help on soft-float

Mon Dec 22 09:50:00 GMT 2003

Hi,

my problem is the  cross compiler could not support
soft float.Since my cpu doesnt have fp,so I need the
soft-float emulation.I have written a test program:

/*
 * mygcc.c
 * Test Program
 */
double x,y,z;

int
main(int argc, char *argv[])
{
	x=128;
	y=234;
	z=x+y;
	return 0;
}
then:
gcc -msoft-float -fomit-frame-pointer -G 0 -g -drx -o mygcc mygcc.c

the RTL insn list are:
/**********************************************/
;; Function main

(note 1 0 2 ("mygcc.c") 12)

(note 2 1 4 "" NOTE_INSN_DELETED)

(insn 4 2 6 (set (mem/f:SI (reg:SI 76) 0)
        (reg:SI 4 a0)) -1 (nil)
    (nil))

(insn 6 4 7 (set (mem/f:SI (plus:SI (reg:SI 76)
                (const_int 4 [0x4])) 0)
        (reg:SI 5 a1)) -1 (nil)
    (nil))

(note 7 6 8 "" NOTE_INSN_FUNCTION_BEG)

(note 8 7 9 "" NOTE_INSN_DELETED)

(note 9 8 10 ("mygcc.c") 13)

(note 10 9 12 "" NOTE_INSN_DELETED)

(insn 12 10 14 (set (reg:DF 81)
        (const_double:DF (const_int 0 [0x0]) 0 [0x0] 1080033280 [0x40600000] [128])) -1 (nil)
    (expr_list:REG_EQUAL (const_double:DF (const_int 0 [0x0]) 0 [0x0] 1080033280 [0x40600000] [128])
        (nil)))

(insn 14 12 15 (set (mem/f:DF (symbol_ref:SI ("x")) 0)
        (reg:DF 81)) -1 (nil)
    (nil))

(note 15 14 17 ("mygcc.c") 14)

(insn 17 15 19 (set (reg:DF 82)
        (const_double:DF (const_int 0 [0x0]) 0 [0x0] 1080901632 [0x406d4000] [234])) -1 (nil)
    (expr_list:REG_EQUAL (const_double:DF (const_int 0 [0x0]) 0 [0x0] 1080901632 [0x406d4000] [234])
        (nil)))

(insn 19 17 20 (set (mem/f:DF (symbol_ref:SI ("y")) 0)
        (reg:DF 82)) -1 (nil)
    (nil))

(note 20 19 22 ("mygcc.c") 15)

(insn 22 20 24 (set (reg:DF 4 a0)
        (mem/f:DF (symbol_ref:SI ("x")) 0)) -1 (nil)
    (insn_list:REG_LIBCALL 28 (nil)))

(insn 24 22 26 (set (reg:DF 6 a2)
        (mem/f:DF (symbol_ref:SI ("y")) 0)) -1 (nil)
    (nil))

(call_insn/u 26 24 28 (parallel[ 
            (set (reg:DF 2 v0)
                (call (mem:SI (symbol_ref:SI ("dpadd")) 0)
                    (const_int 16 [0x10])))
            (clobber (reg:SI 31 ra))
        ] ) -1 (nil)
    (expr_list:REG_EH_REGION (const_int -1 [0xffffffff])
        (nil))
    (expr_list (use (reg:DF 6 a2))
        (expr_list (use (reg:DF 4 a0))
            (nil))))

(insn 28 26 30 (set (reg:DF 83)
        (reg:DF 2 v0)) -1 (nil)
    (insn_list:REG_RETVAL 22 (expr_list:REG_EQUAL (plus:DF (mem/f:DF (symbol_ref:SI ("x")) 0)
                (mem/f:DF (symbol_ref:SI ("y")) 0))
            (nil))))

(insn 30 28 31 (set (mem/f:DF (symbol_ref:SI ("z")) 0)
        (reg:DF 83)) -1 (nil)
    (nil))

(note 31 30 33 ("mygcc.c") 16)

(insn 33 31 34 (set (reg/i:SI 2 v0)
        (const_int 0 [0x0])) -1 (nil)
    (nil))

(insn 34 33 35 (use (reg/i:SI 2 v0)) -1 (nil)
    (nil))

