Mlucas/Glucas for LLR tests?

[mdettweiler]

I was wondering, how easy would it be to tweak Mlucas or Glucas to run LLR tests? Then non-x86 platforms would no longer be limited to sieving for k*2^n-1 projects.

[ewmayer]

I estimate several months' work at least - not horrendous, but given all the stuff already in the near-term [next 6-12 months] to-do list, I won't have any time for such additional work.

Sorry,
-Ernst

[mdettweiler]

Originally Posted by ewmayer

I estimate several months' work at least - not horrendous, but given all the stuff already in the near-term [next 6-12 months] to-do list, I won't have any time for such additional work.

Sorry,
-Ernst

Oh, I see. I thought it would be a trivial task--I guess not.

[Jean Penné]

It is not trivial at all!
I am working for a while(but not continuously...) on a portable, C code only, version of LLR, and wrote a prototype program which works on k*2^n+/-1 numbers, and k up to ~16 bits, but there is still a lot of work to do before having a program reliable and easy to use, enough to be released...

IMHO, the most advanced person on this task is Phil Carmody who wrote a program named "Phrot" :

This program (if my info. is up to date) does PRP tests on k*b^n+/-1, where b is not necesserily 2, and uses the YEAFFT library for the large integer modular multiplications and squarings. I think that to implement LLR and Proth algorithms in this code would not be very difficult, but nevertheless, not really trivial...

Regards,
Jean

[rogue]

Originally Posted by Jean Penné

IMHO, the most advanced person on this task is Phil Carmody who wrote a program named "Phrot" :

This program (if my info. is up to date) does PRP tests on k*b^n+/-1, where b is not necesserily 2, and uses the YEAFFT library for the large integer modular multiplications and squarings. I think that to implement LLR and Proth algorithms in this code would not be very difficult, but nevertheless, not really trivial...

Regards,
Jean

I use phrot frequently, but I limit my searches to Generalized Woodalls, Sierpinski/Riesel Base 5, and Conjectures 'R Us. I have found that his software (on PowerPC) has issues when b = 2^m (for m >= 1) and k > 1. I should clarify that. I don't think the bug is in his software, but in YEAFFT. With those two conditions there appears to be an FPU rounding error in YEAFFT that is somehow masked so that phrot cannot detect it. In other words, phrot thinks that the test is valid (maxerr < 0.5), but the test results are incorrect (numbers that are prime are returned as composite). I have spent a little time researching it, but don't have enough knowledge to debug YEAFFT. It can be difficult to debug because the numbers that fail are over 10000 bits. If I could find one that fails that has only a few hundred bits, I might be able to find the problem. I've tried to contact Guillermo Valor, but he never responded. Phil might have the skill to debug it, but I know he tends to be very busy.

To make this clear, I use his software on PowerPC for all bases where the base is not a power of 2. I have found hundreds of new primes with phrot, but since they tend to be low-hanging fruit for the other projects I participate in, they are not large enough to make it to the Prime Pages. I have tried, but haven't had any luck as of yet.