mersenneforum.org  

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

Closed Thread
 
Thread Tools
Old 2006-03-28, 05:06   #23
Andi47
 
Andi47's Avatar
 
Oct 2004
Austria

2·17·73 Posts
Default

I compiled GMP 4.2 + ecm6.1-beta2 (using ./configure -–with-gmp=/usr/local --enable-asm-redc) yesterday in the evening and was running the benchmark until a few moments ago.

Code:
GMP-ECM 6.1-beta2 [powered by GMP 4.2] [ECM]
Input number is 7697965562235071080146894266187742061373611050088218785901815194731255580083856950577371947531398134889789342805715603682570745833656049440232111949793108909207499508865559739953421824814393825917900810041 (205 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=926057434
Step 1 took 124969ms
Step 2 took 46859ms
Run 2 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=300961870
Step 1 took 125000ms
Step 2 took 46860ms
Run 3 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=1576487558
Step 1 took 125281ms
Step 2 took 46812ms
Run 4 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=3760506750
Step 1 took 125063ms
Step 2 took 47203ms
Run 5 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=2368405546
Step 1 took 124859ms
Step 2 took 47516ms
that's *slow* - even slower than without the --enable-asm-redc option

My steps when compiling were:

tar -xvzf gmp-4.2.tar.gz
cd gmp-4.2/
./configure
make install
make check
cd ~

tar -xvzf ecm-6.1-beta2.tar.gz
cd ecm-6.1-beta2
./configure -–with-gmp=/usr/local --enable-asm-redc
make

Did I do something wrong - or are there still more compile options to use for a Pentium 4?
Andi47 is offline  
Old 2006-03-28, 10:22   #24
Mystwalker
 
Mystwalker's Avatar
 
Jul 2004
Potsdam, Germany

14778 Posts
Default

I still have to benchmark the asm-redc option on my P4, but I also got the impression that it is slower.

I should be able to post results this evening.
Mystwalker is offline  
Old 2006-03-28, 11:34   #25
BotXXX
 
BotXXX's Avatar
 
Aug 2003
Europe

19410 Posts
Default

I also did some tests on a idle Pentium 4 2.93 GHz cpu on the c205 of M1701

GMP-ECM 6.0.1 with GMP 4.2
Code:
BotXXX P4 - GMP-ECM 6.0.1 [powered by GMP 4.2] [ECM]
Input number is 7697965562235071080146894266187742061373611050088218785901815194731255580083856950577371947531398134889789342805715603682570745833656049440232111949793108909207499508865559739953421824814393825917900810041 (205 digits)
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=370079320
Step 1 took 95210ms
Step 2 took 34290ms
Run 2 out of 5:
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=676809121
Step 1 took 94995ms
Step 2 took 34479ms
Run 3 out of 5:
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=3151733463
Step 1 took 95011ms
Step 2 took 34506ms
Run 4 out of 5:
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=1746079971
Step 1 took 95088ms
Step 2 took 34552ms
Run 5 out of 5:
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=2760815317
Step 1 took 95043ms
Step 2 took 34534ms
GMP-ECM 6.1 beta 2 with GMP 4.2
Code:
BotXXX P4 - GMP-ECM 6.1-beta2 [powered by GMP 4.2] [ECM]
Input number is 7697965562235071080146894266187742061373611050088218785901815194731255580083856950577371947531398134889789342805715603682570745833656049440232111949793108909207499508865559739953421824814393825917900810041 (205 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=1386523129
Step 1 took 94531ms
Step 2 took 45203ms
Run 2 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=872759764
Step 1 took 95313ms
Step 2 took 45593ms
Run 3 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=1114418950
Step 1 took 95547ms
Step 2 took 45641ms
Run 4 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=336469270
Step 1 took 95219ms
Step 2 took 46484ms
Run 5 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=3872377383
Step 1 took 95266ms
Step 2 took 46328ms
GMP-ECM 6.1 beta 2 with GMP 4.2 and --enable-asm-redc
Code:
BotXXX P4 - GMP-ECM plus 6.1-beta2 [powered by GMP 4.2] [ECM]
Input number is 7697965562235071080146894266187742061373611050088218785901815194731255580083856950577371947531398134889789342805715603682570745833656049440232111949793108909207499508865559739953421824814393825917900810041 (205 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=141995905
Step 1 took 144844ms
Step 2 took 47828ms
Run 2 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=2723037926
Step 1 took 144937ms
Step 2 took 47766ms
GMP-ECM 6.0.1 with GMP 4.2 and higher B2
Code:
BotXXX P4 - GMP-ECM 6.0.1 [powered by GMP 4.2] [ECM]
Input number is 7697965562235071080146894266187742061373611050088218785901815194731255580083856950577371947531398134889789342805715603682570745833656049440232111949793108909207499508865559739953421824814393825917900810041 (205 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=1770901312
Step 1 took 95594ms
Step 2 took 43465ms
Run 2 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=2895472145
Step 1 took 95109ms
Step 2 took 43899ms
Run 3 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=3492736638
Step 1 took 94750ms
Step 2 took 43696ms
All three were compiled in MinGW with GCC 3.4.5 (mingw special)
and run with the following command line

