Dear Mersenners:

The source and binaries for selected platforms of Mlucas 2.7a are
now available: see

ftp://209.133.33.182/pub/mayer/README.html

Major changes from v2.7z include:

1) v2.7a has a radix-9 FFT pass capability. This allows one to
go from FFT lengths of (say) 256K and 512K to 288K and 576K, rather
than having to jump all the way up to 320K and 640K, respectively.
This will save folks double-checking exponents above 5.15M or first-
time testing above 10.11M anywhere from 12-20% runtime, depending on
the FFT length and the platform. For example, Brian Beesley was able
to do a double-check of an exponent ~5.6M in just 8 days on his
533MHz Alpha PC164; that exponent would have taken 10 days using v2.7z.

2) v2.7a now allows users to concatenate several partially-completed
exponents and their associated savefiles into a single directory; as
the code gets to each new exponent in the worktodo.ini file, it first
checks to see if there are any savefiles (v2.7z did this only at on
program startup, i.e. for the first entry in the .ini file.) Thanks
to Siegmar Szlavik for suggesting this.

3) I've condensed some near-repeat code sequences in the carry and
wrapper/pointwise-square routines in an effort to reduce their code
footprint and thus improve instruction cache usage. Some platforms
(Alpha ev4 and ev6, Sparc) benefit from this; on others (Alpha ev5
and MIPS) it seems to make little or no difference.


V2.7a is a drop-in replacement for 2.7z; I suggest users upgrade at
their earliest convenience, so as to have the radix-9 capability ready
when the PrimeNet server starts handing out exponents in the aforementioned
ranges, which will occur in the near future.

As usual, many thanks to Bill Rea and Brian Beesley, my Sparc and Alpha
Linux "compile czars," for the binaries and tunings for those platforms.

Happy hunting,
-Ernst

p.s.: For those of you confused by my code numbering convention: I use the
suffixes x, y and z to label beta versions of the code, thus z (if I choose
to use it) comes before a.

_________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ      -- http://www.tasam.com/~lrwiman/FAQ-mers

Reply via email to