Re: accuracy ...



The discussion is revisiting a lot of work that was done in the 1950s
and 1960s on floating point arithmetic. The gold standard is to
accumulate in a double length mantissa. On some architectures that was
easy to do. The i386 architecture uses IEEE 754 as far as I am aware (I
have not looked at details for a while). This makes it fairly easy to
double length accumulate reals, but almost all current s/w uses doubles
by default (roughly 15 decimal digits equivalent). This means
accumulations have to be done in quad, which wasn't part of the standard
toolkit.

If I ever get jhbuild to complete, I may even be willing to try.

JN




[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]