mersenneforum.org GMP-ECM crash
 Register FAQ Search Today's Posts Mark Forums Read

 2011-03-03, 23:18 #1 lavalamp     Oct 2007 Manchester, UK 17·79 Posts GMP-ECM crash Here's an unusual ECM crash, anyone know what's going on? Screenshot of error is attached. Specifically I'm using the ecm63mpircore2-64-asmredc binary provided by ATH. It ran fine for (4.7) days on many stage 1 results from P95 before giving up on one just like all the rest. Another instance of ECM that was started at the same time carried on just fine. Each instance tops out at about 4.3 GB of RAM usage, which is fine since I have 12 GB (although I'd like to use more). Should I just put it down to cosmic rays? A small note on the RAM use, I'm currently using as much RAM per curve as I can without ECM crashing. For some reason the binaries from ATH using MPIR let me use more RAM without crashing than Jeff Gilchrist's using GMP. Though I don't know if this is down to version 6.2.3 vs 6.3 or GMP vs MPIR. Another curiosity is that I can't use Jeff's v6.3 binaries for these numbers because it consistently and repeatedly crashes. Attached Thumbnails
2011-03-04, 10:11   #2

May 2008
Worcester, United Kingdom

10000011102 Posts

Quote:
 Originally Posted by lavalamp Here's an unusual ECM crash, anyone know what's going on? Screenshot of error is attached. Specifically I'm using the ecm63mpircore2-64-asmredc binary provided by ATH. It ran fine for (4.7) days on many stage 1 results from P95 before giving up on one just like all the rest. Another instance of ECM that was started at the same time carried on just fine. Each instance tops out at about 4.3 GB of RAM usage, which is fine since I have 12 GB (although I'd like to use more). Should I just put it down to cosmic rays? A small note on the RAM use, I'm currently using as much RAM per curve as I can without ECM crashing. For some reason the binaries from ATH using MPIR let me use more RAM without crashing than Jeff Gilchrist's using GMP. Though I don't know if this is down to version 6.2.3 vs 6.3 or GMP vs MPIR. Another curiosity is that I can't use Jeff's v6.3 binaries for these numbers because it consistently and repeatedly crashes.
If you have a set of parameters that give a consistent, repeatable crash, I can take a look at it for the x64 Visual Studio builds using MPIR.

Last fiddled with by Brian Gladman on 2011-03-04 at 10:12

 2011-03-04, 23:17 #3 lavalamp     Oct 2007 Manchester, UK 134310 Posts I do not, hence the cosmic ray theory. It is, so far anyway, a one time event.
2011-03-05, 03:18   #4
ATH
Einyen

Dec 2003
Denmark

2·3·521 Posts

Quote:
 Originally Posted by lavalamp I do not, hence the cosmic ray theory. It is, so far anyway, a one time event.
I think he meant consistent repeatable crash with Jeff Gilchrist's Visual Studio build ?

Last fiddled with by ATH on 2011-03-05 at 03:19

 2011-03-05, 07:48 #5 Brian Gladman     May 2008 Worcester, United Kingdom 20E16 Posts Any consistent repeatable crash on windows x64 applications using MPIR built with Visual Studio would be worth reporting as I can look at these provided the crash is quickly and easily invoked. For mingw64 builds I probably cannot do much and for GMP based builds nothing at all.
 2011-03-05, 15:01 #6 lavalamp     Oct 2007 Manchester, UK 17×79 Posts ECM does crash if I try to use -k 2 or -k 1, however ATH's binaries were compiled with mingw, not visual studio.
2011-03-05, 17:42   #7
ATH
Einyen

Dec 2003
Denmark

2×3×521 Posts

When I tested speed of Jeff's and my own binary, Jeff's binary crashed in stage2 for numbers above 550 digits.

I use the file "ecm63_win64_core2.zip" from http://gilchrist.ca/jeff/factoring/

I attach a random 600 digit prime I used for speed testing. Ecm crashes with the line: ecm -v 11000000 < 600.txt
Code:
Input number is 8614110783543076121203557864541451475181738140565460140525625406
17672358376042484422011860583124636832511472678571543257381587434765681015776543
33156770671517657847576067258140603313117285353800244201077038607241042767167315
14317643306576014064160281702626518333805205410611222102684018566224812041335033
60231878770646074401401873414133045202765578030814551842776170633787176860533737
17004243526837320128843638133045328864815231888836456721485762882412441457475718
85553127022284648282565867544848504450516488703313021810413614370446650416862100
64387716570260056372555036864857356440482615218335691927 (600 digits)
Using MODMULN
Using B1=11000000, B2=30114149530, polynomial Dickson(12), sigma=2957883585
dF=36864, k=2, d=371280, d2=11, i0=19
Expected number of curves to find a factor of n digits:
35      40      45      50      55      60      65      70      75      80
124     702     4630    34845   292384  2737745 2.8e+007        3.1e+008
3.9e+009        4.6e+010
Step 1 took 507518ms
Estimated memory usage: 516M
Initializing tables of differences for F took 452ms
Computing roots of F took 11997ms
Attached Files
 600.txt (602 Bytes, 188 views)

