X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/18e13a976a3f6f1d695de22d17c3b91befdba176..c84b205ab7658cc51948d8adcca162f2cdd96d51:/src/smpi/smpi_base.c diff --git a/src/smpi/smpi_base.c b/src/smpi/smpi_base.c index 333c590c91..8bb78e14f8 100644 --- a/src/smpi/smpi_base.c +++ b/src/smpi/smpi_base.c @@ -1,270 +1,926 @@ +/* Copyright (c) 2007, 2008, 2009, 2010. 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. */ + #include "private.h" -#include "xbt/time.h" +#include "xbt/virtu.h" +#include "mc/mc.h" +#include "xbt/replay.h" +#include +#include "simix/smx_private.h" +#include "surf/surf.h" +#include "simgrid/sg_config.h" +#include "colls/colls.h" + +XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_base, smpi, "Logging specific to SMPI (base)"); + + +static int match_recv(void* a, void* b, smx_action_t ignored) { + MPI_Request ref = (MPI_Request)a; + MPI_Request req = (MPI_Request)b; + XBT_DEBUG("Trying to match a recv of src %d against %d, tag %d against %d",ref->src,req->src, ref->tag, req->tag); + + xbt_assert(ref, "Cannot match recv against null reference"); + xbt_assert(req, "Cannot match recv against null request"); + if((ref->src == MPI_ANY_SOURCE || req->src == ref->src) + && ((ref->tag == MPI_ANY_TAG && req->tag >=0) || req->tag == ref->tag)){ + //we match, we can transfer some values + // FIXME : move this to the copy function ? + if(ref->src == MPI_ANY_SOURCE)ref->real_src = req->src; + if(ref->tag == MPI_ANY_TAG)ref->real_tag = req->tag; + if(ref->real_size < req->real_size) ref->truncated = 1; + if(req->detached==1){ + ref->detached_sender=req; //tie the sender to the receiver, as it is detached and has to be freed in the receiver + } + XBT_DEBUG("match succeeded"); + return 1; + }else return 0; +} + +static int match_send(void* a, void* b,smx_action_t ignored) { + MPI_Request ref = (MPI_Request)a; + MPI_Request req = (MPI_Request)b; + XBT_DEBUG("Trying to match a send of src %d against %d, tag %d against %d",ref->src,req->src, ref->tag, req->tag); + xbt_assert(ref, "Cannot match send against null reference"); + xbt_assert(req, "Cannot match send against null request"); + + if((req->src == MPI_ANY_SOURCE || req->src == ref->src) + && ((req->tag == MPI_ANY_TAG && ref->tag >=0)|| req->tag == ref->tag)) + { + if(req->src == MPI_ANY_SOURCE)req->real_src = ref->src; + if(req->tag == MPI_ANY_TAG)req->real_tag = ref->tag; + if(req->real_size < ref->real_size) req->truncated = 1; + if(ref->detached==1){ + req->detached_sender=ref; //tie the sender to the receiver, as it is detached and has to be freed in the receiver + } + XBT_DEBUG("match succeeded"); + return 1; + } else return 0; +} + + +typedef struct s_smpi_factor *smpi_factor_t; +typedef struct s_smpi_factor { + long factor; + int nb_values; + double values[4];//arbitrary set to 4 +} s_smpi_factor_t; +xbt_dynar_t smpi_os_values = NULL; +xbt_dynar_t smpi_or_values = NULL; +xbt_dynar_t smpi_ois_values = NULL; + +// Methods used to parse and store the values for timing injections in smpi +// These are taken from surf/network.c and generalized to have more factors +// These methods should be merged with those in surf/network.c (moved somewhere in xbt ?) + +static int factor_cmp(const void *pa, const void *pb) +{ + return (((s_smpi_factor_t*)pa)->factor > ((s_smpi_factor_t*)pb)->factor); +} + -XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_base, smpi, - "Logging specific to SMPI (base)"); -XBT_LOG_EXTERNAL_CATEGORY(smpi_base); -XBT_LOG_EXTERNAL_CATEGORY(smpi_bench); -XBT_LOG_EXTERNAL_CATEGORY(smpi_kernel); -XBT_LOG_EXTERNAL_CATEGORY(smpi_mpi); -XBT_LOG_EXTERNAL_CATEGORY(smpi_mpi_dt); -XBT_LOG_EXTERNAL_CATEGORY(smpi_coll); -XBT_LOG_EXTERNAL_CATEGORY(smpi_receiver); -XBT_LOG_EXTERNAL_CATEGORY(smpi_sender); -XBT_LOG_EXTERNAL_CATEGORY(smpi_util); +static xbt_dynar_t parse_factor(const char *smpi_coef_string) +{ + char *value = NULL; + unsigned int iter = 0; + s_smpi_factor_t fact; + int i=0; + xbt_dynar_t smpi_factor, radical_elements, radical_elements2 = NULL; -void smpi_process_init(int* argc, char*** argv) { - int index; - smpi_process_data_t data; - smx_process_t proc; + smpi_factor = xbt_dynar_new(sizeof(s_smpi_factor_t), NULL); + radical_elements = xbt_str_split(smpi_coef_string, ";"); + xbt_dynar_foreach(radical_elements, iter, value) { + fact.nb_values=0; + radical_elements2 = xbt_str_split(value, ":"); + if (xbt_dynar_length(radical_elements2) <2 || xbt_dynar_length(radical_elements2) > 5) + xbt_die("Malformed radical for smpi factor!"); + for(i =0; i 2) { - free((*argv)[1]); - memmove(&(*argv)[1], &(*argv)[2], sizeof(char *) * (*argc - 2)); - (*argv)[(*argc) - 1] = NULL; + xbt_dynar_push_as(smpi_factor, s_smpi_factor_t, fact); + XBT_DEBUG("smpi_factor:\t%ld : %d values, first: %f", fact.factor, fact.nb_values ,fact.values[0]); + xbt_dynar_free(&radical_elements2); + } + xbt_dynar_free(&radical_elements); + iter=0; + xbt_dynar_sort(smpi_factor, &factor_cmp); + xbt_dynar_foreach(smpi_factor, iter, fact) { + XBT_DEBUG("smpi_factor:\t%ld : %d values, first: %f", fact.factor, fact.nb_values ,fact.values[0]); } - (*argc)--; - DEBUG2("<%d> New process in the game: %p", index, proc); + return smpi_factor; } -void smpi_process_destroy(void) { - int index = smpi_process_index(); +static double smpi_os(double size) +{ + if (!smpi_os_values) { + smpi_os_values = parse_factor(sg_cfg_get_string("smpi/os")); + smpi_register_static(smpi_os_values, xbt_dynar_free_voidp); + } + unsigned int iter = 0; + s_smpi_factor_t fact; + double current=0.0; + xbt_dynar_foreach(smpi_os_values, iter, fact) { + if (size <= fact.factor) { + XBT_DEBUG("os : %lf <= %ld return %f", size, fact.factor, current); + return current; + }else{ + current=fact.values[0]+fact.values[1]*size; + } + } + XBT_DEBUG("os : %lf > %ld return %f", size, fact.factor, current); - DEBUG1("<%d> Process left the game", index); + return current; } -/* MPI Low level calls */ -MPI_Request smpi_mpi_isend(void* buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm) { +static double smpi_ois(double size) +{ + if (!smpi_ois_values) { + smpi_ois_values = parse_factor(sg_cfg_get_string("smpi/ois")); + smpi_register_static(smpi_ois_values, xbt_dynar_free_voidp); + } + unsigned int iter = 0; + s_smpi_factor_t fact; + double current=0.0; + xbt_dynar_foreach(smpi_ois_values, iter, fact) { + if (size <= fact.factor) { + XBT_DEBUG("ois : %lf <= %ld return %f", size, fact.factor, current); + return current; + }else{ + current=fact.values[0]+fact.values[1]*size; + } + } + XBT_DEBUG("ois : %lf > %ld