Re: My PC's cores can synchronize to about 1.000 clock cycles accuracy

#include <iostream>
#include <barrier>
#include <thread>
#include <vector>
#if defined(_WIN32)
     #include <intrin.h>
#elif defined(__linux__)
     #include <x86intrin.h>
#endif
 using namespace std;
 int main()
{
     unsigned hc = thread::hardware_concurrency();
     barrier bar( hc );
     atomic_uint synch( hc );
     atomic_uint64_t zero( 0 );
     atomic_int64_t diffs( 0 );
     auto thr = [&]()
     {
         int64_t sum = 0;
         for( unsigned t = 1'000; t; --t )
         {
             bar.arrive_and_wait();
             if( synch.fetch_sub( 1, memory_order_relaxed ) > 1 )
                 while( synch.load( memory_order_relaxed ) );
             uint64_t tsc = __rdtsc(), expected = 0;
             if( !zero.compare_exchange_weak( expected, tsc, memory_order_relaxed, memory_order_relaxed ) )
                 sum += abs( (int64_t)(expected - tsc) );
             bar.arrive_and_wait();
             synch.store( hc );
             zero.store( 0, memory_order_relaxed );
         }
         diffs.fetch_add( sum, memory_order_relaxed );
     };
     vector<jthread> threads;
     threads.reserve( hc - 1 );
     for( unsigned t = hc - 1; t; --t )
         threads.emplace_back( thr );
     thr();
     threads.resize( 0 );
     cout << (double)diffs.load( memory_order_relaxed ) / (1'000.0 * hc) << endl;
}
 My PC is a AMD 7950X 16-core system.