Re: [linuxsh-dev] Other changes

[NIIBE Y. <gn...@ch...>]

Hi Mitch, 

When I was in Germany last week, I've eaten _real_ Donner Kebab, it's
great.

Mitch Davis wrote:
 > Looking at your mail, this code keeps its dollar:
 > 
 > >                 "mov    $r0, $r4\n\t"   /* return value to arg of exit */
 > > -               "mov    %2, $r3\n\t"    /* exit */
 > > +               "mov    %1, $r3\n\t"    /* exit */
 > >                 "trapa  #0x11\n"

This "$" is used by the assembler only.

 > This code does not:
 > 
 > >  {
 > > -    register long r0 __asm__("$r0") = func;
 > > -    register long r4 __asm__("$r4") = arg0;
 > > -    register long r5 __asm__("$r5") = arg1;
 > > -    register long r6 __asm__("$r6") = arg2;
 > > -    register long r7 __asm__("$r7") = arg3;
 > > +    register long r0 __asm__("r0") = func;
 > > +    register long r4 __asm__("r4") = arg0;
 > > +    register long r5 __asm__("r5") = arg1;
 > > +    register long r6 __asm__("r6") = arg2;
 > > +    register long r7 __asm__("r7") = arg3;
 > >      __asm__ __volatile__("trapa        #0x3f"

This "$" is used for C compiler, which I think not needed, or
it is good not to have.

I'm not sure, which is good.  When I incorporated the patch by Jesper
(the T-flag patch) with "t", I found that it would be more consistent
using such like "r0" for C compiler.  You know, the use of "$t" is
bogus.  We never use such specification with assembler (since
there are specific instruction for T-flag (clrt, movt...).

How do you think?
--