[vtk-developers] Major performance regression in VTK 6.2 to 6.3 with vtkLookupTableMapData().

[Cory Quammen]

The patch at [1] has been merged. Performance should be back to normal.

- Cory

[1] https://gitlab.kitware.com/vtk/vtk/merge_requests/406

On Sun, Jul 12, 2015 at 11:56 PM, Cory Quammen <cory.quammen at kitware.com>
wrote:

> Hi all,
>
> I've created a merge request with a possible fix here:
>
> https://gitlab.kitware.com/vtk/vtk/merge_requests/406
>
> Please try it out and let me know if you find any problems.
>
> Thanks,
> Cory
>
> On Fri, Jul 10, 2015 at 12:19 PM, Cory Quammen <cory.quammen at kitware.com>
> wrote:
>
>> Berk,
>>
>> Yes, that might be possible. The tricky part is that the resize needs to
>> be done correctly in the various places where the original table size is
>> modified, of which there are a few
>> (vtkLookupTable::SetNumberOfTableValues(), ::SetNumberOfColors(),
>> ::SetTable(), etc.). I'll try it out and push a patch when I get the chance.
>>
>> Best,
>> Cory
>>
>>
>>
>> On Wed, Jul 8, 2015 at 2:01 PM, Berk Geveci <berk.geveci at kitware.com>
>> wrote:
>>
>>> It looks like that code can be refactored to do the resizing in some
>>> sort of initialization method. This comment:
>>>
>>>   // Resize the internal table to hold the special colors at the
>>>   // end. When this function is called repeatedly with the same size
>>>   // lookup table, memory reallocation will be done only one the first
>>>   // call if at all.
>>>
>>> indicates that the size does not change dynamically. Maybe it can be
>>> changed to always allocate for the special colors from the beginning?
>>>
>>> -berk
>>>
>>> On Wed, Jul 8, 2015 at 2:40 PM, Cory Quammen <cory.quammen at kitware.com>
>>> wrote:
>>>
>>>> Sean,
>>>>
>>>> I will take a look.
>>>>
>>>> Cory
>>>>
>>>> On Wed, Jul 8, 2015 at 1:11 PM, Sean McBride <sean at rogue-research.com>
>>>> wrote:
>>>>
>>>>> Hi all,
>>>>>
>>>>> As a mentioned in another thread, we've found a big slowdown in our
>>>>> app between 6.2 and 6.3.  Profiling with Instruments reveals lots of time
>>>>> is being spent locking & unlocking a mutex in vtkLookupTableMapData().
>>>>> Looking through git history shows the culprit:
>>>>>
>>>>> ----------------------------
>>>>> commit dac748aedd7ffdc5aca6d555e74c668451ef4772
>>>>> Author: Cory Quammen <cory.quammen at kitware.com>
>>>>> Date:   Sun Mar 15 00:33:11 2015 -0400
>>>>>
>>>>>     BUG 15365: Fixed crash in function called from multiple threads
>>>>>
>>>>>     vtkLookupTableMapData() was not thread safe and could lead to
>>>>>     crashes when accessed from multiple threads. Added a mutex around
>>>>>     the logic to determine if the color table size needed to be
>>>>>     increased.
>>>>>
>>>>>     Added a multi-threaded test that crashes on occasion prior
>>>>>     to this patch, but does not crash with this patch applied.
>>>>> ----------------------------
>>>>>
>>>>> Indeed the added code has the comment:
>>>>>
>>>>> // Since this involves a potential array resize and this function
>>>>> // might be accessed concurently from more than one thread, we need a
>>>>> // mutex here. This shouldn't affect performance much if this function
>>>>> // is used to map many input values, but if it is called repeatedly
>>>>> // with short input arrays, performance may be much worse.
>>>>>
>>>>> So, for my use case, performance is indeed much worse.  :(
>>>>>
>>>>> As a hack, I removed the lock/unlock and performance is back to
>>>>> 6.2-level.
>>>>>
>>>>> Help? :)
>>>>>
>>>>> --
>>>>> ____________________________________________________________
>>>>> Sean McBride, B. Eng                 sean at rogue-research.com
>>>>> Rogue Research                        www.rogue-research.com
>>>>> Mac Software Developer              Montréal, Québec, Canada
>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>> --
>>>> Cory Quammen
>>>> R&D Engineer
>>>> Kitware, Inc.
>>>>
>>>> _______________________________________________
>>>> Powered by www.kitware.com
>>>>
>>>> Visit other Kitware open-source projects at
>>>> http://www.kitware.com/opensource/opensource.html
>>>>
>>>> Search the list archives at:
>>>> http://markmail.org/search/?q=vtk-developers
>>>>
>>>> Follow this link to subscribe/unsubscribe:
>>>> http://public.kitware.com/mailman/listinfo/vtk-developers
>>>>
>>>>
>>>>
>>>
>>
>>
>> --
>> Cory Quammen
>> R&D Engineer
>> Kitware, Inc.
>>
>
>
>
> --
> Cory Quammen
> R&D Engineer
> Kitware, Inc.
>



-- 
Cory Quammen
R&D Engineer
Kitware, Inc.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/vtk-developers/attachments/20150713/006138fc/attachment.html>