On Tue, Aug 2, 2016 at 10:18 PM, Nicolai Hähnle <nhaeh...@gmail.com> wrote: > Bas pointed out some things that mean this patch (and the next) probably > need to be reworked. The tests were probably not thorough enough.
I think that all indirect parameters are read by PFP. Our cache flush sequence already takes that into account and emits PFP_SYNC_ME. Also note that the driver only handles coherency between TC L2 and CB/DB caches. For anything else (e.g. special CP packets), you need to check r600_resource::TC_L2_dirty, which will tell you whether you need to flush TC before you can use the buffer with a block not using TC. See all usages of TC_L2_dirty. Marek _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev