Weird segfault and "fix"?

[Xyzzy]

We have used the same binary for Msieve for a very long time. We think we compiled it with a Debian system a long time ago.

It has always worked perfectly.

It worked today, on a big job, on our Debian 10 Asus "NUC".

But today, trying that big job on our Fedora 33 system, it segfaulted.

$ uname -a 
Linux vi 5.9.15-200.fc33.x86_64 #1 SMP Wed Dec 16 19:14:35 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux

$ ./msieve -nc -p -t 1 -v 
 
Msieve v. 1.54 (SVN 1030) 
Wed Dec 23 18:03:00 2020 
random seeds: bd3abf07 7d0d8f6d 
factoring 146994713613685430462560633054030708553424988945059235528485114251973128261237846710634754720664533019191816640360086976564413805312241389512020336099540339919422287150716423619360302926533658867136193404152798094989961139809454759339483042766147 (246 digits) 
no P-1/P+1/ECM available, skipping 
commencing number field sieve (246-digit input) 
R0: 135637000670178176410246734932581577148578721 
R1: -511324276025564512546607 
A0: 201770283 
A1: 0 
A2: 0 
A3: 0 
A4: 0 
A5: 0 
A6: 1 
skew 24.22, size 9.679e-14, alpha 0.121, combined = 2.205e-14 rroots = 0 
 
commencing relation filtering 
estimated available RAM is 32080.8 MB 
commencing duplicate removal, pass 1 
Segmentation fault (core dumped)

Strangely, it would work for a simple factorization.

$ ./msieve -v 115367564564210182766242534110944507919869313713243756429 
 
Msieve v. 1.54 (SVN 1030) 
Wed Dec 23 18:06:25 2020 
random seeds: e6d43db7 503bfdc1 
factoring 115367564564210182766242534110944507919869313713243756429 (57 digits) 
p7 factor: 3387679 
p50 factor: 34055046113935288073705488067477617542827792631251 
elapsed time 00:00:00

So we recompiled it using the development tools on the Fedora system and now it works.

$ ./msieve -nc -p -t 1 -v 
 
Msieve v. 1.54 (SVN 1035M/VBITS=64/Xyzzy) 
Wed Dec 23 18:00:45 2020 
random seeds: 7a64fe17 6d8eca8e 
factoring 146994713613685430462560633054030708553424988945059235528485114251973128261237846710634754720664533019191816640360086976564413805312241389512020336099540339919422287150716423619360302926533658867136193404152798094989961139809454759339483042766147 (246 digits) 
no P-1/P+1/ECM available, skipping 
commencing number field sieve (246-digit input) 
R0: 135637000670178176410246734932581577148578721 
R1: -511324276025564512546607 
A0: 201770283 
A1: 0 
A2: 0 
A3: 0 
A4: 0 
A5: 0 
A6: 1 
skew 24.22, size 9.679e-14, alpha 0.121, combined = 2.205e-14 rroots = 0 
 
commencing relation filtering 
estimated available RAM is 32080.8 MB 
commencing duplicate removal, pass 1 
^C 
received signal 2; shutting down

Why?

Here is all we could find WRT errors:

[  161.774741] msieve[3252]: segfault at 7fd580928500 ip 000055a6fcfb4dbd sp 00007ffd946d2ee0 error 7 in msieve[55a6fcf9c000+a9000] 
[  161.774749] Code: 56 4c 8d b9 98 00 00 00 41 55 41 54 49 89 cd 55 53 48 89 fd 48 83 ec 08 48 85 f6 8b 19 0f 88 b2 00 00 00 48 8b 81 a0 00 00 00 <48> 89 30 0f 95 c0 0f b6 c0 89 81 9c 00 00 00 4d 8d a5 a8 00 00 00

FWIW, the msieve.dat file (42GB) is not compressed and all duplicates and bad relations have been removed from it.

[EdH]

It might be because there was some Msieve code that didn't play 100% with GMP-6.2.0. Jasonp has since fixed that, so the latest code should be fine. Which GMP is on which machine?

[Xyzzy]

Is this the right information?

Debian 10
libgmp-dev
libgmp-dev_6.1.2+dfsg-4_amd64.deb

Fedora 33
gmp-devel.x86_64
gmp-6.2.0-5.fc33.src.rpm

[EdH]

Originally Posted by Xyzzy

Is this the right information?

Debian 10
libgmp-dev
libgmp-dev_6.1.2+dfsg-4_amd64.deb

Fedora 33
gmp-devel.x86_64
gmp-6.2.0-5.fc33.src.rpm

I would think it was the new GMP-6.2.0 and it looks like you have a later checkout (1035) for Msieve now. You should be good.

If you want to see some details, you can look at (and around) this post in my YAFU "How I ..." thread

[Xyzzy]