Re: [Openexr-devel] performance problem with exrenvmap due to memory consumtion

[Florian Kainz]

Hi Axel,

I will try to reduce exrenvmap's memory footprint.

The resizeErase() method in Array2D is written the
way it is to leave the Array2D object in a usable
state if new[] throws an exception.
The 1D-array version of resizeErase() is incorrect
and should probably be fixed.  If new[] throws an
exception, _data ends up pointing to a deleted memory
block.  Code that tries to access the array after
catching the exception can't tell that the _data
pointer is invalid.
On the other hand, in exrenvmap it really doesn't
matter whether resizeErase() handles exceptions
correctly or not.  The program contains only a
single try block; if an exception is caught,
exrenvmap prints an error message and exits.

Florian


Axel Waggershauser wrote:
> Hi,
>
> I tried to convert a 400M latlong rgb half exr with exrenvmap into a
> cube-map with a width of 2048, that sums up to a memory footprint for
> exrenvmap of about 725M. With running a full featured X desktop and 512M
> of RAM I have to rely on swapping, OK, but there are two things I found
> in the code that seem to unnecessarily demand my machines capabilities:
>
> The loop in makeCubeMap calls repetitively EnvmapImage::resize() and
> EnvmapImage::clear().
>
> 1. EnvmapImage::resize() calls Array2D<T>::resizeErase() which looks
> like:
>
>     T *tmp = new T[sizeX * sizeY];
>     delete [] _data;
>     _data = tmp;
>
> So it first allocates new memory and _then_ deletes the old memory.
> This makes no sens to me, especially since the same method of Array<T>
> makes it the other way around.
>
> 2. EnvmapImage::clear() assign 0 to all pixels, but as far as I can see
> this is for no reason, since they get assigned a new value in
> makeCubeMap.cpp:195 anyway. It only means that I have to swap everything
> in and out three times instead of two.
>
> Any comments on this?
>
> Thanks,
> Axel.
>
>
>
> _______________________________________________
> Openexr-devel mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/openexr-devel