Cool. Do you think something like this is check-in-able behind a ST_PRECOMPILE=1 env var (or something that fits well into the current env var scheme, dunno what it would be)?
On Fri, Feb 7, 2014 at 12:46 PM, Marek Olšák <mar...@gmail.com> wrote: > FYI, I found the patch which compiles some shader variants in glLinkProgram: > > http://cgit.freedesktop.org/~mareko/mesa/commit/?h=precompile&id=8dc586297e4902dfc14cf8001b8202db4fdc9447 > > Marek > > On Tue, Jan 21, 2014 at 12:56 AM, Marek Olšák <mar...@gmail.com> wrote: >> It was pretty simple IIRC. You just need to call st_get_fp(vp)_variant >> to create a pipe shader. I think st_program_string_notify is the right >> place. >> >> Marek >> >> On Mon, Jan 20, 2014 at 7:23 PM, Ilia Mirkin <imir...@alum.mit.edu> wrote: >>> On Mon, Jan 20, 2014 at 10:23 AM, Brian Paul <bri...@vmware.com> wrote: >>>>> I'm having trouble figuring out how to bypass this and make it call >>>>> update_vp/fp/gp which appear like they'll actually call >>>>> pipe->create_fs_state/etc. I see st_link_shader getting called, as >>>>> expected, which creates the TGSI... but what to do from there? >>>> >>>> >>>> If you were to short-circuit the validation steps above and force the VS/FS >>>> to get propagated to the driver ASAP, there's a chance the shader wouldn't >>>> actually do everything that it normally would (ex: invert Y). I don't know >>>> if that matters to you. >>> >>> Not at all. This is for checking various nouveau compiler >>> optimizations against input shaders. I think that as long as the >>> shaders passed to the driver remain basically the same, it should >>> still be useful for comparing emitted code with different (nouveau) >>> compiler versions. >>> >>>> >>>> In any case, there's no simple solution to this. You'll probably have to >>>> hack something up. If it turns out simple, maybe we could enable it with a >>>> debug flag. >>> >>> Could you suggest a way for hacking this up? i.e. what functions might >>> I call from where? I'm definitely having trouble groking the code >>> paths that cause things to be sent right now. >>> >>> Thanks, >>> >>> -ilia >>> _______________________________________________ >>> mesa-dev mailing list >>> mesa-dev@lists.freedesktop.org >>> http://lists.freedesktop.org/mailman/listinfo/mesa-dev _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
