#
…

2023-07-23

00:01fenix144 has quit [Remote host closed the connection]
00:03mikelee has joined #asahi-gpu
00:11mikelee has quit [Ping timeout: 480 seconds]
00:19mikelee has joined #asahi-gpu
00:28mikelee has quit [Ping timeout: 480 seconds]
00:50ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
00:54mikelee has joined #asahi-gpu
00:56ourdumbfuture has joined #asahi-gpu
01:02mikelee has quit [Ping timeout: 480 seconds]
01:13mikelee has joined #asahi-gpu
01:23mikelee has quit [Ping timeout: 480 seconds]
01:28ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
01:32mikelee has joined #asahi-gpu
01:42mikelee has quit [Ping timeout: 480 seconds]
01:45mikelee has joined #asahi-gpu
01:45ourdumbfuture has joined #asahi-gpu
01:53mikelee has quit [Ping timeout: 480 seconds]
02:01mikelee has joined #asahi-gpu
02:06the_sea_peoples has quit [Quit: WeeChat 2.8]
02:09mikelee has quit [Ping timeout: 480 seconds]
02:16mikelee has joined #asahi-gpu
02:19the_sea_peoples has joined #asahi-gpu
02:24mikelee has quit [Ping timeout: 480 seconds]
02:45jeisom has quit [Ping timeout: 480 seconds]
02:46marcanrosefromthedead: you probably need an isb, those counters aren't synchronized
02:46marcanalso there could be an offset on the virtual counter for some reason
02:47marcanalso I assume you're not running this on the hypervisor, right? that *will* screw with timekeeping on purpose by default
02:49mikelee has joined #asahi-gpu
02:52ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
02:57mikelee has quit [Ping timeout: 480 seconds]
03:01c10l56 has joined #asahi-gpu
03:05mikelee has joined #asahi-gpu
03:08c10l5 has quit [Ping timeout: 480 seconds]
03:12Hibyehello has quit [Ping timeout: 480 seconds]
03:15mikelee has quit [Ping timeout: 480 seconds]
03:24mikelee has joined #asahi-gpu
03:32mikelee has quit [Ping timeout: 480 seconds]
03:46tristan2 has joined #asahi-gpu
03:48tristan2_ has quit [Ping timeout: 480 seconds]
03:57mikelee has joined #asahi-gpu
04:32mikelee has quit [Ping timeout: 480 seconds]
04:44mikelee has joined #asahi-gpu
05:18mikelee has quit [Ping timeout: 480 seconds]
05:24mikelee has joined #asahi-gpu
05:34mikelee has quit [Ping timeout: 480 seconds]
05:39mikelee has joined #asahi-gpu
06:25nela has quit [Ping timeout: 480 seconds]
06:31mikelee has quit [Ping timeout: 480 seconds]
06:44mikelee has joined #asahi-gpu
07:32crabbedhaloablut has joined #asahi-gpu
08:57hightower2 has joined #asahi-gpu
09:18deflated8837_ has joined #asahi-gpu
09:18deflated8837 has quit [Read error: Connection reset by peer]
09:40jeisom has joined #asahi-gpu
09:41i509vcb has quit [Quit: Connection closed for inactivity]
09:57jeisom has quit [Ping timeout: 480 seconds]
10:03chadmed has quit [Read error: Connection reset by peer]
10:03chadmed has joined #asahi-gpu
10:15mikelee has quit [Ping timeout: 480 seconds]
10:32mikelee has joined #asahi-gpu
10:33rosefromthedeadmarcan: not on purpose, just a normal asahi install
10:33rosefromthedeadwill try with an extra isb, i already have one before the read but ig it would make sense to need ones either side?
10:35marcanyeah, otherwise it could in principle speculate through the actual work submission to the GPU and all that
10:36marcanI've seen these damn things speculate through an IRQ enough to break some code missing a barrier, it's nuts
10:36marcanfound a years-old Linux core workqueue code bug
10:36marcanno other ARM64 CPU managed to hit it apparently
10:37svensame thing inside the dwc3 driver
10:37rosefromthedeadhmm, still getting some negative durations, but i think less now?
10:37rosefromthedeadi'll check for an offset
10:37sventhere was this one subtle role switch bug where they read the new role outside of the mutex or something
10:38sventriggering that requires both a very fast role switch and a very fast cpu that likes to speculate a lot
10:40mikelee has quit [Ping timeout: 480 seconds]
10:41marcanrosefromthedead: what machine?
10:42marcanno ECV then, so no second offset
10:44marcanrosefromthedead: I wonder if there's just a slight offset with the GPU timer for some reason. I don't really know what the internal architecture is like. I know there's one master AIC timer which I think everything derives from, but not how all the downstream timers are synced from that
10:46marcanhow negative are you getting?
