Register parameters & m68k-unknown-coff

• Date: Sun, 26 Jan 1997 16:22:35 +0200 (EET)
• From:
  Iso-H <jd@ray.fi>
• Reply-To:
  Iso-H <jd@ray.fi>
• Content-Type: TEXT/PLAIN; charset=US-ASCII

 I have problems with mixed register/stack parameter passing.

 I am using registers d0 and d1 for passing first and second parameter
 and stack for the rest.

 To get register parameter passing to work I added these 
 two macro to file "config/m68k/m68k-coff.h":

       #undef TARGET_REGPARM
       #define TARGET_REGPARM (1)

       #undef  FUNCTION_ARG_REGNO_P
       #define FUNCTION_ARG_REGNO_P(N) ((N)==0 || (N)==1)

 Everything seems to be ok until I used varargs; stack
 parameters works (actually everything works just ok, 
 if I use more than two fixed parameter!) , but register
 parameters not. 
 For example from this C-code:

        #include <stdarg.h>

        int tmp1, tmp2, tmp3, tmp4;

        void func( int p_1, ... )
        {
            va_list  args;

            va_start(args, p_1 );

            tmp1 = va_arg(args,int);
            tmp2 = va_arg(args,int);
            tmp3 = va_arg(args,int);
            tmp4 = va_arg(args,int);


        }


 ... gcc generates this assembly code:

        .globl func
        func:
            link.w %a6,#0
            move.l 8(%a6),tmp1
            move.l 12(%a6),tmp2
            move.l 16(%a6),tmp3
            move.l 20(%a6),tmp4
            unlk %a6
            rts


 Obviously gcc should generate something like this:

        .globl func
        func:
            link.w %a6,#0
            move.l %d1,tmp1
            move.l 8(%a6),tmp2
            move.l 12(%a6),tmp3
            move.l 16(%a6),tmp4
            unlk %a6
            rts


 So question is: Do I have to define some other macros 
 in "m68k-coff.h" also? And what are these macros?

• Re: Register parameters & m68k-unknown-coff
• From: Doug Evans <dje@cygnus.com>

• Prev: Problem with gcc 2.7.2.1. Cross compiler host HP UX 9.07 target VXWORKS
• Next: Re: Register parameters & m68k-unknown-coff
• Index: Cross gcc Mailing list
• Thread: Cross gcc Mailing list