ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Split the tuned framework in two. Move all the functions down in the

Просмотр исходного кода 
base, so that they can now be used by all modules. Keep the decision
functions in tuned.
Этот коммит содержится в:
George Bosilca 2015-02-15 14:27:55 -05:00
родитель 855d422e62
Коммит 8fbcdf685d
28 изменённых файлов: 2637 добавлений и 308 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

558
ompi/mca/coll/base/coll_base.h Обычный файл
Просмотреть файл

@ -0,0 +1,558 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2009 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2008 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2008 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2013 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#ifndef MCA_COLL_TUNED_EXPORT_H
#define MCA_COLL_TUNED_EXPORT_H
#include "ompi_config.h"
#include "mpi.h"
#include "opal/mca/mca.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/request/request.h"
/* need to include our own topo prototypes so we can malloc data on the comm correctly */
#include "coll_tuned_topo.h"
/* also need the dynamic rule structures */
#include "coll_tuned_dynamic_rules.h"
/* some fixed value index vars to simplify certain operations */
typedef enum COLLTYPE {
ALLGATHER = 0, /* 0 */
ALLGATHERV, /* 1 */
ALLREDUCE, /* 2 */
ALLTOALL, /* 3 */
ALLTOALLV, /* 4 */
ALLTOALLW, /* 5 */
BARRIER, /* 6 */
BCAST, /* 7 */
EXSCAN, /* 8 */
GATHER, /* 9 */
GATHERV, /* 10 */
REDUCE, /* 11 */
REDUCESCATTER, /* 12 */
SCAN, /* 13 */
SCATTER, /* 14 */
SCATTERV, /* 15 */
COLLCOUNT /* 16 end counter keep it as last element */
} COLLTYPE_T;
/* defined arg lists to simply auto inclusion of user overriding decision functions */
#define ALLGATHER_ARGS void *sbuf, int scount, struct ompi_datatype_t *sdtype, void *rbuf, int rcount, struct ompi_datatype_t *rdtype, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define ALLGATHERV_ARGS void *sbuf, int scount, struct ompi_datatype_t *sdtype, void * rbuf, int *rcounts, int *disps, struct ompi_datatype_t *rdtype, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define ALLREDUCE_ARGS void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype, struct ompi_op_t *op, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define ALLTOALL_ARGS void *sbuf, int scount, struct ompi_datatype_t *sdtype, void* rbuf, int rcount, struct ompi_datatype_t *rdtype, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define ALLTOALLV_ARGS void *sbuf, int *scounts, int *sdisps, struct ompi_datatype_t *sdtype, void *rbuf, int *rcounts, int *rdisps, struct ompi_datatype_t *rdtype, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define ALLTOALLW_ARGS void *sbuf, int *scounts, int *sdisps, struct ompi_datatype_t **sdtypes, void *rbuf, int *rcounts, int *rdisps, struct ompi_datatype_t **rdtypes, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define BARRIER_ARGS struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define BCAST_ARGS void *buff, int count, struct ompi_datatype_t *datatype, int root, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define EXSCAN_ARGS void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype, struct ompi_op_t *op, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define GATHER_ARGS void *sbuf, int scount, struct ompi_datatype_t *sdtype, void *rbuf, int rcount, struct ompi_datatype_t *rdtype, int root, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define GATHERV_ARGS void *sbuf, int scount, struct ompi_datatype_t *sdtype, void *rbuf, int *rcounts, int *disps, struct ompi_datatype_t *rdtype, int root, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define REDUCE_ARGS void *sbuf, void* rbuf, int count, struct ompi_datatype_t *dtype, struct ompi_op_t *op, int root, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define REDUCESCATTER_ARGS void *sbuf, void *rbuf, int *rcounts, struct ompi_datatype_t *dtype, struct ompi_op_t *op, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define SCAN_ARGS void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype, struct ompi_op_t *op, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define SCATTER_ARGS void *sbuf, int scount, struct ompi_datatype_t *sdtype, void *rbuf, int rcount, struct ompi_datatype_t *rdtype, int root, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define SCATTERV_ARGS void *sbuf, int *scounts, int *disps, struct ompi_datatype_t *sdtype, void* rbuf, int rcount, struct ompi_datatype_t *rdtype, int root, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
/* end defined arg lists to simply auto inclusion of user overriding decision functions */
BEGIN_C_DECLS
/* these are the same across all modules and are loaded at component query time */
extern int ompi_coll_tuned_stream;
extern int ompi_coll_tuned_priority;
extern int ompi_coll_tuned_preallocate_memory_comm_size_limit;
extern bool ompi_coll_tuned_use_dynamic_rules;
extern char* ompi_coll_tuned_dynamic_rules_filename;
extern int ompi_coll_tuned_init_tree_fanout;
extern int ompi_coll_tuned_init_chain_fanout;
extern int ompi_coll_tuned_init_max_requests;
extern int ompi_coll_tuned_alltoall_small_msg;
extern int ompi_coll_tuned_alltoall_intermediate_msg;
/* forced algorithm choices */
/* this structure is for storing the indexes to the forced algorithm mca params... */
/* we get these at component query (so that registered values appear in ompi_infoi) */
struct coll_tuned_force_algorithm_mca_param_indices_t {
int algorithm_param_index; /* which algorithm you want to force */
int segsize_param_index; /* segsize to use (if supported), 0 = no segmentation */
int tree_fanout_param_index; /* tree fanout/in to use */
int chain_fanout_param_index; /* K-chain fanout/in to use */
int max_requests_param_index; /* Maximum number of outstanding send or recv requests */
};
typedef struct coll_tuned_force_algorithm_mca_param_indices_t coll_tuned_force_algorithm_mca_param_indices_t;
/* the following type is for storing actual value obtained from the MCA on each tuned module */
/* via their mca param indices lookup in the component */
/* this structure is stored once per collective type per communicator... */
struct coll_tuned_force_algorithm_params_t {
int algorithm; /* which algorithm you want to force */
int segsize; /* segsize to use (if supported), 0 = no segmentation */
int tree_fanout; /* tree fanout/in to use */
int chain_fanout; /* K-chain fanout/in to use */
int max_requests; /* Maximum number of outstanding send or recv requests */
};
typedef struct coll_tuned_force_algorithm_params_t coll_tuned_force_algorithm_params_t;
/* the indices to the MCA params so that modules can look them up at open / comm create time */
extern coll_tuned_force_algorithm_mca_param_indices_t ompi_coll_tuned_forced_params[COLLCOUNT];
/* the actual max algorithm values (readonly), loaded at component open */
extern int ompi_coll_tuned_forced_max_algorithms[COLLCOUNT];
/*
* coll API functions
*/
/* API functions */
int ompi_coll_tuned_init_query(bool enable_progress_threads,
bool enable_mpi_threads);
mca_coll_base_module_t *
ompi_coll_tuned_comm_query(struct ompi_communicator_t *comm, int *priority);
/* API functions of decision functions and any implementations */
/*
* Note this gets long as we have to have a prototype for each
* MPI collective 4 times.. 2 for the comm type and 2 for each decision
* type.
* we might cut down the decision prototypes by conditional compiling
*/
/* All Gather */
int ompi_coll_tuned_allgather_intra_dec_fixed(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_dec_dynamic(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_do_forced(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_do_this(ALLGATHER_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_allgather_intra_check_forced_init(coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_allgather_intra_bruck(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_recursivedoubling(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_ring(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_neighborexchange(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_basic_linear(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_two_procs(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_inter_dec_fixed(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_inter_dec_dynamic(ALLGATHER_ARGS);
/* All GatherV */
int ompi_coll_tuned_allgatherv_intra_dec_fixed(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_intra_dec_dynamic(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_intra_do_forced(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_intra_do_this(ALLGATHERV_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_allgatherv_intra_check_forced_init(coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_allgatherv_intra_bruck(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_intra_ring(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_intra_neighborexchange(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_intra_basic_default(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_intra_two_procs(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_inter_dec_fixed(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_inter_dec_dynamic(ALLGATHERV_ARGS);
/* All Reduce */
int ompi_coll_tuned_allreduce_intra_dec_fixed(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_intra_dec_dynamic(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_intra_do_forced(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_intra_do_this(ALLREDUCE_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_allreduce_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_allreduce_intra_nonoverlapping(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_intra_recursivedoubling(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_intra_ring(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_intra_ring_segmented(ALLREDUCE_ARGS, uint32_t segsize);
int ompi_coll_tuned_allreduce_intra_basic_linear(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_inter_dec_fixed(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_inter_dec_dynamic(ALLREDUCE_ARGS);
/* AlltoAll */
int ompi_coll_tuned_alltoall_intra_dec_fixed(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_intra_dec_dynamic(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_intra_do_forced(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_intra_do_this(ALLTOALL_ARGS, int algorithm, int faninout, int segsize, int max_requests);
int ompi_coll_tuned_alltoall_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_alltoall_intra_pairwise(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_intra_bruck(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_intra_basic_linear(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_intra_linear_sync(ALLTOALL_ARGS, int max_requests);
int ompi_coll_tuned_alltoall_intra_two_procs(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_inter_dec_fixed(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_inter_dec_dynamic(ALLTOALL_ARGS);
/* AlltoAllV */
int ompi_coll_tuned_alltoallv_intra_dec_fixed(ALLTOALLV_ARGS);
int ompi_coll_tuned_alltoallv_intra_dec_dynamic(ALLTOALLV_ARGS);
int ompi_coll_tuned_alltoallv_intra_do_forced(ALLTOALLV_ARGS);
int ompi_coll_tuned_alltoallv_intra_do_this(ALLTOALLV_ARGS, int algorithm);
int ompi_coll_tuned_alltoallv_intra_check_forced_init(coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_alltoallv_intra_pairwise(ALLTOALLV_ARGS);
int ompi_coll_tuned_alltoallv_intra_basic_linear(ALLTOALLV_ARGS);
int ompi_coll_tuned_alltoallv_inter_dec_fixed(ALLTOALLV_ARGS);
int ompi_coll_tuned_alltoallv_inter_dec_dynamic(ALLTOALLV_ARGS);
/* AlltoAllW */
int ompi_coll_tuned_alltoallw_intra_dec_fixed(ALLTOALLW_ARGS);
int ompi_coll_tuned_alltoallw_intra_dec_dynamic(ALLTOALLW_ARGS);
int ompi_coll_tuned_alltoallw_inter_dec_fixed(ALLTOALLW_ARGS);
int ompi_coll_tuned_alltoallw_inter_dec_dynamic(ALLTOALLW_ARGS);
/* Barrier */
int ompi_coll_tuned_barrier_intra_dec_fixed(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_dec_dynamic(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_do_forced(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_do_this(BARRIER_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_barrier_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_barrier_inter_dec_fixed(BARRIER_ARGS);
int ompi_coll_tuned_barrier_inter_dec_dynamic(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_doublering(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_recursivedoubling(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_bruck(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_two_procs(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_linear(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_tree(BARRIER_ARGS);
