Re: [Mesa3d-dev] Support for OpenVG gpus

[Jose F. <jfo...@vm...>]

----- Original Message -----
> Hi,
> I plan to kill the liitle remainig free time with supporting OpenVG
> accelerating gpus. At the moment I have access to the imageon/adreno
> z160 in two different freescale boards. One of the boards has the z160
> next to a 3d Opengl|ES 2.0 gpu. Up to now I only stepped through mesa
> code whenever there were issues in our client code. So I never
> understood the internal structure of mesa.
> 
> I thought that vega would define or use some kind of abstraction
> similar to for example p_context for modern 3d gpus but dedicated to
> vg. But it implements vg on top of a p_context. So Vega seems to be
> the wrong starting point. Looking at what is exposed by the gpu the
> registers values and enums indicate that the level of abstraction is
> close to the vg spec.

Looking at http://en.wikipedia.org/wiki/OpenVG#In_hardware it seems that some hardware has dedicated engine, others use 3d engines.

If the GPU you care is really that close to OpenVG API, then maybe gallium is overkill. Gallium's idea was "one driver to rule them all", that is, to quickly target many graphics APIs with a single and compact driver. If every state trackers passes through its API to gallium interface then that would be lost -- drivers would be huge again.

I also wonder if there's enough momentum behind dedicated OpenVG hardware. Looking at the wikipedia page and https://developer.qualcomm.com/discover/chipsets-and-modems/adreno-gpu it seems like the trend is towards implementing OpenVG on top of generic 3D hardware.

But I admit I haven't been following OpenVG standard closely.

Jose