Help with timing a function!

I want something like this:

//includes

main()
{
//variable declarations

//<record start here>

//function to time

//<record date here>
//<find the difference>

//print total time accurately, at least in milliseconds.
}

I have no idea how to do the parts in bold. I tried gettimeofday(), but then I remembered that it only works under *nix, and I need this to compile and run under Windows. How do I do it?

I suppose a profiler is out of the question?

Quote:

---

Originally Posted by Prelude

I suppose a profiler is out of the question?

---

What's that? I was hoping for something simple (a one-line function like GetTimeOfDay perhaps?)

http://www.infosys.utas.edu.au/info/...C/CStdLib.html

These are all the standard C functions, everything you're looking for is under time.h.

edit: Specifically, you'll probably want to use asctime(), time(), and difftime().

edit: There are also functions which measure how much processor time your program has used, etc... I don't know if these all go as far as milliseconds.

Welp, a profiler is a program that tells you all kinds of useful tidbits about your program as it's run, including time spent in each function. However, if you want a quick and dirty solution:

Code:

---

#include <stdio.h>
#include <time.h>

void long_function(void)
{
  int i;

  for (i = 0; i < 10000000; i++)
    ;
}

int main(void)
{
  clock_t start;
 
  start = clock();
  long_function();
  printf("%f\n", (double)(clock() - start) / CLOCKS_PER_SEC);

  return 0;
}

---

Note that this isn't portable by any means, but I have yet to see it not work, and it serves it's purpose.

What is not portable about it?

Quote:

---

Originally Posted by sand_man

What is not portable about it?

---

I don't know either. The man page for clock says it conforms to the ANSI standard.

Code:

---

CONFORMING TO
      ANSI C.  POSIX requires that CLOCKS_PER_SEC equals 1000000
      independent of the actual resolution.

---

Yeah, its the actual resolution which is the problem. There is nothing to stop the implementation adding 1E6 to the clock once per second. It meets the spec, but its utterly useless at measuring milliseconds.

There is no sub-second standard clock.

>What is not portable about it?
The resolution for CLOCKS_PER_SEC isn't guaranteed to be accurate for fractions of a second and clock_t is a restricted type, just like time_t, so subtraction may not be supported.