Richard Zidlicky wrote:
> that would be somewhat advanced discrete algebra, iirc that exists
> but I was never very good at this abstract stuff. Taking ln/log values
> of bit arrays is of course standard practice but should be reasonable
> robust wrt implementation details change.
Values without a fractional part should never pose a problem. Values
computed with the new QPC routines MAY sometimes be more precise than
the one calculated by the SMSQ/E routines. But that's it. I wouldn't
rely on a certain bit pattern of sin(sqrt(2))anyway ;-)
And the mentioned problem was the only one found so far.
BTW: The problem is fixed now. It was far easier than I thought, I
didn't even have to change one single line of code. Fixed version will
be released soon.
Marcel
