Re: [patch]: New operand parser for i386 intel syntax

[Emmanuel Michon <emmanuel_michon at sdesigns dot com>]

Diego Novillo <dnovillo@cygnus.com> writes:

> I have tested the patch with the GCC regression suite. It passes
> with no new regressions when compiled with -mintel-syntax
> (although I needed to patch the i386 backend to produce the
> correct syntax. I will submit a separate patch to gcc-patches).

1. Seems your patch is ok with ``as'' when you put intel style code inside

intel_syntax noprefix
mov eax,ecx
att_syntax
mov %ecx,%ecx

So, what's the meaning of noprefix?

2. gcc with -m-intel-syntax options just puts a
intel_syntax
at the very beginning of the .s output, what is it for?

3. What about accessing local/parameter variables from 
inside inline assembly?

Let's take a look at the following (working) example:

------------------------
#include <stdio.h>

int bar1;

int substractfive()
{
	asm( ".intel_syntax noprefix\n" );
	// Intel syntax!!!
	asm( "\tpush eax\n");
	asm( "\tpush ebx\n");
	asm( "\tmov eax,bar1\n" );
	asm( "\tmov ebx,5\n" );
	asm( "\tsub eax,ebx\n");
	asm( "\tmov bar1,eax\n");

	asm( "\t.att_syntax\n" );
	// some AT&T for the pleasure.
	asm( "\tpop %ebx\n");
	asm( "\tpop %eax\n");

	return bar1;
}

int main(int argc,char **argv)
{
	bar1=8;
	printf("8-5 = %d\n",substractfive());
	return 0;
}
------------------------

How could we modify this code so that bar1 is
transferred as a parameter to:

int substractfive(int bar1)?

Normal gcc/at&t assembly would use constraints I think
to do that;

is there a way to avoid it?

Sincerely yours,

-- 
Emmanuel Michon
Ingénieur en développement logiciel
REALmagic France       
Mobile: 0662834836 GPGkeyID: D2997E42