/* Bcast */
int ompi_coll_tuned_bcast_intra_generic( BCAST_ARGS, uint32_t count_by_segment, ompi_coll_tree_t* tree );
int ompi_coll_tuned_bcast_intra_dec_fixed(BCAST_ARGS);
int ompi_coll_tuned_bcast_intra_dec_dynamic(BCAST_ARGS);
int ompi_coll_tuned_bcast_intra_do_forced(BCAST_ARGS);
int ompi_coll_tuned_bcast_intra_do_this(BCAST_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_bcast_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_bcast_intra_basic_linear(BCAST_ARGS);
int ompi_coll_tuned_bcast_intra_chain(BCAST_ARGS, uint32_t segsize, int32_t chains);
int ompi_coll_tuned_bcast_intra_pipeline(BCAST_ARGS, uint32_t segsize);
int ompi_coll_tuned_bcast_intra_binomial(BCAST_ARGS, uint32_t segsize);
int ompi_coll_tuned_bcast_intra_bintree(BCAST_ARGS, uint32_t segsize);
int ompi_coll_tuned_bcast_intra_split_bintree(BCAST_ARGS, uint32_t segsize);
int ompi_coll_tuned_bcast_inter_dec_fixed(BCAST_ARGS);
int ompi_coll_tuned_bcast_inter_dec_dynamic(BCAST_ARGS);
/* Exscan */
int ompi_coll_tuned_exscan_intra_dec_fixed(EXSCAN_ARGS);
int ompi_coll_tuned_exscan_intra_dec_dynamic(EXSCAN_ARGS);
int ompi_coll_tuned_exscan_inter_dec_fixed(EXSCAN_ARGS);
int ompi_coll_tuned_exscan_inter_dec_dynamic(EXSCAN_ARGS);
/* Gather */
int ompi_coll_tuned_gather_intra_dec_fixed(GATHER_ARGS);
int ompi_coll_tuned_gather_intra_dec_dynamic(GATHER_ARGS);
int ompi_coll_tuned_gather_intra_do_forced(GATHER_ARGS);
int ompi_coll_tuned_gather_intra_do_this(GATHER_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_gather_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_gather_intra_basic_linear(GATHER_ARGS);
int ompi_coll_tuned_gather_intra_binomial(GATHER_ARGS);
int ompi_coll_tuned_gather_intra_linear_sync(GATHER_ARGS, int first_segment_size);
int ompi_coll_tuned_gather_inter_dec_fixed(GATHER_ARGS);
int ompi_coll_tuned_gather_inter_dec_dynamic(GATHER_ARGS);
/* GatherV */
int ompi_coll_tuned_gatherv_intra_dec_fixed(GATHERV_ARGS);
int ompi_coll_tuned_gatherv_intra_dec_dynamic(GATHER_ARGS);
int ompi_coll_tuned_gatherv_inter_dec_fixed(GATHER_ARGS);
int ompi_coll_tuned_gatherv_inter_dec_dynamic(GATHER_ARGS);
/* Reduce */
int ompi_coll_tuned_reduce_generic( REDUCE_ARGS, ompi_coll_tree_t* tree, int count_by_segment, int max_outstanding_reqs );
int ompi_coll_tuned_reduce_intra_dec_fixed(REDUCE_ARGS);
int ompi_coll_tuned_reduce_intra_dec_dynamic(REDUCE_ARGS);
int ompi_coll_tuned_reduce_intra_do_forced(REDUCE_ARGS);
int ompi_coll_tuned_reduce_intra_do_this(REDUCE_ARGS, int algorithm, int faninout, int segsize, int max_oustanding_reqs);
int ompi_coll_tuned_reduce_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_reduce_intra_basic_linear(REDUCE_ARGS);
int ompi_coll_tuned_reduce_intra_chain(REDUCE_ARGS, uint32_t segsize, int fanout, int max_outstanding_reqs );
int ompi_coll_tuned_reduce_intra_pipeline(REDUCE_ARGS, uint32_t segsize, int max_outstanding_reqs );
int ompi_coll_tuned_reduce_intra_binary(REDUCE_ARGS, uint32_t segsize, int max_outstanding_reqs );
int ompi_coll_tuned_reduce_intra_binomial(REDUCE_ARGS, uint32_t segsize, int max_outstanding_reqs );
int ompi_coll_tuned_reduce_intra_in_order_binary(REDUCE_ARGS, uint32_t segsize, int max_outstanding_reqs );
int ompi_coll_tuned_reduce_inter_dec_fixed(REDUCE_ARGS);
int ompi_coll_tuned_reduce_inter_dec_dynamic(REDUCE_ARGS);
/* Reduce_scatter */
int ompi_coll_tuned_reduce_scatter_intra_dec_fixed(REDUCESCATTER_ARGS);
int ompi_coll_tuned_reduce_scatter_intra_dec_dynamic(REDUCESCATTER_ARGS);
int ompi_coll_tuned_reduce_scatter_intra_do_forced(REDUCESCATTER_ARGS);
int ompi_coll_tuned_reduce_scatter_intra_do_this(REDUCESCATTER_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_reduce_scatter_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_reduce_scatter_intra_nonoverlapping(REDUCESCATTER_ARGS);
int ompi_coll_tuned_reduce_scatter_intra_basic_recursivehalving(REDUCESCATTER_ARGS);
int ompi_coll_tuned_reduce_scatter_intra_ring(REDUCESCATTER_ARGS);
int ompi_coll_tuned_reduce_scatter_inter_dec_fixed(REDUCESCATTER_ARGS);
int ompi_coll_tuned_reduce_scatter_inter_dec_dynamic(REDUCESCATTER_ARGS);
/* Scan */
int ompi_coll_tuned_scan_intra_dec_fixed(SCAN_ARGS);
int ompi_coll_tuned_scan_intra_dec_dynamic(SCAN_ARGS);
int ompi_coll_tuned_scan_inter_dec_fixed(SCAN_ARGS);
int ompi_coll_tuned_scan_inter_dec_dynamic(SCAN_ARGS);
/* Scatter */
int ompi_coll_tuned_scatter_intra_dec_fixed(SCATTER_ARGS);
int ompi_coll_tuned_scatter_intra_dec_dynamic(SCATTER_ARGS);
int ompi_coll_tuned_scatter_intra_do_forced(SCATTER_ARGS);
int ompi_coll_tuned_scatter_intra_do_this(SCATTER_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_scatter_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_scatter_intra_basic_linear(SCATTER_ARGS);
int ompi_coll_tuned_scatter_intra_binomial(SCATTER_ARGS);
int ompi_coll_tuned_scatter_inter_dec_fixed(SCATTER_ARGS);
int ompi_coll_tuned_scatter_inter_dec_dynamic(SCATTER_ARGS);
/* ScatterV */
int ompi_coll_tuned_scatterv_intra_dec_fixed(SCATTERV_ARGS);
int ompi_coll_tuned_scatterv_intra_dec_dynamic(SCATTERV_ARGS);
int ompi_coll_tuned_scatterv_inter_dec_fixed(SCATTERV_ARGS);
int ompi_coll_tuned_scatterv_inter_dec_dynamic(SCATTERV_ARGS);
int mca_coll_tuned_ft_event(int state);
/* Utility functions */
static inline void ompi_coll_tuned_free_reqs(ompi_request_t **reqs, int count)
{
int i;
for (i = 0; i < count; ++i)
ompi_request_free(&reqs[i]);
}
struct mca_coll_tuned_component_t {
/** Base coll component */
mca_coll_base_component_2_0_0_t super;
/** MCA parameter: Priority of this component */
int tuned_priority;
/** global stuff that I need the component to store */
/* MCA parameters first */
/* cached decision table stuff (moved from MCW module) */
ompi_coll_alg_rule_t *all_base_rules;
};
/**
* Convenience typedef
*/
typedef struct mca_coll_tuned_component_t mca_coll_tuned_component_t;
/**
* Global component instance
*/
OMPI_MODULE_DECLSPEC extern mca_coll_tuned_component_t mca_coll_tuned_component;
/*
* Data structure for hanging data off the communicator
* i.e. per module instance
*/
struct mca_coll_tuned_comm_t {
/* standard data for requests and PML usage */
/* Precreate space for requests
* Note this does not effect basic,
* but if in wrong context can confuse a debugger
* this is controlled by an MCA param
*/
ompi_request_t **mcct_reqs;
int mcct_num_reqs;
/*
* tuned topo information caching per communicator
*
* for each communicator we cache the topo information so we can
* reuse without regenerating if we change the root, [or fanout]
* then regenerate and recache this information
*/
/* general tree with n fan out */
ompi_coll_tree_t *cached_ntree;
int cached_ntree_root;
int cached_ntree_fanout;
/* binary tree */
ompi_coll_tree_t *cached_bintree;
int cached_bintree_root;
/* binomial tree */
ompi_coll_tree_t *cached_bmtree;
int cached_bmtree_root;
/* binomial tree */
ompi_coll_tree_t *cached_in_order_bmtree;
int cached_in_order_bmtree_root;
/* chained tree (fanout followed by pipelines) */
ompi_coll_tree_t *cached_chain;
int cached_chain_root;
int cached_chain_fanout;
/* pipeline */
ompi_coll_tree_t *cached_pipeline;
int cached_pipeline_root;
/* in-order binary tree (root of the in-order binary tree is rank 0) */
ompi_coll_tree_t *cached_in_order_bintree;
/* moving to the component */
ompi_coll_com_rule_t *com_rules[COLLCOUNT]; /* the communicator rules for each MPI collective for ONLY my comsize */
/* for forced algorithms we store the information on the module */
/* previously we only had one shared copy, ops, it really is per comm/module */
coll_tuned_force_algorithm_params_t user_forced[COLLCOUNT];
};
typedef struct mca_coll_tuned_comm_t mca_coll_tuned_comm_t;
struct mca_coll_tuned_module_t {
mca_coll_base_module_t super;
mca_coll_tuned_comm_t *tuned_data;
};
typedef struct mca_coll_tuned_module_t mca_coll_tuned_module_t;
OBJ_CLASS_DECLARATION(mca_coll_tuned_module_t);
static inline void mca_coll_tuned_free_reqs(ompi_request_t ** reqs,
int count)
{
int i;
for (i = 0; i < count; ++i)
ompi_request_free(reqs + i);
}
END_C_DECLS
#define COLL_TUNED_UPDATE_BINTREE( OMPI_COMM, TUNED_MODULE, ROOT ) \
do { \
mca_coll_tuned_comm_t* coll_comm = (TUNED_MODULE)->tuned_data; \
if( !( (coll_comm->cached_bintree) \
&& (coll_comm->cached_bintree_root == (ROOT)) ) ) { \
if( coll_comm->cached_bintree ) { /* destroy previous binomial if defined */ \
ompi_coll_tuned_topo_destroy_tree( &(coll_comm->cached_bintree) ); \
} \
coll_comm->cached_bintree = ompi_coll_tuned_topo_build_tree(2,(OMPI_COMM),(ROOT)); \
coll_comm->cached_bintree_root = (ROOT); \
} \
} while (0)
#define COLL_TUNED_UPDATE_BMTREE( OMPI_COMM, TUNED_MODULE, ROOT ) \
do { \
mca_coll_tuned_comm_t* coll_comm = (TUNED_MODULE)->tuned_data; \
if( !( (coll_comm->cached_bmtree) \
&& (coll_comm->cached_bmtree_root == (ROOT)) ) ) { \
if( coll_comm->cached_bmtree ) { /* destroy previous binomial if defined */ \
ompi_coll_tuned_topo_destroy_tree( &(coll_comm->cached_bmtree) ); \
} \
coll_comm->cached_bmtree = ompi_coll_tuned_topo_build_bmtree( (OMPI_COMM), (ROOT) ); \
coll_comm->cached_bmtree_root = (ROOT); \
} \
} while (0)
#define COLL_TUNED_UPDATE_IN_ORDER_BMTREE( OMPI_COMM, TUNED_MODULE, ROOT ) \
do { \
mca_coll_tuned_comm_t* coll_comm = (TUNED_MODULE)->tuned_data; \
if( !( (coll_comm->cached_in_order_bmtree) \
&& (coll_comm->cached_in_order_bmtree_root == (ROOT)) ) ) { \
if( coll_comm->cached_in_order_bmtree ) { /* destroy previous binomial if defined */ \
ompi_coll_tuned_topo_destroy_tree( &(coll_comm->cached_in_order_bmtree) ); \
} \
coll_comm->cached_in_order_bmtree = ompi_coll_tuned_topo_build_in_order_bmtree( (OMPI_COMM), (ROOT) ); \
coll_comm->cached_in_order_bmtree_root = (ROOT); \
} \
} while (0)
#define COLL_TUNED_UPDATE_PIPELINE( OMPI_COMM, TUNED_MODULE, ROOT ) \
do { \
mca_coll_tuned_comm_t* coll_comm = (TUNED_MODULE)->tuned_data; \
if( !( (coll_comm->cached_pipeline) \
&& (coll_comm->cached_pipeline_root == (ROOT)) ) ) { \
if (coll_comm->cached_pipeline) { /* destroy previous pipeline if defined */ \
ompi_coll_tuned_topo_destroy_tree( &(coll_comm->cached_pipeline) ); \
} \
coll_comm->cached_pipeline = ompi_coll_tuned_topo_build_chain( 1, (OMPI_COMM), (ROOT) ); \
coll_comm->cached_pipeline_root = (ROOT); \
} \
} while (0)
#define COLL_TUNED_UPDATE_CHAIN( OMPI_COMM, TUNED_MODULE, ROOT, FANOUT ) \
do { \
mca_coll_tuned_comm_t* coll_comm = (TUNED_MODULE)->tuned_data; \
if( !( (coll_comm->cached_chain) \
&& (coll_comm->cached_chain_root == (ROOT)) \
&& (coll_comm->cached_chain_fanout == (FANOUT)) ) ) { \
if( coll_comm->cached_chain) { /* destroy previous chain if defined */ \
ompi_coll_tuned_topo_destroy_tree( &(coll_comm->cached_chain) ); \
} \
coll_comm->cached_chain = ompi_coll_tuned_topo_build_chain((FANOUT), (OMPI_COMM), (ROOT)); \
coll_comm->cached_chain_root = (ROOT); \
coll_comm->cached_chain_fanout = (FANOUT); \
} \
} while (0)
#define COLL_TUNED_UPDATE_IN_ORDER_BINTREE( OMPI_COMM, TUNED_MODULE ) \
do { \
mca_coll_tuned_comm_t* coll_comm = (TUNED_MODULE)->tuned_data; \
if( !(coll_comm->cached_in_order_bintree) ) { \
/* In-order binary tree topology is defined by communicator size */ \
/* Thus, there is no need to destroy anything */ \
coll_comm->cached_in_order_bintree = \
ompi_coll_tuned_topo_build_in_order_bintree((OMPI_COMM)); \
} \
} while (0)
/**
* This macro give a generic way to compute the best count of
* the segment (i.e. the number of complete datatypes that
* can fit in the specified SEGSIZE). Beware, when this macro
* is called, the SEGCOUNT should be initialized to the count as
* expected by the collective call.
*/
#define COLL_TUNED_COMPUTED_SEGCOUNT(SEGSIZE, TYPELNG, SEGCOUNT) \
if( ((SEGSIZE) >= (TYPELNG)) && \
((SEGSIZE) < ((TYPELNG) * (SEGCOUNT))) ) { \
size_t residual; \
(SEGCOUNT) = (int)((SEGSIZE) / (TYPELNG)); \
residual = (SEGSIZE) - (SEGCOUNT) * (TYPELNG); \
if( residual > ((TYPELNG) >> 1) ) \
(SEGCOUNT)++; \
} \
/**
* This macro gives a generic wait to compute the well distributed block counts
* when the count and number of blocks are fixed.
* Macro returns "early-block" count, "late-block" count, and "split-index"
* which is the block at which we switch from "early-block" count to
* the "late-block" count.
* count = split_index * early_block_count +
* (block_count - split_index) * late_block_count
* We do not perform ANY error checks - make sure that the input values
* make sense (eg. count > num_blocks).
*/
#define COLL_TUNED_COMPUTE_BLOCKCOUNT( COUNT, NUM_BLOCKS, SPLIT_INDEX, \
EARLY_BLOCK_COUNT, LATE_BLOCK_COUNT ) \
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; \
} \
#endif /* MCA_COLL_TUNED_EXPORT_H */