ecm -c 5 3000000 < input.txt

except the last one (6.0.1 with higher B2) which used

ecm -c 5 3000000 5706890290 < input.txt
BotXXX is offline  
Old 2006-03-28, 12:26   #26
BotXXX
 
BotXXX's Avatar
 
Aug 2003
Europe

2·97 Posts
Default

Some more testing. Just to be full ;)

GMP-ECM 6.1 beta 2 with GMP 4.1.4
Code:
BotXXX P4 - GMP-ECM 6.1-beta2 [powered by GMP 4.1.4] [ECM]
Input number is 7697965562235071080146894266187742061373611050088218785901815194731255580083856950577371947531398134889789342805715603682570745833656049440232111949793108909207499508865559739953421824814393825917900810041 (205 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=651263285
Step 1 took 106172ms
Step 2 took 50703ms
Run 2 out of 3:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=612425771
Step 1 took 106078ms
Step 2 took 50547ms
Run 3 out of 3:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=3694284041
Step 1 took 106281ms
Step 2 took 50656ms
GMP-ECM 6.1 beta 2 with GMP 4.1.4 and --enable-asm-redc
Code:
BotXXX P4 - GMP-ECM plus 6.1-beta2 [powered by GMP 4.1.4] [ECM]
Input number is 7697965562235071080146894266187742061373611050088218785901815194731255580083856950577371947531398134889789342805715603682570745833656049440232111949793108909207499508865559739953421824814393825917900810041 (205 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=924551896
Step 1 took 139406ms
Step 2 took 51954ms
Run 2 out of 3:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=3221873555
Step 1 took 139765ms
Step 2 took 52047ms
Run 3 out of 3:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=2106245650
Step 1 took 139453ms
Step 2 took 52000ms
BotXXX is offline  
Old 2006-03-28, 17:42   #27
Mystwalker
 
Mystwalker's Avatar
 
Jul 2004
Potsdam, Germany

3·277 Posts
Default

I've just completed my testing on the P4.

Test setup:

GMP-ECM 6.1-beta2 [powered by GMP 4.2] [ECM]
Input number is 769796556223507108014689426618774206137361105008821878590181519473125558008385695057
7371947531398134889789342805715603682570745833656049440232111949793108909207499508865559739953421824
814393825917900810041 (205 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=579882999

default:

Step 1 took 113109ms
Step 2 took 59000ms

--enable-asm-redc:

Step 1 took 181329ms
Step 2 took 62281ms

Using the -v option, I noticed that each step of stage2 is slower or equally fsat with --enable-asm-redc.
Mystwalker is offline  
Old 2006-03-28, 17:53   #28
Andi47
 
Andi47's Avatar
 
Oct 2004
Austria

2·17·73 Posts
Default

Please post your curves also into the M1701-Thread. When all do benchmarking with the c205, we will complete the 40 digit level just by running benchmark curves

ECM 6.0.1 with GMP 4.2
Code:
GMP-ECM 6.0.1 [powered by GMP 4.2] [ECM]
Input number is 7697965562235071080146894266187742061373611050088218785901815194731255580083856950577371947531398134889789342805715603682570745833656049440232111949793108909207499508865559739953421824814393825917900810041 (205 digits)
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=2577813280
Step 1 took 82376ms
Step 2 took 29635ms
Run 2 out of 5:
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=2350816680
Step 1 took 82366ms
Step 2 took 29762ms
Run 3 out of 5:
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=3230958904
Step 1 took 82414ms
Step 2 took 29808ms
Run 4 out of 5:
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=2788206171
Step 1 took 82462ms
Step 2 took 29829ms
Run 5 out of 5:
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=261214256
Step 1 took 82501ms
Step 2 took 29796ms
No change in speed between 4.1.999 and 4.2

I will time a couple of curves on M1061 tomorrow to compare the timings with Prime95.

