X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/9cca09ad88fcaeb49f2637de97df519c3920aaa4..b6dde5ff6f208b83545a4b4e9e81712d0d8617a1:/src/smpi/colls/smpi_mvapich2_selector_stampede.hpp diff --git a/src/smpi/colls/smpi_mvapich2_selector_stampede.hpp b/src/smpi/colls/smpi_mvapich2_selector_stampede.hpp index a922738a50..8b3928c0d7 100644 --- a/src/smpi/colls/smpi_mvapich2_selector_stampede.hpp +++ b/src/smpi/colls/smpi_mvapich2_selector_stampede.hpp @@ -2,21 +2,26 @@ /* This is the tuning used by MVAPICH for Stampede platform based on (MV2_ARCH_INTEL_XEON_E5_2680_16, * MV2_HCA_MLX_CX_FDR) */ -/* Copyright (c) 2009-2017. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2009-2020. 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. */ /************ Alltoall variables and initializers */ +#ifndef SMPI_MVAPICH2_SELECTOR_STAMPEDE_HPP +#define SMPI_MVAPICH2_SELECTOR_STAMPEDE_HPP + +#include + #define MV2_MAX_NB_THRESHOLDS 32 -XBT_PUBLIC(void) smpi_coll_cleanup_mvapich2(void); +XBT_PUBLIC void smpi_coll_cleanup_mvapich2(void); struct mv2_alltoall_tuning_element { int min; int max; - int (*MV2_pt_Alltoall_function)(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, + int (*MV2_pt_Alltoall_function)(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm_ptr); }; @@ -27,7 +32,7 @@ struct mv2_alltoall_tuning_table { mv2_alltoall_tuning_element in_place_algo_table[MV2_MAX_NB_THRESHOLDS]; }; -int (*MV2_Alltoall_function)(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, +int (*MV2_Alltoall_function)(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm_ptr) = NULL; /* Indicates number of processes per node */ @@ -37,26 +42,23 @@ int mv2_alltoall_num_ppn_conf = 1; int* mv2_size_alltoall_tuning_table = NULL; mv2_alltoall_tuning_table** mv2_alltoall_thresholds_table = NULL; -#define MPIR_Alltoall_bruck_MV2 simgrid::smpi::Coll_alltoall_bruck::alltoall -#define MPIR_Alltoall_RD_MV2 simgrid::smpi::Coll_alltoall_rdb::alltoall -#define MPIR_Alltoall_Scatter_dest_MV2 simgrid::smpi::Coll_alltoall_mvapich2_scatter_dest::alltoall -#define MPIR_Alltoall_pairwise_MV2 simgrid::smpi::Coll_alltoall_pair::alltoall -#define MPIR_Alltoall_inplace_MV2 simgrid::smpi::Coll_alltoall_ring::alltoall +#define MPIR_Alltoall_bruck_MV2 simgrid::smpi::alltoall__bruck +#define MPIR_Alltoall_RD_MV2 simgrid::smpi::alltoall__rdb +#define MPIR_Alltoall_Scatter_dest_MV2 simgrid::smpi::alltoall__mvapich2_scatter_dest +#define MPIR_Alltoall_pairwise_MV2 simgrid::smpi::alltoall__pair +#define MPIR_Alltoall_inplace_MV2 simgrid::smpi::alltoall__ring static void init_mv2_alltoall_tables_stampede() { - int i; int agg_table_sum = 0; mv2_alltoall_tuning_table** table_ptrs = NULL; mv2_alltoall_num_ppn_conf = 3; - if (simgrid::smpi::Colls::smpi_coll_cleanup_callback == NULL) - simgrid::smpi::Colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; - mv2_alltoall_thresholds_table = static_cast( - xbt_malloc(sizeof(mv2_alltoall_tuning_table*) * mv2_alltoall_num_ppn_conf)); - table_ptrs = static_cast( - xbt_malloc(sizeof(mv2_alltoall_tuning_table*) * mv2_alltoall_num_ppn_conf)); - mv2_size_alltoall_tuning_table = static_cast(xbt_malloc(sizeof(int) * mv2_alltoall_num_ppn_conf)); - mv2_alltoall_table_ppn_conf = static_cast(xbt_malloc(mv2_alltoall_num_ppn_conf * sizeof(int))); + if (simgrid::smpi::colls::smpi_coll_cleanup_callback == NULL) + simgrid::smpi::colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; + mv2_alltoall_thresholds_table = new mv2_alltoall_tuning_table*[mv2_alltoall_num_ppn_conf]; + table_ptrs = new mv2_alltoall_tuning_table*[mv2_alltoall_num_ppn_conf]; + mv2_size_alltoall_tuning_table = new int[mv2_alltoall_num_ppn_conf]; + mv2_alltoall_table_ppn_conf = new int[mv2_alltoall_num_ppn_conf]; mv2_alltoall_table_ppn_conf[0] = 1; mv2_size_alltoall_tuning_table[0] = 6; mv2_alltoall_tuning_table mv2_tmp_alltoall_thresholds_table_1ppn[] = { @@ -314,19 +316,16 @@ static void init_mv2_alltoall_tables_stampede() }; table_ptrs[2] = mv2_tmp_alltoall_thresholds_table_16ppn; agg_table_sum = 0; - for (i = 0; i < mv2_alltoall_num_ppn_conf; i++) { + for (int i = 0; i < mv2_alltoall_num_ppn_conf; i++) { agg_table_sum += mv2_size_alltoall_tuning_table[i]; } - mv2_alltoall_thresholds_table[0] = - static_cast(xbt_malloc(agg_table_sum * sizeof(mv2_alltoall_tuning_table))); - memcpy(mv2_alltoall_thresholds_table[0], table_ptrs[0], - (sizeof(mv2_alltoall_tuning_table) * mv2_size_alltoall_tuning_table[0])); - for (i = 1; i < mv2_alltoall_num_ppn_conf; i++) { + mv2_alltoall_thresholds_table[0] = new mv2_alltoall_tuning_table[agg_table_sum]; + std::copy_n(table_ptrs[0], mv2_size_alltoall_tuning_table[0], mv2_alltoall_thresholds_table[0]); + for (int i = 1; i < mv2_alltoall_num_ppn_conf; i++) { mv2_alltoall_thresholds_table[i] = mv2_alltoall_thresholds_table[i - 1] + mv2_size_alltoall_tuning_table[i - 1]; - memcpy(mv2_alltoall_thresholds_table[i], table_ptrs[i], - (sizeof(mv2_alltoall_tuning_table) * mv2_size_alltoall_tuning_table[i])); + std::copy_n(table_ptrs[i], mv2_size_alltoall_tuning_table[i], mv2_alltoall_thresholds_table[i]); } - xbt_free(table_ptrs); + delete[] table_ptrs; } /************ Allgather variables and initializers */ @@ -334,18 +333,18 @@ static void init_mv2_alltoall_tables_stampede() struct mv2_allgather_tuning_element { int min; int max; - int (*MV2_pt_Allgatherction)(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, + int (*MV2_pt_Allgatherction)(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm_ptr); }; struct mv2_allgather_tuning_table { int numproc; - int two_level[MV2_MAX_NB_THRESHOLDS]; + bool two_level[MV2_MAX_NB_THRESHOLDS]; int size_inter_table; mv2_allgather_tuning_element inter_leader[MV2_MAX_NB_THRESHOLDS]; }; -int (*MV2_Allgatherction)(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, +int (*MV2_Allgatherction)(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm); int* mv2_allgather_table_ppn_conf = NULL; @@ -353,38 +352,34 @@ int mv2_allgather_num_ppn_conf = 1; int* mv2_size_allgather_tuning_table = NULL; mv2_allgather_tuning_table** mv2_allgather_thresholds_table = NULL; -static int MPIR_Allgather_RD_Allgather_Comm_MV2(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, +static int MPIR_Allgather_RD_Allgather_Comm_MV2(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm_ptr) { return 0; } -#define MPIR_Allgather_Bruck_MV2 simgrid::smpi::Coll_allgather_bruck::allgather -#define MPIR_Allgather_RD_MV2 simgrid::smpi::Coll_allgather_rdb::allgather -#define MPIR_Allgather_Ring_MV2 simgrid::smpi::Coll_allgather_ring::allgather -#define MPIR_2lvl_Allgather_MV2 simgrid::smpi::Coll_allgather_mvapich2_smp::allgather +#define MPIR_Allgather_Bruck_MV2 simgrid::smpi::allgather__bruck +#define MPIR_Allgather_RD_MV2 simgrid::smpi::allgather__rdb +#define MPIR_Allgather_Ring_MV2 simgrid::smpi::allgather__ring +#define MPIR_2lvl_Allgather_MV2 simgrid::smpi::allgather__mvapich2_smp static void init_mv2_allgather_tables_stampede() { - int i; int agg_table_sum = 0; - if (simgrid::smpi::Colls::smpi_coll_cleanup_callback == NULL) - simgrid::smpi::Colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; - mv2_allgather_tuning_table** table_ptrs = NULL; + if (simgrid::smpi::colls::smpi_coll_cleanup_callback == NULL) + simgrid::smpi::colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; mv2_allgather_num_ppn_conf = 3; - mv2_allgather_thresholds_table = static_cast( - xbt_malloc(sizeof(mv2_allgather_tuning_table*) * mv2_allgather_num_ppn_conf)); - table_ptrs = static_cast( - xbt_malloc(sizeof(mv2_allgather_tuning_table*) * mv2_allgather_num_ppn_conf)); - mv2_size_allgather_tuning_table = static_cast(xbt_malloc(sizeof(int) * mv2_allgather_num_ppn_conf)); - mv2_allgather_table_ppn_conf = static_cast(xbt_malloc(mv2_allgather_num_ppn_conf * sizeof(int))); + mv2_allgather_thresholds_table = new mv2_allgather_tuning_table*[mv2_allgather_num_ppn_conf]; + mv2_allgather_tuning_table** table_ptrs = new mv2_allgather_tuning_table*[mv2_allgather_num_ppn_conf]; + mv2_size_allgather_tuning_table = new int[mv2_allgather_num_ppn_conf]; + mv2_allgather_table_ppn_conf = new int[mv2_allgather_num_ppn_conf]; mv2_allgather_table_ppn_conf[0] = 1; mv2_size_allgather_tuning_table[0] = 6; mv2_allgather_tuning_table mv2_tmp_allgather_thresholds_table_1ppn[] = { { 2, - {0}, + {false}, 1, { {0, -1, &MPIR_Allgather_Ring_MV2}, @@ -392,7 +387,7 @@ static void init_mv2_allgather_tables_stampede() }, { 4, - {0, 0}, + {false, false}, 2, { {0, 262144, &MPIR_Allgather_RD_MV2}, {262144, -1, &MPIR_Allgather_Ring_MV2}, @@ -400,7 +395,7 @@ static void init_mv2_allgather_tables_stampede() }, { 8, - {0, 0}, + {false, false}, 2, { {0, 131072, &MPIR_Allgather_RD_MV2}, {131072, -1, &MPIR_Allgather_Ring_MV2}, @@ -408,7 +403,7 @@ static void init_mv2_allgather_tables_stampede() }, { 16, - {0, 0}, + {false, false}, 2, { {0, 131072, &MPIR_Allgather_RD_MV2}, {131072, -1, &MPIR_Allgather_Ring_MV2}, @@ -416,7 +411,7 @@ static void init_mv2_allgather_tables_stampede() }, { 32, - {0, 0}, + {false, false}, 2, { {0, 65536, &MPIR_Allgather_RD_MV2}, {65536, -1, &MPIR_Allgather_Ring_MV2}, @@ -424,7 +419,7 @@ static void init_mv2_allgather_tables_stampede() }, { 64, - {0, 0}, + {false, false}, 2, { {0, 32768, &MPIR_Allgather_RD_MV2}, {32768, -1, &MPIR_Allgather_Ring_MV2}, @@ -437,7 +432,7 @@ static void init_mv2_allgather_tables_stampede() mv2_allgather_tuning_table mv2_tmp_allgather_thresholds_table_2ppn[] = { { 4, - {0, 0}, + {false, false}, 2, { {0, 524288, &MPIR_Allgather_RD_MV2}, {524288, -1, &MPIR_Allgather_Ring_MV2}, @@ -445,7 +440,7 @@ static void init_mv2_allgather_tables_stampede() }, { 8, - {0, 1, 0}, + {false, true, false}, 2, { {0, 32768, &MPIR_Allgather_RD_MV2}, @@ -455,7 +450,7 @@ static void init_mv2_allgather_tables_stampede() }, { 16, - {0, 1, 0}, + {false, true, false}, 2, { {0, 16384, &MPIR_Allgather_RD_MV2}, @@ -465,7 +460,7 @@ static void init_mv2_allgather_tables_stampede() }, { 32, - {1, 1, 0}, + {true, true, false}, 2, { {0, 65536, &MPIR_Allgather_RD_MV2}, @@ -475,7 +470,7 @@ static void init_mv2_allgather_tables_stampede() }, { 64, - {1, 1, 0}, + {true, true, false}, 2, { {0, 32768, &MPIR_Allgather_RD_MV2}, @@ -485,7 +480,7 @@ static void init_mv2_allgather_tables_stampede() }, { 128, - {1, 1, 0}, + {true, true, false}, 2, { {0, 