On Feb 16 2017, Richard Henderson <r...@twiddle.net> wrote:
> On 02/07/2017 11:59 AM, Laurent Vivier wrote:
>> +static long double floatx80_to_ldouble(floatx80 val)
>> +{
>> + if (floatx80_is_infinity(val)) {
>> + if (floatx80_is_neg(val)) {
>> + return -__builtin_infl();
>> + }
>> + return __builtin_infl();
>> + }
>> + if (floatx80_is_any_nan(val)) {
>> + char low[20];
>> + sprintf(low, "0x%016"PRIx64, val.low);
>> +
>> + return nanl(low);
>> + }
>> +
>> + return *(long double *)&val;
>> +}
>
> This doesn't work except for x86 host.
Not even then.
> You ought to extract the mantissa, convert the 64-bit value to
> long-double, and use ldexpl to scale the result for the exponent.
>
> Similarly converting the other way use frexpl and ldexpl.
There is no guarantee that the host long double has the same range and
precision as floatx80.
Andreas.
--
Andreas Schwab, sch...@linux-m68k.org
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756 01D3 44D5 214B 8276 4ED5
"And now for something completely different."
