X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/8ab4885a856f21049d97a746049b71987d1c695e..93a44592f2e075710723814e39fc3ec3fe93888b:/src/smpi/smpi_mpi.c diff --git a/src/smpi/smpi_mpi.c b/src/smpi/smpi_mpi.c index 70f43ff715..f85caeed2f 100644 --- a/src/smpi/smpi_mpi.c +++ b/src/smpi/smpi_mpi.c @@ -1,295 +1,1026 @@ +/* $Id$tag */ + #include "private.h" +#include "smpi_coll_private.h" +#include "smpi_mpi_dt_private.h" + +XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_mpi, smpi, + "Logging specific to SMPI (mpi)"); -int MPI_Init(int *argc, char ***argv) -{ - smpi_mpi_init(); - smpi_bench_begin(); - return MPI_SUCCESS; +/* MPI User level calls */ + +int MPI_Init(int* argc, char*** argv) { + smpi_process_init(argc, argv); + smpi_bench_begin(MPI_COMM_NULL, NULL); + return MPI_SUCCESS; } -int MPI_Finalize() -{ - smpi_bench_end(); - smpi_mpi_finalize(); - return MPI_SUCCESS; +int MPI_Finalize(void) { + smpi_bench_end(MPI_COMM_NULL, NULL); + smpi_process_destroy(); + return MPI_SUCCESS; } -// right now this just exits the current node, should send abort signal to all -// hosts in the communicator; -int MPI_Abort(MPI_Comm comm, int errorcode) -{ - smpi_exit(errorcode); - return 0; +int MPI_Init_thread(int* argc, char*** argv, int required, int* provided) { + if(provided != NULL) { + *provided = MPI_THREAD_MULTIPLE; + } + return MPI_Init(argc, argv); } -int MPI_Comm_size(MPI_Comm comm, int *size) -{ - int retval = MPI_SUCCESS; +int MPI_Query_thread(int* provided) { + int retval; - smpi_bench_end(); + smpi_bench_end(MPI_COMM_NULL, NULL); + if(provided == NULL) { + retval = MPI_ERR_ARG; + } else { + *provided = MPI_THREAD_MULTIPLE; + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} - if (NULL == comm) { - retval = MPI_ERR_COMM; - } else if (NULL == size) { - retval = MPI_ERR_ARG; - } else { - *size = comm->size; - } +int MPI_Is_thread_main(int* flag) { + int retval; - smpi_bench_begin(); + smpi_bench_end(MPI_COMM_NULL, NULL); + if(flag == NULL) { + retval = MPI_ERR_ARG; + } else { + *flag = smpi_process_index() == 0; + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} - return retval; +int MPI_Abort(MPI_Comm comm, int errorcode) { + smpi_bench_end(comm, NULL); + smpi_process_destroy(); + // FIXME: should kill all processes in comm instead + SIMIX_process_kill(SIMIX_process_self()); + return MPI_SUCCESS; } -int MPI_Comm_rank(MPI_Comm comm, int *rank) -{ - int retval = MPI_SUCCESS; +double MPI_Wtime(void) { + double time; + + smpi_bench_end(MPI_COMM_NULL, NULL); + time = SIMIX_get_clock(); + smpi_bench_begin(MPI_COMM_NULL, NULL); + return time; +} - smpi_bench_end(); +int MPI_Type_size(MPI_Datatype datatype, size_t* size) { + int retval; - if (NULL == comm) { - retval = MPI_ERR_COMM; - } else if (NULL == rank) { - retval = MPI_ERR_ARG; - } else { - *rank = smpi_mpi_comm_rank(comm); - } + smpi_bench_end(MPI_COMM_NULL, NULL); + if(datatype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else if(size == NULL) { + retval = MPI_ERR_ARG; + } else { + *size = smpi_datatype_size(datatype); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} - smpi_bench_begin(); +int MPI_Type_get_extent(MPI_Datatype datatype, MPI_Aint* lb, MPI_Aint* extent) { + int retval; - return retval; + smpi_bench_end(MPI_COMM_NULL, NULL); + if(datatype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else if(lb == NULL || extent == NULL) { + retval = MPI_ERR_ARG; + } else { + retval = smpi_datatype_extent(datatype, lb, extent); + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; } -int MPI_Type_size(MPI_Datatype datatype, size_t *size) -{ - int retval = MPI_SUCCESS; +int MPI_Type_lb(MPI_Datatype datatype, MPI_Aint* disp) { + int retval; - smpi_bench_end(); + smpi_bench_end(MPI_COMM_NULL, NULL); + if(datatype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else if(disp == NULL) { + retval = MPI_ERR_ARG; + } else { + *disp = smpi_datatype_lb(datatype); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} + +int MPI_Type_ub(MPI_Datatype datatype, MPI_Aint* disp) { + int