X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/992f80536a03eb52f1af10a3743de2d227390192..ec3e4ee5f1a7ffeb96e044057809944f364014e6:/src/smpi/colls/allreduce-smp-binomial-pipeline.c diff --git a/src/smpi/colls/allreduce-smp-binomial-pipeline.c b/src/smpi/colls/allreduce-smp-binomial-pipeline.c index 6e829f86c8..06a80a7a0c 100644 --- a/src/smpi/colls/allreduce-smp-binomial-pipeline.c +++ b/src/smpi/colls/allreduce-smp-binomial-pipeline.c @@ -1,3 +1,9 @@ +/* Copyright (c) 2013-2014. The SimGrid Team. + * All rights reserved. */ + +/* This program is free software; you can redistribute it and/or modify it + * under the terms of the license (GNU LGPL) which comes with this package. */ + #include "colls_private.h" /* IMPLEMENTED BY PITCH PATARASUK Non-topoloty-specific (however, number of cores/node need to be changed) @@ -6,11 +12,7 @@ inter-communication The communication are done in a pipeline fashion */ -/* change number of core per smp-node - we assume that number of core per process will be the same for all implementations */ -#ifndef NUM_CORE -#define NUM_CORE 8 -#endif + /* this is a default segment size for pipelining, but it is typically passed as a command line argument */ @@ -46,16 +48,19 @@ int smpi_coll_tuned_allreduce_smp_binomial_pipeline(void *send_buf, int tag = COLL_TAG_ALLREDUCE; int mask, src, dst; MPI_Status status; - int num_core = simcall_host_get_core(SIMIX_host_self()); - // do we use the default one or the number of cores in the platform ? - // if the number of cores is one, the platform may be simulated with 1 node = 1 core - if (num_core == 1) num_core = NUM_CORE; + if(smpi_comm_get_leaders_comm(comm)==MPI_COMM_NULL){ + smpi_comm_init_smp(comm); + } + int num_core=1; + if (smpi_comm_is_uniform(comm)){ + num_core = smpi_comm_size(smpi_comm_get_intra_comm(comm)); + } comm_size = smpi_comm_size(comm); rank = smpi_comm_rank(comm); MPI_Aint extent; extent = smpi_datatype_get_extent(dtype); - tmp_buf = (void *) xbt_malloc(count * extent); + tmp_buf = (void *) smpi_get_tmp_sendbuffer(count * extent); int intra_rank, inter_rank; intra_rank = rank % num_core; @@ -192,6 +197,6 @@ int smpi_coll_tuned_allreduce_smp_binomial_pipeline(void *send_buf, } } // for phase - free(tmp_buf); + smpi_free_tmp_buffer(tmp_buf); return MPI_SUCCESS; }