(jump_insn 35 34 36 (set (pc)
        (label_ref 40)) -1 (nil)
    (nil))

(barrier 36 35 37)

(note 37 36 38 ("mygcc.c") 17)

(note 38 37 39 "" NOTE_INSN_FUNCTION_END)

(note 39 38 40 ("mygcc.c") 17)

(code_label 40 39 0 2 "" [num uses: 0])

/************************************************/

then:
gcc -msoft-float -fomit-frame-pointer -G 0 -g -o mygcc mygcc.c
I use gdb to debug mygcc,I found that z didnt be assigned the
correct value.so I disassembled mygcc,the assembly code is:

int
main(int argc, char *argv[])
{
  40080c:	afa40020 	sw	$a0,32($sp)
  400810:	afa50024 	sw	$a1,36($sp)
	x=128;
  400814:	3c024060 	lui	$v0,0x4060
  400818:	00001821 	move	$v1,$zero
  40081c:	8f81805c 	lw	$at,-32676($gp)
  400820:	00000000 	nop
  400824:	ac220000 	sw	$v0,0($at)
  400828:	8f81805c 	lw	$at,-32676($gp)
  40082c:	00000000 	nop
  400830:	ac230004 	sw	$v1,4($at)
	y=234;
  400834:	8f818018 	lw	$at,-32744($gp)
  400838:	00000000 	nop
  40083c:	8c2230d0 	lw	$v0,12496($at)
  400840:	8c2330d4 	lw	$v1,12500($at)
  400844:	8f818034 	lw	$at,-32716($gp)
  400848:	00000000 	nop
  40084c:	ac220000 	sw	$v0,0($at)
  400850:	8f818034 	lw	$at,-32716($gp)
  400854:	00000000 	nop
  400858:	ac230004 	sw	$v1,4($at)
	z=x+y;
  40085c:	8f84805c 	lw	$a0,-32676($gp)
  400860:	00000000 	nop
  400864:	8c840000 	lw	$a0,0($a0)
  400868:	8f85805c 	lw	$a1,-32676($gp)
  40086c:	00000000 	nop
  400870:	8ca50004 	lw	$a1,4($a1)
  400874:	8f868034 	lw	$a2,-32716($gp)
  400878:	00000000 	nop
  40087c:	8cc60000 	lw	$a2,0($a2)
  400880:	8f878034 	lw	$a3,-32716($gp)
  400884:	00000000 	nop
  400888:	8ce70004 	lw	$a3,4($a3)
  40088c:	8f998040 	lw	$t9,-32704($gp)
  400890:	00000000 	nop
  400894:	0320f809 	jalr	$t9
  400898:	00000000 	nop
  40089c:	8fbc0010 	lw	$gp,16($sp)
  4008a0:	00000000 	nop
  4008a4:	8f818044 	lw	$at,-32700($gp)
  4008a8:	00000000 	nop
  4008ac:	ac220000 	sw	$v0,0($at)
  4008b0:	8f818044 	lw	$at,-32700($gp)
  4008b4:	00000000 	nop
  4008b8:	ac230004 	sw	$v1,4($at)
	return 0;
  4008bc:	00001021 	move	$v0,$zero
  4008c0:	10000001 	b	4008c8 <main+0xd8>
  4008c4:	00000000 	nop
}

My question is: 1) Is the RTL insn list correct?
if it is,then what makes the assembly wrong?
if it is not,how could I fix it?

Since I am a newbie on GCC,I only know a little about gcc's
behavior.

Best Regards,
Alan
>Liu Hongming wrote:
>> I am using gcc cross-compiler for mips bought from some vendors,
>> and it doesnt support -msoft-float option(I dont know why it is a difficulty). 
>
>It isn't clear what your problem is.  What kind of error do you get when
>you try to use the -msoft-float option?  Is this a missing multilib problem?
>
>If you have sources for your cross-compiler, it shouldn't be too hard to
>enable soft-float multilibs and rebuild it.  You need to define the
>MULTILIB_* Makefile variables in the target t-* file.  See for instance
>the example in the gcc/config/mips/t-elf file.
>-- 
>Jim Wilson, GNU Tools Support, http://www.SpecifixInc.com