return %f", size, fact.factor, current); + + return current; +} + +static double smpi_or(double size) +{ + if (!smpi_or_values) { + smpi_or_values = parse_factor(sg_cfg_get_string("smpi/or")); + smpi_register_static(smpi_or_values, xbt_dynar_free_voidp); + } + unsigned int iter = 0; + s_smpi_factor_t fact; + double current=0.0; + xbt_dynar_foreach(smpi_or_values, iter, fact) { + if (size <= fact.factor) { + XBT_DEBUG("or : %lf <= %ld return %f", size, fact.factor, current); + return current; + }else + current=fact.values[0]+fact.values[1]*size; + } + XBT_DEBUG("or : %lf > %ld return %f", size, fact.factor, current); + + return current; +} + +static MPI_Request build_request(void *buf, int count, + MPI_Datatype datatype, int src, int dst, + int tag, MPI_Comm comm, unsigned flags) +{ MPI_Request request; + void *old_buf = NULL; + request = xbt_new(s_smpi_mpi_request_t, 1); - request->comm = comm; - request->src = smpi_comm_rank(comm); + + s_smpi_subtype_t *subtype = datatype->substruct; + + if(datatype->has_subtype == 1){ + // This part handles the problem of non-contiguous memory + old_buf = buf; + buf = count==0 ? NULL : xbt_malloc(count*smpi_datatype_size(datatype)); + if (flags & SEND) { + subtype->serialize(old_buf, buf, count, datatype->substruct); + } + } + + request->buf = buf; + // This part handles the problem of non-contiguous memory (for the + // unserialisation at the reception) + request->old_buf = old_buf; + request->old_type = datatype; + + request->size = smpi_datatype_size(datatype) * count; + request->src = src; request->dst = dst; request->tag = tag; - request->size = smpi_datatype_size(datatype) * count; - request->complete = 0; - smpi_process_post_send(comm, request); - request->pair = SIMIX_network_isend(request->rdv, request->size, -1.0, buf, request->size, request); + request->comm = comm; + request->action = NULL; + request->flags = flags; + request->detached = 0; + request->detached_sender = NULL; + + request->truncated = 0; + request->real_size = 0; + request->real_tag = 0; + + request->refcount=1; +#ifdef HAVE_TRACING + request->send = 0; + request->recv = 0; +#endif + if (flags & SEND) smpi_datatype_unuse(datatype); + return request; } -MPI_Request smpi_mpi_irecv(void* buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm) { - MPI_Request request; - request = xbt_new(s_smpi_mpi_request_t, 1); - request->comm = comm; - request->src = src; - request->dst = smpi_comm_rank(comm); - request->tag = tag; - request->size = smpi_datatype_size(datatype) * count; - request->complete = 0; - smpi_process_post_recv(request); - request->pair = SIMIX_network_irecv(request->rdv, buf, &request->size); +void smpi_empty_status(MPI_Status * status) +{ + if(status != MPI_STATUS_IGNORE) { + status->MPI_SOURCE = MPI_ANY_SOURCE; + status->MPI_TAG = MPI_ANY_TAG; + status->MPI_ERROR = MPI_SUCCESS; + status->count=0; + } +} + +void smpi_action_trace_run(char *path) +{ + char *name; + xbt_dynar_t todo; + xbt_dict_cursor_t cursor; + + action_fp=NULL; + if (path) { + action_fp = fopen(path, "r"); + xbt_assert(action_fp != NULL, "Cannot open %s: %s", path, + strerror(errno)); + } + + if (!xbt_dict_is_empty(action_queues)) { + XBT_WARN + ("Not all actions got consumed. If the simulation ended successfully (without deadlock), you may want to add new processes to your deployment file."); + + + xbt_dict_foreach(action_queues, cursor, name, todo) { + XBT_WARN("Still %lu actions for %s", xbt_dynar_length(todo), name); + } + } + + if (path) + fclose(action_fp); + xbt_dict_free(&action_queues); + action_queues = xbt_dict_new_homogeneous(NULL); +} + +static void smpi_mpi_request_free_voidp(void* request) +{ + MPI_Request req = request; + smpi_mpi_request_free(&req); +} + +/* MPI Low level calls */ +MPI_Request smpi_mpi_send_init(void *buf, int count, MPI_Datatype datatype, + int dst, int tag, MPI_Comm comm) +{ + MPI_Request request = + build_request(buf==MPI_BOTTOM ? (void*)0 : buf, count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag, + comm, PERSISTENT | SEND | PREPARED); + request->refcount++; return request; } -void smpi_mpi_recv(void* buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm, MPI_Status* status) { - MPI_Request request; +MPI_Request smpi_mpi_ssend_init(void *buf, int count, MPI_Datatype datatype, + int dst, int tag, MPI_Comm comm) +{ + MPI_Request request = + build_request(buf==MPI_BOTTOM ? (void*)0 : buf, count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag, + comm, PERSISTENT | SSEND | SEND | PREPARED); + request->refcount++; + return request; +} + +MPI_Request smpi_mpi_recv_init(void *buf, int count, MPI_Datatype datatype, + int src, int tag, MPI_Comm comm) +{ + MPI_Request request = + build_request(buf==MPI_BOTTOM ? (void*)0 : buf, count, datatype, src == MPI_ANY_SOURCE ? MPI_ANY_SOURCE : smpi_group_index(smpi_comm_group(comm), src), smpi_process_index(), tag, + comm, PERSISTENT | RECV | PREPARED); + request->refcount++; + return request; +} + +void smpi_mpi_start(MPI_Request request) +{ + smx_rdv_t mailbox; + + xbt_assert(!request->action, + "Cannot (re)start a non-finished communication"); + if(request->flags & PREPARED)request->flags &= ~PREPARED; + if(request->flags & RECV) { + print_request("New recv", request); + //FIXME: if receive is posted with a large size, but send is smaller, mailboxes may not match ! + if (request->size < sg_cfg_get_int("smpi/async_small_thres")) + mailbox = smpi_process_mailbox_small(); + else + mailbox = smpi_process_mailbox(); + // we make a copy here, as the size is modified by simix, and we may reuse the request in another receive later + request->real_size=request->size; + smpi_datatype_use(request->old_type); + smpi_comm_use(request->comm); + request->action = simcall_comm_irecv(mailbox, request->buf, &request->real_size, &match_recv, request); + + //integrate pseudo-timing for buffering of small messages, do not bother to execute the simcall if 0 + double sleeptime = request->detached ? smpi_or(request->size) : 0.0; + if(sleeptime!