Re: [Gegl-developer] GEGL OpenCL Porting

[Øyvind Kolås <pippin gimp org>]

On Fri, Nov 21, 2014 at 3:59 AM, Nanley Chery <nanleychery gmail com> wrote:

Thanks for the quick fix, it's working on my system. 

I noticed that you've enabled GPU's by default due to some testing. Where can I find these results? According to my tests among two operations Edge-laplace and Video-degradation (currently on my bitbucket branches: edge_upstrm, vid_upstrm), OpenCL on my GPU performs 7.8x slower than my CPU. 

Benchmarking GEGL on CPU vs multithreaded CPU vs OpenCL CPU vs GPU is not easy, since different workloads can have very different performance characteristics. With the multi-threaded work I was happy when there was any performance win over single threaded without crashes/visual glitches; since it meant that the concurrency handling of GeglBuffer had improved as intended.

The type of workload I was using for my testing was many chained point composers / point composer-3s with quite large buffer sizes (4096x4096 buffers) which was meant to be representative of multi-layer compositions in GIMP.

 

Please let me know if you spot anything wrong with my measuring methodology or OpenCL implementation. Also, is the mailing-list and the bugzilla both suitable places to submit patches? 

For patches, bugzilla is best, since things posted there are found again if for some reason forgotten for a while.

/pippin