void run_threaded_string_sends_0(size_t length, size_t numthreads) {
if (rmi.procid() == 1) {
rmi.full_barrier();
return;
}
timer ti;
std::cout << numthreads << " threaded " << SEND_LIMIT_PRINT <<" sends, "
<< length << " bytes\n";
ti.start();
size_t numsends = SEND_LIMIT / (length * numthreads);
size_t rd = rdtsc();
thread_group thrgrp;
for (size_t i = 0; i < numthreads; ++i) {
thrgrp.launch(boost::bind(&teststruct::perform_string_sends_0, this, length, numsends));
}
thrgrp.join();
size_t rd2 = rdtsc();
std::cout << (rd2 - rd) / (numthreads * numsends) << " cycles per call\n";
double t1 = ti.current_time();
rmi.dc().flush();
double t2 = ti.current_time();
rmi.full_barrier();
double t3 = ti.current_time();
print_res(t1,t2,t3);
}
void run_short_sends_0() {
if (rmi.procid() == 1) {
rmi.full_barrier();
return;
}
timer ti;
std::cout << "Single Threaded " << SEND_LIMIT_PRINT << " sends, 4 integer blocks\n";
ti.start();
size_t numsends = SEND_LIMIT / (sizeof(size_t) * 4);
perform_short_sends_0(numsends);
double t1 = ti.current_time();
rmi.dc().flush();
double t2 = ti.current_time();
rmi.full_barrier();
double t3 = ti.current_time();
print_res(t1,t2,t3);
}
void run_threaded_short_pod_sends_0(size_t numthreads) {
if (rmi.procid() == 1) {
rmi.full_barrier();
return;
}
timer ti;
std::cout << numthreads << " threaded "<< SEND_LIMIT_PRINT <<" POD sends, 4 integers\n";
size_t numsends = SEND_LIMIT / (sizeof(size_t) * 4 * numthreads);
ti.start();
thread_group thrgrp;
for (size_t i = 0; i < numthreads; ++i) {
thrgrp.launch(boost::bind(&teststruct::perform_short_pod_sends_0, this, numsends));
}
thrgrp.join();
double t1 = ti.current_time();
rmi.dc().flush();
double t2 = ti.current_time();
rmi.full_barrier();
double t3 = ti.current_time();
print_res(t1,t2,t3);
}
void run_string_sends_0(size_t length) {
if (rmi.procid() == 1) {
rmi.full_barrier();
return;
}
timer ti;
size_t numsends = SEND_LIMIT / (length);
std::cout << "Single Threaded " << SEND_LIMIT_PRINT <<" sends, " << length << " bytes * "<< numsends << "\n";
ti.start();
size_t rd = rdtsc();
perform_string_sends_0(length, numsends);
size_t rd2 = rdtsc();
std::cout << "Completed in: " << ti.current_time() << " seconds\n";
std::cout << (rd2 - rd) / numsends << " cycles per call\n";
double t1 = ti.current_time();
rmi.dc().flush();
std::cout << "Flush in: " << ti.current_time() << " seconds\n";
double t2 = ti.current_time();
rmi.full_barrier();
std::cout << "Receive Complete in: " << ti.current_time() << " seconds\n";
double t3 = ti.current_time();
print_res(t1,t2,t3);
}
/// \copydoc distributed_control::full_barrier()
inline void full_barrier() {
rmi.full_barrier();
}