mersenneforum.org Let's do another nice big GNFS job!
 Register FAQ Search Today's Posts Mark Forums Read

 2010-11-30, 13:58 #34 jasonp Tribal Bullet     Oct 2004 32×5×79 Posts The 64-bit windows binary would require converting all the 64-bit asm that makes the linux binary so fast, and right now only Brian Gladman has the expertise to do that. You may get it to compile on windows, but it won't have all that asm.
2010-11-30, 18:12   #35
Andi47

Oct 2004
Austria

2×17×73 Posts

Quote:
 Originally Posted by Batalov Just to clarify: "mpqs failed" or "mpqs on a prime square" are not critical errors per se: MPQS implementation in the siever is expected only to work to certain limits (which we are pushing more and more) and occasionally fail to factor (but not crash). We of course don't want the siever to hang or crash. If removing the block that starts with mpz_sqrtrem call clears the crash on Windows, then there are some mpz_sqrtrem side-effects that I cannot account for. If the recompiled binary will still crash, then there may be some damage done in mpqs routine after it fails. The three debug cases should be sufficient to examine that. Yes, the sub-topic of crashes could be moved to the Tweaking and compiling the Kleinjung siever and tracked there.
Is it possible that the occasional crashes of the 11e and 12e sievers under windows (I have posted a few of them into the "running GGNFS" thread, IIRC) have the same cause? (one example is this posting, there should be more in this thread)

edit: another crash at q=43274929

Last fiddled with by Andi47 on 2010-11-30 at 18:50

2010-11-30, 19:59   #36
jrk

May 2008

109510 Posts

Quote:
 Originally Posted by Batalov Just to clarify: "mpqs failed" or "mpqs on a prime square" are not critical errors per se: MPQS implementation in the siever is expected only to work to certain limits (which we are pushing more and more) and occasionally fail to factor (but not crash). We of course don't want the siever to hang or crash. If removing the block that starts with mpz_sqrtrem call clears the crash on Windows, then there are some mpz_sqrtrem side-effects that I cannot account for. If the recompiled binary will still crash, then there may be some damage done in mpqs routine after it fails. The three debug cases should be sufficient to examine that. Yes, the sub-topic of crashes could be moved to the Tweaking and compiling the Kleinjung siever and tracked there.
If the "mpqs failed" type of errors is what's causing the crashes on Windows, in the meantime (until the real problem gets fixed in code), the mpqs errors could be reduced by using the more conventional 2LP parameters*.

*OTTOMH, though I don't guarantee these:
Code:
alim: 300000000
rlim: 300000000
lpba: 32
lpbr: 31
mfba: 64
mfbr: 62
alambda: 2.6
rlambda: 2.6

2010-12-02, 11:12   #37
ET_
Banned

"Luigi"
Aug 2002
Team Italia

3·1,619 Posts

Quote:
 Originally Posted by Andi47 what is the name of your outputfile? add -o to the command line.
It worked smoothly, thank you Andi.

Luigi

 2010-12-06, 19:41 #38 bsquared     "Ben" Feb 2007 22·3·311 Posts I see you're requesting .gz files. Is that a general request that the relations be zipped, or can you only deal with gzip specifically? bzip2 produces files about 10-15% smaller than tar -z does.
2010-12-06, 20:18   #39
Andi47

Oct 2004
Austria

1001101100102 Posts

Quote:
 Originally Posted by bsquared I see you're requesting .gz files. Is that a general request that the relations be zipped, or can you only deal with gzip specifically? bzip2 produces files about 10-15% smaller than tar -z does.
I guess bzip2 should be OK - at least fivemack never complained when I uploaded bzip2'd relations in his previous projects

 2010-12-06, 20:36 #40 Batalov     "Serge" Mar 2008 Phi(4,2^7658614+1)/2 2×11×457 Posts Though in an ideal world (all files being error-free) one could simply concat all gz files and get one nice test.dat.gz file, it will fail as soon as the first transfer error in any of the files. In Tom's shoes, I'd gzip -tv and bzip2 -tv all incoming files anyway, so with this in mind the format wouldn't matter as much to me as the file transfer time. But this is for Tom to decide.
 2010-12-07, 01:02 #41 fivemack (loop (#_fork))     Feb 2006 Cambridge, England 2·7·461 Posts I don't mind what format you use; I keep the relations in gzip rather than bzip form on disc because I find it takes a lot longer to bunzip than to gunzip if I want to do a quick filtering test, but if you upload bz2 then I'll convert them to gzip.
 2010-12-07, 02:25 #42 Batalov     "Serge" Mar 2008 Phi(4,2^7658614+1)/2 235068 Posts (...wearing a bioinformatic hat...) For the same reason all fancy 7z methods, albeit packing better and better, are failing the same use case: they are slower and yet slower -- and more and more memory-consuming. There's no free lunch anywhere, including packing - it is a clear tradeoff. Interestingly, if you allow a small tangent: B-W transform recently provided a huge boost to the bioinformatical problem of mapping of billions of short substrings on a larger string with a few errors (a.k.a. genome mapping). There are a dozen programs for this job -- all written within the last two years (because of the new generation of sequencers) and all using BWT, for a representative example see BWA. But this uses the indexing property of B-W with the Ferragina-Manzini clever algorithm. I mean, why is this even interesting? It just goes to show that some algorithms lie dormant for decades because their use is unclear (Wheeler found the transformation in 1983 but it was only in the 90s that Burrows convinced him that it was useful, and they published it; later the .bz2 emerged, and now... (25 years after!) it became used exponentially more.) Last fiddled with by Batalov on 2010-12-07 at 02:33
 2010-12-15, 20:19 #43 Andi47     Oct 2004 Austria 2·17·73 Posts this one got the 16e siever stuck into an endless loop: gnfs-lasieve4I16e -v -M 1 -a 2_956+.poly -o 2_956+_43.44d07.out -f 43953600 -c 46400 Warning: lowering FB_bound to 43953599. FBsize 2658728+0 (deg 5), 8444395+0 (deg 1)
2010-12-19, 11:25   #44
Andi47

Oct 2004
Austria

248210 Posts

Quote:
 Originally Posted by Andi47 reserving 43M-44M
done, 3014307 relations. http://www.sendspace.com/file/8kdmun

Quote:
 Originally Posted by Andi47 Edit: another reproducible crash at q=43515611. Is there a binary for windows 64-bit available, which is newer than svn-374?
I had to restart 51 times due to crashes (lost quite some time, i.e. when I came to the PC and found the error messages on the screen and the threads sitting idle since hours).

After looking into the windows task manager, I noticed that I in fact use a 32-bit binary (svn-374, newest available on Jeff's homepage, hence I never managed to compile the binaries from source). Is there any newer windows binary available? (*hoping for bugfixes*)

 Similar Threads Thread Thread Starter Forum Replies Last Post Prime Monster Lounge 29 2023-01-02 21:46 schickel FactorDB 29 2012-07-18 17:03 Dubslow Forum Feedback 0 2012-05-02 02:13 Raman Puzzles 2 2009-11-01 10:23 Xyzzy Lounge 4 2003-06-28 13:37

All times are UTC. The time now is 01:41.

Thu Feb 2 01:41:00 UTC 2023 up 167 days, 23:09, 1 user, load averages: 0.90, 1.02, 1.06