=0.0){ + simcall_process_sleep(sleeptime); + XBT_DEBUG("receiving size of %zu : sleep %lf ", request->size, smpi_or(request->size)); + } + + } else { + + + int receiver = request->dst;//smpi_group_index(smpi_comm_group(request->comm), request->dst); + + #ifdef HAVE_TRACING + int rank = smpi_process_index(); + if (TRACE_smpi_view_internals()) { + TRACE_smpi_send(rank, rank, receiver,request->size); + } + #endif +/* if(receiver == MPI_UNDEFINED) {*/ +/* XBT_WARN("Trying to send a message to a wrong rank");*/ +/* return;*/ +/* }*/ + print_request("New send", request); + if (request->size < sg_cfg_get_int("smpi/async_small_thres")) { // eager mode + mailbox = smpi_process_remote_mailbox_small(receiver); + }else{ + XBT_DEBUG("Send request %p is not in the permanent receive mailbox (buf: %p)",request,request->buf); + mailbox = smpi_process_remote_mailbox(receiver); + } + if ( (! (request->flags & SSEND)) && (request->size < sg_cfg_get_int("smpi/send_is_detached_thres"))) { + void *oldbuf = NULL; + request->detached = 1; + request->refcount++; + if(request->old_type->has_subtype == 0){ + oldbuf = request->buf; + if (oldbuf && request->size!=0){ + request->buf = xbt_malloc(request->size); + memcpy(request->buf,oldbuf,request->size); + } + } + XBT_DEBUG("Send request %p is detached; buf %p copied into %p",request,oldbuf,request->buf); + } + + // we make a copy here, as the size is modified by simix, and we may reuse the request in another receive later + request->real_size=request->size; + smpi_datatype_use(request->old_type); + smpi_comm_use(request->comm); + + //if we are giving back the control to the user without waiting for completion, we have to inject timings + double sleeptime =0.0; + if(request->detached || (request->flags & (ISEND|SSEND))){// issend should be treated as isend + //isend and send timings may be different + sleeptime = (request->flags & ISEND)? smpi_ois(request->size) : smpi_os(request->size); + } + + if(sleeptime!=0.0){ + simcall_process_sleep(sleeptime); + XBT_DEBUG("sending size of %zu : sleep %lf ", request->size, smpi_os(request->size)); + } + + request->action = + simcall_comm_isend(mailbox, request->size, -1.0, + request->buf, request->real_size, + &match_send, + &smpi_mpi_request_free_voidp, // how to free the userdata if a detached send fails + request, + // detach if msg size < eager/rdv switch limit + request->detached); + +#ifdef HAVE_TRACING + /* FIXME: detached sends are not traceable (request->action == NULL) */ + if (request->action) + simcall_set_category(request->action, TRACE_internal_smpi_get_category()); + +#endif + + } + +} + +void smpi_mpi_startall(int count, MPI_Request * requests) +{ + int i; + + for(i = 0; i < count; i++) { + smpi_mpi_start(requests[i]); + } +} + +void smpi_mpi_request_free(MPI_Request * request) +{ + if((*request) != MPI_REQUEST_NULL){ + (*request)->refcount--; + if((*request)->refcount<0) xbt_die("wrong refcount"); + + if((*request)->refcount==0){ + print_request("Destroying", (*request)); + xbt_free(*request); + *request = MPI_REQUEST_NULL; + }else{ + print_request("Decrementing", (*request)); + + } + }else{ + xbt_die("freeing an already free request"); + } +} + +MPI_Request smpi_isend_init(void *buf, int count, MPI_Datatype datatype, + int dst, int tag, MPI_Comm comm) +{ + MPI_Request request = + build_request(buf==MPI_BOTTOM ? (void*)0 : buf , count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag, + comm, NON_PERSISTENT | SEND | PREPARED); + + return request; +} + +MPI_Request smpi_mpi_isend(void *buf, int count, MPI_Datatype datatype, + int dst, int tag, MPI_Comm comm) +{ + MPI_Request request = + build_request(buf==MPI_BOTTOM?(void*)0:buf, count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag, + comm, NON_PERSISTENT | ISEND | SEND); + + smpi_mpi_start(request); + return request; +} + +MPI_Request smpi_mpi_issend(void *buf, int count, MPI_Datatype datatype, + int dst, int tag, MPI_Comm comm) +{ + MPI_Request request = + build_request(buf==MPI_BOTTOM ? (void*)0 : buf, count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag, + comm, NON_PERSISTENT | ISEND | SSEND | SEND); + smpi_mpi_start(request); + return request; +} + + + +MPI_Request smpi_irecv_init(void *buf, int count, MPI_Datatype datatype, + int src, int tag, MPI_Comm comm) +{ + MPI_Request request = + build_request(buf==MPI_BOTTOM ? (void*)0 : buf, count, datatype, src == MPI_ANY_SOURCE ? MPI_ANY_SOURCE : smpi_group_index(smpi_comm_group(comm), src), smpi_process_index(), tag, + comm, NON_PERSISTENT | RECV | PREPARED); + return request; +} + +MPI_Request smpi_mpi_irecv(void *buf, int count, MPI_Datatype datatype, + int src, int tag, MPI_Comm comm) +{ + MPI_Request request = + build_request(buf==MPI_BOTTOM ? (void*)0 : buf, count, datatype, src == MPI_ANY_SOURCE ? MPI_ANY_SOURCE : smpi_group_index(smpi_comm_group(comm), src), smpi_process_index(), tag, + comm, NON_PERSISTENT | RECV); + + smpi_mpi_start(request); + return request; +} +void smpi_mpi_recv(void *buf, int count, MPI_Datatype datatype, int src, + int tag, MPI_Comm comm, MPI_Status * status) +{ + MPI_Request request; request = smpi_mpi_irecv(buf, count, datatype, src, tag, comm); smpi_mpi_wait(&request, status); } -void smpi_mpi_send(void* buf, int count, MPI_Datatype datatype, int src, int tag, MPI_Comm comm) { - MPI_Request request; - request = smpi_mpi_isend(buf, count, datatype, src, tag, comm); + +void smpi_mpi_send(void *buf, int count, MPI_Datatype datatype, int dst, + int tag, MPI_Comm comm) +{ + MPI_Request request = + build_request(buf==MPI_BOTTOM ? (void*)0 : buf, count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag, + comm, NON_PERSISTENT | SEND); + smpi_mpi_start(request); + smpi_mpi_wait(&request, MPI_STATUS_IGNORE); + +} + +void smpi_mpi_ssend(void *buf, int count, MPI_Datatype datatype, + int dst, int tag, MPI_Comm comm) +{ + MPI_Request request = + build_request(buf==MPI_BOTTOM ? (void*)0 : buf, count, datatype, smpi_process_index(), smpi_group_index(smpi_comm_group(comm), dst), tag, + comm, NON_PERSISTENT | SSEND | SEND); + + smpi_mpi_start(request); smpi_mpi_wait(&request, MPI_STATUS_IGNORE); } -void smpi_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) { +void smpi_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) +{ MPI_Request requests[2]; MPI_Status stats[2]; - - requests[0] = smpi_mpi_isend(sendbuf, sendcount, sendtype, dst, sendtag, comm); - requests[1] = smpi_mpi_irecv(recvbuf, recvcount, recvtype, src, recvtag, comm); + int myid=smpi_process_index(); + if ((smpi_group_index(smpi_comm_group(comm), dst) == myid) && (smpi_group_index(smpi_comm_group(comm), src) == myid)) { + smpi_datatype_copy(sendbuf, sendcount, sendtype, + recvbuf, recvcount, recvtype); + return; + } + requests[0] = + smpi_isend_init(sendbuf, sendcount, sendtype, dst, sendtag, comm); + requests[1] = + smpi_irecv_init(recvbuf, recvcount, recvtype, src, recvtag, comm); + smpi_mpi_startall(2, requests); smpi_mpi_waitall(2, requests, stats); if(status != MPI_STATUS_IGNORE) { // Copy receive status - memcpy(status, &stats[1], sizeof(MPI_Status)); + *status = stats[1]; } } -static void finish_wait(MPI_Request* request, MPI_Status* status) { - MPI_Request data = (MPI_Request)SIMIX_communication_get_data((*request)->pair); +int smpi_mpi_get_count(MPI_Status * status, MPI_Datatype datatype) +{ + return status->count / smpi_datatype_size(datatype); +} + +static void finish_wait(MPI_Request * request, MPI_Status * status) +{ + MPI_Request req = *request; + if(status != MPI_STATUS_IGNORE) + smpi_empty_status(status); + + if(!