X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/e94c2a7fc81a82998524aa55db075be69990d4ea..98bc4480ae41eb74cdc7f10a8223c92866487a2c:/src/smpi/include/smpi_coll.hpp diff --git a/src/smpi/include/smpi_coll.hpp b/src/smpi/include/smpi_coll.hpp index d5fcef4ceb..30f0093fcc 100644 --- a/src/smpi/include/smpi_coll.hpp +++ b/src/smpi/include/smpi_coll.hpp @@ -1,5 +1,5 @@ /*High level handling of collective algorithms*/ -/* Copyright (c) 2009-2018. The SimGrid Team. +/* Copyright (c) 2009-2019. The SimGrid Team. * All rights reserved. */ /* This program is free software; you can redistribute it and/or modify it @@ -14,66 +14,54 @@ /** @brief MPI collective description */ #define COLL_DEFS(cat, ret, args, args2) \ - static void set_##cat(std::string name); \ - static s_mpi_coll_description_t mpi_coll_##cat##_description[]; \ - static int(*cat) args; + void _XBT_CONCAT(set_, cat)(const std::string& name); \ + extern int(*cat) args; -#define COLL_SIG(cat, ret, args, args2)\ - static int cat args; - - -#define COLL_DESCRIPTION(cat, ret, args, name) \ - {# name,\ - # cat " " # name " collective",\ - (void*) Coll_ ## cat ## _ ## name::cat } - -#define COLL_PROTO(cat, ret, args, name) \ -class Coll_ ## cat ## _ ## name : public Coll { \ -public: \ -static ret cat (COLL_UNPAREN args); \ -}; +#define COLL_DESCRIPTION(cat, ret, args, name) \ + { \ + _XBT_STRINGIFY(name) \ + , _XBT_STRINGIFY(cat) " " _XBT_STRINGIFY(name) " collective", (void*)_XBT_CONCAT3(cat, __, name) \ + } #define COLL_UNPAREN(...) __VA_ARGS__ #define COLL_APPLY(action, sig, name) action(sig, name) #define COLL_COMMA , -#define COLL_NOsep -#define COLL_NOTHING(...) #define COLL_GATHER_SIG gather, int, \ - (void *send_buff, int send_count, MPI_Datatype send_type, \ + (const void *send_buff, int send_count, MPI_Datatype send_type, \ void *recv_buff, int recv_count, MPI_Datatype recv_type, \ int root, MPI_Comm comm) #define COLL_ALLGATHER_SIG allgather, int, \ - (void *send_buff, int send_count, MPI_Datatype send_type, \ + (const void *send_buff, int send_count, MPI_Datatype send_type, \ void *recv_buff, int recv_count, MPI_Datatype recv_type, \ MPI_Comm comm) #define COLL_ALLGATHERV_SIG allgatherv, int, \ - (void *send_buff, int send_count, MPI_Datatype send_type, \ - void *recv_buff, int *recv_count, int *recv_disps, \ + (const void *send_buff, int send_count, MPI_Datatype send_type, \ + void *recv_buff, const int *recv_count, const int *recv_disps, \ MPI_Datatype recv_type, MPI_Comm comm) #define COLL_ALLTOALL_SIG alltoall, int, \ - (void *send_buff, int send_count, MPI_Datatype send_type, \ + (const void *send_buff, int send_count, MPI_Datatype send_type, \ void *recv_buff, int recv_count, MPI_Datatype recv_type, \ MPI_Comm comm) #define COLL_ALLTOALLV_SIG alltoallv, int, \ - (void *send_buff, int *send_counts, int *send_disps, MPI_Datatype send_type, \ - void *recv_buff, int *recv_counts, int *recv_disps, MPI_Datatype recv_type, \ + (const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, \ + void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, \ MPI_Comm comm) #define COLL_BCAST_SIG bcast, int, \ (void *buf, int count, MPI_Datatype datatype, \ int root, MPI_Comm comm) #define COLL_REDUCE_SIG reduce, int, \ - (void *buf, void *rbuf, int count, MPI_Datatype datatype, \ + (const void *buf, void *rbuf, int count, MPI_Datatype datatype, \ MPI_Op op, int root, MPI_Comm comm) #define COLL_ALLREDUCE_SIG allreduce, int, \ - (void *sbuf, void *rbuf, int rcount, \ + (const void *sbuf, void *rbuf, int rcount, \ MPI_Datatype dtype, MPI_Op op, MPI_Comm comm) #define COLL_REDUCE_SCATTER_SIG reduce_scatter, int, \ - (void *sbuf, void *rbuf, int *rcounts,\ + (const void *sbuf, void *rbuf, const int *rcounts,\ MPI_Datatype dtype,MPI_Op op,MPI_Comm comm) #define COLL_SCATTER_SIG scatter, int, \ - (void *sendbuf, int sendcount, MPI_Datatype sendtype,\ + (const void *sendbuf, int sendcount, MPI_Datatype sendtype,\ void *recvbuf, int recvcount, MPI_Datatype recvtype,\ int root, MPI_Comm comm) #define COLL_BARRIER_SIG barrier, int, \ @@ -88,57 +76,93 @@ struct s_mpi_coll_description_t { void *coll; }; -class Colls{ - public: - static XBT_PUBLIC void coll_help(const char* category, s_mpi_coll_description_t* table); - static XBT_PUBLIC int find_coll_description(s_mpi_coll_description_t* table, std::string name, const char* desc); - static void set_collectives(); - - // for each collective type, create the set_* prototype, the description array and the function pointer - COLL_APPLY(COLL_DEFS, COLL_GATHER_SIG, ""); - COLL_APPLY(COLL_DEFS, COLL_ALLGATHER_SIG, ""); - COLL_APPLY(COLL_DEFS, COLL_ALLGATHERV_SIG, ""); - COLL_APPLY(COLL_DEFS, COLL_REDUCE_SIG, ""); - COLL_APPLY(COLL_DEFS, COLL_ALLREDUCE_SIG, ""); - COLL_APPLY(COLL_DEFS, COLL_REDUCE_SCATTER_SIG, ""); - COLL_APPLY(COLL_DEFS, COLL_SCATTER_SIG, ""); - COLL_APPLY(COLL_DEFS, COLL_BARRIER_SIG, ""); - COLL_APPLY(COLL_DEFS, COLL_BCAST_SIG, ""); - COLL_APPLY(COLL_DEFS, COLL_ALLTOALL_SIG, ""); - COLL_APPLY(COLL_DEFS, COLL_ALLTOALLV_SIG, ""); - - // These