Error with large number on 32-bit machines

EdH · 2013-01-13, 18:02

How disappointing!

On my 32-bit machines (both antiX and Ubuntu), on which I compiled GMP and ECM, I am receiving the following error:

Code:

$ ecm -c 10 -v 43000000 < ecmIn
GMP-ECM 6.4.3 [configured with GMP 5.0.5] [ECM]
Running on math03
Input number is 16314315482963226490821946080971549450706222359508882207070502298741881754665051765380282665291356409074285196429704092132126707585162244439000837928127809542351956738323750370087594623357586179416038045807359853 (212 digits)
Using MODMULN [mulredc:4, sqrredc:4]
ecm: symbol lookup error: ecm: undefined symbol: __gmpn_add_nc

They work with a smaller input number:

Code:

$ ecm -c 10 -v 43000000 < ecmIn
GMP-ECM 6.4.3 [configured with GMP 5.0.5] [ECM]
Running on math02
Input number is 326286309659264529816438921619430989014124447190177644141410045974837635093301035307605653305827 (96 digits)
Using MODMULN [mulredc:0, sqrredc:0]
Using B1=43000000, B2=240490660426, polynomial Dickson(12), sigma=3200203337
dF=65536, k=5, d=690690, d2=17, i0=46
Expected number of curves to find a factor of n digits:
35    40    45    50    55    60    65    70    75    80
55    246    1286    7557    49831    361851    2844041    2.4e+07    2.2e+08    2.2e+09
Step 1 took 542085ms
********** Factor found in step 1: 86635039
Found composite factor of  8 digits: 86635039
Composite cofactor 3766216457284269587694638443221927666173549563360589520152579904475343227966931893544893 has 88 digits

A 32-bit Ubuntu machine using a repository version seems to be OK with the larger number:

Code:

$ ecm -c 10 -v 43000000 < ecmIn
GMP-ECM 6.3 [configured with GMP 5.0.2] [ECM]
Running on math01
Input number is 16314315482963226490821946080971549450706222359508882207070502298741881754665051765380282665291356409074285196429704092132126707585162244439000837928127809542351956738323750370087594623357586179416038045807359853 (212 digits)
Using MODMULN
Using B1=43000000, B2=240490660426, polynomial Dickson(12), sigma=3523070894
dF=65536, k=5, d=690690, d2=17, i0=46
Expected number of curves to find a factor of n digits:
35    40    45    50    55    60    65    70    75    80
55    246    1286    7557    49831    361851    2844041    2.4e+07    2.2e+08    2.2e+09

Any thoughts are welcome...

xilman · 2013-01-13, 18:35

Quote:

Originally Posted by EdH

How disappointing!

On my 32-bit machines (both antiX and Ubuntu), on which I compiled GMP and ECM, I am receiving the following error:

Code:

$ ecm -c 10 -v 43000000 < ecmIn
GMP-ECM 6.4.3 [configured with GMP 5.0.5] [ECM]
Running on math03
Input number is 16314315482963226490821946080971549450706222359508882207070502298741881754665051765380282665291356409074285196429704092132126707585162244439000837928127809542351956738323750370087594623357586179416038045807359853 (212 digits)
Using MODMULN [mulredc:4, sqrredc:4]
ecm: symbol lookup error: ecm: undefined symbol: __gmpn_add_nc

They work with a smaller input number:

Code:

$ ecm -c 10 -v 43000000 < ecmIn
GMP-ECM 6.4.3 [configured with GMP 5.0.5] [ECM]
Running on math02
Input number is 326286309659264529816438921619430989014124447190177644141410045974837635093301035307605653305827 (96 digits)
Using MODMULN [mulredc:0, sqrredc:0]
Using B1=43000000, B2=240490660426, polynomial Dickson(12), sigma=3200203337
dF=65536, k=5, d=690690, d2=17, i0=46
Expected number of curves to find a factor of n digits:
35    40    45    50    55    60    65    70    75    80
55    246    1286    7557    49831    361851    2844041    2.4e+07    2.2e+08    2.2e+09
Step 1 took 542085ms
********** Factor found in step 1: 86635039
Found composite factor of  8 digits: 86635039
Composite cofactor 3766216457284269587694638443221927666173549563360589520152579904475343227966931893544893 has 88 digits

