allreduce-ompi-ring-segmented.c File Reference

#include "colls_private.h"

Macros
#define	COLL_TUNED_COMPUTED_SEGCOUNT(SEGSIZE, TYPELNG, SEGCOUNT)
#define	COLL_TUNED_COMPUTE_BLOCKCOUNT(COUNT, NUM_BLOCKS, SPLIT_INDEX, EARLY_BLOCK_COUNT, LATE_BLOCK_COUNT)

Functions
int	smpi_coll_tuned_allreduce_ompi_ring_segmented (void *sbuf, void *rbuf, int count, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm)

Macro Definition Documentation

COLL_TUNED_COMPUTED_SEGCOUNT

#define COLL_TUNED_COMPUTED_SEGCOUNT	(		SEGSIZE,
			TYPELNG,
			SEGCOUNT
	)

Value:

if( ((SEGSIZE) >= (TYPELNG)) &&                                     \
        ((SEGSIZE) < ((TYPELNG) * (SEGCOUNT))) ) {                      \
        size_t residual;                                                \
        (SEGCOUNT) = (int)((SEGSIZE) / (TYPELNG));                      \
        residual = (SEGSIZE) - (SEGCOUNT) * (TYPELNG);                  \
        if( residual > ((TYPELNG) >> 1) )                               \
            (SEGCOUNT)++;                                               \
    }                                                                   \

COLL_TUNED_COMPUTE_BLOCKCOUNT

#define COLL_TUNED_COMPUTE_BLOCKCOUNT	(		COUNT,
			NUM_BLOCKS,
			SPLIT_INDEX,
			EARLY_BLOCK_COUNT,
			LATE_BLOCK_COUNT
	)

Value:

EARLY_BLOCK_COUNT = LATE_BLOCK_COUNT = COUNT / NUM_BLOCKS;               \
    SPLIT_INDEX = COUNT % NUM_BLOCKS;                                        \
    if (0 != SPLIT_INDEX) {                                                  \
        EARLY_BLOCK_COUNT = EARLY_BLOCK_COUNT + 1;                           \
    }                                                                        \

Function Documentation

smpi_coll_tuned_allreduce_ompi_ring_segmented()

int smpi_coll_tuned_allreduce_ompi_ring_segmented	(	void *	sbuf,
		void *	rbuf,
		int	count,
		MPI_Datatype	dtype,
		MPI_Op	op,
		MPI_Comm	comm
	)