fairly unused collectives only have one implementation in SMPI - static int gatherv(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int* recvcounts, int* displs, - MPI_Datatype recvtype, int root, MPI_Comm comm); - static int scatterv(void* sendbuf, int* sendcounts, int* displs, MPI_Datatype sendtype, void* recvbuf, - int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm); - static int scan(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm); - static int exscan(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm); - - static void (*smpi_coll_cleanup_callback)(); -}; - -class Coll { - public: - // for each collective type, create a function member - COLL_APPLY(COLL_SIG,COLL_GATHER_SIG,""); - COLL_APPLY(COLL_SIG,COLL_ALLGATHER_SIG,""); - COLL_APPLY(COLL_SIG,COLL_ALLGATHERV_SIG,""); - COLL_APPLY(COLL_SIG,COLL_REDUCE_SIG,""); - COLL_APPLY(COLL_SIG,COLL_ALLREDUCE_SIG,""); - COLL_APPLY(COLL_SIG,COLL_REDUCE_SCATTER_SIG,""); - COLL_APPLY(COLL_SIG,COLL_SCATTER_SIG,""); - COLL_APPLY(COLL_SIG,COLL_BARRIER_SIG,""); - COLL_APPLY(COLL_SIG,COLL_BCAST_SIG,""); - COLL_APPLY(COLL_SIG,COLL_ALLTOALL_SIG,""); - COLL_APPLY(COLL_SIG,COLL_ALLTOALLV_SIG,""); +namespace colls { +XBT_PUBLIC void coll_help(const char* category, s_mpi_coll_description_t* table); +XBT_PUBLIC int find_coll_description(s_mpi_coll_description_t* table, const std::string& name, const char* desc); +void set_collectives(); +XBT_PUBLIC s_mpi_coll_description_t* get_smpi_coll_description(const char* name, int rank); + +// for each collective type, create the set_* prototype and the function pointer +// void set_gather(const std::string& name); +// extern int(*gather)(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, +// MPI_Datatype recv_type, int root, MPI_Comm comm); +COLL_APPLY(COLL_DEFS, COLL_GATHER_SIG, "") +COLL_APPLY(COLL_DEFS, COLL_ALLGATHER_SIG, "") +COLL_APPLY(COLL_DEFS, COLL_ALLGATHERV_SIG, "") +COLL_APPLY(COLL_DEFS, COLL_REDUCE_SIG, "") +COLL_APPLY(COLL_DEFS, COLL_ALLREDUCE_SIG, "") +COLL_APPLY(COLL_DEFS, COLL_REDUCE_SCATTER_SIG, "") +COLL_APPLY(COLL_DEFS, COLL_SCATTER_SIG, "") +COLL_APPLY(COLL_DEFS, COLL_BARRIER_SIG, "") +COLL_APPLY(COLL_DEFS, COLL_BCAST_SIG, "") +COLL_APPLY(COLL_DEFS, COLL_ALLTOALL_SIG, "") +COLL_APPLY(COLL_DEFS, COLL_ALLTOALLV_SIG, "") + +// These fairly unused collectives only have one implementation in SMPI +int gatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int* recvcounts, + const int* displs, MPI_Datatype recvtype, int root, MPI_Comm comm); +int scatterv(const void* sendbuf, const int* sendcounts, const int* displs, MPI_Datatype sendtype, void* recvbuf, + int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm); +int scan(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm); +int exscan(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm); +int alltoallw(const void* sendbuf, const int* sendcounts, const int* senddisps, const MPI_Datatype* sendtypes, + void* recvbuf, const int* recvcounts, const int* recvdisps, const MPI_Datatype* recvtypes, MPI_Comm comm); + +// async collectives +int ibarrier(MPI_Comm comm, MPI_Request* request, int external = 1); +int ibcast(void* buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm, MPI_Request* request, + int external = 1); +int igather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, + MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request, int external = 1); +int igatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int* recvcounts, + const int* displs, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request, int external = 1); +int iallgather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, + MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request, int external = 1); +int iallgatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, const int* recvcounts, + const int* displs, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request, int external = 1); +int iscatter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, + MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request, int external = 1); +int iscatterv(const void* sendbuf, const int* sendcounts, const int* displs, MPI_Datatype sendtype, void* recvbuf, + int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm, MPI_Request* request, int external = 1); +int ireduce(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm, + MPI_Request* request, int external = 1); +int iallreduce(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, + MPI_Request* request, int