On Fri, Sep 25, 2015 at 01:57:55PM +0200, Lucas Stach wrote: > This redefines how we do cache handling in the following way: > > All buffer are pushed into the GPU domain initially, this will only > be done when populating the buffers backing store, but for simplicity > userspace can assume that the buffer is owned by the GPU as soon as it > constructs a GEM handle. The memory is only implicitly pulled back into > the CPU domain when destroying the handle. > > Uncached and writecombined buffers can stay in the GPU domain for their > entire lifetime, as any modification to them will be seen by the GPU > either immediately or latest when the write buffers get flushed when > linking a new submit into the GPUs command stream. > > If any modifications needs to be done to a cached buffer they must be > pulled into the CPU domain before and pushed to the GPU afterwards in > an explicit manner. > > Signed-off-by: Lucas Stach <l.stach@xxxxxxxxxxxxxx> Applied, thanks. -- FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up according to speedtest.net. _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/dri-devel
