GMP-ECM 7.0.5-dev Revision3092 - maxmem not working - Page 2

WraithX · 2022-01-17, 15:28

Ok, I finally tracked it down to gmp-ecm's configure script not recognizing your environment as WIN64 due to your mingw64 in your --build=x86_64-pc-mingw64. I have updated the code to recognize mingw64 as a WIN64 target for the mulredc asm.

Could you try it now and see if it works? The fixed commit is 932a2fe8

ATH · 2022-01-17, 19:30

Quote:

Originally Posted by WraithX

Ok, I finally tracked it down to gmp-ecm's configure script not recognizing your environment as WIN64 due to your mingw64 in your --build=x86_64-pc-mingw64. I have updated the code to recognize mingw64 as a WIN64 target for the mulredc asm.932a2fe8

Thanks, it works now :) and the memory works as well, close enough:
Using lmax = 8388608 with NTT which takes about 13440MB of memory
Peak memory usage: 13635MB

"--build=x86_64-pc-mingw64" is what ./config.guess suggests.
I'm 99% sure though that I tried "--build=x86_64-w64-mingw32" as well yesterday, does that need to be changed as well to be recognized as WIN64?

Maybe "svn://scm.gforge.inria.fr/svn/ecm/trunk" should be shut down now if it is no longer the latest version?

Edit:
Btw in an older version of MSYS2, I get no warnings on "autoreconf -i", so maybe the newest version of autoreconf or libtool is more pedantic? But it works despite of the warnings.

WraithX · 2022-01-17, 20:17

Great! Glad to hear it's working for you.

No extra changes needed, the change I made will treat both x86_64-*-mingw32 and x86_64-*-mingw64 as Win64.

I'll mention to Paul about the old svn site, but it will be up to him if/when to take it down.

Yeah, I was thinking it may be newer autoconf tools causing those warning messages. I'll mention this to Paul as well to see how he wants to handle it.

Thanks again for pointing this out and helping test the fix!

ATH · 2022-01-25, 18:45

Ok, I have a new issue with this binary I compiled, this time with P+1.
First issue is "cosmetic" only but still annoying, it shows huge random k values, but it still runs with only a small normal k-value:
P = 5870865, l = 4194304, s_1 = 816045813760, k = s_2 = 816043786241, m_1 = 8516629
P = 11696685, l = 4194304, s_1 = 523988083712, k = s_2 = 523986010114, m_1 = 4274711

When I use -k 1 or -k 2 instead of -maxmem it still works and still shows a huge k value, but now it is not showing estimated RAM usage before, is that intentional?

Code:

ecm.exe -v -pp1 -maxmem 16000 -resume M4933pp606a.txt 606e8 100e12-120e12 < M4933.txt
GMP-ECM 7.0.5-dev [configured with GMP 6.2.1, --enable-asm-redc] [P+1]
Resuming P+1 residue saved by ATH@5960X with GMP-ECM 7.0.5-dev on Tue Jan 25 05:28:43 2022
Input number is (2^4933-1)/29599/3637821119/25117652081/35368771391/169745595529/90043391515951/740782867588969631/359805386252506578683471/47332662224988461128710060020435516120344835557993/1552544580255765059978618542093161661106227966313804593681333703 (1271 digits)
Using lmax = 4194304 with one pass NTT which takes about 13440MB of memory
Using B1=60600000000-60600000000, B2=99999999596844-125441709555816, polynomial x^1
P = 5870865, l = 4194304, s_1 = 816045813760, k = s_2 = 816043786241, m_1 = 8516629
Can't compute success probabilities for B1 <> B2min
Using special division for factor of 2^4933-1
Step 1 took 0ms
Computing F from factored S_1 took 133468ms
Computing h_x and h_y took 46266ms
Computing DCT-I of h_x took 23219ms
Computing DCT-I of h_y took 23000ms
Multi-point evaluation 1 of 1:
Computing g_x and g_y took 166281ms
Computing forward NTT of g_x took 23641ms
Computing forward NTT of g_y took 23687ms
Adding and computing inverse NTT of sum took 25766ms
Computing gcd of coefficients and N took 29922ms
Step 2 took 495891ms
Peak memory usage: 13633MB


