20081208, 21:57  #1 
Oct 2006
Berlin, Germany
268_{16} Posts 
gmpecm questions
Hello,
I have some questions regarding gmpecm. I downloaded it and compiled it on my Linux32 box according to the description. 1) Can somebody post an input number and a B1 which runs for some minutes (e.g. 30min) and than finds a factor? 2) What is the best way to compile it for the best performance, but it should run on any Linux32 system? 3) How about runtime? I expect, that runtime depends on input number and B1. I will use only "ecm [B1] < in.txt". But how is the relation, double B1 > double runtime? Any indication is very welcome. 4) If a factor with ecm is found, is there a way to verify it? yoyo 
20081208, 23:47  #2 
"Nancy"
Aug 2002
Alexandria
4643_{8} Posts 
1) Whether a factor is found or not depends not only on the B1 value, but also on the sigma value of the curve. One run that finds a factor is, for example
echo "(2^2048+1)/974849/319489"  ecm sigma 2929833171 50000 It takes only a few seconds. 2) Speed depends greatly on how GMP was compiled. If you need the same binary to run on all x86 cpus, you could try a "fat" GMP binary that includes code for different cpus and decides at runtime which code to use for the cpu it's running on. I've never tried building fat GMP binaries, though, so I can't give much advice here. 3) Run time is roughly linear in B1, yes. There's some overhead, but usually it's small compared to the actual curve arithmetic, and time for the latter grows very nearly like B1 (in stage 1 anyway, but stage 2 tries to choose parameters so that it takes no more time than stage 1 so total time still grows roughly linearly) 4) What do you mean? Verify that it divides the input number? If you're on a Unixlike system, you could use "bc" to compute the remainder. Or do you mean: test whether the factor (and/or cofactor) found is a prime? For numbers that are not too large, you could try stuff like Pari/GP (http://pari.math.ubordeaux.fr/) which has a primality proving available in the isprime() function. Alex Last fiddled with by akruppa on 20081211 at 14:47 Reason: typo 
20081209, 00:07  #3 
"Serge"
Mar 2008
Phi(4,2^7658614+1)/2
9436_{10} Posts 
GMP 4.3.0 benchmarks are shocking
I wonder how fast GMPECM will run with the 4.3.0.
The release schedule mentions late 2008 or early 2009. Is there anyone from the inner circles of GMP, who knows more precisely? Anyway, soon we will find out. <S> 
20081209, 10:26  #4  
Oct 2004
Austria
2×17×73 Posts 
Quote:
BTW: I just recognized that I am still running 4.2.2  is there a speedup between 4.2.2 and 4.2.4 for GMPECM? 

20081209, 15:59  #5  
Tribal Bullet
Oct 2004
3538_{10} Posts 
Quote:


20081209, 18:02  #6  
Oct 2006
Berlin, Germany
1150_{8} Posts 
Quote:
Later I need the same for Linux64, and Win 32/64. Has somebody thought about to port GMPECM to cuda and PS3 (with full usage of SPE'S)? yoyo 

20081209, 18:24  #7 
(loop (#_fork))
Feb 2006
Cambridge, England
2^{4}·3·7·19 Posts 
People have thought about running ECM on graphics cards and on PS3; indeed, I have a suspicion that some of the ECM records at EPFL were set on their PS3 cluster.
http://cado.gforge.inria.fr/workshop/slides/bos.pdf and http://cado.gforge.inria.fr/workshop/slides/lange.pdf will probably interest you. (linked from http://cado.gforge.inria.fr/workshop/abstracts.html) The graphicscard implementation is stage 1 only, and about the same speed on a GTX280 as all four processors of a Q6600. Last fiddled with by fivemack on 20081209 at 18:28 
20081209, 18:30  #8  
Nov 2008
2·3^{3}·43 Posts 
Quote:
appeared in my aliquot sequence (www.mersenneforum.org/showthread.php?t=11007, this may interest you). Run curves on this number (401 is normal, run more if you don't find factors) with B1 = 25e4. The factors should be: 423261437229815320090393538611 * 1908632439568485010120359575805448825018466413856309794908853 Last fiddled with by 10metreh on 20081209 at 18:34 

20081209, 20:28  #9 
Tribal Bullet
Oct 2004
2×29×61 Posts 
Just configure GMP with enablefat and then compile. I doubt you can do that with the windows development tools though (Brian Gladman maintains those projects). GMPECM has some SSE2 code that is selected by the configure script, but it does no runtime configuration.

20081209, 21:11  #10  
Oct 2006
Berlin, Germany
268_{16} Posts 
Quote:
yoyo 

20081210, 00:05  #11 
Tribal Bullet
Oct 2004
110111010010_{2} Posts 
Look for the documentation in the source package, in texinfo format. The file is gmp.info1
As far as I know the option to build a fat binary is limited to 32bit x86 only. 
Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Two questions:  Dubslow  GPU Computing  1  20110805 18:22 
Questions about the QS  Carmichael  Factoring  8  20070410 11:30 
Questions  OmbooHankvald  Prime Sierpinski Project  2  20050801 20:18 
LLR questions  OmbooHankvald  Math  6  20050623 11:42 
A few questions :)  xtreme2k  Lounge  59  20021031 06:20 