On Sun, Feb 5, 2017 at 11:06 PM, Kevin Kofler <kevin.kofler@xxxxxxxxx> wrote: >> libglvnd is a solution for this it is a vendor neutral >> implementation of libGL.so.1 which acts as a dispatcher >> to one or more glvnd enabled libGL implementations >> installed on the systems. > > By doing so, it decreases performance for all the users of the Mesa drivers > by adding an unnecessary layer of indirection. I do not see performance > being addressed at all in any of your communication, did you even try to > measure the impact of the added indirection layer? glvnd is basically mesa's gl dispatch layer ripped out and moved into a different git tree. There should be no reason to expect a performance difference. And if there were, do you think nv would be using glvnd?? For some history, the mesa dispatch layer is working in basically the same way as nv blob's dispatch layer. Although nv blob patches the entry points to optimize a bit further for some special cases (single thread, single context). Since it was easier to start w/ mesa code for glvnd, rather than them opening up parts of their own blob, nv added the missing entry point patching to the mesa based glvnd code. So hypothetically if someone was motivated enough, mesa could be made faster with glvnd. Not sure it really matters unless you care about old gl apps) BR, -R _______________________________________________ devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx
