Re: [issue18062] m68k FPU precision issue

[Laurent Vivier <Laurent@Vivier.EU>]

Le 26/05/2013 19:44, Thorsten Glaser a écrit :
> Eero Tamminen dixit:
>
> > > > Now as additional data point, UAE/WinUAE/etc. would be interesting.
> > I built the test with fpu_control.h header from eglibc, using
> > Sparemint GCC 2.9.5 (with 2010 binutils) and MiNTlib.  When it's
> Nice ;)
>
> > I.e. it seems that WinUAE FPU emulation is also lacking FPUCW change
> > handling (for precision).
> >
> > (Hatari's WinUAE CPU core code was synched with upstream last year.)
> OK, thanks. I’d just say let’s say changing FPU precision is not
> part of the target we support. (Funnily enough, ColdFire according
> to the ’net has (unchangeable) 64-bit precision… maybe let’s just
> say precision on m68k in general is not defined.)
m68k has 32, 64 and 80 bit precision [1]. Internally all computing are 
done with a mantissa of 67-bit and a 16-bit exponent (17 on 68882/68881) 
[2].
If I remember correctly, last time I checked, Aranym was using 
internally only 64bit floating point type (like coldfire).

Regards,
Laurent
"MOTOROLA M68000 FAMILY Programmer’s Reference Manual (includes CPU32 
instructions)"
[1] "1.5.2 Binary Floating-Point Formats"
[2] "3.5.1 Intermediate Result"