X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/18b0c9100f020d980f681a3e99eed106fba87804..48af4c8427d171d3e05d39116b02e441f2391374:/src/smpi/smpi_replay.c diff --git a/src/smpi/smpi_replay.c b/src/smpi/smpi_replay.c index 27770ca4cd..df4f3f5b27 100644 --- a/src/smpi/smpi_replay.c +++ b/src/smpi/smpi_replay.c @@ -1,10 +1,9 @@ -/* Copyright (c) 2009-2013. The SimGrid Team. +/* Copyright (c) 2009-2014. 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 #include @@ -27,11 +26,6 @@ static void log_timed_action (const char *const *action, double clock){ } } -typedef struct { - xbt_dynar_t irecvs; /* of MPI_Request */ -} s_smpi_replay_globals_t, *smpi_replay_globals_t; - - /* Helper function */ static double parse_double(const char *string) { @@ -108,14 +102,10 @@ static void action_init(const char *const *action) { int i; XBT_DEBUG("Initialize the counters"); - smpi_replay_globals_t globals = xbt_new(s_smpi_replay_globals_t, 1); - globals->irecvs = xbt_dynar_new(sizeof(MPI_Request),NULL); if(action[2]) MPI_DEFAULT_TYPE= MPI_DOUBLE; // default MPE dataype else MPI_DEFAULT_TYPE= MPI_BYTE; // default TAU datatype - smpi_process_set_user_data((void*) globals); - /* start a simulated timer */ smpi_process_simulated_start(); /*initialize the number of active processes */ @@ -125,21 +115,13 @@ static void action_init(const char *const *action) reqq=xbt_new0(xbt_dynar_t,active_processes); for(i=0;iirecvs)); - xbt_dynar_free_container(&(globals->irecvs)); - } - free(globals); } static void action_comm_size(const char *const *action) @@ -276,6 +258,12 @@ static void action_recv(const char *const *action) { TRACE_smpi_ptp_in(rank, src_traced, rank, __FUNCTION__, extra); #endif + //unknow size from the receiver pov + if(size==-1){ + smpi_mpi_probe(from, 0, MPI_COMM_WORLD, &status); + size=status.count; + } + smpi_mpi_recv(NULL, size, MPI_CURRENT_TYPE, from, 0, MPI_COMM_WORLD, &status); #ifdef HAVE_TRACING @@ -293,9 +281,6 @@ static void action_Irecv(const char *const *action) double clock = smpi_process_simulated_elapsed(); MPI_Request request; - smpi_replay_globals_t globals = - (smpi_replay_globals_t) smpi_process_get_user_data(); - if(action[4]) MPI_CURRENT_TYPE=decode_datatype(action[4]); else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE; @@ -310,6 +295,12 @@ static void action_Irecv(const char *const *action) extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); TRACE_smpi_ptp_in(rank, src_traced, rank, __FUNCTION__, extra); #endif + MPI_Status status; + //unknow size from the receiver pov + if(size==-1){ + smpi_mpi_probe(from, 0, MPI_COMM_WORLD, &status); + size=status.count; + } request = smpi_mpi_irecv(NULL, size, MPI_CURRENT_TYPE, from, 0, MPI_COMM_WORLD); @@ -317,25 +308,59 @@ static void action_Irecv(const char *const *action) TRACE_smpi_ptp_out(rank, src_traced, rank, __FUNCTION__); request->recv = 1; #endif - xbt_dynar_push(globals->irecvs,&request); xbt_dynar_push(reqq[smpi_comm_rank(MPI_COMM_WORLD)],&request); log_timed_action (action, clock); } +static void action_test(const char *const *action){ + double clock = smpi_process_simulated_elapsed(); + MPI_Request request; + MPI_Status status; + int flag = TRUE; + + request = xbt_dynar_pop_as(reqq[smpi_comm_rank(MPI_COMM_WORLD)],MPI_Request); + xbt_assert(request != NULL, "found null request in reqq"); + +#ifdef HAVE_TRACING + int rank = smpi_comm_rank(MPI_COMM_WORLD); + instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); + extra->type=TRACING_TEST; + TRACE_smpi_testing_in(rank, extra); +#endif + flag = smpi_mpi_test(&request, &status); + XBT_DEBUG("MPI_Test result: %d", flag); + /* push back request in dynar to be caught by a subsequent wait. if the test + * did succeed, the request is now NULL. + */ + xbt_dynar_push_as(reqq[smpi_comm_rank(MPI_COMM_WORLD)],MPI_Request, request); + +#ifdef HAVE_TRACING + TRACE_smpi_testing_out(rank); +#endif + + log_timed_action (action, clock); +} + static void action_wait(const char *const *action){ double clock = smpi_process_simulated_elapsed(); MPI_Request request; MPI_Status status; - smpi_replay_globals_t globals = - (smpi_replay_globals_t) smpi_process_get_user_data(); - xbt_assert(xbt_dynar_length(globals->irecvs), - "action wait not preceded by any irecv: %s", + xbt_assert(xbt_dynar_length(reqq[smpi_comm_rank(MPI_COMM_WORLD)]), + "action wait not preceded by any irecv or isend: %s", xbt_str_join_array(action," ")); - request = xbt_dynar_pop_as(globals->irecvs,MPI_Request); + request = xbt_dynar_pop_as(reqq[smpi_comm_rank(MPI_COMM_WORLD)],MPI_Request); + + if (!request){ + /* Assuming that the trace is well formed, this mean the comm might have + * been caught by a MPI_test. Then just return. + */ + return; + } + #ifdef HAVE_TRACING - int rank = request && request->comm != MPI_COMM_NULL + int rank = request->comm != MPI_COMM_NULL ? smpi_comm_rank(request->comm) : -1; @@ -427,7 +452,7 @@ static void action_waitall(const char *const *action){ xbt_dynar_free(&recvs); #endif - xbt_dynar_reset(reqq[smpi_comm_rank(MPI_COMM_WORLD)]); + xbt_dynar_free_container(&(reqq[smpi_comm_rank(MPI_COMM_WORLD)])); } log_timed_action (action, clock); } @@ -620,7 +645,7 @@ static void action_gather(const char *const *action) { MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE; } void *send = calloc(send_size, smpi_datatype_size(MPI_CURRENT_TYPE)); - void *recv = calloc(recv_size, smpi_datatype_size(MPI_CURRENT_TYPE2)); + void *recv = NULL; int root=atoi(action[4]); int rank = smpi_process_index(); @@ -653,6 +678,79 @@ smpi_mpi_gather(send, send_size, MPI_CURRENT_TYPE, } + +static void action_gatherv(const char *const *action) { + /* + The structure of the gatherv action for the rank 0 (total 4 processes) + is the following: + 0 gather 68 68 10 10 10 0 0 0 + + where: + 1) 68 is the sendcount + 2) 68 10 10 10 is the recvcounts + 3) 0 is the root node + 4) 0 is the send datatype id, see decode_datatype() + 5) 0 is the recv datatype id, see decode_datatype() + */ + double clock = smpi_process_simulated_elapsed(); + int comm_size = smpi_comm_size(MPI_COMM_WORLD); + int send_size = parse_double(action[2]); + int *disps = xbt_new0(int, comm_size); + int *recvcounts = xbt_new0(int, comm_size); + int i=0,recv_sum=0; + + MPI_Datatype MPI_CURRENT_TYPE2; + if(action[4+comm_size]) { + MPI_CURRENT_TYPE=decode_datatype(action[4+comm_size]); + MPI_CURRENT_TYPE2=decode_datatype(action[5+comm_size]); + } else { + MPI_CURRENT_TYPE=MPI_DEFAULT_TYPE; + MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE; + } + void *send = calloc(send_size, smpi_datatype_size(MPI_CURRENT_TYPE)); + void *recv = NULL; + for(i=0;itype = TRACING_GATHERV; + extra->send_size = send_size; + extra->recvcounts= xbt_malloc(comm_size*sizeof(int)); + for(i=0; i< comm_size; i++)//copy data to avoid bad free + extra->recvcounts[i] = recvcounts[i]; + extra->root = root; + extra->num_processes = comm_size; + extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); + extra->datatype2 = encode_datatype(MPI_CURRENT_TYPE2); + + TRACE_smpi_collective_in(rank, root, __FUNCTION__, extra); +#endif +smpi_mpi_gatherv(send, send_size, MPI_CURRENT_TYPE, + recv, recvcounts, disps, MPI_CURRENT_TYPE2, + root, MPI_COMM_WORLD); + +#ifdef HAVE_TRACING + TRACE_smpi_collective_out(rank, -1, __FUNCTION__); +#endif + + log_timed_action (action, clock); + xbt_free(recvcounts); + xbt_free(send); + xbt_free(recv); + xbt_free(disps); + +} + static void action_reducescatter(const char *const *action) { /* @@ -713,7 +811,8 @@ static void action_reducescatter(const char *const *action) { #ifdef HAVE_TRACING TRACE_smpi_collective_out(rank, -1, __FUNCTION__); #endif - + xbt_free(recvcounts); + xbt_free(disps); log_timed_action (action, clock); } @@ -835,11 +934,8 @@ static void action_allToAllv(const char *const *action) { #ifdef HAVE_TRACING int rank = smpi_process_index(); - int count=0; - for(i=0;itype = TRACING_ALLTOALLV; - extra->send_size = count; extra->recvcounts= xbt_malloc(comm_size*sizeof(int)); extra->sendcounts= xbt_malloc(comm_size*sizeof(int)); extra->num_processes = comm_size; @@ -847,6 +943,7 @@ static void action_allToAllv(const char *const *action) { for(i=0; i< comm_size; i++){//copy data to avoid bad free extra->send_size += sendcounts[i]; extra->sendcounts[i] = sendcounts[i]; + extra->recv_size += recvcounts[i]; extra->recvcounts[i] = recvcounts[i]; } extra->datatype1 = encode_datatype(MPI_CURRENT_TYPE); @@ -894,6 +991,7 @@ void smpi_replay_init(int *argc, char***argv){ xbt_replay_action_register("Isend", action_Isend); xbt_replay_action_register("recv", action_recv); xbt_replay_action_register("Irecv", action_Irecv); + xbt_replay_action_register("test", action_test); xbt_replay_action_register("wait", action_wait); xbt_replay_action_register("waitAll", action_waitall); xbt_replay_action_register("barrier", action_barrier); @@ -903,6 +1001,7 @@ void smpi_replay_init(int *argc, char***argv){ xbt_replay_action_register("allToAll", action_allToAll); xbt_replay_action_register("allToAllV", action_allToAllv); xbt_replay_action_register("gather", action_gather); + xbt_replay_action_register("gatherV", action_gatherv); xbt_replay_action_register("allGatherV", action_allgatherv); xbt_replay_action_register("reduceScatter", action_reducescatter); xbt_replay_action_register("compute", action_compute); @@ -914,15 +1013,28 @@ void smpi_replay_init(int *argc, char***argv){ int smpi_replay_finalize(){ double sim_time= 1.; /* One active process will stop. Decrease the counter*/ - active_processes--; XBT_DEBUG("There are %lu elements in reqq[*]", xbt_dynar_length(reqq[smpi_comm_rank(MPI_COMM_WORLD)])); - xbt_dynar_free(&reqq[smpi_comm_rank(MPI_COMM_WORLD)]); + if (!xbt_dynar_is_empty(reqq[smpi_comm_rank(MPI_COMM_WORLD)])){ + int count_requests=xbt_dynar_length(reqq[smpi_comm_rank(MPI_COMM_WORLD)]); + MPI_Request requests[count_requests]; + MPI_Status status[count_requests]; + unsigned int i; + + xbt_dynar_foreach(reqq[smpi_comm_rank(MPI_COMM_WORLD)],i,requests[i]); + smpi_mpi_waitall(count_requests, requests, status); + active_processes--; + } else { + active_processes--; + } + + xbt_dynar_free_container(&(reqq[smpi_comm_rank(MPI_COMM_WORLD)])); + if(!active_processes){ /* Last process alive speaking */ /* end the simulated timer */ sim_time = smpi_process_simulated_elapsed(); - XBT_INFO("Simulation time %g", sim_time); + XBT_INFO("Simulation time %f", sim_time); _xbt_replay_action_exit(); xbt_free(reqq); reqq = NULL;