Portable High Performance Timer

I''d appreciate the answer for this, too.

The sdl timer uses unrealiable source and on win32 it causes backwarps, same occurs with certain debian distros.

If gettimeofday() is portable enough for you, perhaps use that.

What kinds of thing did you have in mind? AFAIK gettimeofday is available on a lot of C libraries, but I don''t know about how similar the structure is. Of course its resolution will be OS-dependent (probably 1/18s on DOS )

Mark

Well, on PCs you have access to the rdtsc command (through inline assembler) which pushes the current clock cycle count into registers, which is as high resoluton as you can get. Any non-PC hardware (or pre-Pentium PCs) won''t have this command. The best way to code a portable timer is through use of #define''s and #ifdef for conditional compilation depending on the platform. I''m afraid on different platforms there is no guarantee that a particular timing method will be available. You''ll just need to find the best timer for each platform and add them to your code with the necessary #ifdef / #else / #endif around them.

Not completely sure that on every Pentium-compatible platform, RDTSC is guaranteed to continue to be incremented at the same rate over time.

For instance, on mobile (i.e. laptop) CPUs, which have clock frequency switching, is there a possibility that RDTSC will vary with the clock speed?

I don''t think the OS implementation of gettimeofday / QueryPerformanceCounter relies on RDTSC, instead it asks a timer chip somewhere what time it is?

Mark

It appears the clock() function does what I need, but does it rely of RDTSC? Is markr right about it not being reliable? Anyone know whether it demonstrates consistent behavior across platforms?

clock() does something different from RDTSC.

From the man page:

quote:

---

The clock() function returns an approximation of processor time used by the program.

---

Note "by the program". So clock() isn''t actually looking at a real-time or "wall" clock, it''s looking at the amount of CPU time the program has used. The Linux man page for clock() does not specify how it handles threads, but I''m not sure it would give you the result you expect.

----

I still recommend that you manually code an implementation of your high resolution timer, changing it if necessary for other platforms.

Are there any credible gaming platforms other than win32 which don''t have a useable getttimeofday()? If so, you will have to code another implementation for them too (NB I know nothing about the availability of this in such things as PS2 C library)

Mark

quote:

---

Original post by markr
For instance, on mobile (i.e. laptop) CPUs, which have clock frequency switching, is there a possibility that RDTSC will vary with the clock speed?

---

Yes indeed. This happens with my P4M, even when it''s plugged in. The only way around it is to switch off all power saving features in the BIOS which isn''t really an option. I assume that it won''t work with processors like the Crusoe either.

Fortunately no commercial games I''ve tried use this method, but B5:IFH (http://ifh.firstones.com/) does and everything runs 3-4 times faster than it should.