On 05/11/2018 09:26 AM, Alex Bennée wrote:
>
> Richard Henderson <richard.hender...@linaro.org> writes:
>
>> Shift the NaN fraction to a canonical position, much like we do
>> for the fraction of normal numbers. Immediately, this simplifies
>> the float-to-float conversion. Later, this will facilitate
>> manipulation of NaNs within the shared code paths.
>>
>> Signed-off-by: Richard Henderson <richard.hender...@linaro.org>
>> ---
>> fpu/softfloat.c | 13 ++++++-------
>> 1 file changed, 6 insertions(+), 7 deletions(-)
>>
>> diff --git a/fpu/softfloat.c b/fpu/softfloat.c
>> index 5e4982b035..df377b6314 100644
>> --- a/fpu/softfloat.c
>> +++ b/fpu/softfloat.c
>> @@ -329,10 +329,11 @@ static FloatParts canonicalize(FloatParts part, const
>> FloatFmt *parm,
>> if (part.frac == 0) {
>> part.cls = float_class_inf;
>> } else {
>> + part.frac <<= parm->frac_shift;
>> #ifdef NO_SIGNALING_NANS
>> part.cls = float_class_qnan;
>> #else
>> - int64_t msb = part.frac << (parm->frac_shift + 2);
>> + int64_t msb = part.frac << 2;
>> if ((msb < 0) == status->snan_bit_is_one) {
>> part.cls = float_class_snan;
>> } else {
>> @@ -498,6 +499,7 @@ static FloatParts round_canonical(FloatParts p,
>> float_status *s,
>> case float_class_qnan:
>> case float_class_snan:
>> exp = exp_max;
>> + frac >>= parm->frac_shift;
>
> Unfortunately we now miss this step when we handle:
>
> case float_class_msnan:
> return float64_maybe_silence_nan(float64_pack_raw(p), s);
>
> This seems to fix that:
>
> recanonicalise frac before packing and sending
Good catch. Merged for bisection.
r~