65536, &MPIR_Allgather_RD_MV2}, @@ -500,7 +495,7 @@ static void init_mv2_allgather_tables_stampede() mv2_allgather_tuning_table mv2_tmp_allgather_thresholds_table_16ppn[] = { { 16, - {0, 0}, + {false, false}, 2, { {0, 1024, &MPIR_Allgather_RD_MV2}, {1024, -1, &MPIR_Allgather_Ring_MV2}, @@ -508,7 +503,7 @@ static void init_mv2_allgather_tables_stampede() }, { 32, - {0, 0}, + {false, false}, 2, { {0, 1024, &MPIR_Allgather_RD_Allgather_Comm_MV2}, {1024, -1, &MPIR_Allgather_Ring_MV2}, @@ -516,7 +511,7 @@ static void init_mv2_allgather_tables_stampede() }, { 64, - {0, 0}, + {false, false}, 2, { {0, 1024, &MPIR_Allgather_RD_Allgather_Comm_MV2}, {1024, -1, &MPIR_Allgather_Ring_MV2}, @@ -524,7 +519,7 @@ static void init_mv2_allgather_tables_stampede() }, { 128, - {0, 0}, + {false, false}, 2, { {0, 1024, &MPIR_Allgather_RD_Allgather_Comm_MV2}, {1024, -1, &MPIR_Allgather_Ring_MV2}, @@ -532,7 +527,7 @@ static void init_mv2_allgather_tables_stampede() }, { 256, - {0, 0}, + {false, false}, 2, { {0, 1024, &MPIR_Allgather_RD_Allgather_Comm_MV2}, {1024, -1, &MPIR_Allgather_Ring_MV2}, @@ -540,7 +535,7 @@ static void init_mv2_allgather_tables_stampede() }, { 512, - {0, 0}, + {false, false}, 2, { {0, 1024, &MPIR_Allgather_RD_Allgather_Comm_MV2}, {1024, -1, &MPIR_Allgather_Ring_MV2}, @@ -550,19 +545,16 @@ static void init_mv2_allgather_tables_stampede() }; table_ptrs[2] = mv2_tmp_allgather_thresholds_table_16ppn; agg_table_sum = 0; - for (i = 0; i < mv2_allgather_num_ppn_conf; i++) { + for (int i = 0; i < mv2_allgather_num_ppn_conf; i++) { agg_table_sum += mv2_size_allgather_tuning_table[i]; } - mv2_allgather_thresholds_table[0] = - static_cast(xbt_malloc(agg_table_sum * sizeof(mv2_allgather_tuning_table))); - memcpy(mv2_allgather_thresholds_table[0], table_ptrs[0], - (sizeof(mv2_allgather_tuning_table) * mv2_size_allgather_tuning_table[0])); - for (i = 1; i < mv2_allgather_num_ppn_conf; i++) { + mv2_allgather_thresholds_table[0] = new mv2_allgather_tuning_table[agg_table_sum]; + std::copy_n(table_ptrs[0], mv2_size_allgather_tuning_table[0], mv2_allgather_thresholds_table[0]); + for (int i = 1; i < mv2_allgather_num_ppn_conf; i++) { mv2_allgather_thresholds_table[i] = mv2_allgather_thresholds_table[i - 1] + mv2_size_allgather_tuning_table[i - 1]; - memcpy(mv2_allgather_thresholds_table[i], table_ptrs[i], - (sizeof(mv2_allgather_tuning_table) * mv2_size_allgather_tuning_table[i])); + std::copy_n(table_ptrs[i], mv2_size_allgather_tuning_table[i], mv2_allgather_thresholds_table[i]); } - xbt_free(table_ptrs); + delete[] table_ptrs; } /************ Gather variables and initializers */ @@ -570,7 +562,7 @@ static void init_mv2_allgather_tables_stampede() struct mv2_gather_tuning_element { int min; int max; - int (*MV2_pt_Gather_function)(void* sendbuf, int sendcnt, MPI_Datatype sendtype, void* recvbuf, int recvcnt, + int (*MV2_pt_Gather_function)(const void* sendbuf, int sendcnt, MPI_Datatype sendtype, void* recvbuf, int recvcnt, MPI_Datatype recvtype, int root, MPI_Comm comm_ptr); }; @@ -585,24 +577,23 @@ struct mv2_gather_tuning_table { int mv2_size_gather_tuning_table = 7; mv2_gather_tuning_table* mv2_gather_thresholds_table = NULL; -typedef int (*MV2_Gather_function_ptr)(void* sendbuf, int sendcnt, MPI_Datatype sendtype, void* recvbuf, int recvcnt, +typedef int (*MV2_Gather_function_ptr)(const void* sendbuf, int sendcnt, MPI_Datatype sendtype, void* recvbuf, int recvcnt, MPI_Datatype recvtype, int root, MPI_Comm comm); MV2_Gather_function_ptr MV2_Gather_inter_leader_function = NULL; MV2_Gather_function_ptr MV2_Gather_intra_node_function = NULL; -#define MPIR_Gather_MV2_Direct simgrid::smpi::Coll_gather_ompi_basic_linear::gather -#define MPIR_Gather_MV2_two_level_Direct simgrid::smpi::Coll_gather_mvapich2_two_level::gather -#define MPIR_Gather_intra simgrid::smpi::Coll_gather_mpich::gather +#define MPIR_Gather_MV2_Direct simgrid::smpi::gather__ompi_basic_linear +#define MPIR_Gather_MV2_two_level_Direct simgrid::smpi::gather__mvapich2_two_level +#define MPIR_Gather_intra simgrid::smpi::gather__mpich static void init_mv2_gather_tables_stampede() { - if (simgrid::smpi::Colls::smpi_coll_cleanup_callback == NULL) - simgrid::smpi::Colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; + if (simgrid::smpi::colls::smpi_coll_cleanup_callback == NULL) + simgrid::smpi::colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; mv2_size_gather_tuning_table = 7; - mv2_gather_thresholds_table = - static_cast(xbt_malloc(mv2_size_gather_tuning_table * sizeof(mv2_gather_tuning_table))); + mv2_gather_thresholds_table = new mv2_gather_tuning_table[mv2_size_gather_tuning_table]; mv2_gather_tuning_table mv2_tmp_gather_thresholds_table[] = { {16, 2, @@ -653,8 +644,7 @@ static void init_mv2_gather_tables_stampede() {{0, -1, &MPIR_Gather_intra}}}, }; - memcpy(mv2_gather_thresholds_table, mv2_tmp_gather_thresholds_table, - mv2_size_gather_tuning_table * sizeof(mv2_gather_tuning_table)); + std::copy_n(mv2_tmp_gather_thresholds_table, mv2_size_gather_tuning_table, mv2_gather_thresholds_table); } /************ Allgatherv variables and initializers */ @@ -662,8 +652,8 @@ static void