Unexpected termination of PM-1

[Viliam Furik]

Originally Posted by tuckerkao

It depends on the available memories of the user's PC. Many larger factors require the bounds to be up to 2^14 or greater which can be out of reach by many models. ECM has the higher chance to find such the useful factors, but require many curves to be run before able to find 1.

You do realize, that 2^14 = 16384, right? We are talking about B2 size, which is usually in the range of millions and more.

[kruoli]

You should be able to manually convert the Prime95 save file into a GMP-ECM save state. IIRC, there should be a How-To on this somewhere on this forum (by kriesel?). I have done this for the old save file format at least once successfully. GMP-ECM with OpenMP will be much more performant for stage 2 and multiple cores.

[VBCurtis]

Originally Posted by Miszka

I made several attempts to determine the value of B2 to complete my task.
Only B2=B1=40000000000 was successful.
In this experimental way I found out that for small Mersenne numbers the P-1 method is not a good method.

No. You found out what George already told you: Prime95 is not the right software for small Mersennes and large bounds.

[nordi]

Originally Posted by kruoli

You should be able to manually convert the Prime95 save file into a GMP-ECM save state. IIRC, there should be a How-To on this somewhere on this forum (by kriesel?)

I know there's instructions for doing this with ECM, but didn't know it would be possible for P-1. It sounds really interesting, but all I could find from Kriesel was this which does not mention gmp-ecm. Any clues?

[ATH]

Originally Posted by Viliam Furik

You do realize, that 2^14 = 16384, right? We are talking about B2 size, which is usually in the range of millions and more.

I think he meant 10¹⁴.

[kruoli]

Originally Posted by nordi

I know there's instructions for doing this with ECM, but didn't know it would be possible for P-1. It sounds really interesting, but all I could find from Kriesel was this which does not mention gmp-ecm. Any clues?

You may be right, there might be no such post. Maybe it was from GP2. But at least I am sure that it works. I did some testing. This worked for me (curly brackets indicate that something needs to be replaced):

1. Run Prime95/mprime with Pminus1=1,2,{exponent},-1,{B1},0.
2. Now, a save file exists m{zero padded exponent}.
3. Open the file in a Hex Editor.
   1. Remove the first 88 bytes.
   2. Reverse all bytes.
   3. Copy the hexadecimal representation.
   4. Remove leading zeros.
   5. Convert this to lower case.
4. Create a new file with this content:

   METHOD=P-1; B1={B1}; N=2^{exponent}-1; X=0x{modified hex data from the last step}; CHECKSUM=0; PROGRAM=; Y=0x0; X0=0x3; Y0=0x0; WHO=; TIME=;

5. Execute ecm -resume {file name from the previous step} {B1}-{B1} {wanted B2}. It will state something like

   GMP-ECM 7.0.4 [configured with GMP 6.2.1, --enable-asm-redc] [ECM]
   Resume file line has bad checksum 0, expected 729905639

6. Modify the file you created to use the checksum GMP-ECM expected.
7. Repeat the same command again. Now it will run.

I have not tested with known factors, but I guess we could work something out from here.

[Miszka]

Originally Posted by VBCurtis

No. You found out what George already told you: Prime95 is not the right software for small Mersennes and large bounds.

Maybe I misspoke, but I understood it as you wrote.

[Viliam Furik]

Originally Posted by ATH

I think he meant 10¹⁴.

Yes, probably. But it's not what he wrote.

[kriesel]

Originally Posted by nordi

all I could find from Kriesel was this which does not mention gmp-ecm.

Thanks for doing your homework first. (Not everyone does.) I've updated that page considerably since your post. I knew of no way to move P-1 work in progress among applications without giving up all the ground previously gained. (And I've searched, and looked at writing a translator or two.)

Recently kruoli previously posted about a manual method for prime95 P-1 -> gmp-ecm.

What I was looking for was CUDAPm1 -> anything reliable. If someone knows of a way, please contact me by PM or post it here.


edit: There's a maintained list of some P-1 self test candidates with known factors scattered across a wide range of exponents and therefore many fft lengths here
"(Not everyone does.)" was about certain folks who start a new thread asking a question that's easily answered by looking in the program documentation, or reference info, or with a quick forum or web search. And who continue to do so after repeatedly drawing flak from both ordinary users and moderators. And often it's the same few who don't bother to do enough independent thinking or testing. Not related to the previous posters in this thread.

[kruoli]

Originally Posted by kriesel

(Not everyone does.)

It is really hard to search for posts regarding P-1 because the forum search does not accept "P-1" as a keyword. I looked at your blog site map, there was no mention. But that does not mean such a post does not exist. So I did a reasonable amount of "homework".

[kriesel]

Originally Posted by tuckerkao

Many larger factors require the bounds to be up to 2^14 or greater

2¹⁴=16384, which is a rather small bound for most exponents. Perhaps you meant 10¹⁴?