external = 1); +int iscan(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, + MPI_Request* request, int external = 1); +int iexscan(const void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm, + MPI_Request* request, int external = 1); +int ireduce_scatter(const void* sendbuf, void* recvbuf, const int* recvcounts, MPI_Datatype datatype, MPI_Op op, + MPI_Comm comm, MPI_Request* request, int external = 1); +int ireduce_scatter_block(const void* sendbuf, void* recvbuf, int recvcount, MPI_Datatype datatype, MPI_Op op, + MPI_Comm comm, MPI_Request* request, int external = 1); +int ialltoall(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, + MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request, int external = 1); +int ialltoallv(const void* sendbuf, const int* sendcounts, const int* senddisps, MPI_Datatype sendtype, void* recvbuf, + const int* recvcounts, const int* recvdisps, MPI_Datatype recvtype, MPI_Comm comm, MPI_Request* request, + int external = 1); +int ialltoallw(const void* sendbuf, const int* sendcounts, const int* senddisps, const MPI_Datatype* sendtypes, + void* recvbuf, const int* recvcounts, const int* recvdisps, const MPI_Datatype* recvtypes, MPI_Comm comm, + MPI_Request* request, int external = 1); + +extern void (*smpi_coll_cleanup_callback)(); }; - /************* * GATHER * *************/ +int gather__default(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, int root, MPI_Comm comm); +int gather__ompi(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, int root, MPI_Comm comm); +int gather__ompi_basic_linear(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, int root, MPI_Comm comm); +int gather__ompi_binomial(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, int root, MPI_Comm comm); +int gather__ompi_linear_sync(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, int root, MPI_Comm comm); +int gather__mpich(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, int root, MPI_Comm comm); +int gather__mvapich2(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, int root, MPI_Comm comm); +int gather__mvapich2_two_level(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, int root, MPI_Comm comm); +int gather__impi(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, int root, MPI_Comm comm); +int gather__automatic(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, int root, MPI_Comm comm); + #define COLL_GATHERS(action, COLL_sep) \ COLL_APPLY(action, COLL_GATHER_SIG, default) COLL_sep \ COLL_APPLY(action, COLL_GATHER_SIG, ompi) COLL_sep \ @@ -151,11 +175,31 @@ COLL_APPLY(action, COLL_GATHER_SIG, mvapich2_two_level) COLL_sep \ COLL_APPLY(action, COLL_GATHER_SIG, impi) COLL_sep \ COLL_APPLY(action, COLL_GATHER_SIG, automatic) -COLL_GATHERS(COLL_PROTO, COLL_NOsep) - /************* * ALLGATHER * *************/ +int allgather__default(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__2dmesh(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__3dmesh(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__bruck(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__GB(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__loosely_lr(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__NTSLR(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__NTSLR_NB(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__pair(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__rdb(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__rhv(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__ring(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__SMP_NTS(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__smp_simple(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__spreading_simple(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__ompi(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__ompi_neighborexchange (const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__mvapich2(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__mvapich2_smp(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__mpich(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__impi(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int allgather__automatic(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); #define COLL_ALLGATHERS(action, COLL_sep) \ COLL_APPLY(action, COLL_ALLGATHER_SIG, default) COLL_sep \ @@ -181,8 +225,6 @@ COLL_APPLY(action, COLL_ALLGATHER_SIG, mpich) COLL_sep \ COLL_APPLY(action, COLL_ALLGATHER_SIG, impi) COLL_sep \ COLL_APPLY(action, COLL_ALLGATHER_SIG, automatic) -COLL_ALLGATHERS(COLL_PROTO, COLL_NOsep) - /************** * ALLGATHERV * **************/ @@ -202,7 +244,19 @@ COLL_APPLY(action, COLL_ALLGATHERV_SIG, mvapich2) COLL_sep \ COLL_APPLY(action, COLL_ALLGATHERV_SIG, impi) COLL_sep \ COLL_APPLY(action, COLL_ALLGATHERV_SIG, automatic) -COLL_ALLGATHERVS(COLL_PROTO, COLL_NOsep) +int allgatherv__default(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, const int *recv_count, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int allgatherv__GB(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, const int *recv_count, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int allgatherv__pair(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, const int *recv_count, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int allgatherv__ring(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, const int *recv_count, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int allgatherv__ompi(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, const int *recv_count, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int allgatherv__ompi_neighborexchange(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, const int *recv_count, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int allgatherv__ompi_bruck (const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, const int *recv_count, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int allgatherv__mpich(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, const int *recv_count, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int allgatherv__mpich_rdb(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, const int *recv_count, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int allgatherv__mpich_ring(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, const int *recv_count, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int allgatherv__mvapich2(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, const int *recv_count, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int allgatherv__impi(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, const int *recv_count, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int allgatherv__automatic(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, const int *recv_count, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); /************* * ALLREDUCE * @@ -232,7 +286,28 @@ COLL_APPLY(action, COLL_ALLREDUCE_SIG, impi) COLL_sep \ COLL_APPLY(action, COLL_ALLREDUCE_SIG, rab) COLL_sep \ COLL_APPLY(action, COLL_ALLREDUCE_SIG, automatic) -COLL_ALLREDUCES(COLL_PROTO, COLL_NOsep) +int allreduce__default(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__lr(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__rab1(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__rab2(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__rab_rdb(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__rdb(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__smp_binomial(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__smp_binomial_pipeline(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__smp_rdb(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__smp_rsag(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__smp_rsag_lr(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__smp_rsag_rab(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__redbcast(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__ompi(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__ompi_ring_segmented(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__mpich(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__mvapich2(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__mvapich2_rs(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__mvapich2_two_level(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__impi(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__rab(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); +int allreduce__automatic(const void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm); /************ * ALLTOALL * @@ -261,7 +336,27 @@ COLL_APPLY(action, COLL_ALLTOALL_SIG, mpich) COLL_sep \ COLL_APPLY(action, COLL_ALLTOALL_SIG, impi) COLL_sep \ COLL_APPLY(action, COLL_ALLTOALL_SIG, automatic) -COLL_ALLTOALLS(COLL_PROTO, COLL_NOsep) +int alltoall__default(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__2dmesh(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__3dmesh(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__basic_linear(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__bruck(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__pair(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__pair_rma(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__pair_light_barrier(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__pair_mpi_barrier(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__pair_one_barrier(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__rdb(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__ring(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__ring_light_barrier(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__ring_mpi_barrier(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__ring_one_barrier(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__mvapich2(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__mvapich2_scatter_dest(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__ompi(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__mpich(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__impi(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); +int