Re: [Gnash-dev] Relation between hardware acceleration and renderers

[Rob Savoye]

On 09/07/11 10:36, Sandro Santilli wrote:
> Now that the hardware acceleration branch was merged to master,
> can we have an overview of what changes in terms of architecture ?

  This has been answered in great detail on this list in the past, so I
see no need to waste time duplicating all those posts. The only recent
change worth mentioning is I refactored the OpenVG renderer to use the
"StyleHandler" for gradients and bitmaps. (both of which OpenVG has
hardware acceleration for)

  As a note, our existing OpenGL renderer is total crap, and needs to be
either fixed, or depreciated. It doesn't even use most of the hardware
acceleration features of OpenGL anyway. I think using OpenGLES2 is the
best long-term solution, as that is now supported by many distributions,
plus works well on mobile too. Course we still need an OpenGLES2
renderer to be written. Which is on the wish list...

  Our existing dependency on AGG is a performance killer on many
platforms. Not only that, AGG itself appears to be unmaintained.

> In particular, a recently opened task [1] is about making
> double-buffering work for both AGG _and_ hardware accelerated GPU:
> does it mean that hardware acceleration is a new renderer or do we
> get hardware acceleration for any existing renderer ?

  Gnash has had hardware acceleration support since the very beginning,
so I don't quite see your point. We now have a new renderer, OpenVG
(with OpenGLES as a future task), that can use the GPU for both double
buffering, image transformations, hardware gradient support, and video
decoding. Many of the changes made for OpenVG (the hwaccel branch) are
to take full advantage of hardware acceleration where ever possible.

  The current double buffering code was AGG specific, whereas the task
is to refactor this so it also works with hardware based double
buffering, initially OpenVG.

        - rob -