A 32-bit Ubuntu machine using a repository version seems to be OK with the larger number:

Code:

$ ecm -c 10 -v 43000000 < ecmIn
GMP-ECM 6.3 [configured with GMP 5.0.2] [ECM]
Running on math01
Input number is 16314315482963226490821946080971549450706222359508882207070502298741881754665051765380282665291356409074285196429704092132126707585162244439000837928127809542351956738323750370087594623357586179416038045807359853 (212 digits)
Using MODMULN
Using B1=43000000, B2=240490660426, polynomial Dickson(12), sigma=3523070894
dF=65536, k=5, d=690690, d2=17, i0=46
Expected number of curves to find a factor of n digits:
35    40    45    50    55    60    65    70    75    80
55    246    1286    7557    49831    361851    2844041    2.4e+07    2.2e+08    2.2e+09

Any thoughts are welcome...

When you installed GMP and ECM did you run "make check" as recommended?

WraithX · 2013-01-13, 19:19

Quote:

Originally Posted by EdH

On my 32-bit machines (both antiX and Ubuntu), on which I compiled GMP and ECM, I am receiving the following error:

Code:

$ ecm -c 10 -v 43000000 < ecmIn
<snip>
ecm: symbol lookup error: ecm: undefined symbol: __gmpn_add_nc

Any thoughts are welcome...

Can you give us the output of "ecm -printconfig". I'm curious to see if you have the following line:
HAVE___GMPN_ADD_NC = 1
Also, if you could post a zipped up copy of your config.log, that might help us track this down.

Now, I'm not sure what would cause this problem, but, since you are building these things yourself, could you try the following as a temporary work-around:
1) In your ecm directory run "make clean"
2) Then run your "./configure" again
3) Then, go into the config.h and if you find the following line:
#define HAVE___GMPN_ADD_NC 1
replace it with:
/* #define HAVE___GMPN_ADD_NC 1 */
4) Then run "make" and "make check" to be sure those complete successfully
5) Then try your number again and let us know the results

Hopefully we can figure this out.

ET_ · 2013-01-13, 20:15

Quote:

Originally Posted by WraithX

Can you give us the output of "ecm -printconfig". I'm curious to see if you have the following line:
HAVE___GMPN_ADD_NC = 1
Also, if you could post a zipped up copy of your config.log, that might help us track this down.

Now, I'm not sure what would cause this problem, but, since you are building these things yourself, could you try the following as a temporary work-around:
1) In your ecm directory run "make clean"
2) Then run your "./configure" again
3) Then, go into the config.h and if you find the following line:
#define HAVE___GMPN_ADD_NC 1
replace it with:
/* #define HAVE___GMPN_ADD_NC 1 */
4) Then run "make" and "make check" to be sure those complete successfully
5) Then try your number again and let us know the results

Hopefully we can figure this out.

Hey, that workatound resolved MY compile problem with
#define HAVE___GMPN_REDC_1 1

Thank you!

Luigi

WraithX · 2013-01-13, 22:36

Quote:

Originally Posted by ET_

Hey, that workatound resolved MY compile problem with
#define HAVE___GMPN_REDC_1 1

Thank you!

Luigi

Hmmm, if so then that means that ./configure might be missing something or messing something up. Can you post a zipped up copy of your config.log so we can see what might be going wrong?

ET_ · 2013-01-13, 23:22

Quote:

Originally Posted by WraithX

Hmmm, if so then that means that ./configure might be missing something or messing something up. Can you post a zipped up copy of your config.log so we can see what might be going wrong?

