
static void	simd::faddmul (const Vec< Tin, SIMD_WIDTH > vecsIn[numInVecs< Tout, Tin >()], dont_deduce< Tfloat > off, dont_deduce< Tfloat > fac, Vec< Tout, SIMD_WIDTH > vecsOut[numOutVecs< Tout, Tin >()])
	Adds a floating point constant to the elements of Vec's, then multiplies with a floating point constant in floating point arithmetic.

template<size_t DIM, size_t NVEC, typename Tout , typename Tin , size_t SIMD_WIDTH>
static void	simd::fdivMsigmoidmul (const Vec< Tin, SIMD_WIDTH > vecsNum[DIM][NVEC], const Vec< Tin, SIMD_WIDTH > vecsDenom[DIM][NVEC], const double w[DIM], const double w0[DIM], double fac, Vec< Tout, SIMD_WIDTH > vecsOut[numOutVecs< Tout, Tin >()])
	Special function used in MinWarping.

template<typename Tout , typename Tin , typename Tfloat = BigEnoughFloat<Tout, Tin>, size_t SIMD_WIDTH>
static void	simd::fdivmul (const Vec< Tin, SIMD_WIDTH > vecsNum[numInVecs< Tout, Tin >()], const Vec< Tin, SIMD_WIDTH > vecsDenom[numInVecs< Tout, Tin >()], dont_deduce< Tfloat > fac, Vec< Tout, SIMD_WIDTH > vecsOut[numOutVecs< Tout, Tin >()])
	Divides Vec's element-wise, then multiplies with a constant factor in floating point arithmetic.

template<typename Tout , typename Tin , typename Tfloat = BigEnoughFloat<Tout, Tin>, size_t SIMD_WIDTH>
static void	simd::fmul (const Vec< Tin, SIMD_WIDTH > vecsIn[numInVecs< Tout, Tin >()], dont_deduce< Tfloat > fac, Vec< Tout, SIMD_WIDTH > vecsOut[numOutVecs< Tout, Tin >()])
	Multiplies Vec's element-wise with a floating point constant in floating point arithmetic.

template<typename Tout , typename Tin , typename Tfloat = BigEnoughFloat<Tout, Tin>, size_t SIMD_WIDTH>
static void	simd::fmuladd (const Vec< Tin, SIMD_WIDTH > vecsIn[numInVecs< Tout, Tin >()], dont_deduce< Tfloat > fac, dont_deduce< Tfloat > off, Vec< Tout, SIMD_WIDTH > vecsOut[numOutVecs< Tout, Tin >()])
	Multiplies the elements of Vec's with a floating point constant, then adds a floating point constant in floating point arithmetic.

template<typename Tout , typename Tin , typename Tfloat = BigEnoughFloat<Tout, Tin>, size_t SIMD_WIDTH>
static void	simd::fwaddmul (const Vec< Tin, SIMD_WIDTH > vecsIn1[numInVecs< Tout, Tin >()], const Vec< Tin, SIMD_WIDTH > vecsIn2[numInVecs< Tout, Tin >()], dont_deduce< Tfloat > w, dont_deduce< Tfloat > fac, Vec< Tout, SIMD_WIDTH > vecsOut[numOutVecs< Tout, Tin >()])
	Linearly interpolates Vec's element-wise with a constant weight and then scales by a constant factor in floating point arithmetic.

Typedef Documentation

◆ BigEnoughFloat

template<typename Tout , typename Tin >

using simd::BigEnoughFloat

Smallest floating point type that is at least as big as the input and output types.

Template Parameters

Tout	output type
Tin	input type

Function Documentation

◆ faddmul()

template<typename Tout , typename Tin , typename Tfloat = BigEnoughFloat<Tout, Tin>, size_t SIMD_WIDTH>

static void simd::faddmul	(	const Vec< Tin, SIMD_WIDTH >	vecsIn[numInVecs< Tout, Tin >()],
		dont_deduce< Tfloat >	off,
		dont_deduce< Tfloat >	fac,
		Vec< Tout, SIMD_WIDTH >	vecsOut[numOutVecs< Tout, Tin >()] )

inlinestatic

Adds a floating point constant to the elements of Vec's, then multiplies with a floating point constant in floating point arithmetic.

Template Parameters

Tfloat the floating point type to perform the calculation in. Defaults to BigEnoughFloat<Tout, Tin>

Parameters

[in]	vecsIn	input Vec's
	off	float constant to add
	fac	float constant to multiply with
[out]	vecsOut	output Vec's

◆ fdivMsigmoidmul()

template<size_t DIM, size_t NVEC, typename Tout , typename Tin , size_t SIMD_WIDTH>

static void simd::fdivMsigmoidmul	(	const Vec< Tin, SIMD_WIDTH >	vecsNum[DIM][NVEC],
		const Vec< Tin, SIMD_WIDTH >	vecsDenom[DIM][NVEC],
		const double	w[DIM],
		const double	w0[DIM],
		double	fac,
		Vec< Tout, SIMD_WIDTH >	vecsOut[numOutVecs< Tout, Tin >()] )

inlinestatic

Special function used in MinWarping.

