GMP-ECM 7.0-dev assertion failed!

[Antonio]

Was working my way through factoring 85*2^67692-1 when this occurred: -

F:\ECM>
F:\ECM>ecm -v -pm1 -maxmem 12288 -inp num2ecm.txt 500000
GMP-ECM 7.0-dev [configured with GMP 6.0.0, --enable-asm-redc] [P-1]
Input number is (85*2^67692-1)/(3*7*179*486271481*563592288920887*6600048163432517) (20337 digits)
Using REDC
Using lmax = 16384 with NTT which takes about 409MB of memory
Using B1=500000, B2=658722972, polynomial x^1, x0=2500857976
P = 39585, l = 16384, s_1 = 8064, k = s_2 = 2, m_1 = 3
Probability of finding a factor of n digits:
(Use -go parameter to specify known factors in P-1)
20      25      30      35      40      45      50      55      60      65
0.14    0.026   0.0034  0.00036 3e-005  2.1e-006        1.3e-007        6.9e-009        3.1e-010        1.3e-011
Step 1 took 213565ms
Computing F from factored S_1Assertion failed!
Program: F:\ECM\ecm.exe
File: pm1fs2.c, Line 1857
Expression: (__builtin_constant_p (1UL) && (1UL) == 0 ? ((F[deg])->_mp_size < 0 ? -1 : (F[deg])->_mp_size > 0) : __gmpz_cmp_ui (F[deg],1UL)) == 0
This application has requested the Runtime to terminate it in an unusual way.
Please contact the application's support team for more information.
F:\ECM>

Since this had worked up to finding the last factor I reran the previous P-1 test: -

F:\ECM>ecm -v -pm1 -maxmem 12288 -inp num2ecm.txt 500000
GMP-ECM 7.0-dev [configured with GMP 6.0.0, --enable-asm-redc] [P-1]
Input number is (85*2^67692-1)/(3*7*179*486271481*563592288920887) (20353 digits)
Using REDC
Using lmax = 16384 with NTT which takes about 409MB of memory
Using B1=500000, B2=658722972, polynomial x^1, x0=1998205761
P = 39585, l = 16384, s_1 = 8064, k = s_2 = 2, m_1 = 3
Probability of finding a factor of n digits:
(Use -go parameter to specify known factors in P-1)
20      25      30      35      40      45      50      55      60      65
0.14    0.026   0.0034  0.00036 3e-005  2.1e-006        1.3e-007        6.9e-009        3.1e-010        1.3e-011
Step 1 took 140697ms
Computing F from factored S_1 took 61449ms
Computing h took 7426ms
Computing DCT-I of h took 780ms
Multi-point evaluation 1 of 2:
Computing g_i took 26427ms
Computing g*h took 1607ms
Computing gcd of coefficients and N took 15958ms
Multi-point evaluation 2 of 2:
Computing g_i took 26427ms
Computing g*h took 1607ms
Computing gcd of coefficients and N took 15943ms
Step 2 took 176094ms

So everything appears to be OK up to removing the last factor found, very strange!


I am running ECM on the number without a problem at the moment.

GMP-ECM 7.0-dev [configured with GMP 6.0.0, --enable-asm-redc] [ECM]
Using B1=50000, B2=15446350, polynomial x^2, 3 threads
____________________________________________________________________________
 Curves Complete |   Average seconds/curve   |    Runtime    |      ETA
-----------------|---------------------------|---------------|--------------
    36 of    220 | Stg1 187.6s | Stg2 42.10s |   0d 00:49:45 |   0d 03:56:40

(Running on Windows 7 sp1, 16GB ram on a i5 3570k)

[xilman]

Originally Posted by Antonio

Was working my way through factoring 85*2^67692-1 when this occurred: -
...
(Running on Windows 7 sp1, 16GB ram on a i5 3570k)

Please report this to the ECM dev team.

Paul

Along with reporting to the ECM development team, what build version were you using please? There seem to have been a few different versions over the last little while.

[Antonio]

Originally Posted by jwaltos

Along with reporting to the ECM development team, what build version were you using please? There seem to have been a few different versions over the last little while.

