Re: Why doesn't -fomit-frame-pointer work(very well)?

[Peter Barada <pbarada at mail dot wm dot sps dot mot dot com>]

>> 1) Bump FIRST_PSEUDO_REGISTER by two.
>> 2) define FRAME_POINTER_REGNUM and ARG_POINTER_REGNUM to be the
>>    new pseudo-registers
>
>Actually, you can leave FRAME_POINTER_REGNUM == ARG_POINTER_REGNUM
>as you do currently.  

Ok, I've got:

/* Base register for access to local variables of the function.  */
#define HARD_FRAME_POINTER_REGNUM 14

/* Base register for access to local variables of the function.  */
#ifndef SUPPORT_SUN_FPA
#define FRAME_POINTER_REGNUM 24
#else
#define FRAME_POINTER_REGNUM 56
#endif

#define REGISTER_NAMES \
{"%d0",   "%d1",   "%d2",   "%d3",   "%d4",   "%d5",   "%d6",   "%d7",	     \
 "%a0",   "%a1",   "%a2",   "%a3",   "%a4",   "%a5",   "%a6",   "%sp",	     \
 "%fp0",  "%fp1",  "%fp2",  "%fp3",  "%fp4",  "%fp5",  "%fp6",  "%fp7",	\
 ".fp" }

#define ELIMINABLE_REGS					\
{{ ARG_POINTER_REGNUM, STACK_POINTER_REGNUM},		\
 { ARG_POINTER_REGNUM, HARD_FRAME_POINTER_REGNUM},	\
 { FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM},		\
 { FRAME_POINTER_REGNUM, HARD_FRAME_POINTER_REGNUM}}	\

#define CAN_ELIMINATE(FROM, TO) \
  ((TO) == STACK_POINTER_REGNUM ? ! frame_pointer_needed : 1)

#define INITIAL_ELIMINATION_OFFSET(FROM, TO, OFFSET)			\
  (OFFSET) = 0 /* Total hack for now, fix later... */

1) I'll assume the above is correct.
2) Is there anything esle that has to be done?

I've tried:

/tmp/crap/bin/m68k-elf-gcc -fomit-frame-pointer -mcfv4e -O -S -o /tmp/z.s /tmp/z.c

On the code:

typedef __builtin_va_list __gnuc_va_list;
typedef __gnuc_va_list va_list;
extern void putzi(char c, char **p);
extern int _vformat (void (*putsub)(char c, char **p),
                     char **putbuf, const char *fmt, va_list argp);
int printf(const char *template, ...)
{
  int retval;
  va_list args;

  __builtin_stdarg_start((args),template);
  retval = _vformat(putzi, (char **)0, template, args);
  __builtin_va_end(args);

  return (retval);
}

And it produces:

printf:
	link.w %a6,#0
	pea 12(%a6)
	move.l 8(%a6),-(%sp)
	clr.l -(%sp)
	pea putzi
	jbsr _vformat
	lea (16,%sp),%sp
	unlk %a6
	rts

Which didn't eliminate the frame pointer.  I see from dumps that the
".fp" register(24) *is* eliminated in favor of "%a6"(14), and from gdb I see:

(gdb) c
Continuing.
Hardware watchpoint 18: *$4

Old value = 1
New value = 0
eliminate_regs_in_insn (insn=0x40023820, replace=0)
    at /home/pbarada/work/cvs-wavemark/cross-linux-tools/gcc-304/gcc/reload1.c:3237
(gdb) p *ep
$5 = {from = 14, to = 15, initial_offset = 4, can_eliminate = 0, 
  can_eliminate_previous = 1, offset = 8, previous_offset = 4, 
  ref_outside_mem = 1, from_rtx = 0x4001c020, to_rtx = 0x4001c000}
(gdb) call debug_rtx(insn)
(insn 19 16 21 (set (mem/f:SI (pre_dec:SI (reg/f:SI 15 %sp)) 0)
        (plus:SI (reg/f:SI 15 %sp)
            (const_int 16 [0x10]))) 60 {pushasi} (nil)
    (nil))
(gdb) 

So it looks like it tried to eliminate the frame pointer and then
failed.  I'm assuming that my defintion of INITIAL_ELIMINATION_OFFSET
needs to be correct for all of this to work(since ep->can_eliminate is
set to 0 since ep->previous_offset doesn't match ep->offset), right?

Thanks.

-- 
Peter Barada                                   Peter.Barada@motorola.com
Wizard                                         781-852-2768 (direct)
WaveMark Solutions(wholly owned by Motorola)   781-270-0193 (fax)