Sender: bill AT delorie DOT com
Message-ID: <3779C9A1.35125F69@taniwha.org>
Date: Wed, 30 Jun 1999 19:39:13 +1200
From: Bill Currie <bill AT taniwha DOT org>
X-Mailer: Mozilla 4.05 [en] (X11; I; Linux 2.2.9 i486)
MIME-Version: 1.0
To: djgpp-workers AT delorie DOT com
Subject: Re: Regparm and asm statements.. what now?
References: <3778E043 DOT D8D5F4 AT inti DOT gov DOT ar> <Pine DOT SUN DOT 3 DOT 91 DOT 990629184054 DOT 18841S-100000 AT is> <19990629185744 DOT C4792 AT tabor DOT ta DOT jcu DOT cz>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Reply-To: djgpp-workers AT delorie DOT com

Jan Hubicka wrote:
>
> >
  Eli:
> > What happens if someone compiles a program without including the headers
> > that declare the prototype?
> >
> > According to ANSI C, users can legitimately do that, and still assume
> > they get a working program.

IMHO, broken spec.  Nuff said.

> You will get broken program with regparm anyways, because the function
> with variable number of arguments are called using stack convention.
> So you need to include stdio to get regparmized hello world working.
> I am not sure if this is OK and later this can be fixed in gcc2.96.
> It will require some nasty hacks to varrargs stuff.

On archs that use N (usually 4) regs by default (eg i860, pa-risc, most
risc?), the convention is that the first N parameter words (not
necessarily parameters themselves, depends on param size) are *ALWAYS*
passed in the registers (with FP parms being sent in both int and fp
regs), no matter what, with any additional parameters passed on the
stack.  The varargs function then has prologue code that saves the
incoming regs on the stack.  The va_args `macro' does the right thing
for accessing the incoming params.  I don't know how of even if this can
be made to work for the i386 (though I don't see why not), but do we
want to?

Anyway, varargs stuff need not be a problem.  So long as the calling
conventions match, no prototype is truely needed.

Bill
-- 
Leave others their otherness.