/* * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana * University Research and Technology * Corporation. All rights reserved. * Copyright (c) 2004-2005 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. * $COPYRIGHT$ * * Additional copyrights may follow * * $HEADER$ */ #include "ompi_config.h" #include "coll_hierarch.h" #include #include "mpi.h" #include "ompi/communicator/communicator.h" #include "ompi/group/group.h" #include "ompi/proc/proc.h" #include "ompi/op/op.h" #include "ompi/mca/coll/coll.h" #include "ompi/mca/coll/base/base.h" #include "ompi/mca/coll/base/coll_tags.h" #include "ompi/class/ompi_bitmap.h" #include "ompi/mca/bml/bml.h" #include "ompi/mca/bml/base/base.h" #include "ompi/mca/pml/pml.h" #include "ompi/mca/btl/btl.h" /* Local functions and data */ #define HIER_MAXPROTOCOL 7 static int mca_coll_hierarch_max_protocol=HIER_MAXPROTOCOL; static char hier_prot[HIER_MAXPROTOCOL][7]={"0","tcp","gm","mx","mvapi","openib","sm"}; static void mca_coll_hierarch_checkfor_component (struct ompi_communicator_t *comm, int component_level, char *component_name, int *key, int *ncount); static void mca_coll_hierarch_dump_struct ( struct mca_coll_base_comm_t *c); /* * Linear set of collective algorithms */ static const mca_coll_base_module_1_0_0_t intra = { /* Initialization / finalization functions */ mca_coll_hierarch_module_init, mca_coll_hierarch_module_finalize, /* Collective function pointers */ /* function pointers marked with NULL are not yet implemented and will use the functions provided in the basic module */ NULL, /* allgather */ NULL, /* allgatherv */ mca_coll_hierarch_allreduce_intra, NULL, /* alltoall */ NULL, /* alltoallv */ NULL, /* alltoallw */ mca_coll_hierarch_barrier_intra, mca_coll_hierarch_bcast_intra, NULL, /* exscan */ NULL, /* gather */ NULL, /* gatherv */ mca_coll_hierarch_reduce_intra, NULL, /* reduce_scatter */ NULL, /* scan */ NULL, /* scatter */ NULL /* scatterv */ }; static const mca_coll_base_module_1_0_0_t null_intra = { /* Initialization / finalization functions */ mca_coll_hierarch_module_init, mca_coll_hierarch_module_finalize, /* Collective function pointers */ /* function pointers marked with NULL are not yet implemented and will use the functions provided in the basic module */ NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL }; /* * Initial query function that is invoked during MPI_INIT, allowing * this module to indicate what level of thread support it provides. */ int mca_coll_hierarch_init_query(bool allow_hierarch_user_threads, bool have_hidden_user_threads) { /* Don't ask. All done */ return OMPI_SUCCESS; } /* * Invoked when there's a new communicator that has been created. * Look at the communicator and decide which set of functions and * priority we want to return. */ const mca_coll_base_module_1_0_0_t * mca_coll_hierarch_comm_query(struct ompi_communicator_t *comm, int *priority, struct mca_coll_base_comm_t **data) { int size, rank; int color, ncount, maxncount; struct mca_coll_base_comm_t *tdata=NULL; int level; int ret=OMPI_SUCCESS; int ignore_sm=0; int symmetric=0; /* This module only works for intra-communicators at the moment */ if ( OMPI_COMM_IS_INTER(comm) ) { return NULL; } /* Get the priority level attached to this module. If priority = 0, we assume that we won't be chosen anyway, so we quit and improve therefore the startup time. */ *priority = mca_coll_hierarch_priority_param; if ( 0 >= mca_coll_hierarch_priority_param ) { return NULL; } /* Check