X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/1b4e20873a7bd405842f3d8d1e6dd1d518db5535..bf9897b3f75fba374088221ba582819e2aa503e0:/src/smpi/smpi_replay.c diff --git a/src/smpi/smpi_replay.c b/src/smpi/smpi_replay.c index 507957f0e9..9b5ff76cc8 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 @@ -19,6 +18,11 @@ xbt_dynar_t *reqq = NULL; MPI_Datatype MPI_DEFAULT_TYPE; MPI_Datatype MPI_CURRENT_TYPE; +static int sendbuffer_size=0; +char* sendbuffer=NULL; +static int recvbuffer_size=0; +char* recvbuffer=NULL; + static void log_timed_action (const char *const *action, double clock){ if (XBT_LOG_ISENABLED(smpi_replay, xbt_log_priority_verbose)){ char *name = xbt_str_join_array(action, " "); @@ -27,6 +31,32 @@ static void log_timed_action (const char *const *action, double clock){ } } +//allocate a single buffer for all sends, growing it if needed +void* smpi_get_tmp_sendbuffer(int size){ + if (!_xbt_replay_is_active()) + return xbt_malloc(size); + if (sendbuffer_sizetype=TRACING_COMPUTING; extra->comp_size=flops; @@ -179,7 +209,7 @@ static void action_send(const char *const *action) } #ifdef HAVE_TRACING - int rank = smpi_comm_rank(MPI_COMM_WORLD); + int rank = smpi_process_index(); int dst_traced = smpi_group_rank(smpi_comm_group(MPI_COMM_WORLD), to); instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); @@ -213,7 +243,7 @@ static void action_Isend(const char *const *action) else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE; #ifdef HAVE_TRACING - int rank = smpi_comm_rank(MPI_COMM_WORLD); + int rank = smpi_process_index(); int dst_traced = smpi_group_rank(smpi_comm_group(MPI_COMM_WORLD), to); instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); extra->type = TRACING_ISEND; @@ -232,7 +262,7 @@ static void action_Isend(const char *const *action) request->send = 1; #endif - xbt_dynar_push(reqq[smpi_comm_rank(MPI_COMM_WORLD)],&request); + xbt_dynar_push(reqq[smpi_process_index()],&request); log_timed_action (action, clock); } @@ -247,7 +277,7 @@ static void action_recv(const char *const *action) { else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE; #ifdef HAVE_TRACING - int rank = smpi_comm_rank(MPI_COMM_WORLD); + int rank = smpi_process_index(); int src_traced = smpi_group_rank(smpi_comm_group(MPI_COMM_WORLD), from); instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); @@ -259,6 +289,12 @@ 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 @@ -280,7 +316,7 @@ static void action_Irecv(const char *const *action) else MPI_CURRENT_TYPE= MPI_DEFAULT_TYPE; #ifdef HAVE_TRACING - int rank = smpi_comm_rank(MPI_COMM_WORLD); + int rank = smpi_process_index(); int src_traced = smpi_group_rank(smpi_comm_group(MPI_COMM_WORLD), from); instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); extra->type = TRACING_IRECV; @@ -290,6 +326,12 @@ 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); @@ -297,7 +339,36 @@ static void action_Irecv(const char *const *action) TRACE_smpi_ptp_out(rank, src_traced, rank, __FUNCTION__); request->recv = 1; #endif - xbt_dynar_push(reqq[smpi_comm_rank(MPI_COMM_WORLD)],&request); + xbt_dynar_push(reqq[smpi_process_index()],&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_process_index()],MPI_Request); + xbt_assert(request != NULL, "found null request in reqq"); + +#ifdef HAVE_TRACING + int rank = smpi_process_index(); + 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_process_index()],MPI_Request, request); + +#ifdef HAVE_TRACING + TRACE_smpi_testing_out(rank); +#endif log_timed_action (action, clock); } @@ -307,11 +378,18 @@ static void