> Hi everyone, > > This is a cleanup in a series of cleanups I am going to make to improve the > horribly over-engineered and slow state management of r600g. This particular > patch series brings nothing new or special, it's merely a preparation for > future work. > > This series removes any uses of the register mask, which was used for partial > updates of registers. These patches show that sometimes we used it even when > we didn't have to, and the few cases that are valid can be handled separately > without much effort anyway. The majority of registers don't need this feature. > > Later on when we start using a more direct way to emit immutable states, we > should be able to bypass r600_pipe_state and memcpy state vectors into the > command stream directly. Getting rid of the register mask is a prerequisite > for that. > > There are no piglit regressions. Tested with RV670, RV730, and REDWOOD. > > The patches are also available at: > git://people.freedesktop.org/~mareko/mesa r600-kill-regmask >
I've looked over this and I've wanted to do something like this since I first ran callgrind on r600g, reducing the CPU overhead here is a good plan. Reviewed-by: Dave Airlie <[email protected]> Dave. _______________________________________________ mesa-dev mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/mesa-dev
