On Nov 5, 2016 1:19 PM, "Christian König" <deathsim...@vodafone.de> wrote: > > Am 04.11.2016 um 17:43 schrieb Marek Olšák: >> >> On Fri, Nov 4, 2016 at 5:23 PM, Alex Deucher <alexdeuc...@gmail.com> wrote: >>> >>> On Fri, Nov 4, 2016 at 10:57 AM, Marek Olšák <mar...@gmail.com> wrote: >>>> >>>> On Nov 4, 2016 3:29 PM, "Emil Velikov" <emil.l.veli...@gmail.com> wrote: >>>>> >>>>> On 31 October 2016 at 15:12, Marek Olšák <mar...@gmail.com> wrote: >>>>>> >>>>>> First, DRI_PRIME should work OK on radeon & amdgpu. I've been testing >>>>>> it with 2 GPUs in 2 PCIe slots. (only OpenGL though) >>>>>> >>>>>> I think the thread got derailed at the beginning and nobody asked the >>>>>> important question: >>>>>> >>>>>> Does radeonsi need PIPE_BIND_VIDEO_DECODE_OUTPUT that it would >>>>>> translate into linear? >>>>>> >>>>> Agreed things got derailed a bit. >>>>> >>>>> On the patch itself, can we get an ack/r-b, please? >>>>> It'll be nice to have nouveau working again, until a >>>>> better/alternative solution is found. >>>> >>>> I'm okay with the patch. My only concern is decoding performance. If the >>>> decoder works in 64x1 blocks, tiling will hurt. That's the theory. I don't >>>> know how the decoder works. >>> >>> Linear definitely performs better. Some early UVD blocks didn't even >>> support tiling IIRC. >> >> I've looked at the code and it looks like VdpOutputSurface is only >> used for final 3D passes and not as a decoding output, while >> VdpVideoSurface is used as a decoding output. Is that correct? > > > Yeah, that is correct. We don't use tiling with the VdpVideoSurface either because it isn't really beneficial for video decoding.
In that case, this whole discussion was pointless, because the linear layout is inefficient if it's only used by 3D and shouldn't have been used here in the first place. Marek > > Regards, > Christian. > >> >> Marek > > >
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
