Update on 8014320

Tim Buktu tbuktu at hotmail.com
Mon Dec 23 19:37:53 UTC 2013


I've been quiet because real life needed my attention.
Obviously, the Schoenhage-Strassen and Barrett algorithms are not going
to make it into JDK8. I hope that is not a problem for anyone. We
shouldn't use the word "problem" anyway but rather think of it as an
opportunity :-) for more refinement and testing.

Improvements I have planned are:
 * Compact storage of DFT vectors. At the moment, they are stored as
numbers modulo 2^2^(n+1) when (2^2^n)+1 would suffice. This will save
memory and may speed up the DFT and IDFT steps.
 * Use Bailey's algorithm recursively if it improves cache locality and
doesn't add too much complexity to the code.
 * Maybe add a multiplyParallel() method, now that dft() and idft() use
Bailey's algorithm which should make them easy to parallelize.
 * See if dft() and idft() can be made faster on 64-bit CPUs by using
long[][] arrays instead of int[][].

As always, comments are welcome.
Merry Chris... I mean, Holidays and a happy new year to everyone!


More information about the core-libs-dev mailing list