(req->detached && req->flags & SEND) && !(req->flags & PREPARED)){ + if(status != MPI_STATUS_IGNORE) { + int src = req->src == MPI_ANY_SOURCE ? req->real_src : req->src; + status->MPI_SOURCE = smpi_group_rank(smpi_comm_group(req->comm), src); + status->MPI_TAG = req->tag == MPI_ANY_TAG ? req->real_tag : req->tag; + status->MPI_ERROR = req->truncated ? MPI_ERR_TRUNCATE : MPI_SUCCESS; + // this handles the case were size in receive differs from size in send + // FIXME: really this should just contain the count of receive-type blocks, + // right? + status->count = req->real_size; + } + + print_request("Finishing", req); + MPI_Datatype datatype = req->old_type; + + if(datatype->has_subtype == 1){ + // This part handles the problem of non-contignous memory + // the unserialization at the reception + s_smpi_subtype_t *subtype = datatype->substruct; + if(req->flags & RECV) { + subtype->unserialize(req->buf, req->old_buf, req->real_size/smpi_datatype_size(datatype) , datatype->substruct); + } + if(req->detached == 0) free(req->buf); + } + smpi_comm_unuse(req->comm); + smpi_datatype_unuse(datatype); - xbt_assert0(data != MPI_REQUEST_NULL, "Erroneous situation"); - if(status != MPI_STATUS_IGNORE) { - status->MPI_SOURCE = (*request)->src; - status->MPI_TAG = (*request)->tag; - status->MPI_ERROR = MPI_SUCCESS; } - DEBUG3("finishing wait for %p [data = %p, complete = %d]", *request, data, data->complete); - // data == *request if sender is first to finish its wait - // data != *request if receiver is first to finish its wait - if(data->complete == 0) { - // first arrives here - data->complete = 1; - if(data != *request) { - // receveiver cleans its part - xbt_free(*request); + +#ifdef HAVE_TRACING + if (TRACE_smpi_view_internals()) { + if(req->flags & RECV){ + int rank = smpi_process_index(); + int src_traced = (req->src == MPI_ANY_SOURCE ? req->real_src : req->src); + TRACE_smpi_recv(rank, src_traced, rank); + } } +#endif + + if(req->detached_sender!=NULL){ + smpi_mpi_request_free(&(req->detached_sender)); + } + + if(req->flags & NON_PERSISTENT) { + smpi_mpi_request_free(request); } else { - // second arrives here - if(data != *request) { - // receiver cleans everything - xbt_free(data); - } - xbt_free(*request); + req->action = NULL; } - *request = MPI_REQUEST_NULL; } -int smpi_mpi_test(MPI_Request* request, MPI_Status* status) { - MPI_Request data = (MPI_Request)SIMIX_communication_get_data((*request)->pair); - int flag = data && data->complete == 1; +int smpi_mpi_test(MPI_Request * request, MPI_Status * status) { + int flag; + //assume that request is not MPI_REQUEST_NULL (filtered in PMPI_Test or smpi_mpi_testall before) + if ((*request)->action == NULL) + flag = 1; + else + flag = simcall_comm_test((*request)->action); if(flag) { finish_wait(request, status); + request=MPI_REQUEST_NULL; + }else{ + smpi_empty_status(status); } return flag; } -int smpi_mpi_testany(int count, MPI_Request requests[], int* index, MPI_Status* status) { - MPI_Request data; - int i, flag; +int smpi_mpi_testany(int count, MPI_Request requests[], int *index, + MPI_Status * status) +{ + xbt_dynar_t comms; + int i, flag, size; + int* map; *index = MPI_UNDEFINED; flag = 0; + comms = xbt_dynar_new(sizeof(smx_action_t), NULL); + map = xbt_new(int, count); + size = 0; for(i = 0; i < count; i++) { - if(requests[i] != MPI_REQUEST_NULL) { - data = (MPI_Request)SIMIX_communication_get_data(requests[i]->pair); - if(data != MPI_REQUEST_NULL && data->complete == 1) { - finish_wait(&requests[i], status); - *index = i; - flag = 1; - break; + if((requests[i]!=MPI_REQUEST_NULL) && requests[i]->action) { + xbt_dynar_push(comms, &requests[i]->action); + map[size] = i; + size++; + } + } + if(size > 0) { + i = simcall_comm_testany(comms); + // not MPI_UNDEFINED, as this is a simix return code + if(i != -1) { + *index = map[i]; + finish_wait(&requests[*index], status); + flag = 1; + } + }else{ + //all requests are null or inactive, return true + flag=1; + smpi_empty_status(status); + } + xbt_free(map); + xbt_dynar_free(&comms); + + return flag; +} + + +int smpi_mpi_testall(int count, MPI_Request requests[], + MPI_Status status[]) +{ + MPI_Status stat; + MPI_Status *pstat = status == MPI_STATUSES_IGNORE ? MPI_STATUS_IGNORE : &stat; + int flag=1; + int i; + for(i=0; ipair); +void smpi_mpi_probe(int source, int tag, MPI_Comm comm, MPI_Status* status){ + int flag=0; + //FIXME find another wait to avoid busy waiting ? + // the issue here is that we have to wait on a nonexistent comm + while(flag==0){ + smpi_mpi_iprobe(source, tag, comm, &flag, status); + XBT_DEBUG("Busy Waiting on probing : %d", flag); + } +} + +void smpi_mpi_iprobe(int source, int tag, MPI_Comm comm, int* flag, MPI_Status* status){ + + MPI_Request request =build_request(NULL, 0, MPI_CHAR, source == MPI_ANY_SOURCE ? MPI_ANY_SOURCE : smpi_group_index(smpi_comm_group(comm), source), smpi_comm_rank(comm), tag, + comm, NON_PERSISTENT | RECV); + + //to avoid deadlock, we have to sleep some time here, or the timer won't advance and we will only do iprobe simcalls + double sleeptime= sg_cfg_get_double("smpi/iprobe"); + //multiplier to the sleeptime, to increase speed of execution, each failed iprobe will increase it + static int nsleeps = 1; + + simcall_process_sleep(sleeptime); + + // behave like a receive, but don't do it + smx_rdv_t mailbox; + + print_request("New iprobe", request); + // We have to test both mailboxes as we don't know if we will receive one one or another + if (sg_cfg_get_int("smpi/async_small_thres")>0){ + mailbox = smpi_process_mailbox_small(); + XBT_DEBUG("trying to probe the perm recv mailbox"); + request->action = simcall_comm_iprobe(mailbox, request->src, request->tag, &match_recv, (void*)request); + } + if (request->action==NULL){ + mailbox = smpi_process_mailbox(); + XBT_DEBUG("trying to probe the other mailbox"); + request->action = simcall_comm_iprobe(mailbox, request->src, request->tag, &match_recv, (void*)request); + } + + if(request->action){ + MPI_Request req = (MPI_Request)SIMIX_comm_get_src_data(request->action); + *flag = 1; + if(status != MPI_STATUS_IGNORE && !