Python Driver for GGNFS and MSIEVE

[jrk]

Originally Posted by jrk

Try building with CUDA=1

Also, please use the v1.48 release version of msieve instead of the interim SVN version you got. There was a lot of code-churn in this release, so some stuff might have been broken that was fixed prior to the finished release.

[firejuggler]

the main problem is that i do not know *how* to compile msieve. i'm on Win XP.

>make -C 1.48 x86 ECM=1 CUDA=1

make: Entering directory `H:/Documents and Settings/Vincent/Mes documents/Téléchargements/msieve/msieve/tags/1.48'
gcc -D_FILE_OFFSET_BITS=64 -O3 -fomit-frame-pointer -march=k8 -DNDEBUG -D_LARGEF
ILE64_SOURCE -Wall -W -I. -Iinclude -Ignfs -Ignfs/poly -Ignfs/poly/stage1 -DHAV
E_GMP_ECM -I"$CUDA_PATHinclude" -DHAVE_CUDA -c -o common/filter/clique.o common/
filter/clique.c
In file included from include\msieve.h:24,
from include\common.h:18,
from common\filter\filter.h:23,
from common\filter\filter_priv.h:23,
from common/filter/clique.c:15:
include\util.h:26: windows.h: No such file or directory
include\util.h:27: process.h: No such file or directory
include\util.h:55: stdio.h: No such file or directory
include\util.h:56: stdlib.h: No such file or directory
include\util.h:57: string.h: No such file or directory
include\util.h:58: ctype.h: No such file or directory
include\util.h:59: stdarg.h: No such file or directory
include\util.h:60: time.h: No such file or directory
include\util.h:61: sys\types.h: No such file or directory
include\util.h:62: sys\stat.h: No such file or directory
include\util.h:64: inttypes.h: No such file or directory
include\util.h:69: math.h: No such file or directory
In file included from include\common.h:19,
from common\filter\filter.h:23,
from common\filter\filter_priv.h:23,
from common/filter/clique.c:15:
include\gmp_xface.h:19: gmp.h: No such file or directory
make: *** [common/filter/clique.o] Error 1
make: Leaving directory `H:/Documents and Settings/Vincent/Mes documents/Téléchargements/msieve/msieve/tags/1.48'

edit : ok, I realise that i need visual studio.

[jrk]

Originally Posted by firejuggler

the main problem is that i do not know *how* to compile msieve. i'm on Win XP.

include\gmp_xface.h:19: gmp.h: No such file or directory

edit : ok, I realise that i need visual studio.

You need GMP installed for msieve to build. I'm not a Windows user, so can't detail the steps you need, but perhaps others can.

Brian Gladman maintains the Visual Studio (2010) build files for msieve so it should be able to build in VS without too much trouble.

[Brian Gladman]

Provided you have built MPIR, Visual Studio 2010 will build the non-CUDA version of msieve 'out of the box'. The free Visual Studio C++ Express 2010 can build msieve for win32 and, if the free Windows 7.1 SDK is installed, for x64 as well.

The CUDA version of msieve can only be built with the full Visual Studio 2010 product and also requires Nvidia Parallel Nsight.

Brian

[firejuggler]

As I can't afford it, i'll guess i'll wait for someone to compile it for me. Or continue to use 1.44 gpu. Or try again tommorow.

[jrk]

Actually, aren't there already v1.48 Windows builds (with and without CUDA) on the SF files page?

Yes, see here: http://sourceforge.net/projects/msie.../Msieve v1.48/

The Windows CUDA build is in msieve148_gpu.zip and the non-CUDA build is in msieve148.zip. The CUDA build has both the regular and "fermi-tuned" GPU kernels available, so use what you need.

[firejuggler]

Ah. Thank you. I didn't notice that page (up till now i only seen 'normal' 1.48) . Many, many thanks

[em99010pepe]

Under your script Brian, what's the relation between polynomial search time with the composite size?

Thank you.

Carlos

[Brian Gladman]

If I recall correctly, I didn't intentionally change the way this is done in the original Perl script so it should be the same as it used to be. I didn't look in any detail at the sematics of the original Perl so I can't answer your question without going back and analysing how this works.

[jasonp]

Unless the scripts force a shutdown early, the internal time limit for poly selection is a linear interpolation between the entries of the following table.

Note that there is no consideration for parallel poly selection or for the number of sieving machines that are waiting for the poly selection to complete. Also, if you run with '-np X,Y' to search for a polynomial in a particular range, then there is no limit on the time spent (but a limit on the amount of time given to each of the leading coefficients between X and Y).

static const poly_deadline_t time_limits[] = {
	{MIN_NFS_BITS, 4 * 60},
	{304, 8 * 60},
	{320, 15 * 60},
	{348, 30 * 60},
	{365, 1 * 3600},
	{383, 2 * 3600},
	{399, 4 * 3600},
	{416, 8 * 3600},
	{433, 16 * 3600},
	{449, 32 * 3600},
	{466, 64 * 3600},
	{482, 100 * 3600},
	{498, 200 * 3600},
	{514, 300 * 3600},
};

[chris2be8]

What would you recommend for the ratio between time searching for polynomials and time sieving? I'm doing a c145 (with my trick to pipeline polynomial selection), stage 1 said "searching leading coefficients from 1 to 1614000" and it's got to 3132 in 1 day. Which will take much longer to reach 1614000 that the sieving will.

It also said "time limit set to 93.25 hours" but does it stop then when just running stage 1?

I've also noticed that stage 1 has used 7 CPU hours in 1 day, stage 2 used 1 day per day. This is on a dual core AMD.

Chris K