[Bug c/42076] regression on float representation

[vaillant dot etienne at gmail dot com]

------- Comment #2 from vaillant dot etienne at gmail dot com  2009-11-18 11:17 -------
(In reply to comment #1)
> You are likely running into excess precision issues of i?86.  Try using
> -mpc64 or -mfpmath=sse.
> 

I try :
$ gcc-4.4 -msse2 y.c -o y-4.4
$ ./y-4.4
ration=0.210000 ratio*10000000=2099999

same problem :(

I watch asm code and :

c code :
double ratio = 0.21;

asm code on 4.3 :
fldl   0x8048580
fstpl  -0x10(%ebp)

asm code en 4.4 :
fldl   0x8048590
fstpl  0x38(%esp)

I think the is on 0x8048580 and 0x8048590


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42076