0
ompi/mca/coll/tuned/coll_tuned_allgather.c → ompi/mca/coll/base/coll_base_allgather.c
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_allgatherv.c → ompi/mca/coll/base/coll_base_allgatherv.c
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_allreduce.c → ompi/mca/coll/base/coll_base_allreduce.c
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_alltoall.c → ompi/mca/coll/base/coll_base_alltoall.c
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_alltoallv.c → ompi/mca/coll/base/coll_base_alltoallv.c
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_barrier.c → ompi/mca/coll/base/coll_base_barrier.c
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_bcast.c → ompi/mca/coll/base/coll_base_bcast.c
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_gather.c → ompi/mca/coll/base/coll_base_gather.c
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_reduce.c → ompi/mca/coll/base/coll_base_reduce.c
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_reduce_scatter.c → ompi/mca/coll/base/coll_base_reduce_scatter.c
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_scatter.c → ompi/mca/coll/base/coll_base_scatter.c
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_topo.c → ompi/mca/coll/base/coll_base_topo.c
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_topo.h → ompi/mca/coll/base/coll_base_topo.h
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_util.c → ompi/mca/coll/base/coll_base_util.c
Просмотреть файл

0
ompi/mca/coll/tuned/coll_tuned_util.h → ompi/mca/coll/base/coll_base_util.h
Просмотреть файл

319
ompi/mca/coll/tuned/coll_tuned.h
Просмотреть файл

@ -1,19 +1,8 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2009 The University of Tennessee and The University
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2008 Sun Microsystems, Inc. All rights reserved.
* Copyright (c) 2008 Cisco Systems, Inc. All rights reserved.
* Copyright (c) 2013 Los Alamos National Security, LLC. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -28,55 +17,12 @@
#include "mpi.h"
#include "opal/mca/mca.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/request/request.h"
/* need to include our own topo prototypes so we can malloc data on the comm correctly */
#include "coll_tuned_topo.h"
#include "ompi/mca/coll/base/coll_base.h"
/* also need the dynamic rule structures */
#include "coll_tuned_dynamic_rules.h"
/* some fixed value index vars to simplify certain operations */
typedef enum COLLTYPE {
ALLGATHER = 0, /* 0 */
ALLGATHERV, /* 1 */
ALLREDUCE, /* 2 */
ALLTOALL, /* 3 */
ALLTOALLV, /* 4 */
ALLTOALLW, /* 5 */
BARRIER, /* 6 */
BCAST, /* 7 */
EXSCAN, /* 8 */
GATHER, /* 9 */
GATHERV, /* 10 */
REDUCE, /* 11 */
REDUCESCATTER, /* 12 */
SCAN, /* 13 */
SCATTER, /* 14 */
SCATTERV, /* 15 */
COLLCOUNT /* 16 end counter keep it as last element */
} COLLTYPE_T;
/* defined arg lists to simply auto inclusion of user overriding decision functions */
#define ALLGATHER_ARGS void *sbuf, int scount, struct ompi_datatype_t *sdtype, void *rbuf, int rcount, struct ompi_datatype_t *rdtype, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define ALLGATHERV_ARGS void *sbuf, int scount, struct ompi_datatype_t *sdtype, void * rbuf, int *rcounts, int *disps, struct ompi_datatype_t *rdtype, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define ALLREDUCE_ARGS void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype, struct ompi_op_t *op, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define ALLTOALL_ARGS void *sbuf, int scount, struct ompi_datatype_t *sdtype, void* rbuf, int rcount, struct ompi_datatype_t *rdtype, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define ALLTOALLV_ARGS void *sbuf, int *scounts, int *sdisps, struct ompi_datatype_t *sdtype, void *rbuf, int *rcounts, int *rdisps, struct ompi_datatype_t *rdtype, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define ALLTOALLW_ARGS void *sbuf, int *scounts, int *sdisps, struct ompi_datatype_t **sdtypes, void *rbuf, int *rcounts, int *rdisps, struct ompi_datatype_t **rdtypes, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define BARRIER_ARGS struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define BCAST_ARGS void *buff, int count, struct ompi_datatype_t *datatype, int root, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define EXSCAN_ARGS void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype, struct ompi_op_t *op, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define GATHER_ARGS void *sbuf, int scount, struct ompi_datatype_t *sdtype, void *rbuf, int rcount, struct ompi_datatype_t *rdtype, int root, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define GATHERV_ARGS void *sbuf, int scount, struct ompi_datatype_t *sdtype, void *rbuf, int *rcounts, int *disps, struct ompi_datatype_t *rdtype, int root, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define REDUCE_ARGS void *sbuf, void* rbuf, int count, struct ompi_datatype_t *dtype, struct ompi_op_t *op, int root, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define REDUCESCATTER_ARGS void *sbuf, void *rbuf, int *rcounts, struct ompi_datatype_t *dtype, struct ompi_op_t *op, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define SCAN_ARGS void *sbuf, void *rbuf, int count, struct ompi_datatype_t *dtype, struct ompi_op_t *op, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define SCATTER_ARGS void *sbuf, int scount, struct ompi_datatype_t *sdtype, void *rbuf, int rcount, struct ompi_datatype_t *rdtype, int root, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
#define SCATTERV_ARGS void *sbuf, int *scounts, int *disps, struct ompi_datatype_t *sdtype, void* rbuf, int rcount, struct ompi_datatype_t *rdtype, int root, struct ompi_communicator_t *comm, mca_coll_base_module_t *module
/* end defined arg lists to simply auto inclusion of user overriding decision functions */
BEGIN_C_DECLS
/* these are the same across all modules and are loaded at component query time */
@ -148,12 +94,6 @@ int ompi_coll_tuned_allgather_intra_dec_dynamic(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_do_forced(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_do_this(ALLGATHER_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_allgather_intra_check_forced_init(coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_allgather_intra_bruck(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_recursivedoubling(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_ring(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_neighborexchange(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_basic_linear(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_intra_two_procs(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_inter_dec_fixed(ALLGATHER_ARGS);
int ompi_coll_tuned_allgather_inter_dec_dynamic(ALLGATHER_ARGS);
@ -163,11 +103,6 @@ int ompi_coll_tuned_allgatherv_intra_dec_dynamic(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_intra_do_forced(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_intra_do_this(ALLGATHERV_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_allgatherv_intra_check_forced_init(coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_allgatherv_intra_bruck(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_intra_ring(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_intra_neighborexchange(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_intra_basic_default(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_intra_two_procs(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_inter_dec_fixed(ALLGATHERV_ARGS);
int ompi_coll_tuned_allgatherv_inter_dec_dynamic(ALLGATHERV_ARGS);
@ -177,11 +112,6 @@ int ompi_coll_tuned_allreduce_intra_dec_dynamic(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_intra_do_forced(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_intra_do_this(ALLREDUCE_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_allreduce_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_allreduce_intra_nonoverlapping(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_intra_recursivedoubling(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_intra_ring(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_intra_ring_segmented(ALLREDUCE_ARGS, uint32_t segsize);
int ompi_coll_tuned_allreduce_intra_basic_linear(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_inter_dec_fixed(ALLREDUCE_ARGS);
int ompi_coll_tuned_allreduce_inter_dec_dynamic(ALLREDUCE_ARGS);
@ -191,11 +121,6 @@ int ompi_coll_tuned_alltoall_intra_dec_dynamic(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_intra_do_forced(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_intra_do_this(ALLTOALL_ARGS, int algorithm, int faninout, int segsize, int max_requests);
int ompi_coll_tuned_alltoall_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_alltoall_intra_pairwise(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_intra_bruck(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_intra_basic_linear(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_intra_linear_sync(ALLTOALL_ARGS, int max_requests);
int ompi_coll_tuned_alltoall_intra_two_procs(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_inter_dec_fixed(ALLTOALL_ARGS);
int ompi_coll_tuned_alltoall_inter_dec_dynamic(ALLTOALL_ARGS);
@ -205,8 +130,6 @@ int ompi_coll_tuned_alltoallv_intra_dec_dynamic(ALLTOALLV_ARGS);
int ompi_coll_tuned_alltoallv_intra_do_forced(ALLTOALLV_ARGS);
int ompi_coll_tuned_alltoallv_intra_do_this(ALLTOALLV_ARGS, int algorithm);
int ompi_coll_tuned_alltoallv_intra_check_forced_init(coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_alltoallv_intra_pairwise(ALLTOALLV_ARGS);
int ompi_coll_tuned_alltoallv_intra_basic_linear(ALLTOALLV_ARGS);
int ompi_coll_tuned_alltoallv_inter_dec_fixed(ALLTOALLV_ARGS);
int ompi_coll_tuned_alltoallv_inter_dec_dynamic(ALLTOALLV_ARGS);
@ -224,12 +147,6 @@ int ompi_coll_tuned_barrier_intra_do_this(BARRIER_ARGS, int algorithm, int fanin
int ompi_coll_tuned_barrier_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_barrier_inter_dec_fixed(BARRIER_ARGS);
int ompi_coll_tuned_barrier_inter_dec_dynamic(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_doublering(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_recursivedoubling(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_bruck(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_two_procs(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_linear(BARRIER_ARGS);
int ompi_coll_tuned_barrier_intra_tree(BARRIER_ARGS);
/* Bcast */
int ompi_coll_tuned_bcast_intra_generic( BCAST_ARGS, uint32_t count_by_segment, ompi_coll_tree_t* tree );
@ -238,12 +155,6 @@ int ompi_coll_tuned_bcast_intra_dec_dynamic(BCAST_ARGS);
int ompi_coll_tuned_bcast_intra_do_forced(BCAST_ARGS);
int ompi_coll_tuned_bcast_intra_do_this(BCAST_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_bcast_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_bcast_intra_basic_linear(BCAST_ARGS);
int ompi_coll_tuned_bcast_intra_chain(BCAST_ARGS, uint32_t segsize, int32_t chains);
int ompi_coll_tuned_bcast_intra_pipeline(BCAST_ARGS, uint32_t segsize);
int ompi_coll_tuned_bcast_intra_binomial(BCAST_ARGS, uint32_t segsize);
int ompi_coll_tuned_bcast_intra_bintree(BCAST_ARGS, uint32_t segsize);
int ompi_coll_tuned_bcast_intra_split_bintree(BCAST_ARGS, uint32_t segsize);
int ompi_coll_tuned_bcast_inter_dec_fixed(BCAST_ARGS);
int ompi_coll_tuned_bcast_inter_dec_dynamic(BCAST_ARGS);
@ -259,9 +170,6 @@ int ompi_coll_tuned_gather_intra_dec_dynamic(GATHER_ARGS);
int ompi_coll_tuned_gather_intra_do_forced(GATHER_ARGS);
int ompi_coll_tuned_gather_intra_do_this(GATHER_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_gather_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_gather_intra_basic_linear(GATHER_ARGS);
int ompi_coll_tuned_gather_intra_binomial(GATHER_ARGS);
int ompi_coll_tuned_gather_intra_linear_sync(GATHER_ARGS, int first_segment_size);
int ompi_coll_tuned_gather_inter_dec_fixed(GATHER_ARGS);
int ompi_coll_tuned_gather_inter_dec_dynamic(GATHER_ARGS);
@ -278,12 +186,6 @@ int ompi_coll_tuned_reduce_intra_dec_dynamic(REDUCE_ARGS);
int ompi_coll_tuned_reduce_intra_do_forced(REDUCE_ARGS);
int ompi_coll_tuned_reduce_intra_do_this(REDUCE_ARGS, int algorithm, int faninout, int segsize, int max_oustanding_reqs);
int ompi_coll_tuned_reduce_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_reduce_intra_basic_linear(REDUCE_ARGS);
int ompi_coll_tuned_reduce_intra_chain(REDUCE_ARGS, uint32_t segsize, int fanout, int max_outstanding_reqs );
int ompi_coll_tuned_reduce_intra_pipeline(REDUCE_ARGS, uint32_t segsize, int max_outstanding_reqs );
int ompi_coll_tuned_reduce_intra_binary(REDUCE_ARGS, uint32_t segsize, int max_outstanding_reqs );
int ompi_coll_tuned_reduce_intra_binomial(REDUCE_ARGS, uint32_t segsize, int max_outstanding_reqs );
int ompi_coll_tuned_reduce_intra_in_order_binary(REDUCE_ARGS, uint32_t segsize, int max_outstanding_reqs );
int ompi_coll_tuned_reduce_inter_dec_fixed(REDUCE_ARGS);
int ompi_coll_tuned_reduce_inter_dec_dynamic(REDUCE_ARGS);
@ -293,10 +195,6 @@ int ompi_coll_tuned_reduce_scatter_intra_dec_dynamic(REDUCESCATTER_ARGS);
int ompi_coll_tuned_reduce_scatter_intra_do_forced(REDUCESCATTER_ARGS);
int ompi_coll_tuned_reduce_scatter_intra_do_this(REDUCESCATTER_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_reduce_scatter_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_reduce_scatter_intra_nonoverlapping(REDUCESCATTER_ARGS);
int ompi_coll_tuned_reduce_scatter_intra_basic_recursivehalving(REDUCESCATTER_ARGS);
int ompi_coll_tuned_reduce_scatter_intra_ring(REDUCESCATTER_ARGS);
int ompi_coll_tuned_reduce_scatter_inter_dec_fixed(REDUCESCATTER_ARGS);
int ompi_coll_tuned_reduce_scatter_inter_dec_dynamic(REDUCESCATTER_ARGS);
@ -312,8 +210,6 @@ int ompi_coll_tuned_scatter_intra_dec_dynamic(SCATTER_ARGS);
int ompi_coll_tuned_scatter_intra_do_forced(SCATTER_ARGS);
int ompi_coll_tuned_scatter_intra_do_this(SCATTER_ARGS, int algorithm, int faninout, int segsize);
int ompi_coll_tuned_scatter_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices);
int ompi_coll_tuned_scatter_intra_basic_linear(SCATTER_ARGS);
int ompi_coll_tuned_scatter_intra_binomial(SCATTER_ARGS);
int ompi_coll_tuned_scatter_inter_dec_fixed(SCATTER_ARGS);
int ompi_coll_tuned_scatter_inter_dec_dynamic(SCATTER_ARGS);
@ -325,16 +221,6 @@ int ompi_coll_tuned_scatterv_inter_dec_dynamic(SCATTERV_ARGS);
int mca_coll_tuned_ft_event(int state);
/* Utility functions */
static inline void ompi_coll_tuned_free_reqs(ompi_request_t **reqs, int count)
{
int i;
for (i = 0; i < count; ++i)
ompi_request_free(&reqs[i]);
}
struct mca_coll_tuned_component_t {
/** Base coll component */
mca_coll_base_component_2_0_0_t super;
@ -359,200 +245,17 @@ typedef struct mca_coll_tuned_component_t mca_coll_tuned_component_t;
*/
OMPI_MODULE_DECLSPEC extern mca_coll_tuned_component_t mca_coll_tuned_component;
/*
* Data structure for hanging data off the communicator
* i.e. per module instance
*/
struct mca_coll_tuned_comm_t {
/* standard data for requests and PML usage */
/* Precreate space for requests
* Note this does not effect basic,
* but if in wrong context can confuse a debugger
* this is controlled by an MCA param
*/
ompi_request_t **mcct_reqs;
int mcct_num_reqs;
/*
* tuned topo information caching per communicator
*
* for each communicator we cache the topo information so we can
* reuse without regenerating if we change the root, [or fanout]
* then regenerate and recache this information
*/
/* general tree with n fan out */
ompi_coll_tree_t *cached_ntree;
int cached_ntree_root;
int cached_ntree_fanout;
/* binary tree */
ompi_coll_tree_t *cached_bintree;
int cached_bintree_root;
/* binomial tree */
ompi_coll_tree_t *cached_bmtree;
int cached_bmtree_root;
/* binomial tree */
ompi_coll_tree_t *cached_in_order_bmtree;
int cached_in_order_bmtree_root;
/* chained tree (fanout followed by pipelines) */
ompi_coll_tree_t *cached_chain;
int cached_chain_root;
int cached_chain_fanout;
/* pipeline */
ompi_coll_tree_t *cached_pipeline;
int cached_pipeline_root;
/* in-order binary tree (root of the in-order binary tree is rank 0) */
ompi_coll_tree_t *cached_in_order_bintree;
/* moving to the component */
ompi_coll_com_rule_t *com_rules[COLLCOUNT]; /* the communicator rules for each MPI collective for ONLY my comsize */
/* for forced algorithms we store the information on the module */
/* previously we only had one shared copy, ops, it really is per comm/module */
coll_tuned_force_algorithm_params_t user_forced[COLLCOUNT];
};
typedef struct mca_coll_tuned_comm_t mca_coll_tuned_comm_t;
struct mca_coll_tuned_module_t {
mca_coll_base_module_t super;
mca_coll_tuned_comm_t *tuned_data;
mca_coll_base_module_t super;
/* for forced algorithms we store the information on the module */
/* previously we only had one shared copy, ops, it really is per comm/module */
coll_tuned_force_algorithm_params_t user_forced[COLLCOUNT];
/* the communicator rules for each MPI collective for ONLY my comsize */
ompi_coll_com_rule_t *com_rules[COLLCOUNT];
};
typedef struct mca_coll_tuned_module_t mca_coll_tuned_module_t;
OBJ_CLASS_DECLARATION(mca_coll_tuned_module_t);
static inline void mca_coll_tuned_free_reqs(ompi_request_t ** reqs,
int count)
{
int i;
for (i = 0; i < count; ++i)
ompi_request_free(reqs + i);
}
END_C_DECLS
#define COLL_TUNED_UPDATE_BINTREE( OMPI_COMM, TUNED_MODULE, ROOT ) \
do { \
mca_coll_tuned_comm_t* coll_comm = (TUNED_MODULE)->tuned_data; \
if( !( (coll_comm->cached_bintree) \
&& (coll_comm->cached_bintree_root == (ROOT)) ) ) { \
if( coll_comm->cached_bintree ) { /* destroy previous binomial if defined */ \
ompi_coll_tuned_topo_destroy_tree( &(coll_comm->cached_bintree) ); \
} \
coll_comm->cached_bintree = ompi_coll_tuned_topo_build_tree(2,(OMPI_COMM),(ROOT)); \
coll_comm->cached_bintree_root = (ROOT); \
} \
} while (0)
#define COLL_TUNED_UPDATE_BMTREE( OMPI_COMM, TUNED_MODULE, ROOT ) \
do { \
mca_coll_tuned_comm_t* coll_comm = (TUNED_MODULE)->tuned_data; \
if( !( (coll_comm->cached_bmtree) \
&& (coll_comm->cached_bmtree_root == (ROOT)) ) ) { \
if( coll_comm->cached_bmtree ) { /* destroy previous binomial if defined */ \
ompi_coll_tuned_topo_destroy_tree( &(coll_comm->cached_bmtree) ); \
} \
coll_comm->cached_bmtree = ompi_coll_tuned_topo_build_bmtree( (OMPI_COMM), (ROOT) ); \
coll_comm->cached_bmtree_root = (ROOT); \
} \
} while (0)
#define COLL_TUNED_UPDATE_IN_ORDER_BMTREE( OMPI_COMM, TUNED_MODULE, ROOT ) \
do { \
mca_coll_tuned_comm_t* coll_comm = (TUNED_MODULE)->tuned_data; \
if( !( (coll_comm->cached_in_order_bmtree) \
&& (coll_comm->cached_in_order_bmtree_root == (ROOT)) ) ) { \
if( coll_comm->cached_in_order_bmtree ) { /* destroy previous binomial if defined */ \
ompi_coll_tuned_topo_destroy_tree( &(coll_comm->cached_in_order_bmtree) ); \
} \
coll_comm->cached_in_order_bmtree = ompi_coll_tuned_topo_build_in_order_bmtree( (OMPI_COMM), (ROOT) ); \
coll_comm->cached_in_order_bmtree_root = (ROOT); \
} \
} while (0)
#define COLL_TUNED_UPDATE_PIPELINE( OMPI_COMM, TUNED_MODULE, ROOT ) \
do { \
mca_coll_tuned_comm_t* coll_comm = (TUNED_MODULE)->tuned_data; \
if( !( (coll_comm->cached_pipeline) \
&& (coll_comm->cached_pipeline_root == (ROOT)) ) ) { \
if (coll_comm->cached_pipeline) { /* destroy previous pipeline if defined */ \
ompi_coll_tuned_topo_destroy_tree( &(coll_comm->cached_pipeline) ); \
} \
coll_comm->cached_pipeline = ompi_coll_tuned_topo_build_chain( 1, (OMPI_COMM), (ROOT) ); \
coll_comm->cached_pipeline_root = (ROOT); \
} \
} while (0)
#define COLL_TUNED_UPDATE_CHAIN( OMPI_COMM, TUNED_MODULE, ROOT, FANOUT ) \
do { \
mca_coll_tuned_comm_t* coll_comm = (TUNED_MODULE)->tuned_data; \
if( !( (coll_comm->cached_chain) \
&& (coll_comm->cached_chain_root == (ROOT)) \
&& (coll_comm->cached_chain_fanout == (FANOUT)) ) ) { \
if( coll_comm->cached_chain) { /* destroy previous chain if defined */ \
ompi_coll_tuned_topo_destroy_tree( &(coll_comm->cached_chain) ); \
} \
coll_comm->cached_chain = ompi_coll_tuned_topo_build_chain((FANOUT), (OMPI_COMM), (ROOT)); \
coll_comm->cached_chain_root = (ROOT); \
coll_comm->cached_chain_fanout = (FANOUT); \
} \
} while (0)
#define COLL_TUNED_UPDATE_IN_ORDER_BINTREE( OMPI_COMM, TUNED_MODULE ) \
do { \
mca_coll_tuned_comm_t* coll_comm = (TUNED_MODULE)->tuned_data; \
if( !(coll_comm->cached_in_order_bintree) ) { \
/* In-order binary tree topology is defined by communicator size */ \
/* Thus, there is no need to destroy anything */ \
coll_comm->cached_in_order_bintree = \
ompi_coll_tuned_topo_build_in_order_bintree((OMPI_COMM)); \
} \
} while (0)
/**
* This macro give a generic way to compute the best count of
* the segment (i.e. the number of complete datatypes that
* can fit in the specified SEGSIZE). Beware, when this macro
* is called, the SEGCOUNT should be initialized to the count as
* expected by the collective call.
*/
#define COLL_TUNED_COMPUTED_SEGCOUNT(SEGSIZE, TYPELNG, SEGCOUNT) \
if( ((SEGSIZE) >= (TYPELNG)) && \
((SEGSIZE) < ((TYPELNG) * (SEGCOUNT))) ) { \
size_t residual; \
(SEGCOUNT) = (int)((SEGSIZE) / (TYPELNG)); \
residual = (SEGSIZE) - (SEGCOUNT) * (TYPELNG); \
if( residual > ((TYPELNG) >> 1) ) \
(SEGCOUNT)++; \
} \
/**
* This macro gives a generic wait to compute the well distributed block counts
* when the count and number of blocks are fixed.
* Macro returns "early-block" count, "late-block" count, and "split-index"
* which is the block at which we switch from "early-block" count to
* the "late-block" count.
* count = split_index * early_block_count +
* (block_count - split_index) * late_block_count
* We do not perform ANY error checks - make sure that the input values
* make sense (eg. count > num_blocks).
*/
#define COLL_TUNED_COMPUTE_BLOCKCOUNT( COUNT, NUM_BLOCKS, SPLIT_INDEX, \
EARLY_BLOCK_COUNT, LATE_BLOCK_COUNT ) \
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; \
} \
#endif /* MCA_COLL_TUNED_EXPORT_H */
#endif /* MCA_COLL_TUNED_EXPORT_H */

