Re: [patch RFC] SH: Use FRAME_GROWS_DOWNWARD

[Kaz Kojima <kkojima at rr dot iij4u dot or dot jp>]

Joern RENNECKE <joern.rennecke@st.com> wrote:
> Oops, I analyzed the problem with apply_result_mode /
> apply_result_size, but when I intended to return to the function to 
> patch it,
> I ended up in apply_args_size.  FUNCTION_ARG_REGNO_P is true for
> to all function argument passing registers, so apply_args_mode  s hould
> be fine.

builtin-apply4.c passes for sh-elf and sh-elf/-m4 with the new
patch.  The tweaked testcase in
  http://gcc.gnu.org/ml/gcc-patches/2005-07/msg01282.html
passes for sh-elf but fails for sh-elf/-m4.  The assembler output
for the last case looks like:

	...
	jsr	@r11		; <- call foo
	fsts	fpul,fr11
	flds	fr0,fpul
	sts	fpul,r1
	mov.l	r0,@r14
	flds	fr1,fpul
	mov.l	r1,@r10
	mov.l	r1,@(4,r14)
	sts	fpul,r1
	flds	fr2,fpul
	mov.l	r1,@(4,r10)
	sts	fpul,r1
	mov.l	r2,@(8,r14)
	flds	fr3,fpul
	mov.l	r1,@(8,r10)
	sts	fpul,r1
	mov.l	r3,@(12,r14)
	mov.l	r1,@(12,r10)
	mov.l	r4,@(16,r14)
	mov.l	r5,@(20,r14)
	mov.l	r6,@(24,r14)
	mov.l	r7,@(28,r14)
	...

and r1 is used to save fr1 before saving itself to the block.

Regards,
	kaz