whether we should ignore sm. This might be necessary to take advantage of the some ib or gm collectives. */ ignore_sm = mca_coll_hierarch_ignore_sm_param; /* Check whether we can assume a symmetric configuration. This can save commmunication and improve the startup time, since we can conclude from our configuration onto the configuration of every other process. */ symmetric = mca_coll_hierarch_symmetric_param; size = ompi_comm_size(comm); if ( size < 3 ) { /* No need for hierarchical collectives for 1 or 2 procs. */ return NULL; } rank = ompi_comm_rank(comm); /* allocate the data structure holding all information */ tdata = calloc ( 1, sizeof(struct mca_coll_base_comm_t)); if ( NULL == tdata ) { *priority = 0; return NULL; } tdata->hier_num_colorarr = size; tdata->hier_colorarr = (int *) malloc ( sizeof(int) * size); if ( NULL == tdata->hier_colorarr ) { *priority = 0; return NULL; } /* * walk through the list of registered protocols, and check which one * is feasable. * Later we start with level=0, and introduce the multi-cell check */ if ( ignore_sm ) { mca_coll_hierarch_max_protocol = HIER_MAXPROTOCOL - 1; } for ( level = mca_coll_hierarch_max_protocol - 1; level >0 ; level--) { mca_coll_hierarch_checkfor_component ( comm, level, hier_prot[level], &color, &ncount); /* This is probably a no-no! but for the moment we agreed with Jeff, that this might be the best solution. These functions emulate an allreduce and an allgather. */ if ( symmetric ) { maxncount = ncount; } else { ret = mca_coll_hierarch_allreduce_tmp (&ncount, &maxncount, 1, MPI_INT, MPI_MAX, comm ); if ( OMPI_SUCCESS != ret ) { return NULL; } } if ( 0 == maxncount ) { if ( mca_coll_hierarch_verbose_param ) { printf("%s:%d: nobody talks with %s. Continuing to next level.\n", comm->c_name, rank, hier_prot[level]); } continue; } else if ( maxncount == (size-1) ) { /* * everybody can talk to every other process with this protocol, * no need to continue in the hierarchy tree and for the * hierarchical component. * Its (size-1) because we do not count ourselves. * maxncount[1] should be zero. */ if ( mca_coll_hierarch_verbose_param ) { printf("%s:%d: everybody talks with %s. No need to continue\n", comm->c_name, rank, hier_prot[level]); } goto exit; } else { if ( mca_coll_hierarch_verbose_param ) { printf("%s:%d: %d procs talk with %s. Use this protocol, key %d\n", comm->c_name, rank, maxncount, hier_prot[level], color); } ret = mca_coll_hierarch_allgather_tmp (&color, 1, MPI_INT, tdata->hier_colorarr, 1, MPI_INT, comm ); if ( OMPI_SUCCESS != ret ) { return NULL; } tdata->hier_level = level; *data = tdata; return &null_intra; } } exit: if ( NULL != tdata->hier_colorarr ) { free ( tdata->hier_colorarr ) ; } if ( NULL != tdata ) { free ( tdata ); } *priority = 0; return NULL; } /* * Init module on the communicator */ const struct mca_coll_base_module_1_0_0_t * mca_coll_hierarch_module_init(struct ompi_communicator_t *comm) { int color; int size, rank, ret=OMPI_SUCCESS; int *llr=NULL; struct ompi_communicator_t *lcomm=NULL; struct ompi_communicator_t *llcomm=NULL; struct mca_coll_base_comm_t *data=NULL; struct mca_coll_hierarch_llead_t *llead=NULL; rank = ompi_comm_rank(comm); size = ompi_comm_size(comm); data = comm->c_coll_selected_data; color = data->hier_colorarr[rank]; /* Generate the subcommunicator based on the color returned by the previous function. */ ret = ompi_comm_split ( comm, color, rank, &lcomm, 0 ); if ( OMPI_SUCCESS != ret ) { goto