(req->flags & PREPARED)) { + status->MPI_SOURCE = smpi_group_rank(smpi_comm_group(comm), req->src); + status->MPI_TAG = req->tag; + status->MPI_ERROR = MPI_SUCCESS; + status->count = req->real_size; + } + nsleeps=1;//reset the number of sleeps we will do next time + } + else { + *flag = 0; + nsleeps++; + } + smpi_mpi_request_free(&request); - DEBUG6("wait for request %p (%p: %p) [src = %d, dst = %d, tag = %d]", - *request, (*request)->pair, data, (*request)->src, (*request)->dst, (*request)->tag); - // data is null if receiver waits before sender enters the rdv - if(data == MPI_REQUEST_NULL || data->complete == 0) { - SIMIX_network_wait((*request)->pair, -1.0); + return; +} + +void smpi_mpi_wait(MPI_Request * request, MPI_Status * status) +{ + print_request("Waiting", *request); + if ((*request)->action != NULL) { // this is not a detached send + simcall_comm_wait((*request)->action, -1.0); } finish_wait(request, status); + request=MPI_REQUEST_NULL; + // FIXME for a detached send, finish_wait is not called: } -int smpi_mpi_waitany(int count, MPI_Request requests[], MPI_Status* status) { +int smpi_mpi_waitany(int count, MPI_Request requests[], + MPI_Status * status) +{ xbt_dynar_t comms; - MPI_Request data; int i, size, index; - int* map; + int *map; index = MPI_UNDEFINED; if(count > 0) { - // First check for already completed requests + // Wait for a request to complete + comms = xbt_dynar_new(sizeof(smx_action_t), NULL); + map = xbt_new(int, count); + size = 0; + XBT_DEBUG("Wait for one of %d", count); for(i = 0; i < count; i++) { if(requests[i] != MPI_REQUEST_NULL) { - data = (MPI_Request)SIMIX_communication_get_data(requests[i]->pair); - if(data != MPI_REQUEST_NULL && data->complete == 1) { - index = i; - break; - } - } - } - if(index == MPI_UNDEFINED) { - // Otherwise, wait for a request to complete - comms = xbt_dynar_new(sizeof(smx_comm_t), NULL); - map = xbt_new(int, count); - size = 0; - DEBUG0("Wait for one of"); - for(i = 0; i < count; i++) { - if(requests[i] != MPI_REQUEST_NULL && requests[i]->complete == 0) { - DEBUG4(" request %p [src = %d, dst = %d, tag = %d]", - requests[i], requests[i]->src, requests[i]->dst, requests[i]->tag); - xbt_dynar_push(comms, &requests[i]->pair); + if (requests[i]->action != NULL) { + XBT_DEBUG("Waiting any %p ", requests[i]); + xbt_dynar_push(comms, &requests[i]->action); map[size] = i; size++; - } + }else{ + //This is a finished detached request, let's return this one + size=0;//so we free the dynar but don't do the waitany call + index=i; + finish_wait(&requests[i], status);//cleanup if refcount = 0 + requests[i]=MPI_REQUEST_NULL;//set to null + break; + } } - if(size > 0) { - index = SIMIX_network_waitany(comms); - index = map[index]; - } - xbt_free(map); - xbt_dynar_free_container(&comms); } - if(index != MPI_UNDEFINED) { - finish_wait(&requests[index], status); + if(size > 0) { + i = simcall_comm_waitany(comms); + + // not MPI_UNDEFINED, as this is a simix return code + if (i != -1) { + index = map[i]; + finish_wait(&requests[index], status); + } } + xbt_free(map); + xbt_dynar_free(&comms); } + + if (index==MPI_UNDEFINED) + smpi_empty_status(status); + return index; } -void smpi_mpi_waitall(int count, MPI_Request requests[], MPI_Status status[]) { - int index; +int smpi_mpi_waitall(int count, MPI_Request requests[], + MPI_Status status[]) +{ + int index, c; MPI_Status stat; - - while(count > 0) { - index = smpi_mpi_waitany(count, requests, &stat); - if(index == MPI_UNDEFINED) { - break; + MPI_Status *pstat = status == MPI_STATUSES_IGNORE ? MPI_STATUS_IGNORE : &stat; + int retvalue = MPI_SUCCESS; + //tag invalid requests in the set + if (status != MPI_STATUSES_IGNORE) { + for (c = 0; c < count; c++) { + if (requests[c] == MPI_REQUEST_NULL || requests[c]->dst == MPI_PROC_NULL) { + smpi_empty_status(&status[c]); + } else if (requests[c]->src == MPI_PROC_NULL) { + smpi_empty_status(&status[c]); + status[c].MPI_SOURCE = MPI_PROC_NULL; + } } - if(status != MPI_STATUS_IGNORE) { - memcpy(&status[index], &stat, sizeof(stat)); + } + for(c = 0; c < count; c++) { + if (MC_is_active()) { + smpi_mpi_wait(&requests[c], pstat); + index = c; + } else { + index = smpi_mpi_waitany(count, requests, pstat); + if (index == MPI_UNDEFINED) + break; + requests[index]=MPI_REQUEST_NULL; + } + if (status != MPI_STATUSES_IGNORE) { + status[index] = *pstat; + if (status[index].MPI_ERROR == MPI_ERR_TRUNCATE) + retvalue = MPI_ERR_IN_STATUS; } - // Move the last request to the found position - requests[index] = requests[count - 1]; - requests[count - 1] = MPI_REQUEST_NULL; - count--; } + + return retvalue; } -int smpi_mpi_waitsome(int incount, MPI_Request requests[], int* indices, MPI_Status status[]) { - MPI_Request data; - int i, count; +int smpi_mpi_waitsome(int incount, MPI_Request requests[], int *indices, + MPI_Status status[]) +{ + int i, count, index; + MPI_Status stat; + MPI_Status *pstat = status == MPI_STATUSES_IGNORE ? MPI_STATUS_IGNORE : &stat; count = 0; - for(i = 0; i < count; i++) { - if(requests[i] != MPI_REQUEST_NULL) { - data = (MPI_Request)SIMIX_communication_get_data(requests[i]->pair); - if(data != MPI_REQUEST_NULL && data->complete == 1) { - finish_wait(&requests[i], status != MPI_STATUS_IGNORE ? &status[i] : MPI_STATUS_IGNORE); - indices[count] = i; - count++; + for(i = 0; i < incount; i++) + { + index=smpi_mpi_waitany(incount, requests, pstat); + if(index!=MPI_UNDEFINED){ + indices[count] = index; + count++; + if(status != MPI_STATUSES_IGNORE) { + status[index] = *pstat; } + requests[index]=MPI_REQUEST_NULL; + }else{ + return MPI_UNDEFINED; } } return count; } -void smpi_mpi_bcast(void* buf, int count, MPI_Datatype datatype, int root, MPI_Comm comm) { +int smpi_mpi_testsome(int incount, MPI_Request requests[], int *indices, + MPI_Status status[]) +{ + int i, count, count_dead; + MPI_Status stat; + MPI_Status *pstat = status == MPI_STATUSES_IGNORE ? MPI_STATUS_IGNORE : &stat; + + count = 0; + count_dead = 0; + for(i = 0; i < incount; i++) { + if((requests[i] != MPI_REQUEST_NULL)) { + if(smpi_mpi_test(&requests[i], pstat)) { + indices[i] = 1; + count++; + if(status != MPI_STATUSES_IGNORE) { + status[i] = *pstat; + } + requests[i]=MPI_REQUEST_NULL; + + } + }else{ + count_dead++; + } + } + if(count_dead==incount)return MPI_UNDEFINED; + else return count; +} + +void smpi_mpi_bcast(void *buf, int count, MPI_Datatype datatype, int root, + MPI_Comm comm) +{ // arity=2: a binary tree, arity=4 seem