Date: Thu, 24 Jun 1999 12:53:16 -0500
From: Eric Rudd <rudd AT cyberoptics DOT com>
Subject: Re: libm sources from cyberoptics
To: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
Cc: djgpp-workers AT delorie DOT com
Message-id: <3772708C.C1E87C92@cyberoptics.com>
Organization: CyberOptics
MIME-version: 1.0
X-Mailer: Mozilla 4.05 [en] (Win95; U)
Content-type: text/plain; charset=us-ascii
Content-transfer-encoding: 7bit
References: <Pine DOT SUN DOT 3 DOT 91 DOT 990624173154 DOT 15836F-100000 AT is>
Reply-To: djgpp-workers AT delorie DOT com

Eli Zaretskii wrote:

> I don't see what will need to be changed for C9X, except moving the
> prototypes to the ANSI section of the headers.  I don't think users will
> mind to pay an extra jmp, even when exp2 and exp10 are ANSI.
>
> Or am I missing something?

Once C9x issues, I thought it would be more straightforward (especially for debugging)
to make direct calls to the new ANSI functions exp2 and exp10, and reserve the stubs
for the aliases pow2 and pow10.  But the jumps are fast, so it doesn't make a lot of
difference.

> No, I didn't test it at all.  I simply looked at the source.  I guess I
> overlooked something, but it seemed to me that the result of fscale is
> not checked against the maximum possible number that a double can hold.

Actually, about 80% of the code in ldexp is devoted to error checking the input and
output values.  If you were looking for an FTST or FXAM, however, you wouldn't have
found them, because I use integer tests in most places, for speed.

-Eric Rudd