mersenneforum.org  

Go Back   mersenneforum.org > Factoring Projects > GMP-ECM

Reply
 
Thread Tools
Old 2006-05-23, 18:31   #1
Mystwalker
 
Mystwalker's Avatar
 
Jul 2004
Potsdam, Germany

3×277 Posts
Default Benchmarking suite discussion

In one or two other threads, there were already some speed comparisons between different versions of gmp and/or ecm.

But I think we should come up with a more thorough investigation. In my opinion, we at least need various test composites (various lengths, some of a special form) and several bound combinations for every architecture, compile parameter setting and gmp resp. ecm version (I guess we can eliminate a lot of the overall possible combinations, when we see that certain combinations just don't seem to make any sense.).

So, we have the following dimensions:

Composites: Are there any founded candidates?
Bounds: I guess at least 40-55 digits...
Architectures: All there are (interest for).
Compile parameter settings: asm-redc, ntt (which did not work for me), tune, gwnum, ...
gmp: I haven't seen a case where 4.1.4 is faster than 4.2 - maybe we can stick to the latter here
ecm: 6.0.1, 6.1, maybe 6.1-beta2

Opinions? Recommondations? Alternatives? Critics?
Mystwalker is offline   Reply With Quote
Old 2006-05-23, 20:41   #2
Phil MjX
 
Phil MjX's Avatar
 
Sep 2004

101110012 Posts
Default

hi !
on the ecm-dev archives, P Zimmermann wrote :
"I just noticed that using Pierrick Gaudry's new assembly code for redc does
not work in ecm-6.1. Therefore version 6.1.1 will soon appear. Sorry, my fault."

I think we should wait for this release before benchmarking version 6.1.x

Regards
Philippe.
Phil MjX is offline   Reply With Quote
Old 2006-05-25, 10:28   #3
Mystwalker
 
Mystwalker's Avatar
 
Jul 2004
Potsdam, Germany

14778 Posts
Default

Sounds reasonable. But I think we can discuss the benchmarking procedure now, so we have it complete when ecm-6.1.1 is out.
Mystwalker is offline   Reply With Quote
Old 2006-05-25, 15:20   #4
Andi47
 
Andi47's Avatar
 
Oct 2004
Austria

46628 Posts
Default

Quote:
Originally Posted by Phil MjX
hi !
on the ecm-dev archives, P Zimmermann wrote :
"I just noticed that using Pierrick Gaudry's new assembly code for redc does
not work in ecm-6.1. Therefore version 6.1.1 will soon appear. Sorry, my fault."

I think we should wait for this release before benchmarking version 6.1.x

Regards
Philippe.
Hmmmm... Mayby that's why 6.1 is slower than 6.0.1 on my computer...?
Andi47 is offline   Reply With Quote
Old 2006-05-26, 21:08   #5
Mystwalker
 
Mystwalker's Avatar
 
Jul 2004
Potsdam, Germany

14778 Posts
Default

What CPU do you have?

On the Pentium-M I've tested so far, 6.1 was always faster than 6.0.1. The only noteworthy difference in the curve completion times was the asm-redc-enabled version with tuned gmp-params. Interestingly, it was 4-5% slower for a 205 digit number in step2, but 4-5% faster for a 507 digit number in step2...

Well, that and a 1% slowdown for the asm-redc-enabled 6.1-beta2 in step1 for the 507 digit number (747 instead of 738 seconds).
Mystwalker is offline   Reply With Quote
Old 2006-06-01, 00:13   #6
geoff
 
geoff's Avatar
 
Mar 2003
New Zealand

13·89 Posts
Default

Quote:
Originally Posted by Mystwalker
Bounds: I guess at least 40-55 digits...
The important thing with comparing stage two of the curve is to base the comparison on the actual bound used by gmp-ecm (reported as B2' when run with the -v switch), not on the bound given on the command line. The actual bound used changes with gmp-ecm version, and with the -k value used, and maybe with other options used. It is probably OK just to choose a standard B2 bound for each digit level and make a linear adjustment for curves that are different from the standard.
geoff is offline   Reply With Quote
Old 2006-06-10, 13:16   #7
Mystwalker
 
Mystwalker's Avatar
 
Jul 2004
Potsdam, Germany

14778 Posts
Default

I've just seen that GMP 4.2.1 is out (already since over a month, it seems...). Do the mentioned speedup/bugfixes affect ecm?

Last fiddled with by Mystwalker on 2006-06-10 at 13:20
Mystwalker is offline   Reply With Quote
Old 2006-06-11, 10:08   #8
akruppa
 
akruppa's Avatar
 
"Nancy"
Aug 2002
Alexandria

2,467 Posts
Default

According to the announcement, the only speedup is for mpz_divexact() which we rarely use in GMP-ECM and never in speed-critical sections of the code.

Alex
akruppa is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Looking for benchmarking help with a Phenom or PhenomII X6 mrolle Software 25 2012-03-14 14:15
Photoshop Creative Suite 5 and CUDA Rodrigo GPU Computing 1 2011-07-04 10:51
GMP 5.0.1 vs GMP 4.1.4 benchmarking unconnected GMP-ECM 5 2011-04-03 16:16
Benchmarking dual-CPU machines garo Software 2 2010-09-27 20:33
Benchmarking challenge! Xyzzy Software 17 2003-08-26 15:43

All times are UTC. The time now is 17:01.


Thu Oct 21 17:01:41 UTC 2021 up 90 days, 11:30, 1 user, load averages: 2.21, 2.21, 2.06

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.