ecm.exe -v -pp1 -maxmem 16000 -resume M4933pp606a.txt 606e8 100e12-130e12 < M4933.txt
GMP-ECM 7.0.5-dev [configured with GMP 6.2.1, --enable-asm-redc] [P+1]
Resuming P+1 residue saved by ATH@5960X with GMP-ECM 7.0.5-dev on Tue Jan 25 05:28:43 2022
Input number is (2^4933-1)/29599/3637821119/25117652081/35368771391/169745595529/90043391515951/740782867588969631/359805386252506578683471/47332662224988461128710060020435516120344835557993/1552544580255765059978618542093161661106227966313804593681333703 (1271 digits)
Using lmax = 4194304 with one pass NTT which takes about 13440MB of memory
Using B1=60600000000-60600000000, B2=99999979989968-149610225474652, polynomial x^1
P = 11696685, l = 4194304, s_1 = 523988083712, k = s_2 = 523986010114, m_1 = 4274711
Can't compute success probabilities for B1 <> B2min
Using special division for factor of 2^4933-1
Step 1 took 0ms
Computing F from factored S_1 took 136703ms
Computing h_x and h_y took 46844ms
Computing DCT-I of h_x took 23047ms
Computing DCT-I of h_y took 23265ms
Multi-point evaluation 1 of 2:
Computing g_x and g_y took 165985ms
Computing forward NTT of g_x took 23750ms
Computing forward NTT of g_y took 23828ms
Adding and computing inverse NTT of sum took 25890ms
Computing gcd of coefficients and N took 29469ms
Multi-point evaluation 2 of 2:
Computing g_x and g_y took 166031ms
Computing forward NTT of g_x took 23719ms
Computing forward NTT of g_y took 23688ms
Adding and computing inverse NTT of sum took 25812ms
Computing gcd of coefficients and N took 29391ms
Step 2 took 768031ms
Peak memory usage: 13633MB








ecm.exe -v -pp1 -k 1 -resume M4933pp606a.txt 606e8 100e12-125e12 <M4933.txt
GMP-ECM 7.0.5-dev [configured with GMP 6.2.1, --enable-asm-redc] [P+1]
Resuming P+1 residue saved by ATH@5960X with GMP-ECM 7.0.5-dev on Tue Jan 25 05:28:43 2022
Input number is (2^4933-1)/29599/3637821119/25117652081/35368771391/169745595529/90043391515951/740782867588969631/359805386252506578683471/47332662224988461128710060020435516120344835557993/1552544580255765059978618542093161661106227966313804593681333703 (1271 digits)
Using B1=60600000000-60600000000, B2=99999999596844-125441709555816, polynomial x^1
P = 5870865, l = 4194304, s_1 = 833225682944, k = s_2 = 833223655425, m_1 = 8516629
Can't compute success probabilities for B1 <> B2min
Using special division for factor of 2^4933-1
Step 1 took 0ms
Computing F from factored S_1 took 133016ms
Computing h_x and h_y took 45625ms
Computing DCT-I of h_x took 23172ms
Computing DCT-I of h_y took 22938ms
Multi-point evaluation 1 of 1:
Computing g_x and g_y took 166078ms
Computing forward NTT of g_x took 23812ms
Computing forward NTT of g_y took 23672ms
Adding and computing inverse NTT of sum took 25875ms
Computing gcd of coefficients and N took 29797ms
Step 2 took 494609ms
Peak memory usage: 13633MB

