Subj : Re: A question about atomic_ptr
To   : comp.programming.threads
From : Chris Thomasson
Date : Mon Apr 18 2005 09:49 am

> "Should" and "possibly" aside, which one is faster

mfence is faster than LOCK.




> and by what margin? ;-)


Go ahead and try running the following code on your system:




#include <windows.h>
#include <stdlib.h>
#include <stdio.h>


#define LOOPS 500000000




int main()
{
  int i;
  DWORD f, s;



  /* lock prefix */
  s = GetTickCount();

  for ( i = 0; i < LOOPS; ++i )
  {
    __asm__ __volatile__ ("lock; addl $0, 0(%esp)");
  }

  f = GetTickCount();

  printf( "lock prefix time: %d\n", f - s );



  /* mfence */
  s = GetTickCount();

  for ( i = 0; i < LOOPS; ++i )
  {
    __asm__ __volatile__ ("addl $0, 0(%esp)\r\n mfence");
  }

  f = GetTickCount();

  printf( "mfence time: %d\n", f - s );




  system( "pause" );

  return 0;
}

.