Discussion related to cado-nfs

[Paul Zimmermann <Paul.Zimmermann@inria.fr>]

       Dear Hermann,

thank you, I am able to reproduce the issue with your files on machine gcc117
from the GCC compile farm, and I investigate in branch bug_characters_arm:

zimmerma@gcc117:~/cado-nfs$ ./build/gcc117/linalg/characters -poly 
/tmp/cado.4x9eajs3/c110.poly -purged /tmp/cado.4x9eajs3/c110.purged.gz -index 
/tmp/cado.4x9eajs3/c110.index.gz -heavyblock 
/tmp/cado.4x9eajs3/c110.dense.bin -out /tmp/cado.4x9eajs3/c110.kernel -ker 
/tmp/cado.4x9eajs3/c110.bwc/W -lpb0 25 -lpb1 26 -nchar 50 -t 1
...
Computing 64 characters for 1469231 (a,b) pairs
BUG: A[bs]==0 && B[bs]==0 i=524288 bs=8669 nrows=1469231
eval_64chars_batch_thread: a==0&&b==0 z=8669
enter eval_64chars with a=b=0
Error, Jacobi symbol is 0 for a = 0, b = 0, p = 67108879, r = 28784789
Please check lpb0/lpb1 are large enough
terminate called after throwing an instance of 'std::runtime_error'
  what():  code BUG() : condition res != 0 failed at 
/home/zimmerma/cado-nfs/linalg/characters.cpp:195

However it is not fully deterministic, sometimes it works:

done building big character matrix at wct=16.4s
done building small character matrix at wct=18.1s
done reading heavy block of size 292975 x 32 at wct=18.4s
/tmp/cado.4x9eajs3/c110.bwc/W: 64 vectors
Total: 64 kernel vectors
done reading 64 kernel vectors at wct=18.4s
Info: input kernel vectors reduced to dimension 61
done multiplying matrices at wct=18.5s
Computing left nullspace of 61 x 96 matrix
dim of ker = 23
Wrote 23 non-zero dependencies to /tmp/cado.4x9eajs3/c110.kernel
This includes 0 discarded zero dependencies, as well as 41 padding zeros
Total usage: time 17s (cpu), 19s (wct) ; memory 114MiB, peak 201MiB

It appears for some reason the values a=b=0 are given to eval_64chars,
these values come from the tables all_A[] and all_B[] in
big_character_matrix, which are initialized by filter_rels().

At this point I cannot debug further, I will pass on our local expert.

Best regards,
Paul

> Date: Mon, 17 Jul 2023 21:12:09 +0200
> From: hermann@stamm-wilbrandt.de
> Cc: cado-nfs@inria.fr
> 
> On 2023-07-16 07:30, Paul Zimmermann wrote:
> > What I wanted you to try is only run the characters command, after 
> > copying the
> > files computed on the Pi400.
> > 
> Hi Paul,
> 
> PC was rebooted, so /tmp/cado... was gone.
> 
> 
> I extracted the tar file and ran the characters command, completed 
> without issues in 4 seconds, see below.
> 
> 
> You can try yourself, I uploaded that (1.1GB) .tar.gz file:
> https://stamm-wilbrandt.de/en/forum/cado.4x9eajs3.tar.gz
> 
> You need to create the two links I described in this posting
> https://sympa.inria.fr/sympa/arc/cado-nfs/2023-07/msg00009.html
> to be able to run.
> 
> Here is sha256sum just to be sure you will get unmodified file:
> hermann@7600x:/tmp$ ssh web23.web138@stamm-wilbrandt.de
> web23.web138@stamm-wilbrandt.de's password:
> Last failed login: Mon Jul 17 21:09:33 CEST 2023 from 
> ip-078-042-191-222.um17.pools.vodafone-ip.de on ssh:notty
> There was 1 failed login attempt since the last successful login.
> -chrootsh-4.2$ sha256sum httpdocs/en/forum/cado.4x9eajs3.tar.gz
> 81f0b1e2320b908d49b856fd7b55c9262cd1c67c20b696c3eb8be8c5a14e3fb0  
> httpdocs/en/forum/cado.4x9eajs3.tar.gz
> -chrootsh-4.2$
> 
> 
> 
> The character command on same data succeeding on x86 Linux indicates, 
> that the cado-nfs on arm CPU has a problem.
> 
> 
> hermann@7600x:~$ cd /tmp
> hermann@7600x:/tmp$ ls -l cado*
> ls: cannot access 'cado*': No such file or directory
> hermann@7600x:/tmp$ tar -zxf ~/cado.4x9eajs3.tar.gz
> hermann@7600x:/tmp$ time 
> /home/pi/cado-nfs/build/pi400-64/linalg/characters -poly 
> /tmp/cado.4x9eajs3/c110.poly -purged /tmp/cado.4x9eajs3/c110.purged.gz 
> -index /tmp/cado.4x9eajs3/c110.index.gz -heavyblock 
> /tmp/cado.4x9eajs3/c110.dense.bin -out /tmp/cado.4x9eajs3/c110.kernel 
> -ker /tmp/cado.4x9eajs3/c110.bwc/W -lpb0 25 -lpb1 26 -nchar 50 -t 1 > 
> /tmp/cado.4x9eajs3/c110.characters.stdout.1 2> 
> /tmp/cado.4x9eajs3/c110.characters.stderr.1
> 
> real  0m4.248s
> user  0m4.535s
> sys   0m0.052s
> hermann@7600x:/tmp$ cat /tmp/cado.4x9eajs3/c110.characters.stderr.1
> /home/pi/cado-nfs/build/pi400-64/linalg/characters.re9cc3be15 -poly 
> /tmp/cado.4x9eajs3/c110.poly -purged /tmp/cado.4x9eajs3/c110.purged.gz 
> -index /tmp/cado.4x9eajs3/c110.index.gz -heavyblock 
> /tmp/cado.4x9eajs3/c110.dense.bin -out /tmp/cado.4x9eajs3/c110.kernel 
> -ker /tmp/cado.4x9eajs3/c110.bwc/W -lpb0 25 -lpb1 26 -nchar 50 -t 1
> # Warning: parameter nratchars is checked by this program but is 
> undocumented.
> Note: total 64 characters, including 14 trivial padding characters
> Reading 1469231 (a,b) pairs from /tmp/cado.4x9eajs3/c110.purged.gz
> Computing 64 characters for 1469231 (a,b) pairs
> done building big character matrix at wct=4.0s
> done building small character matrix at wct=4.2s
> done reading heavy block of size 292975 x 32 at wct=4.2s
> /tmp/cado.4x9eajs3/c110.bwc/W: 64 vectors
> Total: 64 kernel vectors
> done reading 64 kernel vectors at wct=4.2s
> Info: input kernel vectors reduced to dimension 61
> done multiplying matrices at wct=4.2s
> Computing left nullspace of 61 x 96 matrix
> dim of ker = 23
> Wrote 23 non-zero dependencies to /tmp/cado.4x9eajs3/c110.kernel
> This includes 0 discarded zero dependencies, as well as 41 padding zeros
> hermann@7600x:/tmp$
> 
> 
> Regards,
> 
> Hermann.
>