What type of rounding is used when an integer is divided?

[hhh3h]

I read that floating point division uses the "round half to even" method.

+23.5 becomes +24, +22.5 becomes +22, -22.5 becomes -22, and -23.5 becomes -24

I was wondering what method integer division uses?  Someone said it was truncated, which is equivalent to always rounding "down" (toward negative infinity).  Is that correct?

Thank you.

EDIT: If this matters, I suppose this question is directed towards the x86-64 platform.

[Frank Kotler]

Yeah, integer division truncates, but that's towards zero, not towards negative infinity. The remainder is in ah, dx, edx, or rdx depending on the size.

The FPU is configurable via the "control word". "Round to nearest even" is probably a common default (useful for statistics, supposedly), but I wouldn't count on it!

Best,
Frank

[hhh3h]

Thanks, I appreciate your reply and the correction on truncate vs down / towards negative infinity

Why would I not want to count on the configurable control word?  Are you saying it might not be accurate for statistics purposes, or do you mean that it might not always be defaulted to that setting?

[Frank Kotler]

Merely that it might not always default to the same value. I was joking about the signal to noise ratio on a newsgroup ( news:alt.lang.asm ) being in danger of causing an underflow exception, and we fooled around with an example. Turned out it behaved differently in Dos vs Linux - apparently someone changed the exception handling. (I forget details) So you "might" discover rounding mode changed, too. If it's "mission critical", you might want to set the control word to a known value (and put it back).

But this has nothing to do with integer division, of course...

Best,
Frank

[hhh3h]

Ahh ok good to know