Athlon Optimized GFN Deep Sieve

SSE2 Update

Build 213 - Add logging all output to athgfnsv.log in the same directory as the executable. Add control of the priority class under the options menu. The worker thread is always below normal priority in whatever class is selected.

Build 212 - No changes to the wide sieve logic or implementation. Some speed increases were added to the standard trial division sieve (SSE2 version) when processing small numbers of candidates. Also there was an occasional consistency warning generated in this section. The cause of the warning was fixed. Lower the minimum number of candidates required in a standard sieve to 1. The packed K file logic was enabled. There are various updates to specialized features and report generation currently used only by the author.

Build 206 - Removed the GFN lower limit of 8192. N can now be any GFN exponent. Also updated the resume sieve for the non-wide mode to allow for a resume of a standard sieve without a factoring input file. Leave this input blank (empty) to resume a standard sieve from only a sieve output file.

Build 204 - Added SSE2 support for Pentium 4 computers. The P4 and Athlon versions are now both very fast. Build 204 runs on PIII, P4, or Athlon. It is optimized only for Athlon and P4.

Credits Update

Phil Carmody lead the research into the math required for a wide GFN sieve algorithm. His ideas made wide sieving possible.

Yves Gallot provided some key ideas into optimizing the SSE2 implementation. I am very grateful for the time and effort he contributed to this effort.

The speed is measured on the range of p values calculated per day at N=65536, P=600 trillion. Larger P values will be a little faster. Smaller P values will be a little slower.

Wide Sieve Update

There is no reason for users to seive their own GFN ranges. The work has already been completed by Phil Carmody, and is available by following the links from the status page of the Generalized Fermat Prime Search web site at http://perso.wanadoo.fr/yves.gallot/primes/gfn.html

You can go directly to Phil's pages at http://fatphil.org/maths/GFN/index.html

Warning, if you forget to select the Wide Range Math option and specify a wide range, you may try to allocate hundreds of megabytes of RAM to do a normal process. Expect 1 MB for each 10,000 b values in the range. With Wide Range Math, this is not a consideration.

New Wide Sieve - Select Start New Sieve, check Wide Range Math, select factors output file and candidates output file, exponent and b range, and hit start. In wide range math, the factors are appended to the factors output file. If wide range math is not selected, the factors output file is overwritten (after a warning and a chance to abort before actually overwriting it.) The factors file is not sorted by b values.

Continue Wide Sieve - Select Resume Old Sieve, check Wide Range Math, select factors output file, in candidate file and out candidate file. Again, in wide range math, the factors are appended to the factors output file. If wide range is not selected, a warning will be given with an abort option before overwriting the factors file.

To regenerate a sorted factor and candidate file from a factors file generated by Wide Range Math, select the Build From Factors option. All the factors are double checked, sorted, and written using the format of the normal factoring options. Candidates are sorted and written in standard format (which is the same for wide or normal mode). It does not matter whether the Wide Range Math checkbox is selected. Use this for only small ranges, 40,000 b ranges or less since the memory requirements and sorting times are not optimized.

The Generate Report option takes a standard factors file and a proth.log file (verbose with date format selected) and generates a report with the factors and composite tests. It checks to make sure all b values are either in the factors file or the proth.log file. This routine is very dependent upon data format and is a quick and dirty method basically for the author only and not generally supported.

Credits

I would like to thank Yves Gallot for assistance and for an independent verification routine to help double check that factors identified are really factors.

Disclaimer

The software on this page is free to the general public for usage in their prime number searches. Each user must understand that they use this software at their own risk. If PRP, Proth, or Prime95 make your computer run too hot, so will this software. The only guarantee I can provide is that I personally use this software in my prime searches.

David Underbakke - g144 g205 g216 g294 g295 p53 p57 p101 x22 x35

david@underbakke.com