mesa glx/dri3 offloading support

Hi,

do these changes to mesa tree make ‘tearing’ problem disappear on nvidia randr1.4 offloading feature?

http://cgit.freedesktop.org/mesa/mesa/commit/?id=9320c8fea947fd0f6eb723c67f0bdb947e45c4c3

No.

Only way to get rid of tearing currently is to use bumblebee+[primus,virtualgl].
You can also use tripple buffering to mitigate effects somewhat, but it aint perfect either.

As far I know randr offloading is “in most cases” also less performant than primus (which is on the other hand limited to opengl apps only).

To my best knowledge, some kernel dev (david airlie @airlied) had some kernel interfaces marked as GPL only (GEM - dmabuf) so native experience is unlikely unless you use open source drivers (which still lag behind performancewise).

On a sidenote, primus is best we can get atm with regards to opengl offloading, unfortunatedly it isn’t zero copy and it requires double buffered apps to work. But other than that it uses CPU/GPU asynchronous transfers with PBOs and is fairly efficient (on fermi and newer cards).

To my best knowledge, on windows they have zero copy offloading between nvidia/intel cards (intel card buffer is stored in memory), but I haven’t heard of a way to do this on linux (would require mmapping intel framebuffer and using opengl to render directly into that).

Probably this problem will get fixed when we get to Wayland… I hope