problem while building arm vfp softfloat gcc

[add]

sorry for my testcase, I had rewriten it to easy to understand, but on the
contrary...
Here is my true test case using printf:
----main.c
#include <stdio.h>
#include <stdlib.h>

int main()
{
        float i=1.0;
        printf("%f",i);
        return 0;
}
----main.c

here is generated by arm-linux-gcc -S main.c -o main.s
----main.s
<sniz>
main:
          @ args = 0, pretend = 0, frame = 4
          @ frame_needed = 1, uses_anonymous_args = 0
          mov     ip, sp
          stmfd   sp!, {r4, fp, ip, lr, pc}
          sub     fp, ip, #4
          sub     sp, sp, #4
          ldr     r3, .L2 @ float
          str     r3, [fp, #-20]  @ float
          ldr     r0, [fp, #-20]  @ float
          bl      __extendsfdf2
          mov     r4, r1
          mov     r3, r0
          ldr     r0, .L2+4
          mov     r2, r4
          mov     r1, r3
          bl      printf
          mov     r3, #0
          mov     r0, r3
<sniz>
-----main.s

if I arm-linux-gcc main.s -o test
then execute test on pxa board, it will output: -1.999744

but if I modified the two lines before "bl printf" in main.s to:
mov	r2, r3
mov	r3, r4
then compile main.s and execute,
it will print right result: 1.00000

so conclusion:
my toolchain must has a problem of float param passing, seemly related to
float word order. so I want to dig into my gcc source tree to find where is
responsible for generating this wrong asm code.
I hope this time it's clearer :)

and Dan, I think the most recently snapshot ptxdist still has this problem
:(


in your letter you wrote:
>add <addsub@eyou.com> wrote:
>> I'm building a arm-linux-gcc toolchain with soft-vfp support by
>> crosstool-0.27(Thanks to Dan). But generated code cannot execute
>> "printf("%f",1.00)" well, it will output 0.00000.
>> 
>> Then I found gcc doesn't produce right code, it uses wrong registers
sequence
>> to pass float argument to function()
>
>Richard Earnshaw wrote:
>> To the best of my knowledge the only configuration of a released version 
>> of gcc that supports soft-vfp is arm-netbsdelf.
>
>He's using a patch that adds that, I think.  The patch is taken from
>http://www.pengutronix.de/software/ptxdist/patches-cvs/gcc-3.3.2/generic/gcc-
3.3.2-arm-softfloat.diff
>This is mostly just Nicholas Pitre's patch:
>http://lists.arm.linux.org.uk/pipermail/linux-arm/2003-October/006436.html
>
>(The script he's using to build gcc is
>http://kegel.com/crosstool/crosstool-0.27/demo-arm-softfloat.sh
>and the complete set of gcc patches he's using is
>http://kegel.com/crosstool/crosstool-0.27/patches/gcc-3.3.2/
>most of which just deal with the regression test suite,
>or fix minor sh4 problems.)
>
>Anyway, he says in his little "float i=1.0; foo(i);" test program, gcc
generates
>         mov     r4, r1
>         mov     r3, r0
>         mov     r1, r4           /* this line wrong -by addsub*/
>         mov     r0, r3           /* this line wrong -by addsub*/
>         bl      foo
>when it should generate
>         mov     r4, r1
>         mov     r3, r0
>         mov	r2, r3
>         mov	r3, r4
>         bl      foo
>I don't know anything about arm assembly, and it'll be a while before
>I can dig into this.  I'd probably start by just building the latest cvs
>version of ptxdist's toolchain, and see where its gcc build procedure
>differs from mine... although I could also try running the gcc regression
>tests on the arm hardware I have.  Anyway, perhaps Robert or Nicolas will
>rescue me and tell me what I forgot :-)
>- Dan
>
> 





--http://www.eyou.com
--Îȶ¨¿É¿¿µÄµç×ÓÐÅÏä  ÓïÒôÓʼþ  Òƶ¯ÊéÇ©  ÈÕÀú·þÎñ  ÍøÂç´æ´¢...ÒÚÓÊδ¾¡

--http://vip.eyou.com
--¿ì¿ìµÇ¼ÒÚÓÊVIPÐÅÏä  ×¢²áÄúÖÐÒâµÄÓû§Ãû



------
Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
Want to unsubscribe? Send a note to crossgcc-unsubscribe@sources.redhat.com