On Sun, Nov 16, 2014 at 7:33 PM, Matt Turner <matts...@gmail.com> wrote: > On Sun, Nov 16, 2014 at 5:51 PM, Thomas Helland > <thomashellan...@gmail.com> wrote: >> This allows the backend to decide if it does not >> want saturates, or if it wants to combine min/max >> together by itself. >> >> Usefull for drivers that implement saturate with min/max >> as it can allow for some optimizations by min/max-pruning. >> Drivers like freedreno and vc4 will benefit. >> --- >> I have not made any drivers set the option. >> Probably some pipe-cap or something is needed >> in gallium to allow this? >> I could possibly find time to do a follow-up >> patch for this if someone leads me in the right direction. >> >> saturate() in ir_builder generates ir_unop_saturate unconditionally. >> If someone comes along later and does saturate() >> in some optimization pass, whithout checking the options, >> they will mess up possible optimizations for some backends. >> The cleanest way I could find for always doing the right >> thing was to write a lowering pass in opt_algebraic to >> lower ir_unop_sature into max(min(x, 1), 0); > > We already have that, so no need for work in that area :) > > See lower_instructions_visitor::sat_to_clamp().
... and as a result the second hunk of this patch isn't needed. _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