init_mv2_gather_tables_stampede() struct mv2_allgatherv_tuning_element { int min; int max; - int (*MV2_pt_Allgatherv_function)(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int* recvcounts, - int* displs, MPI_Datatype recvtype, MPI_Comm commg); + int (*MV2_pt_Allgatherv_function)(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int* recvcounts, + const int* displs, MPI_Datatype recvtype, MPI_Comm commg); }; struct mv2_allgatherv_tuning_table { @@ -672,23 +662,22 @@ struct mv2_allgatherv_tuning_table { mv2_allgatherv_tuning_element inter_leader[MV2_MAX_NB_THRESHOLDS]; }; -int (*MV2_Allgatherv_function)(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int* recvcounts, - int* displs, MPI_Datatype recvtype, MPI_Comm comm); +int (*MV2_Allgatherv_function)(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int* recvcounts, + const int* displs, MPI_Datatype recvtype, MPI_Comm comm); int mv2_size_allgatherv_tuning_table = 0; mv2_allgatherv_tuning_table* mv2_allgatherv_thresholds_table = NULL; -#define MPIR_Allgatherv_Rec_Doubling_MV2 simgrid::smpi::Coll_allgatherv_mpich_rdb::allgatherv -#define MPIR_Allgatherv_Bruck_MV2 simgrid::smpi::Coll_allgatherv_ompi_bruck::allgatherv -#define MPIR_Allgatherv_Ring_MV2 simgrid::smpi::Coll_allgatherv_mpich_ring::allgatherv +#define MPIR_Allgatherv_Rec_Doubling_MV2 simgrid::smpi::allgatherv__mpich_rdb +#define MPIR_Allgatherv_Bruck_MV2 simgrid::smpi::allgatherv__ompi_bruck +#define MPIR_Allgatherv_Ring_MV2 simgrid::smpi::allgatherv__mpich_ring static void init_mv2_allgatherv_tables_stampede() { - if (simgrid::smpi::Colls::smpi_coll_cleanup_callback == NULL) - simgrid::smpi::Colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; + if (simgrid::smpi::colls::smpi_coll_cleanup_callback == NULL) + simgrid::smpi::colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; mv2_size_allgatherv_tuning_table = 6; - mv2_allgatherv_thresholds_table = static_cast( - xbt_malloc(mv2_size_allgatherv_tuning_table * sizeof(mv2_allgatherv_tuning_table))); + mv2_allgatherv_thresholds_table = new mv2_allgatherv_tuning_table[mv2_size_allgatherv_tuning_table]; mv2_allgatherv_tuning_table mv2_tmp_allgatherv_thresholds_table[] = { { 16, @@ -734,8 +723,7 @@ static void init_mv2_allgatherv_tables_stampede() }, }; - memcpy(mv2_allgatherv_thresholds_table, mv2_tmp_allgatherv_thresholds_table, - mv2_size_allgatherv_tuning_table * sizeof(mv2_allgatherv_tuning_table)); + std::copy_n(mv2_tmp_allgatherv_thresholds_table, mv2_size_allgatherv_tuning_table, mv2_allgatherv_thresholds_table); } /************ Allreduce variables and initializers */ @@ -743,71 +731,70 @@ static void init_mv2_allgatherv_tables_stampede() struct mv2_allreduce_tuning_element { int min; int max; - int (*MV2_pt_Allreducection)(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, + int (*MV2_pt_Allreducection)(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm); }; struct mv2_allreduce_tuning_table { int numproc; - int mcast_enabled; - int is_two_level_allreduce[MV2_MAX_NB_THRESHOLDS]; + bool mcast_enabled; + bool is_two_level_allreduce[MV2_MAX_NB_THRESHOLDS]; int size_inter_table; mv2_allreduce_tuning_element inter_leader[MV2_MAX_NB_THRESHOLDS]; int size_intra_table; mv2_allreduce_tuning_element intra_node[MV2_MAX_NB_THRESHOLDS]; }; -int (*MV2_Allreducection)(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, +int (*MV2_Allreducection)(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) = NULL; -int (*MV2_Allreduce_intra_function)(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, +int (*MV2_Allreduce_intra_function)(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) = NULL; int mv2_size_allreduce_tuning_table = 0; mv2_allreduce_tuning_table* mv2_allreduce_thresholds_table = NULL; -static int MPIR_Allreduce_mcst_reduce_two_level_helper_MV2(void* sendbuf, void* recvbuf, int count, +static int MPIR_Allreduce_mcst_reduce_two_level_helper_MV2(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) { return 0; } -static int MPIR_Allreduce_mcst_reduce_redscat_gather_MV2(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, +static int MPIR_Allreduce_mcst_reduce_redscat_gather_MV2(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) { return 0; } -static int MPIR_Allreduce_reduce_p2p_MV2(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, +static int MPIR_Allreduce_reduce_p2p_MV2(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) { - simgrid::smpi::Colls::reduce(sendbuf, recvbuf, count, datatype, op, 0, comm); + simgrid::smpi::colls::reduce(sendbuf, recvbuf, count, datatype, op, 0, comm); return MPI_SUCCESS; } -static int MPIR_Allreduce_reduce_shmem_MV2(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, +static int MPIR_Allreduce_reduce_shmem_MV2(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) { - simgrid::smpi::Colls::reduce(sendbuf, recvbuf, count, datatype, op, 0, comm); + simgrid::smpi::colls::reduce(sendbuf, recvbuf, count, datatype, op, 0, comm); return MPI_SUCCESS; } -#define MPIR_Allreduce_pt2pt_rd_MV2 simgrid::smpi::Coll_allreduce_rdb::allreduce -#define MPIR_Allreduce_pt2pt_rs_MV2 simgrid::smpi::Coll_allreduce_mvapich2_rs::allreduce -#define MPIR_Allreduce_two_level_MV2 simgrid::smpi::Coll_allreduce_mvapich2_two_level::allreduce +#define MPIR_Allreduce_pt2pt_rd_MV2 simgrid::smpi::allreduce__rdb +#define MPIR_Allreduce_pt2pt_rs_MV2 simgrid::smpi::allreduce__mvapich2_rs +#define MPIR_Allreduce_two_level_MV2 simgrid::smpi::allreduce__mvapich2_two_level static void init_mv2_allreduce_tables_stampede() { - if (simgrid::smpi::Colls::smpi_coll_cleanup_callback == NULL) - simgrid::smpi::Colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; + if (simgrid::smpi::colls::smpi_coll_cleanup_callback == NULL) + simgrid::smpi::colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; mv2_size_allreduce_tuning_table = 8; - mv2_allreduce_thresholds_table = static_cast( - xbt_malloc(mv2_size_allreduce_tuning_table * sizeof(mv2_allreduce_tuning_table))); + mv2_allreduce_thresholds_table = new mv2_allreduce_tuning_table[mv2_size_allreduce_tuning_table]; mv2_allreduce_tuning_table mv2_tmp_allreduce_thresholds_table[] = { { 16, - 0, - {1, 0}, + false, + {true, false}, 2, { {0, 1024, &MPIR_Allreduce_pt2pt_rd_MV2}, {1024, -1, &MPIR_Allreduce_pt2pt_rs_MV2}, @@ -819,8 +806,8 @@ static void init_mv2_allreduce_tables_stampede() }, { 32, - 0, - {1, 1, 0}, + false, + {true, true, false}, 3, { {0, 1024, &MPIR_Allreduce_pt2pt_rd_MV2}, @@ -834,8 +821,8 @@ static void init_mv2_allreduce_tables_stampede() }, { 64, - 0, - {1, 1, 0}, + false, + {true, true, false}, 3, { {0, 512, &MPIR_Allreduce_pt2pt_rd_MV2}, @@ -849,8 +836,8 @@ static void init_mv2_allreduce_tables_stampede() }, { 128, - 0, - {1, 1, 0}, + false, + {true, true, false}, 3, { {0, 512, &MPIR_Allreduce_pt2pt_rd_MV2}, @@ -864,8 +851,8 @@ static void init_mv2_allreduce_tables_stampede() }, { 256, - 0, - {1, 1, 0}, + false, + {true, true, false}, 3, { {0, 512, &MPIR_Allreduce_pt2pt_rd_MV2}, @@ -879,8 +866,8 @@ static void init_mv2_allreduce_tables_stampede() }, { 512, - 0, - {1, 1, 0}, + false, + {true, true, false}, 3, { {0, 512, &MPIR_Allreduce_pt2pt_rd_MV2}, @@ -894,8 +881,8 @@ static void init_mv2_allreduce_tables_stampede() }, { 1024, - 0, - {1, 1, 1, 0}, + false, + {true, true, true, false}, 4, { {0, 512, &MPIR_Allreduce_pt2pt_rd_MV2}, @@ -910,8 +897,8 @@ static void init_mv2_allreduce_tables_stampede() }, { 2048, - 0, - {1, 1, 1, 0}, + false, + {true, true, true, false}, 4, { {0, 64, &MPIR_Allreduce_pt2pt_rd_MV2}, @@ -927,8 +914,7 @@ static void init_mv2_allreduce_tables_stampede() }, }; - memcpy(mv2_allreduce_thresholds_table, mv2_tmp_allreduce_thresholds_table, - mv2_size_allreduce_tuning_table * sizeof(mv2_allreduce_tuning_table)); + std::copy_n(mv2_tmp_allreduce_thresholds_table, mv2_size_allreduce_tuning_table, mv2_allreduce_thresholds_table); } struct mv2_bcast_tuning_element { @@ -943,7 +929,7 @@ struct mv2_bcast_tuning_table { int bcast_segment_size; int intra_node_knomial_factor; int inter_node_knomial_factor; - int is_two_level_bcast[MV2_MAX_NB_THRESHOLDS]; + bool is_two_level_bcast[MV2_MAX_NB_THRESHOLDS]; int size_inter_table; mv2_bcast_tuning_element inter_leader[MV2_MAX_NB_THRESHOLDS]; int size_intra_table; @@ -969,33 +955,32 @@ int mv2_intra_node_knomial_factor = 4; #define INTRA_NODE_ROOT 0 -#define MPIR_Pipelined_Bcast_Zcpy_MV2 simgrid::smpi::Coll_bcast_mpich::bcast -#define MPIR_Pipelined_Bcast_MV2 simgrid::smpi::Coll_bcast_mpich::bcast -#define MPIR_Bcast_binomial_MV2 simgrid::smpi::Coll_bcast_binomial_tree::bcast -#define MPIR_Bcast_scatter_ring_allgather_shm_MV2 simgrid::smpi::Coll_bcast_scatter_LR_allgather::bcast -#define MPIR_Bcast_scatter_doubling_allgather_MV2 simgrid::smpi::Coll_bcast_scatter_rdb_allgather::bcast -#define MPIR_Bcast_scatter_ring_allgather_MV2 simgrid::smpi::Coll_bcast_scatter_LR_allgather::bcast -#define MPIR_Shmem_Bcast_MV2 simgrid::smpi::Coll_bcast_mpich::bcast -#define MPIR_Bcast_tune_inter_node_helper_MV2 simgrid::smpi::Coll_bcast_mvapich2_inter_node::bcast -#define MPIR_Bcast_inter_node_helper_MV2 simgrid::smpi::Coll_bcast_mvapich2_inter_node::bcast -#define MPIR_Knomial_Bcast_intra_node_MV2 simgrid::smpi::Coll_bcast_mvapich2_knomial_intra_node::bcast -#define MPIR_Bcast_intra_MV2 simgrid::smpi::Coll_bcast_mvapich2_intra_node::bcast +#define MPIR_Pipelined_Bcast_Zcpy_MV2 simgrid::smpi::bcast__mpich +#define MPIR_Pipelined_Bcast_MV2 simgrid::smpi::bcast__mpich +#define MPIR_Bcast_binomial_MV2 simgrid::smpi::bcast__binomial_tree +#define MPIR_Bcast_scatter_ring_allgather_shm_MV2 simgrid::smpi::bcast__scatter_LR_allgather +#define MPIR_Bcast_scatter_doubling_allgather_MV2 simgrid::smpi::bcast__scatter_rdb_allgather +#define MPIR_Bcast_scatter_ring_allgather_MV2 simgrid::smpi::bcast__scatter_LR_allgather +#define MPIR_Shmem_Bcast_MV2 simgrid::smpi::bcast__mpich +#define MPIR_Bcast_tune_inter_node_helper_MV2 simgrid::smpi::bcast__mvapich2_inter_node +#define MPIR_Bcast_inter_node_helper_MV2 simgrid::smpi::bcast__mvapich2_inter_node +#define MPIR_Knomial_Bcast_intra_node_MV2 simgrid::smpi::bcast__mvapich2_knomial_intra_node +#define MPIR_Bcast_intra_MV2 simgrid::smpi::bcast__mvapich2_intra_node static void init_mv2_bcast_tables_stampede() { // Stampede, - if (simgrid::smpi::Colls::smpi_coll_cleanup_callback == NULL) - simgrid::smpi::Colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; + if (simgrid::smpi::colls::smpi_coll_cleanup_callback == NULL) + simgrid::smpi::colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; mv2_size_bcast_tuning_table = 8; - mv2_bcast_thresholds_table = - static_cast(xbt_malloc(mv2_size_bcast_tuning_table * sizeof(mv2_bcast_tuning_table))); + mv2_bcast_thresholds_table = new mv2_bcast_tuning_table[mv2_size_bcast_tuning_table]; mv2_bcast_tuning_table mv2_tmp_bcast_thresholds_table[] = { {16, 8192, 4, 4, - {1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1}, + {true, true, true, true, true, true, true, true, true, true, true}, 11, {{0, 8, &MPIR_Pipelined_Bcast_Zcpy_MV2, 2}, {8, 16, &MPIR_Pipelined_Bcast_Zcpy_MV2, 4}, @@ -1024,7 +1009,7 @@ static void init_mv2_bcast_tables_stampede() 8192, 4, 4, - {1, 1, 1, 1, 1, 1, 1, 1}, + {true, true, true, true, true, true, true, true}, 8, {{0, 128, &MPIR_Pipelined_Bcast_Zcpy_MV2, 2}, {128, 256, &MPIR_Pipelined_Bcast_Zcpy_MV2, 4}, @@ -1047,7 +1032,7 @@ static void init_mv2_bcast_tables_stampede() 8192, 4, 4, - {1, 1, 1, 1, 1, 1, 1, 1, 1}, + {true, true, true, true, true, true, true, true, true}, 9, {{0, 2, &MPIR_Pipelined_Bcast_Zcpy_MV2, 4}, {2, 4, &MPIR_Pipelined_Bcast_Zcpy_MV2, 8}, @@ -1072,7 +1057,7 @@ static void init_mv2_bcast_tables_stampede() 8192, 4, 4, - {1, 1, 1, 0}, + {true, true, true, false}, 4, {{0, 8192, &MPIR_Pipelined_Bcast_Zcpy_MV2, 8}, {8192, 16384, &MPIR_Pipelined_Bcast_Zcpy_MV2, 4}, @@ -1087,7 +1072,7 @@ static void init_mv2_bcast_tables_stampede() 8192, 4, 4, - {1, 1, 1, 1, 1}, + {true, true, true, true, true}, 5, {{0, 16384, &MPIR_Pipelined_Bcast_Zcpy_MV2, 4}, {16384, 131072, &MPIR_Pipelined_Bcast_Zcpy_MV2, 2}, @@ -1104,7 +1089,7 @@ static void init_mv2_bcast_tables_stampede() 8192, 4, 4, - {1, 1, 1, 1, 1}, + {true, true, true, true, true}, 5, {{0, 4096, &MPIR_Pipelined_Bcast_Zcpy_MV2, 8}, {4096, 16384, &MPIR_Pipelined_Bcast_Zcpy_MV2, 4}, @@ -1121,7 +1106,7 @@ static void init_mv2_bcast_tables_stampede() 8192, 4, 4, - {1, 1, 1, 1, 1}, + {true, true, true, true, true}, 5, {{0, 8192, &MPIR_Pipelined_Bcast_Zcpy_MV2, 8}, {8192, 16384, &MPIR_Pipelined_Bcast_Zcpy_MV2, 4}, @@ -1138,7 +1123,7 @@ static void init_mv2_bcast_tables_stampede() 8192, 4, 4, - {1, 1, 1, 1, 1, 1, 1}, + {true, true, true, true, true, true, true}, 7, {{0, 16, &MPIR_Pipelined_Bcast_Zcpy_MV2, 8}, {16, 32, &MPIR_Pipelined_Bcast_Zcpy_MV2, 4}, @@ -1156,8 +1141,7 @@ static void init_mv2_bcast_tables_stampede() {32768, 524288, &MPIR_Shmem_Bcast_MV2, -1}, {524288, -1, &MPIR_Shmem_Bcast_MV2, -1}}}}; - memcpy(mv2_bcast_thresholds_table, mv2_tmp_bcast_thresholds_table, - mv2_size_bcast_tuning_table * sizeof(mv2_bcast_tuning_table)); + std::copy_n(mv2_tmp_bcast_thresholds_table, mv2_size_bcast_tuning_table, mv2_bcast_thresholds_table); } /************ Reduce variables and initializers */ @@ -1165,7 +1149,7 @@ static void init_mv2_bcast_tables_stampede() struct mv2_reduce_tuning_element { int min; int max; - int (*MV2_pt_Reduce_function)(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, + int (*MV2_pt_Reduce_function)(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm_ptr); }; @@ -1173,7 +1157,7 @@ struct mv2_reduce_tuning_table { int numproc; int inter_k_degree; int intra_k_degree; - int is_two_level_reduce[MV2_MAX_NB_THRESHOLDS]; + bool is_two_level_reduce[MV2_MAX_NB_THRESHOLDS]; int size_inter_table; mv2_reduce_tuning_element inter_leader[MV2_MAX_NB_THRESHOLDS]; int size_intra_table; @@ -1186,33 +1170,32 @@ mv2_reduce_tuning_table* mv2_reduce_thresholds_table = NULL; int mv2_reduce_intra_knomial_factor = -1; int mv2_reduce_inter_knomial_factor = -1; -int (*MV2_Reduce_function)(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, +int (*MV2_Reduce_function)(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm_ptr) = NULL; -int (*MV2_Reduce_intra_function)(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, +int (*MV2_Reduce_intra_function)(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm_ptr) = NULL; -#define MPIR_Reduce_inter_knomial_wrapper_MV2 simgrid::smpi::Coll_reduce_mvapich2_knomial::reduce -#define MPIR_Reduce_intra_knomial_wrapper_MV2 simgrid::smpi::Coll_reduce_mvapich2_knomial::reduce -#define MPIR_Reduce_binomial_MV2 simgrid::smpi::Coll_reduce_binomial::reduce -#define MPIR_Reduce_redscat_gather_MV2 simgrid::smpi::Coll_reduce_scatter_gather::reduce -#define MPIR_Reduce_shmem_MV2 simgrid::smpi::Coll_reduce_ompi_basic_linear::reduce -#define MPIR_Reduce_two_level_helper_MV2 simgrid::smpi::Coll_reduce_mvapich2_two_level::reduce +#define MPIR_Reduce_inter_knomial_wrapper_MV2 simgrid::smpi::reduce__mvapich2_knomial +#define MPIR_Reduce_intra_knomial_wrapper_MV2 simgrid::smpi::reduce__mvapich2_knomial +#define MPIR_Reduce_binomial_MV2 simgrid::smpi::reduce__binomial +#define MPIR_Reduce_redscat_gather_MV2 simgrid::smpi::reduce__scatter_gather +#define MPIR_Reduce_shmem_MV2 simgrid::smpi::reduce__ompi_basic_linear +#define MPIR_Reduce_two_level_helper_MV2 simgrid::smpi::reduce__mvapich2_two_level static void init_mv2_reduce_tables_stampede() { - if (simgrid::smpi::Colls::smpi_coll_cleanup_callback == NULL) - simgrid::smpi::Colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; + if (simgrid::smpi::colls::smpi_coll_cleanup_callback == NULL) + simgrid::smpi::colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; /*Stampede*/ mv2_size_reduce_tuning_table = 8; - mv2_reduce_thresholds_table = - static_cast(xbt_malloc(mv2_size_reduce_tuning_table * sizeof(mv2_reduce_tuning_table))); + mv2_reduce_thresholds_table = new mv2_reduce_tuning_table[mv2_size_reduce_tuning_table]; mv2_reduce_tuning_table mv2_tmp_reduce_thresholds_table[] = { { 16, 4, 4, - {1, 0, 0}, + {true, false, false}, 3, { {0, 262144, &MPIR_Reduce_inter_knomial_wrapper_MV2}, @@ -1228,7 +1211,7 @@ static void init_mv2_reduce_tables_stampede() 32, 4, 4, - {1, 1, 1, 1, 0, 0, 0}, + {true, true, true, true, false, false, false}, 7, { {0, 8192, &MPIR_Reduce_inter_knomial_wrapper_MV2}, @@ -1253,7 +1236,7 @@ static void init_mv2_reduce_tables_stampede() 64, 4, 4, - {1, 1, 1, 1, 0}, + {true, true, true, true, false}, 5, { {0, 8192, &MPIR_Reduce_inter_knomial_wrapper_MV2}, @@ -1275,7 +1258,7 @@ static void init_mv2_reduce_tables_stampede() 128, 4, 4, - {1, 0, 1, 0, 1, 0}, + {true, false, true, false, true, false}, 6, { {0, 8192, &MPIR_Reduce_inter_knomial_wrapper_MV2}, @@ -1298,7 +1281,7 @@ static void init_mv2_reduce_tables_stampede() 256, 4, 4, - {1, 1, 1, 0, 1, 1, 0}, + {true, true, true, false, true, true, false}, 7, { {0, 8192, &MPIR_Reduce_inter_knomial_wrapper_MV2}, @@ -1323,7 +1306,7 @@ static void init_mv2_reduce_tables_stampede() 512, 4, 4, - {1, 0, 1, 1, 1, 0}, + {true, false, true, true, true, false}, 6, { {0, 8192, &MPIR_Reduce_inter_knomial_wrapper_MV2}, @@ -1346,7 +1329,7 @@ static void init_mv2_reduce_tables_stampede() 1024, 4, 4, - {1, 0, 1, 1, 1}, + {true, false, true, true, true}, 5, { {0, 8192, &MPIR_Reduce_inter_knomial_wrapper_MV2}, @@ -1368,7 +1351,7 @@ static void init_mv2_reduce_tables_stampede() 2048, 4, 4, - {1, 0, 1, 1, 1, 1}, + {true, false, true, true, true, true}, 6, { {0, 2048, &MPIR_Reduce_inter_knomial_wrapper_MV2}, @@ -1390,8 +1373,7 @@ static void init_mv2_reduce_tables_stampede() }, }; - memcpy(mv2_reduce_thresholds_table, mv2_tmp_reduce_thresholds_table, - mv2_size_reduce_tuning_table * sizeof(mv2_reduce_tuning_table)); + std::copy_n(mv2_tmp_reduce_thresholds_table, mv2_size_reduce_tuning_table, mv2_reduce_thresholds_table); } /************ Reduce scatter variables and initializers */ @@ -1399,7 +1381,7 @@ static void init_mv2_reduce_tables_stampede() struct mv2_red_scat_tuning_element { int min; int max; - int (*MV2_pt_Red_scat_function)(void* sendbuf, void* recvbuf, int* recvcnts, MPI_Datatype datatype, MPI_Op op, + int (*MV2_pt_Red_scat_function)(const void* sendbuf, void* recvbuf, const int* recvcnts, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm_ptr); }; @@ -1412,27 +1394,25 @@ struct mv2_red_scat_tuning_table { int mv2_size_red_scat_tuning_table = 0; mv2_red_scat_tuning_table* mv2_red_scat_thresholds_table = NULL; -int (*MV2_Red_scat_function)(void* sendbuf, void* recvbuf, int* recvcnts, MPI_Datatype datatype, MPI_Op op, +int (*MV2_Red_scat_function)(const void* sendbuf, void* recvbuf, const int* recvcnts, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm_ptr); -static int MPIR_Reduce_Scatter_Basic_MV2(void* sendbuf, void* recvbuf, int* recvcnts, MPI_Datatype datatype, MPI_Op op, +static int MPIR_Reduce_Scatter_Basic_MV2(const void* sendbuf, void* recvbuf, const int* recvcnts, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) { - simgrid::smpi::Coll_reduce_scatter_default::reduce_scatter(sendbuf, recvbuf, recvcnts, datatype, op, comm); + simgrid::smpi::reduce_scatter__default(sendbuf, recvbuf, recvcnts, datatype, op, comm); return MPI_SUCCESS; } -#define MPIR_Reduce_scatter_non_comm_MV2 simgrid::smpi::Coll_reduce_scatter_mpich_noncomm::reduce_scatter -#define MPIR_Reduce_scatter_Rec_Halving_MV2 \ - simgrid::smpi::Coll_reduce_scatter_ompi_basic_recursivehalving::reduce_scatter -#define MPIR_Reduce_scatter_Pair_Wise_MV2 simgrid::smpi::Coll_reduce_scatter_mpich_pair::reduce_scatter +#define MPIR_Reduce_scatter_non_comm_MV2 simgrid::smpi::reduce_scatter__mpich_noncomm +#define MPIR_Reduce_scatter_Rec_Halving_MV2 simgrid::smpi::reduce_scatter__ompi_basic_recursivehalving +#define MPIR_Reduce_scatter_Pair_Wise_MV2 simgrid::smpi::reduce_scatter__mpich_pair static void init_mv2_reduce_scatter_tables_stampede() { - if (simgrid::smpi::Colls::smpi_coll_cleanup_callback == NULL) - simgrid::smpi::Colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; + if (simgrid::smpi::colls::smpi_coll_cleanup_callback == NULL) + simgrid::smpi::colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; mv2_size_red_scat_tuning_table = 6; - mv2_red_scat_thresholds_table = static_cast( - xbt_malloc(mv2_size_red_scat_tuning_table * sizeof(mv2_red_scat_tuning_table))); + mv2_red_scat_thresholds_table = new mv2_red_scat_tuning_table[mv2_size_red_scat_tuning_table]; mv2_red_scat_tuning_table mv2_tmp_red_scat_thresholds_table[] = { { 16, @@ -1484,8 +1464,7 @@ static void