Am 08.10.2015 um 16:46 schrieb Marek Olšák: > > On Oct 8, 2015 4:03 PM, "Roland Scheidegger" <srol...@vmware.com > <mailto:srol...@vmware.com>> wrote: >> >> Was that because stencil used sampler 1 but sview 0 which looks >> definitely wrong as there needs o be a 1:1 mapping between these when >> the tex opcodes are used? > > I don't remember how the SVIEW declarations work. It was TEX and there > were 2 declarations, one was float and the other one was int. I think > it's only required for freedreno with TEX.
I was just saying the sampler index (for stencil texturing) was 1, but the sview declaration was only for index 0. Thus probably the reason why it broke (llvmpipe should handle sview declarations just fine even with tex opcodes, but if you have them they MUST be correct and they must be present for either all or none of the samplers). Thus what probably happened was since there was a sview declaration, llvmpipe tried to figure out the texture from it rather than using the sampler decl, but there wasn't any for index 1 (thus skipping texturing completely). But in any case, I'm fine without using them (though I forgot if freedreno would require it for this particular case). Roland > Marek > >> >> Roland >> >> Am 08.10.2015 um 10:45 schrieb Marek Olšák: >> > The problem was that it declared SVIEW but llvmpipe didn't like that for >> > some reason. >> > >> > Marek >> > >> > On Oct 8, 2015 5:12 AM, "Roland Scheidegger" <srol...@vmware.com > <mailto:srol...@vmware.com> >> > <mailto:srol...@vmware.com <mailto:srol...@vmware.com>>> wrote: >> > >> > Am 08.10.2015 um 02:02 schrieb Marek Olšák: >> > > Hi Brian, >> > > >> > > This is the commit that broke DrawPixels on llvmpipe, but not > radeonsi >> > > and softpipe. It's fixed in v2, which is attached. >> > > >> > > Marek >> > > >> > > On Tue, Oct 6, 2015 at 3:26 AM, Marek Olšák <mar...@gmail.com > <mailto:mar...@gmail.com> >> > <mailto:mar...@gmail.com <mailto:mar...@gmail.com>>> wrote: >> > >> From: Marek Olšák <marek.ol...@amd.com > <mailto:marek.ol...@amd.com> <mailto:marek.ol...@amd.com > <mailto:marek.ol...@amd.com>>> >> > >> >> > >> - there is no connection to user fragment shaders, so having > these as >> > >> shader variants makes no sense >> > >> - don't use Mesa IR, use TGSI >> > >> - don't create gl_fragment_program, just create the shader CSO >> > >> --- >> > >> >> > >> > I don't quite get from the differences, was the shader actually >> > incorrect or is this something which should be fixed in llvmpipe? >> > >> > Roland >> > >> > _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
