meta poly for Cxxx ?

[cubaq]

I am working on C156,and at least two other C156 are in line (none of them are special form).
I am using yafu without Msieve (ini file below).
Is it possible to use some existing poly dedicated for C156 (I.E. this one) and decrease sieve time
instead of spending another ~35 CPU-hours to find one.
If not - how it will fail: at once, during sieve, atlinear, or (not this one please) in sqrt stage.
If yes - how to create nfs.dat.p entry when values of norm,alpha and rroots are unknown; I would create nfs.job and nfs.fb myself but why the poly's in those files are different.


Thanks in advance,

cubaq

PS. yafu.ini:
B1ecm=11000
B1pm1=100000
B1pp1=20000
rhomax=1000
threads=8
pretest_ratio=0.25
%ggnfs_dir=D:\ggnfs\
ggnfs_dir=D:/ggnfs/
R=1
%ecm_path=..\Gmp-Ecm\ecm.exe
ecm_path=..\Gmp-Ecm\ecm
tune_info= Intel(R) Xeon(R) CPU E5-4650 0 @2.70GHz,LINUX64,1.73786e-05,0.200412,0.400046,0.0987873,98.8355,2699.98


session.log (partial)
01/31/19 11:53:07 v1.34.5 @ WLODEK-KOMPUTER,System/Build Info:
Using GMP-ECM 6.3, Powered by GMP 5.1.1
cached 78498 primes. pmax = 999983
detected Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz
detected L1 = 32768 bytes, L2 = 8388608 bytes, CL = 64bytes
measured cpu frequency ~= 3565.667100
using 20 random witnesses for Rabin-Miller PRP checks

[firejuggler]

Unfortunately, poly are composite specific. So it will crash imediately.
As for finding a poly, you can ask for one in this thread

https://mersenneforum.org/showthread.php?t=18368 The trouble is that it will take around 24H to get a 'decent' usable poly. As for how to create nfs.dat.p and the other file, I do not know.

[VBCurtis]

Pretest ratio of 0.25 is really low; for a C156 you ought to do at *minimum* a T45 worth of ECM. I'd do half a T50 before GNFS, personally. (Or maybe I misunderstand what pretest ratio is)

That's about a 6-desktop-core-week job to run through GNFS; spending 1.5 days on poly select is likely about right. If you get a score within 10% of the record poly score for that size, it's "good enough". If you don't have a GPU to use with msieve, the CADO tools produce very nice polys from CPU searches.

Trying to use a poly from a different input will crash GNFS immediately with error "poly does not divide N!" or similar.

[cubaq]

Thank You both for reply


@firejuggler:

While using 'other' poly will crash, that knowlege seems to be pointless.




@VBCurtis:

Pretest ratio does not control upper limit of ECM stage. From the other hand, I am using ecm directly (yes up to ~T50 for C156).

How do You measure that kind of percentage (i.e. i have e 1.952e-012 for C156 - is it good enough?).
As far CADO I've just start to learn "How To …" so it'l take some weeks: do I really may ask for proper poly for C156,C157 on forum?


Thanks once more,


cubaq

[cubaq]

@firejuggler:



I am terrible sorry, I've just realize how that may be read. I ment the last sentence of Your reply. Sorry once more.


cubaq.

[VBCurtis]

Originally Posted by cubaq

How do You measure that kind of percentage (i.e. i have e 1.952e-012 for C156 - is it good enough?).
As far CADO I've just start to learn "How To …" so it'l take some weeks: do I really may ask for proper poly for C156,C157 on forum?

The record for a C156 is 2.893e-12; for C157 it's 2.776e-12. That means 2.5 to 2.6e-12 should be reachable with a day or two of GPU-msieve.

The poly you've found would take roughly 25% longer to sieve than one with the score I've estimated (the ratio of the scores is roughly the inverse ratio of sieving times). It's worth spending a few days on poly select to save ~1.5 core-weeks of sieving!