Re: [Mesa3d-dev] [mesa patch 1/3] dri2: Event driven buffer validation.

[Francisco J. <cur...@ri...>]

Michel Dänzer <mi...@da...> writes:

> On Fri, 2010-01-22 at 00:10 +0100, Francisco Jerez wrote: 
>> Michel Dänzer <mi...@da...> writes:
>> 
>> > On Sat, 2010-01-16 at 23:01 +0100, Francisco Jerez wrote: 
>> >> When a buffer invalidation event is received from the X server, the
>> >> "invalidate" hook of the DRI2 flush extension is executed: A generic
>> >> implementation (dri2InvalidateDrawable) is provided that just bumps
>> >> the "pStamp" sequence number in __DRIdrawableRec.
>> >> 
>> >> For old servers not supporting buffer invalidation events, the
>> >> invalidate hook will be called after flushing the fake front/back
>> >> buffer (that's typically once per frame -- not a lot worse than the
>> >> situation we were in before).
>> >
>> > When the invalidate event is available, IMO the buffers should also only
>> > be invalidated after a buffer swap / front buffer flush, not 'in the
>> > middle of a frame'.
>> >
>> Event-driven applications want the new buffers immediately after getting
>> the resize event, otherwise they could try to draw outside the bounds of
>> the old buffers, and that rendering would be lost.
>
> I still don't see any benefit to changing the buffers while the app is
> rendering a frame, but I guess the problem is that the DRI2 code in
> libGL doesn't know if that's the case or not...
>
>> Patches 2 and 3 make the affected drivers fabricate a flush before
>> switching buffers, so the previous rendering shouldn't be lost either.
>
> That may help for the fake front but not for the back buffer. Then again
> I guess the worst case is intermittent incomplete frames during a window
> resize, which may be acceptable.

That shouldn't happen, we only get InvalidateBuffers events when the
client is actively processing X events: sane clients don't do that "in
the middle of a frame" -- and whoever is actually doing that is likely
to be prepared to deal with a resize and some invalidated buffers at the
same point.