How big is the OpenCL language compared to e.g. OpenGL 3.2?
Phoronix: OpenCL Over Mesa, Gallium3D Discussion
While NVIDIA and ATI/AMD have OpenCL support within their binary drivers, the open-source Mesa / Gallium3D stack is still lacking open-source support for the Open Computing Language on Linux. But the discussion surrounding OpenCL in Gallium3D has been renewed on the mailing list today.Since August there has been a branch that provides Clover, which is an OpenCL state tracker ("CL over [Gallium3D]"), but unlike other state trackers it hasn't been progressing at a brisk pace...
http://www.phoronix.com/vr.php?view=Nzc4NQ
How big is the OpenCL language compared to e.g. OpenGL 3.2?
In terms of language OpenCL seems to be smaller than OpenGL. On the other hand, the implementation is more work than the language complexity suggests because of the need to run well on a wider range of targets than OpenGL.
I wonder why "coal" was suggested as a name rather than "chlorine" (Cl), which seemed like the obvious choice![]()
Reading up a bit on lex and yacc might be useful :
http://dinosaur.compilertools.net/
This URL is made of win.
OpenCL has a pretty clean, object-oriented API, which is a welcome change of direction compared to OpenGL. On the other hand, it looks pretty difficult to implement in a performant fashion - each GPU will need its own optimizations, more or less.
On the other hand, a Gallium implementation that runs on the CPU would be sweet. Not only would this improve portability / interoperability, it would also make math-heavy applications much more feasible on VM-based languages like Python, C# and the like.
Some preliminary testing here indicates a performance improvement between 5-50x when using AMD's CPU implementation instead of .Net code. More testing is necessary, but the result isn't too shabby!