The original failure was noticed with GMP-ECM 7.0.0 svn2540 configured with GMP 6.0.0

I went back and tried an earlier version with the following results: -

f:\yafu\ecm>ecmx -pm1 -maxmem 12288 -inp num2ecm.txt 110000
GMP-ECM 6.4.4 [configured with MPIR 2.6.0] [P-1]
Input number is (85*2^67692-1)/(3*7*179*486271481*563592288920887*6600048163432517) (20337 digits)
Using B1=110000, B2=44251182, polynomial x^1, x0=452485962
Step 1 took 56628ms
Step 2 took 30062ms
 
 
f:\yafu\ecm>ecmy -pm1 -maxmem 12288 -inp num2ecm.txt 110000
GMP-ECM 6.4.4 [configured with GMP 6.0.0, --enable-asm-redc] [P-1]
Input number is (85*2^67692-1)/(3*7*179*486271481*563592288920887*6600048163432517) (20337 digits)
Using B1=110000, B2=44251182, polynomial x^1, x0=3499163316
Step 1 took 47876ms
Assertion failed!
Program: f:\yafu\ecm\ecmy.exe
File: pm1fs2.c, Line 1851
Expression: (__builtin_constant_p (1UL) && (1UL) == 0 ? ((F[deg])->_mp_size < 0 ? -1 : (F[deg])->_mp_size > 0) : __gmpz_cmp_ui (F[deg],1UL)) == 0
This application has requested the Runtime to terminate it in an unusual way.
Please contact the application's support team for more information.

So it looks like it may be a problem with GMP 6.0.0

You are probably correct. I recall in an earlier post that GMP works best within a Linux environment whereas MPIR works best within Windows.

[wreck]

I met some strange behavior when using B1=3e9 on R323 with -maxmem 1000
Here is the log

GMP-ECM 7.0-dev [configured with MPIR 2.6.0,--enable--openmp] [ECM]
Input number is (10^323-1)/(9*207...) (271 digits)
[Fri Nov 28 17:52:02 2014]
Computing batch product (of zu bits) of primes below B1=0 took -1073741824ms
Using MODMULN [mulredc:0,sqrredc:1]
Using B1=3000000000,B2=84996584711230,polynomial Dickson(30),sigma=2:2092172894349633986
dF=131072,k=438,d=1345890,d2=11,i0=2219
...
Step 1 took 209618ms
Using 32 small primes for NTT
Estimated memory usage: 735M
...
Step 2 took 25928380ms
...

It took 1000 seconds when using B1=11e7, so there seems something wrong of the B1=3e9.

[xilman]

Originally Posted by wreck

I met some strange behavior when using B1=3e9 on R323 with -maxmem 1000
Here is the log

GMP-ECM 7.0-dev [configured with MPIR 2.6.0,--enable--openmp] [ECM]
Input number is (10^323-1)/(9*207...) (271 digits)
[Fri Nov 28 17:52:02 2014]
Computing batch product (of zu bits) of primes below B1=0 took -1073741824ms
Using MODMULN [mulredc:0,sqrredc:1]
Using B1=3000000000,B2=84996584711230,polynomial Dickson(30),sigma=2:2092172894349633986
dF=131072,k=438,d=1345890,d2=11,i0=2219
...
Step 1 took 209618ms
Using 32 small primes for NTT
Estimated memory usage: 735M
...
Step 2 took 25928380ms
...

It took 1000 seconds when using B1=11e7, so there seems something wrong of the B1=3e9.

What took 1000 seconds? Stage 1, stage 2 or the entire computation?

If stage 1 alone took 1000 seconds something indeed looks strange as the subsequent run took only 1/5 as long whereas 3e9/11e7 = 27.3

OTOH, 25928380 / 1000 / 27.3 = 950 which is in line with expectations.

Likewise (25928380 + 209618) / 1000 / 27.3 = 957, also consistent.


Paul

[wreck]

I mean 1000 seconds when using B1=11e7 for stage 1.
I suppose the stage 1 takes longer time when B1 get larger,this is why I think there are something wrong.
The binary is download from gilchrist's site.

[ATH]

Originally Posted by Antonio

So it looks like it may be a problem with GMP 6.0.0

Your number worked for me with SVN2655 both with GMP 6.0.0a and MPIR 2.7.0. You can try these if you want, they are compiled for a Sandy Bridge cpu:

ecm2655-mpir.zip
ecm2655-gmp.zip

[Antonio]

Originally Posted by ATH

Your number worked for me with SVN2655 both with GMP 6.0.0a and MPIR 2.7.0. You can try these if you want, they are compiled for a Sandy Bridge cpu:

ecm2655-mpir.zip
ecm2655-gmp.zip

Thanks - those worked for me too.
Taking the hint, I recompiled and tuned for my Ivy Bridge using the latest version (SVN2657) + GMP 6.0.0a and that's working as well.