Functions
int	m2m (double vA1, int A1, int n1, double vA2, int A2, int n2)
int	m2m_sum (double vA1, int A1, int n1, double vA2, int A2, int n2)
int	card_or (int A1, int n1, int A2, int n2)
int	card_and (int A1, int n1, int A2, int n2)
int	set_or (int A1, int n1, int A2, int n2, int *A1orA2)
int	set_and (int A1, int n1, int A2, int n2, int *A1andA2)
int	butterfly_init (int indices, int count, int R, int nR, int *S, int nS, int *com_indices, int com_count, int steps, MPI_Comm comm)
	Initialize tables for butterfly-like communication scheme This routine set up needed tables for the butterfly communication scheme. Sending and receiving tabs should be well allocated(at least size of number of steps in butterfly scheme). Double pointer are partially allocated, the last allocation is performed inside the routine. com_indices and com_count are also allocated inside the routine, thus they are passing by reference. They represent indices which have to be communicated an their number. Algotithm is based 2 parts. The first one identify intersection between processors indices, using 3 successives butterfly communication schemes : bottom up, top down, and top down again. The second part works locally to build sets of indices to communicate.
int	butterfly_reduce (int *R, int nR, int nRmax, int *S, int nS, int nSmax, double *val, int steps, MPI_Comm comm)
	Perform a sparse sum reduction (or mapped reduction) using a butterfly-like communication scheme.
double	truebutterfly_reduce (int *R, int nR, int nRmax, int *S, int nS, int nSmax, double *val, int steps, MPI_Comm comm)
	Perform a sparse sum reduction (or mapped reduction) using a butterfly-like communication scheme.
int	sindex (int T, int nT, int A, int nA)
int	omp_pindex (int T, int nT, int A, int nA)
int	ssort (int *indices, int count, int flag)
int	omp_psort (int *A, int nA, int flag)
int	ring_init (int indices, int count, int R, int nR, int *S, int nS, int steps, MPI_Comm comm)
	Initialize tables for ring-like communication scheme.
int	ring_reduce (int *R, int nR, int nRmax, int *S, int nS, int nSmax, double val, double res_val, int steps, MPI_Comm comm)
	Perform a sparse sum reduction (or mapped reduction) using a ring-like communication scheme.
int	ring_nonblocking_reduce (int *R, int nR, int *S, int nS, double val, double res_val, int steps, MPI_Comm comm)
	Perform a sparse sum reduction (or mapped reduction) using a ring-like non-blocking communication scheme.
int	ring_noempty_reduce (int *R, int nR, int nneR, int *S, int nS, int nneS, double val, double res_val, int steps, MPI_Comm comm)
	Perform a sparse sum reduction (or mapped reduction) using a ring-like non-blocking no-empty communication scheme.
int	truebutterfly_init (int indices, int count, int R, int nR, int *S, int nS, int *com_indices, int com_count, int steps, MPI_Comm comm)
	Initialize tables for butterfly-like communication scheme (true means pair wise) This routine set up needed tables for the butterfly communication scheme. Sending and receiving tabs should be well allocated(at least size of number of steps in butterfly scheme). Double pointer are partially allocated, the last allocation is performed inside the routine. com_indices and com_count are also allocated inside the routine, thus they are passing by reference. They represent indices which have to be communicated an their number. Algotithm is based 2 parts. The first one identify intersection between processors indices, using 3 successives butterfly communication schemes : bottom up, top down, and top down again. The second part works locally to build sets of indices to communicate.

Detailed Description

These are low level internal routines. These are generally not to be used by external users.

Function Documentation

int m2m	(	double *	vA1,
		int *	A1,
		int	n1,
		double *	vA2,
		int *	A2,
		int	n2
	)

Function m2m for "map to map" Extract values from one map (A1, vA1), and for each pixel shared with an other map (A2, vA2), assign pixel value in vA1 and to pixel value in vA2.

Returns:: a number of elements shared between A1 and A2

See also:: m2m_sum

Definition at line 93 of file alm.c.

int m2m_sum	(	double *	vA1,
		int *	A1,
		int	n1,
		double *	vA2,
		int *	A2,
		int	n2
	)

Function m2m_sum for "sum map to map" Extract values from one map (A1, vA1), and for each pixel shared with an other map (A2, vA2), sum pixel value in vA1 to pixel value in vA2.

Returns:: a number of elements shared between A1 and A2

See also:: m2m

Definition at line 118 of file alm.c.

int card_or	(	int *	A1,
		int	n1,
		int *	A2,
		int	n2
	)

Compute $card(A_1 \cup A_2)$ A1 and A2 should be two ascending ordered monotmony sets. of size n1 and n2.

Parameters:

n1	number of elemnets in A1
A1	set of indices
n2	number of elemnets in A2
A2	set of indices

Returns:: size of the union

Definition at line 60 of file als.c.

int card_and	(	int *	A1,
		int	n1,
		int *	A2,
		int	n2
	)

Compute $card(A_1 \cap A_2)$ A1 and A2 should be two ascending ordered monotony sets, of size n1 and n2.

