P1 on non-x86 architecture.

[Lorenzo]

Is it possible to do P1 on non-x86 architecture (something like mLucas)?

[paulunderwood]

What is "P1"?

[Lorenzo]

Originally Posted by paulunderwood

What is "P1"?

P-1 Factoring

[paulunderwood]

https://en.wikipedia.org/wiki/Pollar...mplementations

[Lorenzo]

Originally Posted by paulunderwood

https://en.wikipedia.org/wiki/Pollar...mplementations

Sorry, but i little bit confused. How to use GMP-ECM for
PFactor=AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA,1,2,333001717,-1,76,2 ???

[paulunderwood]

If it is not using FFT arithmetic it is hopeless

[Lorenzo]

Originally Posted by paulunderwood

If it is not using FFT arithmetic it is hopeless

Sorry. Did not understand you. So P-1 factoring for M333001717 possible in mprime and cudapm (GPU version for nVidia card). Implemintation for another architecture (Linux on ARM, Power, SPARC etc.) does not exist?

[paulunderwood]

Originally Posted by Lorenzo

Sorry. Did not understand you. So P-1 factoring for M333001717 possible in mprime and cudapm (GPU version for nVidia card). Implemintation for another architecture (Linux on ARM, Power, SPARC etc.) does not exist?

I think it does not exist for those architectures. Somebody correct me if I am wrong.

[ET_]

Originally Posted by paulunderwood

I think it does not exist for those architectures. Somebody correct me if I am wrong.

AFAIK, there was at least an ECM version for ARM, don't know about P-1 though...

[ewmayer]

I intend to add p-1 support to a not-too-distant Mlucas release, but no specific timeline. Key sticking point is a subquadratic gcd - I could just go with an off-the-shelf implementation of such, but have been putting in place my own implementation of such over the past couple years, and prefer to finish that work. That will avoid 3rd-part library dependencies and also allow me to compare/improve my own code versus GMP's and George's tuned-up version of the Crandall giants.c subquadratic gcd used in prime95, which is interesting from a research angle.

I actually had a working FFT-mul-based p-1 algo in the old days of Mlucas-fortran which used a highly-tuned int64-based quadratic gcd, but that was before the M-numbers being tested got so large that the timings for using said gcd started going up into many hours, as opposed to the few minutes a good subquadratic gcd needs.

[Lorenzo]

Originally Posted by ewmayer

I intend to add p-1 support to a not-too-distant Mlucas release, but no specific timeline. Key sticking point is a subquadratic gcd - I could just go with an off-the-shelf implementation of such, but have been putting in place my own implementation of such over the past couple years, and prefer to finish that work. That will avoid 3rd-part library dependencies and also allow me to compare/improve my own code versus GMP's and George's tuned-up version of the Crandall giants.c subquadratic gcd used in prime95, which is interesting from a research angle.

I actually had a working FFT-mul-based p-1 algo in the old days of Mlucas-fortran which used a highly-tuned int64-based quadratic gcd, but that was before the M-numbers being tested got so large that the timings for using said gcd started going up into many hours, as opposed to the few minutes a good subquadratic gcd needs.

ohhh, God. Sad to hear ((

So it means that is one way to use mainframe or POWER ... and this way is double check jobs.