Greg Walker wrote:
>
> A mere curiosity ...
>
> I have a class that performs many calcs. The values are *slightly*
> (after about 14 significant figures) different depending on the jvm I
> use (either ibm's 1.1.6 or blackdown's 1.1.7). Is there a difference
> between algorithms to solve for trig functions or logarithms between
> jvm's? If so, which is more accurate? Isn't there a standard for math
> operations (IEEE) that both jvm's use? I am not sure I see why this
> happens. jit or nojit does not make a difference (ibm) and native or
> green does not make a difference (blackdown).
IEEE754 does a lot of good things, but it doesn't guarantee what you're
asking for here. Take a look at
http://www.validgh.com/goldberg/addendum.html for gory levels of detail.
I'm not ruling out that something odd is happening, but floating-point
math is a difficult area in which simple qualifications like "more
accurate" are not easy to characterize.
Nathan
> Note: I have NOT been able to reproduce this with single math
> operations.
>
> --
> Greg Walker
> [EMAIL PROTECTED]
>
> ----------------------------------------------------------------------
> To UNSUBSCRIBE, email to [EMAIL PROTECTED]
> with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
----------------------------------------------------------------------
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
