Buffer Damage

[Vlad Zahorodnii]

On 1/4/21 2:05 PM, Simon Ser wrote:
> I don't think these would benefit from the optimization you're
> suggesting. They don't cycle between pre-rendered buffers, for
> instance.

No, no, it was just an example when the surface damage and the buffer 
damage might be different, I don't necessarily want to optimize that 
extremely rare case. The issue is that the compositor has simply no idea 
how exactly the client utilizes the buffers. Keeping the last N surface 
damages and accumulating them is a good heuristic, but it may produce 
incorrect results in some cases + it is redundant if the app uses 
hardware acceleration.

The bottom line is that shared memory client buffers tank performance 
and something should be probably done about it.

Cheers,
Vlad

> I don't think it's worth implementing an optimization if it only
> improves performance in synthetic edge cases that never happen in
> real-world scenarios.