LLR 3.8.5 Development version

[Jean Penné]

Hi All,

LLR 3.8.5 Development version is available to download from my development directory :

http://jpenne.free.fr/Development/cllr385dev.zip Windows console application

http://jpenne.free.fr/Development/llr385dev.zip Windows GUI application

http://jpenne.free.fr/Development/llr385devlinux.zip Linux, dynamic link

http://jpenne.free.fr/Development/llr385devslinux.zip Linux, static link

http://jpenne.free.fr/Development/llr385devsrc.zip Compressed source directory

- One added feature : The Baillie-PSW PRP test is executed in place of the standard Lucas PRP test if the new option -oBPSW=1 is set in command line.
(see http://www.trnicely.net/misc/bpsw.html for details).

- In this Development version, three new bugs have been fixed, concerning
Fermat, Lucas and Frobenius PRP tests on non base two numbers and/or abs(c)!=1

Regards,
Jean

[MrRepunit]

Hello,

I wanted to point out a bug concerning the llr 3.8.5 development version (normal version 3.8.5 is working fine), running under Ubuntu 64 bit:
Running it on all known repunits with following the input file (starting with ./llr -d worktodo.txt)

ABC(10^$a-1)/9
19
23
317
1031
49081
86453
109297
270343

it gives the following output (here just for R(1031)):

Starting probable prime test of (10^1031-1)/9

(10^1031-1)/9 is base 3-Fermat PRP!  Time : 39.723 ms.
Starting Lucas sequence
(10^1031-1)/9 is strong-Fermat and Lucas PRP, Starting Frobenius test sequence
(10^1031-1)/9 is strong-Fermat and Lucas PSP (P = 6, Q = 2), but composite!!.
Frobenius RES64: 1C71C71C71C71C71  Time : 114.774 ms.

Same happens with the -oBPSW=1 flag.
So it seems there is a problem with just the Frobenius test.

Cheers,
Danilo

[Dead J. Dona]

Good day.

I want to know - does llr have any self-tests to check that everything is going fine?

like ./llr -selftest ----- to make all the tests or
./llr -selftest Fermat --- to check only one...

[henryzz]

Originally Posted by Dead J. Dona

Good day.

I want to know - does llr have any self-tests to check that everything is going fine?

like ./llr -selftest ----- to make all the tests or
./llr -selftest Fermat --- to check only one...

Not AFAIK. If you want to check your pc is working properly you need to use Prime95

[Jean Penné]

Originally Posted by MrRepunit

Hello,

I wanted to point out a bug concerning the llr 3.8.5 development version (normal version 3.8.5 is working fine), running under Ubuntu 64 bit:
Running it on all known repunits with following the input file (starting with ./llr -d worktodo.txt)

ABC(10^$a-1)/9
19
23
317
1031
49081
86453
109297
270343

it gives the following output (here just for R(1031)):

Starting probable prime test of (10^1031-1)/9

(10^1031-1)/9 is base 3-Fermat PRP!  Time : 39.723 ms.
Starting Lucas sequence
(10^1031-1)/9 is strong-Fermat and Lucas PRP, Starting Frobenius test sequence
(10^1031-1)/9 is strong-Fermat and Lucas PSP (P = 6, Q = 2), but composite!!.
Frobenius RES64: 1C71C71C71C71C71  Time : 114.774 ms.

Same happens with the -oBPSW=1 flag.
So it seems there is a problem with just the Frobenius test.

Cheers,
Danilo

This bug is fixed (a final modular reduction was missing) and the binaries
and source of LLR 3.8.5 development version are now updated accordingly.

Regards,
Jean

[Jean Penné]

Hi,

I updated again today LLR 3.8.5 development version source and binaries.

Neither new feature nor bug fixed, but roundoff error recovery has been improved :

Now, after more than 5 recoverable roundoff errors reached in a test, square_carefully and/or mul_carefully is used for all following iterations in this test.

This allows to complete the test successfully, even in these very rare but difficult cases!

I encountered this problem while PRP testing this large repunit : (10^270343-1)/9 and the BPSW option is set ; The Frobenius test get repeatedly 0.5 Roundoff errors while using gwsquare, although gwsquare_carefully works fine!

Regards,
Jean

[siegert81]

Is there a 64-bit version of LLR available?

If not, will there ever be a 64-bit version?