[Radiance-general] Re: rtrace
Greg Ward
gregoryjward at gmail.com
Sun Jul 18 23:14:44 PDT 2010
Just back from a week in the woods, and playing catch-up....
> From: Thomas Bleicher <tbleicher at googlemail.com>
> Date: July 14, 2010 2:31:46 AM PDT
>
> I have to speculate about the connection between "-x" and "-n" here
> but i think that if you allow more processes (-n) than points in a
> scan line (-x) then rtrace simply can't start enough processes. This
> would mean that each point creates it's own process. Greg?
This is more or less correct. What actually happens is that rays are
sent to multiple processes, but we need to preserve the first-in/first-
out (FIFO) order of the rays sent to rtrace, which means buffering up
results on the output side. Logically, if you have set -x to 5, then
we need to take 5 ordered input ray origins and directions and run in
through N processes until we get all 5 results out so we can put them
back in the same order. In other words, throwing additional processes
over the -x setting buys you nothing, because your queue is never
deeper than this number. The default of -x 0 (infinite) is the best,
because it means rtrace can continue to slurp in new rays and buffer
them indefinitely, outputting ordered results as they become available.
In any case, you aren't likely to see much benefit to the rtrace -n
option if you are only computing a handful of rays.
Cheers,
-Greg
More information about the Radiance-general
mailing list