Parameters

[in]	vecsNum	Numerator vectors.
[in]	vecsDenom	Denominator vectors.
[in]	w	Weights.
[in]	w0	Weights.
	fac	Factor.
[out]	vecsOut	Output vectors.

◆ fdivmul()

template<typename Tout , typename Tin , typename Tfloat = BigEnoughFloat<Tout, Tin>, size_t SIMD_WIDTH>

static void simd::fdivmul	(	const Vec< Tin, SIMD_WIDTH >	vecsNum[numInVecs< Tout, Tin >()],
		const Vec< Tin, SIMD_WIDTH >	vecsDenom[numInVecs< Tout, Tin >()],
		dont_deduce< Tfloat >	fac,
		Vec< Tout, SIMD_WIDTH >	vecsOut[numOutVecs< Tout, Tin >()] )

inlinestatic

Divides Vec's element-wise, then multiplies with a constant factor in floating point arithmetic.

Template Parameters

Tfloat the floating point type to perform the calculation in. Defaults to BigEnoughFloat<Tout, Tin>

Parameters

[in]	vecsNum	numerator Vec's
[in]	vecsDenom	denominator Vec's
	fac	factor to multiply with
[out]	vecsOut	output Vec's

See also: numInVecs(), numOutVecs()

◆ fmul()

template<typename Tout , typename Tin , typename Tfloat = BigEnoughFloat<Tout, Tin>, size_t SIMD_WIDTH>

static void simd::fmul	(	const Vec< Tin, SIMD_WIDTH >	vecsIn[numInVecs< Tout, Tin >()],
		dont_deduce< Tfloat >	fac,
		Vec< Tout, SIMD_WIDTH >	vecsOut[numOutVecs< Tout, Tin >()] )

inlinestatic

Multiplies Vec's element-wise with a floating point constant in floating point arithmetic.

Template Parameters

Tfloat the floating point type to perform the calculation in. Defaults to BigEnoughFloat<Tout, Tin>

Parameters

[in]	vecsIn	input Vec's
	fac	factor
[out]	vecsOut	output Vec's

◆ fmuladd()

template<typename Tout , typename Tin , typename Tfloat = BigEnoughFloat<Tout, Tin>, size_t SIMD_WIDTH>

static void simd::fmuladd	(	const Vec< Tin, SIMD_WIDTH >	vecsIn[numInVecs< Tout, Tin >()],
		dont_deduce< Tfloat >	fac,
		dont_deduce< Tfloat >	off,
		Vec< Tout, SIMD_WIDTH >	vecsOut[numOutVecs< Tout, Tin >()] )

inlinestatic

Multiplies the elements of Vec's with a floating point constant, then adds a floating point constant in floating point arithmetic.

Template Parameters

Tfloat the floating point type to perform the calculation in. Defaults to BigEnoughFloat<Tout, Tin>

Parameters

[in]	vecsIn	input Vec's
	fac	float constant to multiply with
	off	float constant to add
[out]	vecsOut	output Vec's

◆ fwaddmul()

template<typename Tout , typename Tin , typename Tfloat = BigEnoughFloat<Tout, Tin>, size_t SIMD_WIDTH>

static void simd::fwaddmul	(	const Vec< Tin, SIMD_WIDTH >	vecsIn1[numInVecs< Tout, Tin >()],
		const Vec< Tin, SIMD_WIDTH >	vecsIn2[numInVecs< Tout, Tin >()],
		dont_deduce< Tfloat >	w,
		dont_deduce< Tfloat >	fac,
		Vec< Tout, SIMD_WIDTH >	vecsOut[numOutVecs< Tout, Tin >()] )

inlinestatic

Linearly interpolates Vec's element-wise with a constant weight and then scales by a constant factor in floating point arithmetic.

The result is calculated as: out = fac * (w * v1 + (1 - w) * v2) (implemented as out = fac * (v2 + w * (v1 - v2))).

Template Parameters

Tfloat the floating point type to perform the calculation in. Defaults to BigEnoughFloat<Tout, Tin>

Parameters

[in]	vecsIn1	first input Vec's
[in]	vecsIn2	second input Vec's
	w	interpolation weight
	fac	scaling factor
[out]	vecsOut	output Vec's

Description

Typedefs

Functions

Typedef Documentation

◆ BigEnoughFloat

Function Documentation

◆ faddmul()

◆ fdivMsigmoidmul()

◆ fdivmul()

◆ fmul()

◆ fmuladd()

◆ fwaddmul()