20130806, 17:48  #23 
"Dana Jacobsen"
Feb 2011
Bangkok, TH
38A_{16} Posts 
This is more a mathofECPP question, but to the best of my knowledge, no. We can combine the proofs, but only really in the sense that for any proof that reads "N is prime if Q [and Q2 and Q3...] are prime" we can use ECPP as an option for proving the Q(s) prime. ECPP doesn't operate on n1 or n+1. My program, like many others, does a quick n1 and n+1 proof attempt (BLS75 theorems 3 and 15) for each Q, but it is done purely as a time saver, and if we don't use the n1 or n+1 proof, we don't use the factors found. We use ECPP to give us a list of other numbers we try to partially factor (it's a little different than the BLS75, Pocklington, etc. style proofs in that we're trying to get a single large prime factor, rather than a product of primes).
Last fiddled with by danaj on 20130806 at 17:50 Reason: clarify that factors found in n1/n+1 are used only if those proofs are used 
20130810, 21:00  #24 
Oct 2006
Berlin, Germany
3×197 Posts 
Hello,
I did some tests with the 1.01 verions an have some questions: 1) Should it be noticeable faster if I compile and run it on a 64 bit system instead a 32 bit system? Currently I run it on e 32 bit system. 2) How to generate a prime certificate out of it? yoyo 
20130810, 22:29  #25  
"Dana Jacobsen"
Feb 2011
Bangkok, TH
1110001010_{2} Posts 
Quote:
Here's something I just tried. It's a 64bit Windows 7 machine (Intel 3770K) but the compilers I have are using 32bit for unsigned long. I compiled with Cygwin and under DOS using MinGW (whatever is included with the Strawberry Perl distribution). Cygwin comes out about 34x slower than the MinGW compiled code, for the same code on the same machine. This looks pretty repeatable. For a 500digit number I just ran, the MinGWcompiled code finished the entire proof by the time the Cygwincompiled version got to the second Q value. So definitely try some different systems out to see which works best. That isn't a pure 32bit machine, so in theory GMP would be smart enough to do the right thing in most cases regardless of the compiler's choice. Still, clearly the result is very different. The one machine I do have which is pure 32bit is an Intel Atom, and it's just sluggish on everything, so I'm not sure how much I'd learn. Another thing I could do is spend a little more time working on the GMPECM integration. It's a bit slower than my p1 on x86_64 machines for the typical depths I'm using, but I imagine it's well tuned on a larger range of machines and it'll be far better once we need deeper factoring. If I were making a distributed system that was going after quite large numbers, I'd probably just go directly to GMPECM as I'd be more interested in getting its factoring prowess on the larger numbers than about it taking 2s vs 1s on the small numbers. Quote:
Code:
Usage: standalone/ecppdj [options] <number> Options: v set verbose V set extra verbose c print certificate nm1 use n1 proof only aks use AKS for proof help this message Use c to get a certificate (from the default ECPP or nm1, AKS and APRCL don't give certificates). The tar file includes a verification program written in Perl, though it requires some additional CPAN modules. It also works for Primo certificates. Alternately you can get my pure GMP verifier here: vcert.c which also does both Primo and MPU certificates. All the development is done on github, and you can bring up issues there also. 

20131223, 22:14  #26 
Oct 2006
Berlin, Germany
3×197 Posts 
The smallest PRP in factorDB has currently 2357 digits. How long will it need to calculate a prime certificate for it with ecppdj? Does this make sense?
yoyo 
20131223, 23:01  #27 
"Dana Jacobsen"
Feb 2011
Bangkok, TH
1110001010_{2} Posts 
It does, but I'll first say that I discussed the Primo proof for that number with Marcel Martin and got a reply. He is going to update his verifier, and I'll update mine and send to Syd so we can handle the proof recent versions of Primo generate for it.

20131224, 12:50  #28 
Mar 2006
111011000_{2} Posts 
I saw your email to Marcel, but I never saw a reply to from him. Could you let me know what was discussed so I can make sure my verifier will work with recent versions of Primo?

20131224, 18:11  #29 
"Dana Jacobsen"
Feb 2011
Bangkok, TH
2·3·151 Posts 
I forwarded it to you. S=1 is allowed if all other conditions are met. His email has more details.

20140821, 00:53  #30 
"Dana Jacobsen"
Feb 2011
Bangkok, TH
2·3·151 Posts 
Link to latest version: ecppdj.tar.gz
Permission to edit the first post would be nice to allow it to show the latest information. Version 1.04 has been released. 1.00 June 2013 1.01 July 2013 1.02 <not released> 1.03 April 2014 1.04 August 2014 1.03 added performance enhancements to ECPP. 1.04 has no ECPP changes, but includes spelling changes and an expression parser (thanks to Mathew for corrections and suggestions). Let me know any suggestions or comments. Latest performance chart [i4770K, 4.3GHz, DDR32133, GMP 6.0.0a, gcc 4.8.2, Fedora; all single core; average time; 100 random primes to 150 digits, 5 random primes to 2000 digits; GMP versions even for 1020 digits; ecppdj run with large 28M poly set] 
20140821, 03:46  #31 
Dec 2012
100010110_{2} Posts 
Thanks for the update! Performance looks nice under 500 digits. I'll see if I can make it and try it out a bit.

20140821, 05:08  #32 
"Dana Jacobsen"
Feb 2011
Bangkok, TH
1612_{8} Posts 
I find it very useful for my own work, but mpz_aprcl, Pari, and Primo are other good choices. Also note the graph is loglog, so at 500 digits these are all in the 1015 second range. The graph does a nice job of showing the differences, just keep in mind the scale  the actual time difference on the low end is quite small, and on the high end what looks like a small difference can be the difference between 1 hour and 3 hours. Primo also comes with multithreading out of the box.

20141020, 05:31  #33 
I moo ablest echo power!
May 2013
5×347 Posts 
Apologies for the dumb question, but does ecppdj produce a certificate that can be uploaded directly to FactorDB? I generated a certificate for this number (http://factorization.ath.cx/index.ph...00000718032814) and verified it with vcert.exe, but when I tried to upload it, FactorDB said it couldn't find the input number. I've attached the cert file.

Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
New ECPP record  mjm  Computer Science & Computational Number Theory  33  20200213 14:50 
ECPP on Windows?  CRGreathouse  Software  10  20150914 12:32 
Can I just leave this here? (ECPP)  trhabib  Miscellaneous Math  6  20110819 16:34 
Looking for ECPP software  nuggetprime  Software  14  20100307 17:09 
new ECPP article  R. Gerbicz  GMPECM  2  20060913 16:24 