The P-1 factoring CUDA program

[owftheevil]

And I thank you for yours.

[chalsall]

Originally Posted by owftheevil

And I thank you for yours.

Thank you.

[xilman]

Originally Posted by kracker

Now give us your half-first born child. Naow!

Top half or bottom half?

[ET_]

Originally Posted by xilman

Top half or bottom half?

Axial symmetry?

Luigi

[owftheevil]

Here's the code if anyone wants to play with it. It builds without problems on Ubuntu 12.04 with either Cuda4.2 or Cuda5.0 and gmp5.1.0. Its esssentially a slightly modified CUDALucas, so if you can build CuLu on Windows, you have a good start on building this.


Edit to add: run it with e.g.

./CUDA-pm1 60593041, -b1 1000, [-f 3360k]

If you don't specify b1 it defaults to 1. Specifying the fft is only necessary if b1 is small (< ~ 690). In that case, the test will finish before it knows if the fft is big enough and will sometimes give invalid results.

[Batalov]

Very impressive and very "evil"! "The lesser of two weevils".

[ET_]

Originally Posted by owftheevil

Here's the code if anyone wants to play with it. It builds without problems on Ubuntu 12.04 with either Cuda4.2 or Cuda5.0 and gmp5.1.0. Its esssentially a slightly modified CUDALucas, so if you can build CuLu on Windows, you have a good start on building this.


Edit to add: run it with e.g.

./CUDA-pm1 60593041, -b1 1000, [-f 3360k]

If you don't specify b1 it defaults to 1. Specifying the fft is only necessary if b1 is small (< ~ 690). In that case, the test will finish before it knows if the fft is big enough and will sometimes give invalid results.

Version 0.00, I love it!

Thanks a lot!

Luigi

P.S. the auto-correction works like a charm...

luigi@luigi-ubuntu:~/luigi/CUDA/cudapm1-0.00$ ./CUDA-Pm1 60593041, -b1 1000

Starting Stage 1 P-1, M60593041, B1 = 1000, fft length = 3200K
Doing 1475 iterations
Running careful round off test for 1000 iterations. If average error >= 0.25, the test will restart with a longer FFT.
Iteration  100, average error = 0.14299, max error = 0.34317
Iteration  200, average error = 0.13821, max error = 0.32842
Iteration = 284 < 1000 && err = 0.35178 >= 0.35, increasing n from 3200K
Starting Stage 1 P-1, M60593041, B1 = 1000, fft length = 3360K
Doing 1475 iterations
Running careful round off test for 1000 iterations. If average error >= 0.25, the test will restart with a longer FFT.
Iteration  100, average error = 0.05615, max error = 0.12774
Iteration  200, average error = 0.05472, max error = 0.13322
Iteration  300, average error = 0.05601, max error = 0.13491
Iteration  400, average error = 0.05518, max error = 0.14575
Iteration  500, average error = 0.05690, max error = 0.12720
Iteration  600, average error = 0.05746, max error = 0.13921
Iteration  700, average error = 0.05881, max error = 0.13656
Iteration  800, average error = 0.05901, max error = 0.13856
Iteration  900, average error = 0.05955, max error = 0.14368
Iteration 1000, average error = 0.05916 < 0.25 (max error = 0.14575), continuing test.
M60593041, 0x962b95049cafb7d9, offset = 0, n = 3360K, CUDA-P-1 v0.00
Stage 1 complete, estimated total time = 0:58
M60593041 has a factor: 2105528336291622770155712978260232660484461209

P.S.2: M1257787 has a factor: 1

[owftheevil]

I'm not sure, but I think Dubslow is responsible for the roundoff test part. Its hard to tell who did what on CuLu.

Edit: I didn't see the PS. I have so far been too lazy to make a different message for no factor found. I was thinking of just adding "but you already knew that, didn't you."

[henryzz]

I can see that cpus are going to become obsolete for P-1 stage 1 soon. This should help kill the P-1 deficit.

[firejuggler]

Thanks for your work owftheevil.

[frmky]

Great! I very much look forward to this being polished. Stage-1 only is certainly much better than no P-1. Once this uses GIMPS input files and saves results in GIMPS output format, I'll switch my 4 C1060's from double-checks to P-1 Stage-1.