#
#asahi-gpu · 2023-07-23 17:07
ella-0alyssa: the agxv queue stuff uses multiple commands per ioctl
#asahi-gpu · 2023-07-23 15:41
alyssathis isn't a work day for me it's a guilty agx pleasure day, no review :>
#asahi-gpu · 2023-07-23 15:41
linaOK ^^
#asahi-gpu · 2023-07-23 15:40
alyssayeah, tomorrow
#asahi-gpu · 2023-07-23 15:40
linaAlso can you look at the shadowing MR? It's kind of a nasty bug/thing but the fix is easy, I want to get it fixed ^^
#asahi-gpu · 2023-07-23 15:40
alyssatake your time, I have work tomorrow and will have to put this branch down :V
#asahi-gpu · 2023-07-23 15:39
linaI should sleep but I can try looking at this later ^^
#asahi-gpu · 2023-07-23 15:39
alyssaPVR can't do it though, I asked
#asahi-gpu · 2023-07-23 15:39
alyssayou just.. merge the VDM and CDM into one big queue. that's it.
#asahi-gpu · 2023-07-23 15:38
alyssaor particularly magic
#asahi-gpu · 2023-07-23 15:38
alyssaMali can do it, it's not fundamentally hard
#asahi-gpu · 2023-07-23 15:38
linaYeah ^^
#asahi-gpu · 2023-07-23 15:38
alyssait would be really nice if they added it
#asahi-gpu · 2023-07-23 15:38
alyssainterleaved compute/vertex?
#asahi-gpu · 2023-07-23 15:38
alyssamake which work?
#asahi-gpu · 2023-07-23 15:38
linaI should trace that...
#asahi-gpu · 2023-07-23 15:38
linaAssuming we don't find a magic mesh shader thing to make it work ^^
#asahi-gpu · 2023-07-23 15:37
alyssaI've got a use case :>
#asahi-gpu · 2023-07-23 15:37
linawwwwwwwwwwww
#asahi-gpu · 2023-07-23 15:37
alyssa"Right now, the Gallium driver only issues single-command submissions and always uses [C0, R0] barriers to fully serialize everything. This might be improved in the future, but the main use case for all this is Vulkan queues."
#asahi-gpu · 2023-07-23 15:37
alyssawhistles
#asahi-gpu · 2023-07-23 15:37
alyssaclipboard fail
#asahi-gpu · 2023-07-23 15:37
alyssaerm
#asahi-gpu · 2023-07-23 15:37
alyssa> https://ok.reddit.com/r/egg_irl/comments/157bmhk/egg_irl/
#asahi-gpu · 2023-07-23 15:36
alyssagot it got it
#asahi-gpu · 2023-07-23 15:36
alyssathey're relative to the ioctl ooookay
#asahi-gpu · 2023-07-23 15:36
alyssaOhh
#asahi-gpu · 2023-07-23 15:35
linaand don't wait for anything render
#asahi-gpu · 2023-07-23 15:35
linait means wait for 1 compute command to be done
#asahi-gpu · 2023-07-23 15:35
alyssareads
#asahi-gpu · 2023-07-23 15:35
alyssaI don't quite understand what {NONE, 1} does
#asahi-gpu · 2023-07-23 15:35
alyssaright, ok
#asahi-gpu · 2023-07-23 15:31
linaAlso I did document all this: https://github.com/AsahiLinux/docs/wiki/SW:AGX-driver-notes#queues ^^
#asahi-gpu · 2023-07-23 15:27
linaAlso even with two separate ioctls you shouldn't need syncobjs as long as you still have the {0, 0} barrier (which maps to BARRIER_RENDER | BARRIER_COMPUTE in the existing code)
#asahi-gpu · 2023-07-23 15:23
linaSo not the compute command that precedes it in the same submission
#asahi-gpu · 2023-07-23 15:23
linaIf you were passing 0,0 for both then that wouldn't work, that would tell the render command to wait for any render/compute commands *prior* to the submission you're making right now
#asahi-gpu · 2023-07-23 15:22
linaalyssa: The first (compute) command needs to specify the barriers as {0, 0} to wait for any previous submission (implicitly, we always do this in GL, no need for sync objects), and the second (render) command needs to specify the barriers as {NONE, 1} to wait for the compute command to complete
#asahi-gpu · 2023-07-23 15:18
cyka_deliktest
#asahi-gpu · 2023-07-23 15:18
linaalyssa: It should work. You need to specify the barriers correctly for it to work, yes.
#asahi-gpu · 2023-07-23 14:56
alyssaif this is a kernel bug (rather than something that's just not supported yet) I can push a mesa branch to test against
#asahi-gpu · 2023-07-23 14:50
alyssaI've hacked around it in Mesa for now (submitting two ioctls back to back, putting the in syncs on the first and the out syncs on the second) but I shouldn't have to, I don't think
#asahi-gpu · 2023-07-23 14:47
alyssaIf not under the current uapi, can that be added? Working around this in Mesa would be annoying because it splits up the syncobjs and such across multiple ioctls when ideally it would be just one
#asahi-gpu · 2023-07-23 14:47
alyssaI want a barrier between my second command and my first. Can I express that?
#asahi-gpu · 2023-07-23 14:46
alyssaspecifically, I don't understnad the drm_asahi_cmd barriers
#asahi-gpu · 2023-07-23 14:45
alyssaIs this.. supposed to work though?
#asahi-gpu · 2023-07-23 14:45
alyssa(this requires a big refactor of how we submit batches in Mesa, but it seems far less ugly than the alternatives since it avoids messing with delicate batch tracking logic. and its probably faster.)
#asahi-gpu · 2023-07-23 14:44
alyssaspecifically, I want to submit both a compute job and a render job in the same ioctl, where the render job depends on the compute
#asahi-gpu · 2023-07-23 14:43
alyssa(When the two commands are in separate ioctls with syncing it works)
#asahi-gpu · 2023-07-23 14:43
alyssaunsure if this is my bug or yours
#asahi-gpu · 2023-07-23 14:43
alyssaI have a branch using it but am getting a race condition, second command isn't waiting for the first one to complete

plus ›