[yt-users] YT's treecode
Andrew Davis
andrew.davis at yale.edu
Fri May 27 11:30:36 PDT 2011
Hello again,
First, thanks for the code fix. I am, however having some problems with the
speed up -- at the moment, I'm not seeing any, unfortunately.
I'm testing this with ~ 140000 cells of a zoom in AMR simulation. The number of
levels in the region is 11, so it's very similar to what is posted on the YT
clump-finding help page
(http://yt.enzotools.org/doc/analysis_modules/clump_finding.html#treecode-optimization)
-- namely the first figure for L=10 dataset with slightly fewer cells.
When I run:
> sp = pf.h.sphere(c,0.065/pf['pc'])
> pot = sp.quantities["IsBound"](truncate=False, include_thermal_energy=False,
treecode=False, opening_angle=0.0)
the timing is just ~230 seconds.
However, just changing to opening_angle=1.0 and treecode=True, the time is
roughly 900 seconds.
Higher opening angles only decrease the time by a little bit (opening_angle =
8.0 takes rougly 850 seconds). The error is still very small, so it is
calculating the right number at least, just slower than I think it should, as
the figures on the help page suggest around a factor of two speedup is
expected.
Is there anything that you see that I'm doing wrong?
Thanks!
Andrew
*************************************************************
** Andrew J. Davis andrew.davis at yale.edu **
** Dept. of Astronomy 203-432-5119 **
** Yale University www.astro.yale.edu/adavis **
*************************************************************
More information about the yt-users
mailing list