Externally Generated Polynomials Don't Seem To Be Being Considered

[EdH]

I need some help in my setup, I guess...

I have two machines running poly select on older GPUs using msieve revision 979. Their commands to search are:

./msieve -i number -np

These appear to be working as expected and generate lots of candidates in their respective msieve.dat.p files. When interrupted, an msieve.fb file is created.

I have a third machine running factmsieve.py with a much more current version of msieve that is also tasked with polynomial selection, but it is using its CPU. I have wall time limited this polynomial run. It generates the file number.dat.p as it is running and then number.fb when the interrupt occurs. All of this is running OK.

However, I have tried to incorporate the two GPU generated groups of candidates into the CPU generated one, but the chosen polynomial never appears to be from the GPU generated lists, even when they have candidates with better scores.

I have tried concatenating the msieve.dat.p contents onto the number.dat.p file after it is initiated and I have tried starting out by generating the new number.dat.p file from the two msieve.dat.p files prior to running factmsieve.py.

What might I be missing?

Thanks.

[chris2be8]

Here's a perl script to find the best poly in several NAME.*.msieve.dat.p files and generate NAME.poly from it. It assumes NAME.n exists in the current directory.

I wrote it to be called from a shell script which would have run several msieve tasks in parallel. But it should work if you manually put them all in 1 dir, then call this script (with 1 parm of NAME).

Chris

[EdH]

Originally Posted by chris2be8

Here's a perl script to find the best poly in several NAME.*.msieve.dat.p files and generate NAME.poly from it. It assumes NAME.n exists in the current directory.

I wrote it to be called from a shell script which would have run several msieve tasks in parallel. But it should work if you manually put them all in 1 dir, then call this script (with 1 parm of NAME).

Chris

I'll look it over, but I'd need to modify factmsieve.py to make use of it, since everything is running automated and scripted to that right now. I'm not too opposed to modifying it - I already have for mpi and other changes, but I thought perhaps there was a simple way to have the msieve thread consider already generated polys. Maybe I need to separate the selection steps. Of course, that's a modification, as well.

Thanks.
-Ed

[jasonp]

Polynomial selection will not consider polynomials that have already been generated in the .p file that is the output from the stage 2 root sieve. That's true even if you manually choose the best polynomial and make the .fb file out of it; then polynomial selection sees you have already chosen a polynomial, and decides its job is finished :)

[EdH]

Originally Posted by jasonp

Polynomial selection will not consider polynomials that have already been generated in the .p file that is the output from the stage 2 root sieve. That's true even if you manually choose the best polynomial and make the .fb file out of it; then polynomial selection sees you have already chosen a polynomial, and decides its job is finished :)

Thanks! I had noticed the second part. I'm currently experimenting with having the better GPU machine supply the .fb file, instead of additions to the .p file, which causes the poly selection to be skipped altogether in the factmsieve.py machine. That fits with your explanation. The only wrinkle is that I had hoped to use two GPU machines, without having to manually select anything.

Thanks for all...