Date: Wed, 3 Sep 1997 08:31:30 -0400 (EDT)
Message-Id: <199709031231.IAA29822@delorie.com>
From: DJ Delorie <dj AT delorie DOT com>
To: eldredge AT ap DOT net
CC: djgpp AT delorie DOT com
In-reply-to: <199709030259.TAA10342@adit.ap.net> (message from Nate Eldredge
	on Tue, 2 Sep 1997 19:59:21 -0700 (PDT))
Subject: Re: Optimization of math.h functions
Precedence: bulk


> If there's a disadvantage to this, somebody please let me know. Otherwise, I
> think it should be done.

Most of the i387 FPU instructions are only valid on a limited range of
input values.  The functions have to perform other math on the input
to coerce it to the right value before doing the actual operation.
For example, sin(x) is only valid for one or two quadrants, so you
have to do some divides and subtracts (and maybe call cos(pi/2-x)) to
get a valid number.

The functions also have to deal with invalid input, like sqrt(-1),
which sets errno to EDOM.

If there are functions that can be inlined with 100% compatibility
with published standards, we should inline them.  We cannot inline if
it means incompatibility.