butterfly_extra.h File Reference

Go to the source code of this file.

Functions
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.
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.
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.
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.
double	butterfly_reduce_b (int **R, int *nR, int nRmax, int **S, int *nS, int nSmax, double *val, int b, int steps, MPI_Comm comm)

---

Function Documentation

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