retval; - if (NULL == datatype) { - retval = MPI_ERR_TYPE; - } else if (NULL == size) { - retval = MPI_ERR_ARG; - } else { - *size = datatype->size; - } + smpi_bench_end(MPI_COMM_NULL, NULL); + if(datatype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else if(disp == NULL) { + retval = MPI_ERR_ARG; + } else { + *disp = smpi_datatype_ub(datatype); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} - smpi_bench_begin(); +int MPI_Op_create(MPI_User_function* function, int commute, MPI_Op* op) { + int retval; - return retval; + smpi_bench_end(MPI_COMM_NULL, NULL); + if(function == NULL || op == NULL) { + retval = MPI_ERR_ARG; + } else { + *op = smpi_op_new(function, commute); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; } -int MPI_Barrier(MPI_Comm comm) -{ - int retval = MPI_SUCCESS; +int MPI_Op_free(MPI_Op* op) { + int retval; + + smpi_bench_end(MPI_COMM_NULL, NULL); + if(op == NULL) { + retval = MPI_ERR_ARG; + } else if(*op == MPI_OP_NULL) { + retval = MPI_ERR_OP; + } else { + smpi_op_destroy(*op); + *op = MPI_OP_NULL; + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} - smpi_bench_end(); +int MPI_Group_free(MPI_Group *group) { + int retval; - if (NULL == comm) { - retval = MPI_ERR_COMM; - } else { - retval = smpi_mpi_barrier(comm); - } + smpi_bench_end(MPI_COMM_NULL, NULL); + if(group == NULL) { + retval = MPI_ERR_ARG; + } else { + smpi_group_destroy(*group); + *group = MPI_GROUP_NULL; + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} - smpi_bench_begin(); +int MPI_Group_size(MPI_Group group, int* size) { + int retval; - return retval; + smpi_bench_end(MPI_COMM_NULL, NULL); + if(group == MPI_GROUP_NULL) { + retval = MPI_ERR_GROUP; + } else if(size == NULL) { + retval = MPI_ERR_ARG; + } else { + *size = smpi_group_size(group); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; } -int MPI_Irecv(void *buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm, MPI_Request *request) -{ - int retval = MPI_SUCCESS; +int MPI_Group_rank(MPI_Group group, int* rank) { + int retval; - smpi_bench_end(); + smpi_bench_end(MPI_COMM_NULL, NULL); + if(group == MPI_GROUP_NULL) { + retval = MPI_ERR_GROUP; + } else if(rank == NULL) { + retval = MPI_ERR_ARG; + } else { + *rank = smpi_group_rank(group, smpi_process_index()); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} - if (NULL == request) { - retval = MPI_ERR_ARG; - } else { - int dst = 0; - retval = smpi_create_request(buf, count, datatype, src, dst, tag, comm, request); - if (NULL != *request && MPI_SUCCESS == retval) { - retval = smpi_mpi_irecv(*request); - } - } +int MPI_Group_translate_ranks (MPI_Group group1, int n, int* ranks1, MPI_Group group2, int* ranks2) { + int retval, i, index; - smpi_bench_begin(); + smpi_bench_end(MPI_COMM_NULL, NULL); + if(group1 == MPI_GROUP_NULL || group2 == MPI_GROUP_NULL) { + retval = MPI_ERR_GROUP; + } else { + for(i = 0; i < n; i++) { + index = smpi_group_index(group1, ranks1[i]); + ranks2[i] = smpi_group_rank(group2, index); + } + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} - return retval; +int MPI_Group_compare(MPI_Group group1, MPI_Group group2, int* result) { + int retval; + + smpi_bench_end(MPI_COMM_NULL, NULL); + if(group1 == MPI_GROUP_NULL || group2 == MPI_GROUP_NULL) { + retval = MPI_ERR_GROUP; + } else if(result == NULL) { + retval = MPI_ERR_ARG; + } else { + *result = smpi_group_compare(group1, group2); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} + +int MPI_Group_union(MPI_Group group1, MPI_Group group2, MPI_Group* newgroup) { + int retval, i, proc1, proc2, size, size2; + + smpi_bench_end(MPI_COMM_NULL, NULL); + if(group1 == MPI_GROUP_NULL || group2 == MPI_GROUP_NULL) { + retval = MPI_ERR_GROUP; + } else if(newgroup == NULL) { + retval = MPI_ERR_ARG; + } else { + size = smpi_group_size(group1); + size2 = smpi_group_size(group2); + for(i = 0; i < size2; i++) { + proc2 = smpi_group_index(group2, i); + proc1 = smpi_group_rank(group1, proc2); + if(proc1 == MPI_UNDEFINED) { + size++; + } + } + if(size == 0) { + *newgroup = MPI_GROUP_EMPTY; + } else { + *newgroup = smpi_group_new(size); + size2 = smpi_group_size(group1); + for(i = 0; i < size2; i++) { + proc1 = smpi_group_index(group1, i); + smpi_group_set_mapping(*newgroup, proc1, i); + } + for(i = size2; i < size; i++) { + proc2 = smpi_group_index(group2, i - size2); + smpi_group_set_mapping(*newgroup, proc2, i); + } + } + smpi_group_use(*newgroup); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; } -int MPI_Recv(void *buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm, MPI_Status *status) -{ - int retval = MPI_SUCCESS; - int dst = 0; - smpi_mpi_request_t request; +int MPI_Group_intersection(MPI_Group group1, MPI_Group group2, MPI_Group* newgroup) { + int retval, i, proc1, proc2, size, size2; - smpi_bench_end(); + smpi_bench_end(MPI_COMM_NULL, NULL); + if(group1 == MPI_GROUP_NULL || group2 == MPI_GROUP_NULL) { + retval = MPI_ERR_GROUP; + } else if(newgroup == NULL) { + retval = MPI_ERR_ARG; + } else { + size = smpi_group_size(group1); + size2 = smpi_group_size(group2); + for(i = 0; i < size2; i++) { + proc2 = smpi_group_index(group2, i); + proc1 = smpi_group_rank(group1, proc2); + if(proc1 == MPI_UNDEFINED) { + size--; + } + } + if(size == 0) { + *newgroup = MPI_GROUP_EMPTY; + } else { + *newgroup = smpi_group_new(size); + size2 = smpi_group_size(group1); + for(i = 0; i < size2; i++) { + proc1 = smpi_group_index(group1, i); + proc2 = smpi_group_rank(group2, proc1); + if(proc2 != MPI_UNDEFINED) { + smpi_group_set_mapping(*newgroup, proc1, i); + } + } + } + smpi_group_use(*newgroup); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} + +int MPI_Group_difference(MPI_Group group1, MPI_Group group2, MPI_Group* newgroup) { + int retval, i, proc1, proc2, size, size2; + + smpi_bench_end(MPI_COMM_NULL, NULL); + if(group1 == MPI_GROUP_NULL || group2 == MPI_GROUP_NULL) { + retval = MPI_ERR_GROUP; + } else if(newgroup == NULL) { + retval = MPI_ERR_ARG; + } else { + size = size2 = smpi_group_size(group1); + for(i = 0; i < size2; i++) { + proc1 = smpi_group_index(group1, i); + proc2 = smpi_group_rank(group2, proc1); + if(proc2 != MPI_UNDEFINED) { + size--; + } + } + if(size == 0) { + *newgroup = MPI_GROUP_EMPTY; + } else { + *newgroup = smpi_group_new(size); + for(i = 0; i < size2; i++) { + proc1 = smpi_group_index(group1, i); + proc2 = smpi_group_rank(group2, proc1); + if(proc2 == MPI_UNDEFINED) { + smpi_group_set_mapping(*newgroup, proc1, i); + } + } + } + smpi_group_use(*newgroup); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} - retval = smpi_create_request(buf, count, datatype, src, dst, tag, comm, &request); - if (NULL != request && MPI_SUCCESS == retval) { - retval = smpi_mpi_irecv(request); - if (MPI_SUCCESS == retval) { - retval = smpi_mpi_wait(request, status); - } - xbt_mallocator_release(smpi_global->request_mallocator, request); - } +int MPI_Group_incl(MPI_Group group, int n, int* ranks, MPI_Group* newgroup) { + int retval, i, index; - smpi_bench_begin(); + smpi_bench_end(MPI_COMM_NULL, NULL); + if(group == MPI_GROUP_NULL) { + retval = MPI_ERR_GROUP; + } else if(newgroup == NULL) { + retval = MPI_ERR_ARG; + } else { + if(n == 0) { + *newgroup = MPI_GROUP_EMPTY; + } else if(n == smpi_group_size(group)) { + *newgroup = group; + } else { + *newgroup = smpi_group_new(n); + for(i = 0; i < n; i++) { + index = smpi_group_index(group, ranks[i]); + smpi_group_set_mapping(*newgroup, index, i); + } + } + smpi_group_use(*newgroup); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} + +int MPI_Group_excl(MPI_Group group, int n, int* ranks, MPI_Group* newgroup) { + int retval, i, size, rank, index; + + smpi_bench_end(MPI_COMM_NULL, NULL); + if(group == MPI_GROUP_NULL) { + retval = MPI_ERR_GROUP; + } else if(newgroup == NULL) { + retval = MPI_ERR_ARG; + } else { + if(n == 0) { + *newgroup = group; + } else if(n == smpi_group_size(group)) { + *newgroup = MPI_GROUP_EMPTY; + } else { + size = smpi_group_size(group) - n; + *newgroup = smpi_group_new(size); + rank = 0; + while(rank < size) { + for(i = 0; i < n; i++) { + if(ranks[i] == rank) { + break; + } + } + if(i >= n) { + index = smpi_group_index(group, rank); + smpi_group_set_mapping(*newgroup, index, rank); + rank++; + } + } + } + smpi_group_use(*newgroup); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} + +int MPI_Group_range_incl(MPI_Group group, int n, int ranges[][3], MPI_Group* newgroup) { + int retval, i, j, rank, size, index; + + smpi_bench_end(MPI_COMM_NULL, NULL); + if(group == MPI_GROUP_NULL) { + retval = MPI_ERR_GROUP; + } else if(newgroup == NULL) { + retval = MPI_ERR_ARG; + } else { + if(n == 0) { + *newgroup = MPI_GROUP_EMPTY; + } else { + size = 0; + for(i = 0; i < n; i++) { + for(rank = ranges[i][0]; /* First */ + rank >= 0 && rank <= ranges[i][1]; /* Last */ + rank += ranges[i][2] /* Stride */) { + size++; + } + } + if(size == smpi_group_size(group)) { + *newgroup = group; + } else { + *newgroup = smpi_group_new(size); + j = 0; + for(i = 0; i < n; i++) { + for(rank = ranges[i][0]; /* First */ + rank >= 0 && rank <= ranges[i][1]; /* Last */ + rank += ranges[i][2] /* Stride */) { + index = smpi_group_index(group, rank); + smpi_group_set_mapping(*newgroup, index, j); + j++; + } + } + } + } + smpi_group_use(*newgroup); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} + +int MPI_Group_range_excl(MPI_Group group, int n, int ranges[][3], MPI_Group* newgroup) { + int retval, i, newrank, rank, size, index, add; + + smpi_bench_end(MPI_COMM_NULL, NULL); + if(group == MPI_GROUP_NULL) { + retval = MPI_ERR_GROUP; + } else if(newgroup == NULL) { + retval = MPI_ERR_ARG; + } else { + if(n == 0) { + *newgroup = group; + } else { + size = smpi_group_size(group); + for(i = 0; i < n; i++) { + for(rank = ranges[i][0]; /* First */ + rank >= 0 && rank <= ranges[i][1]; /* Last */ + rank += ranges[i][2] /* Stride */) { + size--; + } + } + if(size == 0) { + *newgroup = MPI_GROUP_EMPTY; + } else { + *newgroup = smpi_group_new(size); + newrank = 0; + while(newrank < size) { + for(i = 0; i < n; i++) { + add = 1; + for(rank = ranges[i][0]; /* First */ + rank >= 0 && rank <= ranges[i][1]; /* Last */ + rank += ranges[i][2] /* Stride */) { + if(rank == newrank) { + add = 0; + break; + } + } + if(add == 1) { + index = smpi_group_index(group, newrank); + smpi_group_set_mapping(*newgroup, index, newrank); + } + } + } + } + } + smpi_group_use(*newgroup); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} + +int MPI_Comm_rank(MPI_Comm comm, int* rank) { + int retval; + + smpi_bench_end(comm, "Comm_rank"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else { + *rank = smpi_comm_rank(comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Comm_rank"); + return retval; +} + +int MPI_Comm_size(MPI_Comm comm, int* size) { + int retval; + + smpi_bench_end(comm, "Comm_size"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(size == NULL) { + retval = MPI_ERR_ARG; + } else { + *size = smpi_comm_size(comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Comm_size"); + return retval; +} + +int MPI_Comm_group(MPI_Comm comm, MPI_Group* group) { + int retval; + + smpi_bench_end(comm, "Comm_group"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(group == NULL) { + retval = MPI_ERR_ARG; + } else { + *group = smpi_comm_group(comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Comm_group"); + return retval; +} + +int MPI_Comm_compare(MPI_Comm comm1, MPI_Comm comm2, int* result) { + int retval; + + smpi_bench_end(MPI_COMM_NULL, NULL); + if(comm1 == MPI_COMM_NULL || comm2 == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(result == NULL) { + retval = MPI_ERR_ARG; + } else { + if(comm1 == comm2) { /* Same communicators means same groups */ + *result = MPI_IDENT; + } else { + *result = smpi_group_compare(smpi_comm_group(comm1), smpi_comm_group(comm2)); + if(*result == MPI_IDENT) { + *result = MPI_CONGRUENT; + } + } + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} + +int MPI_Comm_dup(MPI_Comm comm, MPI_Comm* newcomm) { + int