Here it is!

I can add that all the tests done with make check were ok after the update.

Luigi

EdH · 2013-01-14, 03:11

Thanks, everyone!

@xilman: Yes!

Code:

...
NOTE: NEXT TEST WILL FAIL ON 32BIT MACHINES, THIS IS EXPECTED.
GMP-ECM 6.4.3 [configured with GMP 5.0.5] [P+1]
Input number is 8589934621 (10 digits)
Using B1=4294967310-4294967311, B2=1, polynomial x^1
Step 1 took 48ms
********** Factor found in step 1: 8589934621
Found input number N
GMP-ECM 6.4.3 [configured with GMP 5.0.5] [P+1]
Input number is 6054018161*10^400+417727253109 (410 digits)
Using B1=2000, B2=2352760, polynomial x^1, x0=4
./ecm: symbol lookup error: ./ecm: undefined symbol: __gmpn_add_nc
############### ERROR ###############
Expected return code 14 but got 127
make: *** [check] Error 1

I took the NOTE above to mean the error I received.

@WraithX:

Code:

$ ecm -printconfig
Compilation options:
Included GMP header files version 5.0.5
GWNUM_VERSION undefined
HAVE_SSE2 = 1
$HAVE___GMPN_ADD_NC = 1
HAVE___GMPN_MOD_34LSUB1 = 1
HAVE___GMPN_REDC_1 = 1
MEMORY_DEBUG undefined
USE_ASM_REDC undefined
WANT_ASSERT undefined
WANT_SHELLCMD undefined
_OPENMP undefined
MPZMOD_THRESHOLD = 84
REDC_THRESHOLD = 119
MUL_NTT_THRESHOLD = 4096
NTT_GFP_TWIDDLE_DIF_BREAKOVER = 17
NTT_GFP_TWIDDLE_DIT_BREAKOVER = 17
PREREVERTDIVISION_NTT_THRESHOLD = 64
POLYINVERT_NTT_THRESHOLD = 1024
POLYEVALT_NTT_THRESHOLD = 512
MPZSPV_NORMALISE_STRIDE = 2048

That line has a preceding "$" instead of just the line. The config.log is zipped below.

Here is the result of commenting out the "#define HAVE___GMPN_ADD_NC 1" line:

Code:

...
NOTE: NEXT 3 TESTS WILL FAIL ON SOME 32BIT MACHINES, THIS IS EXPECTED.
GMP-ECM 6.4.3 [configured with GMP 5.0.5] [ECM]
Input number is 10000000000000000000000000000000000000121 (41 digits)
Using B1=2950307, B2=4281751120, polynomial Dickson(6), sigma=61
Step 1 took 20229ms
********** Factor found in step 1: 10000000000000000000000000000000000000121
Found input number N
GMP-ECM 6.4.3 [configured with GMP 5.0.5] [ECM]
Input number is 10000000000000000000000000000000000000121 (41 digits)
Using B1=9405629, B2=23422152400, polynomial Dickson(12), sigma=266
Step 1 took 65384ms
********** Factor found in step 1: 10000000000000000000000000000000000000121
Found input number N
GMP-ECM 6.4.3 [configured with GMP 5.0.5] [ECM]
Input number is 10000000000000000000000000000000000000121 (41 digits)
Using B1=4372759, B2=8561918830, polynomial Dickson(6), sigma=291
Step 1 took 30265ms
********** Factor found in step 1: 10000000000000000000000000000000000000121
Found input number N
All ECM tests are ok.

