GMP-ECM 6.2.3 exit with 0xc0000005

[yoyo]

Hello,
some of my Boinc useres have crashing gmp-ecm on Windows 32 bit systems. This mostly happens on systems with less memory e.g. 3GB and high ecm predicted memory, e.g. 2GB.
It happend already with GMP-ECM 6.2.1.

Here is an example result:
http://www.rechenkraft.net/yoyo/resu...sultid=3052022
on this host:
http://www.rechenkraft.net/yoyo/show...p?hostid=12108
The predicted RAM was 2GB.
Exactly the same happend here:
http://www.rechenkraft.net/yoyo/resu...sultid=3030183
host:
http://www.rechenkraft.net/yoyo/show...p?hostid=27840
Both crashed after 1,7h of runtime, I think in phase2.

The user posted also a stack where it crashed:
http://www.rechenkraft.net/phpBB/vie...105109#p105109
maybe it helps.
yoyo

[Jeff Gilchrist]

Which binary are you using a MingW32 or the MSVC (my) binary?

Jeff.

[yoyo]

Your binary.

[Jeff Gilchrist]

Unfortunately I don't have a Windows 32bit OS with enough memory to actually do any tests on. Hopefully Alex can take a look at that stack dump and maybe see something that points to the problem or recognize the error code.

It is hitting both a Core2 and AMD system so doesn't seem to be an architecture specific bug.

Is that repeatable for them? Maybe you can get them to try one of Yamato's Mingw32 win32 binaries (with muldrec asm):
http://www.kay-schoenberger.de/eng/m...m623_win32.zip
or (no muldrec asm)
http://www.kay-schoenberger.de/eng/m...23_win32_2.zip

Jeff.

[10metreh]

Doesn't Win32 only allow 2GB per process?

[akruppa]

The stack trace does not tell much... looks like it tried to dereference the value in eax which happened to be 0. This might be an out-of-memory situation in the Schönhage-Strassen code. Is the output from the aborted run available, preferably with -v? How much memory does GMP-ECM think it'll need, and how much is actually getting allocated? If it needs more than 2GB, does the machine allow more than 2GB VM size per process?

Alex

[yoyo]

If 2GB is a limit on 32bit windows (also on 32 bit Linux), is it than possible to run it with -maxmem 2000000?
yoyo

[akruppa]

I'm not sure how accurate the memory estimation is under Windows, better try first. If the estimates aren't too far off, then you could use -maxmem 2000 (the unit is megabytes!) or maybe 1800 or so to leave a safety margin.

Alex

[yoyo]

This is the output before the crash:

GMP-ECM 6.2.3 [powered by GMP 4.2.1_MPIR_1.1.1] [ECM]
Input number is 9498541323...7059673777 (1210 digits)
[Tue Apr 28 18:31:26 2009]
Using special division for factor of 2^4253+1
Using B1=43000000, B2=198654756318, polynomial Dickson(12), sigma=4258974300
dF=92160, k=2, d=1021020, d2=19, i0=24
Expected number of curves to find a factor of n digits:
20   25   30   35   40   45   50   55   60   65
2   5   15   56   253   1323   7837   51385   373591   2960912
Step 1 took 8725000ms
Estimated memory usage: 2494M
Initializing tables of differences for F took 2312ms
Computing roots of F took 191547ms

yoyo

[Andi_HB]

To increase the Limit from 32bit Windows you can try this in the comand prompt as admin. But thats without guarantee!
This German link http://www.pc-experience.de/wbb2/thr...threadid=24883 have more Informations about this (only German).

bcdedit /set IncreaseUserVA 3072

[yoyo]

What would be the better option for win32 if I think it will require more than 2GB:
-k 8
or
-maxmem 1800 ?
yoyo