X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/c5ad8ca1a68bbaa9152471c8d0eeb99d762f0d86..8b9fbe95d804fdef03fd9e09419d568dddcbec3f:/src/smpi/smpi_base.c diff --git a/src/smpi/smpi_base.c b/src/smpi/smpi_base.c index d06a471653..9ff9da7e0a 100644 --- a/src/smpi/smpi_base.c +++ b/src/smpi/smpi_base.c @@ -7,20 +7,14 @@ #include "private.h" #include "xbt/time.h" #include "mc/mc.h" +#include "xbt/replay.h" +#include +#include "surf/surf.h" 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 int match_recv(void* a, void* b) { + +static int match_recv(void* a, void* b, smx_action_t ignored) { MPI_Request ref = (MPI_Request)a; MPI_Request req = (MPI_Request)b; @@ -30,7 +24,7 @@ static int match_recv(void* a, void* b) { && (ref->tag == MPI_ANY_TAG || req->tag == ref->tag); } -static int match_send(void* a, void* b) { +static int match_send(void* a, void* b,smx_action_t ignored) { MPI_Request ref = (MPI_Request)a; MPI_Request req = (MPI_Request)b; @@ -63,9 +57,39 @@ static MPI_Request build_request(void *buf, int count, return request; } +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) { - smpi_mpi_request_free(request); + MPI_Request req = request; + smpi_mpi_request_free(&req); } /* MPI Low level calls */ @@ -98,42 +122,58 @@ void smpi_mpi_start(MPI_Request request) "Cannot (re)start a non-finished communication"); if(request->flags & RECV) { print_request("New recv", request); + if (request->size < xbt_cfg_get_int(_surf_cfg_set, "smpi/async_small_thres")) + mailbox = smpi_process_mailbox_small(); + else mailbox = smpi_process_mailbox(); + // FIXME: SIMIX does not yet support non-contiguous datatypes request->action = simcall_comm_irecv(mailbox, request->buf, &request->size, &match_recv, request); } else { print_request("New send", request); - mailbox = smpi_process_remote_mailbox( - smpi_group_index(smpi_comm_group(request->comm), request->dst)); - // FIXME: SIMIX does not yet support non-contiguous datatypes - if (request->size < 64*1024 ) { // eager mode => detached send (FIXME: this limit should be configurable) - void *oldbuf = request->buf; - detached = 1; - request->buf = 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); - } else { - XBT_DEBUG("Send request %p is not detached (buf: %p)",request,request->buf); + if (request->size < xbt_cfg_get_int(_surf_cfg_set, "smpi/async_small_thres")) { // eager mode => detached send (FIXME: this limit should be configurable) + mailbox = smpi_process_remote_mailbox_small( + smpi_group_index(smpi_comm_group(request->comm), request->dst)); + }else{ + XBT_DEBUG("Send request %p is not in the permanent receive mailbox (buf: %p)",request,request->buf); + mailbox = smpi_process_remote_mailbox( + smpi_group_index(smpi_comm_group(request->comm), request->dst)); + } + if (request->size < 64*1024 ) { //(FIXME: this limit should be configurable) + void *oldbuf = request->buf; + detached = 1; + request->buf = 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); + }else{ + XBT_DEBUG("Send request %p is not detached (buf: %p)",request,request->buf); + mailbox = smpi_process_remote_mailbox( + smpi_group_index(smpi_comm_group(request->comm), request->dst)); + } + + request->action = + simcall_comm_isend(mailbox, request->size, -1.0, + request->buf, request->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 + 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 + } - request->action = - simcall_comm_isend(mailbox, request->size, -1.0, - request->buf, request->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 - detached); -#ifdef HAVE_TRACING - simcall_set_category (request->action, TRACE_internal_smpi_get_category()); -#endif - } } void smpi_mpi_startall(int count, MPI_Request * requests) { - int i; + int i; for(i = 0; i < count; i++) { smpi_mpi_start(requests[i]); @@ -195,13 +235,15 @@ void smpi_mpi_recv(void *buf, int count, MPI_Datatype datatype, int src, smpi_mpi_wait(&request, status); } + + void smpi_mpi_send(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm) { - MPI_Request request; + MPI_Request request; - request = smpi_mpi_isend(buf, count, datatype, dst, tag, comm); - smpi_mpi_wait(&request, MPI_STATUS_IGNORE); + request = smpi_mpi_isend(buf, count, datatype, dst, tag, comm); + smpi_mpi_wait(&request, MPI_STATUS_IGNORE); } void smpi_mpi_sendrecv(void *sendbuf, int sendcount, MPI_Datatype sendtype, @@ -253,13 +295,13 @@ int smpi_mpi_test(MPI_Request * request, MPI_Status * status) { int flag; if ((*request)->action == NULL) - flag = 1; + flag = 1; else flag = simcall_comm_test((*request)->action); if(flag) { - smpi_mpi_wait(request, status); - } - return flag; + smpi_mpi_wait(request, status); + } + return flag; } int smpi_mpi_testany(int count, MPI_Request requests[], int *index, @@ -323,7 +365,7 @@ int smpi_mpi_waitany(int count, MPI_Request requests[], XBT_DEBUG("Wait for one of"); for(i = 0; i < count; i++) { if((requests[i] != MPI_REQUEST_NULL) && (requests[i]->action != NULL)) { - print_request(" ", requests[i]); + print_request("Waiting any ", requests[i]); xbt_dynar_push(comms, &requests[i]->action); map[size] = i; size++; @@ -331,10 +373,12 @@ int smpi_mpi_waitany(int count, MPI_Request requests[], } if(size > 0) { i = simcall_comm_waitany(comms); + // FIXME: MPI_UNDEFINED or does SIMIX have a return code? if (i != MPI_UNDEFINED) { index = map[i]; finish_wait(&requests[index], status); + } } xbt_free(map); @@ -667,6 +711,7 @@ void smpi_mpi_reduce(void *sendbuf, void *recvbuf, int count, smpi_mpi_startall(size - 1, requests); for(src = 0; src < size - 1; src++) { index = smpi_mpi_waitany(size - 1, requests, MPI_STATUS_IGNORE); + XBT_VERB("finished waiting any request with index %d", index); if(index == MPI_UNDEFINED) { break; }