Last fiddled with by Andi47 on 2006-03-28 at 17:56
Andi47 is offline  
Old 2006-03-28, 19:19   #29
Phil MjX
 
Phil MjX's Avatar
 
Sep 2004

5×37 Posts
Default

Hello !
I have read something about it,

I think in the ecm-dev list (or P Zimmermann paper "20 years of ECM"):

the --enable-asm-redc works for "smaller" numbers (<180 digits ??
the bigger ones being squarred with another method in gmp-ecm)
that's why you won't see no effect with such a "large c205 number",
try a bench with the c126 I have provided (currently ggnfs'ing)...

quoting the ecm-dev archives :
Quote:
I have svn-commited some asm functions for Redc and combined Mul/Redc.
For stage2, there is essentially no effect. For stage1, the gain is 0 to
30% depending on the size of the input and the architecture (for large
input, the gain is 0). Below is some benchs for the architectures I've
been working on.

All the best,
Pierrick.

-----------------------------------------------------------------

ECM.rev908 powered by GMP-4.1.4 (+ PG's patch for AMD64).

Timings are for input number: c155

Stage1 Stage2

Pentium M at 1.7 GHz: B1=1e5
asm: 2850 2510
noasm: 3630 2560

Opteron 250 at 2.4 GHz: B1=1e6
asm: 4900 5400
noasm: 7000 5620

Pentium 4 at 3 GHz: B1=1e5
asm: 1880 2410
noasm: 2260 2410

Athlon MP 2200+ at 1.8 GHz: B1=1e5 [cpu with high loadavg]
asm: 2110 2180
noasm: 2330 2250
I can send my pentium M binaries for testing purposes to anyone who PM me
(caution, currently compiled with gmp 4.1.99)

Best regards.
Philippe.

some benchs with the Spock number :

Quote:
GMP-ECM 6.0.1 [powered by GMP 4.2] [ECM]
Input number is 7697965562235071080146894266187742061373611050088218785901815194731255580083856950577371947531398134889789342805715603682570745833656049440232111949793108909207499508865559739953421824814393825917900810041 (205 digits)
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=131644919
Step 1 took 191044ms
Step 2 took 56208ms

GMP-ECM 6.1-beta2 [powered by GMP 4.2] [ECM]
Input number is 7697965562235071080146894266187742061373611050088218785901815194731255580083856950577371947531398134889789342805715603682570745833656049440232111949793108909207499508865559739953421824814393825917900810041 (205 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=1734072750
Step 1 took 188313ms
Step 2 took 60578ms

No great improvement in step 1, worse in step 2


GMP-ECM 6.0.1 [powered by GMP 4.2] [ECM]
Input number is 491853986948777594673218825453230605859928447069055088395227463891218667208716253406266089004846828871229829486993713464642129 (126 digits)
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=3332598043
Step 1 took 90834ms
Step 2 took 33105ms

GMP-ECM 6.1-beta2 [powered by GMP 4.2] [ECM]
Input number is 491853986948777594673218825453230605859928447069055088395227463891218667208716253406266089004846828871229829486993713464642129 (126 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=1859033613
Step 1 took 77282ms
Step 2 took 33234ms

Better in step 1, same time in step 2

Last fiddled with by Phil MjX on 2006-03-28 at 19:22
Phil MjX is offline  
Old 2006-03-28, 20:02   #30
VJS
 
VJS's Avatar
 
Dec 2004

13×23 Posts
Default

Again guys watch your b2 bounds! They are different...

Increasing from 4e9 to 5.7e9 between the clients, about 40%. try setting the b2 bounds in the command line for a accurate comparision or simply multiply the time for the lesser bound by 140%.

Any idea when a windows athlon exe will be availabe?
I'd test it on composites between 150-180 digits at the 40 and 45-digit levels.

tlc4tvm at yahoo if someone has one.
VJS is offline  
Old 2006-03-28, 21:24   #31
Phil MjX
 
Phil MjX's Avatar
 
Sep 2004

5×37 Posts
Default

Quote:
Originally Posted by VJS
Again guys watch your b2 bounds! They are different...

Increasing from 4e9 to 5.7e9 between the clients, about 40%. try setting the b2 bounds in the command line for a accurate comparision or simply multiply the time for the lesser bound by 140%.
I do partially agree :

One should use the -v option and multiply the time taken by one curve by the number of curves needed to "complete" a given range (eg the 40 digits level with B1=3e6) and then make comparisons between versions.

But in an other hand, increasing the B2 bound by 40% doesn't mean you'll increase the probability of finding a factor by 40% (by far)...so multiplying the time by 140% doesn't mean anything...

example with 6.0.1 : 2440 curves @ 3e6 needed for the 40 digits level
with 6.1.beta2 : 2350 curves @ 3e6

time to complete one curve with star trek number with my pc and 6.0.1 :
247.2 seconds
time for one curve with 6.1.beta2
248.9 seconds

time to complete the 40 digits level with 6.0.1 :
167 hours and a half
with 6.1.beta2 :
162 hours and a half

the winner is gmp-ecm 6.1beta2...