retval; + + smpi_bench_end(comm, "Comm_dup"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(newcomm == NULL) { + retval = MPI_ERR_ARG; + } else { + *newcomm = smpi_comm_new(smpi_comm_group(comm)); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Comm_dup"); + return retval; +} + +int MPI_Comm_create(MPI_Comm comm, MPI_Group group, MPI_Comm* newcomm) { + int retval; + + smpi_bench_end(comm, "Comm_create"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(group == MPI_GROUP_NULL) { + retval = MPI_ERR_GROUP; + } else if(newcomm == NULL) { + retval = MPI_ERR_ARG; + } else { + *newcomm = smpi_comm_new(group); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Comm_create"); + return retval; +} + +int MPI_Comm_free(MPI_Comm* comm) { + int retval; + + smpi_bench_end(*comm, "Comm_free"); + if(comm == NULL) { + retval = MPI_ERR_ARG; + } else if(*comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else { + smpi_comm_destroy(*comm); + *comm = MPI_COMM_NULL; + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, NULL); + return retval; +} + +int MPI_Irecv(void* buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm, MPI_Request* request) { + int retval; + + smpi_bench_end(comm, "Irecv"); + if(request == NULL) { + retval = MPI_ERR_ARG; + } else { + *request = smpi_mpi_irecv(buf, count, datatype, src, tag, comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Irecv"); + return retval; +} + +int MPI_Isend(void* buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm, MPI_Request* request) { + int retval; + + smpi_bench_end(comm, "Isend"); + if(request == NULL) { + retval = MPI_ERR_ARG; + } else { + *request = smpi_mpi_isend(buf, count, datatype, dst, tag, comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Isend"); + return retval; +} - return retval; +int MPI_Recv(void* buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm, MPI_Status* status) { + smpi_bench_end(comm, "Recv"); + smpi_mpi_recv(buf, count, datatype, src, tag, comm, status); + smpi_bench_begin(comm, "Recv"); + return MPI_SUCCESS; } -int MPI_Isend(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm, MPI_Request *request) -{ - int retval = MPI_SUCCESS; +int MPI_Send(void* buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm) { + smpi_bench_end(comm, "Send"); + smpi_mpi_send(buf, count, datatype, dst, tag, comm); + smpi_bench_begin(comm, "Send"); + return MPI_SUCCESS; +} + +int MPI_Sendrecv(void* sendbuf, int sendcount, MPI_Datatype sendtype, int dst, int sendtag, void* recvbuf, int recvcount, MPI_Datatype recvtype, int src, int recvtag, MPI_Comm comm, MPI_Status* status) { + smpi_bench_end(comm, "Sendrecv"); + smpi_mpi_sendrecv(sendbuf, sendcount, sendtype, dst, sendtag, recvbuf, recvcount, recvtype, src, recvtag, comm, status); + smpi_bench_begin(comm, "Sendrecv"); + return MPI_SUCCESS; +} + +int MPI_Sendrecv_replace(void* buf, int count, MPI_Datatype datatype, int dst, int sendtag, int src, int recvtag, MPI_Comm comm, MPI_Status* status) { + //TODO: suboptimal implementation + void* recvbuf; + int retval, size; + + size = smpi_datatype_size(datatype) * count; + recvbuf = xbt_new(char, size); + retval = MPI_Sendrecv(buf, count, datatype, dst, sendtag, recvbuf, count, datatype, src, recvtag, comm, status); + memcpy(buf, recvbuf, size * sizeof(char)); + xbt_free(recvbuf); + return retval; +} + +int MPI_Test(MPI_Request* request, int* flag, MPI_Status* status) { + int retval; + + smpi_bench_end((*request)->comm, "Test"); + if(request == NULL || flag == NULL) { + retval = MPI_ERR_ARG; + } else if(*request == MPI_REQUEST_NULL) { + retval = MPI_ERR_REQUEST; + } else { + *flag = smpi_mpi_test(request, status); + retval = MPI_SUCCESS; + } + smpi_bench_begin((*request)->comm, "Test"); + return retval; +} + +int MPI_Testany(int count, MPI_Request requests[], int* index, int* flag, MPI_Status* status) { + int retval; + + smpi_bench_end(MPI_COMM_NULL, "Testany"); //FIXME + if(index == NULL || flag == NULL) { + retval = MPI_ERR_ARG; + } else { + *flag = smpi_mpi_testany(count, requests, index, status); + retval = MPI_SUCCESS; + } + smpi_bench_begin(*flag ? requests[*index]->comm : MPI_COMM_NULL, "Testany"); + return retval; +} + +int MPI_Wait(MPI_Request* request, MPI_Status* status) { + int retval; + + smpi_bench_end((*request)->comm, "Wait"); + if(request == NULL) { + retval = MPI_ERR_ARG; + } else if(*request == MPI_REQUEST_NULL) { + retval = MPI_ERR_REQUEST; + } else { + smpi_mpi_wait(request, status); + retval = MPI_SUCCESS; + } + smpi_bench_begin((*request)->comm, "Wait"); + return retval; +} + +int MPI_Waitany(int count, MPI_Request requests[], int* index, MPI_Status* status) { + int retval; + + smpi_bench_end(MPI_COMM_NULL, "Waitany"); //FIXME + if(index == NULL) { + retval = MPI_ERR_ARG; + } else { + *index = smpi_mpi_waitany(count, requests, status); + retval = MPI_SUCCESS; + } + smpi_bench_begin(*index > 0 ? requests[*index]->comm : MPI_COMM_NULL, "Waitany"); + return retval; +} + +int MPI_Waitall(int count, MPI_Request requests[], MPI_Status status[]) { + smpi_bench_end(MPI_COMM_NULL, "Waitall"); //FIXME + smpi_mpi_waitall(count, requests, status); + smpi_bench_begin(MPI_COMM_NULL, "Waitall"); //FIXME + return MPI_SUCCESS; +} + +int MPI_Waitsome(int incount, MPI_Request requests[], int* outcount, int* indices, MPI_Status status[]) { + int retval; + + smpi_bench_end(MPI_COMM_NULL, "Waitsome"); //FIXME + if(outcount == NULL || indices == NULL) { + retval = MPI_ERR_ARG; + } else { + *outcount = smpi_mpi_waitsome(incount, requests, indices, status); + retval = MPI_SUCCESS; + } + smpi_bench_begin(MPI_COMM_NULL, "Waitsome"); //FIXME + return retval; +} + +int MPI_Bcast(void* buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm) { + int retval; + + smpi_bench_end(comm, "Bcast"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else { + smpi_mpi_bcast(buf, count, datatype, root, comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Bcast"); + return retval; +} + +int MPI_Barrier(MPI_Comm comm) { + int retval; + + smpi_bench_end(comm, "Barrier"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else { + smpi_mpi_barrier(comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Barrier"); + return retval; +} + +int MPI_Gather(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm) { + int retval; + + smpi_bench_end(comm, "Gather"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else { + smpi_mpi_gather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Gather"); + return retval; +} + +int MPI_Gatherv(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int* recvcounts, int* displs, MPI_Datatype recvtype, int root, MPI_Comm comm) { + int retval; + + smpi_bench_end(comm, "Gatherv"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else if(recvcounts == NULL || displs == NULL) { + retval = MPI_ERR_ARG; + } else { + smpi_mpi_gatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, root, comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Gatherv"); + return retval; +} + +int MPI_Allgather(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm) { + int retval; + + smpi_bench_end(comm, "Allgather"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else { + smpi_mpi_allgather(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Allgather"); + return retval; +} + +int MPI_Allgatherv(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int* recvcounts, int* displs, MPI_Datatype recvtype, MPI_Comm comm) { + int retval; + + smpi_bench_end(comm, "Allgatherv"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else if(recvcounts == NULL || displs == NULL) { + retval = MPI_ERR_ARG; + } else { + smpi_mpi_allgatherv(sendbuf, sendcount, sendtype, recvbuf, recvcounts, displs, recvtype, comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Allgatherv"); + return retval; +} + +int MPI_Scatter(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm) { + int retval; + + smpi_bench_end(comm, "Scatter"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else { + smpi_mpi_scatter(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, root, comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Scatter"); + return