action_wait(const char *const *action){ MPI_Request request; MPI_Status status; - xbt_assert(xbt_dynar_length(reqq[smpi_comm_rank(MPI_COMM_WORLD)]), + xbt_assert(xbt_dynar_length(reqq[smpi_process_index()]), "action wait not preceded by any irecv or isend: %s", xbt_str_join_array(action," ")); - request = xbt_dynar_pop_as(reqq[smpi_comm_rank(MPI_COMM_WORLD)],MPI_Request); - xbt_assert(request != NULL, "found null request in reqq"); + request = xbt_dynar_pop_as(reqq[smpi_process_index()],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->comm != MPI_COMM_NULL ? smpi_comm_rank(request->comm) @@ -341,7 +419,7 @@ static void action_waitall(const char *const *action){ int count_requests=0; unsigned int i=0; - count_requests=xbt_dynar_length(reqq[smpi_comm_rank(MPI_COMM_WORLD)]); + count_requests=xbt_dynar_length(reqq[smpi_process_index()]); if (count_requests>0) { MPI_Request requests[count_requests]; @@ -349,7 +427,7 @@ static void action_waitall(const char *const *action){ /* The reqq is an array of dynars. Its index corresponds to the rank. Thus each rank saves its own requests to the array request. */ - xbt_dynar_foreach(reqq[smpi_comm_rank(MPI_COMM_WORLD)],i,requests[i]); + xbt_dynar_foreach(reqq[smpi_process_index()],i,requests[i]); #ifdef HAVE_TRACING //save information from requests @@ -405,7 +483,7 @@ static void action_waitall(const char *const *action){ xbt_dynar_free(&recvs); #endif - xbt_dynar_free_container(&(reqq[smpi_comm_rank(MPI_COMM_WORLD)])); + xbt_dynar_free_container(&(reqq[smpi_process_index()])); } log_timed_action (action, clock); } @@ -413,12 +491,12 @@ static void action_waitall(const char *const *action){ static void action_barrier(const char *const *action){ double clock = smpi_process_simulated_elapsed(); #ifdef HAVE_TRACING - int rank = smpi_comm_rank(MPI_COMM_WORLD); + int rank = smpi_process_index(); instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); extra->type = TRACING_BARRIER; TRACE_smpi_collective_in(rank, -1, __FUNCTION__, extra); #endif - smpi_mpi_barrier(MPI_COMM_WORLD); + mpi_coll_barrier_fun(MPI_COMM_WORLD); #ifdef HAVE_TRACING TRACE_smpi_collective_out(rank, -1, __FUNCTION__); #endif @@ -446,7 +524,7 @@ static void action_bcast(const char *const *action) } #ifdef HAVE_TRACING - int rank = smpi_comm_rank(MPI_COMM_WORLD); + int rank = smpi_process_index(); int root_traced = smpi_group_index(smpi_comm_group(MPI_COMM_WORLD), root); instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); @@ -482,7 +560,7 @@ static void action_reduce(const char *const *action) } #ifdef HAVE_TRACING - int rank = smpi_comm_rank(MPI_COMM_WORLD); + int rank = smpi_process_index(); int root_traced = smpi_group_rank(smpi_comm_group(MPI_COMM_WORLD), root); instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); extra->type = TRACING_REDUCE; @@ -511,7 +589,7 @@ static void action_allReduce(const char *const *action) { double clock = smpi_process_simulated_elapsed(); #ifdef HAVE_TRACING - int rank = smpi_comm_rank(MPI_COMM_WORLD); + int rank = smpi_process_index(); instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); extra->type = TRACING_ALLREDUCE; extra->send_size = comm_size; @@ -520,9 +598,8 @@ static void action_allReduce(const char *const *action) { TRACE_smpi_collective_in(rank, -1, __FUNCTION__,extra); #endif - mpi_coll_reduce_fun(NULL, NULL, comm_size, MPI_CURRENT_TYPE, MPI_OP_NULL, 0, MPI_COMM_WORLD); + mpi_coll_allreduce_fun(NULL, NULL, comm_size, MPI_CURRENT_TYPE, MPI_OP_NULL, MPI_COMM_WORLD); smpi_execute_flops(comp_size); - mpi_coll_bcast_fun(NULL, comm_size, MPI_CURRENT_TYPE, 