X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/ae486da8d6f00286ea3bf487c83978598c2f8bc6..6d004c352f7b26fba38486001f874e65466b5bee:/src/smpi/internals/smpi_bench.cpp diff --git a/src/smpi/internals/smpi_bench.cpp b/src/smpi/internals/smpi_bench.cpp index aa676ec268..ac32fa2ee6 100644 --- a/src/smpi/internals/smpi_bench.cpp +++ b/src/smpi/internals/smpi_bench.cpp @@ -11,7 +11,6 @@ #include "smpi_comm.hpp" #include "src/internal_config.h" #include "src/mc/mc_replay.hpp" -#include "src/simix/ActorImpl.hpp" #include "xbt/config.hpp" #include "src/smpi/include/smpi_actor.hpp" @@ -39,25 +38,12 @@ double smpi_host_speed; SharedMallocType smpi_cfg_shared_malloc = SharedMallocType::GLOBAL; double smpi_total_benched_time = 0; -extern "C" XBT_PUBLIC void smpi_execute_flops_(double* flops); - -void smpi_execute_flops_(double *flops) -{ - smpi_execute_flops(*flops); -} - -extern "C" XBT_PUBLIC void smpi_execute_(double* duration); -void smpi_execute_(double *duration) -{ - smpi_execute(*duration); -} - void smpi_execute_flops(double flops) { xbt_assert(flops >= 0, "You're trying to execute a negative amount of flops (%f)!", flops); XBT_DEBUG("Handle real computation time: %f flops", flops); simgrid::s4u::this_actor::exec_init(flops) ->set_name("computation") - ->set_tracing_category(TRACE_internal_smpi_get_category()) + ->set_tracing_category(smpi_process()->get_tracing_category()) ->start() ->wait(); smpi_switch_data_segment(simgrid::s4u::Actor::self()); @@ -190,7 +176,7 @@ static unsigned int private_sleep(double secs) int rank = simgrid::s4u::this_actor::get_pid(); TRACE_smpi_sleeping_in(rank, secs); - simcall_process_sleep(secs); + simgrid::s4u::this_actor::sleep_for(secs); TRACE_smpi_sleeping_out(rank); @@ -202,14 +188,14 @@ unsigned int smpi_sleep(unsigned int secs) { if (not smpi_process()) return sleep(secs); - return private_sleep(static_cast(secs)); + return private_sleep(secs); } int smpi_usleep(useconds_t usecs) { if (not smpi_process()) return usleep(usecs); - return static_cast(private_sleep(static_cast(usecs) / 1000000.0)); + return static_cast(private_sleep(usecs / 1000000.0)); } #if _POSIX_TIMERS > 0 @@ -217,7 +203,7 @@ int smpi_nanosleep(const struct timespec* tp, struct timespec* t) { if (not smpi_process()) return nanosleep(tp,t); - return static_cast(private_sleep(static_cast(tp->tv_sec + tp->tv_nsec / 1000000000.0))); + return static_cast(private_sleep(tp->tv_sec + tp->tv_nsec / 1000000000.0)); } #endif @@ -237,7 +223,7 @@ int smpi_gettimeofday(struct timeval* tv, struct timezone* tz) #endif } if (smpi_wtime_sleep > 0) - simcall_process_sleep(smpi_wtime_sleep); + simgrid::s4u::this_actor::sleep_for(smpi_wtime_sleep); smpi_bench_begin(); return 0; } @@ -255,7 +241,7 @@ int smpi_clock_gettime(clockid_t clk_id, struct timespec* tp) tp->tv_nsec = static_cast((now - tp->tv_sec) * 1e9); } if (smpi_wtime_sleep > 0) - simcall_process_sleep(smpi_wtime_sleep); + simgrid::s4u::this_actor::sleep_for(smpi_wtime_sleep); smpi_bench_begin(); return 0; } @@ -268,7 +254,7 @@ double smpi_mpi_wtime() smpi_bench_end(); time = SIMIX_get_clock(); if (smpi_wtime_sleep > 0) - simcall_process_sleep(smpi_wtime_sleep); + simgrid::s4u::this_actor::sleep_for(smpi_wtime_sleep); smpi_bench_begin(); } else { time = SIMIX_get_clock(); @@ -373,12 +359,11 @@ void smpi_sample_1(int global, const char *file, int line, int iters, double thr } } -int smpi_sample_2(int global, const char *file, int line) +int smpi_sample_2(int global, const char *file, int line, int iter_count) { SampleLocation loc(global, file, line); - int res; - XBT_DEBUG("sample2 %s", loc.c_str()); + XBT_DEBUG("sample2 %s %d", loc.c_str(), iter_count); auto sample = samples.find(loc); if (sample == samples.end()) xbt_die("Y U NO use SMPI_SAMPLE_* macros? Stop messing directly with smpi_sample_* functions!"); @@ -389,31 +374,25 @@ int smpi_sample_2(int global, const char *file, int line) XBT_DEBUG("benchmarking: count:%d iter:%d stderr:%f thres:%f; mean:%f; total:%f", data.count, data.iters, data.relstderr, data.threshold, data.mean, data.sum); smpi_bench_begin(); - res = 1; } else { // Enough data, no more bench (either we got enough data from previous visits to this benched nest, or we just //ran one bench and need to bail out now that our job is done). Just sleep instead if (not data.need_more_benchs()){ XBT_DEBUG("No benchmark (either no need, or just ran one): count >= iter (%d >= %d) or stderrset_sampling(0); - smpi_execute(sleep); + smpi_execute(data.mean*iter_count); smpi_bench_begin(); + return 0; } else { XBT_DEBUG("Skipping - Benchmark already performed - accumulating time"); xbt_os_threadtimer_start(smpi_process()->timer()); } - res = 0; // prepare to capture future, unrelated computations } - - return res; + return 1; } void smpi_sample_3(int global, const char *file, int line) @@ -437,7 +416,7 @@ void smpi_sample_3(int global, const char *file, int line) double period = xbt_os_timer_elapsed(smpi_process()->timer()); data.sum += period; data.sum_pow2 += period * period; - double n = static_cast(data.count); + double n = data.count; data.mean = data.sum / n; data.relstderr = sqrt((data.sum_pow2 / n - data.mean * data.mean) / n) / data.mean; @@ -448,7 +427,7 @@ void smpi_sample_3(int global, const char *file, int line) data.benching = false; } -void smpi_sample_exit(int global, const char *file, int line){ +int smpi_sample_exit(int global, const char *file, int line, int iter_count){ if (smpi_process()->sampling()){ SampleLocation loc(global, file, line); @@ -459,16 +438,12 @@ void smpi_sample_exit(int global, const char *file, int line){ LocalData& data = sample->second; if (smpi_process()->sampling()){//end of loop, but still sampling needed - double sleep = data.mean; - if (data.sum != 0.0){ //we finished benching, sum is unecessary after the first injection, we can reset it. - sleep = data.sum; - data.sum = 0.0; - } smpi_process()->set_sampling(0); - smpi_execute(sleep); + smpi_execute(data.mean*iter_count); smpi_bench_begin(); } } + return 0; } smpi_trace_call_location_t* smpi_trace_get_call_location()