Msieve 1.40 Beta Feedback - Page 4

Batalov · 2009-03-04, 20:11

I've spied in some recent package the -march=core2 option and tried it on beta2 (on x86_64). Is it useful (different from nocona?)? Don't know yet, decided to try. Compiled fine (there were some assembly warnings though), ran some tests, ok. Doesn't link with gmp-ecm svn (a.k.a 6.2.2) library, hmm. Fell back to 6.2.1 for msieve. Linking to mpir, of course, had no problems.

(The previously PM'd crash on small numbers is gone.)

It finishes an snfs-175 at home as we speak.

EDIT: core2 option is added since gcc-4.3 (not listed in 4.2.x docs)

Batalov · 2009-03-05, 02:49

P.S. The following has nothing to do with the thread

...except that the number was finished by msieve-1.40beta2. So the program works fine.

The aforementioned snfs-175 was a c155 and it split to p78.p78.

So, Use ~~sunscreen~~ msieve-1.40!
The long-term benefits of ~~sunscreen~~ msieve-1.40 have been proved by scientists, whereas the rest of my advice has no basis more reliable than my own meandering experience.

10metreh · 2009-03-06, 20:22

Jason, could you remove the lower limit on NFS in the final version?

jasonp · 2009-03-06, 20:32

Quote:

Originally Posted by 10metreh

Jason, could you remove the lower limit on NFS in the final version?

The beta should have no lower limit on the input size for NFS postprocessing. Is that not the case for you?

10metreh · 2009-03-07, 07:45

Quote:

Originally Posted by jasonp

The beta should have no lower limit on the input size for NFS postprocessing. Is that not the case for you?

Whoops, just noticed I was testing with 1.39...

jasonp · 2009-03-07, 14:49

Quote:

Originally Posted by 10metreh

Whoops, just noticed I was testing with 1.39...

v1.39 is so last year

Jeff Gilchrist · 2009-03-07, 18:52

Hi Jason,

Right now I'm testing the MSVC builds with QS and so far so good, the 1.40 is a nice bit faster than 1.39 for 32bit (probably your CMOV fix).

But there is still the problem that Brian's MSVC project file seems to be outdated in that it tries to build two executables, one GGNFS.exe and one MPQS.exe file. These executables are identical except they have different names. I'm not sure if Brian regularly reads these forums but I can e-mail and maybe either he or I could go and modify it so it just creates 1 executable called msieve.exe.

What do you think?

jasonp · 2009-03-08, 03:25

Quote:

Originally Posted by Jeff Gilchrist

But there is still the problem that Brian's MSVC project file seems to be outdated in that it tries to build two executables, one GGNFS.exe and one MPQS.exe file. These executables are identical except they have different names. I'm not sure if Brian regularly reads these forums but I can e-mail and maybe either he or I could go and modify it so it just creates 1 executable called msieve.exe.

The NFS code used to be optional, because the numerical integrations involved started off getting solved using the GSL library, which I didn't want users to have to compile if all they wanted was QS (the NFS postprocessing wasn't even written back then). Once msieve used its own numerical integration code that wasn't necessary, and the two subprojects were made to generate the same binary. No reason that should remain the case, if you want to go ahead and make the change. Just email me the new project files (or a patch to the old project files, if the changes are small) and I'll include them in the final release of v1.40

mklasson · 2009-03-08, 04:49

Jason,

just encountered a problem with msieve v1.40b2, though it's apparently also present in ye olde v1.16. :)

Code:

>msieve 3955675543118959886598527
error: tiny factoring failed

10metreh · 2009-03-08, 07:33

Quote:

Originally Posted by mklasson

Jason,

just encountered a problem with msieve v1.40b2, though it's apparently also present in ye olde v1.16. :)

Code:

>msieve 3955675543118959886598527
error: tiny factoring failed

There is a thread related to this error: http://www.mersenneforum.org/showthread.php?t=11384

Jeff Gilchrist · 2009-03-08, 19:29

Quote:

Originally Posted by jasonp

No reason that should remain the case, if you want to go ahead and make the change. Just email me the new project files (or a patch to the old project files, if the changes are small) and I'll include them in the final release of v1.40

Ok I will go ahead and make the changes then and send you the new project files.

I have done some testing in Windows and where 1.39 used to crash for the MSVC build in QS it now works fine with 1.40beta2. I'm also linking against MPIR instead of GMP so I'm not sure if that makes any difference. Is it just the poly selection that uses GMP code?

