Thanks to some great patches from Dana Jacobsen (and some version numbering snafu from me), we have arrived at Math::Prime::FastSieve v0.12.
Don't bother with v0.11 (I uploaded with inconsistent version numbering). v0.12 provides the following enhancements over v0.10: The prime sieve's bit vector only contains odds, cutting the bit vector's memory footprint in half, and improving speed performance of building a sieve considerably (20-30% in some cases). This patch came from Dana Jacobsen. I've also taken Dana's suggestion and used unsigned longs internally (and longs as parameters). This allows Perl's built with long support to handle (theoretically) sieves up to 9.2e18. ( 2^64 / 2 --- if we were passing unsigned longs as params it would be 2^64, but Inline::CPP's typemap doesn't yet support unsigned longs, so we pass longs). Generating a sieve of 8_000_000_000 takes 8_000_000_000 / 8 / 2 bytes, so about 477MB of RAM, and on one of my systems, generates the sieve in about 40 seconds. A sieve of 3_000_000 can be produced in 1/240th of a second on the same machine. To my knowledge, M::P::FS is the fastest primes generator and toolkit on CPAN. As before, this module depends on Inline::CPP and Inline::MakeMaker. -- David Oswald daosw...@gmail.com
