Linear algebra using all available cores

[bur]

I recently installed a fresh copy of CADO and it uses all available cores (including logical cores) for the LA step. Since it's a smaller number I simply ran ./cado-nfs.py $a -t 10.

I tried -t 10, "tasks.linalg.characters.threads = 2" and even played around with different values for "tasks.linalg.bwc.threads = 2x2" on the command line but nothing changed.

Polyselect and sieving is done with the number of threads I specified.

In the snapshot file it says tasks.threads = 10 and tasks.linalg.bwc.threads = 2x2


I would simply use msieve, but for this specific factorisation it spit out an "error -1" for every single relation. remdups also gave 99% bad relations. This never happened before, so I doubt it's related to the threading problem. And especially for small numbers it's much more convenient to just have CADO handle everything.

[VBCurtis]

When I use the command-line flag --server-threads 10, all phases of the job use 10 threads or fewer. I haven't tried -t, and I'm not sure if they're two invocations of the same flag.

Why are you trying "2x2" for bwc instead of 4?

[bur]

When I use the command-line flag --server-threads 10, all phases of the job use 10 threads or fewer. I haven't tried -t, and I'm not sure if they're two invocations of the same flag.

I assume it's the same, -t is from the readme.

Why are you trying "2x2" for bwc instead of 4?

Because I don't understand the option... I just took it from params.c90 to see if it would change anything. Then I'd have investigated further.