218
ompi/mca/coll/tuned/coll_tuned_allgather_decision.c Обычный файл
Просмотреть файл

@ -0,0 +1,218 @@
/*
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "opal/util/bit_ops.h"
#include "ompi/constants.h"
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/coll/base/coll_base_topo.h"
#include "ompi/mca/coll/base/coll_base_util.h"
#include "coll_tuned.h"
/* allgather algorithm variables */
static int coll_tuned_allgather_forced_algorithm = 0;
static int coll_tuned_allgather_segment_size = 0;
static int coll_tuned_allgather_tree_fanout;
static int coll_tuned_allgather_chain_fanout;
/* valid values for coll_tuned_allgather_forced_algorithm */
static mca_base_var_enum_value_t allgather_algorithms[] = {
{0, "ignore"},
{1, "linear"},
{2, "bruck"},
{3, "recursive_doubling"},
{4, "ring"},
{5, "neighbor"},
{6, "two_proc"},
{0, NULL}
};
/* The following are used by dynamic and forced rules */
/* publish details of each algorithm and if its forced/fixed/locked in */
/* as you add methods/algorithms you must update this and the query/map
routines */
/* this routine is called by the component only */
/* this makes sure that the mca parameters are set to their initial values
and perms */
/* module does not call this they call the forced_getvalues routine instead */
int
ompi_coll_tuned_allgather_intra_check_forced_init(coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
{
mca_base_var_enum_t *new_enum;
int cnt;
for( cnt = 0; NULL != allgather_algorithms[cnt].string; cnt++ );
ompi_coll_tuned_forced_max_algorithms[ALLGATHER] = cnt;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allgather_algorithm_count",
"Number of allgather algorithms available",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,
&cnt);
/* MPI_T: This variable should eventually be bound to a communicator */
coll_tuned_allgather_forced_algorithm = 0;
(void) mca_base_var_enum_create("coll_tuned_allgather_algorithms", allgather_algorithms, &new_enum);
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allgather_algorithm",
"Which allallgather algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 bruck, 3 recursive doubling, 4 ring, 5 neighbor exchange, 6: two proc only.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_allgather_forced_algorithm);
OBJ_RELEASE(new_enum);
if (mca_param_indices->algorithm_param_index < 0) {
return mca_param_indices->algorithm_param_index;
}
coll_tuned_allgather_segment_size = 0;
mca_param_indices->segsize_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allgather_algorithm_segmentsize",
"Segment size in bytes used by default for allgather algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation. Currently, available algorithms do not support segmentation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_allgather_segment_size);
coll_tuned_allgather_tree_fanout = ompi_coll_tuned_init_tree_fanout; /* get system wide default */
mca_param_indices->tree_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allgather_algorithm_tree_fanout",
"Fanout for n-tree used for allgather algorithms. Only has meaning if algorithm is forced and supports n-tree topo based operation. Currently, available algorithms do not support n-tree topologies.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_allgather_tree_fanout);
coll_tuned_allgather_chain_fanout = ompi_coll_tuned_init_chain_fanout; /* get system wide default */
mca_param_indices->chain_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allgather_algorithm_chain_fanout",
"Fanout for chains used for allgather algorithms. Only has meaning if algorithm is forced and supports chain topo based operation. Currently, available algorithms do not support chain topologies.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_allgather_chain_fanout);
return (MPI_SUCCESS);
}
int ompi_coll_tuned_allgather_intra_do_forced(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:allgather_intra_do_forced selected algorithm %d",
tuned_module->user_forced[ALLGATHER].algorithm));
switch (tuned_module->user_forced[ALLGATHER].algorithm) {
case (0):
return ompi_coll_tuned_allgather_intra_dec_fixed(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
case (1):
return ompi_coll_base_allgather_intra_basic_linear(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
case (2):
return ompi_coll_base_allgather_intra_bruck(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
case (3):
return ompi_coll_base_allgather_intra_recursivedoubling(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
case (4):
return ompi_coll_base_allgather_intra_ring(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
case (5):
return ompi_coll_base_allgather_intra_neighborexchange(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
case (6):
return ompi_coll_base_allgather_intra_two_procs(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:allgather_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
tuned_module->user_forced[ALLGATHER].algorithm,
ompi_coll_tuned_forced_max_algorithms[ALLGATHER]));
return (MPI_ERR_ARG);
}
int ompi_coll_tuned_allgather_intra_do_this(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm, int faninout, int segsize)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:allgather_intra_do_this selected algorithm %d topo faninout %d segsize %d",
algorithm, faninout, segsize));
switch (algorithm) {
case (0):
return ompi_coll_tuned_allgather_intra_dec_fixed(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
case (1):
return ompi_coll_base_allgather_intra_basic_linear(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
case (2):
return ompi_coll_base_allgather_intra_bruck(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
case (3):
return ompi_coll_base_allgather_intra_recursivedoubling(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
case (4):
return ompi_coll_base_allgather_intra_ring(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
case (5):
return ompi_coll_base_allgather_intra_neighborexchange(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
case (6):
return ompi_coll_base_allgather_intra_two_procs(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:allgather_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[ALLGATHER]));
return (MPI_ERR_ARG);
}

