Sender: nate AT cartsys DOT com
Message-ID: <37781E66.A5CBE238@cartsys.com>
Date: Mon, 28 Jun 1999 18:16:22 -0700
From: Nate Eldredge <nate AT cartsys DOT com>
X-Mailer: Mozilla 4.08 [en] (X11; I; Linux 2.2.10 i586)
MIME-Version: 1.0
To: djgpp AT delorie DOT com
Subject: Re: Some nice questions!! (one of them is silly!)
References: <01bec0de$10cd1120$LocalHost AT thendren>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Reply-To: djgpp AT delorie DOT com

Christopher Nelson wrote:

> >   2) what are registers that i can safely change in
> >my assembler function without having to push its
> >values?
> 
> there are no "safe" registers to manipulate.  the compiler uses all the
> registers available for it in an optimized register allocation algorithm.
> if you mess with the registers in an unexpected way, you should save them,
> then restore them.  if you are using inline assembly, you can do this after
> the third ':' so that the compiler knows not to expect anything in the
> registers that you thrash.  finally, if you write a function in assembly
> that's prototyped as returning an int or an unsigned int, you may usually do
> anything you want with EAX because the compiler will expect the return value
> to be passed in that register.  of course, if your program doesn't use the
> return value you can't assume what the compiler will do with the register.
> (most often it will just mark it dirty and reload it with whatever value it
> needs for the next expression resolution.)

When calling a standalone function, eax, ecx and edx are expected to be
changed.  The compiler knows this.  (Ref line 490 or so of
gcc-2.81/config/i386/i386.h (#define CALL_USED_REGISTERS)).  Therefore,
these are safe to change in a pure asm function.



> >   3) in djgpp progs, does djgpp make the register es
> >equal to ds?? ALWAYS??
> 
> no.  DJGPP usually ignores the segment registers.  after the initial
> initialization code, it never addresses the segment registers again, unless
> you do it in assembler -- or use the "farptr" routines.

Except that it's done implicitly by GCC.  Obviously ds must be correct,
and since GCC does generate things like `rep movsl' sometimes, es needs
to be the same.

I guess that could be a semantic confusion.  DJGPP *requires* that they
be the same, but it does not *make* that the case under all
circumstances (it won't fix it if you break it).
-- 

Nate Eldredge
nate AT cartsys DOT com