0, MPI_COMM_WORLD); #ifdef HAVE_TRACING TRACE_smpi_collective_out(rank, -1, __FUNCTION__); #endif @@ -545,8 +622,8 @@ static void action_allToAll(const char *const *action) { MPI_CURRENT_TYPE=MPI_DEFAULT_TYPE; MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE; } - void *send = calloc(send_size*comm_size, smpi_datatype_size(MPI_CURRENT_TYPE)); - void *recv = calloc(recv_size*comm_size, smpi_datatype_size(MPI_CURRENT_TYPE2)); + void *send = smpi_get_tmp_sendbuffer(send_size*comm_size* smpi_datatype_size(MPI_CURRENT_TYPE)); + void *recv = smpi_get_tmp_recvbuffer(recv_size*comm_size* smpi_datatype_size(MPI_CURRENT_TYPE2)); #ifdef HAVE_TRACING int rank = smpi_process_index(); @@ -567,8 +644,7 @@ static void action_allToAll(const char *const *action) { #endif log_timed_action (action, clock); - xbt_free(send); - xbt_free(recv); + } @@ -597,14 +673,14 @@ static void action_gather(const char *const *action) { MPI_CURRENT_TYPE=MPI_DEFAULT_TYPE; MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE; } - void *send = calloc(send_size, smpi_datatype_size(MPI_CURRENT_TYPE)); + void *send = smpi_get_tmp_sendbuffer(send_size* smpi_datatype_size(MPI_CURRENT_TYPE)); void *recv = NULL; int root=atoi(action[4]); int rank = smpi_process_index(); if(rank==root) - recv = calloc(recv_size*comm_size, smpi_datatype_size(MPI_CURRENT_TYPE2)); + recv = smpi_get_tmp_recvbuffer(recv_size*comm_size* smpi_datatype_size(MPI_CURRENT_TYPE2)); #ifdef HAVE_TRACING instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1); @@ -617,7 +693,7 @@ static void action_gather(const char *const *action) { TRACE_smpi_collective_in(rank, root, __FUNCTION__, extra); #endif -smpi_mpi_gather(send, send_size, MPI_CURRENT_TYPE, + mpi_coll_gather_fun(send, send_size, MPI_CURRENT_TYPE, recv, recv_size, MPI_CURRENT_TYPE2, root, MPI_COMM_WORLD); @@ -626,8 +702,7 @@ smpi_mpi_gather(send, send_size, MPI_CURRENT_TYPE, #endif log_timed_action (action, clock); - xbt_free(send); - xbt_free(recv); + } @@ -660,7 +735,7 @@ static void action_gatherv(const char *const *action) { MPI_CURRENT_TYPE=MPI_DEFAULT_TYPE; MPI_CURRENT_TYPE2=MPI_DEFAULT_TYPE; } - void *send = calloc(send_size, smpi_datatype_size(MPI_CURRENT_TYPE)); + void *send = smpi_get_tmp_sendbuffer(send_size* smpi_datatype_size(MPI_CURRENT_TYPE)); void *recv = NULL; for(i=0;i2){ + char *endptr; + double value = strtod((*argv)[2], &endptr); + if (*endptr != '\0') + THROWF(unknown_error, 0, "%s is not a double", (*argv)[2]); + XBT_VERB("Delayed start for instance - Sleeping for %f flops ",value ); + smpi_execute_flops(value); + } xbt_replay_action_runner(*argc, *argv); } @@ -966,32 +1041,34 @@ int smpi_replay_finalize(){ double sim_time= 1.; /* One active process will stop. Decrease the counter*/ XBT_DEBUG("There are %lu elements in reqq[*]", - xbt_dynar_length(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)]); + xbt_dynar_length(reqq[smpi_process_index()])); + if (!xbt_dynar_is_empty(reqq[smpi_process_index()])){ + int count_requests=xbt_dynar_length(reqq[smpi_process_index()]); 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]); + xbt_dynar_foreach(reqq[smpi_process_index()],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)])); + xbt_dynar_free_container(&(reqq[smpi_process_index()])); 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(sendbuffer); + xbt_free(recvbuffer); xbt_free(reqq); reqq = NULL; } - smpi_mpi_barrier(MPI_COMM_WORLD); + mpi_coll_barrier_fun(MPI_COMM_WORLD); #ifdef HAVE_TRACING int rank = smpi_process_index(); instr_extra_data extra = xbt_new0(s_instr_extra_data_t,1);