Re: [Qemu-devel] [PATCH v2 2/5] target-tricore: Added MADD.F and MSUB.F instructions

[Richard Henderson]

On 11/08/2016 04:08 PM, Bastian Koppelmann wrote:
> On 11/08/2016 12:42 PM, Richard Henderson wrote:
> > On 11/07/2016 03:44 PM, Bastian Koppelmann wrote:
> > > Multiplies D[a] and D[b] and adds/subtracts the result to/from D[d].
> > > The result is put in D[c]. All operands are floating-point numbers.
> > >
> > > Signed-off-by: Bastian Koppelmann <address@hidden>
> > > ---
> > >  target-tricore/fpu_helper.c | 93
> > > ++++++++++++++++++++++++++++++++++++++++++++-
> > >  target-tricore/helper.h     |  2 +
> > >  target-tricore/translate.c  |  8 ++++
> > >  3 files changed, 102 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/target-tricore/fpu_helper.c b/target-tricore/fpu_helper.c
> > > index f717b53..d530a0b 100644
> > > --- a/target-tricore/fpu_helper.c
> > > +++ b/target-tricore/fpu_helper.c
> > > @@ -21,7 +21,8 @@
> > >  #include "cpu.h"
> > >  #include "exec/helper-proto.h"
> > >
> > > -#define ADD_NAN   0x7cf00001
> > > +#define QUIET_NAN 0x7fc00000
> > > +#define ADD_NAN   0x7fc00001
> > >  #define DIV_NAN   0x7fc00008
> > >  #define MUL_NAN   0x7fc00002
> > >  #define FPU_FS PSW_USB_C
> > > @@ -47,6 +48,42 @@ static inline bool f_is_denormal(float32 arg)
> > >      return float32_is_zero_or_denormal(arg) && !float32_is_zero(arg);
> > >  }
> > >
> > > +static inline float32 f_maddsub_nan_result(float32 arg1, float32 arg2,
> > > +                                           float32 arg3, float32 result,
> > > +                                           uint32_t flags)
> > > +{
> > > +    uint32_t aSign, bSign, cSign;
> > > +    uint32_t aExp, bExp, cExp;
> > > +
> > > +    if (float32_is_any_nan(arg1) || float32_is_any_nan(arg2) ||
> > > +        float32_is_any_nan(arg3)) {
> > > +        return QUIET_NAN;
> > > +    } else if (float32_is_infinity(arg1) && float32_is_zero(arg2)) {
> > > +        return MUL_NAN;
> > > +    } else if (float32_is_zero(arg1) && float32_is_infinity(arg2)) {
> > > +        return MUL_NAN;
> > > +    } else {
> > > +        aSign = arg1 >> 31;
> > > +        bSign = arg2 >> 31;
> > > +        cSign = arg3 >> 31;
> > > +
> > > +        aExp = (arg1 >> 23) & 0xff;
> > > +        bExp = (arg2 >> 23) & 0xff;
> > > +        cExp = (arg3 >> 23) & 0xff;
> > > +
> > > +        if (flags & float_muladd_negate_c) {
> >
> > flags here is muladd argument,
>
> I don't know what you mean by that.

I mean that the type you pass is not the type you check here.
They're different sets of bitmasks.

> I thought of extending softfloat's muladd() to return a specializable
> ADD_NAN to catch this case in the first place, but TriCore would be the
> only target using it and I'm not sure if it's worth while.
>
> Any comments on this idea?

I think just fixing up tricore is the way to go.


r~