init_mv2_reduce_scatter_tables_stampede() }, }; - memcpy(mv2_red_scat_thresholds_table, mv2_tmp_red_scat_thresholds_table, - mv2_size_red_scat_tuning_table * sizeof(mv2_red_scat_tuning_table)); + std::copy_n(mv2_tmp_red_scat_thresholds_table, mv2_size_red_scat_tuning_table, mv2_red_scat_thresholds_table); } /************ Scatter variables and initializers */ @@ -1493,7 +1472,7 @@ static void init_mv2_reduce_scatter_tables_stampede() struct mv2_scatter_tuning_element { int min; int max; - int (*MV2_pt_Scatter_function)(void* sendbuf, int sendcnt, MPI_Datatype sendtype, void* recvbuf, int recvcnt, + int (*MV2_pt_Scatter_function)(const void* sendbuf, int sendcnt, MPI_Datatype sendtype, void* recvbuf, int recvcnt, MPI_Datatype recvtype, int root, MPI_Comm comm); }; @@ -1510,40 +1489,36 @@ int mv2_scatter_num_ppn_conf = 1; int* mv2_size_scatter_tuning_table = NULL; mv2_scatter_tuning_table** mv2_scatter_thresholds_table = NULL; -int (*MV2_Scatter_function)(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, +int (*MV2_Scatter_function)(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm) = NULL; -int (*MV2_Scatter_intra_function)(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, +int (*MV2_Scatter_intra_function)(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm) = NULL; -int MPIR_Scatter_mcst_wrap_MV2(void* sendbuf, int sendcnt, MPI_Datatype sendtype, void* recvbuf, int recvcnt, +int MPIR_Scatter_mcst_wrap_MV2(const void* sendbuf, int sendcnt, MPI_Datatype sendtype, void* recvbuf, int recvcnt, MPI_Datatype recvtype, int root, MPI_Comm comm_ptr); -int MPIR_Scatter_mcst_wrap_MV2(void* sendbuf, int sendcnt, MPI_Datatype sendtype, void* recvbuf, int recvcnt, +int MPIR_Scatter_mcst_wrap_MV2(const void* sendbuf, int sendcnt, MPI_Datatype sendtype, void* recvbuf, int recvcnt, MPI_Datatype recvtype, int root, MPI_Comm comm_ptr) { return 0; } -#define MPIR_Scatter_MV2_Binomial simgrid::smpi::Coll_scatter_ompi_binomial::scatter -#define MPIR_Scatter_MV2_Direct simgrid::smpi::Coll_scatter_ompi_basic_linear::scatter -#define MPIR_Scatter_MV2_two_level_Binomial simgrid::smpi::Coll_scatter_mvapich2_two_level_binomial::scatter -#define MPIR_Scatter_MV2_two_level_Direct simgrid::smpi::Coll_scatter_mvapich2_two_level_direct::scatter +#define MPIR_Scatter_MV2_Binomial simgrid::smpi::scatter__ompi_binomial +#define MPIR_Scatter_MV2_Direct simgrid::smpi::scatter__ompi_basic_linear +#define MPIR_Scatter_MV2_two_level_Binomial simgrid::smpi::scatter__mvapich2_two_level_binomial +#define MPIR_Scatter_MV2_two_level_Direct simgrid::smpi::scatter__mvapich2_two_level_direct static void init_mv2_scatter_tables_stampede() { - if (simgrid::smpi::Colls::smpi_coll_cleanup_callback == NULL) - simgrid::smpi::Colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; + if (simgrid::smpi::colls::smpi_coll_cleanup_callback == NULL) + simgrid::smpi::colls::smpi_coll_cleanup_callback = &smpi_coll_cleanup_mvapich2; int agg_table_sum = 0; - int i; - mv2_scatter_tuning_table** table_ptrs = NULL; mv2_scatter_num_ppn_conf = 3; - mv2_scatter_thresholds_table = - static_cast(xbt_malloc(sizeof(mv2_scatter_tuning_table*) * mv2_scatter_num_ppn_conf)); - table_ptrs = - static_cast(xbt_malloc(sizeof(mv2_scatter_tuning_table*) * mv2_scatter_num_ppn_conf)); - mv2_size_scatter_tuning_table = static_cast(xbt_malloc(sizeof(int) * mv2_scatter_num_ppn_conf)); - mv2_scatter_table_ppn_conf = static_cast(xbt_malloc(mv2_scatter_num_ppn_conf * sizeof(int))); + mv2_scatter_thresholds_table = new mv2_scatter_tuning_table*[mv2_scatter_num_ppn_conf]; + mv2_scatter_tuning_table** table_ptrs = new mv2_scatter_tuning_table*[mv2_scatter_num_ppn_conf]; + mv2_size_scatter_tuning_table = new int[mv2_scatter_num_ppn_conf]; + mv2_scatter_table_ppn_conf = new int[mv2_scatter_num_ppn_conf]; mv2_scatter_table_ppn_conf[0] = 1; mv2_size_scatter_tuning_table[0] = 6; mv2_scatter_tuning_table mv2_tmp_scatter_thresholds_table_1ppn[] = { @@ -1822,17 +1797,16 @@ static void init_mv2_scatter_tables_stampede() }; table_ptrs[2] = mv2_tmp_scatter_thresholds_table_16ppn; agg_table_sum = 0; - for (i = 0; i < mv2_scatter_num_ppn_conf; i++) { + for (int i = 0; i < mv2_scatter_num_ppn_conf; i++) { agg_table_sum += mv2_size_scatter_tuning_table[i]; } - mv2_scatter_thresholds_table[0] = - static_cast(xbt_malloc(agg_table_sum * sizeof(mv2_scatter_tuning_table))); - memcpy(mv2_scatter_thresholds_table[0], table_ptrs[0], - (sizeof(mv2_scatter_tuning_table) * mv2_size_scatter_tuning_table[0])); - for (i = 1; i < mv2_scatter_num_ppn_conf; i++) { + mv2_scatter_thresholds_table[0] = new mv2_scatter_tuning_table[agg_table_sum]; + std::copy_n(table_ptrs[0], mv2_size_scatter_tuning_table[0], mv2_scatter_thresholds_table[0]); + for (int i = 1; i < mv2_scatter_num_ppn_conf; i++) { mv2_scatter_thresholds_table[i] = mv2_scatter_thresholds_table[i - 1] + mv2_size_scatter_tuning_table[i - 1]; - memcpy(mv2_scatter_thresholds_table[i], table_ptrs[i], - (sizeof(mv2_scatter_tuning_table) * mv2_size_scatter_tuning_table[i])); + std::copy_n(table_ptrs[i], mv2_size_scatter_tuning_table[i], mv2_scatter_thresholds_table[i]); } - xbt_free(table_ptrs); + delete[] table_ptrs; } + +#endif