10:47rosefromthedeadonly on startup of tinywl for some reason, -3150us, -94us, and everything positive from there
10:48rosefromthedeadit's also possible that my code is just mixing up values from different batches or something
10:48rosefromthedeadnot sure how conscious i was when i wrote it
10:49mikelee has joined #asahi-gpu
10:58ourdumbfuture has joined #asahi-gpu
10:59ourdumbfuture has quit []
11:00mikelee has quit [Ping timeout: 480 seconds]
11:09mikelee has joined #asahi-gpu
11:18mikelee has quit [Ping timeout: 480 seconds]
11:26mikelee has joined #asahi-gpu
11:34mikelee has quit [Ping timeout: 480 seconds]
11:44mikelee has joined #asahi-gpu
11:52mikelee has quit [Ping timeout: 480 seconds]
12:05Z750 has quit [Quit: Ping timeout (120 seconds)]
12:08Z750 has joined #asahi-gpu
12:15ourdumbfuture has joined #asahi-gpu
12:18mikelee has joined #asahi-gpu
12:26mikelee has quit [Ping timeout: 480 seconds]
12:37mikelee has joined #asahi-gpu
12:49mikelee has quit [Ping timeout: 480 seconds]
13:13mikelee has joined #asahi-gpu
13:24mikelee has quit [Ping timeout: 480 seconds]
13:51mikelee has joined #asahi-gpu
13:59mikelee has quit [Ping timeout: 480 seconds]
14:01nela has joined #asahi-gpu
14:08mikelee has joined #asahi-gpu
14:16mikelee has quit [Ping timeout: 480 seconds]
14:17mikelee has joined #asahi-gpu
14:25jeisom has joined #asahi-gpu
14:26nela has quit [Ping timeout: 480 seconds]
14:26mikelee has quit [Ping timeout: 480 seconds]
14:34ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
14:35nela has joined #asahi-gpu
14:42alyssa has joined #asahi-gpu
14:42alyssalina: I see the uapi lets you submit multiple commands in one ioctl .. is this tested?
14:43alyssaMesa doesn't currently use it so IDK if that kernel path was ever exercised
14:43alyssaI have a branch using it but am getting a race condition, second command isn't waiting for the first one to complete
14:43alyssaunsure if this is my bug or yours
14:43nela has quit [Ping timeout: 480 seconds]
14:43alyssa(When the two commands are in separate ioctls with syncing it works)
14:44alyssaspecifically, I want to submit both a compute job and a render job in the same ioctl, where the render job depends on the compute
14:45alyssa(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.)
14:45alyssaIs this.. supposed to work though?
14:46alyssaspecifically, I don't understnad the drm_asahi_cmd barriers
14:46ourdumbfuture has joined #asahi-gpu
14:47alyssaI want a barrier between my second command and my first. Can I express that?
14:47alyssaIf 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
14:50alyssaI'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
14:51mikelee has joined #asahi-gpu
14:56alyssaif this is a kernel bug (rather than something that's just not supported yet) I can push a mesa branch to test against
15:08nela has joined #asahi-gpu
15:10mikelee has quit [Ping timeout: 480 seconds]
15:12lurk has joined #asahi-gpu
15:16lurker has joined #asahi-gpu
15:16lurker has quit []
15:16cyka_delik has joined #asahi-gpu
15:17lurk has quit [Quit: Page closed]
15:18linaalyssa: It should work. You need to specify the barriers correctly for it to work, yes.
15:21mikelee has joined #asahi-gpu
15:22linaalyssa: 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
15:23linaIf 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
15:23linaSo not the compute command that precedes it in the same submission
15:27linaAlso 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)
15:30mikelee has quit [Ping timeout: 480 seconds]
15:35alyssaright, ok
15:35alyssaI don't quite understand what {NONE, 1} does
15:35linait means wait for 1 compute command to be done
15:35linaand don't wait for anything render
15:36Z750 has quit [Quit: Ping timeout (120 seconds)]
15:36alyssathey're relative to the ioctl ooookay
15:36alyssagot it got it
15:37alyssaclipboard fail
15:37* alyssawhistles
15:37alyssa"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."
15:37linawwwwwwwwwwww
15:37alyssaI've got a use case :>
15:38linaAssuming we don't find a magic mesh shader thing to make it work ^^
15:38linaI should trace that...
15:38alyssamake which work?
15:38alyssainterleaved compute/vertex?
15:38alyssait would be really nice if they added it
15:38linaYeah ^^
15:38alyssaMali can do it, it's not fundamentally hard
15:38alyssaor particularly magic
15:39mikelee has joined #asahi-gpu
15:39alyssayou just.. merge the VDM and CDM into one big queue. that's it.
