To: djgpp AT delorie DOT com
Subject: Re: DJGPP Optimizing
Message-ID: <19961231.163855.4983.3.chambersb@juno.com>
From: chambersb AT juno DOT com (Benjamin D Chambers)
Date: Mon, 30 Dec 1996 19:35:29 EST

>Hi!
Hi :)

>I have some troubles using the optimizing features in the 
>compiler

>1) When the compiler optimizes a tight inner loop, that 
>uses only integer math
>it inserts a bunch of floating point instructions (fxch) 
>that doesn't seem to do
Check to make sure _everything_ is integer math, and that you aren't
accidentally typecasting to a float/double (ie using trig functions, etc)

>anything at all (I can remove them and my code works fine 
>without). As these
>instructions are inside the tight loop they eat up about 
>50% of the time spent here..
If it's really unnecessary floating math, I'm surprised it doesn't eat up
more...

>2) When using the trigonomic functions cos/sin in 3D 
>graphics it seems that whenever
>I use the -O# compiler option the functions are treated 
>differently than without
>the -O# option. Result is that the 3D space is skew'ed. 
>Am I assuming something
>about the trigonomic function (ie. angles can be outside 
>[0 : 2*PI]) that does not 
>hold when optimizing?
I believe the 387 (and 487 and 587) require trig functions to be within
2pi.  However, I can't test this for sure, since I'm FPUless.

>Kind regards, Henning Semler
Good luck,

...Chambers