to be a good setting (see P2P-MPI)) nary_tree_bcast(buf, count, datatype, root, comm, 4); } -void smpi_mpi_barrier(MPI_Comm comm) { +void smpi_mpi_barrier(MPI_Comm comm) +{ // arity=2: a binary tree, arity=4 seem to be a good setting (see P2P-MPI)) nary_tree_barrier(comm, 4); } -void smpi_mpi_gather(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm) { - int system_tag = 666; - int rank, size, src, index, sendsize, recvsize; - MPI_Request* requests; +void smpi_mpi_gather(void *sendbuf, int sendcount, MPI_Datatype sendtype, + void *recvbuf, int recvcount, MPI_Datatype recvtype, + int root, MPI_Comm comm) +{ + int system_tag = COLL_TAG_GATHER; + int rank, size, src, index; + MPI_Aint lb = 0, recvext = 0; + MPI_Request *requests; rank = smpi_comm_rank(comm); size = smpi_comm_size(comm); @@ -272,29 +928,62 @@ void smpi_mpi_gather(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* // Send buffer to root smpi_mpi_send(sendbuf, sendcount, sendtype, root, system_tag, comm); } else { - sendsize = smpi_datatype_size(sendtype); - recvsize = smpi_datatype_size(recvtype); + // FIXME: check for errors + smpi_datatype_extent(recvtype, &lb, &recvext); // Local copy from root - memcpy(&((char*)recvbuf)[root * recvcount * recvsize], sendbuf, sendcount * sendsize * sizeof(char)); + smpi_datatype_copy(sendbuf, sendcount, sendtype, + (char *)recvbuf + root * recvcount * recvext, recvcount, recvtype); // Receive buffers from senders requests = xbt_new(MPI_Request, size - 1); index = 0; for(src = 0; src < size; src++) { if(src != root) { - requests[index] = smpi_mpi_irecv(&((char*)recvbuf)[src * recvcount * recvsize], recvcount, recvtype, src, system_tag, comm); + requests[index] = smpi_irecv_init((char *)recvbuf + src * recvcount * recvext, + recvcount, recvtype, + src, system_tag, comm); index++; } } // Wait for completion of irecv's. + smpi_mpi_startall(size - 1, requests); smpi_mpi_waitall(size - 1, requests, MPI_STATUS_IGNORE); xbt_free(requests); } } -void smpi_mpi_gatherv(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int* recvcounts, int* displs, MPI_Datatype recvtype, int root, MPI_Comm comm) { - int system_tag = 666; - int rank, size, src, index, sendsize; - MPI_Request* requests; + +void smpi_mpi_reduce_scatter(void *sendbuf, void *recvbuf, int *recvcounts, + MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) +{ + int i, size, count; + int *displs; + int rank = smpi_process_index(); + void *tmpbuf; + + /* arbitrarily choose root as rank 0 */ + size = smpi_comm_size(comm); + count = 0; + displs = xbt_new(int, size); + for (i = 0; i < size; i++) { + displs[i] = count; + count += recvcounts[i]; + } + tmpbuf=(void*)xbt_malloc(count*smpi_datatype_get_extent(datatype)); + mpi_coll_reduce_fun(sendbuf, tmpbuf, count, datatype, op, 0, comm); + smpi_mpi_scatterv(tmpbuf, recvcounts, displs, datatype, recvbuf, + recvcounts[rank], datatype, 0, comm); + xbt_free(displs); + xbt_free(tmpbuf); +} + +void smpi_mpi_gatherv(void *sendbuf, int sendcount, MPI_Datatype sendtype, + void *recvbuf, int *recvcounts, int *displs, + MPI_Datatype recvtype, int root, MPI_Comm comm) +{ + int system_tag = COLL_TAG_GATHERV; + int rank, size, src, index; + MPI_Aint lb = 0, recvext = 0; + MPI_Request *requests; rank = smpi_comm_rank(comm); size = smpi_comm_size(comm); @@ -302,226 +991,402 @@ void smpi_mpi_gatherv(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* // Send buffer to root smpi_mpi_send(sendbuf, sendcount, sendtype, root, system_tag, comm); } else { - sendsize = smpi_datatype_size(sendtype); + // FIXME: check for errors + smpi_datatype_extent(recvtype, &lb, &recvext); // Local copy from root - memcpy(&((char*)recvbuf)[displs[root]], sendbuf, sendcount * sendsize * sizeof(char)); + smpi_datatype_copy(sendbuf, sendcount, sendtype, + (char *)recvbuf + displs[root] * recvext, + recvcounts[root], recvtype); // Receive buffers from senders requests = xbt_new(MPI_Request, size - 1); index = 0; for(src = 0; src < size; src++) { if(src != root) { - requests[index] = smpi_mpi_irecv(&((char*)recvbuf)[displs[src]], recvcounts[src], recvtype, src, system_tag, comm); + requests[index] = + smpi_irecv_init((char *)recvbuf + displs[src] * recvext, + recvcounts[src], recvtype, src, system_tag, comm); index++; } } // Wait for completion of irecv's. + smpi_mpi_startall(size - 1, requests); smpi_mpi_waitall(size - 1, requests, MPI_STATUS_IGNORE); xbt_free(requests); } } -void smpi_mpi_allgather(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, MPI_Comm comm) { - int system_tag = 666; - int rank, size, other, index, sendsize, recvsize; - MPI_Request* requests; +void smpi_mpi_allgather(void *sendbuf, int sendcount, + MPI_Datatype sendtype, void *recvbuf, + int recvcount, MPI_Datatype recvtype, + MPI_Comm comm) +{ + int system_tag = COLL_TAG_ALLGATHER; + int rank, size, other, index; + MPI_Aint lb = 0, recvext = 0; + MPI_Request *requests; rank = smpi_comm_rank(comm); size = smpi_comm_size(comm); - sendsize = smpi_datatype_size(sendtype); - recvsize = smpi_datatype_size(recvtype); + // FIXME: check for errors + smpi_datatype_extent(recvtype, &lb, &recvext); // Local copy from self - memcpy(&((char*)recvbuf)[rank * recvcount * recvsize], sendbuf, sendcount * sendsize * sizeof(char)); + smpi_datatype_copy(sendbuf, sendcount, sendtype, + (char *)recvbuf + rank * recvcount * recvext, recvcount, + recvtype); // Send/Recv buffers to/from others; requests = xbt_new(MPI_Request, 2 * (size - 1)); index = 0; for(other = 0; other < size; other++) { if(other != rank) { - requests[index] = smpi_mpi_isend(sendbuf, sendcount, sendtype, other, system_tag, comm); + requests[index] = + smpi_isend_init(sendbuf, sendcount, sendtype, other, system_tag, + comm); index++; - requests[index] = smpi_mpi_irecv(&((char*)recvbuf)[other * recvcount * recvsize], recvcount, recvtype, other, system_tag, comm); + requests[index] = smpi_irecv_init((char *)recvbuf + other * recvcount * recvext, + recvcount, recvtype, other, + system_tag, comm); index++; } } // Wait for completion of all comms. + smpi_mpi_startall(2 * (size - 1), requests); smpi_mpi_waitall(2 * (size - 1), requests, MPI_STATUS_IGNORE); xbt_free(requests); } -void smpi_mpi_allgatherv(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int* recvcounts, int* displs, MPI_Datatype recvtype, MPI_Comm comm) { - int system_tag = 666; - int rank, size, other, index, sendsize, recvsize; - MPI_Request* requests; +void smpi_mpi_allgatherv(void *sendbuf, int sendcount, + MPI_Datatype sendtype, void *recvbuf, + int *recvcounts, int *displs, + MPI_Datatype recvtype, MPI_Comm comm) +{ + int system_tag = COLL_TAG_ALLGATHERV; + int rank, size, other, index; + MPI_Aint lb = 0, recvext = 0; + MPI_Request *requests; rank = smpi_comm_rank(comm); size = smpi_comm_size(comm); - sendsize = smpi_datatype_size(sendtype); - recvsize = smpi_datatype_size(recvtype); + // FIXME: check for errors + smpi_datatype_extent(recvtype, &lb, &recvext); // Local copy from self - memcpy(&((char*)recvbuf)[displs[rank]], sendbuf, sendcount * sendsize * sizeof(char)); + smpi_datatype_copy(sendbuf, sendcount, sendtype, + (char *)recvbuf + displs[rank] * recvext, + recvcounts[rank], recvtype); // Send buffers to others; requests = xbt_new(MPI_Request, 2 * (size - 1)); index = 0; for(other = 0; other < size; other++) { if(other != rank) { - requests[index] = smpi_mpi_isend(sendbuf, sendcount, sendtype, other, system_tag, comm); + requests[index] = + smpi_isend_init(sendbuf, sendcount, sendtype, other, system_tag, + comm); index++; - requests[index] = smpi_mpi_irecv(&((char*)recvbuf)[displs[other]], recvcounts[other], recvtype, other, system_tag, comm); + requests[index] = + smpi_irecv_init((char *)recvbuf + displs[other] * recvext, recvcounts[other], + recvtype, other, system_tag, comm); index++; } } // Wait for completion of all comms. + smpi_mpi_startall(2 * (size - 1), requests); smpi_mpi_waitall(2 * (size - 1), requests, MPI_STATUS_IGNORE); xbt_free(requests); } -void smpi_mpi_scatter(void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm) { - int system_tag = 666; - int rank, size, dst, index, sendsize, recvsize; - MPI_Request* requests; +void smpi_mpi_scatter(void *sendbuf, int sendcount, MPI_Datatype sendtype, + void *recvbuf, int recvcount, MPI_Datatype recvtype, + int root, MPI_Comm comm) +{ + int system_tag = COLL_TAG_SCATTER; + int rank, size, dst, index; + MPI_Aint lb = 0, sendext = 0; + MPI_Request *requests; rank = smpi_comm_rank(comm); size = smpi_comm_size(comm); if(rank != root) { // Recv buffer from root - smpi_mpi_recv(recvbuf, recvcount, recvtype, root, system_tag, comm, MPI_STATUS_IGNORE); + smpi_mpi_recv(recvbuf, recvcount, recvtype, root, system_tag, comm, + MPI_STATUS_IGNORE); } else { - sendsize = smpi_datatype_size(sendtype); - recvsize = smpi_datatype_size(recvtype); + // FIXME: check for errors + smpi_datatype_extent(sendtype, &lb, &sendext); // Local copy from root - memcpy(recvbuf, &((char*)sendbuf)[root * sendcount * sendsize], recvcount * recvsize * sizeof(char)); + if(recvbuf!=MPI_IN_PLACE){ + smpi_datatype_copy((char *)sendbuf + root * sendcount * sendext, + sendcount, sendtype, recvbuf, recvcount, recvtype); + } // Send buffers to receivers requests = xbt_new(MPI_Request, size - 1); index = 0; for(dst = 0; dst < size; dst++) { if(dst != root) { - requests[index] = smpi_mpi_isend(&((char*)sendbuf)[dst * sendcount * sendsize], sendcount, sendtype, dst, system_tag, comm); + requests[index] = smpi_isend_init((char *)sendbuf + dst * sendcount * sendext, + sendcount, sendtype, dst, + system_tag, comm); index++; } } // Wait for completion of isend's. + smpi_mpi_startall(size - 1, requests); smpi_mpi_waitall(size - 1, requests, MPI_STATUS_IGNORE); xbt_free(requests); } } -void smpi_mpi_scatterv(void* sendbuf, int* sendcounts, int* displs, MPI_Datatype sendtype, void* recvbuf, int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm) { - int system_tag = 666; - int rank, size, dst, index, sendsize, recvsize; - MPI_Request* requests; +void smpi_mpi_scatterv(void *sendbuf, int *sendcounts, int *displs, + MPI_Datatype sendtype, void *recvbuf, int recvcount, + MPI_Datatype recvtype, int root, MPI_Comm comm) +{ + int system_tag = COLL_TAG_SCATTERV; + int rank, size, dst, index; + MPI_Aint lb = 0, sendext = 0; + MPI_Request *requests; rank = smpi_comm_rank(comm); size = smpi_comm_size(comm); if(rank != root) { // Recv buffer from root - smpi_mpi_recv(recvbuf, recvcount, recvtype, root, system_tag, comm, MPI_STATUS_IGNORE); + smpi_mpi_recv(recvbuf, recvcount, recvtype, root, system_tag, comm, + MPI_STATUS_IGNORE); } else { - sendsize = smpi_datatype_size(sendtype); - recvsize = smpi_datatype_size(recvtype); + // FIXME: check for errors + smpi_datatype_extent(sendtype, &lb, &sendext); // Local copy from root - memcpy(recvbuf, &((char*)sendbuf)[displs[root]], recvcount * recvsize * sizeof(char)); + if(recvbuf!=MPI_IN_PLACE){ + smpi_datatype_copy((char *)sendbuf + displs[root] * sendext, sendcounts[root], + sendtype, recvbuf, recvcount, recvtype); + } // Send buffers to receivers requests = xbt_new(MPI_Request, size - 1); index = 0; for(dst = 0; dst < size; dst++) { if(dst != root) { - requests[index] = smpi_mpi_isend(&((char*)sendbuf)[displs[dst]], sendcounts[dst], sendtype, dst, system_tag, comm); + requests[index] = + smpi_isend_init((char *)sendbuf + displs[dst] * sendext, sendcounts[dst], + sendtype, dst, system_tag, comm); index++; } } // Wait for completion of isend's. + smpi_mpi_startall(size - 1, requests); smpi_mpi_waitall(size - 1, requests, MPI_STATUS_IGNORE); xbt_free(requests); } } -void smpi_mpi_reduce(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, MPI_Comm comm) { - int system_tag = 666; - int rank, size, src, index, datasize; - MPI_Request* requests; - void** tmpbufs; - +void smpi_mpi_reduce(void *sendbuf, void *recvbuf, int count, + MPI_Datatype datatype, MPI_Op op, int root, + MPI_Comm comm) +{ + int system_tag = COLL_TAG_REDUCE; + int rank, size, src, index; + MPI_Aint lb = 0, dataext = 0; + MPI_Request *requests; + void **tmpbufs; + + + char* sendtmpbuf = (char*) sendbuf; + if( sendbuf == MPI_IN_PLACE ) { + sendtmpbuf = (char *)xbt_malloc(count*smpi_datatype_get_extent(datatype)); + smpi_datatype_copy(recvbuf, count, datatype,sendtmpbuf, count, datatype); + } + rank = smpi_comm_rank(comm); size = smpi_comm_size(comm); + //non commutative case, use a working algo from openmpi + if(!smpi_op_is_commute(op)){ + smpi_coll_tuned_reduce_ompi_basic_linear(sendtmpbuf, recvbuf, count, + datatype, op, root, comm); + return; + } + if(rank != root) { // Send buffer to root - smpi_mpi_send(sendbuf, count, datatype, root, system_tag, comm); + smpi_mpi_send(sendtmpbuf, count, datatype, root, system_tag, comm); } else { - datasize = smpi_datatype_size(datatype); + // FIXME: check for errors + smpi_datatype_extent(datatype, &lb, &dataext); // Local copy from root - memcpy(recvbuf, sendbuf, count * datasize * sizeof(char)); + if (sendtmpbuf && recvbuf) + smpi_datatype_copy(sendtmpbuf, count, datatype, recvbuf, count, datatype); // Receive