[yt-users] problem with fields that use ghost zones

John Wise jwise at physics.gatech.edu
Wed Jan 18 08:41:28 PST 2012


Hi Matt,

Thanks very much for the quick fix!  I've tested on one of my datasets, 
and it works now in serial and parallel.

John

On 01/18/2012 11:29 AM, Matthew Turk wrote:
> Hi John,
>
> I isolated the problem to when you have>  1 level jump in refinement
> at an Enzo root grid tile boundary.  I believe the fix I just pushed
> (33df4ef9f06c) should fix this case without resulting in a performance
> degradation for other situations (i.e., FLASH simulations or nested
> hierarchy simulations.)
>
> Thanks for bringing this up ...
>
> -Matt
>
> On Wed, Jan 18, 2012 at 12:02 AM, Sam Skillman<samskillman at gmail.com>  wrote:
>> Hi John,
>>
>> I just verified that it breaks on my even simpler data with the
>> MosaicFisheyeCamera with no_ghost=False (getting the vertex centered data).
>>   However, it does not fail when doing a normal rendering.  I'm out of steam
>> for tonight, but hopefully we'll track this down right away.  If you want to
>> do stuff for now, you can do hg backout e699c92663c2 and stay on the latest
>> tip without that ghost zone change.  I'm sure this'll be a high priority to
>> fix.
>>
>> Best,
>> Sam
>>
>> On Tue, Jan 17, 2012 at 9:24 PM, John Wise<jwise at physics.gatech.edu>  wrote:
>>>
>>> I forgot to include the debugging output from the crash.
>>>
>>> yt : [INFO     ] 2012-01-17 23:16:58,301 Rendering fisheye of 100^2
>>> vz = nan nan -26.912 nan, dv = 0.604703, ds = 14 6 16, dp = 0.911364 0.9
>>> 0.395297
>>> OINDEX(0,0,0) = nan, OINDEX(0,0,1) = nan
>>> vz = nan nan -26.9203 nan, dv = 0.414264, ds = 14 6 16, dp = 0.934091 -0.1
>>> 0.585736
>>> OINDEX(0,0,0) = nan, OINDEX(0,0,1) = nan
>>> Segmentation fault
>>>
>>> On 17 Jan 2012, at 23:21, John Wise wrote:
>>>
>>>> Hi,
>>>>
>>>> I tried to make some volume renderings with the latest tip
>>>> (f7e39b91cc6b), and it was crashing because there were NaNs in the data when
>>>> offset_interpolate was called.  I could make it crash in a small fisheye
>>>> rendering (100x100) of a 128^3 AMR simulation.  I've uploaded the dataset
>>>> (538MB) and script to
>>>>
>>>> http://www.physics.gatech.edu/~jw254/scpics/RD0009.tar
>>>> http://paste.yt-project.org/show/2042/
>>>>
>>>> I uploaded a diff of my debugging statement to here.
>>>>
>>>> http://paste.yt-project.org/show/2041/
>>>>
>>>> When I reverted back to the changeset (1558cb36d03b) before the ghost
>>>> zone update, this problem when away.  Could someone look at this or tell me
>>>> where to search for the bug?
>>>>
>>>> Thanks!
>>>> John
>>>>
>>>> On 17 Jan 2012, at 17:29, Matthew Turk wrote:
>>>>
>>>>> Hi Britton,
>>>>>
>>>>> Looks great.  Thanks.
>>>>>
>>>>> As a sidenote, I added the functionality to auto-detect which fields
>>>>> are needed; supplying them in ValidateSpatial should be faster,
>>>>> though.  Thanks!
>>>>>
>>>>> -Matt
>>>>>
>>>>> On Tue, Jan 17, 2012 at 5:22 PM, Britton Smith<brittonsmith at gmail.com>
>>>>> wrote:
>>>>>> Hi everyone,
>>>>>>
>>>>>> I'm finally following up on this thread.  I just submitted a pull
>>>>>> request
>>>>>> for the docs that includes some documentation of how to create fields
>>>>>> that
>>>>>> use ghost zones.  Some of it was a little tricky to explain, so let me
>>>>>> know
>>>>>> if the wording can be improved.
>>>>>>
>>>>>> Britton
>>>>>>
>>>>>>
>>>>>> On Wed, Jan 4, 2012 at 2:03 PM, Matthew Turk<matthewturk at gmail.com>
>>>>>> wrote:
>>>>>>>
>>>>>>> Hi Britton,
>>>>>>>
>>>>>>> On Wed, Jan 4, 2012 at 1:32 PM, Britton Smith
>>>>>>> <brittonsmith at gmail.com>
>>>>>>> wrote:
>>>>>>>> Hi Matt,
>>>>>>>>
>>>>>>>> I'm not sure I understand the change that you're talking about, but
>>>>>>>> it
>>>>>>>> seems
>>>>>>>> that they were auto-detected in the past, since the example without
>>>>>>>> them
>>>>>>>> there used to work.  I'm in favor of returning to that if possible.
>>>>>>>> Regardless, I think it would be worthwhile to add something to the
>>>>>>>> derived
>>>>>>>> fields documentation discussing this.  I could add that if we wanted
>>>>>>>> it,
>>>>>>>> but
