Message-ID: <32D16662.76FE@pobox.oleane.com>
Date: Mon, 06 Jan 1997 21:53:54 +0100
From: Francois Charton <deef AT pobox DOT oleane DOT com>
Organization: CCMSA
MIME-Version: 1.0
To: Michael Phelps <morphine AT hops DOT cs DOT jhu DOT edu>
CC: djgpp AT delorie DOT com
Subject: Re: FP vs int timings
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

Michael Phelps wrote:
> 
>                 b = (double)random();
>                 c = (double)random();
>                 a = b / c;[snip]
>                 y = (int)random();
>                 z = (int)random();
>                 x = y / z;[resnip]
> results on my 5x86/100 under Win95:
>         doubles:  2.29s for 1000000 divides
>         ints:     2.12s for 1000000 divides
> 
> conclusions:  on my machine, at least, there is not a significant
> difference in speed between integer and floating-point math.  (The 

If you want to compare their speed, you have to test fixed vs float on 
more than just a divide...

As an example, fixed point math usually are good at addition and 
substraction, and I think the compares could also be quite fast.

OTOH, fixed by fixed multiplies and divisions are slow (as you have 
noticed, in fact their are even worse than in your program, because there 
is one more shift involved)...

But fixed by integer multiplies and divides are fast, and cast from int 
to fixed also are... and with fixed point numbers, you can index 
arrays...

The list could go on. 

To me, on most modern machines (anything with a built in FPU basically), 
integer calculations are not significantly faster than floating point 
instructions *on the average*. However, if speed is a concern, they are 
worth experimenting with, they can help a lot at times...


Francois