rixed
Member
Im still working on my software renderer gpu940 (that I used for the 'code alone' little demo).
Lately, I was adding a GL-ES API to ease the use of the renderer.
So far, I've just ported glxgears, which is not a very good example because of its waste of
ressources (well, the purpose of glxgears is only to benchmark GL implementations).
Anyway, with proper overclocking it runs at 20 fps fullscreen with clipping and zbuffer, which is
twice as much as previous atempts based on vincent-gl and tinyGL, IIRC.
Some part of the renderer can still be optimized, so that I think it's possible to actually use
it for well-optimized GL apps (well optimized means : no over-usage of zbuffer and low-poly models).
And the 920 CPU is free to run anything else (the GL-ES layer runs on the 920, and performs only 3d
transformations and API convertion to the GPU, which then performs all clipping, projections and rendering).
With recent optimizations, the code-alone demo now runs at approx 50fps.
This is definitivelly false to state that the 940 can not handle rendering because of low RAM bandwidth.
[cut]
I am so happy I felt I had to share these news
Roadmap for future weeks :
- fix and optimize GL API lighting computations ;
- add textures to the GL API ;
- try to port a little game to see how far we can get ;
After that, probably design a dedicated 3d game for the GP2X, but this is another story.
--
link to gpu940 : http://gna.org/projects/gpu940/[/cut]
Lately, I was adding a GL-ES API to ease the use of the renderer.
So far, I've just ported glxgears, which is not a very good example because of its waste of
ressources (well, the purpose of glxgears is only to benchmark GL implementations).
Anyway, with proper overclocking it runs at 20 fps fullscreen with clipping and zbuffer, which is
twice as much as previous atempts based on vincent-gl and tinyGL, IIRC.
Some part of the renderer can still be optimized, so that I think it's possible to actually use
it for well-optimized GL apps (well optimized means : no over-usage of zbuffer and low-poly models).
And the 920 CPU is free to run anything else (the GL-ES layer runs on the 920, and performs only 3d
transformations and API convertion to the GPU, which then performs all clipping, projections and rendering).
With recent optimizations, the code-alone demo now runs at approx 50fps.
This is definitivelly false to state that the 940 can not handle rendering because of low RAM bandwidth.
[cut]
I am so happy I felt I had to share these news
Roadmap for future weeks :
- fix and optimize GL API lighting computations ;
- add textures to the GL API ;
- try to port a little game to see how far we can get ;
After that, probably design a dedicated 3d game for the GP2X, but this is another story.
--
link to gpu940 : http://gna.org/projects/gpu940/[/cut]