>>>>>>>> it will be good to have it reviewed, since I'm not very familiar
>>>>>>>> with
>>>>>>>> how it
>>>>>>>> works.
>>>>>>>
>>>>>>> I think the best solution would be to have it simply auto-detect the
>>>>>>> fields necessary, rather than mandating they be specified (which may
>>>>>>> not always give the correct results.)  I'll implement this tomorrow
>>>>>>> morning.
>>>>>>>
>>>>>>> After some digging, it seems to me that this situation arose because
>>>>>>> we fixed a bug which had silently allowed this to occur, related to
>>>>>>> checking for field parameters in fields requiring ghost zones.
>>>>>>>
>>>>>>> -Matt
>>>>>>>
>>>>>>>>
>>>>>>>> Britton
>>>>>>>>
>>>>>>>>
>>>>>>>> On Wed, Jan 4, 2012 at 1:26 PM, Matthew Turk<matthewturk at gmail.com>
>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>> Hi Britton,
>>>>>>>>>
>>>>>>>>> It may not be documented, but I think we can actually auto-detect
>>>>>>>>> them; this would add on a list of lists of strings to the
>>>>>>>>> hierarchy,
>>>>>>>>> but I think that is manageable.  Would this be worthwhile?
>>>>>>>>>
>>>>>>>>> -Matt
>>>>>>>>>
>>>>>>>>> On Wed, Jan 4, 2012 at 1:24 PM, Britton Smith
>>>>>>>>> <brittonsmith at gmail.com>
>>>>>>>>> wrote:
>>>>>>>>>> Hi Sam,
>>>>>>>>>>
>>>>>>>>>> That fixed it, thanks!  I actually encountered this problem while
>>>>>>>>>> working on
>>>>>>>>>> my own derived field that used ghost zones and was using
>>>>>>>>>> VorticitySquared as
>>>>>>>>>> my example for how to do it.  I think in the past it was not
>>>>>>>>>> required
>>>>>>>>>> to
>>>>>>>>>> list the fields with ValidateSpatial, which is why it was working
>>>>>>>>>> as
>>>>>>>>>> is
>>>>>>>>>> in
>>>>>>>>>> older versions.  I wasn't able to find documentation on how to
>>>>>>>>>> make
>>>>>>>>>> fields
>>>>>>>>>> that use ghost_zones.  If it's in there and I just missed it,
>>>>>>>>>> could
>>>>>>>>>> someone
>>>>>>>>>> point me toward it?  If not, I could add something to the Creating
>>>>>>>>>> Derived
>>>>>>>>>> Fields section.
>>>>>>>>>>
>>>>>>>>>> Britton
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> On Wed, Jan 4, 2012 at 11:46 AM, Sam Skillman
>>>>>>>>>> <samskillman at gmail.com>
>>>>>>>>>> wrote:
>>>>>>>>>>>
>>>>>>>>>>> Hey Britton,
>>>>>>>>>>>
>>>>>>>>>>> It looks like VorticitySquared wasn't specifying the necessary
>>>>>>>>>>> fields
>>>>>>>>>>> (x,y,z velocity) in the definition.  I will push a change
>>>>>>>>>>> momentarily
>>>>>>>>>>> after
>>>>>>>>>>> I look around at any other ghost zone requiring fields to make
>>>>>>>>>>> sure
>>>>>>>>>>> they
>>>>>>>>>>> work.  DivV, for example, does the right thing.
>>>>>>>>>>>
>>>>>>>>>>> Sam
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> On Wed, Jan 4, 2012 at 9:32 AM, Britton Smith
>>>>>>>>>>> <brittonsmith at gmail.com>
>>>>>>>>>>> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>> Hi everyone,
>>>>>>>>>>>>
>>>>>>>>>>>> I'm having a problem using fields that use ghost zones.  The
>>>>>>>>>>>> following
>>>>>>>>>>>> simple script:
>>>>>>>>>>>> http://paste.yt-project.org/show/2010/
>>>>>>>>>>>>
>>>>>>>>>>>> gives this error:
>>>>>>>>>>>> http://paste.yt-project.org/show/bOikDPScBBtDiUGvH11X/
>>>>>>>>>>>>
>>>>>>>>>>>> I am working from the tip, but I get the same behavior from
>>>>>>>>>>>> yt/2.3.
>>>>>>>>>>>> In
>>>>>>>>>>>> yt/2.2, everything is working.  I am working now to narrow that
>>>>>>>>>>>> range
>>>>>>>>>>>> down a
>>>>>>>>>>>> bit, but does anyone have an idea?
>>>>>>>>>>>>
>>>>>>>>>>>> Britton
>>>
>>> _______________________________________________
>>> yt-users mailing list
>>> yt-users at lists.spacepope.org
>>> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
>>
>>
>>
>> _______________________________________________
>> yt-users mailing list
>> yt-users at lists.spacepope.org
>> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
>>
> _______________________________________________
> yt-users mailing list
> yt-users at lists.spacepope.org
> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org

-- 
John Wise
Assistant Professor of Physics
Center for Relativistic Astrophysics, Georgia Tech



More information about the yt-users mailing list