X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/2455643f96166346bf10d16ddc289cc44f1fe78a..1cce0801abcad7c884f9e72ceda87e36d6635104:/src/smpi/smpi_global.cpp diff --git a/src/smpi/smpi_global.cpp b/src/smpi/smpi_global.cpp index 1da63088db..d140c1aeeb 100644 --- a/src/smpi/smpi_global.cpp +++ b/src/smpi/smpi_global.cpp @@ -5,6 +5,7 @@ * under the terms of the license (GNU LGPL) which comes with this package. */ #include "private.h" +#include "private.hpp" #include "smpi_mpi_dt_private.h" #include "mc/mc.h" #include "src/mc/mc_record.h" @@ -14,13 +15,20 @@ #include "simgrid/sg_config.h" #include "src/mc/mc_replay.h" #include "src/msg/msg_private.h" +#include "src/simix/SynchroComm.hpp" + #include /* DBL_MAX */ #include #include #include +#include XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_kernel, smpi, "Logging specific to SMPI (kernel)"); +#include +#include /* trim_right / trim_left */ + +std::unordered_map location2speedup; typedef struct s_smpi_process_data { double simulated; @@ -38,34 +46,34 @@ typedef struct s_smpi_process_data { char state; int sampling; /* inside an SMPI_SAMPLE_ block? */ char* instance_id; - int replaying; /* is the process replaying a trace */ + bool replaying; /* is the process replaying a trace */ xbt_bar_t finalization_barrier; int return_value; + smpi_trace_call_location_t* trace_call_loc; } s_smpi_process_data_t; -static smpi_process_data_t *process_data = NULL; +static smpi_process_data_t *process_data = nullptr; int process_count = 0; int smpi_universe_size = 0; -int* index_to_process_data = NULL; +int* index_to_process_data = nullptr; extern double smpi_total_benched_time; xbt_os_timer_t global_timer; MPI_Comm MPI_COMM_WORLD = MPI_COMM_UNINITIALIZED; - -MPI_Errhandler *MPI_ERRORS_RETURN = NULL; -MPI_Errhandler *MPI_ERRORS_ARE_FATAL = NULL; -MPI_Errhandler *MPI_ERRHANDLER_NULL = NULL; +MPI_Errhandler *MPI_ERRORS_RETURN = nullptr; +MPI_Errhandler *MPI_ERRORS_ARE_FATAL = nullptr; +MPI_Errhandler *MPI_ERRHANDLER_NULL = nullptr; #define MAILBOX_NAME_MAXLEN (5 + sizeof(int) * 2 + 1) static char *get_mailbox_name(char *str, int index) { - snprintf(str, MAILBOX_NAME_MAXLEN, "SMPI-%0*x", (int) (sizeof(int) * 2), index); + snprintf(str, MAILBOX_NAME_MAXLEN, "SMPI-%0*x", static_cast (sizeof(int) * 2), index); return str; } static char *get_mailbox_name_small(char *str, int index) { - snprintf(str, MAILBOX_NAME_MAXLEN, "small%0*x", (int) (sizeof(int) * 2), index); + snprintf(str, MAILBOX_NAME_MAXLEN, "small%0*x", static_cast (sizeof(int) * 2), index); return str; } @@ -75,16 +83,15 @@ void smpi_process_init(int *argc, char ***argv) smpi_process_data_t data; smx_process_t proc; - if (argc && argv) { + if (argc != nullptr && argv != nullptr) { proc = SIMIX_process_self(); - //FIXME: dirty cleanup method to avoid using msg cleanup functions on these processes when using MSG+SMPI - SIMIX_process_set_cleanup_function(proc, SIMIX_process_cleanup); + SIMIX_process_set_cleanup_function(proc, MSG_process_cleanup_from_SIMIX); char* instance_id = (*argv)[1]; int rank = xbt_str_parse_int((*argv)[2], "Invalid rank: %s"); index = smpi_process_index_of_smx_process(proc); - if(!index_to_process_data){ - index_to_process_data=(int*)xbt_malloc(SIMIX_process_count()*sizeof(int)); + if(index_to_process_data == nullptr){ + index_to_process_data=static_cast(xbt_malloc(SIMIX_process_count()*sizeof(int))); } if(smpi_privatize_global_variables){ @@ -99,20 +106,19 @@ void smpi_process_init(int *argc, char ***argv) smpi_deployment_register_process(instance_id, rank, index, &temp_comm_world, &temp_bar); data = smpi_process_remote_data(index); data->comm_world = temp_comm_world; - if(temp_bar != NULL) data->finalization_barrier = temp_bar; + if(temp_bar != nullptr) + data->finalization_barrier = temp_bar; data->index = index; data->instance_id = instance_id; - data->replaying = 0; - //xbt_free(simcall_process_get_data(proc)); + data->replaying = false; simdata_process_t simdata = static_cast(simcall_process_get_data(proc)); simdata->data = data; if (*argc > 3) { - free((*argv)[1]); memmove(&(*argv)[0], &(*argv)[2], sizeof(char *) * (*argc - 2)); - (*argv)[(*argc) - 1] = NULL; - (*argv)[(*argc) - 2] = NULL; + (*argv)[(*argc) - 1] = nullptr; + (*argv)[(*argc) - 2] = nullptr; } (*argc)-=2; data->argc = argc; @@ -122,11 +128,11 @@ void smpi_process_init(int *argc, char ***argv) XBT_DEBUG("<%d> New process in the game: %p", index, proc); } xbt_assert(smpi_process_data(), - "smpi_process_data() returned NULL. You probably gave a NULL parameter to MPI_Init. Although it's required by " + "smpi_process_data() returned nullptr. You probably gave a nullptr parameter to MPI_Init. Although it's required by " "MPI-2, this is currently not supported by SMPI."); } -void smpi_process_destroy(void) +void smpi_process_destroy() { int index = smpi_process_index(); if(smpi_privatize_global_variables){ @@ -137,7 +143,7 @@ void smpi_process_destroy(void) } /** @brief Prepares the current process for termination. */ -void smpi_process_finalize(void) +void smpi_process_finalize() { // This leads to an explosion of the search graph which cannot be reduced: if(MC_is_active() || MC_record_replay_is_active()) @@ -159,9 +165,9 @@ int smpi_process_finalized() } /** @brief Check if a process is initialized */ -int smpi_process_initialized(void) +int smpi_process_initialized() { - if (!index_to_process_data){ + if (index_to_process_data == nullptr){ return false; } else{ int index = smpi_process_index(); @@ -170,27 +176,27 @@ int smpi_process_initialized(void) } /** @brief Mark a process as initialized (=MPI_Init called) */ -void smpi_process_mark_as_initialized(void) +void smpi_process_mark_as_initialized() { int index = smpi_process_index(); if ((index != MPI_UNDEFINED) && (process_data[index_to_process_data[index]]->state != SMPI_FINALIZED)) process_data[index_to_process_data[index]]->state = SMPI_INITIALIZED; } -void smpi_process_set_replaying(int value){ +void smpi_process_set_replaying(bool value){ int index = smpi_process_index(); if ((index != MPI_UNDEFINED) && (process_data[index_to_process_data[index]]->state != SMPI_FINALIZED)) process_data[index_to_process_data[index]]->replaying = value; } -int smpi_process_get_replaying(){ +bool smpi_process_get_replaying(){ int index = smpi_process_index(); if (index != MPI_UNDEFINED) return process_data[index_to_process_data[index]]->replaying; - else return _xbt_replay_is_active(); + else return (_xbt_replay_is_active() != 0); } -int smpi_global_size(void) +int smpi_global_size() { char *value = getenv("SMPI_GLOBAL_SIZE"); xbt_assert(value,"Please set env var SMPI_GLOBAL_SIZE to the expected number of processes."); @@ -198,7 +204,7 @@ int smpi_global_size(void) return xbt_str_parse_int(value, "SMPI_GLOBAL_SIZE contains a non-numerical value: %s"); } -smpi_process_data_t smpi_process_data(void) +smpi_process_data_t smpi_process_data() { simdata_process_t simdata = static_cast(SIMIX_process_self_get_data()); return static_cast(simdata->data); @@ -221,38 +227,50 @@ void *smpi_process_get_user_data() return process_data->data; } -int smpi_process_count(void) +int smpi_process_count() { return process_count; } -int smpi_process_index(void) +/** + * \brief Returns a structure that stores the location (filename + linenumber) + * of the last calls to MPI_* functions. + * + * \see smpi_trace_set_call_location + */ +smpi_trace_call_location_t* smpi_process_get_call_location() +{ + smpi_process_data_t process_data = smpi_process_data(); + return process_data->trace_call_loc; +} + +int smpi_process_index() { smpi_process_data_t data = smpi_process_data(); //return -1 if not initialized - return data ? data->index : MPI_UNDEFINED; + return data != nullptr ? data->index : MPI_UNDEFINED; } -MPI_Comm smpi_process_comm_world(void) +MPI_Comm smpi_process_comm_world() { smpi_process_data_t data = smpi_process_data(); //return MPI_COMM_NULL if not initialized - return data ? *data->comm_world : MPI_COMM_NULL; + return data != nullptr ? *data->comm_world : MPI_COMM_NULL; } -smx_mailbox_t smpi_process_mailbox(void) +smx_mailbox_t smpi_process_mailbox() { smpi_process_data_t data = smpi_process_data(); return data->mailbox; } -smx_mailbox_t smpi_process_mailbox_small(void) +smx_mailbox_t smpi_process_mailbox_small() { smpi_process_data_t data = smpi_process_data(); return data->mailbox_small; } -xbt_mutex_t smpi_process_mailboxes_mutex(void) +xbt_mutex_t smpi_process_mailboxes_mutex() { smpi_process_data_t data = smpi_process_data(); return data->mailboxes_mutex; @@ -276,37 +294,37 @@ xbt_mutex_t smpi_process_remote_mailboxes_mutex(int index) return data->mailboxes_mutex; } -xbt_os_timer_t smpi_process_timer(void) +xbt_os_timer_t smpi_process_timer() { smpi_process_data_t data = smpi_process_data(); return data->timer; } -void smpi_process_simulated_start(void) +void smpi_process_simulated_start() { smpi_process_data_t data = smpi_process_data(); data->simulated = SIMIX_get_clock(); } -double smpi_process_simulated_elapsed(void) +double smpi_process_simulated_elapsed() { smpi_process_data_t data = smpi_process_data(); return SIMIX_get_clock() - data->simulated; } -MPI_Comm smpi_process_comm_self(void) +MPI_Comm smpi_process_comm_self() { smpi_process_data_t data = smpi_process_data(); if(data->comm_self==MPI_COMM_NULL){ MPI_Group group = smpi_group_new(1); - data->comm_self = smpi_comm_new(group, NULL); + data->comm_self = smpi_comm_new(group, nullptr); smpi_group_set_mapping(group, smpi_process_index(), 0); } return data->comm_self; } -MPI_Comm smpi_process_get_comm_intra(void) +MPI_Comm smpi_process_get_comm_intra() { smpi_process_data_t data = smpi_process_data(); return data->comm_intra; @@ -324,7 +342,7 @@ void smpi_process_set_sampling(int s) data->sampling = s; } -int smpi_process_get_sampling(void) +int smpi_process_get_sampling() { smpi_process_data_t data = smpi_process_data(); return data->sampling; @@ -336,34 +354,37 @@ void print_request(const char *message, MPI_Request request) message, request, request->buf, request->size, request->src, request->dst, request->tag, request->flags); } -void smpi_comm_copy_buffer_callback(smx_synchro_t comm, void *buff, size_t buff_size) +void smpi_comm_copy_buffer_callback(smx_synchro_t synchro, void *buff, size_t buff_size) { XBT_DEBUG("Copy the data over"); void* tmpbuff=buff; + simgrid::simix::Comm *comm = dynamic_cast(synchro); - if((smpi_privatize_global_variables) && ((char*)buff >= smpi_start_data_exe) - && ((char*)buff < smpi_start_data_exe + smpi_size_data_exe ) + if((smpi_privatize_global_variables) && (static_cast(buff) >= smpi_start_data_exe) + && (static_cast(buff) < smpi_start_data_exe + smpi_size_data_exe ) ){ XBT_DEBUG("Privatization : We are copying from a zone inside global memory... Saving data to temp buffer !"); - smpi_switch_data_segment(((smpi_process_data_t)(((simdata_process_t)SIMIX_process_get_data(comm->comm.src_proc))->data))->index); - tmpbuff = (void*)xbt_malloc(buff_size); + + + smpi_switch_data_segment((static_cast((static_cast(SIMIX_process_get_data(comm->src_proc))->data))->index)); + tmpbuff = static_cast(xbt_malloc(buff_size)); memcpy(tmpbuff, buff, buff_size); } - if((smpi_privatize_global_variables) && ((char*)comm->comm.dst_buff >= smpi_start_data_exe) - && ((char*)comm->comm.dst_buff < smpi_start_data_exe + smpi_size_data_exe )){ + if((smpi_privatize_global_variables) && ((char*)comm->dst_buff >= smpi_start_data_exe) + && ((char*)comm->dst_buff < smpi_start_data_exe + smpi_size_data_exe )){ XBT_DEBUG("Privatization : We are copying to a zone inside global memory - Switch data segment"); - smpi_switch_data_segment(((smpi_process_data_t)(((simdata_process_t)SIMIX_process_get_data(comm->comm.dst_proc))->data))->index); + smpi_switch_data_segment((static_cast((static_cast(SIMIX_process_get_data(comm->dst_proc))->data))->index)); } - memcpy(comm->comm.dst_buff, tmpbuff, buff_size); - if (comm->comm.detached) { + memcpy(comm->dst_buff, tmpbuff, buff_size); + if (comm->detached) { // if this is a detached send, the source buffer was duplicated by SMPI // sender to make the original buffer available to the application ASAP xbt_free(buff); //It seems that the request is used after the call there this should be free somewhere else but where??? //xbt_free(comm->comm.src_data);// inside SMPI the request is kept inside the user data and should be free - comm->comm.src_buff = NULL; + comm->src_buff = nullptr; } if(tmpbuff!=buff)xbt_free(tmpbuff); @@ -387,11 +408,11 @@ static void smpi_check_options(){ } } -int smpi_enabled(void) { - return process_data != NULL; +int smpi_enabled() { + return process_data != nullptr; } -void smpi_global_init(void) +void smpi_global_init() { int i; MPI_Group group; @@ -402,6 +423,28 @@ void smpi_global_init(void) global_timer = xbt_os_timer_new(); xbt_os_walltimer_start(global_timer); } + + if (xbt_cfg_get_string("smpi/comp-adjustment-file")[0] != '\0') { + std::string filename {xbt_cfg_get_string("smpi/comp-adjustment-file")}; + std::ifstream fstream(filename); + if (!fstream.is_open()) { + xbt_die("Could not open file %s. Does it exist?", filename.c_str()); + } + + std::string line; + typedef boost::tokenizer< boost::escaped_list_separator> Tokenizer; + std::getline(fstream, line); // Skip the header line + while (std::getline(fstream, line)) { + Tokenizer tok(line); + Tokenizer::iterator it = tok.begin(); + Tokenizer::iterator end = std::next(tok.begin()); + + std::string location = *it; + boost::trim(location); + location2speedup.insert(std::pair(location, std::stod(*end))); + } + } + if (process_count == 0){ process_count = SIMIX_process_count(); smpirun=1; @@ -410,9 +453,8 @@ void smpi_global_init(void) process_data = xbt_new0(smpi_process_data_t, process_count); for (i = 0; i < process_count; i++) { process_data[i] = xbt_new(s_smpi_process_data_t, 1); - //process_data[i]->index = i; - process_data[i]->argc = NULL; - process_data[i]->argv = NULL; + process_data[i]->argc = nullptr; + process_data[i]->argv = nullptr; process_data[i]->mailbox = simcall_mbox_create(get_mailbox_name(name, i)); process_data[i]->mailbox_small = simcall_mbox_create(get_mailbox_name_small(name, i)); process_data[i]->mailboxes_mutex = xbt_mutex_init(); @@ -421,18 +463,22 @@ void smpi_global_init(void) MC_ignore_heap(process_data[i]->timer, xbt_os_timer_size()); process_data[i]->comm_self = MPI_COMM_NULL; process_data[i]->comm_intra = MPI_COMM_NULL; - process_data[i]->comm_world = NULL; + process_data[i]->comm_world = nullptr; process_data[i]->state = SMPI_UNINITIALIZED; process_data[i]->sampling = 0; - process_data[i]->finalization_barrier = NULL; + process_data[i]->finalization_barrier = nullptr; process_data[i]->return_value = 0; + + if (xbt_cfg_get_boolean("smpi/trace-call-location")) { + process_data[i]->trace_call_loc = xbt_new(smpi_trace_call_location_t, 1); + } } //if the process was launched through smpirun script we generate a global mpi_comm_world //if not, we let MPI_COMM_NULL, and the comm world will be private to each mpi instance if(smpirun){ group = smpi_group_new(process_count); - MPI_COMM_WORLD = smpi_comm_new(group, NULL); - MPI_Attr_put(MPI_COMM_WORLD, MPI_UNIVERSE_SIZE, (void *)(MPI_Aint)process_count); + MPI_COMM_WORLD = smpi_comm_new(group, nullptr); + MPI_Attr_put(MPI_COMM_WORLD, MPI_UNIVERSE_SIZE, reinterpret_cast(process_count)); xbt_bar_t bar=xbt_barrier_init(process_count); for (i = 0; i < process_count; i++) { @@ -442,7 +488,7 @@ void smpi_global_init(void) } } -void smpi_global_destroy(void) +void smpi_global_destroy() { int count = smpi_process_count(); int i; @@ -463,14 +509,19 @@ void smpi_global_destroy(void) } xbt_os_timer_free(process_data[i]->timer); xbt_mutex_destroy(process_data[i]->mailboxes_mutex); + if (xbt_cfg_get_boolean("smpi/trace-call-location")) { + xbt_free(process_data[i]->trace_call_loc); + } xbt_free(process_data[i]); } xbt_free(process_data); - process_data = NULL; + process_data = nullptr; if (MPI_COMM_WORLD != MPI_COMM_UNINITIALIZED){ smpi_comm_cleanup_smp(MPI_COMM_WORLD); smpi_comm_cleanup_attributes(MPI_COMM_WORLD); + if(smpi_coll_cleanup_callback!=nullptr) + smpi_coll_cleanup_callback(); xbt_free(MPI_COMM_WORLD); } @@ -543,56 +594,57 @@ static void smpi_init_logs(){ static void smpi_init_options(){ int gather_id = find_coll_description(mpi_coll_gather_description, xbt_cfg_get_string("smpi/gather"),"gather"); - mpi_coll_gather_fun = (int (*)(void *, int, MPI_Datatype, void *, int, MPI_Datatype, int, MPI_Comm)) - mpi_coll_gather_description[gather_id].coll; + mpi_coll_gather_fun = reinterpret_cast + (mpi_coll_gather_description[gather_id].coll); int allgather_id = find_coll_description(mpi_coll_allgather_description, xbt_cfg_get_string("smpi/allgather"),"allgather"); - mpi_coll_allgather_fun = (int (*)(void *, int, MPI_Datatype, void *, int, MPI_Datatype, MPI_Comm)) - mpi_coll_allgather_description[allgather_id].coll; + mpi_coll_allgather_fun = reinterpret_cast + (mpi_coll_allgather_description[allgather_id].coll); int allgatherv_id = find_coll_description(mpi_coll_allgatherv_description, xbt_cfg_get_string("smpi/allgatherv"),"allgatherv"); - mpi_coll_allgatherv_fun = (int (*)(void *, int, MPI_Datatype, void *, int *, int *, MPI_Datatype, MPI_Comm)) - mpi_coll_allgatherv_description[allgatherv_id].coll; + mpi_coll_allgatherv_fun = reinterpret_cast + (mpi_coll_allgatherv_description[allgatherv_id].coll); int allreduce_id = find_coll_description(mpi_coll_allreduce_description, xbt_cfg_get_string("smpi/allreduce"),"allreduce"); - mpi_coll_allreduce_fun = (int (*)(void *sbuf, void *rbuf, int rcount, MPI_Datatype dtype, MPI_Op op, MPI_Comm comm)) - mpi_coll_allreduce_description[allreduce_id].coll; + mpi_coll_allreduce_fun = reinterpret_cast + (mpi_coll_allreduce_description[allreduce_id].coll); int alltoall_id = find_coll_description(mpi_coll_alltoall_description, xbt_cfg_get_string("smpi/alltoall"),"alltoall"); - mpi_coll_alltoall_fun = (int (*)(void *, int, MPI_Datatype, void *, int, MPI_Datatype, MPI_Comm)) - mpi_coll_alltoall_description[alltoall_id].coll; + mpi_coll_alltoall_fun = reinterpret_cast + (mpi_coll_alltoall_description[alltoall_id].coll); int alltoallv_id = find_coll_description(mpi_coll_alltoallv_description, xbt_cfg_get_string("smpi/alltoallv"),"alltoallv"); - mpi_coll_alltoallv_fun = (int (*)(void *, int *, int *, MPI_Datatype, void *, int *, int *, MPI_Datatype, MPI_Comm)) - mpi_coll_alltoallv_description[alltoallv_id].coll; + mpi_coll_alltoallv_fun = reinterpret_cast + (mpi_coll_alltoallv_description[alltoallv_id].coll); int bcast_id = find_coll_description(mpi_coll_bcast_description, xbt_cfg_get_string("smpi/bcast"),"bcast"); - mpi_coll_bcast_fun = (int (*)(void *buf, int count, MPI_Datatype datatype, int root, MPI_Comm com)) - mpi_coll_bcast_description[bcast_id].coll; + mpi_coll_bcast_fun = reinterpret_cast + (mpi_coll_bcast_description[bcast_id].coll); int reduce_id = find_coll_description(mpi_coll_reduce_description, xbt_cfg_get_string("smpi/reduce"),"reduce"); - mpi_coll_reduce_fun = (int (*)(void *buf, void *rbuf, int count, MPI_Datatype datatype, MPI_Op op, int root, - MPI_Comm comm)) mpi_coll_reduce_description[reduce_id].coll; + mpi_coll_reduce_fun = reinterpret_cast + (mpi_coll_reduce_description[reduce_id].coll); int reduce_scatter_id = find_coll_description(mpi_coll_reduce_scatter_description, xbt_cfg_get_string("smpi/reduce-scatter"),"reduce_scatter"); - mpi_coll_reduce_scatter_fun = (int (*)(void *sbuf, void *rbuf, int *rcounts,MPI_Datatype dtype, MPI_Op op, - MPI_Comm comm)) mpi_coll_reduce_scatter_description[reduce_scatter_id].coll; + mpi_coll_reduce_scatter_fun = reinterpret_cast + (mpi_coll_reduce_scatter_description[reduce_scatter_id].coll); int scatter_id = find_coll_description(mpi_coll_scatter_description, xbt_cfg_get_string("smpi/scatter"),"scatter"); - mpi_coll_scatter_fun = (int (*)(void *sendbuf, int sendcount, MPI_Datatype sendtype, void *recvbuf, - int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm)) - mpi_coll_scatter_description[scatter_id].coll; + mpi_coll_scatter_fun = reinterpret_cast + (mpi_coll_scatter_description[scatter_id].coll); int barrier_id = find_coll_description(mpi_coll_barrier_description, xbt_cfg_get_string("smpi/barrier"),"barrier"); - mpi_coll_barrier_fun = (int (*)(MPI_Comm comm)) mpi_coll_barrier_description[barrier_id].coll; + mpi_coll_barrier_fun = reinterpret_cast + (mpi_coll_barrier_description[barrier_id].coll); + smpi_coll_cleanup_callback=nullptr; smpi_cpu_threshold = xbt_cfg_get_double("smpi/cpu-threshold"); smpi_running_power = xbt_cfg_get_double("smpi/running-power"); smpi_privatize_global_variables = xbt_cfg_get_boolean("smpi/privatize-global-variables"); @@ -604,7 +656,7 @@ int smpi_main(int (*realmain) (int argc, char *argv[]), int argc, char *argv[]) { srand(SMPI_RAND_SEED); - if (getenv("SMPI_PRETEND_CC") != NULL) { + if (getenv("SMPI_PRETEND_CC") != nullptr) { /* Hack to ensure that smpicc can pretend to be a simple compiler. Particularly handy to pass it to the * configuration tools */ return 0;