A possible bug in LLR/PFGW while using GWNUM (no bug in P95) - Page 5

pinhodecarlos · 2015-01-02, 13:45

Do you guys know if mprime still uses the old RES? If not I made a test on a few candidates that a few months ago had the same issue as posted by "AG5BPilot".
First result of the candidate was using the latest mprime version under linux with an avx processor. Second result for the same candidate was run under windows with avx2.0 processor. RES don't match so I ask "using a slower, more reliable method" means using a different FFT length? Will it cause different RES?

Code:

5*2^4861684-1 is not prime.  RES64: 070B09EDC7FC155B.
5*2^4861684-1 is not prime.  LLR Res64: CB26DEA1F0727791  Time : 4924.949 sec.
5*2^4862454-1 is not prime.  RES64: AFC485010F1E4C9C
5*2^4862454-1 is not prime.  LLR Res64: CC06F5511F59EC44  Time : 4943.288 sec.
5*2^4853964-1 is not prime.  RES64: E19D921AD6DECF13.
5*2^4853964-1 is not prime.  LLR Res64: BD204C54267596C2  Time : 4897.556 sec.
5*2^4855108-1 is not prime.  RES64: F1900AB229F963A8
5*2^4855108-1 is not prime.  LLR Res64: 0A6441632949AA97  Time : 4927.116 sec.
5*2^4847622-1 is not prime.  RES64: 72E5D1B0411002F8
5*2^4847622-1 is not prime.  LLR Res64: 397F6DD1CB898DA3  Time : 4882.750 sec.
5*2^4847650-1 is not prime.  RES64: 9DC9B1C69A7696CF.
5*2^4847650-1 is not prime.  LLR Res64: 382A1A7D2D638E94  Time : 4916.712 sec.

Thank you in advance,
Carlos

AG5BPilot · 2015-01-02, 15:51

Quote:

Originally Posted by pinhodecarlos

Do you guys know if mprime still uses the old RES? If not I made a test on a few candidates that a few months ago had the same issue as posted by "AG5BPilot".

My issue turned out to not be an issue at all -- the non-matching residue was a hardware error on the AMD. The AVX (or FMA) results were correct.

AG5BPilot · 2015-02-12, 12:56

Quote:

Originally Posted by Batalov

No progress since the debug case (yep... holidays!).

I lean even more towards gwtogiant / gmod suspicion. I have some ideas + some from George, but haven't run them.

Any news?

Batalov · 2015-02-12, 17:20

No news.

rogue · 2015-03-23, 22:04

I have run a test with pfgw 3.7.8 (on 1024*3^1877301+1) and can report that I have reproduced the primality test results. -a1 and -a2 yield composite, but -a3 yields prime. I'm going to run a PRP test with the different settings for -a and compare the results. If they do differ, then I will run a test with different values of k to see if that have an impact on the results. With k being a power of 2, I wonder if that has any bearing on the results. I will then change to use gwsquare_carefully for every step.

If the PRP test results are the same regardless of -a, then it could be problem with pfgw, but since LLR has the same problem and uses a different algorithm, it points to gwnum as the possible culprit.

rogue · 2015-03-24, 00:38

Hmmm. The PRP test came back as 3-PRP without specifying -a.

I'm running a test that will do error checking on all steps and I'm running a test with a 32-bit build of pfgw.

Can someone please answer this question? Is LLR exhibiting the same problem on the primality test of this number?

Batalov · 2015-03-24, 01:01

I've written up all that I've gathered at the UTM page (C.C. didn't try to mess with -a values and marked it 'External', so I had to write some explanation). What you wrote agrees with what I observed; in particular PRP test is immune to the bug at default (-a0).

In LLR, I didn't try to force "PRP only" but such option exists afair, so I guess it can be done. I'll try both the latest (.14) and then current (.13) versions.
_______________________

...ok, that's where "FBase=" parameter "goes" in LLR! I figured it out by running
./sllr64 -d -w4 -oForcePRP=1 -oFermatBase=7 -oPBase=5 -oFBase=11 c.npg > & 4/out &
and in that directory ("4/"), the PRP base is 11.
Output:
Starting probable prime test of 1024*3^1877301+1
Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, a = 11
1024*3^1877301+1 is base 11-Strong Fermat PRP! (895704 decimal digits) Time : 5270.521 sec.
Starting Lucas sequence
Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, P = 5, Q = 3
...
In directory with default (FBase=3):
Starting probable prime test of 1024*3^1877301+1
Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, a = 3
1024*3^1877301+1 is base 3-Strong Fermat PRP! (895704 decimal digits) Time : 5347.999 sec.
Starting Lucas sequence
Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, P = 5, Q = 3
...

rogue · 2015-03-24, 11:42

32-bit pfgw returned composite on the N-1 test. The one with error checking every step is still running after 8 hours.

Batalov · 2015-03-24, 16:55

The PRP tests by LLR finished. They are three-fold (strong-Fermat, Lucas and Frobenius) and they all seem to pass (in test bases 3 and 11 and with two FFT sizes AVX FFT length 240K and AVX FFT length 256K)

Code:

Starting probable prime test of 1024*3^1877301+1
Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, a = 11
1024*3^1877301+1 is base 11-Strong Fermat PRP! (895704 decimal digits)  Time : 5270.521 sec.
Starting Lucas sequence
Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, P = 5, Q = 3
1024*3^1877301+1 is strong-Fermat and Lucas PRP, Starting Frobenius test sequence
Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, Q = 3
1024*3^1877301+1 is strong-Fermat, Lucas and Frobenius PRP! (P = 5, Q = 3, D = 13)  Time : 21495.646 sec.
(Factorized part = 100.00%)