212
ompi/mca/coll/tuned/coll_tuned_allgatherv_decision.c Обычный файл
Просмотреть файл

@ -0,0 +1,212 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "ompi/constants.h"
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "coll_tuned.h"
#include "ompi/mca/coll/base/coll_base_topo.h"
#include "ompi/mca/coll/base/coll_base_util.h"
/* allgatherv algorithm variables */
static int coll_tuned_allgatherv_forced_algorithm = 0;
static int coll_tuned_allgatherv_segment_size = 0;
static int coll_tuned_allgatherv_tree_fanout;
static int coll_tuned_allgatherv_chain_fanout;
/* valid values for coll_tuned_allgatherv_forced_algorithm */
static mca_base_var_enum_value_t allgatherv_algorithms[] = {
{0, "ignore"},
{1, "default"},
{2, "bruck"},
{3, "ring"},
{4, "neighbor"},
{5, "two_proc"},
{0, NULL}
};
/* The following are used by dynamic and forced rules */
/* publish details of each algorithm and if its forced/fixed/locked in */
/* as you add methods/algorithms you must update this and the query/map
routines */
/* this routine is called by the component only */
/* this makes sure that the mca parameters are set to their initial values
and perms */
/* module does not call this they call the forced_getvalues routine instead */
int
ompi_coll_tuned_allgatherv_intra_check_forced_init(coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
{
mca_base_var_enum_t *new_enum;
int cnt;
for( cnt = 0; NULL != allgatherv_algorithms[cnt].string; cnt++ );
ompi_coll_tuned_forced_max_algorithms[ALLGATHERV] = cnt;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allgatherv_algorithm_count",
"Number of allgatherv algorithms available",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,
&cnt);
/* MPI_T: This variable should eventually be bound to a communicator */
coll_tuned_allgatherv_forced_algorithm = 0;
(void) mca_base_var_enum_create("coll_tuned_allgatherv_algorithms", allgatherv_algorithms, &new_enum);
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allgatherv_algorithm",
"Which allallgatherv algorithm is used. Can be locked down to choice of: 0 ignore, 1 default (allgathervv + bcast), 2 bruck, 3 ring, 4 neighbor exchange, 5: two proc only.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_allgatherv_forced_algorithm);
OBJ_RELEASE(new_enum);
if (mca_param_indices->algorithm_param_index < 0) {
return mca_param_indices->algorithm_param_index;
}
coll_tuned_allgatherv_segment_size = 0;
mca_param_indices->segsize_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allgatherv_algorithm_segmentsize",
"Segment size in bytes used by default for allgatherv algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation. Currently, available algorithms do not support segmentation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_allgatherv_segment_size);
coll_tuned_allgatherv_tree_fanout = ompi_coll_tuned_init_tree_fanout; /* get system wide default */
mca_param_indices->tree_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allgatherv_algorithm_tree_fanout",
"Fanout for n-tree used for allgatherv algorithms. Only has meaning if algorithm is forced and supports n-tree topo based operation. Currently, available algorithms do not support n-tree topologies.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_allgatherv_tree_fanout);
coll_tuned_allgatherv_chain_fanout = ompi_coll_tuned_init_chain_fanout; /* get system wide default */
mca_param_indices->chain_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allgatherv_algorithm_chain_fanout",
"Fanout for chains used for allgatherv algorithms. Only has meaning if algorithm is forced and supports chain topo based operation. Currently, available algorithms do not support chain topologies.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_allgatherv_chain_fanout);
return (MPI_SUCCESS);
}
int ompi_coll_tuned_allgatherv_intra_do_forced(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void *rbuf, int *rcounts,
int *rdispls,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:allgatherv_intra_do_forced selected algorithm %d",
tuned_module->user_forced[ALLGATHERV].algorithm));
switch (tuned_module->user_forced[ALLGATHERV].algorithm) {
case (0):
return ompi_coll_tuned_allgatherv_intra_dec_fixed(sbuf, scount, sdtype,
rbuf, rcounts, rdispls, rdtype,
comm, module);
case (1):
return ompi_coll_base_allgatherv_intra_basic_default(sbuf, scount, sdtype,
rbuf, rcounts, rdispls, rdtype,
comm, module);
case (2):
return ompi_coll_base_allgatherv_intra_bruck(sbuf, scount, sdtype,
rbuf, rcounts, rdispls, rdtype,
comm, module);
case (3):
return ompi_coll_base_allgatherv_intra_ring(sbuf, scount, sdtype,
rbuf, rcounts, rdispls, rdtype,
comm, module);
case (4):
return ompi_coll_base_allgatherv_intra_neighborexchange(sbuf, scount, sdtype,
rbuf, rcounts, rdispls, rdtype,
comm, module);
case (5):
return ompi_coll_base_allgatherv_intra_two_procs(sbuf, scount, sdtype,
rbuf, rcounts, rdispls, rdtype,
comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:allgatherv_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
tuned_module->user_forced[ALLGATHERV].algorithm,
ompi_coll_tuned_forced_max_algorithms[ALLGATHERV]));
return (MPI_ERR_ARG);
}
int ompi_coll_tuned_allgatherv_intra_do_this(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void *rbuf, int *rcounts,
int *rdispls,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm, int faninout,
int segsize)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:allgatherv_intra_do_this selected algorithm %d topo faninout %d segsize %d",
algorithm, faninout, segsize));
switch (algorithm) {
case (0):
return ompi_coll_tuned_allgatherv_intra_dec_fixed(sbuf, scount, sdtype,
rbuf, rcounts, rdispls, rdtype,
comm, module);
case (1):
return ompi_coll_base_allgatherv_intra_basic_default(sbuf, scount, sdtype,
rbuf, rcounts, rdispls, rdtype,
comm, module);
case (2):
return ompi_coll_base_allgatherv_intra_bruck(sbuf, scount, sdtype,
rbuf, rcounts, rdispls, rdtype,
comm, module);
case (3):
return ompi_coll_base_allgatherv_intra_ring(sbuf, scount, sdtype,
rbuf, rcounts, rdispls, rdtype,
comm, module);
case (4):
return ompi_coll_base_allgatherv_intra_neighborexchange(sbuf, scount, sdtype,
rbuf, rcounts, rdispls, rdtype,
comm, module);
case (5):
return ompi_coll_base_allgatherv_intra_two_procs(sbuf, scount, sdtype,
rbuf, rcounts, rdispls, rdtype,
comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:allgatherv_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[ALLGATHERV]));
return (MPI_ERR_ARG);
}

182
ompi/mca/coll/tuned/coll_tuned_allreduce_decision.c Обычный файл
Просмотреть файл

@ -0,0 +1,182 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "opal/util/bit_ops.h"
#include "ompi/constants.h"
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/pml/pml.h"
#include "ompi/op/op.h"
#include "coll_tuned.h"
#include "ompi/mca/coll/base/coll_base_topo.h"
#include "ompi/mca/coll/base/coll_base_util.h"
/* allreduce algorithm variables */
static int coll_tuned_allreduce_algorithm_count = 5;
static int coll_tuned_allreduce_forced_algorithm = 0;
static int coll_tuned_allreduce_segment_size = 0;
static int coll_tuned_allreduce_tree_fanout;
static int coll_tuned_allreduce_chain_fanout;
/* valid values for coll_tuned_allreduce_forced_algorithm */
static mca_base_var_enum_value_t allreduce_algorithms[] = {
{0, "ignore"},
{1, "basic_linear"},
{2, "nonoverlapping"},
{3, "recursive_doubling"},
{4, "ring"},
{5, "segmented_ring"},
{0, NULL}
};
/* The following are used by dynamic and forced rules */
/* publish details of each algorithm and if its forced/fixed/locked in */
/* as you add methods/algorithms you must update this and the query/map routines */
/* this routine is called by the component only */
/* this makes sure that the mca parameters are set to their initial values and perms */
/* module does not call this they call the forced_getvalues routine instead */
int ompi_coll_tuned_allreduce_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
{
mca_base_var_enum_t *new_enum;
int cnt;
for( cnt = 0; NULL != allreduce_algorithms[cnt].string; cnt++ );
ompi_coll_tuned_forced_max_algorithms[ALLREDUCE] = cnt;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allreduce_algorithm_count",
"Number of allreduce algorithms available",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,
&cnt);
/* MPI_T: This variable should eventually be bound to a communicator */
coll_tuned_allreduce_forced_algorithm = 0;
(void) mca_base_var_enum_create("coll_tuned_allreduce_algorithms", allreduce_algorithms, &new_enum);
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allreduce_algorithm",
"Which allreduce algorithm is used. Can be locked down to any of: 0 ignore, 1 basic linear, 2 nonoverlapping (tuned reduce + tuned bcast), 3 recursive doubling, 4 ring, 5 segmented ring",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_allreduce_forced_algorithm);
OBJ_RELEASE(new_enum);
if (mca_param_indices->algorithm_param_index < 0) {
return mca_param_indices->algorithm_param_index;
}
coll_tuned_allreduce_segment_size = 0;
mca_param_indices->segsize_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allreduce_algorithm_segmentsize",
"Segment size in bytes used by default for allreduce algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_allreduce_segment_size);
coll_tuned_allreduce_tree_fanout = ompi_coll_tuned_init_tree_fanout; /* get system wide default */
mca_param_indices->tree_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allreduce_algorithm_tree_fanout",
"Fanout for n-tree used for allreduce algorithms. Only has meaning if algorithm is forced and supports n-tree topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_allreduce_tree_fanout);
coll_tuned_allreduce_chain_fanout = ompi_coll_tuned_init_chain_fanout; /* get system wide default */
mca_param_indices->chain_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allreduce_algorithm_chain_fanout",
"Fanout for chains used for allreduce algorithms. Only has meaning if algorithm is forced and supports chain topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_allreduce_chain_fanout);
return (MPI_SUCCESS);
}
int ompi_coll_tuned_allreduce_intra_do_forced(void *sbuf, void *rbuf, int count,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:allreduce_intra_do_forced selected algorithm %d, segment size %d",
tuned_module->user_forced[ALLREDUCE].algorithm,
tuned_module->user_forced[ALLREDUCE].segsize));
switch (tuned_module->user_forced[ALLREDUCE].algorithm) {
case (0):
return ompi_coll_tuned_allreduce_intra_dec_fixed(sbuf, rbuf, count, dtype, op, comm, module);
case (1):
return ompi_coll_base_allreduce_intra_basic_linear(sbuf, rbuf, count, dtype, op, comm, module);
case (2):
return ompi_coll_base_allreduce_intra_nonoverlapping(sbuf, rbuf, count, dtype, op, comm, module);
case (3):
return ompi_coll_base_allreduce_intra_recursivedoubling(sbuf, rbuf, count, dtype, op, comm, module);
case (4):
return ompi_coll_base_allreduce_intra_ring(sbuf, rbuf, count, dtype, op, comm, module);
case (5):
return ompi_coll_base_allreduce_intra_ring_segmented(sbuf, rbuf, count, dtype, op, comm, module, tuned_module->user_forced[ALLREDUCE].segsize);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:allreduce_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
tuned_module->user_forced[ALLREDUCE].algorithm,
ompi_coll_tuned_forced_max_algorithms[ALLREDUCE]));
return (MPI_ERR_ARG);
}
int ompi_coll_tuned_allreduce_intra_do_this(void *sbuf, void *rbuf, int count,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm, int faninout, int segsize)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:allreduce_intra_do_this algorithm %d topo fan in/out %d segsize %d",
algorithm, faninout, segsize));
switch (algorithm) {
case (0):
return ompi_coll_tuned_allreduce_intra_dec_fixed(sbuf, rbuf, count, dtype, op, comm, module);
case (1):
return ompi_coll_base_allreduce_intra_basic_linear(sbuf, rbuf, count, dtype, op, comm, module);
case (2):
return ompi_coll_base_allreduce_intra_nonoverlapping(sbuf, rbuf, count, dtype, op, comm, module);
case (3):
return ompi_coll_base_allreduce_intra_recursivedoubling(sbuf, rbuf, count, dtype, op, comm, module);
case (4):
return ompi_coll_base_allreduce_intra_ring(sbuf, rbuf, count, dtype, op, comm, module);
case (5):
return ompi_coll_base_allreduce_intra_ring_segmented(sbuf, rbuf, count, dtype, op, comm, module, segsize);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:allreduce_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[ALLREDUCE]));
return (MPI_ERR_ARG);
}

