07:19 benjaminl: curious whether people have tried to measure perfetto clock synchronization accuracy in non-panfrost drivers, and whether they ran into a problem like this on?
07:19 benjaminl: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34390
07:19 benjaminl: *one
18:23 Lynne: suppose I have a buffer which gets written and read like: Dispatch; r+w barrier; Dispatch; r+w barrier; Dispatch
18:23 Lynne: for some reason the second dispatch doesn't see the changes the first dispatch made, unless I put a barrier before the first dispatch
18:24 Lynne: a barrier defines a clear dependency, doesn't it?
23:02 Company: apparently llvmpipe and lavapipe disagree on how to convert certain YUV values and now I get off-by one differences
23:09 Company: looks like it's https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/compiler/nir/nir_lower_tex.c#L50 vs https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/vulkan/runtime/vk_nir_convert_ycbcr.c#L86