Bye.
Philippe.

Last fiddled with by Phil MjX on 2006-03-28 at 21:30
Phil MjX is offline  
Old 2006-03-28, 22:51   #32
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

2×4,079 Posts
Default

Quote:
Originally Posted by Phil MjX
the winner is gmp-ecm 6.1beta2....
Have you tried the use gwnum option? GPL compatible libraries are available at ftp://mersenne.org/gimps
Prime95 is online now  
Old 2006-03-29, 04:59   #33
Andi47
 
Andi47's Avatar
 
Oct 2004
Austria

9B216 Posts
Default

Quote:
Originally Posted by BotXXX
I also did some tests on a idle Pentium 4 2.93 GHz cpu on the c205 of M1701

GMP-ECM 6.1 beta 2 with GMP 4.2
Code:
BotXXX P4 - GMP-ECM 6.1-beta2 [powered by GMP 4.2] [ECM]
Input number is 7697965562235071080146894266187742061373611050088218785901815194731255580083856950577371947531398134889789342805715603682570745833656049440232111949793108909207499508865559739953421824814393825917900810041 (205 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=1386523129
Step 1 took 94531ms
Step 2 took 45203ms
Run 2 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=872759764
Step 1 took 95313ms
Step 2 took 45593ms
Run 3 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=1114418950
Step 1 took 95547ms
Step 2 took 45641ms
Run 4 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=336469270
Step 1 took 95219ms
Step 2 took 46484ms
Run 5 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=3872377383
Step 1 took 95266ms
Step 2 took 46328ms
GMP-ECM 6.0.1 with GMP 4.2 and higher B2
Code:
BotXXX P4 - GMP-ECM 6.0.1 [powered by GMP 4.2] [ECM]
Input number is 7697965562235071080146894266187742061373611050088218785901815194731255580083856950577371947531398134889789342805715603682570745833656049440232111949793108909207499508865559739953421824814393825917900810041 (205 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=1770901312
Step 1 took 95594ms
Step 2 took 43465ms
Run 2 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=2895472145
Step 1 took 95109ms
Step 2 took 43899ms
Run 3 out of 5:
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=3492736638
Step 1 took 94750ms
Step 2 took 43696ms
All three were compiled in MinGW with GCC 3.4.5 (mingw special)
and run with the following command line

ecm -c 5 3000000 < input.txt

except the last one (6.0.1 with higher B2) which used

ecm -c 5 3000000 5706890290 < input.txt
For this one the winner is ecm-6.0.1
Andi47 is offline  
Closed Thread

Thread Tools


All times are UTC. The time now is 22:31.


Tue Jan 31 22:31:52 UTC 2023 up 166 days, 20 hrs, 0 users, load averages: 1.10, 0.87, 0.94

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2023, 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.

≠ ± ∓ ÷ × · − √ ‰ ⊗ ⊕ ⊖ ⊘ ⊙ ≤ ≥ ≦ ≧ ≨ ≩ ≺ ≻ ≼ ≽ ⊏ ⊐ ⊑ ⊒ ² ³ °
∠ ∟ ° ≅ ~ ‖ ⟂ ⫛
≡ ≜ ≈ ∝ ∞ ≪ ≫ ⌊⌋ ⌈⌉ ∘ ∏ ∐ ∑ ∧ ∨ ∩ ∪ ⨀ ⊕ ⊗ 𝖕 𝖖 𝖗 ⊲ ⊳
∅ ∖ ∁ ↦ ↣ ∩ ∪ ⊆ ⊂ ⊄ ⊊ ⊇ ⊃ ⊅ ⊋ ⊖ ∈ ∉ ∋ ∌ ℕ ℤ ℚ ℝ ℂ ℵ ℶ ℷ ℸ 𝓟
¬ ∨ ∧ ⊕ → ← ⇒ ⇐ ⇔ ∀ ∃ ∄ ∴ ∵ ⊤ ⊥ ⊢ ⊨ ⫤ ⊣ … ⋯ ⋮ ⋰ ⋱
∫ ∬ ∭ ∮ ∯ ∰ ∇ ∆ δ ∂ ℱ ℒ ℓ
𝛢𝛼 𝛣𝛽 𝛤𝛾 𝛥𝛿 𝛦𝜀𝜖 𝛧𝜁 𝛨𝜂 𝛩𝜃𝜗 𝛪𝜄 𝛫𝜅 𝛬𝜆 𝛭𝜇 𝛮𝜈 𝛯𝜉 𝛰𝜊 𝛱𝜋 𝛲𝜌 𝛴𝜎𝜍 𝛵𝜏 𝛶𝜐 𝛷𝜙𝜑 𝛸𝜒 𝛹𝜓 𝛺𝜔