Thread: Execution Time in Microseconds?

> Is here a way I can measure time in processor ticks?
Maybe, but you're not going to find out if you stick to standard C++.
You need to tell us which platform, OS and compiler you're using (eg Intel P4, XP-SP2, VC6).

The smallest unit of measuring absolute time is the resolution of the highest-performance timer component you can lay your hands on. Which that is is always system-dependent.

Doesn't boost have some timing library?

Yes, but it's not very accurate, I think.

Windows has some millisecond-precise functions. See this post: http://cboard.cprogramming.com/showp...7&postcount=27

Again, state your compiler and OS.

Well i've used this in my programs and it works great:

Code:

#include <iostream>
#include <ctime>
...
//Global Variables
unsigned int timer;
unsigned in timer2;
...
void startTimer()
{
	timer = clock();
}

int stopTimer()
{
	timer2 = clock();
	timer = timer2 - timer;
	return timer;
}

Then again i'm on Windows XP using MSVC++ Express 2005


GOod luck!

Mmm, well . . . clock() returns a clock_t, not an unsigned int and certainly not an unsigned in. Global variables are a bad idea (I know my sample code used them too, but that code is from like a year and a half ago). timer2 could just be a local variable. clock() (and clock_t) either need to be prefixed with std:: or you need a using statement.

> I'm using XP+ MSVC++
Check out the use of QueryPerformanceCounter in the link dwks posted in post #6

> XP+ GCC (CYGWIN), and Slackware Linux+ GCC
Use this asm to read the time stamp counter common to all pentium processors
http://cboard.cprogramming.com/showt...ighlight=rdtsc

Even if you have only millisecond timing precision, you can always run your function/code/whatever half a million times, time the entire thing, then divide by how many times you ran it. Something like:

Code:

StartTimer()
for(x = 0; x < 500000; ++x) RunCode();
EndTimer();

You take on the time needed for the for() loop, however, but it may/may not be insignificant...
Be aware of compiler optimizations and settings as well - the compiler may very well remove the code you're testing if it realizes it doesn't effectively do anything.

Well if you are using Unix/Linux...there are some POSIX functions out there that should get you nanosecond time precision. You can do something like this:

struct timespec tp;
clock_gettime ( CLOCK_REALTIME, &tp );

seconds = tp.tv_sec
nanoseconds = tp.tv_nsec

I honestly can't remember the include file though....it's probably time.h or ctime...depending on the compiler you are using. Once again, this is only for POSIX standard Linux/Unix systems...nothing else.