alltoall__automatic(const void *send_buff, int send_count, MPI_Datatype send_type, void *recv_buff, int recv_count, MPI_Datatype recv_type, MPI_Comm comm); /************* * ALLTOALLV * @@ -285,7 +380,22 @@ COLL_APPLY(action, COLL_ALLTOALLV_SIG, mvapich2) COLL_sep \ COLL_APPLY(action, COLL_ALLTOALLV_SIG, impi) COLL_sep \ COLL_APPLY(action, COLL_ALLTOALLV_SIG, automatic) -COLL_ALLTOALLVS(COLL_PROTO, COLL_NOsep) +int alltoallv__default(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__bruck(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__pair(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__pair_light_barrier(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__pair_mpi_barrier(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__pair_one_barrier(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__ring(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__ring_light_barrier(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__ring_mpi_barrier(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__ring_one_barrier(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__ompi(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__mpich(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__ompi_basic_linear(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__mvapich2(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__impi(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); +int alltoallv__automatic(const void *send_buff, const int *send_counts, const int *send_disps, MPI_Datatype send_type, void *recv_buff, const int *recv_counts, const int *recv_disps, MPI_Datatype recv_type, MPI_Comm comm); /********* * BCAST * @@ -318,7 +428,31 @@ COLL_APPLY(action, COLL_BCAST_SIG, mvapich2_knomial_intra_node) COLL_sep \ COLL_APPLY(action, COLL_BCAST_SIG, impi) COLL_sep \ COLL_APPLY(action, COLL_BCAST_SIG, automatic) -COLL_BCASTS(COLL_PROTO, COLL_NOsep) +int bcast__default(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__arrival_pattern_aware(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__arrival_pattern_aware_wait(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__arrival_scatter(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__binomial_tree(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__flattree(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__flattree_pipeline(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__NTSB(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__NTSL(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__NTSL_Isend(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__scatter_LR_allgather(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__scatter_rdb_allgather(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__SMP_binary(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__SMP_binomial(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__SMP_linear(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__ompi(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__ompi_split_bintree(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__ompi_pipeline(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__mpich(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__mvapich2(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__mvapich2_inter_node(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__mvapich2_intra_node(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__mvapich2_knomial_intra_node(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__impi(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); +int bcast__automatic(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm); /********** * REDUCE * @@ -346,7 +480,26 @@ COLL_APPLY(action, COLL_REDUCE_SIG, impi) COLL_sep \ COLL_APPLY(action, COLL_REDUCE_SIG, rab) COLL_sep \ COLL_APPLY(action, COLL_REDUCE_SIG, automatic) -COLL_REDUCES(COLL_PROTO, COLL_NOsep) +int