exit; } data->hier_comm = comm; data->hier_lcomm = lcomm; data->hier_num_reqs = 2 * size; data->hier_reqs = (ompi_request_t **) malloc (sizeof(ompi_request_t)*size*2); if ( NULL == data->hier_reqs ) { goto exit; } /* allocate a certain number of the hierarch_llead structures, which store information about local leader and the according subcommunicators */ llead = (struct mca_coll_hierarch_llead_t * ) malloc ( sizeof(struct mca_coll_hierarch_llead_t)); if ( NULL == llead ) { goto exit; } /* These two routines set all relevant entries in the mca_coll_base_comm_t * structure. The first one makes all entries which are independent of the * offset (and have to be done only once per module. The second one is * depending on the offset, and has to be called therefore every time we need * a new llcomm */ mca_coll_hierarch_get_llr ( data ); mca_coll_hierarch_get_all_lleaders ( rank, data, llead, 1 ); /* Generate the lleader communicator assuming that all lleaders are the first process in the list of processes with the same color. A function generating other lleader-comms will follow soon. */ ret = ompi_comm_split ( comm, llead->am_lleader, rank, &llcomm, 0); if ( OMPI_SUCCESS != ret ) { goto exit; } llead->llcomm = llcomm; /* Store it now on the data structure */ OBJ_CONSTRUCT(&(data->hier_llead), ompi_pointer_array_t); ompi_pointer_array_add ( &(data->hier_llead), llead); if ( mca_coll_hierarch_verbose_param ) { mca_coll_hierarch_dump_struct (data); } exit: if ( NULL != llr ) { free ( llr ); } if ( OMPI_SUCCESS != ret ) { ompi_comm_free ( &lcomm ); if ( NULL != data ) { if ( NULL != data->hier_reqs ) { free ( data->hier_reqs); } if ( NULL != data->hier_colorarr ) { free ( data->hier_colorarr ) ; } if ( NULL != llead->lleaders ) { free ( llead->lleaders); } if ( NULL != data->hier_llr ) { free ( data->hier_llr ); } free ( data ); } return NULL; } return &intra; } /* * Finalize module on the communicator */ int mca_coll_hierarch_module_finalize(struct ompi_communicator_t *comm) { int i; int size; struct ompi_communicator_t *lcomm=NULL; struct mca_coll_base_comm_t *data=NULL; struct mca_coll_hierarch_llead_t *current=NULL; data = comm->c_coll_selected_data; lcomm = data->hier_lcomm; ompi_comm_free (&lcomm); free ( data->hier_reqs ); size = ompi_pointer_array_get_size ( &(data->hier_llead)); for ( i=0; ihier_llead), i ) ; if ( current->lleaders != NULL ) { ompi_comm_free ( &(current->llcomm)); free ( current->lleaders ); } free ( current ); } ompi_pointer_array_remove_all ( &(data->hier_llead)); OBJ_DESTRUCT (&(data->hier_llead)); free ( data->hier_colorarr ); free ( data->hier_llr); free ( data ); comm->c_coll_selected_data = NULL; return OMPI_SUCCESS; } int mca_coll_hierarch_comm_unquery ( struct ompi_communicator_t *comm, struct mca_coll_base_comm_t *data ) { free ( data->hier_colorarr ); free ( data ); return OMPI_SUCCESS; } int mca_coll_hierarch_get_all_lleaders ( int rank, struct mca_coll_base_comm_t *data, struct mca_coll_hierarch_llead_t * llead, int offset ) { int i, j, ret=OMPI_SUCCESS; int *cntarr=NULL; int mycolor; cntarr = (int *)calloc (1, sizeof (int)* data->hier_num_lleaders ); if ( NULL == cntarr ) { return OMPI_ERR_OUT_OF_RESOURCE; } llead->lleaders = (int *) malloc (sizeof(int) * data->hier_num_lleaders); if ( NULL == llead->lleaders ) { ret = OMPI_ERR_OUT_OF_RESOURCE; goto exit; } llead->offset = offset; for ( i=0; i < data->hier_num_lleaders; i++ ) { if ( data->hier_llr[i] == MPI_UNDEFINED ) { cntarr[i] = 