GEGL developer sponsored to improve hardware acceleration support
Victor Oliveira, a former Googler Summer of Code 2011 student for GIMP/GEGL, has just announced that his further work on getting GEGL to use OpenCL based hardware acceleration will be supported by AMD.
This summer Victor worked on optional use of OpenCL in GEGL — a new GIMP's node based image processing core. The news got around, and in mid-September he was contacted by Eric Lundgren, marketing and alliances manager from AMD, who was interested in having GEGL playing well with the company's APUs. The two eventually agreed on getting some contracted work done by Victor to get GEGL's OpenCL support into a better shape by March 2012.
Victor is going to focus on infrastructure behind the scenes to make it work efficiently. What it means is that color conversions should be done on GPU whenever possible, and composition operations (Porter-Duff, arithmetic ones like divide and multiply etc.) as well as some essential filters like gaussian blur should be done on GPU as well. When OpenCL is not available, GEGL will use fallback to the current implementation of the operations.
Victor is also planning to continue his experiments with GPU-side buffer management and he's also going to implement a simple API for writing OpenCL operations for GEGL, so that new developers could easily write their own new hardware accelerated filters or port existing ones.
The work is being done in a public branch of the main GEGL repository. Victor is also going to update his blog as his project progresses. AMD is interested to see GIMP actually using OpenCL by the end of the project which is entirely possible thanks to optional GEGL based projection (coming in 2.8) and the experimental GEGL operation tool (available since 2.6).