204
ompi/mca/coll/tuned/coll_tuned_alltoall_decision.c Обычный файл
Просмотреть файл

@ -0,0 +1,204 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "ompi/constants.h"
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/pml/pml.h"
#include "coll_tuned.h"
#include "ompi/mca/coll/base/coll_base_topo.h"
#include "ompi/mca/coll/base/coll_base_util.h"
/* alltoall algorithm variables */
static int coll_tuned_alltoall_forced_algorithm = 0;
static int coll_tuned_alltoall_segment_size = 0;
static int coll_tuned_alltoall_max_requests;
static int coll_tuned_alltoall_tree_fanout;
static int coll_tuned_alltoall_chain_fanout;
/* valid values for coll_tuned_alltoall_forced_algorithm */
static mca_base_var_enum_value_t alltoall_algorithms[] = {
{0, "ignore"},
{1, "linear"},
{2, "pairwise"},
{3, "modified_bruck"},
{4, "linear_sync"},
{5, "two_proc"},
{0, NULL}
};
/* The following are used by dynamic and forced rules */
/* publish details of each algorithm and if its forced/fixed/locked in */
/* as you add methods/algorithms you must update this and the query/map routines */
/* this routine is called by the component only */
/* this makes sure that the mca parameters are set to their initial values and perms */
/* module does not call this they call the forced_getvalues routine instead */
int ompi_coll_tuned_alltoall_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
{
mca_base_var_enum_t*new_enum;
int cnt;
for( cnt = 0; NULL != alltoall_algorithms[cnt].string; cnt++ );
ompi_coll_tuned_forced_max_algorithms[ALLTOALL] = cnt;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"alltoall_algorithm_count",
"Number of alltoall algorithms available",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,
&cnt);
/* MPI_T: This variable should eventually be bound to a communicator */
coll_tuned_alltoall_forced_algorithm = 0;
(void) mca_base_var_enum_create("coll_tuned_alltoall_algorithms", alltoall_algorithms, &new_enum);
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"alltoall_algorithm",
"Which alltoall algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 pairwise, 3: modified bruck, 4: linear with sync, 5:two proc only.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_alltoall_forced_algorithm);
OBJ_RELEASE(new_enum);
if (mca_param_indices->algorithm_param_index < 0) {
return mca_param_indices->algorithm_param_index;
}
coll_tuned_alltoall_segment_size = 0;
mca_param_indices->segsize_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"alltoall_algorithm_segmentsize",
"Segment size in bytes used by default for alltoall algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_alltoall_segment_size);
coll_tuned_alltoall_tree_fanout = ompi_coll_tuned_init_tree_fanout; /* get system wide default */
mca_param_indices->tree_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"alltoall_algorithm_tree_fanout",
"Fanout for n-tree used for alltoall algorithms. Only has meaning if algorithm is forced and supports n-tree topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_alltoall_tree_fanout);
coll_tuned_alltoall_chain_fanout = ompi_coll_tuned_init_chain_fanout; /* get system wide default */
mca_param_indices->chain_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"alltoall_algorithm_chain_fanout",
"Fanout for chains used for alltoall algorithms. Only has meaning if algorithm is forced and supports chain topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_alltoall_chain_fanout);
coll_tuned_alltoall_max_requests = 0; /* no limit for alltoall by default */
mca_param_indices->max_requests_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"alltoall_algorithm_max_requests",
"Maximum number of outstanding send or recv requests. Only has meaning for synchronized algorithms.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_alltoall_max_requests);
if (mca_param_indices->max_requests_param_index < 0) {
return mca_param_indices->max_requests_param_index;
}
if (coll_tuned_alltoall_max_requests < 0) {
if( 0 == ompi_comm_rank( MPI_COMM_WORLD ) ) {
opal_output( 0, "Maximum outstanding requests must be positive number greater than 1. Switching to system level default %d \n",
ompi_coll_tuned_init_max_requests );
}
coll_tuned_alltoall_max_requests = 0;
}
return (MPI_SUCCESS);
}
int ompi_coll_tuned_alltoall_intra_do_forced(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:alltoall_intra_do_forced selected algorithm %d",
tuned_module->user_forced[ALLTOALL].algorithm));
switch (tuned_module->user_forced[ALLTOALL].algorithm) {
case (0):
return ompi_coll_tuned_alltoall_intra_dec_fixed (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (1):
return ompi_coll_base_alltoall_intra_basic_linear (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (2):
return ompi_coll_base_alltoall_intra_pairwise (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (3):
return ompi_coll_base_alltoall_intra_bruck (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (4):
return ompi_coll_base_alltoall_intra_linear_sync (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module,
tuned_module->user_forced[ALLTOALL].max_requests);
case (5):
return ompi_coll_base_alltoall_intra_two_procs (sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:alltoall_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
tuned_module->user_forced[ALLTOALL].algorithm, ompi_coll_tuned_forced_max_algorithms[ALLTOALL]));
return (MPI_ERR_ARG);
}
int ompi_coll_tuned_alltoall_intra_do_this(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm, int faninout, int segsize,
int max_requests)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:alltoall_intra_do_this selected algorithm %d topo faninout %d segsize %d",
algorithm, faninout, segsize));
switch (algorithm) {
case (0):
return ompi_coll_tuned_alltoall_intra_dec_fixed(sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (1):
return ompi_coll_base_alltoall_intra_basic_linear(sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (2):
return ompi_coll_base_alltoall_intra_pairwise(sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (3):
return ompi_coll_base_alltoall_intra_bruck(sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
case (4):
return ompi_coll_base_alltoall_intra_linear_sync(sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module, max_requests);
case (5):
return ompi_coll_base_alltoall_intra_two_procs(sbuf, scount, sdtype, rbuf, rcount, rdtype, comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:alltoall_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[ALLTOALL]));
return (MPI_ERR_ARG);
}

156
ompi/mca/coll/tuned/coll_tuned_alltoallv_decision.c Обычный файл
Просмотреть файл

@ -0,0 +1,156 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "ompi/constants.h"
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/pml/pml.h"
#include "coll_tuned.h"
#include "ompi/mca/coll/base/coll_base_topo.h"
#include "ompi/mca/coll/base/coll_base_util.h"
/* alltoallv algorithm variables */
static int coll_tuned_alltoallv_algorithm_count = 2;
static int coll_tuned_alltoallv_forced_algorithm = 0;
/* valid values for coll_tuned_alltoallv_forced_algorithm */
static mca_base_var_enum_value_t alltoallv_algorithms[] = {
{0, "ignore"},
{1, "basic_linear"},
{2, "pairwise"},
{0, NULL}
};
/*
* The following are used by dynamic and forced rules. Publish
* details of each algorithm and if its forced/fixed/locked in as you add
* methods/algorithms you must update this and the query/map routines.
* This routine is called by the component only. This makes sure that
* the mca parameters are set to their initial values and perms.
* Module does not call this. They call the forced_getvalues routine
* instead.
*/
int ompi_coll_tuned_alltoallv_intra_check_forced_init(coll_tuned_force_algorithm_mca_param_indices_t
*mca_param_indices)
{
mca_base_var_enum_t *new_enum;
int cnt;
for( cnt = 0; NULL != alltoallv_algorithms[cnt].string; cnt++ );
ompi_coll_tuned_forced_max_algorithms[ALLTOALLV] = cnt;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"alltoallv_algorithm_count",
"Number of alltoallv algorithms available",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,
&cnt);
/* MPI_T: This variable should eventually be bound to a communicator */
coll_tuned_alltoallv_forced_algorithm = 0;
(void) mca_base_var_enum_create("coll_tuned_alltoallv_algorithms", alltoallv_algorithms, &new_enum);
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"alltoallv_algorithm",
"Which alltoallv algorithm is used. "
"Can be locked down to choice of: 0 ignore, "
"1 basic linear, 2 pairwise.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_alltoallv_forced_algorithm);
OBJ_RELEASE(new_enum);
if (mca_param_indices->algorithm_param_index < 0) {
return mca_param_indices->algorithm_param_index;
}
return (MPI_SUCCESS);
}
int ompi_coll_tuned_alltoallv_intra_do_forced(void *sbuf, int *scounts, int *sdisps,
struct ompi_datatype_t *sdtype,
void* rbuf, int *rcounts, int *rdisps,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:alltoallv_intra_do_forced selected algorithm %d",
tuned_module->user_forced[ALLTOALLV].algorithm));
switch (tuned_module->user_forced[ALLTOALLV].algorithm) {
case (0):
return ompi_coll_tuned_alltoallv_intra_dec_fixed(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module);
case (1):
return ompi_coll_base_alltoallv_intra_basic_linear(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module);
case (2):
return ompi_coll_base_alltoallv_intra_pairwise(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:alltoallv_intra_do_forced attempt to "
"select algorithm %d when only 0-%d is valid.",
tuned_module->user_forced[ALLTOALLV].algorithm,
ompi_coll_tuned_forced_max_algorithms[ALLTOALLV]));
return (MPI_ERR_ARG);
}
/* If the user selects dynamic rules and specifies the algorithm to
* use, then this function is called. */
int ompi_coll_tuned_alltoallv_intra_do_this(void *sbuf, int *scounts, int *sdisps,
struct ompi_datatype_t *sdtype,
void* rbuf, int *rcounts, int *rdisps,
struct ompi_datatype_t *rdtype,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:alltoallv_intra_do_this selected algorithm %d ",
algorithm));
switch (algorithm) {
case (0):
return ompi_coll_tuned_alltoallv_intra_dec_fixed(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module);
case (1):
return ompi_coll_base_alltoallv_intra_basic_linear(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module);
case (2):
return ompi_coll_base_alltoallv_intra_pairwise(sbuf, scounts, sdisps, sdtype,
rbuf, rcounts, rdisps, rdtype,
comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:alltoall_intra_do_this attempt to select "
"algorithm %d when only 0-%d is valid.",
algorithm, ompi_coll_tuned_forced_max_algorithms[ALLTOALLV]));
return (MPI_ERR_ARG);
}

135
ompi/mca/coll/tuned/coll_tuned_barrier_decision.c Обычный файл
Просмотреть файл

@ -0,0 +1,135 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "opal/util/bit_ops.h"
#include "ompi/constants.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/pml/pml.h"
#include "coll_tuned.h"
#include "ompi/mca/coll/base/coll_base_topo.h"
#include "ompi/mca/coll/base/coll_base_util.h"
/* barrier algorithm variables */
static int coll_tuned_barrier_forced_algorithm = 0;
/* valid values for coll_tuned_barrier_forced_algorithm */
static mca_base_var_enum_value_t barrier_algorithms[] = {
{0, "ignore"},
{1, "linear"},
{2, "double_ring"},
{3, "recursive_doubling"},
{4, "bruck"},
{5, "two_proc"},
{6, "tree"},
{0, NULL}
};
/* The following are used by dynamic and forced rules */
/* publish details of each algorithm and if its forced/fixed/locked in */
/* as you add methods/algorithms you must update this and the query/map */
/* routines */
/* this routine is called by the component only */
/* this makes sure that the mca parameters are set to their initial values */
/* and perms */
/* module does not call this they call the forced_getvalues routine instead */
int ompi_coll_tuned_barrier_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
{
mca_base_var_enum_t *new_enum;
int cnt;
for( cnt = 0; NULL != barrier_algorithms[cnt].string; cnt++ );
ompi_coll_tuned_forced_max_algorithms[BARRIER] = cnt;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"barrier_algorithm_count",
"Number of barrier algorithms available",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,
&cnt);
/* MPI_T: This variable should eventually be bound to a communicator */
coll_tuned_barrier_forced_algorithm = 0;
(void) mca_base_var_enum_create("coll_tuned_barrier_algorithms", barrier_algorithms, &new_enum);
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"barrier_algorithm",
"Which barrier algorithm is used. Can be locked down to choice of: 0 ignore, 1 linear, 2 double ring, 3: recursive doubling 4: bruck, 5: two proc only, 6: tree",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_barrier_forced_algorithm);
OBJ_RELEASE(new_enum);
if (mca_param_indices->algorithm_param_index < 0) {
return mca_param_indices->algorithm_param_index;
}
return (MPI_SUCCESS);
}
int ompi_coll_tuned_barrier_intra_do_forced(struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:barrier_intra_do_forced selected algorithm %d",
tuned_module->user_forced[BARRIER].algorithm));
switch (tuned_module->user_forced[BARRIER].algorithm) {
case (0): return ompi_coll_tuned_barrier_intra_dec_fixed(comm, module);
case (1): return ompi_coll_base_barrier_intra_linear(comm, module);
case (2): return ompi_coll_base_barrier_intra_doublering(comm, module);
case (3): return ompi_coll_base_barrier_intra_recursivedoubling(comm, module);
case (4): return ompi_coll_base_barrier_intra_bruck(comm, module);
case (5): return ompi_coll_base_barrier_intra_two_procs(comm, module);
case (6): return ompi_coll_base_barrier_intra_tree(comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:barrier_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
tuned_module->user_forced[BARRIER].algorithm,
ompi_coll_tuned_forced_max_algorithms[BARRIER]));
return (MPI_ERR_ARG);
}
int ompi_coll_tuned_barrier_intra_do_this (struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm, int faninout, int segsize)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:barrier_intra_do_this selected algorithm %d topo fanin/out%d",
algorithm, faninout));
switch (algorithm) {
case (0): return ompi_coll_tuned_barrier_intra_dec_fixed(comm, module);
case (1): return ompi_coll_base_barrier_intra_linear(comm, module);
case (2): return ompi_coll_base_barrier_intra_doublering(comm, module);
case (3): return ompi_coll_base_barrier_intra_recursivedoubling(comm, module);
case (4): return ompi_coll_base_barrier_intra_bruck(comm, module);
case (5): return ompi_coll_base_barrier_intra_two_procs(comm, module);
case (6): return ompi_coll_base_barrier_intra_tree(comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:barrier_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[BARRIER]));
return (MPI_ERR_ARG);
}

