Parallel Algorithms

Functions
template<class T >
T	tinker::reduceSum (const T *gpu_a, size_t nelem, int queue)
	Sum over all of the elements of an 1D array. More...
template<class HT , size_t HN, class DPTR >
void	tinker::reduceSum2 (HT(&h_ans)[HN], DPTR v, size_t nelem, int queue)
	Sum over all of the elements of a 2D array. More...
template<class T >
void	tinker::reduceSumOnDevice (T *dp_ans, const T *a, size_t nelem, int queue)
	Sum over all of the elements of an 1D array. This routine will save the result on the device memory in an asynchronous/non-blocking manner. A valid device pointer is required for the result on the device. More...
template<class HT , size_t HN, class DPTR >
void	tinker::reduceSum2OnDevice (HT(&dref)[HN], DPTR v, size_t nelem, int queue)
	Sum over all of the elements of a 2D array. This routine will save the result on the device memory in an asynchronous/non-blocking manner. A valid device array is required for the result on device. More...
template<class T >
T	tinker::dotProd (const T *a, const T *b, size_t nelem, int queue)
	Dot product of two linear arrays. More...
template<class T >
void	tinker::dotProd (T *ans, const T *a, const T *b, size_t nelem, int queue)
	Dot product of two linear arrays. More...
template<class T >
void	tinker::scaleArray (T *dst, T scal, size_t nelem, int queue)
	Multiply all of the elements in an 1D array by a scalar. More...

Detailed Description

Function Documentation

dotProd() [1/2]

template<class T >

T tinker::dotProd	(	const T *	a,
		const T *	b,
		size_t	nelem,
		int	queue
	)

Dot product of two linear arrays.

\[ DotProduct = \sum_i^n a_i \cdot b_i \]

Returns

The dot product to the host thread.

dotProd() [2/2]

template<class T >

void tinker::dotProd	(	T *	ans,
		const T *	a,
		const T *	b,
		size_t	nelem,
		int	queue
	)

Dot product of two linear arrays.

reduceSum()

template<class T >

T tinker::reduceSum	(	const T *	gpu_a,
		size_t	nelem,
		int	queue
	)

Sum over all of the elements of an 1D array.

\[ Sum = \sum_i^n a_i \]

Returns

The sum.

reduceSum2()

template<class HT , size_t HN, class DPTR >

void tinker::reduceSum2	(	HT(&)	h_ans[HN],
		DPTR	v,
		size_t	nelem,
		int	queue
	)

Sum over all of the elements of a 2D array.

In Fortran syntax:

\[ Ans(k) = \sum_i^n v(k,i), 1 \le k \le HN \]

In C++ syntax:

\[ Ans[k] = \sum_i^n v[i][k], 0 \le k < HN \]

reduceSum2OnDevice()

template<class HT , size_t HN, class DPTR >

void tinker::reduceSum2OnDevice	(	HT(&)	dref[HN],
		DPTR	v,
		size_t	nelem,
		int	queue
	)

Sum over all of the elements of a 2D array. This routine will save the result on the device memory in an asynchronous/non-blocking manner. A valid device array is required for the result on device.

Fortran syntax:

\[ Ans(k) = \sum_i^n v(k,i), 1 \le k \le HN \]

C++ syntax:

\[ Ans[k] = \sum_i^n v[i][k], 0 \le k < HN \]

Template Parameters

HT	Type of the array element.
HN	Length of the result array.
DPTR	Type of the 2D array.

Parameters

dref	Reference to the device array that stores the reduced result.
v	Device pointer to the 2D array.
nelem	Number of elements.
queue	OpenACC queue.

reduceSumOnDevice()

template<class T >

void tinker::reduceSumOnDevice	(	T *	dp_ans,
		const T *	a,
		size_t	nelem,
		int	queue
	)

Sum over all of the elements of an 1D array. This routine will save the result on the device memory in an asynchronous/non-blocking manner. A valid device pointer is required for the result on the device.

\[ Sum = \sum_i^n a_i \]

Parameters

dp_ans	Device pointer used to store the reduction result.
a	Device pointer to the array.
nelem	Number of elements.
queue	OpenACC queue.

scaleArray()

template<class T >

void tinker::scaleArray	(	T *	dst,
		T	scal,
		size_t	nelem,
		int	queue
	)

Multiply all of the elements in an 1D array by a scalar.

\[ a_i = c \cdot a_i \]