X-Authentication-Warning:	ieva01.lanet.lv: pavenis owned process doing -bs
Date:	Thu, 4 Mar 1999 10:13:40 +0200 (WET)
From:	Andris Pavenis <pavenis AT lanet DOT lv>
To:	Nate Eldredge <nate AT cartsys DOT com>
cc:	djgpp-workers AT delorie DOT com
Subject:	Re: egcs-1.1.X (Was: Re: Problems with collect2 in egcs-1.1.1 and
	1.1.2preX)
In-Reply-To:	<36DDA6AB.F06CDFDE@cartsys.com>
Message-ID:	<Pine.A41.4.05.9903040959150.50756-100000@ieva01.lanet.lv>
MIME-Version:	1.0
Reply-To:	djgpp-workers AT delorie DOT com

On Wed, 3 Mar 1999, Nate Eldredge wrote:

> Andris Pavenis wrote:
> > 
> > On 2 Mar 99, at 15:23, DJ Delorie wrote:
> > 
> > >
> > > > Looking at the configure.in in egcs, I noticed just above *-pc-
> > > > msdosdjgpp is *-msdos-go32. Would it be OK to delete the entries for
> > > > *-msdos-go32 since *-pc-msdosdjgpp is the preferred triplet?
> > >
> > > No.  You may issue a warning for those cases, though.
> > >
> > > The current official triplet is i586-pc-msdosdjgpp - it's the only one
> > > that works right for both gcc and binutils.
> > >
> > > But, others may still be using other variants in specific cases.
> > >
> > 
> > I don't need  target  i[34567]-pc-go32. For me i[34567]-pc-msdosdjgpp is
> > enough.
> > 
> > config-guess detects DJGPP as i386-pc-msdosdjgpp for all
> > i[3456]86. I override target (--target=i586-pc-msdosdjgpp) when I built egcs.
> > One posible problem is that then compiler options defaults to
> > -march=pentium -mcpu=pentium. It may seem to be dangerous but:
> [snip]
> 
> I actually can't imagine any differences.  I don't have a Pentium manual
> here to check but AFAIK only a few instructions were added between 386
> and Pentium, and I doubt GCC generates any of them:
> 
> invlpg (for system software only)
> cmpxchg
> cmpxchg8 (I envy the compiler that can use them)
> cpuid (ha!!)
> 
> I also looked at the egcs 1.1.1 source in config/i386, and didn't see
> any arch=pentium dependencies.  So it looks like it's safe for current
> versions.

It's easy to test that DJGPP port of egcs-1.1.1 defaults to 
-march=pentium -mcpu=pentium


cat /dev/null | gcc -S -x c -fverbose-asm -o - -


I also was slightly startled when I found that. But seems that 486 and
pentium related instructions are actually not generated by egcs.
At least I have verified that I'm getting the same effect with egcs-1.0.3
binaries included in Slackware-3.5 and 3.6 Linux distribution. 

> 
> Of course, that won't necessarily be true in the future.
> -- 

Maybe. At least binaries are no more compatible when compiled with
-march=pentiumpro AFAIK. But I haven't heard about problems with 
-march=pentium.

Andris

- Raw text -