#include "src/internal_config.h"
#include "src/mc/mc_replay.hpp"
#include "xbt/config.hpp"
+#include "xbt/file.hpp"
#include "src/smpi/include/smpi_actor.hpp"
#include <unordered_map>
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) {
+// Private execute_flops used by smpi_execute and smpi_execute_benched
+void private_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)
smpi_switch_data_segment(simgrid::s4u::Actor::self());
}
+void smpi_execute_flops(double flops) {
+ int rank = simgrid::s4u::this_actor::get_pid();
+ TRACE_smpi_computing_in(rank, flops);
+
+ private_execute_flops(flops);
+
+ TRACE_smpi_computing_out(rank);
+}
+
void smpi_execute(double duration)
{
if (duration >= smpi_cpu_threshold) {
int rank = simgrid::s4u::this_actor::get_pid();
TRACE_smpi_computing_in(rank, flops);
- smpi_execute_flops(flops);
+ private_execute_flops(flops);
TRACE_smpi_computing_out(rank);
xbt_os_threadtimer_start(smpi_process()->timer());
}
+double smpi_adjust_comp_speed(){
+ double speedup=1;
+ if (simgrid::config::get_value<std::string>("smpi/comp-adjustment-file")[0] != '\0') {
+
+ smpi_trace_call_location_t* loc = smpi_process()->call_location();
+ std::string key = loc->get_composed_key();
+ std::unordered_map<std::string, double>::const_iterator it = location2speedup.find(key);
+ if (it != location2speedup.end()) {
+ speedup = it->second;
+ }
+ }
+ return speedup;
+}
+
void smpi_bench_end()
{
if (MC_is_active() || MC_record_replay_is_active())
return;
- double speedup = 1;
xbt_os_timer_t timer = smpi_process()->timer();
xbt_os_threadtimer_stop(timer);
}
// Maybe we need to artificially speed up or slow down our computation based on our statistical analysis.
- if (simgrid::config::get_value<std::string>("smpi/comp-adjustment-file")[0] != '\0') {
-
- smpi_trace_call_location_t* loc = smpi_process()->call_location();
- std::string key = loc->get_composed_key();
- std::unordered_map<std::string, double>::const_iterator it = location2speedup.find(key);
- if (it != location2speedup.end()) {
- speedup = it->second;
- }
- }
-
// Simulate the benchmarked computation unless disabled via command-line argument
if (simgrid::config::get_value<bool>("smpi/simulate-computation")) {
- smpi_execute(xbt_os_timer_elapsed(timer)/speedup);
+ smpi_execute(xbt_os_timer_elapsed(timer)/smpi_adjust_comp_speed());
}
#if HAVE_PAPI
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);
{
if (not smpi_process())
return sleep(secs);
- return private_sleep(static_cast<double>(secs));
+ return private_sleep(secs);
}
int smpi_usleep(useconds_t usecs)
{
if (not smpi_process())
return usleep(usecs);
- return static_cast<int>(private_sleep(static_cast<double>(usecs) / 1000000.0));
+ return static_cast<int>(private_sleep(usecs / 1000000.0));
}
#if _POSIX_TIMERS > 0
{
if (not smpi_process())
return nanosleep(tp,t);
- return static_cast<int>(private_sleep(static_cast<double>(tp->tv_sec + tp->tv_nsec / 1000000000.0)));
+ return static_cast<int>(private_sleep(tp->tv_sec + tp->tv_nsec / 1000000000.0));
}
#endif
#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;
}
tp->tv_nsec = static_cast<long int>((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;
}
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();
0, // count
true // benching (if we have no data, we need at least one)
});
- LocalData& data = insert.first->second;
if (insert.second) {
XBT_DEBUG("XXXXX First time ever on benched nest %s.", loc.c_str());
xbt_assert(threshold > 0 || iters > 0,
"You should provide either a positive amount of iterations to bench, or a positive maximal stderr (or both)");
} else {
+ LocalData& data = insert.first->second;
if (data.iters != iters || data.threshold != threshold) {
XBT_ERROR("Asked to bench block %s with different settings %d, %f is not %d, %f. "
"How did you manage to give two numbers at the same line??",
double period = xbt_os_timer_elapsed(smpi_process()->timer());
data.sum += period;
data.sum_pow2 += period * period;
- double n = static_cast<double>(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;
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!");
- LocalData& data = sample->second;
if (smpi_process()->sampling()){//end of loop, but still sampling needed
- smpi_process()->set_sampling(0);
- smpi_execute(data.mean*iter_count);
- smpi_bench_begin();
+ LocalData& data = sample->second;
+ smpi_process()->set_sampling(0);
+ smpi_execute(data.mean * iter_count);
+ smpi_bench_begin();
}
}
return 0;
loc->previous_filename = loc->filename;
loc->previous_linenumber = loc->linenumber;
- loc->filename = file;
- loc->linenumber = line;
+ if(not simgrid::config::get_value<bool>("smpi/trace-call-use-absolute-path"))
+ loc->filename = simgrid::xbt::Path(file).get_base_name();
+ else
+ loc->filename = file;
+ loc->linenumber = line;
}
/** Required for Fortran bindings */