Tweaking polynomial search for C197

[Batalov]

An incremental improvement; still a long way to go to catch up with the 5th degree:
# norm 2.027687e-14 alpha -10.660168 e 4.031e-15 rroots 4
skew: 18649788.79
c0: -1054693895229737111542912875645712585798874585840
c1: -648953488177000550863606724453595751154052
c2: 46873012550011122605269225426938000
c3: 5344436734176828120536762249
c4: -178225159341792183379
c5: -5869354067069
c6: 205716
Y0: -81698919261232456726361244988593
Y1: 4025080561718281307

[em99010pepe]

I never used msieve to find a poly, what's the flag? How I start the search? I will try it on 4 cores...

[bdodson]

Originally Posted by em99010pepe

I never used msieve to find a poly, what's the flag? How I start the search? I will try it on 4 cores...

I'm using a minor variation of the instructions Jason gives on the RSA768
thread. -Bruce

[em99010pepe]

Originally Posted by bdodson

I'm using a minor variation of the instructions Jason gives on the RSA768
thread. -Bruce

The command worked there but not here...

[jasonp]

Originally Posted by fivemack

Making an msieve.dat.m file containing only that line and running -np2 on it with 3e29/3e-15 got a 3.629e-15 hit in 55 seconds. Going all the way to stage2_norm=1e30 gave only a 3.249e-15, in 187 seconds; are some internal buckets getting full, and can they be enlarged?

Missed this question...

There are a lot of internal buckets in the root sieve, because it's possible to spend weeks on a single stage 1 hit if you're not careful. For degree 5 the number of lines given to the actual sieve is controlled by XLINE_HEAP_SIZE in gnfs/poly/stage2/root_sieve_deg45_x.c

If the sieving region is large enough, the code in gnfs/poly/stage2/root_sieve_deg5_xy.c restricts the choice of lines to those that have the highest root score modulo the smallest few p^k (not even the second highest; for degree 5 the limit on p^k here is quite small, so I suspect there isn't much advantage to being less restrictive).

[em99010pepe]

Here's my msieve output. What am I doing wrong?

msieve0.exe -v -l msieve.log -nf msieve.log -s msieve.fb -s msieve.dat -np1 50000,60000

Msieve v. 1.48
Sat May 07 17:26:48 2011
random seeds: bd90aeec 73de0051
factoring 6117378187980081398706225420896908215238102943841526255679961994389507
96154227403439943437705346896479335277911619430806081130583094865607915507328093
40767016424578028793088881479019117986214217881 (197 digits)
searching for 15-digit factors
commencing number field sieve (197-digit input)
warning: factor base file uninitialized
commencing number field sieve polynomial selection
error: no parameters for 197 digit inputs

[jasonp]

The current code has no parameters for 195- or 200-digit degree-5 inputs, and there's currently no way to add them without having to recompile.

[chris2be8]

Originally Posted by em99010pepe

Here's my msieve output. What am I doing wrong?

msieve0.exe -v -l msieve.log -nf msieve.log -s msieve.fb -s msieve.dat -np1 50000,60000

Are the parameters correct? Unless you had a cut and paste error you are using msieve.log for both log (-l) and factor base (-nf). And you have -s twice so I don't know where any relations would go.

The following might be what you intended:
msieve0.exe -v -l msieve.log -nf msieve.fb -s msieve.dat -np1 50000,60000

Chris K

[fivemack]

Originally Posted by em99010pepe

Here's my msieve output. What am I doing wrong?

Please download the most recent SVN, add

	{195, 1.00E+030, 1.00E+029, 2.00E-015},
	{197, 1.00E+030, 1.00E+029, 2.00E-015},

to the end of the definition of prebuilt_params_deg5 in gnfs/poly/poly_skew.c

and recompile. It should then work.

[fivemack]

Bruce Dodson has run one pass over 5000000 - 6000000.

Best score is 5.430e-15 from

1.56274e+27 5439060 2166888684226480343737 102378390127366473853535596376254933061

(that hit has the fourth best stage1 score out of 300,000 or so for 5300e3 .. 5500e3 ... if anyone wants to try attacking it with different stage 2 parameters and see if they can get a better stage 2 hit, fine by me! )

I see that 6000000 - 6800000 has appeared, so I'll run stage2 over that next.

I've finished stage1 on 2000000 - 3000000 but haven't quite finished stage 2.

[fivemack]

I've finished one pass over 2100000-3000000

Best is 5.500e-15 from

2.28335e+27 2885580 2513149667468253200263 116216138797108534231486454637868571506

18th-best of 3300 stage-1 hits that went to stage 2. Looking as if I could cut off at 5e27 rather than 1.5e28, but only worth doing that if I have a way of usefully spending longer on stage 2.

6M-7M done by bdodson and currently going through stage 2; 3M-4M stage 1 currently running at my machine.