Here are some NFS post-processing benchmarks:

Code:

Intel Core2 Q9550 @ 3.4GHz (Vista 64bit)

MSIEVE GNFS Post-Processing
===========================
C121 = 8996941959382577683409613171454174240738275788293429353288678806601664747011544951714674576925430397444054300751795281737
prp58 factor: 7113268608388826628041175633889105572982840043845203074777
prp64 factor: 1264811221773952923237987117254409784246941529158530073219882481
Benchmark: msieve -nc -v -t 4

MSIEVE 1.40 64bit MSVC =  32m 51.095s
MSIEVE 1.39 64bit MSVC =  33m 41.596s
MSIEVE 1.40 32bit gcc  =  35m 31.417s
MSIEVE 1.39 32bit gcc  =  35m 49.944s
MSIEVE 1.40 32bit MSVC =  36m 37.735s

Some QS benchmarks:

Code:

Intel Core2 Q9550 @ 3.4GHz (Vista 64bit)

SIQS (C85 = 1877138824359859508015524119652506869600959721781289179190693027302028679377371001561)
==================================================================================================
MSIEVE 1.39 64bit MSVC = 19m 44.561s
MSIEVE 1.40 64bit MSVC = 20m 03.693s
MSIEVE 1.40 32bit MSVC = 14m 00.927s
MSIEVE 1.40 32bit gcc  = 14m 59.053s
MSIEVE 1.39 32bit gcc  = 17m 13.576s

SIQS (C80 = 43756152090407155008788902702412144383525640641502974083054213255054353547943661)
=============================================================================================
MSIEVE 1.39 64bit MSVC =   6m 08.020s
MSIEVE 1.40 64bit MSVC =   6m 11.059s
MSIEVE 1.40 32bit MSVC =   4m 31.999s
MSIEVE 1.40 32bit gcc  =   4m 56.035s
MSIEVE 1.39 32bit gcc  =   5m 32.956s

SIQS (C75 = 281396163585532137380297959872159569353696836686080935550459706878100362721)
========================================================================================
MSIEVE 1.40 64bit MSVC =  2m 18.664s
MSIEVE 1.39 64bit MSVC =  2m 20.431s
MSIEVE 1.40 32bit MSVC =  1m 45.507s
MSIEVE 1.39 32bit MSVC =  1m 52.508s
MSIEVE 1.40 32bit gcc  =  1m 53.237s
MSIEVE 1.39 32bit gcc  =  2m 02.585s

SIQS (C65 = 34053408309992030649212497354061832056920539397279047809781589871)
==============================================================================
MSIEVE 1.39 64bit MSVC =  0m 15.522s
MSIEVE 1.40 64bit MSVC =  0m 15.978s
MSIEVE 1.39 32bit MSVC =  0m 14.242s
MSIEVE 1.40 32bit MSVC =  0m 14.423s
MSIEVE 1.39 32bit gcc  =  0m 15.023s
MSIEVE 1.40 32bit gcc  =  0m 15.244s

2009-03-04, 20:11	#34
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 2505₁₆ Posts	I've spied in some recent package the -march=core2 option and tried it on beta2 (on x86_64). Is it useful (different from nocona?)? Don't know yet, decided to try. Compiled fine (there were some assembly warnings though), ran some tests, ok. Doesn't link with gmp-ecm svn (a.k.a 6.2.2) library, hmm. Fell back to 6.2.1 for msieve. Linking to mpir, of course, had no problems. (The previously PM'd crash on small numbers is gone.) It finishes an snfs-175 at home as we speak. EDIT: core2 option is added since gcc-4.3 (not listed in 4.2.x docs) Last fiddled with by Batalov on 2009-03-04 at 20:16 Reason: rtfm'd a little

2009-03-08, 04:49	#42
mklasson Feb 2004 2·3·43 Posts	Jason, just encountered a problem with msieve v1.40b2, though it's apparently also present in ye olde v1.16. :) Code: >msieve 3955675543118959886598527 error: tiny factoring failed

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Msieve 1.53 feedback	xilman	Msieve	149	2018-11-12 06:37
Msieve 1.51 feedback	xilman	Msieve	256	2014-01-26 22:06
Msieve v1.46 feedback	em99010pepe	Msieve	153	2010-12-12 14:21
Msieve 1.44 feedback	xilman	Msieve	111	2010-09-14 21:50
Msieve 1.43 feedback	Jeff Gilchrist	Msieve	47	2009-11-24 15:53

2009-03-05, 02:49	#35
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 3⁶×13 Posts	P.S. The following has nothing to do with the thread ...except that the number was finished by msieve-1.40beta2. So the program works fine. The aforementioned snfs-175 was a c155 and it split to p78.p78. So, Use ~~sunscreen~~ msieve-1.40! The long-term benefits of ~~sunscreen~~ msieve-1.40 have been proved by scientists, whereas the rest of my advice has no basis more reliable than my own meandering experience.

2009-03-06, 20:22	#36
10metreh Nov 2008 2322₁₀ Posts	Jason, could you remove the lower limit on NFS in the final version?

2009-03-07, 18:52	#40
Jeff Gilchrist Jun 2003 Ottawa, Canada 3×17×23 Posts	Hi Jason, Right now I'm testing the MSVC builds with QS and so far so good, the 1.40 is a nice bit faster than 1.39 for 32bit (probably your CMOV fix). But there is still the problem that Brian's MSVC project file seems to be outdated in that it tries to build two executables, one GGNFS.exe and one MPQS.exe file. These executables are identical except they have different names. I'm not sure if Brian regularly reads these forums but I can e-mail and maybe either he or I could go and modify it so it just creates 1 executable called msieve.exe. What do you think?