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

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

I've noticed that gcc-3.0.4 doesn't eliminate the frame pointer very
well when asked to with -fomit-frame-pointer, at least for a compiler
configured for m68k-elf.

Here's a simple function that should easily have its frame pointer
elimintated, and when built with -m68000 -O -fomit-frame-pointer, doesn't:

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);
}

	.file	"z.c"
	.text
	.globl	printf
	.type	printf,@function
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

The reason in this case is the instruction:

(gdb) where 3
#0  elimination_effects (x=0x4001c010, mem_mode=VOIDmode)
    at /home/pbarada/work/cvs-wavemark/cross-linux-tools/gcc-304/gcc/reload1.c:2711
#1  0x08166133 in elimination_effects (x=0x401af9f0, mem_mode=VOIDmode)
    at /home/pbarada/work/cvs-wavemark/cross-linux-tools/gcc-304/gcc/reload1.c:2852
#2  0x081665fe in eliminate_regs_in_insn (insn=0x400237e0, replace=0)
    at /home/pbarada/work/cvs-wavemark/cross-linux-tools/gcc-304/gcc/reload1.c:3093
(More stack frames follow...)
(gdb) fra 2
#2  0x081665fe in eliminate_regs_in_insn (insn=0x400237e0, replace=0)
    at /home/pbarada/work/cvs-wavemark/cross-linux-tools/gcc-304/gcc/reload1.c:3093
(gdb) call debug_rtx(insn)

(insn 17 14 19 (set (mem/f:SI (pre_dec:SI (reg/f:SI 15 %sp)) 0)
        (plus:SI (reg/f:SI 14 %a6)
            (const_int 12 [0xc]))) 60 {pushasi} (nil)
    (nil))

SInce it references both the stack and the frame pointer then the code
declares that its impossible to eliminate it(via
reg_eliminate[0].ref_outside_mem being set to true which later causes
reg_eliminate[0].can_eliminate to be set to ).

Unfortunately passing paramters to child functions is a common
construct...  Any ideas how best to fix this?  Does frame pointer
elimination work in other ports?

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