Gary E. Miller writes: > That because a float is only 21 significant digits, the same significant > digits as in 7FFFFC00
Nope, a float has 24 binary digits for the mantissa (including the hidden bit), unless they changed the standard while I wasn't looking. So there really isn't an explanation of where the extra three missing bits went. > So somewhere netBSD is using a float instead of a double... > Or something else that is netBSD specific... Not necessarily. The printf conversion specifier in that test is bunk, use %g or even better %a and specify the correct precision. Also, add the index into the table somewhere in the output so one can see what test was actually performed and output _all_ intermediates. > TEST(lfpfunc, FDF_RoundTrip)../../tests/libntp/lfpfunc.c:268::FAIL: > Expected 0.0 > Was 2147482624.0. 2147483647.500000 diff 2147482624.000000 not within > 2.384186e-07 The operative number here is actually the third one (diff) and it looks to me like op3 was zero (the conversion back from double was'#t performed correctly). You'd likely gain more insight if you looked at the bit pattern from temp before converting it back to double. Regards, Achim. -- +<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+ Samples for the Waldorf Blofeld: http://Synth.Stromeko.net/Downloads.html#BlofeldSamplesExtra _______________________________________________ devel mailing list [email protected] http://lists.ntpsec.org/mailman/listinfo/devel
