Msieve 1.42 feedback

[jrk]

Originally Posted by jasonp

Grr, still no crash when run locally. Also no crash on my 64-bit system. What platform and compiler version are you using?

Linux x86_64 GCC 4.3.2

[henryzz]

would it make any difference how much memory the pc had for this crash of jrk's?

[jasonp]

Still can't get it to crash. Try the following:

make clean
make x86_64 GMP=1 OPT_FLAGS=-g
gdb msieve

Within gdb, try the following:

b common/filter/singleton.c:305
cond 1 counts[ideal_list[k]] > 10000000
run -v -nc -mb 20 <your_number>

and see if it completes normally. If not, please PM me the output from 'bt full'.
(Apologies if you know how to use gdb already)

Edit: never mind, valgrind caught something. Do you still get a crash if you take "filter->lp_file_size = get_file_size(buf2);" on line 321 of common/filter/singleton.c, move it to the bottom of the function, and replace 'buf2' with 'buf' ? This gets the size of a file for later code to use, and apparently there is still data buffered to be written to the file that is not counted until the file is closed, so later code reads off the end of a too-small array.

[jrk]

Originally Posted by jasonp

Edit: never mind, valgrind caught something. Do you still get a crash if you take "filter->lp_file_size = get_file_size(buf2);" on line 321 of common/filter/singleton.c, move it to the bottom of the function, and replace 'buf2' with 'buf' ? This gets the size of a file for later code to use, and apparently there is still data buffered to be written to the file that is not counted until the file is closed, so later code reads off the end of a too-small array.

I tried that and it doesn't crash.

[frmky]

Attached is the log for 7,319+. The factorization was completed with version 1.41, but I reran the filtering with 1.42b2 (plus the file size change) specifying 8 GB of memory. 1.42 produced a somewhat smaller and lighter matrix, and took 2 fewer hours to do so.

[Andi47]

when typing "msieve -h" to list all flags, the output line for the -mb flag misses a ")".


Edit: Are the parameters for poly search for larger inputs (say c140+) the same as in beta 1?

[10metreh]

Originally Posted by Andi47

when typing "msieve -h" to list all flags, the output line for the -mb flag misses a ")".

That's not a "very minor bug", it's just a "very minor typo"

[Andi47]

Originally Posted by 10metreh

That's not a "very minor bug", it's just a "very minor typo"

corrected in the post above.

[jasonp]

Originally Posted by frmky

Attached is the log for 7,319+. The factorization was completed with version 1.41, but I reran the filtering with 1.42b2 (plus the file size change) specifying 8 GB of memory. 1.42 produced a somewhat smaller and lighter matrix, and took 2 fewer hours to do so.

Thanks. The singleton removal is 4x faster than before but the clique processing still takes four hours. I have to do something about that. Also, when reading everything from disk in one pass the code still pulls in loads of ideals that the merge phase really cannot do anything about (i.e. they occur too often).

valgrind also found a nasty memory leak in the posted source, as well as a few other minor problems, which I've fixed. Beta 3 will be out later today.

[Shaopu Lin]

When I factored rsa100 by msieve 1.42 beta 3, it crashed due to segmentation fault.

math@linux-0rdr:~/Desktop/application/nfs/ggnfs/gnfs> gdb msieve
GNU gdb (GDB; openSUSE 11.1) 6.8.50.20081120-cvs
Copyright (C) 2008 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "x86_64-suse-linux".
For bug reporting instructions, please see:
<http://bugs.opensuse.org/>...
(gdb) run  -s gnfs.dat -l ggnfs.log -i gnfs.ini -v -nf gnfs.fb -t 2 -nc1
Starting program: /usr/local/bin/msieve -s gnfs.dat -l ggnfs.log -i gnfs.ini -v -nf gnfs.fb -t 2 -nc1
[Thread debugging using libthread_db enabled]


Msieve v. 1.42
Sun Jul 12 15:26:00 2009
random seeds: ef636916 ce4d7035
factoring 8152076978298033032614464108763248275638533543184212928005190196721776176660059482366247964158442259 (100 digits)
searching for 15-digit factors
[New Thread 0x7ffff6fbe950 (LWP 7285)]
commencing number field sieve (100-digit input)
R0: -1051744772620805140007329
R1:  137992068071417
A0:  3371890176974270511868545840
A1: -10886188273030830462761
A2: -26456360310836702
A3:  17842052158
A4:  6660
skew 1256178.72, size 1.013689e-13, alpha -5.324659, combined = 1.211124e-08

commencing relation filtering
estimated available RAM is 1010.6 MB
commencing duplicate removal, pass 1
found 153747 hash collisions in 2488365 relations
commencing duplicate removal, pass 2
found 77110 duplicates and 2411255 unique relations
memory use: 8.2 MB
reading ideals above 100000
commencing singleton removal, initial pass
memory use: 74.6 MB
reading all ideals from disk
memory use: 74.0 MB
keeping 3475453 ideals with weight <= 200, target excess is 19293

Program received signal SIGSEGV, Segmentation fault.
0x000000000043a7cc in filter_read_lp_file ()
Current language:  auto; currently asm
(gdb) bt full
#0  0x000000000043a7cc in filter_read_lp_file ()
No locals.
#1  0x0000000000424307 in nfs_filter_relations ()
No locals.
#2  0x000000000041a577 in factor_gnfs ()
No locals.
#3  0x0000000000406799 in msieve_run ()
No locals.
#4  0x0000000000404557 in factor_integer ()
No locals.
#5  0x0000000000404f91 in main ()
No locals.
(gdb)

My msieve is compiled with ECM=1 by gcc 4.3.2 on 64-bit Linux.

[10metreh]

Here's a degree 4 run that failed to produce a poly:

Tue Jul 14 09:51:11 2009  
Tue Jul 14 09:51:11 2009  
Tue Jul 14 09:51:11 2009  Msieve v. 1.42
Tue Jul 14 09:51:11 2009  random seeds: a57d8d40 c56ea5b0
Tue Jul 14 09:51:11 2009  factoring 215462567442955338756702599130397733703866085260479837730988052460823759983714539036267990403 (93 digits)
Tue Jul 14 09:51:14 2009  searching for 15-digit factors
Tue Jul 14 09:51:16 2009  commencing number field sieve (93-digit input)
Tue Jul 14 09:51:17 2009  commencing number field sieve polynomial selection
Tue Jul 14 09:51:17 2009  time limit set to 0.16 hours
Tue Jul 14 09:51:17 2009  searching leading coefficients from 1 to 1649401
Tue Jul 14 10:05:10 2009  polynomial selection complete
Tue Jul 14 10:05:10 2009  warning: polynomial is corrupt
Tue Jul 14 10:05:10 2009  elapsed time 00:13:59

The DOS window output at the end was:

polynomial selection complete
warning: polynomial is corrupt
error generating or reading NFS polynomials
elapsed time 00:13:59