183
ompi/mca/coll/tuned/coll_tuned_bcast_decision.c Обычный файл
Просмотреть файл

@ -0,0 +1,183 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "ompi/constants.h"
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/pml/pml.h"
#include "coll_tuned.h"
#include "ompi/mca/coll/base/coll_base_topo.h"
#include "ompi/mca/coll/base/coll_base_util.h"
/* bcast algorithm variables */
static int coll_tuned_bcast_algorithm_count = 6;
static int coll_tuned_bcast_forced_algorithm = 0;
static int coll_tuned_bcast_segment_size = 0;
static int coll_tuned_bcast_tree_fanout;
static int coll_tuned_bcast_chain_fanout;
/* valid values for coll_tuned_bcast_forced_algorithm */
static mca_base_var_enum_value_t bcast_algorithms[] = {
{0, "ignore"},
{1, "basic_linear"},
{2, "chain"},
{3, "pipeline"},
{4, "split_binary_tree"},
{5, "binary_tree"},
{6, "binomial"},
{0, NULL}
};
/* The following are used by dynamic and forced rules */
/* publish details of each algorithm and if its forced/fixed/locked in */
/* as you add methods/algorithms you must update this and the query/map routines */
/* this routine is called by the component only */
/* this makes sure that the mca parameters are set to their initial values and perms */
/* module does not call this they call the forced_getvalues routine instead */
int ompi_coll_tuned_bcast_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
{
mca_base_var_enum_t *new_enum;
int cnt;
for( cnt = 0; NULL != bcast_algorithms[cnt].string; cnt++ );
ompi_coll_tuned_forced_max_algorithms[BCAST] = cnt;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"bcast_algorithm_count",
"Number of bcast algorithms available",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,
&cnt);
/* MPI_T: This variable should eventually be bound to a communicator */
coll_tuned_bcast_forced_algorithm = 0;
(void) mca_base_var_enum_create("coll_tuned_bcast_algorithms", bcast_algorithms, &new_enum);
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"bcast_algorithm",
"Which bcast algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 chain, 3: pipeline, 4: split binary tree, 5: binary tree, 6: binomial tree.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_bcast_forced_algorithm);
OBJ_RELEASE(new_enum);
if (mca_param_indices->algorithm_param_index < 0) {
return mca_param_indices->algorithm_param_index;
}
coll_tuned_bcast_segment_size = 0;
mca_param_indices->segsize_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"bcast_algorithm_segmentsize",
"Segment size in bytes used by default for bcast algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_bcast_segment_size);
coll_tuned_bcast_tree_fanout = ompi_coll_tuned_init_tree_fanout; /* get system wide default */
mca_param_indices->tree_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"bcast_algorithm_tree_fanout",
"Fanout for n-tree used for bcast algorithms. Only has meaning if algorithm is forced and supports n-tree topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_bcast_tree_fanout);
coll_tuned_bcast_chain_fanout = ompi_coll_tuned_init_chain_fanout; /* get system wide default */
mca_param_indices->chain_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"bcast_algorithm_chain_fanout",
"Fanout for chains used for bcast algorithms. Only has meaning if algorithm is forced and supports chain topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_bcast_chain_fanout);
return (MPI_SUCCESS);
}
int ompi_coll_tuned_bcast_intra_do_forced(void *buf, int count,
struct ompi_datatype_t *dtype,
int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_do_forced algorithm %d",
tuned_module->user_forced[BCAST].algorithm));
switch (tuned_module->user_forced[BCAST].algorithm) {
case (0): return ompi_coll_tuned_bcast_intra_dec_fixed( buf, count, dtype, root, comm, module );
case (1): return ompi_coll_base_bcast_intra_basic_linear( buf, count, dtype, root, comm, module );
case (2): return ompi_coll_base_bcast_intra_chain( buf, count, dtype, root, comm, module,
tuned_module->user_forced[BCAST].segsize,
tuned_module->user_forced[BCAST].chain_fanout );
case (3): return ompi_coll_base_bcast_intra_pipeline( buf, count, dtype, root, comm, module,
tuned_module->user_forced[BCAST].segsize );
case (4): return ompi_coll_base_bcast_intra_split_bintree( buf, count, dtype, root, comm, module,
tuned_module->user_forced[BCAST].segsize );
case (5): return ompi_coll_base_bcast_intra_bintree( buf, count, dtype, root, comm, module,
tuned_module->user_forced[BCAST].segsize );
case (6): return ompi_coll_base_bcast_intra_binomial( buf, count, dtype, root, comm, module,
tuned_module->user_forced[BCAST].segsize );
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
tuned_module->user_forced[BCAST].algorithm, ompi_coll_tuned_forced_max_algorithms[BCAST]));
return (MPI_ERR_ARG);
}
int ompi_coll_tuned_bcast_intra_do_this(void *buf, int count,
struct ompi_datatype_t *dtype,
int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm, int faninout, int segsize)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_do_this algorithm %d topo faninout %d segsize %d",
algorithm, faninout, segsize));
switch (algorithm) {
case (0):
return ompi_coll_tuned_bcast_intra_dec_fixed( buf, count, dtype, root, comm, module );
case (1):
return ompi_coll_base_bcast_intra_basic_linear( buf, count, dtype, root, comm, module );
case (2):
return ompi_coll_base_bcast_intra_chain( buf, count, dtype, root, comm, module, segsize, faninout );
case (3):
return ompi_coll_base_bcast_intra_pipeline( buf, count, dtype, root, comm, module, segsize );
case (4):
return ompi_coll_base_bcast_intra_split_bintree( buf, count, dtype, root, comm, module, segsize );
case (5):
return ompi_coll_base_bcast_intra_bintree( buf, count, dtype, root, comm, module, segsize );
case (6):
return ompi_coll_base_bcast_intra_binomial( buf, count, dtype, root, comm, module, segsize );
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:bcast_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[BCAST]));
return (MPI_ERR_ARG);
}

198
ompi/mca/coll/tuned/coll_tuned_gather_decision.c Обычный файл
Просмотреть файл

@ -0,0 +1,198 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "ompi/constants.h"
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/pml/pml.h"
#include "coll_tuned.h"
#include "ompi/mca/coll/base/coll_base_topo.h"
#include "ompi/mca/coll/base/coll_base_util.h"
/* gather algorithm variables */
static int coll_tuned_gather_forced_algorithm = 0;
static int coll_tuned_gather_segment_size = 0;
static int coll_tuned_gather_tree_fanout;
static int coll_tuned_gather_chain_fanout;
/* valid values for coll_tuned_gather_forced_algorithm */
static mca_base_var_enum_value_t gather_algorithms[] = {
{0, "ignore"},
{1, "basic_linear"},
{2, "binomial"},
{3, "linear_sync"},
{0, NULL}
};
/* The following are used by dynamic and forced rules */
/* publish details of each algorithm and if its forced/fixed/locked in */
/* as you add methods/algorithms you must update this and the query/map
routines */
/* this routine is called by the component only */
/* this makes sure that the mca parameters are set to their initial values
and perms */
/* module does not call this they call the forced_getvalues routine instead */
int
ompi_coll_tuned_gather_intra_check_forced_init(coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
{
mca_base_var_enum_t *new_enum;
int cnt;
for( cnt = 0; NULL != gather_algorithms[cnt].string; cnt++ );
ompi_coll_tuned_forced_max_algorithms[GATHER] = cnt;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"gather_algorithm_count",
"Number of gather algorithms available",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,
&cnt);
/* MPI_T: This variable should eventually be bound to a communicator */
coll_tuned_gather_forced_algorithm = 0;
(void) mca_base_var_enum_create("coll_tuned_gather_algorithms", gather_algorithms, &new_enum);
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"gather_algorithm",
"Which gather algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 binomial, 3 linear with synchronization.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_gather_forced_algorithm);
OBJ_RELEASE(new_enum);
if (mca_param_indices->algorithm_param_index < 0) {
return mca_param_indices->algorithm_param_index;
}
coll_tuned_gather_segment_size = 0;
mca_param_indices->segsize_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"gather_algorithm_segmentsize",
"Segment size in bytes used by default for gather algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation. Currently, available algorithms do not support segmentation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_gather_segment_size);
coll_tuned_gather_tree_fanout = ompi_coll_tuned_init_tree_fanout; /* get system wide default */
mca_param_indices->tree_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"gather_algorithm_tree_fanout",
"Fanout for n-tree used for gather algorithms. Only has meaning if algorithm is forced and supports n-tree topo based operation. Currently, available algorithms do not support n-tree topologies.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_gather_tree_fanout);
coll_tuned_gather_chain_fanout = ompi_coll_tuned_init_chain_fanout; /* get system wide default */
mca_param_indices->chain_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"gather_algorithm_chain_fanout",
"Fanout for chains used for gather algorithms. Only has meaning if algorithm is forced and supports chain topo based operation. Currently, available algorithms do not support chain topologies.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_gather_chain_fanout);
return (MPI_SUCCESS);
}
int
ompi_coll_tuned_gather_intra_do_forced(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:gather_intra_do_forced selected algorithm %d",
tuned_module->user_forced[GATHER].algorithm));
switch (tuned_module->user_forced[GATHER].algorithm) {
case (0):
return ompi_coll_tuned_gather_intra_dec_fixed(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
case (1):
return ompi_coll_base_gather_intra_basic_linear(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
case (2):
return ompi_coll_base_gather_intra_binomial(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
case (3):
return ompi_coll_base_gather_intra_linear_sync(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module,
tuned_module->user_forced[GATHER].segsize);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:gather_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
tuned_module->user_forced[GATHER].algorithm,
ompi_coll_tuned_forced_max_algorithms[GATHER]));
return (MPI_ERR_ARG);
}
int
ompi_coll_tuned_gather_intra_do_this(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm, int faninout, int segsize)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:gather_intra_do_this selected algorithm %d topo faninout %d segsize %d",
algorithm, faninout, segsize));
switch (algorithm) {
case (0):
return ompi_coll_tuned_gather_intra_dec_fixed(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
case (1):
return ompi_coll_base_gather_intra_basic_linear(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
case (2):
return ompi_coll_base_gather_intra_binomial(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
case (3):
return ompi_coll_base_gather_intra_linear_sync(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module,
segsize);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:gather_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[GATHER]));
return (MPI_ERR_ARG);
}

222
ompi/mca/coll/tuned/coll_tuned_reduce_decision.c Обычный файл
Просмотреть файл

