X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/45c3f1cfee86fb48c96d53f8267f99b6db6e3d7a..954676b700e711f38ec4d286d33d5427d3f4ca46:/src/smpi/colls/allreduce-smp-rdb.c diff --git a/src/smpi/colls/allreduce-smp-rdb.c b/src/smpi/colls/allreduce-smp-rdb.c index 5b67e5a975..9770d2993f 100644 --- a/src/smpi/colls/allreduce-smp-rdb.c +++ b/src/smpi/colls/allreduce-smp-rdb.c @@ -11,11 +11,6 @@ It uses 2-layer communication: binomial for intra-communication and rdb for inter-communication*/ -/* 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 /* ** NOTE ** Use -DMPICH2 if this code does not compile. @@ -40,10 +35,13 @@ int smpi_coll_tuned_allreduce_smp_rdb(void *send_buf, void *recv_buf, int count, 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)); + } /* #ifdef MPICH2_REDUCTION MPI_User_function * uop = MPIR_Op_table[op % 16 - 1]; @@ -58,7 +56,7 @@ int smpi_coll_tuned_allreduce_smp_rdb(void *send_buf, void *recv_buf, int count, 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); /* compute intra and inter ranking */ int intra_rank, inter_rank; @@ -184,6 +182,6 @@ int smpi_coll_tuned_allreduce_smp_rdb(void *send_buf, void *recv_buf, int count, mask >>= 1; } - free(tmp_buf); + smpi_free_tmp_buffer(tmp_buf); return MPI_SUCCESS; }