Parameters:

n1	number of elemnets in A1
A1	set of indices
n2	number of elemnets in A2
A2	set of indices

Returns:: size of the intersection

Definition at line 89 of file als.c.

int set_or	(	int *	A1,
		int	n1,
		int *	A2,
		int	n2,
		int *	A1orA2
	)

Compute $A1 \cup A_2$ A1 and A2 should be two ascending ordered sets. It requires the sizes of these two sets, n1 and n2. A1andA2 has to be previouly allocated.

Parameters:

n1	number of elemnets in A1
A1	set of indices
n2	number of elemnets in A2
A2	set of indices
address	to the set A1orA2

Returns:: number of elements in A1orA2

Definition at line 118 of file als.c.

int set_and	(	int *	A1,
		int	n1,
		int *	A2,
		int	n2,
		int *	A1andA2
	)

Compute $A_1 \cap A_2$ A1 and A2 should be two monotony sets in ascending order. It requires the sizes of these two sets, n1 and n2. A1andA2 has to be previously allocated.

Parameters:

n1	number of elemnets in A1
A1	set of indices
n2	number of elemnets in A2
A2	set of indices
address	to the set A1andA2

Returns:: number of elements in A1andA2

Definition at line 162 of file als.c.

int butterfly_init	(	int *	indices,
		int	count,
		int **	R,
		int *	nR,
		int **	S,
		int *	nS,
		int **	com_indices,
		int *	com_count,
		int	steps,
		MPI_Comm	comm
	)

Initialize tables for butterfly-like communication scheme This routine set up needed tables for the butterfly communication scheme. Sending and receiving tabs should be well allocated(at least size of number of steps in butterfly scheme). Double pointer are partially allocated, the last allocation is performed inside the routine. com_indices and com_count are also allocated inside the routine, thus they are passing by reference. They represent indices which have to be communicated an their number. Algotithm is based 2 parts. The first one identify intersection between processors indices, using 3 successives butterfly communication schemes : bottom up, top down, and top down again. The second part works locally to build sets of indices to communicate.

Parameters:

indices	set of indices(monotony) handle by a process.
count	number of elements
R	pointer to receiving maps
nR	array of number of elements in each receiving map
S	pointer to sending maps
nS	array of number of elements in each sending map
com_indices	set of indices(monotony) communicated by a process
com_count	number of elements
steps	number of communication exchange in the butterfly scheme
comm	MPI communicator

Returns:: 0 if no error

Definition at line 37 of file butterfly.c.

double butterfly_reduce	(	int **	R,
		int *	nR,
		int	nRmax,
		int **	S,
		int *	nS,
		int	nSmax,
		double *	val,
		int	steps,
		MPI_Comm	comm
	)

Perform a sparse sum reduction (or mapped reduction) using a butterfly-like communication scheme.

Parameters:

R	pointer to receiving maps
nR	array of number of elements in each receiving map
nRmax	maximum size of received message
S	pointer to sending maps
nS	array of number of elements in each sending map
nSmax	maximum size of sent message
val	set of values (typically values associated to communicated indices)
steps	number of communication exchange in the butterfly scheme
comm	MPI communicator

Returns:: 0 if no error

Definition at line 209 of file butterfly.c.

int truebutterfly_reduce	(	int **	R,
		int *	nR,
		int	nRmax,
		int **	S,
		int *	nS,
		int	nSmax,
		double *	val,
		int	steps,
		MPI_Comm	comm
	)

Perform a sparse sum reduction (or mapped reduction) using a butterfly-like communication scheme.

Perform a sparse sum reduction (or mapped reduction) using a butterfly-like communication scheme (true means pairwise)

Parameters:

R	pointer to receiving maps
nR	array of number of elements in each receiving map
nRmax	maximum size of received message
S	pointer to sending maps
nS	array of number of elements in each sending map
nSmax	maximum size of sent message
val	set of values (typically values associated to communicated indices)
steps	number of communication exchange in the butterfly scheme
comm	MPI communicator

Returns:: 0 if no error

Definition at line 430 of file butterfly_extra.c.

int sindex	(	int *	T,
		int	nT,
		int *	A,
		int	nA
	)

Sequential reindexing

Parameters:

T	monotony array
nT	number of index
A	tab to reindex
nA	number of element to reindex

Returns:: array of indices

Definition at line 18 of file cindex.c.

int omp_pindex	(	int *	T,
		int	nT,
		int *	A,
		int	nA
	)

Multithread (OpenMP) reindexing

Parameters:

T	monotony array
nT	number of index
A	tab to reindex
nA	inumber of element to reindex

Returns:: array of indices

Definition at line 36 of file cindex.c.

int ssort	(	int *	indices,
		int	count,
		int	flag
	)

Sort and merge redundant elements of a set of indices using a specified method. The indices tab, initially an arbitrary set of integers, becomes a monotony set. Available methods :

quick sort
bubble sort
insertion sort
counting sort
shell sort

Parameters:

indices	tab (modified)
count	number of indices
flag	method

Returns:: number of sorted elements

Definition at line 161 of file csort.c.

int omp_psort	(	int *	A,
		int	nA,
		int	flag
	)

Sort and merge redundant elements of a set of indices, using openMP. The indices tab, initially an arbitrary set of integers, becomes a monotony set. Algorithm is diivided in two steps :

each thread sorts, in parallel, a subpart of the set using a specified method.
subsets obtained are merged successively in a binary tree manner.

