Welcome to the double Mersenne and Fermat factoring program. It is a fast GPU factoring program to find new factors
of small Fermat numbers (F26 to F223) and small double Mersenne numbers (MM31, MM61, MM89, MM107, and MM127).
It requires a Nvidia GPU with compute capabilities 2.0 or higher.
Visit http://www.fermatsearch.org to grab a range of k values for Fermat numbers.
Visit http://www.doublemersennes.org for the double Mersenne numbers.
Here is some of what you need to know:
mmff.exe is derived from mfaktc.exe. Like mfaktc.exe, mmff.exe uses a .ini file to control its behavior (mmff.ini).
You'll want to adjust GPUSievePrimes and GPUSieveSize for best performance.
Mmff.exe also uses a worktodo.txt file to control what numbers to factor.
For double-Mersennes, we tell mmff what double-Mersenne to try and factor.
For Fermat factoring, we tell mmff which factor candidates to try since a factor candidate might divide one of several Fermat numbers.
Add this line to factor a double-Mersenne number:
MMFactor=exponent,minimum-bit-level-or-k-value,maximum-bit-level-or-k-value
FermatFactor=N,minimum-k-value,maximum-k-value
Since Mersenne factors are of the form 2*k*p + 1, the k-values for double-Mersennes yield factors 2*k*M(exponent)+1.
Fermat factors are of the form k*2N + 1 where k is odd.
A factor can divide any Fermat number from F12 up to F(N-2).
Example worktodo.txt entries:
MMFactor=127,179,180
FermatFactor=33,5460e9,5470e9
The first example looks for factors of MM127 from 2179 to 2180 (a big job that really should be split up into many k ranges.
The second example looks for any Fermat number with a factor between 5460000000000*233+1 and 5470000000000*233+1
(there is a known factor of F31 in that range).
To run the program once you've setup worktodo.txt and mmff.ini is very easy. On linux type:
./mmff.exe >> my_output_file
Be sure to check the output file or results.txt for any new factors!
Mark Rodenkirch developed a program to search for bigger DoubleMersenne factors it's called dmdsieve and can be found on SourceForge
It can handle values from MM521 to MM86243.
Mark Rodenkirch also developed a suite of sievers called mtsieve.
The suit contains two sievers devoted to double mersenne numbers, called dmdsieve and dmdsievecl.
The first uses a multi-threaded CPU approach to sieve out all k of the factor 2kp+1 that make the potential factor composite.
dmdsievecl adopts the same approach, but using OpenCL on GPUs, and it is up to 50+ faster depending on the GPU.
For those who prefer using other sievers, Mark developed dmd_tester, a program that takes a presieved file in inut and tests each candidate. Source for Linux and executable for Windows.