On 04/08/2015 01:46 PM, Jason Ekstrand wrote: > On Wed, Apr 8, 2015 at 12:53 PM, Ian Romanick <i...@freedesktop.org> wrote: >> On 04/08/2015 02:25 AM, Martin Peres wrote: >>> On 08/04/15 10:06, Kenneth Graunke wrote: >>>> Previously, we translated into NIR and did all the optimizations and >>>> lowering as part of running fs_visitor. This meant that we did all of >>>> that work twice for fragment shaders - once for SIMD8, and again for >>>> SIMD16. We also had to redo it every time we hit a state based >>>> recompile. >>>> >>>> We now generate NIR once at link time. ARB programs don't have linking, >>>> so we instead generate it at ProgramStringNotify time. >>>> >>>> Mesa's fixed function vertex program handling doesn't bother to inform >>>> the driver about new programs at all (which is rather mean), so we >>>> generate NIR at the last minute, if it hasn't happened already. >>>> >>>> shader-db runs ~9.4% faster on my i7-5600U, with a release build. >>> >>> Nice speed improvement but wouldn't it affect negatively programs using >>> SSO to recombine shaders at run time? >> >> Hm... that's a fair question. Does NIR do any cross-stage optimization? > > Not at the moment. We probably should since NIR can probably > dead-code things better.
Okay. Then this shouldn't impact SSO. _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