Available methods for the fully parallel step :

quick sort
bubble sort
insertion sort
counting sort
shell sort

Parameters:

indices	tab (modified)
count	number of elements to sort

Returns:: flag method

Definition at line 291 of file csort.c.

int ring_init	(	int *	indices,
		int	count,
		int **	R,
		int *	nR,
		int **	S,
		int *	nS,
		int	steps,
		MPI_Comm	comm
	)

Initialize tables for ring-like communication scheme.

This routine set up needed tables for the ring communication scheme. Sending and receiving tabs should be well allocated(at least size of number of steps in ring scheme). Double pointer are partially allocated, the last allocation is performed inside the routine (only for R S are just pointer).

Parameters:

indices	set of indices(monotony) handle by a process.
count	number of elements
R	pointer to receiving maps
nR	array of number of elements in each receiving map
S	pointer to sending maps
nS	array of number of elements in each sending map
com_indices	set of indices(monotony) communicated by a process
com_count	number of elements
steps	number of communication exchange in the ring scheme
comm	MPI communicator

Todo:: Ring loop and ring table are set from index 1 to size. Should be shift and be set from index 0 to size-1.

Returns:: 0 if no error

Definition at line 32 of file ring.c.

int ring_reduce	(	int **	R,
		int *	nR,
		int	nRmax,
		int **	S,
		int *	nS,
		int	nSmax,
		double *	val,
		double *	res_val,
		int	steps,
		MPI_Comm	comm
	)

Perform a sparse sum reduction (or mapped reduction) using a ring-like communication scheme.

Parameters:

R	pointer to receiving maps
nR	array of number of elements in each receiving map
nRmax	maximum size of received message
S	pointer to sending maps
nS	array of number of elements in each sending map
nSmax	maximum size of sent message
val	set of values (typically values associated to communicated indices)
steps	number of communication exchange in the butterfly scheme
comm	MPI communicator

Returns:: 0 if no error

Definition at line 82 of file ring.c.

int ring_nonblocking_reduce	(	int **	R,
		int *	nR,
		int **	S,
		int *	nS,
		double *	val,
		double *	res_val,
		int	steps,
		MPI_Comm	comm
	)

Perform a sparse sum reduction (or mapped reduction) using a ring-like non-blocking communication scheme.

Parameters:

R	pointer to receiving maps
nR	array of number of elements in each receiving map
S	pointer to sending maps
nS	array of number of elements in each sending map
val	set of values (typically values associated to communicated indices)
steps	number of communication exchange in the butterfly scheme
comm	MPI communicator

Returns:: 0 if no error

Definition at line 126 of file ring.c.

int ring_noempty_reduce	(	int **	R,
		int *	nR,
		int	nneR,
		int **	S,
		int *	nS,
		int	nneS,
		double *	val,
		double *	res_val,
		int	steps,
		MPI_Comm	comm
	)

Perform a sparse sum reduction (or mapped reduction) using a ring-like non-blocking no-empty communication scheme.

Parameters:

R	pointer to receiving maps
nR	array of number of elements in each receiving map
nneR	number of no-empty receiving messages
S	pointer to sending maps
nS	array of number of elements in each sending map
nneS	number of no-empty sending messages
val	set of values (typically values associated to communicated indices)
steps	number of communication exchange in the butterfly scheme
comm	MPI communicator

Returns:: 0 if no error

Definition at line 185 of file ring.c.

int truebutterfly_init	(	int *	indices,
		int	count,
		int **	R,
		int *	nR,
		int **	S,
		int *	nS,
		int **	com_indices,
		int *	com_count,
		int	steps,
		MPI_Comm	comm
	)

Initialize tables for butterfly-like communication scheme (true means pair wise) This routine set up needed tables for the butterfly communication scheme. Sending and receiving tabs should be well allocated(at least size of number of steps in butterfly scheme). Double pointer are partially allocated, the last allocation is performed inside the routine. com_indices and com_count are also allocated inside the routine, thus they are passing by reference. They represent indices which have to be communicated an their number. Algotithm is based 2 parts. The first one identify intersection between processors indices, using 3 successives butterfly communication schemes : bottom up, top down, and top down again. The second part works locally to build sets of indices to communicate.

Parameters:

indices	set of indices(monotony) handle by a process.
count	number of elements
R	pointer to receiving maps
nR	array of number of elements in each receiving map
S	pointer to sending maps
nS	array of number of elements in each sending map
com_indices	set of indices(monotony) communicated by a process
com_count	number of elements
steps	number of communication exchange in the butterfly scheme
comm	MPI communicator

Returns:: 0 if no error

Definition at line 37 of file truebutterfly.c.

Functions

Detailed Description

Function Documentation