ecm.exe -v -pp1 -k 2 -resume M4933pp606a.txt 606e8 100e12-125e12 <M4933.txt
GMP-ECM 7.0.5-dev [configured with GMP 6.2.1, --enable-asm-redc] [P+1]
Resuming P+1 residue saved by ATH@5960X with GMP-ECM 7.0.5-dev on Tue Jan 25 05:28:43 2022
Input number is (2^4933-1)/29599/3637821119/25117652081/35368771391/169745595529/90043391515951/740782867588969631/359805386252506578683471/47332662224988461128710060020435516120344835557993/1552544580255765059978618542093161661106227966313804593681333703 (1271 digits)
Using B1=60600000000-60600000000, B2=99999979989968-149610225474652, polynomial x^1
P = 11696685, l = 4194304, s_1 = 833225729024, k = s_2 = 833223655426, m_1 = 4274711
Can't compute success probabilities for B1 <> B2min
Using special division for factor of 2^4933-1
Step 1 took 0ms
Computing F from factored S_1 took 135390ms
Computing h_x and h_y took 47313ms
Computing DCT-I of h_x took 23078ms
Computing DCT-I of h_y took 23062ms
Multi-point evaluation 1 of 2:
Computing g_x and g_y took 166563ms
Computing forward NTT of g_x took 23687ms
Computing forward NTT of g_y took 23766ms
Adding and computing inverse NTT of sum took 25891ms
Computing gcd of coefficients and N took 29250ms
Multi-point evaluation 2 of 2:
Computing g_x and g_y took 166656ms
Computing forward NTT of g_x took 23969ms
Computing forward NTT of g_y took 23812ms
Adding and computing inverse NTT of sum took 25985ms
Computing gcd of coefficients and N took 29453ms
Step 2 took 768453ms
Peak memory usage: 13633MB

Gordon · 2022-02-10, 00:47

Quote:

Originally Posted by ATH

It is using more RAM than maxmem-setting, I'm not sure if the number is too big at 1271 digits?

I also tried different -k settings and it also used more than it initially informed.

With: -maxmem 16000:
Using lmax = 8388608 with NTT which takes about 13440MB of memory
Peak memory usage: 23520MB

[snipped quite a bit]

I am doing lots of ECM work on small (sub 5k) candidates.

PC has 64gb ram, if I allow ECM to have say 32gb ram, then when it tries to allocate another 4 gig to take it to more than 16 gig it fails with out of memory, despite there being over 30gb still free.

How do you get it to use 25 gig?

Line 1 from the output file...

GMP-ECM 7.0.5-dev [configured with GMP 6.1.2, --enable-asm-redc] [ECM]

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
CPU working 24/7	BaptisteTesson	Information & Answers	5	2016-08-26 15:41
ssh -X not working?	Dubslow	Linux	3	2012-05-11 14:44
Log Out not working?	cheesehead	Forum Feedback	1	2012-03-19 17:13
DST not working?	Dubslow	Forum Feedback	2	2012-03-19 06:53
How is working on 44M-45M ?	hbock	Lone Mersenne Hunters	0	2005-04-06 17:16

2022-01-17, 15:28	#12
WraithX Mar 2006 3·173 Posts	Ok, I finally tracked it down to gmp-ecm's configure script not recognizing your environment as WIN64 due to your mingw64 in your --build=x86_64-pc-mingw64. I have updated the code to recognize mingw64 as a WIN64 target for the mulredc asm. Could you try it now and see if it works? The fixed commit is 932a2fe8

2022-01-17, 20:17	#14
WraithX Mar 2006 3×173 Posts	Great! Glad to hear it's working for you. No extra changes needed, the change I made will treat both x86_64--mingw32 and x86_64--mingw64 as Win64. I'll mention to Paul about the old svn site, but it will be up to him if/when to take it down. Yeah, I was thinking it may be newer autoconf tools causing those warning messages. I'll mention this to Paul as well to see how he wants to handle it. Thanks again for pointing this out and helping test the fix!