On Tuesday, September 26, 2017 4:31:46 AM PDT Eero Tamminen wrote: > Hi, > > On 26.09.2017 12:04, Timothy Arceri wrote: > > On 26/09/17 17:50, Kenneth Graunke wrote: > [...] > >> 2. It would be great to use this in anv as well, especially given that > >> it has no GLSL IR linking to do this sort of stuff. Plus, the SPIR-V > >> might be compiled independently...and when you build the pipeline, > >> you have all the stages and can do cross-stage linking optimizations > >> like this. > > > > Yeah the reason for doing this was to eventually add it to radv. i965 > > was just a nicer place to test it, same goes for the passes that will > > follow. Right now the biggest blocked is justifying the added complexity > > as there isn't a standard shader-db like tool, and it's not making any > > noticeable differences in any benchmarks. > > I think the old GLBenchmark v2.5 / v2.7 tests (e.g. Egypt) might be > something where cross-stage linking optimizations impact could show up. > > > - Eero
Those aren't Vulkan benchmarks, though.
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev