Local astrometry.net "solve-field" - multicore...?

[furrysocks2]

I've just successfully "solved my first plate" locally - quite exciting the first time.

It failed on the first few attempts so I ran it through the online solver to give it low/high scale hints. It solved but took ages so looked at the online logs and saw "--downsample 2" so tried that. Much better - succeeds on first pass.

 

However, I've looked at the options, man page and searched online and found nothing - can it run multithreaded to take advantage of multiple cores? It looks like it's saturating just a single core and it would obviously be nice to be able to speed things up a little.

Does anyone know whether this can be configured?

[Davey-T]

No idea, I use Pinpoint on Win10, .5, offline and it solves in seconds.

Dave

[Marci]

May need to delve into GCC's ARCH_FLAGS (see 'Optimizing...' section at http://astrometry.net/doc/readme.html) and enable during compile for your cpu architecture.

 

I can't find anything that says yay or nay either way in astrometry docs, and the source is a tad large to check by hand. May need to check compilation flags for dependencies also. Big job.

[Marci]

Aha! Ask in the 'issues' section over at https://github.com/dstndstn/astrometry.net - I believe that's now the most direct support route these days. (Forgot it'd been gitted)

[furrysocks2]

Thanks, Marci.

Aye, funny that it's not mentioned one way or the other... seems like the kind of task that would lend itself well to concurrency.

[Marci]

Can confirm the solver doesn't do multi threading... shame.

https://groups.google.com/d/topic/astrometry/npvCpu6xUnw/discussion

[furrysocks2]

18 minutes ago, Marci said:

Can confirm the solver doesn't do multi threading... shame.

https://groups.google.com/d/topic/astrometry/npvCpu6xUnw/discussion

Well found, thank you.

Shouldn't matter when I know my kit/scale, but could easily enough write a wrapper to partition the search across a few scale ranges.

[Marci]

When running it as a server, it multithreads simultaneous jobs across cpus... e.g.: separate instance of solve-field per core - 4 cores - will batch process the same image on 4 scales simultaneously. That's built in, somewhere... not looked into running beyond standalone yet.