A common mistake with AccurateTimeElapsed()

[Lenn]

I've seen a few threads asking about why AccurateTimeElapsed seems to not give accurate values. I read through some of the replies and most often people seem to forget to advise on one first, most basic thing.

AccurateTimeElapsed returns a SINGLE not an INTEGER value.

If you have it return the time elapsed to an integer variable, it will simply round it up or down, and the decimal part is very important in keeping accurate time. Otherwise, each rounding adds or substracts a little bit of time and you get wrong results.

I just thought I'd post this here instead of several old threads where I've found these problems without proper advice given.

[NetLizard]

Floating-point division(decimal remainder)
------>  / Foward slash <--------


Integer division ( No remainder )
------->  \ Backslash <----------

[Mietze]

Also keep in mind that this is not accurate all the time. This will return the time difference between two RenderToScreen(). If you have vertical synchronization activated or use multhithreading, it's better to use your own performance timer for this!

[TecnoBacon]

Just to note: it is almost twice as fast to multiply then divide in the proc, even if your storing a single for the multiply.

[Mietze]

Just to note: it is almost twice as fast to multiply then divide in the proc, even if your storing a single for the multiply.

*nod*

[beyonder]

Just to note: it is almost twice as fast to multiply then divide in the proc, even if your storing a single for the multiply.

Why? Why? Why? I know this is true... I ran some tests and I believe its even faster than twice.

Sounds to me a division needs to be 'flipped' into a multiplication adding a few more steps to the process...

[TecnoBacon]

Actually it has to do with addressing, a div is a multiply within the proc so several iterations are done to get the same result as a single mult would accomplish.