2011-03-05, 18:15   #8

May 2008
Worcester, United Kingdom

52610 Posts

Quote:
 Originally Posted by ATH When I tested speed of Jeff's and my own binary, Jeff's binary crashed in stage2 for numbers above 550 digits. I use the file "ecm63_win64_core2.zip" from http://gilchrist.ca/jeff/factoring/ I attach a random 600 digit prime I used for speed testing. Ecm crashes with the line: ecm -v 11000000 < 600.txt Code: Input number is 8614110783543076121203557864541451475181738140565460140525625406 17672358376042484422011860583124636832511472678571543257381587434765681015776543 33156770671517657847576067258140603313117285353800244201077038607241042767167315 14317643306576014064160281702626518333805205410611222102684018566224812041335033 60231878770646074401401873414133045202765578030814551842776170633787176860533737 17004243526837320128843638133045328864815231888836456721485762882412441457475718 85553127022284648282565867544848504450516488703313021810413614370446650416862100 64387716570260056372555036864857356440482615218335691927 (600 digits) Using MODMULN Using B1=11000000, B2=30114149530, polynomial Dickson(12), sigma=2957883585 dF=36864, k=2, d=371280, d2=11, i0=19 Expected number of curves to find a factor of n digits: 35 40 45 50 55 60 65 70 75 80 124 702 4630 34845 292384 2737745 2.8e+007 3.1e+008 3.9e+009 4.6e+010 Step 1 took 507518ms Estimated memory usage: 516M Initializing tables of differences for F took 452ms Computing roots of F took 11997ms
===========================
I have just run this and got:

GMP-ECM 6.3 [configured with MPIR 2.3.0, --enable-asm-redc] [ECM]
Input number is 86141107835430761212035578645414514751817381405654601405256254061767235837604248442201186058312463683251147267857154
325738158743476568101577654333156770671517657847576067258140603313117285353800244201077038607241042767167315143176433065760140641602
817026265183338052054106112221026840185662248120413350336023187877064607440140187341413304520276557803081455184277617063378717686053
373717004243526837320128843638133045328864815231888836456721485762882412441457475718855531270222846482825658675448485044505164887033
1302181041361437044665041686210064387716570260056372555036864857356440482615218335691927 (600 digits)
Using MODMULN
Using B1=11000000, B2=30114149530, polynomial Dickson(12), sigma=21267629
dF=36864, k=2, d=371280, d2=11, i0=19
Expected number of curves to find a factor of n digits:
35 40 45 50 55 60 65 70 75 80
124 702 4630 34845 292384 2737745 2.8e+007 3.1e+008 3.9e+009 4.6e+010
Step 1 took 521495ms
Estimated memory usage: 516M
Initializing tables of differences for F took 453ms
Computing roots of F took 12808ms
Building F from its roots took 9454ms
Computing 1/F took 5179ms
Initializing table of differences for G took 359ms
Computing roots of G took 10623ms
Building G from its roots took 9360ms
Computing roots of G took 10639ms
Building G from its roots took 9376ms
Computing G * H took 2948ms
Reducing G * H mod F took 4478ms
Computing polyeval(F,G) took 23072ms
Computing product of all F(g_i) took 140ms
Step 2 took 99045ms
Expected time to find a factor of n digits:
35 40 45 50 55 60 65 70 75 80
21.43h 5.05d 33.26d 250.26d 5.75y 53.87y 554.11y 6031y 76148y 902836y

 2011-03-05, 18:53 #9 henryzz Just call me Henry     "David" Sep 2007 Cambridge (GMT/BST) 2×5×587 Posts ATH isn't the only one getting a crash on that number with Jeff's binaries. Just got crash here as well.
 2011-03-05, 19:13 #10 Brian Gladman     May 2008 Worcester, United Kingdom 10000011102 Posts I'll have to leave Jeff to sort that out since my version doesn't crash. Last fiddled with by Brian Gladman on 2011-03-05 at 19:15
 2011-03-06, 01:03 #11 lavalamp     Oct 2007 Manchester, UK 17·79 Posts With regards to the crashing with high RAM use, I'm running stage 2 on a 1000 digit number with a large B2 value. Jeff's 6.2.3 GMP binary crashes with high RAM use and ATH's 6.3 MPIR binary crashes with much higher RAM use. This is repeatable and happens every time. I say it's due to high RAM use, but it could just be that it is coincident with the RAM use spiking, it always crashes just as ECM is gobbling up RAM during the "Computing polyeval(F,G)" step. The inputs come from Prime95 output and look like this: Code: N=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he command line I use for ECM with ATH's binaries is: Code: ecm -resume "results.txt" -k 3 -treefile ecm -nn -timestamp 1 260000000-3000000000000 > "save.txt" If I change that 3 to a 2 or 1, it will crash every single time. Currently the RAM use is ~4.1 GB (though the estimated usage by ECM is ~4.6 GB), if I lower the value for k that increases significantly, but I have the additional capacity to handle it, so that is not the problem. Three months ago I had a look at this problem in more detail to see the exact parameters that caused ECM to crash (when I was still using Jeff's ECM-GMP binaries that crashed with lower RAM use): http://www.mersenneforum.org/showpos...72&postcount=7 It seems I made a mistake in the first post, I didn't realise that Jeff used MPIR in his binaries instead of GMP, unfortunately the crash happens after a fair amount of time (though it always happens at the same point).

 Similar Threads Thread Thread Starter Forum Replies Last Post yoyo GMP-ECM 26 2011-06-01 06:31 storm5510 Software 8 2009-08-31 02:07 yoyo GMP-ECM 2 2009-01-26 19:22 remaker Software 4 2007-05-03 16:39 coastertux Software 2 2004-03-16 20:49

All times are UTC. The time now is 11:28.

Sun May 9 11:28:35 UTC 2021 up 31 days, 6:09, 0 users, load averages: 2.66, 2.70, 2.62