Aliqueit.exe discussion - Page 12

10metreh · 2009-04-02, 06:42

Quote:

Originally Posted by Joshua2

Change in 1.40 msieve
- Modified the main driver to allow NFS on any size input, no matter
how small, if only the postprocessing is desired
So do you think aliq 1.07 might be smart enough to be able to set use msieve for poly to false under c98, and true over? And msieve can still do the post-proc. I would like to test the breakeven point on my box.

Msieve's NFS lower limit is (probably) 2^320, because Andi47 tried using msieve poly selection on a low C97 and failed, but higher C97s worked.

Andi47 · 2009-04-02, 06:48

Quote:

Originally Posted by 10metreh

Msieve's NFS lower limit is (probably) 2^320, because Andi47 tried using msieve poly selection on a low C97 and failed, but higher C97s worked.

That's also my guess. I think this guess has also made it into my gnfs.ub script - I set the treshold there to do QS below 2^320 and GNFS above.

mklasson · 2009-04-02, 06:52

Quote:

Originally Posted by mdettweiler

Based on my personal experience (and others' that I've read about, too) the crossover point with the current array of applications is around 96 digits. I personally do anything 95 digits or less with QS, and 96 digits or more with GNFS.

But that's on linux, right? Having a twice as fast gnfs-lasieve makes a big difference.

Quote:

Originally Posted by Joshua2

C98 took just over 2 hours using msieve to poly select. Its funny it used msieve auto w/ my C108, but not the C98. If I were you I would make aliquot use 96 as the crossover, and if its a 96 or 97 use the other pearl script that doesn't use msieve. That should speed things up right?

See above. On my windows x64 i7 machine the crossover point seems to be ~99, maybe 98 digits. That's why its customisable though. Just use whatever's best on your machine.

Quote:

Originally Posted by Joshua2

Edit: Think I need to put "" in like msieve_cmd = "C:\Program Files\ggnfs\msieve" - that seems to have fixed it... thats weird

Not weird at all if you don't have "c:\program files\ggnfs" in your PATH envvar like I mentioned a couple of posts back. If you do, then yes, that's kinda weird.

Quote:

Originally Posted by Joshua2

Change in 1.40 msieve
- Modified the main driver to allow NFS on any size input, no matter
how small, if only the postprocessing is desired
So do you think aliq 1.07 might be smart enough to be able to set use msieve for poly to false under c98, and true over? And msieve can still do the post-proc. I would like to test the breakeven point on my box.

Sounds possible. I'll just force use_msieve_polyfind=false for c97-.

Joshua2 · 2009-04-02, 06:55

Quote:

Originally Posted by 10metreh

Msieve's NFS lower limit is (probably) 2^320, because Andi47 tried using msieve poly selection on a low C97 and failed, but higher C97s worked.

http://mersenneforum.org/showpost.ph...&postcount=107
It is 2^320. I am trying to recompile a 2^300 version, but its complaining it needs gmp.h. Any ideas?

10metreh · 2009-04-02, 06:57

Quote:

Originally Posted by Joshua2

http://mersenneforum.org/showpost.ph...&postcount=107
It is 2^320. I am trying to recompile a 2^300 version, but its complaining it needs gmp.h. Any ideas?

It needs to be compiled with GMP=1 for the new poly selection to work. I'm not a compiling person, but I guess it needs gmp.h as well.

mklasson · 2009-04-02, 06:57

Quote:

Originally Posted by Joshua2

http://mersenneforum.org/showpost.ph...&postcount=107
It is 2^320. I am trying to recompile a 2^300 version, but its complaining it needs gmp.h. Any ideas?

Get gmp.

Joshua2 · 2009-04-02, 07:00

I downloaded the latest gmp source and stuck the folder in the msieve folder, but that didn't do anything. I do know what it means set GMP=1, where do I do that? I've got very little experience in c++, but I had a summer job in c#.

mdettweiler · 2009-04-02, 07:16

Quote:

Originally Posted by mklasson

But that's on linux, right? Having a twice as fast gnfs-lasieve makes a big difference.