15:39alyssaPVR can't do it though, I asked
15:39linaI should sleep but I can try looking at this later ^^
15:40alyssatake your time, I have work tomorrow and will have to put this branch down :V
15:40linaAlso 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 ^^
15:40alyssayeah, tomorrow
15:41alyssathis isn't a work day for me it's a guilty agx pleasure day, no review :>
15:43Z750 has joined #asahi-gpu
15:52mikelee has quit [Ping timeout: 480 seconds]
15:53cyka_delik has quit [Remote host closed the connection]
16:12i509vcb has joined #asahi-gpu
16:17mikelee has joined #asahi-gpu
16:23jeisom has quit [Quit: Leaving]
16:24jeisom has joined #asahi-gpu
16:27mikelee has quit [Ping timeout: 480 seconds]
16:31mikelee has joined #asahi-gpu
16:39jeisom has quit [Ping timeout: 480 seconds]
16:40cylm has quit [Quit: WeeChat 3.6]
16:41mikelee has quit [Ping timeout: 480 seconds]
16:46mikelee has joined #asahi-gpu
16:54mikelee has quit [Ping timeout: 480 seconds]
17:02mikelee has joined #asahi-gpu
17:07ella-0alyssa: the agxv queue stuff uses multiple commands per ioctl
17:11mikelee has quit [Ping timeout: 480 seconds]
17:17mikelee has joined #asahi-gpu
17:28mikelee has quit [Ping timeout: 480 seconds]
17:35ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
17:35zalyx0 has quit [Quit: later alligator]
17:37cylm has joined #asahi-gpu
17:38mikelee has joined #asahi-gpu
17:44Cyrinux947 has joined #asahi-gpu
17:45puyanlotfi has joined #asahi-gpu
17:46Cyrinux94 has quit [Ping timeout: 480 seconds]
17:48mikelee has quit [Ping timeout: 480 seconds]
18:15mikelee has joined #asahi-gpu
18:23mikelee has quit [Ping timeout: 480 seconds]
18:25nsklaus has joined #asahi-gpu
18:32mikelee has joined #asahi-gpu
18:40mikelee has quit [Ping timeout: 480 seconds]
18:49mikelee has joined #asahi-gpu
18:50ourdumbfuture has joined #asahi-gpu
19:00ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
19:00mikelee has quit [Ping timeout: 480 seconds]
19:01ourdumbfuture has joined #asahi-gpu
19:11mikelee has joined #asahi-gpu
19:19mikelee has quit [Ping timeout: 480 seconds]
19:27mikelee has joined #asahi-gpu
19:41mikelee has quit [Ping timeout: 480 seconds]
19:44possiblemeatball has joined #asahi-gpu
20:01possiblemeatball has quit [Quit: Quit]
20:06mikelee has joined #asahi-gpu
20:15mikelee has quit [Ping timeout: 480 seconds]
20:20jeisom has joined #asahi-gpu
20:29mkurz has joined #asahi-gpu
20:30merry has joined #asahi-gpu
20:40fenix144 has joined #asahi-gpu
20:41mikelee has joined #asahi-gpu
20:49mikelee has quit [Ping timeout: 480 seconds]
21:03fenix144 has quit [Ping timeout: 480 seconds]
21:08zalyx0 has joined #asahi-gpu
21:09ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
21:12zalyx0 has quit []
21:12ourdumbfuture has joined #asahi-gpu
21:12fenix144 has joined #asahi-gpu
21:13zalyx0 has joined #asahi-gpu
21:28fenix144 has quit [Ping timeout: 480 seconds]
21:29bluetail has left #asahi-gpu [The Lounge - https://thelounge.chat]
21:33mikelee has joined #asahi-gpu
21:51mikelee has quit [Ping timeout: 480 seconds]
21:55ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
21:59ourdumbfuture has joined #asahi-gpu
22:09ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
22:18mikelee has joined #asahi-gpu
22:22ourdumbfuture has joined #asahi-gpu
22:28mikelee has quit [Ping timeout: 480 seconds]
22:34puyanlotfi has quit [Quit: Connection closed for inactivity]
23:01ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
23:03ourdumbfuture has joined #asahi-gpu
23:14mikelee has joined #asahi-gpu
23:23mikelee has quit [Ping timeout: 480 seconds]
23:30ourdumbfuture has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
23:35ourdumbfuture has joined #asahi-gpu
23:43mikelee has joined #asahi-gpu
23:51mikelee has quit [Ping timeout: 480 seconds]
23:52nsklaus has quit [Ping timeout: 480 seconds]
23:59mikelee has joined #asahi-gpu
…