reduce__default(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__arrival_pattern_aware(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__binomial(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__flat_tree(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__NTSL(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__scatter_gather(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__ompi(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__ompi_chain(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__ompi_pipeline(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__ompi_basic_linear(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__ompi_in_order_binary(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__ompi_binary(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__ompi_binomial(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__mpich(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__mvapich2(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__mvapich2_knomial(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__mvapich2_two_level(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__impi(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__rab(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); +int reduce__automatic(const void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm); /************* * REDUCE_SCATTER * @@ -365,7 +518,17 @@ COLL_APPLY(action, COLL_REDUCE_SCATTER_SIG, mvapich2) COLL_sep \ COLL_APPLY(action, COLL_REDUCE_SCATTER_SIG, impi) COLL_sep \ COLL_APPLY(action, COLL_REDUCE_SCATTER_SIG, automatic) -COLL_REDUCE_SCATTERS(COLL_PROTO, COLL_NOsep) +int reduce_scatter__default(const void *sbuf, void *rbuf, const int *rcounts, MPI_Datatype dtype,MPI_Op op,MPI_Comm comm); +int reduce_scatter__ompi(const void *sbuf, void *rbuf, const int *rcounts, MPI_Datatype dtype,MPI_Op op,MPI_Comm comm); +int reduce_scatter__ompi_basic_recursivehalving(const void *sbuf, void *rbuf, const int *rcounts, MPI_Datatype dtype,MPI_Op op,MPI_Comm comm); +int reduce_scatter__ompi_ring(const void *sbuf, void *rbuf, const int *rcounts, MPI_Datatype dtype,MPI_Op op,MPI_Comm comm); +int reduce_scatter__mpich(const void *sbuf, void *rbuf, const int *rcounts, MPI_Datatype dtype,MPI_Op op,MPI_Comm comm); +int reduce_scatter__mpich_pair(const void *sbuf, void *rbuf, const int *rcounts, MPI_Datatype dtype,MPI_Op op,MPI_Comm comm); +int reduce_scatter__mpich_rdb(const void *sbuf, void *rbuf, const int *rcounts, MPI_Datatype dtype,MPI_Op op,MPI_Comm comm); +int reduce_scatter__mpich_noncomm(const void *sbuf, void *rbuf, const int *rcounts, MPI_Datatype dtype,MPI_Op op,MPI_Comm comm); +int reduce_scatter__mvapich2(const void *sbuf, void *rbuf, const int *rcounts, MPI_Datatype dtype,MPI_Op op,MPI_Comm comm); +int reduce_scatter__impi(const void *sbuf, void *rbuf, const int *rcounts, MPI_Datatype dtype,MPI_Op op,MPI_Comm comm); +int reduce_scatter__automatic(const void *sbuf, void *rbuf, const int *rcounts, MPI_Datatype dtype,MPI_Op op,MPI_Comm comm); /************* * SCATTER * @@ -383,7 +546,16 @@ COLL_APPLY(action, COLL_SCATTER_SIG, mvapich2_two_level_direct) COLL_sep \ COLL_APPLY(action, COLL_SCATTER_SIG, impi) COLL_sep \ COLL_APPLY(action, COLL_SCATTER_SIG, automatic) -COLL_SCATTERS(COLL_PROTO, COLL_NOsep) +int scatter__default(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm); +int scatter__ompi(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm); +int scatter__ompi_basic_linear(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm); +int scatter__ompi_binomial(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm); +int scatter__mpich(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm); +int scatter__mvapich2(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm); +int scatter__mvapich2_two_level_binomial(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm); +int scatter__mvapich2_two_level_direct(const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm); +int scatter__impi (const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm); +int scatter__automatic (const void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm); /************* * BARRIER * @@ -405,7 +577,20 @@ COLL_APPLY(action, COLL_BARRIER_SIG, mvapich2) COLL_sep \ COLL_APPLY(action, COLL_BARRIER_SIG, impi) COLL_sep \ COLL_APPLY(action, COLL_BARRIER_SIG, automatic) -COLL_BARRIERS(COLL_PROTO, COLL_NOsep) +int barrier__default(MPI_Comm comm); +int barrier__ompi(MPI_Comm comm); +int barrier__ompi_basic_linear(MPI_Comm comm); +int barrier__ompi_two_procs(MPI_Comm comm); +int barrier__ompi_tree(MPI_Comm comm); +int barrier__ompi_bruck(MPI_Comm comm); +int barrier__ompi_recursivedoubling(MPI_Comm comm); +int barrier__ompi_doublering(MPI_Comm comm); +int barrier__mpich_smp(MPI_Comm comm); +int barrier__mpich(MPI_Comm comm); +int barrier__mvapich2_pair(MPI_Comm comm); +int barrier__mvapich2 (MPI_Comm comm); +int barrier__impi(MPI_Comm comm); +int barrier__automatic(MPI_Comm comm); } }