Sender: nate AT cartsys DOT com
Message-ID: <35C355CA.40E48B1@cartsys.com>
Date: Sat, 01 Aug 1998 10:52:10 -0700
From: Nate Eldredge <nate AT cartsys DOT com>
MIME-Version: 1.0
To: "Salvador Eduardo Tropea (SET)" <salvador AT inti DOT gov DOT ar>
CC: lubaldo AT adinet DOT com DOT uy, djgpp AT delorie DOT com
Subject: Re: -m and -march= switches and a 5x86...
References: <m0z2EqC-000S4TC AT inti DOT gov DOT ar>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Precedence: bulk

Salvador Eduardo Tropea (SET) wrote:
> 
> lubaldo AT adinet DOT com DOT uy wrote:
> 
> >       I have GCC 2.8.1 installed.
> >       I have tryed the -m386, -m486, -mpentium and -mpentiumpro switches and
> > they generate the exact same code... why is that?
> 
> I tried -mpentiumpro and the code miserably(?) died with SIGILL because gcc
> used the conditional assigment instruction (P6 core only).
> Are you sure the assembler generated doesn't have even 1 line of difference?
> What are you compiling?
> 
> >       I have tryed the -march=586 switch. It produces Pentium code that it is
> > not compatible with earlyer CPU's, like 486, 386... but it works on my
> > machine!
> 
> Who said GCC produces code that doesn't run in 486 when you use 586? Is in
> the manuals? I think gcc only changes the instruction order and aligment.
> What P5 instruction is used?

`-mcpu=CPU TYPE'
     Assume the defaults for the machine type CPU TYPE when scheduling
     instructions.  The choices for CPU TYPE are: `i386', `i486',
     `i586' (`pentium'), `pentium', `i686' (`pentiumpro') and
     `pentiumpro'. While picking a specific CPU TYPE will schedule
     things appropriately for that particular chip, the compiler will
     not generate any code that does not run on the i386 without the
     `-march=CPU TYPE' option being used.

`-march=CPU TYPE'
     Generate instructions for the machine type CPU TYPE.  The choices
     for CPU TYPE are: `i386', `i486', `pentium', and `pentiumpro'.
     Specifying `-march=CPU TYPE' implies `-mcpu=CPU TYPE'.

So `-mcpu' changes ordering, alignment, etc; while `-march' changes the
actual instructions used.
 
> > I have an AMD 5x86 processor, to my knowledge, this is a 486DX
> > compatible processor but it is not meant to be Pentium compatible...
> 
> True.

I haven't looked at a manual or opcode ref lately, but I can't think of
anything at the application level a Pentium can do that a 486 can't
(with the possible exception of the probably-useless CMPXCHG8B
instruction).

> > I have tryed also the -march=686 and that it doesn't work and thats
> > Ok... And the code produced by GCC with the -march=586 directive is
> > different from the one produced with the -march=486 directive... so: I
> > don't understand why it works... someone knows something?
> 
> I think is just reordered to exploit the P5 pipes, that's all.

I suspect you are right.

As a further red herring, note that `-march=686' has some bugs which can
make it generate code that won't work even on a real PPro.  (Apparently
it doesn't handle the FICOMP instructions correctly.)
-- 

Nate Eldredge
nate AT cartsys DOT com