Re: [hatari-devel] FPU update

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

Le 08/01/2017 à 13:56, Toni Wilen a écrit :
> > - Adds an option to use software floating point arithmetics (via
> > SoftFloat library) - Fixes lots of bugs, mostly related to rounding
> > and normalization. At least one rounding bug was critical (crashed
> > Mac OS from many applications) - Improved handling of the FPSR
> > register, although this needs more testing and comparing to real
> > systems
>
> Thanks. Looks more cleaner (and much more functional, I don't have any
> good Amiga test cases) than my original and not really finished
> softfloat support.
>
> But I really want to have "normal" and softfloat FPU emulation on the
> fly switchable (not compile time), I moved your new low level functions
> back to fpu.cpp and combined both modes.
>
> I can do testing if needed, I have 68030+68882, 68040 and 68060 based
> Amigas. I don't have any 68881 systems.

Hi

I committed Andreas's changes that Toni merged into WinUAE to Hatari 
devel version ; note that in Hatari there's no setting yet to switch 
between "normal" and softfloat version, but at least everything compiles 
fine (ie softfload is not used yet)


I saw a few warning in fpp.c :

To clear fxsizes, we should use :

static floatx80 fxsizes[6] = { { 0 } };
  instead of
static floatx80 fxsizes[6] = { 0 };


Some printf warnings :

/home/npomarede/src/hatari-cpu/src/cpu/fpp.c: In function 'fp_print':
/home/npomarede/src/hatari-cpu/src/cpu/fpp.c:723:4: attention : format 
'%Le' expects argument of type 'long double', but argument 4 has type 
'double' [-Wformat=]
    _stprintf(fs, _T("%c%#.17Le%s%s"), n?'-':'+', (fptype) 0.0, u ? 
_T("U") : _T(""), d ? _T("D") : _T(""));
    ^
/home/npomarede/src/hatari-cpu/src/cpu/fpp.c:737:4: attention : format 
'%Le' expects argument of type 'long double', but argument 4 has type 
'fptype' [-Wformat=]
    _stprintf(fs, _T("%c%#.17Le%s%s"), n?'-':'+', result, u ? _T("U") : 
_T(""), d ? _T("D") : _T(""));


This is because Andreas' patch also set USE_LONG_DOUBLE, so %Le is OK. 
But in WinUAE sources, USE_LONG_DOUBLE is not set, so fptype will just 
be "double", hence the warning.
Should we use long double ?


Also, winuae has a file "include/fpp-softfloat.h", but it's never 
included, it seems the code was moved directly in fpp.c ?

Nicolas