X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/7f04fa5306f52960db709bb520c49e9f4ecec000..3f54b3aa219f110121bb75de9209c94f3d1022bd:/src/smpi/colls/allreduce-rab-rsag.c diff --git a/src/smpi/colls/allreduce-rab-rsag.c b/src/smpi/colls/allreduce-rab-rsag.c index 54149d4db4..34e23d6369 100644 --- a/src/smpi/colls/allreduce-rab-rsag.c +++ b/src/smpi/colls/allreduce-rab-rsag.c @@ -1,27 +1,25 @@ -#include "colls.h" +#include "colls_private.h" //#include int smpi_coll_tuned_allreduce_rab_rsag(void *sbuff, void *rbuff, int count, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm) { - int nprocs, rank, type_size, tag = 543; + int nprocs, rank, tag = 543; int mask, dst, pof2, newrank, rem, newdst, i, send_idx, recv_idx, last_idx, send_cnt, recv_cnt, *cnts, *disps; MPI_Aint extent; MPI_Status status; void *tmp_buf = NULL; - MPI_Comm_size(comm, &nprocs); - MPI_Comm_rank(comm, &rank); + nprocs = smpi_comm_size(comm); + rank = smpi_comm_rank(comm); - MPI_Type_extent(dtype, &extent); + extent = smpi_datatype_get_extent(dtype); tmp_buf = (void *) xbt_malloc(count * extent); - MPI_Sendrecv(sbuff, count, dtype, rank, tag, rbuff, count, dtype, rank, tag, + smpi_mpi_sendrecv(sbuff, count, dtype, rank, tag, rbuff, count, dtype, rank, tag, comm, &status); - MPI_Type_size(dtype, &type_size); - // find nearest power-of-two less than or equal to comm_size pof2 = 1; while (pof2 <= nprocs) @@ -40,7 +38,7 @@ int smpi_coll_tuned_allreduce_rab_rsag(void *sbuff, void *rbuff, int count, // even if (rank % 2 == 0) { - MPI_Send(rbuff, count, dtype, rank + 1, tag, comm); + smpi_mpi_send(rbuff, count, dtype, rank + 1, tag, comm); // temporarily set the rank to -1 so that this // process does not pariticipate in recursive @@ -48,11 +46,11 @@ int smpi_coll_tuned_allreduce_rab_rsag(void *sbuff, void *rbuff, int count, newrank = -1; } else // odd { - MPI_Recv(tmp_buf, count, dtype, rank - 1, tag, comm, &status); + smpi_mpi_recv(tmp_buf, count, dtype, rank - 1, tag, comm, &status); // do the reduction on received data. since the // ordering is right, it doesn't matter whether // the operation is commutative or not. - star_reduction(op, tmp_buf, rbuff, &count, &dtype); + smpi_op_apply(op, tmp_buf, rbuff, &count, &dtype); // change the rank newrank = rank / 2; @@ -76,8 +74,8 @@ int smpi_coll_tuned_allreduce_rab_rsag(void *sbuff, void *rbuff, int count, // reduce-scatter, calculate the count that each process receives // and the displacement within the buffer - cnts = (int *) malloc(pof2 * sizeof(int)); - disps = (int *) malloc(pof2 * sizeof(int)); + cnts = (int *) xbt_malloc(pof2 * sizeof(int)); + disps = (int *) xbt_malloc(pof2 * sizeof(int)); for (i = 0; i < (pof2 - 1); i++) cnts[i] = count / pof2; @@ -111,7 +109,7 @@ int smpi_coll_tuned_allreduce_rab_rsag(void *sbuff, void *rbuff, int count, } // Send data from recvbuf. Recv into tmp_buf - MPI_Sendrecv((char *) rbuff + disps[send_idx] * extent, send_cnt, + smpi_mpi_sendrecv((char *) rbuff + disps[send_idx] * extent, send_cnt, dtype, dst, tag, (char *) tmp_buf + disps[recv_idx] * extent, recv_cnt, dtype, dst, tag, comm, &status); @@ -121,7 +119,7 @@ int smpi_coll_tuned_allreduce_rab_rsag(void *sbuff, void *rbuff, int count, // This algorithm is used only for predefined ops // and predefined ops are always commutative. - star_reduction(op, (char *) tmp_buf + disps[recv_idx] * extent, + smpi_op_apply(op, (char *) tmp_buf + disps[recv_idx] * extent, (char *) rbuff + disps[recv_idx] * extent, &recv_cnt, &dtype); @@ -162,7 +160,7 @@ int smpi_coll_tuned_allreduce_rab_rsag(void *sbuff, void *rbuff, int count, recv_cnt += cnts[i]; } - MPI_Sendrecv((char *) rbuff + disps[send_idx] * extent, send_cnt, + smpi_mpi_sendrecv((char *) rbuff + disps[send_idx] * extent, send_cnt, dtype, dst, tag, (char *) rbuff + disps[recv_idx] * extent, recv_cnt, dtype, dst, tag, comm, &status); @@ -183,9 +181,9 @@ int smpi_coll_tuned_allreduce_rab_rsag(void *sbuff, void *rbuff, int count, if (rank < 2 * rem) { if (rank % 2) // odd - MPI_Send(rbuff, count, dtype, rank - 1, tag, comm); + smpi_mpi_send(rbuff, count, dtype, rank - 1, tag, comm); else // even - MPI_Recv(rbuff, count, dtype, rank + 1, tag, comm, &status); + smpi_mpi_recv(rbuff, count, dtype, rank + 1, tag, comm, &status); } free(tmp_buf);