retval; +} + +int MPI_Scatterv(void* sendbuf, int* sendcounts, int* displs, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm) { + int retval; + + smpi_bench_end(comm, "Scatterv"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else if(sendcounts == NULL || displs == NULL) { + retval = MPI_ERR_ARG; + } else { + smpi_mpi_scatterv(sendbuf, sendcounts, displs, sendtype, recvbuf, recvcount, recvtype, root, comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Scatterv"); + return retval; +} + +int MPI_Reduce(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm) { + int retval; + + smpi_bench_end(comm, "Reduce"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(datatype == MPI_DATATYPE_NULL || op == MPI_OP_NULL) { + retval = MPI_ERR_ARG; + } else { + smpi_mpi_reduce(sendbuf, recvbuf, count, datatype, op, root, comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Reduce"); + return retval; +} + +int MPI_Allreduce(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) { + int retval; + + smpi_bench_end(comm, "Allreduce"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(datatype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else if(op == MPI_OP_NULL) { + retval = MPI_ERR_OP; + } else { + smpi_mpi_allreduce(sendbuf, recvbuf, count, datatype, op, comm); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Allreduce"); + return retval; +} + +int MPI_Reduce_scatter(void* sendbuf, void* recvbuf, int* recvcounts, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) { + int retval, i, rank, size, count; + int* displs; + + smpi_bench_end(comm, "Reduce_scatter"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(datatype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else if(op == MPI_OP_NULL) { + retval = MPI_ERR_OP; + } else if(recvcounts == NULL) { + retval = MPI_ERR_ARG; + } else { + /* arbitrarily choose root as rank 0 */ + /* TODO: faster direct implementation ? */ + rank = smpi_comm_rank(comm); + size = smpi_comm_size(comm); + count = 0; + displs = xbt_new(int, size); + for(i = 0; i < size; i++) { + count += recvcounts[i]; + displs[i] = 0; + } + smpi_mpi_reduce(sendbuf, recvbuf, count, datatype, op, 0, comm); + smpi_mpi_scatterv(recvbuf, recvcounts, displs, datatype, recvbuf, recvcounts[rank], datatype, 0, comm); + xbt_free(displs); + retval = MPI_SUCCESS; + } + smpi_bench_begin(comm, "Reduce_scatter"); + return retval; +} + +/** + * MPI_Alltoall user entry point + * + * Uses the logic of OpenMPI (upto 1.2.7 or greater) for the optimizations + * ompi/mca/coll/tuned/coll_tuned_module.c + **/ + +int MPI_Alltoall(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm) { + int retval, size, sendsize; + + smpi_bench_end(comm, "Alltoall"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else { + size = smpi_comm_size(comm); + sendsize = smpi_datatype_size(sendtype) * sendcount; + if(sendsize < 200 && size > 12) { + retval = smpi_coll_tuned_alltoall_bruck(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm); + } else if(sendsize < 3000) { + retval = smpi_coll_tuned_alltoall_basic_linear(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm); + } else { + retval = smpi_coll_tuned_alltoall_pairwise(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm); + } + } + smpi_bench_begin(comm, "Alltoall"); + return retval; +} - smpi_bench_end(); +int MPI_Alltoallv(void* sendbuf, int* sendcounts, int* senddisps, MPI_Datatype sendtype, void* recvbuf, int *recvcounts, int* recvdisps, MPI_Datatype recvtype, MPI_Comm comm) { + int retval; - if (NULL == request) { - retval = MPI_ERR_ARG; - } else { - int src = 0; - retval = smpi_create_request(buf, count, datatype, src, dst, tag, comm, request); - if (NULL != *request && MPI_SUCCESS == retval) { - retval = smpi_mpi_isend(*request); - } - } - - smpi_bench_begin(); - - return retval; -} - -int MPI_Send(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm) -{ - int retval = MPI_SUCCESS; - int src = 0; - smpi_mpi_request_t