@ -0,0 +1,222 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "ompi/constants.h"
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_base_topo.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/pml/pml.h"
#include "ompi/op/op.h"
#include "coll_tuned.h"
/* reduce algorithm variables */
static int coll_tuned_reduce_forced_algorithm = 0;
static int coll_tuned_reduce_segment_size = 0;
static int coll_tuned_reduce_max_requests;
static int coll_tuned_reduce_tree_fanout;
static int coll_tuned_reduce_chain_fanout;
/* valid values for coll_tuned_reduce_forced_algorithm */
static mca_base_var_enum_value_t reduce_algorithms[] = {
{0, "ignore"},
{1, "linear"},
{2, "chain"},
{3, "pipeline"},
{4, "binary"},
{5, "binomial"},
{6, "in-order_binary"},
{0, NULL}
};
/**
* The following are used by dynamic and forced rules
*
* publish details of each algorithm and if its forced/fixed/locked in
* as you add methods/algorithms you must update this and the query/map routines
*
* this routine is called by the component only
* this makes sure that the mca parameters are set to their initial values and
* perms module does not call this they call the forced_getvalues routine
* instead.
*/
int ompi_coll_tuned_reduce_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
{
mca_base_var_enum_t*new_enum;
int cnt;
for( cnt = 0; NULL != reduce_algorithms[cnt].string; cnt++ );
ompi_coll_tuned_forced_max_algorithms[REDUCE] = cnt;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_algorithm_count",
"Number of reduce algorithms available",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,
&cnt);
/* MPI_T: This variable should eventually be bound to a communicator */
coll_tuned_reduce_forced_algorithm = 0;
(void) mca_base_var_enum_create("coll_tuned_reduce_algorithms", reduce_algorithms, &new_enum);
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_algorithm",
"Which reduce algorithm is used. Can be locked down to choice of: 0 ignore, 1 linear, 2 chain, 3 pipeline, 4 binary, 5 binomial, 6 in-order binary",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_forced_algorithm);
OBJ_RELEASE(new_enum);
if (mca_param_indices->algorithm_param_index < 0) {
return mca_param_indices->algorithm_param_index;
}
coll_tuned_reduce_segment_size = 0;
mca_param_indices->segsize_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_algorithm_segmentsize",
"Segment size in bytes used by default for reduce algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_segment_size);
coll_tuned_reduce_tree_fanout = ompi_coll_tuned_init_tree_fanout; /* get system wide default */
mca_param_indices->tree_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_algorithm_tree_fanout",
"Fanout for n-tree used for reduce algorithms. Only has meaning if algorithm is forced and supports n-tree topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_tree_fanout);
coll_tuned_reduce_chain_fanout = ompi_coll_tuned_init_chain_fanout; /* get system wide default */
mca_param_indices->chain_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_algorithm_chain_fanout",
"Fanout for chains used for reduce algorithms. Only has meaning if algorithm is forced and supports chain topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_chain_fanout);
coll_tuned_reduce_max_requests = 0; /* no limit for reduce by default */
mca_param_indices->max_requests_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_algorithm_max_requests",
"Maximum number of outstanding send requests on leaf nodes. 0 means no limit.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_max_requests);
if (mca_param_indices->max_requests_param_index < 0) {
return mca_param_indices->max_requests_param_index;
}
if (coll_tuned_reduce_max_requests < 0) {
if( 0 == ompi_comm_rank( MPI_COMM_WORLD ) ) {
opal_output( 0, "Maximum outstanding requests must be positive number or 0. Initializing to 0 (no limit).\n" );
}
coll_tuned_reduce_max_requests = 0;
}
return (MPI_SUCCESS);
}
int ompi_coll_tuned_reduce_intra_do_forced(void *sbuf, void* rbuf, int count,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op, int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
const int segsize = tuned_module->user_forced[REDUCE].segsize;
const int chain_fanout = tuned_module->user_forced[REDUCE].chain_fanout;
const int max_requests = tuned_module->user_forced[REDUCE].max_requests;
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_do_forced selected algorithm %d",
tuned_module->user_forced[REDUCE].algorithm));
switch (tuned_module->user_forced[REDUCE].algorithm) {
case (0): return ompi_coll_tuned_reduce_intra_dec_fixed(sbuf, rbuf, count, dtype,
op, root, comm, module);
case (1): return ompi_coll_base_reduce_intra_basic_linear(sbuf, rbuf, count, dtype,
op, root, comm, module);
case (2): return ompi_coll_base_reduce_intra_chain(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, chain_fanout, max_requests);
case (3): return ompi_coll_base_reduce_intra_pipeline(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (4): return ompi_coll_base_reduce_intra_binary(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (5): return ompi_coll_base_reduce_intra_binomial(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (6): return ompi_coll_base_reduce_intra_in_order_binary(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
tuned_module->user_forced[REDUCE].algorithm, ompi_coll_tuned_forced_max_algorithms[REDUCE]));
return (MPI_ERR_ARG);
}
int ompi_coll_tuned_reduce_intra_do_this(void *sbuf, void* rbuf, int count,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op, int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm, int faninout,
int segsize, int max_requests )
{
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_do_this selected algorithm %d topo faninout %d segsize %d",
algorithm, faninout, segsize));
switch (algorithm) {
case (0): return ompi_coll_tuned_reduce_intra_dec_fixed(sbuf, rbuf, count, dtype,
op, root, comm, module);
case (1): return ompi_coll_base_reduce_intra_basic_linear(sbuf, rbuf, count, dtype,
op, root, comm, module);
case (2): return ompi_coll_base_reduce_intra_chain(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, faninout, max_requests);
case (3): return ompi_coll_base_reduce_intra_pipeline(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (4): return ompi_coll_base_reduce_intra_binary(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (5): return ompi_coll_base_reduce_intra_binomial(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (6): return ompi_coll_base_reduce_intra_in_order_binary(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[REDUCE]));
return (MPI_ERR_ARG);
}

173
ompi/mca/coll/tuned/coll_tuned_reduce_scatter_decision.c Обычный файл
Просмотреть файл

@ -0,0 +1,173 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "opal/util/bit_ops.h"
#include "ompi/constants.h"
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_base_topo.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/pml/pml.h"
#include "ompi/op/op.h"
#include "coll_tuned.h"
/* reduce_scatter algorithm variables */
static int coll_tuned_reduce_scatter_forced_algorithm = 0;
static int coll_tuned_reduce_scatter_segment_size = 0;
static int coll_tuned_reduce_scatter_tree_fanout;
static int coll_tuned_reduce_scatter_chain_fanout;
/* valid values for coll_tuned_reduce_scatter_forced_algorithm */
static mca_base_var_enum_value_t reduce_scatter_algorithms[] = {
{0, "ignore"},
{1, "non-overlapping"},
{2, "recursive_halfing"},
{3, "ring"},
{0, NULL}
};
/**
* The following are used by dynamic and forced rules
*
* publish details of each algorithm and if its forced/fixed/locked in
* as you add methods/algorithms you must update this and the query/map routines
*
* this routine is called by the component only
* this makes sure that the mca parameters are set to their initial values and
* perms module does not call this they call the forced_getvalues routine
* instead
*/
int ompi_coll_tuned_reduce_scatter_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
{
mca_base_var_enum_t *new_enum;
int cnt;
for( cnt = 0; NULL != reduce_scatter_algorithms[cnt].string; cnt++ );
ompi_coll_tuned_forced_max_algorithms[REDUCESCATTER] = cnt;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_scatter_algorithm_count",
"Number of reduce_scatter algorithms available",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,
&cnt);
/* MPI_T: This variable should eventually be bound to a communicator */
coll_tuned_reduce_scatter_forced_algorithm = 0;
(void) mca_base_var_enum_create("coll_tuned_reduce_scatter_algorithms", reduce_scatter_algorithms, &new_enum);
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_scatter_algorithm",
"Which reduce reduce_scatter algorithm is used. Can be locked down to choice of: 0 ignore, 1 non-overlapping (Reduce + Scatterv), 2 recursive halving, 3 ring",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_scatter_forced_algorithm);
OBJ_RELEASE(new_enum);
if (mca_param_indices->algorithm_param_index < 0) {
return mca_param_indices->algorithm_param_index;
}
coll_tuned_reduce_scatter_segment_size = 0;
mca_param_indices->segsize_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_scatter_algorithm_segmentsize",
"Segment size in bytes used by default for reduce_scatter algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_scatter_segment_size);
coll_tuned_reduce_scatter_tree_fanout = ompi_coll_tuned_init_tree_fanout; /* get system wide default */
mca_param_indices->tree_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_scatter_algorithm_tree_fanout",
"Fanout for n-tree used for reduce_scatter algorithms. Only has meaning if algorithm is forced and supports n-tree topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_scatter_tree_fanout);
coll_tuned_reduce_scatter_chain_fanout = ompi_coll_tuned_init_chain_fanout; /* get system wide default */
mca_param_indices->chain_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_scatter_algorithm_chain_fanout",
"Fanout for chains used for reduce_scatter algorithms. Only has meaning if algorithm is forced and supports chain topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_scatter_chain_fanout);
return (MPI_SUCCESS);
}
int ompi_coll_tuned_reduce_scatter_intra_do_forced(void *sbuf, void* rbuf,
int *rcounts,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_scatter_intra_do_forced selected algorithm %d",
tuned_module->user_forced[REDUCESCATTER].algorithm));
switch (tuned_module->user_forced[REDUCESCATTER].algorithm) {
case (0): return ompi_coll_tuned_reduce_scatter_intra_dec_fixed(sbuf, rbuf, rcounts,
dtype, op, comm, module);
case (1): return ompi_coll_base_reduce_scatter_intra_nonoverlapping(sbuf, rbuf, rcounts,
dtype, op, comm, module);
case (2): return ompi_coll_base_reduce_scatter_intra_basic_recursivehalving(sbuf, rbuf, rcounts,
dtype, op, comm, module);
case (3): return ompi_coll_base_reduce_scatter_intra_ring(sbuf, rbuf, rcounts,
dtype, op, comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_scatter_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
tuned_module->user_forced[REDUCESCATTER].algorithm, ompi_coll_tuned_forced_max_algorithms[REDUCESCATTER]));
return (MPI_ERR_ARG);
}
int ompi_coll_tuned_reduce_scatter_intra_do_this(void *sbuf, void* rbuf,
int *rcounts,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm, int faninout, int segsize)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_scatter_intra_do_this selected algorithm %d topo faninout %d segsize %d",
algorithm, faninout, segsize));
switch (algorithm) {
case (0): return ompi_coll_tuned_reduce_scatter_intra_dec_fixed(sbuf, rbuf, rcounts,
dtype, op, comm, module);
case (1): return ompi_coll_base_reduce_scatter_intra_nonoverlapping(sbuf, rbuf, rcounts,
dtype, op, comm, module);
case (2): return ompi_coll_base_reduce_scatter_intra_basic_recursivehalving(sbuf, rbuf, rcounts,
dtype, op, comm, module);
case (3): return ompi_coll_base_reduce_scatter_intra_ring(sbuf, rbuf, rcounts,
dtype, op, comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_scatter_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[REDUCESCATTER]));
return (MPI_ERR_ARG);
}

185
ompi/mca/coll/tuned/coll_tuned_scatter_decision.c Обычный файл
Просмотреть файл

@ -0,0 +1,185 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "ompi/constants.h"
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/coll/base/coll_base_topo.h"
#include "ompi/mca/coll/base/coll_base_util.h"
#include "ompi/mca/pml/pml.h"
#include "coll_tuned.h"
/* scatter algorithm variables */
static int coll_tuned_scatter_forced_algorithm = 0;
static int coll_tuned_scatter_segment_size = 0;
static int coll_tuned_scatter_tree_fanout;
static int coll_tuned_scatter_chain_fanout;
/* valid values for coll_tuned_scatter_forced_algorithm */
static mca_base_var_enum_value_t scatter_algorithms[] = {
{0, "ignore"},
{1, "basic_linear"},
{2, "binomial"},
{0, NULL}
};
/* The following are used by dynamic and forced rules */
/* publish details of each algorithm and if its forced/fixed/locked in */
/* as you add methods/algorithms you must update this and the query/map
routines */
/* this routine is called by the component only */
/* this makes sure that the mca parameters are set to their initial values
and perms */
/* module does not call this they call the forced_getvalues routine instead */
int
ompi_coll_tuned_scatter_intra_check_forced_init(coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
{
mca_base_var_enum_t *new_enum;
int cnt;
for( cnt = 0; NULL != scatter_algorithms[cnt].string; cnt++ );
ompi_coll_tuned_forced_max_algorithms[SCATTER] = cnt;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"scatter_algorithm_count",
"Number of scatter algorithms available",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,
&cnt);
/* MPI_T: This variable should eventually be bound to a communicator */
coll_tuned_scatter_forced_algorithm = 0;
(void) mca_base_var_enum_create("coll_tuned_scatter_algorithms", scatter_algorithms, &new_enum);
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"scatter_algorithm",
"Which scatter algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 binomial.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_scatter_forced_algorithm);
OBJ_RELEASE(new_enum);
if (mca_param_indices->algorithm_param_index < 0) {
return mca_param_indices->algorithm_param_index;
}
coll_tuned_scatter_segment_size = 0;
mca_param_indices->segsize_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"scatter_algorithm_segmentsize",
"Segment size in bytes used by default for scatter algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation. Currently, available algorithms do not support segmentation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_scatter_segment_size);
coll_tuned_scatter_tree_fanout = ompi_coll_tuned_init_tree_fanout; /* get system wide default */
mca_param_indices->tree_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"scatter_algorithm_tree_fanout",
"Fanout for n-tree used for scatter algorithms. Only has meaning if algorithm is forced and supports n-tree topo based operation. Currently, available algorithms do not support n-tree topologies.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_scatter_tree_fanout);
coll_tuned_scatter_chain_fanout = ompi_coll_tuned_init_chain_fanout; /* get system wide default */
mca_param_indices->chain_fanout_param_index=
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"scatter_algorithm_chain_fanout",
"Fanout for chains used for scatter algorithms. Only has meaning if algorithm is forced and supports chain topo based operation. Currently, available algorithms do not support chain topologies.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_scatter_chain_fanout);
return (MPI_SUCCESS);
}
int
ompi_coll_tuned_scatter_intra_do_forced(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:scatter_intra_do_forced selected algorithm %d",
tuned_module->user_forced[SCATTER].algorithm));
switch (tuned_module->user_forced[SCATTER].algorithm) {
case (0):
return ompi_coll_tuned_scatter_intra_dec_fixed(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
case (1):
return ompi_coll_base_scatter_intra_basic_linear(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
case (2):
return ompi_coll_base_scatter_intra_binomial(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:scatter_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
tuned_module->user_forced[SCATTER].algorithm, ompi_coll_tuned_forced_max_algorithms[SCATTER]));
return MPI_ERR_ARG;
}
int
ompi_coll_tuned_scatter_intra_do_this(void *sbuf, int scount,
struct ompi_datatype_t *sdtype,
void* rbuf, int rcount,
struct ompi_datatype_t *rdtype,
int root,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module,
int algorithm, int faninout, int segsize)
{
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:scatter_intra_do_this selected algorithm %d topo faninout %d segsize %d",
algorithm, faninout, segsize));
switch (algorithm) {
case (0):
return ompi_coll_tuned_scatter_intra_dec_fixed(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
case (1):
return ompi_coll_base_scatter_intra_basic_linear(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
case (2):
return ompi_coll_base_scatter_intra_binomial(sbuf, scount, sdtype,
rbuf, rcount, rdtype,
root, comm, module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:scatter_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[SCATTER]));
return MPI_ERR_ARG;
}