The prime case is interesting ... esp since it's all completely manufactured. NVIDIA can render to both rbg/bgr orderings, and I suspect other GPUs can too. However for scanout, only one is supported. I don't have any great suggestions for how to resolve this cleanly, but figured I'd point it out.
On Fri, May 18, 2018 at 11:32 PM, Mario Kleiner <mario.kleiner...@gmail.com> wrote: > Hi, > > this series contains all the remaining patches for depth 30 support > on nouveau and possible future driver with xBGR2101010 ordering > instead of the xRGB2101010 ordering already supported since Mesa 18.0. > These are the ones i obsessively tested with success. > > First two patches are Daniel's, which have been reviewed and tested > by myself, Ilia and Eric, but for some reason Patchwork didn't detect > the R-b and t-b statements, so i added them to make it easier for it. > > The egl/x11 patch is identical to v2, just more tested on ati-ddx to > make sure the machine agrees with my explanation to v2 of why it should > work fine. > > Patches 4 and 5 are new: 4 Replaces the hard-coding of 2101010 formats > by actually detecting which formats are supported. Together with patch 5 > that is not only nicer, but may also provide more opportunity to page-flip > in some cases. > > Patch 5 is needed for Prime render offload from a Wayland server to a > Wayland client running on a different gpu, on at least Intel + NVidia, > but also helps to enable page-flipping more often. I tested this with > Weston + depth 30 enablement patches. While render offload always works > now, for page flipping (zero-copy present) i needed to disable the new > dmabuf-import path atm. on some gpu's, as Weston's current dmabuf import > and/or atomic modesetting is not yet ready to always allow page-flipping. > But that's a todo for Weston, this patch just enables it in principle. > > What is still missing in Mesa is proper Prime offload under regular X > between gpu's with mismatched channel ordering like Intel + NVidia, ie. > the equivalent of patches 4 and 5 for X11. Atm. render offload would simply > display with swapped red and blue channels in such a case. I probably > won't have time or ideas to look into this within the next month or two. > > Thanks, > -mario > _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