1; llead->lleaders[i] = MPI_UNDEFINED; } } for ( i=0; ihier_num_colorarr; i++) { if ( data->hier_colorarr[i] == MPI_UNDEFINED ) { continue; } for ( j=0; jhier_num_lleaders; j++) { if ( cntarr[j] >= offset ) { continue; } if ( data->hier_colorarr[i] == data->hier_llr[j]) { cntarr[j]++; llead->lleaders[j] = i; break; } } } mycolor = data->hier_colorarr[rank]; if ( mycolor == MPI_UNDEFINED ) { llead->am_lleader = 1; llead->my_lleader = MPI_UNDEFINED; } else { llead->am_lleader = 0; for ( i=0; i< data->hier_num_lleaders; i++ ) { if ( data->hier_llr[i] == mycolor ) { llead->my_lleader = cntarr[i]-1; if ( llead->lleaders[i] == rank ) { llead->am_lleader = 1; } break; } } } exit: if ( NULL != cntarr ) { free ( cntarr ); } return ret; } int mca_coll_hierarch_get_llr ( struct mca_coll_base_comm_t *data ) { int i, j, cnt, found; int ncount; ncount = mca_coll_hierarch_count_lleaders ( data->hier_num_colorarr, data->hier_colorarr); data->hier_num_lleaders = ncount; data->hier_llr = (int *) malloc ( data->hier_num_lleaders * sizeof(int)); data->hier_max_offset = (int *) calloc ( 1, data->hier_num_lleaders * sizeof(int)); if ( ( NULL == data->hier_llr) || ( NULL == data->hier_max_offset )) { return OMPI_ERR_OUT_OF_RESOURCE; } data->hier_llr[0] = data->hier_colorarr[0]; data->hier_max_offset[0]=1; for ( cnt=1, i=1; ihier_num_colorarr; i++ ) { if ( data->hier_colorarr[i] == MPI_UNDEFINED ) { data->hier_llr[cnt] = data->hier_colorarr[i]; data->hier_max_offset[cnt] = 1; cnt++; continue; } for ( found=0, j=0; jhier_llr[j] == data->hier_colorarr[i]) { data->hier_max_offset[j]++; found = 1; break; } } if ( !found ) { data->hier_llr[cnt] = data->hier_colorarr[i]; data->hier_max_offset[cnt]++; cnt++; } } return OMPI_SUCCESS; } struct ompi_communicator_t* mca_coll_hierarch_get_llcomm (int root, struct mca_coll_base_comm_t *data, int* llroot, int* lroot) { struct ompi_communicator_t *llcomm=NULL; struct ompi_group_t *llgroup=NULL; struct ompi_group_t *group=NULL; struct mca_coll_hierarch_llead_t *llead=NULL; int found, i, rc, num_llead, offset; int rank = ompi_comm_rank (data->hier_comm); /* determine what our offset of root is in the colorarr */ offset = mca_coll_hierarch_get_offset ( root, data->hier_num_colorarr, data->hier_colorarr ); num_llead = ompi_pointer_array_get_size ( &(data->hier_llead) ); for ( found=0, i=0; i < num_llead; i++ ) { llead = (struct mca_coll_hierarch_llead_t *) ompi_pointer_array_get_item ( &(data->hier_llead), i ); if ( llead == NULL ) { continue; } if (llead->offset == offset ) { found = 1; break; } #if 0 else if () { /* the offset of root = maxoffset of this color and * the offset on llead is larger then offset of root. * then we can also use this llead structure */ } #endif } if ( !found ) { /* allocate a new llead element */ llead = (struct mca_coll_hierarch_llead_t *) malloc ( sizeof(struct mca_coll_hierarch_llead_t)); if ( NULL == llead ) { return NULL; } /* generate the list of lleaders with this offset */ mca_coll_hierarch_get_all_lleaders ( rank, data, llead, offset ); /* create new lleader subcommunicator */ rc = ompi_comm_split ( data->hier_comm, llead->am_lleader, root, &llcomm, 0); if ( OMPI_SUCCESS != rc ) { return NULL; } llead->llcomm = llcomm; /* Store the new element on the data struct */ ompi_pointer_array_add ( &(data->hier_llead), llead); } llcomm = llead->llcomm; *lroot = llead->my_lleader; *llroot = MPI_UNDEFINED; if ( MPI_COMM_NULL != llcomm ) { rc = ompi_comm_group ( data->hier_comm, &group); if ( OMPI_SUCCESS != rc ) { return NULL; } rc = ompi_comm_group ( llcomm, &llgroup); if ( OMPI_SUCCESS != rc ) { return NULL; } rc = ompi_group_translate_ranks ( group, 1, &root, llgroup, llroot); if ( OMPI_SUCCESS != rc ) { return NULL; } /* ompi_group_free (&llgroup) */ /* ompi_group_free (&group); */ } return llcomm; } /**********************************************************************/ /**********************************************************************/ /**********************************************************************/ /* This function checks how many processes are using the component 'component_name' for communication and returns this count in 'ncount'. Furthermore it returns a 'key', which can be used to split the communicator into subgroups, such that the new communicators will definitly have all processes communicate with this component. Oct 13: the algorithm has been modified such that it returns the number of processes using the specified component and the number of processes to which an even 'faster' protocol is being used. (Faster specified in this context as being further up in the list of hier_prot protocols specified at the beginning of this file). */ static void mca_coll_hierarch_checkfor_component ( struct ompi_communicator_t *comm, int component_level, char *component_name, int *key, int *ncount ) { ompi_bitmap_t reachable; ompi_proc_t **procs=NULL; struct mca_bml_base_endpoint_t **bml_endpoints=NULL; struct mca_bml_base_btl_array_t *bml_btl_array=NULL; mca_bml_base_btl_t *bml_btl=NULL; mca_btl_base_component_t *btl=NULL; int i, size, rc; int counter=0; int firstproc=999999; int rank = -1; int use_rdma=0; /* default values in case an error occurs */ *ncount=0; *key=MPI_UNDEFINED; /* Shall we check the the rdma list instead of send-list in the endpoint-structure? */ use_rdma = mca_coll_hierarch_use_rdma_param; size = ompi_comm_size ( comm ); rank = ompi_comm_rank ( comm ); OBJ_CONSTRUCT(&reachable, ompi_bitmap_t); rc = ompi_bitmap_init(&reachable, size); if(OMPI_SUCCESS != rc) { return; } bml_endpoints = (struct mca_bml_base_endpoint_t **) malloc ( size * sizeof(struct mca_bml_base_endpoint_t*)); if ( NULL == bml_endpoints ) { return; } procs = comm->c_local_group->grp_proc_pointers; rc = mca_bml.bml_add_procs ( size, procs, bml_endpoints, &reachable ); if(OMPI_SUCCESS != rc) { return; } for ( i=0; ibtl_rdma); } else { bml_btl_array = &(bml_endpoints[i]->btl_send); } bml_btl = mca_bml_base_btl_array_get_index ( bml_btl_array, 0 ); btl = bml_btl->btl->btl_component; /* sanity check */ if ( strcmp(btl->btl_version.mca_type_name,"btl") ) { printf("Oops, got the wrong component! type_name = %s\n", btl->btl_version.mca_type_name ); } /* check for the required component */ if (! strcmp (btl->btl_version.mca_component_name, component_name)){ counter++; if (ihier_comm ); printf("%d: Dump of hier-struct for comm %s cid %u\n", rank, c->hier_comm->c_name, c->hier_comm->c_contextid); printf("%d: No of llead communicators: %d No of lleaders: %d\n", rank, ompi_pointer_array_get_size ( &(c->hier_llead)), c->hier_num_lleaders ); for ( i=0; i < ompi_pointer_array_get_size(&(c->hier_llead)); i++ ) { current = ompi_pointer_array_get_item (&(c->hier_llead), i); if ( current == NULL ) { continue; } printf("%d: my_leader %d am_leader %d\n", rank, current->my_lleader, current->am_lleader ); for (j=0; jhier_num_lleaders; j++ ) { printf("%d: lleader[%d] = %d\n", rank, j, current->lleaders[j]); } } return; }