04:21wens: just putting it out there, I'm working on converting the st7789v panel driver to use mipi-dbi for read/write (mostly done) and adding tinydrm support
23:12karolherbst: gallium question: when I have a resource that is used by multiple pipe_contexts and written to by shaders. Then if I want to map that resource, do I have to wait on all contexts to be done writing to it?
23:15airlied: karolherbst: yes unless it's unsync map
23:16karolherbst: mhhhh
23:16airlied: though maybe you just have to wait for the kernel instead of other contexts
23:16airlied: since they should fence it
23:17karolherbst: yeah.. tho the issue I'm debugging is that each CPU thread maps different parts of the resource
23:18karolherbst: on a different pipe_context
23:18karolherbst: also there is no expectations to see the writes from different contexts
23:21karolherbst: looks like using "PIPE_MAP_DIRECTLY" fixes the issue for me.. mhh
23:24zmike: every other type of map will probably use a staging resource
23:24zmike: so you'd need to wait and/or trigger a barrier for the readback copy
23:28karolherbst: yeah but my fallback path should be equally reliable, which it isn't...
23:32karolherbst: okay a flush + wait before the map also fixes it mhh..
23:34airlied: if you have internal references to it in a cmd buffer you are recording, then flush/wait is usually needed, but drivers should already be doing that
23:36karolherbst: yeah.. that's what I was assuming..
23:36karolherbst: but it's also only broken with iris/xe
23:56alyssa: there could easily be an iris/xe bug