[visit-developers] VisIt compositing problem on Franklin

tom fogal tfogal at alumni.unh.edu
Wed May 27 20:52:40 EDT 2009


"Gunther H. Weber" <ghweber at lbl.gov> writes:
> someone here at LBNL has found an instance where VisIt crashes on
> Franklin when at least one of the plots has transparent geometry. It
> seems that the crash occurs during the compositing stage, even though
> I am not 100% sure of that. The job status file generated after the
> engine is terminated indicates that the problem may be due to an "out
> of memory" condition, even though I am even less sure that this is
> true. Is there any chance that using Ice-T instead of VisIt's regular
> compositor may solve the problem? (Tom?)

Unfortunately, no -- IceT has trouble with transparent geometry at the
moment.  I'm working on it...

> Does anyone have an idea how to go about identifying the exact
> problem.

Can you run a recent trunk?  I've recently fixed some memory issues
that come up during compositing (`recently' == very recently, as in the
last 24 hours).

> I guess the next step is to run VisIt with debug logs, but in the
> past I had problems doing so on Franklin since not all directories
> are visible from the compute nodes.  Is there a way to tell VisIt
> to generate the debug logs in a specific directory instead of the
> current directory?

A quick/incomplete grep makes me think `no'... you could probably get
it going quickly by hacking common/misc/DebugStreamFull::Initialize,
though.  It might be nice to add a new command line option / env
variable for this.

-tom


More information about the visit-developers mailing list