Well, I compiled my (32-bit) Linux binaries myself, and didn't do anything to mess with asm code...so it's possible that I'm using slower binaries, too. At any rate, I seem to get similar speeds on 32-bit Windows (using the GGNFS precompiled binaries) vs. 32-bit Linux on the same machine.

BTW, this reminded me to ask: is there anything special I need to do in order to get the asm code to be compiled in on 32-bit Linux? Or do I already have it after compiling with a normal "./configure", "make"?

mklasson · 2009-04-02, 07:20

Quote:

Originally Posted by Joshua2

I downloaded the latest gmp source and stuck the folder in the msieve folder, but that didn't do anything. I do know what it means set GMP=1, where do I do that? I've got very little experience in c++, but I had a summer job in c#.

Err, I think I woke up too early. GMP is not GMP-ECM ffs... Get gmp at http://gmplib.org/.

Nevertheless: if you're compiling msieve with msvc I think GMP=1 is already set. (project settings->C/C++->preprocessor->defines or something like that). By default I think it expects your folder structure to have a very specific layout though, with a folder called "gmp-ecm" (containing gmp-ecm obv.) in the same directory as your "msieve" dir. Same goes for the "gmp" dir.

mklasson · 2009-04-02, 07:21

Quote:

Originally Posted by mdettweiler

Well, I compiled my (32-bit) Linux binaries myself, and didn't do anything to mess with asm code...so it's possible that I'm using slower binaries, too. At any rate, I seem to get similar speeds on 32-bit Windows (using the GGNFS precompiled binaries) vs. 32-bit Linux on the same machine.

BTW, this reminded me to ask: is there anything special I need to do in order to get the asm code to be compiled in on 32-bit Linux? Or do I already have it after compiling with a normal "./configure", "make"?

I think it's already taken care of. You should really get the fast 64-bit gnfs-lasieve if you're using a 64-bit machine though. Twice the speed, twice the fun.

Joshua2 · 2009-04-02, 07:33

I did download the gmp souce from there. It does appear to be wanting gmp. I don't see any gmp-ecm separate from gmp though, although the msieve source talks about it...Maybe I should leave this for the experts, but I wish I could do it too. :) (I can provide the fix on the 2^320, someone else provide compiling :)
Edit: the time searching for polynomials is just arbitrary it seems and the ranges of bits are pretty large. I bet they could be fine tuned a lot if there weren't so hard to get at.

2009-04-02, 07:33	#132
Joshua2 Sep 2004 13×41 Posts	I did download the gmp souce from there. It does appear to be wanting gmp. I don't see any gmp-ecm separate from gmp though, although the msieve source talks about it...Maybe I should leave this for the experts, but I wish I could do it too. :) (I can provide the fix on the 2^320, someone else provide compiling :) Edit: the time searching for polynomials is just arbitrary it seems and the ranges of bits are pretty large. I bet they could be fine tuned a lot if there weren't so hard to get at. Last fiddled with by Joshua2 on 2009-04-02 at 07:47

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Resuming aliqueit	johnadam74	Aliquot Sequences	4	2016-03-28 12:32
Apparent aliqueit issue with specifying factors	pakaran	Aliquot Sequences	2	2015-09-12 23:10
Using Several Instances of Aliqueit for a large gnfs job	EdH	Aliquot Sequences	6	2011-12-13 18:58
Setting up aliqueit	science_man_88	Aliquot Sequences	185	2011-11-08 12:18
Tried out aliqueit.exe: ggnfs failing	Greebley	Aliquot Sequences	35	2010-02-13 15:23

2009-04-02, 07:00	#128
Joshua2 Sep 2004 13×41 Posts	I downloaded the latest gmp source and stuck the folder in the msieve folder, but that didn't do anything. I do know what it means set GMP=1, where do I do that? I've got very little experience in c++, but I had a summer job in c#.