# this run with -oFFT_Increment=1
Starting probable prime test of 1024*3^1877301+1
Using all-complex AVX FFT length 256K, Pass1=256, Pass2=1K, a = 11
1024*3^1877301+1 is base 11-Strong Fermat PRP! (895704 decimal digits)  Time : 5303.134 sec.
Starting Lucas sequence
Using all-complex AVX FFT length 256K, Pass1=256, Pass2=1K, P = 5, Q = 3
1024*3^1877301+1 is strong-Fermat and Lucas PRP, Starting Frobenius test sequence
Using all-complex AVX FFT length 256K, Pass1=256, Pass2=1K, Q = 3
1024*3^1877301+1 is strong-Fermat, Lucas and Frobenius PRP! (P = 5, Q = 3, D = 13)  Time : 22297.196 sec.
(Factorized part = 100.00%)

Starting probable prime test of 1024*3^1877301+1
Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, a = 3
1024*3^1877301+1 is base 3-Strong Fermat PRP! (895704 decimal digits)  Time : 5347.999 sec.
Starting Lucas sequence
Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, P = 5, Q = 3
1024*3^1877301+1 is strong-Fermat and Lucas PRP, Starting Frobenius test sequence
Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, Q = 3
1024*3^1877301+1 is strong-Fermat, Lucas and Frobenius PRP! (P = 5, Q = 3, D = 13)  Time : 23277.896 sec.
(Factorized part = 100.00%)

rogue · 2015-03-24, 19:19

Quote:

Originally Posted by rogue

32-bit pfgw returned composite on the N-1 test. The one with error checking every step is still running after 8 hours.

The one with -a1 and error checking returned composite. I'm not certain what to try next.

Batalov · 2015-03-24, 19:50

I may revive the debugging path that I'd dropped around New Year. That is, in short, following on a hunch that something is happening in the very last step, I think. I've attached the almost done computation's savefile (but this is for LLR). You may want to create a similar .pfr file (with some manipulations to instruct the code to dump it exactly at the iteration you want, if needed), and try something similar.

2015-03-24, 01:01	#51
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 19×23² Posts	I've written up all that I've gathered at the UTM page (C.C. didn't try to mess with -a values and marked it 'External', so I had to write some explanation). What you wrote agrees with what I observed; in particular PRP test is immune to the bug at default (-a0). In LLR, I didn't try to force "PRP only" but such option exists afair, so I guess it can be done. I'll try both the latest (.14) and then current (.13) versions. _______________________ ...ok, that's where "FBase=" parameter "goes" in LLR! I figured it out by running ./sllr64 -d -w4 -oForcePRP=1 -oFermatBase=7 -oPBase=5 -oFBase=11 c.npg > & 4/out & and in that directory ("4/"), the PRP base is 11. Output: Starting probable prime test of 10243^1877301+1 Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, a = 11 10243^1877301+1 is base 11-Strong Fermat PRP! (895704 decimal digits) Time : 5270.521 sec. Starting Lucas sequence Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, P = 5, Q = 3 ... In directory with default (FBase=3): Starting probable prime test of 10243^1877301+1 Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, a = 3 10243^1877301+1 is base 3-Strong Fermat PRP! (895704 decimal digits) Time : 5347.999 sec. Starting Lucas sequence Using all-complex AVX FFT length 240K, Pass1=1280, Pass2=192, P = 5, Q = 3 ... Last fiddled with by Batalov on 2015-03-24 at 03:11

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
PFGW 4.0.4 (with gwnum v30.10) Released	rogue	Software	545	2023-01-20 14:12
LLR V3.8.2 using gwnum 26.2 is available!	Jean Penné	Software	25	2010-11-01 15:18
PFGW 3.3.6 or PFGW 3.4.2 Please update now!	Joe O	Sierpinski/Riesel Base 5	5	2010-09-30 14:07
GWNUM?	Unregistered	Information & Answers	3	2010-09-12 19:52
GWNUM as DLL?	Cyclamen Persicum	Software	1	2007-01-02 20:53

2015-02-12, 17:20	#48
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 19·23² Posts	No news.

2015-03-23, 22:04	#49
rogue "Mark" Apr 2003 Between here and the 2⁴×3×5×29 Posts	I have run a test with pfgw 3.7.8 (on 1024*3^1877301+1) and can report that I have reproduced the primality test results. -a1 and -a2 yield composite, but -a3 yields prime. I'm going to run a PRP test with the different settings for -a and compare the results. If they do differ, then I will run a test with different values of k to see if that have an impact on the results. With k being a power of 2, I wonder if that has any bearing on the results. I will then change to use gwsquare_carefully for every step. If the PRP test results are the same regardless of -a, then it could be problem with pfgw, but since LLR has the same problem and uses a different algorithm, it points to gwnum as the possible culprit.

2015-03-24, 00:38	#50
rogue "Mark" Apr 2003 Between here and the 2⁴×3×5×29 Posts	Hmmm. The PRP test came back as 3-PRP without specifying -a. I'm running a test that will do error checking on all steps and I'm running a test with a 32-bit build of pfgw. Can someone please answer this question? Is LLR exhibiting the same problem on the primality test of this number?

2015-03-24, 11:42	#52
rogue "Mark" Apr 2003 Between here and the 2⁴×3×5×29 Posts	32-bit pfgw returned composite on the N-1 test. The one with error checking every step is still running after 8 hours.

2015-03-24, 19:50	#55
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 19·23² Posts	I may revive the debugging path that I'd dropped around New Year. That is, in short, following on a hunch that something is happening in the very last step, I think. I've attached the almost done computation's savefile (but this is for LLR). You may want to create a similar .pfr file (with some manipulations to instruct the code to dump it exactly at the iteration you want, if needed), and try something similar.