Re: Bignums, object grind, and garbage collection
John W. Kennedy wrote:
John Ersatznom wrote:
John W. Kennedy wrote:
John Ersatznom wrote:
* I'm not sure the implementation is as fast as
possible. In particular, it doesn't seem to be
documented anywhere whether these classes switch to
Karatsuba multiplication and then to FFTs at
appropriate bit lengths or just stick with slow, O(n^2)
long multiplication throughout.
For whatever it's worth, Ruby finds the first 14 perfect numbers a
/lot/ faster than Java does. (Perl is a lot slower. GCLISP is fastest.)
Is Ruby a broadly-ported system with free* development tools, a
similar range of standard classes and APIs including decent GUI and
other functionality, with a decent free* IDE available, and one that's
somehow on-topic for comp.lang.java.programmer? If so, I may look into
it.
It's got most of that, in fact. But my point is that, even though Ruby
is interpreted -- not even bytecoded -- its big-number performance is
about five times faster than Java's, HotSpot and all, which rather
implies that Java's big numbers could be a lot sleeker than they are.
Sounds like I need to go with my own bignums then.
Any suggestions on implementing the basic
array-of-primitive-integer-type case are welcome, as well as any
pointers to freely reusable libraries for the purpose. (They must be
freely distributable with the calling code. LGPL preferred, and failing
that some kind of open source preferred, but anything that can be called
from GPL and LGPL code and binary distributed with same would do.
Requiring a nonstandard native library is a big minus.)