[ncl-talk] Plotting unstructured mesh

Mathieu Dutour mathieu.dutour at gmail.com
Tue Dec 16 09:25:52 MST 2014


On Mon, Dec 15, 2014 at 11:15 PM, David Brown <dbrown at ucar.edu> wrote:
>
> Hi Mathieu,
> I am interested in your comment that "the speed is still not
> absolutely excellent at 2.8s". What kind of performance do you think
> would be "excellent"? Are there other tools that achieve the
> performance you want? We are interested in any actual comparisons you
> want to give us. And of course we are always seeking to improve the
> performance of NCL as long as it does not compromise accuracy, which
> is our over-arching goal.
>
Ok, I understand your concern. As I said, the 2s is acceptable though not
absolutely excellent.

I will write some test for the plot of unstructured grid and compare with
other
programs. I trully want to have unstructured mesh plots and the situation
is not excellent right now.

However, I can already give you a test where ncl is problematic.
I want to draw a finite element mesh on the plane. So basically 80000
edges.

I use "gsn_add_polyline" for each edge which I know is problematic for
ncl, but I do not see any obvious alternative in ncl for the task. However,
a same call to SVG (Scalable Vector Graphics) works correctly with
"inkscape" with similar syntax.

Test files are available via
http://mathieudutour.altervista.org/Sending/PLOT_MESH.tar.gz

  Mathieu

 -dave
>
> On Thu, Dec 11, 2014 at 7:21 AM, Mathieu Dutour
> <mathieu.dutour at gmail.com> wrote:
> > Dear Mary,
> >
> > thank you very much! Yes with "RasterFill" it works correctly.
> > I used version 6.2.1.
> >
> > However the speed is still not absolutely excellent at 2.8s for a
> > plot. Is there a list of standard practice to make plots faster?
> >
> >   Mathieu
> >
> > On Thu, Dec 11, 2014 at 7:13 AM, Mary Haley <haley at ucar.edu> wrote:
> >>
> >> Mathieu,
> >>
> >> What version of NCL are you running?
> >>
> >> ncl -V
> >>
> >> I was able to run your script just fine, but I changed cnFillMode from
> >> "AreaFill" to "RasterFill". See attached png.
> >>
> >> --Mary
> >>
> >>
> >> On Mon, Dec 8, 2014 at 8:17 AM, Mathieu Dutour <
> mathieu.dutour at gmail.com>
> >> wrote:
> >>>
> >>> Dear ncl users,
> >>>
> >>> I have some errors for some contour plots on unstructured
> >>> grids (the script is in attachment and the relevant netcdf file
> >>> is available from
> >>> http://mathieudutour.altervista.org/Sending/test.nc   )
> >>>
> >>> Below is exact error:
> >>> fatal:ContourPlotPreDraw: ARSCAM/ARPRAM - ALGORITHM FAILURE
> >>> fatal:ContourPlotDraw: draw error
> >>> fatal:PlotManagerPreDraw: error in plot pre-draw
> >>> fatal:_NhlPlotManagerDraw: PreDraw error
> >>>
> >>> I believe that the data is too irregular for the algorithm to plot it.
> >>> But is there any way to alleviate this problem?
> >>>
> >>> Secondly, the plots sometime takes extremely long time to be done.
> >>> The mesh has only 110804 nodes which is nowhere near big. What
> >>> would be possible to avoid that problem?
> >>>
> >>>   Mathieu
> >>>
> >>> _______________________________________________
> >>> ncl-talk mailing list
> >>> List instructions, subscriber options, unsubscribe:
> >>> http://mailman.ucar.edu/mailman/listinfo/ncl-talk
> >>>
> >>
> >
> >
> > _______________________________________________
> > ncl-talk mailing list
> > List instructions, subscriber options, unsubscribe:
> > http://mailman.ucar.edu/mailman/listinfo/ncl-talk
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/ncl-talk/attachments/20141216/68bd8bb1/attachment.html 


More information about the ncl-talk mailing list