2014-03-02, 15:49
The 3d graphics driver for the Pi currently runs mostly on the VPU (a processor on the GPU).
There's not much point in releasing that source, as the compiler/debugger tools for the VPU are not freely available.
What's been released is the 3D graphics driver source code for a similar chip (with the same 3d hardware) where the driver runs on the ARM.
Running on the ARM makes building/debugging/developing much easier and so this is an easier starting point than releasing the Pi's VPU code.
The hope is that the ARM side driver is made functional and can be used as a replacement for the VPU based one.
This driver may have better or worse performance than the closed source one (probably worse initially).
The key thing is if someone wants to improve it, or add new feature, or fix bugs, then that will be possible.
There's also now a full register level spec for the 3D hardware (including the SIMD processors that run GL ES shaders),
which means people could run other (non-3D) code on it - e.g. an OpenCL compiler could now be written,
which could be useful for various optimisation tasks.
We recently released FFT code that ran on this SIMD procesor, which is about 10 times faster than FFTW running on the ARM.
In theory other functions could now be produced to accelerate parts of (currently unsupported) codecs, or do audio processing, or video processing or whatever.
Whether people will choose to work on features like this, I don't know, but it is now a possibility.
There's not much point in releasing that source, as the compiler/debugger tools for the VPU are not freely available.
What's been released is the 3D graphics driver source code for a similar chip (with the same 3d hardware) where the driver runs on the ARM.
Running on the ARM makes building/debugging/developing much easier and so this is an easier starting point than releasing the Pi's VPU code.
The hope is that the ARM side driver is made functional and can be used as a replacement for the VPU based one.
This driver may have better or worse performance than the closed source one (probably worse initially).
The key thing is if someone wants to improve it, or add new feature, or fix bugs, then that will be possible.
There's also now a full register level spec for the 3D hardware (including the SIMD processors that run GL ES shaders),
which means people could run other (non-3D) code on it - e.g. an OpenCL compiler could now be written,
which could be useful for various optimisation tasks.
We recently released FFT code that ran on this SIMD procesor, which is about 10 times faster than FFTW running on the ARM.
In theory other functions could now be produced to accelerate parts of (currently unsupported) codecs, or do audio processing, or video processing or whatever.
Whether people will choose to work on features like this, I don't know, but it is now a possibility.