And, trying the 212 digit number again:

Code:

$ ecm -c 10 -v 43000000 < ecmIn
GMP-ECM 6.4.3 [configured with GMP 5.0.5] [ECM]
Running on math02
Input number is 16314315482963226490821946080971549450706222359508882207070502298741881754665051765380282665291356409074285196429704092132126707585162244439000837928127809542351956738323750370087594623357586179416038045807359853 (212 digits)
Using MODMULN [mulredc:4, sqrredc:4]
Using B1=43000000, B2=240490660426, polynomial Dickson(12), sigma=2155976667
dF=65536, k=5, d=690690, d2=17, i0=46
Expected number of curves to find a factor of n digits:
35    40    45    50    55    60    65    70    75    80
55    246    1286    7557    49831    361851    2844041    2.4e+07    2.2e+08    2.2e+09

It appears to be working, but I suppose I'll have to wait quite some time for a result to show up.

WraithX · 2013-01-15, 03:11

Quote:

Originally Posted by EdH

That line has a preceding "$" instead of just the line. The config.log is zipped below.

It appears to be working, but I suppose I'll have to wait quite some time for a result to show up.

@EdH
Did you put the "$" in front of it? I don't see anything in the code that would print out a "$" in front of that line.

I'm wondering if there might be multiple gmp.h or libgmp.a or libgmp.so* files on your computer? We can try to find out by seeing what all paths are defined in your environment with the following command:
env | grep -i path
(At this point I'm not sure how to search a PATH [or LD_LIBRARY_PATH] variable for a certain file, any linux gurus know???)

Then, we need to find out what library your binary is linking to. Can you run ldd on your ecm binary? Like so:
ldd ecm
We're interested in the line with libgmp.so, it could look like:
libgmp.so.10 => /usr/local/lib/libgmp.so.10
Once you have that, can you try running nm on the lib from above? Like:
nm /usr/local/lib/libgmp.so.10 | grep -i add_nc

Theoretically, that might come up empty. My guess at that point will be that you have another libgmp (.a, .so), perhaps in /usr/lib/. If so, could you try the nm+grep command on that libgmp too? I think one will have the add_nc, and the other won't. Let's see where this gets us.

@ET_
What problem were you running into, specifically? If you're using a recent GMP (5.0+) that you've compiled yourself, it should have mpn_redc_1. Was GMP-ECM not compiling for you? Or were you getting errors while running? What version of GMP are you using? Is it pre-compiled or did you build it yourself? I'd like to fix any problems rather than spread the above "work-around" to too many people.

Dubslow · 2013-01-15, 03:52

A better command might be `locate libgmp.a libgmp.so`, and then comparing that to what ldd says.

EdH · 2013-01-15, 06:07

Quote:

Originally Posted by WraithX

@EdH
Did you put the "$" in front of it? I don't see anything in the code that would print out a "$" in front of that line.

Nope! That's how it appeared when I opened it.

Quote:

Originally Posted by WraithX

I'm wondering if there might be multiple gmp.h or libgmp.a or libgmp.so* files on your computer? We can try to find out by seeing what all paths are defined in your environment with the following command:
env | grep -i path
(At this point I'm not sure how to search a PATH [or LD_LIBRARY_PATH] variable for a certain file, any linux gurus know???)

Code:

$ env | grep -i path
PATH=/usr/local/bin:/usr/bin:/bin:/usr/local/games:/usr/games

Quote:

Originally Posted by WraithX

Then, we need to find out what library your binary is linking to. Can you run ldd on your ecm binary? Like so:
ldd ecm

Code:

$ ldd ecm
ldd: ./ecm: No such file or directory

Code:

$ whereis libecm
libecm: /usr/local/lib/libecm.a /usr/local/lib/libecm.la

Code:

$ whereis libgmp
libgmp: /usr/local/lib/libgmp.so /usr/local/lib/libgmp.la /usr/local/lib/libgmp.a

Quote:

Originally Posted by WraithX

We're interested in the line with libgmp.so, it could look like:
libgmp.so.10 => /usr/local/lib/libgmp.so.10
Once you have that, can you try running nm on the lib from above? Like:
nm /usr/local/lib/libgmp.so.10 | grep -i add_nc

Code:

$ nm /usr/local/lib/libgmp.so | grep -i add_nc
00021340 T __gmpn_add_nc

Code:

