Post-Processing Fails at Cycle Optimization

wombatman · 2013-10-11, 21:19

Code:

Fri Oct 11 16:15:26 2013  weight of 4401231 cycles is about 308463782 (70.09/cycle)
Fri Oct 11 16:15:26 2013  distribution of cycle lengths:
Fri Oct 11 16:15:26 2013  1 relations: 707738
Fri Oct 11 16:15:26 2013  2 relations: 587582
Fri Oct 11 16:15:26 2013  3 relations: 515398
Fri Oct 11 16:15:26 2013  4 relations: 434760
Fri Oct 11 16:15:26 2013  5 relations: 371133
Fri Oct 11 16:15:26 2013  6 relations: 309905
Fri Oct 11 16:15:26 2013  7 relations: 263835
Fri Oct 11 16:15:26 2013  8 relations: 222328
Fri Oct 11 16:15:26 2013  9 relations: 185425
Fri Oct 11 16:15:26 2013  10+ relations: 803127
Fri Oct 11 16:15:26 2013  heaviest cycle: 28 relations
Fri Oct 11 16:15:27 2013  commencing cycle optimization
failed to reallocate 163840000 bytes
Return value 0. Terminating...

Looking around for this error suggested that memory might be the case, but I watched the memory used and there was still nearly 4GB of RAM free. There are also a few hundred GB of disk space, so it's not that either.

This is using MSieve 1.52 SVN945 on Windows 7 64-bit. Any ideas?

Edit: The number is the cofactor of 11^238+5^238 (C149, SNFS difficulty = 213). I've done a C142 myself, so unless I was right at the edge of something, I don't know why this would fail.
Edit 2: Using the 1.51 official version from Jeff Gilchrist's site (the 64 bit i7/i5 one) works.

jasonp · 2013-10-12, 02:01

No idea why v1.52 wouldn't work; the changes in the latest SVN should have nothing to do with the filtering. I wonder if your current binary has some kind of 32-bit limit even if your system does not...

wombatman · 2013-10-12, 03:00

Yeah, it's a 32-bit file. However, I tried the 64-bit versions from Jeff's page (SVN 942 and SVN 939). Both of them throw errors on the actual linear algebra:

Code:

lanczos error: submatrix is not invertiblemensions (0.0%, ETA 14h39m)

lanczos halted after 5 iterations (dim = 250)
linear algebra failed; retrying...

I'm going to try rebuilding a 64-bit version from the latest SVN.

Edit: And now I'm reminded why I didn't have a 64-bit version yet...errors at the literal last step of compiling about not being able to find correct calls to a couple of GMP-based functions. In the meantime, using the 1.51 version that works.

wombatman · 2013-10-12, 04:51

Well, some fiddling around and not including ECM=1 in the build, and I have a successfully working V1.52 SVN945 64-bit version of Msieve. I guess I screwed up the previous build. It's lovely to watch the expected completion time drop from 34 hours to 13.5 hours.

Now to try and get a version with ECM built into it...

2013-10-12, 03:00	#3
wombatman I moo ablest echo power! May 2013 1,741 Posts	Yeah, it's a 32-bit file. However, I tried the 64-bit versions from Jeff's page (SVN 942 and SVN 939). Both of them throw errors on the actual linear algebra: Code: lanczos error: submatrix is not invertiblemensions (0.0%, ETA 14h39m) lanczos halted after 5 iterations (dim = 250) linear algebra failed; retrying... I'm going to try rebuilding a 64-bit version from the latest SVN. Edit: And now I'm reminded why I didn't have a 64-bit version yet...errors at the literal last step of compiling about not being able to find correct calls to a couple of GMP-based functions. In the meantime, using the 1.51 version that works. Last fiddled with by wombatman on 2013-10-12 at 03:25

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
16e Post Processing Progress	pinhodecarlos	NFS@Home	8	2018-11-28 13:45
NFS@Home Post-Processing Rack Build	pinhodecarlos	NFS@Home	1	2016-09-27 12:34
Crash doing large post-processing job	wombatman	Msieve	22	2013-12-04 01:37
Update on 7^254+1 post processing	dleclair	NFSNET Discussion	4	2005-04-05 09:51
Post processing for 2,757-	xilman	NFSNET Discussion	3	2003-11-06 14:23

2013-10-12, 02:01	#2
jasonp Tribal Bullet Oct 2004 2×3×19×31 Posts	No idea why v1.52 wouldn't work; the changes in the latest SVN should have nothing to do with the filtering. I wonder if your current binary has some kind of 32-bit limit even if your system does not...

2013-10-12, 04:51	#4
wombatman I moo ablest echo power! May 2013 1,741 Posts	Well, some fiddling around and not including ECM=1 in the build, and I have a successfully working V1.52 SVN945 64-bit version of Msieve. I guess I screwed up the previous build. It's lovely to watch the expected completion time drop from 34 hours to 13.5 hours. Now to try and get a version with ECM built into it...