llrnet on base exponent of 2

[tnerual]

hello,

i work on the conjecture are us project ...

we actually work on the base 16 sierpinski part

so we test number like 31347*16^36249+1


i got these timings (same numbers tested):
LLRNET: 31347*16^36249+1 is not prime. RES64: C9DDFFA46F2BB6FB Time: 490.180 sec.
LLR 3.7.1C : 31347*2^144996+1 is not prime. Proth RES64: F25ECE59387BB41F Time: 138.144 sec.

the llrnet version is the on from rieselbase 5 modified to accept something else than k*2^n+1

LLR is almost 4 time faster than LLRnet (probably due to the fact that 2^4=16 and that llr test in base 2 and llrnet test in base 16)

is there a way to modify it or to find a trick to do the job in the fast way ?

thanks

[ET_]

Originally Posted by tnerual

hello,

i work on the conjecture are us project ...

we actually work on the base 16 sierpinski part

so we test number like 31347*16^36249+1


i got these timings (same numbers tested):
LLRNET: 31347*16^36249+1 is not prime. RES64: C9DDFFA46F2BB6FB Time: 490.180 sec.
LLR 3.7.1C : 31347*2^144996+1 is not prime. Proth RES64: F25ECE59387BB41F Time: 138.144 sec.

the llrnet version is the on from rieselbase 5 modified to accept something else than k*2^n+1

LLR is almost 4 time faster than LLRnet (probably due to the fact that 2^4=16 and that llr test in base 2 and llrnet test in base 16)

is there a way to modify it or to find a trick to do the job in the fast way ?

thanks

31347*16^36249+1 = 31347*2^(36249*4)+1 ?

Luigi

[tnerual]

Originally Posted by ET_

31347*16^36249+1 = 31347*2^(36249*4)+1 ?

Luigi

yes i know ... i'm testing exactly the same number but LLrnet is 4 times slower ...

[ET_]

Originally Posted by tnerual

yes i know ... i'm testing exactly the same number but LLrnet is 4 times slower ...

I think that Jean Penne' used optimized assembly routines in his LLR.

Luigi

[axn]

Originally Posted by ET_

I think that Jean Penne' used optimized assembly routines in his LLR.

Luigi

I think what he is saying that the newer LLR can automatically recognize that a power-of-two base can be tested as a base-2 number, thus using the more efficient base-2 FFTs rather than the generic PRP FFT. Thus, if the LLRNet (which uses older LLR 3.5) can be patched to do automatic conversion, it can gain from this principle.

[tnerual]

Originally Posted by axn1

I think what he is saying that the newer LLR can automatically recognize that a power-of-two base can be tested as a base-2 number, thus using the more efficient base-2 FFTs rather than the generic PRP FFT. Thus, if the LLRNet (which uses older LLR 3.5) can be patched to do automatic conversion, it can gain from this principle.

thanks axn for translating what i was trying to explain ...

at this moment, i just convert base 16 exponent to base 2 exponent with excel ... it's not too difficult (for small amount of pairs), but it would be more user friendly if it was done "automagically"

it's not really a bug but something really annoying (4 times slower)

[axn]

I have a patched LLRNet which can fudge base and n before calling the prp test. However, I can't attach it here, since the zip file is around 640KB (forum limit is 244KB). If you PM me your email, I can send it to you by mail.

If you prefer to build your own LLRNet, I have attached the modified file that needs to be inserted into the LLRNet source code.

[mdettweiler]

Originally Posted by axn1

I have a patched LLRNet which can fudge base and n before calling the prp test. However, I can't attach it here, since the zip file is around 640KB (forum limit is 244KB). If you PM me your email, I can send it to you by mail.

If you prefer to build your own LLRNet, I have attached the modified file that needs to be inserted into the LLRNet source code.

Does it do an LLR or Proth test, rather than a PRP test, after it's fudged the base and n? The newer versions of manual LLR will do an LLR/Proth test after converting the numbers to base 2, as long as they can be converted to base 2.

[axn]

Originally Posted by Anonymous

Does it do an LLR or Proth test, rather than a PRP test, after it's fudged the base and n? The newer versions of manual LLR will do an LLR/Proth test after converting the numbers to base 2, as long as they can be converted to base 2.

It should. However, I haven't tested it yet, so I don't even know whether it works at all

Any volunteers?

[mdettweiler]

Originally Posted by axn1

It should. However, I haven't tested it yet, so I don't even know whether it works at all

Any volunteers?

I'll try. (It shouldn't even need to be set up to communicate with a server--all I'll need to do is drop one or two sierp. base 16 candidates in the workfile.txt file.) My email is bugmesticky *at* googlemail *dot* com.

BTW, since it would appear that you've been at least able to build LLRnet successfully on your machine, could you possibly try building a copy of LLRnet with the LLR 3.7.1c source dropped in for me? I was going to try it myself, but for some reason the gcc on my machine (Ubuntu 7.10) couldn't find all the required files to build LUA (which is apparently part of LLRnet). I could send you my copy of the LLRnet source files that already have the 3.7.1c source code dropped in (in place of the 3.5.1 code), if you want.

[axn]

Originally Posted by Anonymous

I'll try. (It shouldn't even need to be set up to communicate with a server--all I'll need to do is drop one or two sierp. base 16 candidates in the workfile.txt file.) My email is bugmesticky *at* googlemail *dot* com.

YGM.

I'll try to build it with LLR 3.7, but my past attempts with 3.6 has not been very successful so I am not very hopeful.