buffers from senders //TODO: make a MPI_barrier here ? requests = xbt_new(MPI_Request, size - 1); - tmpbufs = xbt_new(void*, size - 1); + tmpbufs = xbt_new(void *, size - 1); index = 0; for(src = 0; src < size; src++) { if(src != root) { - tmpbufs[index] = xbt_malloc(count * datasize); - requests[index] = smpi_mpi_irecv(tmpbufs[index], count, datatype, src, system_tag, comm); + // FIXME: possibly overkill we we have contiguous/noncontiguous data + // mapping... + tmpbufs[index] = xbt_malloc(count * dataext); + requests[index] = + smpi_irecv_init(tmpbufs[index], count, datatype, src, + system_tag, comm); index++; } } // Wait for completion of irecv's. + smpi_mpi_startall(size - 1, requests); for(src = 0; src < size - 1; src++) { index = smpi_mpi_waitany(size - 1, requests, MPI_STATUS_IGNORE); + XBT_DEBUG("finished waiting any request with index %d", index); if(index == MPI_UNDEFINED) { break; } - smpi_op_apply(op, tmpbufs[index], recvbuf, &count, &datatype); + if(op) /* op can be MPI_OP_NULL that does nothing */ + smpi_op_apply(op, tmpbufs[index], recvbuf, &count, &datatype); } for(index = 0; index < size - 1; index++) { xbt_free(tmpbufs[index]); } xbt_free(tmpbufs); xbt_free(requests); + + if( sendbuf == MPI_IN_PLACE ) { + xbt_free(sendtmpbuf); + } } } -void smpi_mpi_allreduce(void* sendbuf, void* recvbuf, int count, MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) { +void smpi_mpi_allreduce(void *sendbuf, void *recvbuf, int count, + MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) +{ smpi_mpi_reduce(sendbuf, recvbuf, count, datatype, op, 0, comm); smpi_mpi_bcast(recvbuf, count, datatype, 0, comm); +} -/* -FIXME: buggy implementation +void smpi_mpi_scan(void *sendbuf, void *recvbuf, int count, + MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) +{ + int system_tag = -888; + int rank, size, other, index; + MPI_Aint lb = 0, dataext = 0; + MPI_Request *requests; + void **tmpbufs; - int system_tag = 666; - int rank, size, other, index, datasize; - MPI_Request* requests; - void** tmpbufs; - rank = smpi_comm_rank(comm); size = smpi_comm_size(comm); - datasize = smpi_datatype_size(datatype); + + // FIXME: check for errors + smpi_datatype_extent(datatype, &lb, &dataext); + // Local copy from self - memcpy(recvbuf, sendbuf, count * datasize * sizeof(char)); + smpi_datatype_copy(sendbuf, count, datatype, recvbuf, count, datatype); + // Send/Recv buffers to/from others; - //TODO: make a MPI_barrier here ? - requests = xbt_new(MPI_Request, 2 * (size - 1)); - tmpbufs = xbt_new(void*, size - 1); + requests = xbt_new(MPI_Request, size - 1); + tmpbufs = xbt_new(void *, rank); index = 0; - for(other = 0; other < size; other++) { - if(other != rank) { - tmpbufs[index / 2] = xbt_malloc(count * datasize); - requests[index] = smpi_mpi_isend(sendbuf, count, datatype, other, system_tag, comm); - requests[index + 1] = smpi_mpi_irecv(tmpbufs[index / 2], count, datatype, other, system_tag, comm); - index += 2; + for(other = 0; other < rank; other++) { + // FIXME: possibly overkill we we have contiguous/noncontiguous data + // mapping... + tmpbufs[index] = xbt_malloc(count * dataext); + requests[index] = + smpi_irecv_init(tmpbufs[index], count, datatype, other, system_tag, + comm); + index++; + } + for(other = rank + 1; other < size; other++) { + requests[index] = + smpi_isend_init(sendbuf, count, datatype, other, system_tag, comm); + index++; + } + // Wait for completion of all comms. + smpi_mpi_startall(size - 1, requests); + + if(smpi_op_is_commute(op)){ + for(other = 0; other < size - 1; other++) { + index = smpi_mpi_waitany(size - 1, requests, MPI_STATUS_IGNORE); + if(index == MPI_UNDEFINED) { + break; + } + if(index < rank) { + // #Request is below rank: it's a irecv + smpi_op_apply(op, tmpbufs[index], recvbuf, &count, &datatype); + } + } + }else{ + //non commutative case, wait in order + for(other = 0; other < size - 1; other++) { + smpi_mpi_wait(&(requests[other]), MPI_STATUS_IGNORE); + if(index < rank) { + smpi_op_apply(op, tmpbufs[other], recvbuf, &count, &datatype); + } } } + for(index = 0; index < rank; index++) { + xbt_free(tmpbufs[index]); + } + xbt_free(tmpbufs); + xbt_free(requests); +} + +void smpi_mpi_exscan(void *sendbuf, void *recvbuf, int count, + MPI_Datatype datatype, MPI_Op op, MPI_Comm comm) +{ + int system_tag = -888; + int rank, size, other, index; + MPI_Aint lb = 0, dataext = 0; + MPI_Request *requests; + void **tmpbufs; + int recvbuf_is_empty=1; + rank = smpi_comm_rank(comm); + size = smpi_comm_size(comm); + + // FIXME: check for errors + smpi_datatype_extent(datatype, &lb, &dataext); + + // Send/Recv buffers to/from others; + requests = xbt_new(MPI_Request, size - 1); + tmpbufs = xbt_new(void *, rank); + index = 0; + for(other = 0; other < rank; other++) { + // FIXME: possibly overkill we we have contiguous/noncontiguous data + // mapping... + tmpbufs[index] = xbt_malloc(count * dataext); + requests[index] = + smpi_irecv_init(tmpbufs[index], count, datatype, other, system_tag, + comm); + index++; + } + for(other = rank + 1; other < size; other++) { + requests[index] = + smpi_isend_init(sendbuf, count, datatype, other, system_tag, comm); + index++; + } // Wait for completion of all comms. - for(other = 0; other < 2 * (size - 1); other++) { - index = smpi_mpi_waitany(size - 1, requests, MPI_STATUS_IGNORE); - if(index == MPI_UNDEFINED) { - break; + smpi_mpi_startall(size - 1, requests); + if(smpi_op_is_commute(op)){ + for(other = 0; other < size - 1; other++) { + index = smpi_mpi_waitany(size - 1, requests, MPI_STATUS_IGNORE); + if(index == MPI_UNDEFINED) { + break; + } + if(index < rank) { + if(recvbuf_is_empty){ + smpi_datatype_copy(tmpbufs[index], count, datatype, recvbuf, count, datatype); + recvbuf_is_empty=0; + }else + // #Request is below rank: it's a irecv + smpi_op_apply(op, tmpbufs[index], recvbuf, &count, &datatype); + } } - if((index & 1) == 1) { - // Request is odd: it's a irecv - smpi_op_apply(op, tmpbufs[index / 2], recvbuf, &count, &datatype); + }else{ + //non commutative case, wait in order + for(other = 0; other < size - 1; other++) { + smpi_mpi_wait(&(requests[other]), MPI_STATUS_IGNORE); + if(index < rank) { + if(recvbuf_is_empty){ + smpi_datatype_copy(tmpbufs[other], count, datatype, recvbuf, count, datatype); + recvbuf_is_empty=0; + }else smpi_op_apply(op, tmpbufs[other], recvbuf, &count, &datatype); + } } } - for(index = 0; index < size - 1; index++) { + for(index = 0; index < rank; index++) { xbt_free(tmpbufs[index]); } xbt_free(tmpbufs); xbt_free(requests); -*/ }