Another way is to count the bits, as in the following:
.sub _main
N1 = 1
N2 = 0.5
I0 = 0
REPEAT:
I0 = I0 + 1
N2 = N2 / 2
N3 = N1 + N2
ne N1, N3, REPEAT
print I0
print " bits precision\n"
end
.end
On Wed, 17 Nov 2004 01:13:21 +1300, Adam Warner <[EMAIL PROTECTED]> wrote:
> Hi Leopold Toetsch,
>
> >> PI is (very) approximately:
> > 3.141590653589694692726652647252194583415985107421875000000000
> > ^^^^^
> > 3.141592653589793238462643383279502884197169399375105820974944
> >
> > You might probably want to run more iterations ;) And you'll never get
> > 60 digits out of long doubles.
>
> I appreciate that Leopold. I thought the zeros indicated where the stored
> value cuts off. But with 52 significant figures at ~3.3 bits per figure
> that's about 172 bits, which indicates they would have to be 256 bit
> floats when including the exponent (my mistake in calling them 128 bit)!
>
> In other words sprintf is printing trailing garbage:
>
> .sub _main
> set N1, 2.0
> set N2, 3.0
> div N1, N1, N2
> new P0, .PerlArray
> set P0, 1
> set P0[0], N1
> sprintf S0, "%.60f\n", P0
> print S0
> end
> .end
>
> On the CVS version I compiled this prints:
> 0.666666666666666629659232512494781985878944396972656250000000
>
> On this win32 release (running on a different machine):
> http://www.jwcs.net/developers/perl/pow/download/pow-0.1.1-release.zip
> parrot prints:
> 0.666666666666666630000000000000000000000000000000000000000000
>
> Regards,
> Adam
>
>