$ nm /usr/local/lib/libgmp.la | grep -i add_nc
nm: /usr/local/lib/libgmp.la: File format not recognized

Code:

$ nm /usr/local/lib/libgmp.a | grep -i add_nc
00000000 T __gmpn_add_nc
         U __gmpn_add_nc
         U __gmpn_add_nc
         U __gmpn_add_nc
         U __gmpn_add_nc

Quote:

Originally Posted by WraithX

Theoretically, that might come up empty. My guess at that point will be that you have another libgmp (.a, .so), perhaps in /usr/lib/. If so, could you try the nm+grep command on that libgmp too? I think one will have the add_nc, and the other won't. Let's see where this gets us.

Here's the locate command, but ldd didn't work (above), so there's nothing to compare:

Code:

$ locate libgmp
/home/math02/Math/ggnfs/src/lasieve4/libgmp-aux.a
/home/math02/Math/gmp/libgmp.la
/home/math02/Math/gmp/.libs/libgmp.a
/home/math02/Math/gmp/.libs/libgmp.la
/home/math02/Math/gmp/.libs/libgmp.lai
/home/math02/Math/gmp/.libs/libgmp.so
/home/math02/Math/gmp/.libs/libgmp.so.10
/home/math02/Math/gmp/.libs/libgmp.so.10.0.5
/usr/lib/i386-linux-gnu/libgmp.so.10
/usr/lib/i386-linux-gnu/libgmp.so.10.0.5
/usr/lib/i386-linux-gnu/openssl-1.0.0/engines/libgmp.so
/usr/local/lib/libgmp.a
/usr/local/lib/libgmp.la
/usr/local/lib/libgmp.so
/usr/local/lib/libgmp.so.10
/usr/local/lib/libgmp.so.10.0.5
/usr/share/doc/libgmp10
/usr/share/doc/libgmp10/README.Debian
/usr/share/doc/libgmp10/TODO.Debian
/usr/share/doc/libgmp10/changelog.Debian.gz
/usr/share/doc/libgmp10/copyright
/var/lib/dpkg/info/libgmp10:i386.list
/var/lib/dpkg/info/libgmp10:i386.md5sums
/var/lib/dpkg/info/libgmp10:i386.postinst
/var/lib/dpkg/info/libgmp10:i386.postrm
/var/lib/dpkg/info/libgmp10:i386.shlibs

Thanks for all the help...

ET_ · 2013-01-15, 07:55

Quote:

Originally Posted by WraithX

@ET_
What problem were you running into, specifically? If you're using a recent GMP (5.0+) that you've compiled yourself, it should have mpn_redc_1. Was GMP-ECM not compiling for you? Or were you getting errors while running? What version of GMP are you using? Is it pre-compiled or did you build it yourself? I'd like to fix any problems rather than spread the above "work-around" to too many people.

IIRC, there had been a "duplicate definition" issue between ecm-gmp.h and gmp.h while making the source, after [auto]configurations ran ok.

I was using MPIR 2.5.1 compiled to use GMP 5.0.2 configuration files. Yes I configured and compiled it [by] myself.

I already uploaded the config.log file to the forum, on another subthread of this section.

Luigi

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Error Prone Machines	PageFault	Data	17	2012-04-10 01:40
Scalability of Glucas on large machines: A future //ed prime 95 ?	T.Rex	Software	29	2006-12-13 21:40
Start and Stop Prime 95 on Large Groups of Windows XP Machines	MarcGetty	Software	3	2006-03-07 07:54
Putting prime 95 on a large number of machines	moo	Software	10	2004-12-15 13:25
Team_Prime_Rib error-prone machines	GP2	Data	10	2003-10-05 18:34

2013-01-15, 03:52	#9
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 3×29×83 Posts	A better command might be `locate libgmp.a libgmp.so`, and then comparing that to what ldd says.