request; - - smpi_bench_end(); - - retval = smpi_create_request(buf, count, datatype, src, dst, tag, comm, &request); - if (NULL != request && MPI_SUCCESS == retval) { - retval = smpi_mpi_isend(request); - if (MPI_SUCCESS == retval) { - smpi_mpi_wait(request, MPI_STATUS_IGNORE); - } - xbt_mallocator_release(smpi_global->request_mallocator, request); - } - - smpi_bench_begin(); - - return retval; -} - -// FIXME: needs to return null in event of MPI_UNDEFINED color... -int MPI_Comm_split(MPI_Comm comm, int color, int key, MPI_Comm *comm_out) -{ - int retval = MPI_SUCCESS; - - int index, rank; - smpi_mpi_request_t request; - int colorkey[2]; - smpi_mpi_status_t status; - - smpi_bench_end(); - - // FIXME: need to test parameters - - index = smpi_host_index(); - rank = comm->index_to_rank_map[index]; - - if (0 == rank) { - int *colors = xbt_new(int, comm->size); - int *keys = xbt_new(int, comm->size); - int i, j, k; - smpi_mpi_communicator_t tempcomm = NULL; - int colortmp; - int keycount; - int *keystmp = xbt_new(int, comm->size); - int *rankstmp = xbt_new(int, comm->size); - int tmpval; - int indextmp; - - colors[0] = color; - keys[0] = key; - - // FIXME: not efficient - for (i = 1; i < comm->size; i++) { - retval = smpi_create_request(colorkey, 2, MPI_INT, MPI_ANY_SOURCE, rank, MPI_ANY_TAG, comm, &request); - smpi_mpi_irecv(request); - smpi_mpi_wait(request, &status); - xbt_mallocator_release(smpi_global->request_mallocator, request); - colors[i] = colorkey[0]; - keys[i] = colorkey[1]; - } - - for (i = 0; i < comm->size; i++) { - if (-1 == colors[i]) { - continue; - } - colortmp = colors[i]; - keycount = 0; - for (j = i; j < comm->size; j++) { - if(colortmp == colors[j]) { - colors[j] = -1; - keystmp[keycount] = keys[j]; - rankstmp[keycount] = j; - keycount++; - } - } - if (0 < keycount) { - // FIXME: yes, mock me, bubble sort... - for (j = 0; j < keycount; j++) { - for (k = keycount - 1; k > j; k--) { - if (keystmp[k] < keystmp[k - 1]) { - tmpval = keystmp[k]; - keystmp[k] = keystmp[k - 1]; - keystmp[k - 1] = tmpval; - - tmpval = rankstmp[k]; - rankstmp[k] = rankstmp[k - 1]; - rankstmp[k - 1] = tmpval; - } - } - } - tempcomm = xbt_new(s_smpi_mpi_communicator_t, 1); - tempcomm->barrier_count = 0; - tempcomm->barrier_mutex = SIMIX_mutex_init(); - tempcomm->barrier_cond = SIMIX_cond_init(); - tempcomm->rank_to_index_map = xbt_new(int, keycount); - tempcomm->index_to_rank_map = xbt_new(int, smpi_global->host_count); - for (j = 0; j < smpi_global->host_count; j++) { - tempcomm->index_to_rank_map[j] = -1; - } - for (j = 0; j < keycount; j++) { - indextmp = comm->rank_to_index_map[rankstmp[j]]; - tempcomm->rank_to_index_map[j] = indextmp; - tempcomm->index_to_rank_map[indextmp] = j; - } - for (j = 0; j < keycount; j++) { - retval = smpi_create_request(&j, 1, MPI_INT, 0, rankstmp[j], 0, comm, &request); - request->data = tempcomm; - smpi_mpi_isend(request); - smpi_mpi_wait(request, &status); - xbt_mallocator_release(smpi_global->request_mallocator, request); - } - } - } - } else { - colorkey[0] = color; - colorkey[1] = key; - retval = smpi_create_request(colorkey, 2, MPI_INT, rank, 0, 0, comm, &request); - smpi_mpi_isend(request); - smpi_mpi_wait(request, &status); - xbt_mallocator_release(smpi_global->request_mallocator, request); - retval = smpi_create_request(colorkey, 1, MPI_INT, 0, rank, 0, comm, &request); - smpi_mpi_irecv(request); - smpi_mpi_wait(request, &status); - comm_out = request->data; - } - - smpi_bench_begin(); - - return retval; + smpi_bench_end(comm, "Alltoallv"); + if(comm == MPI_COMM_NULL) { + retval = MPI_ERR_COMM; + } else if(sendtype == MPI_DATATYPE_NULL || recvtype == MPI_DATATYPE_NULL) { + retval = MPI_ERR_TYPE; + } else if(sendcounts == NULL || senddisps == NULL || recvcounts == NULL || recvdisps == NULL) { + retval = MPI_ERR_ARG; + } else { + retval = smpi_coll_basic_alltoallv(sendbuf, sendcounts, senddisps, sendtype, recvbuf, recvcounts, recvdisps, recvtype, comm); + } + smpi_bench_begin(comm, "Alltoallv"); + return retval; }