12:16dabrain34[m]: hello, is there a runner for macos M1 available on FDO ?
12:33ystreet00: I don't believe so
12:34ystreet00: there isn't any technical limitation to that, just no hardware available
13:30mupuf: DavidHeidelberg[m]: seems like the renaming of jobs broke running Mesa CI in forks? https://gitlab.freedesktop.org/hakzsam/mesa/-/jobs/43826233#L100
13:30mupuf: that leads to many more failures
13:31mupuf: wait, no, the problem is that we get an error 500 and kept going
13:49dabrain34[m]: <ystreet00> "there isn't any technical..." <- ylatuya: mentioned we can use rosetta to run a x86 app on M1
13:51slomo: dabrain34[m]: that's the other direction though (and there are limitations)
13:53__tim: you want to run it or just build it?
14:38DavidHeidelberg[m]: mupuf: u know I take this accusations really hard :D I'm always in 1second like "oh damn!"
14:39mupuf: Sorry about that :D
14:42eric_engestrom: dabrain34[m]: gitlab supports macos on apple silicon (https://docs.gitlab.com/runner/install/osx.html), but we don't have any such machine set up right now
14:43eric_engestrom: the two issues are that we need someone were to volunteer 1) the machine and 2) the time to maintain it
14:43eric_engestrom: s/were//
14:44dabrain34[m]: I dont know if there is overhead in term of performance to use x86 app on mac M1 ?
14:44__tim: the problem is also the VM setup, you don't really want a shell runner I think
14:44eric_engestrom: __tim: ah right, I didn't think about that, but that's a good point
14:45dabrain34[m]: for GStreamer we need the best performance we can get from the machine
14:45eric_engestrom: dabrain34[m]: there is overhead, but I have no idea how much 🤷
14:45dabrain34[m]: where is the x86 machine ?
14:45mupuf: The? There are hundreds :p
14:45eric_engestrom: dabrain34[m]: "where is the x86 machine?" -> what do you mean?
14:46dabrain34[m]: oki where are the machines ?
14:46mupuf: The main runners are in packet, a cloud provider
14:46__tim: in the cloud ;)
14:47__tim: the gstreamer mac runners are on macstadium in the Atlanta DC
14:47dabrain34[m]: ok i feel old school here :p sorry about that .. so this providers (cloud) can not provide M1 machine ?
14:48__tim: the hardware is not the problem, the CI setup with VMs is
14:48eric_engestrom: dabrain34[m]: there's basically 2 kinds of runners: the ones that have a specific hardware (eg. GPU) installed on them, and the "generic" ones
14:48eric_engestrom: generic ones are all in the cloud with various providers, and hardware-specific ones are almost all in some company's office or someone's home office
14:49mupuf: The cloud providers tend to limit the hardware they deploy, for ease of maintenance
14:49dabrain34[m]: (i'm not that old school then, remembering of old BuildBot times )
14:49mupuf: And they go for big core counts, not consumer hardware
14:49dabrain34[m]: but isnt it a copy of the x86 ones ?
14:50dabrain34[m]: the VM I mean
14:50mupuf: We use containers on linux
14:50mupuf: Do you guys have a Macos runner?
14:51mupuf: May also be a hackintosh
14:51__tim: gstreamer has mac runners yes (old x86 mac minis)
14:53dabrain34[m]: they are in macstadium ?
14:53__tim: correct
14:53dabrain34[m]: ok
14:53__tim: there are aspirations/plans to move to M1, but it's not trivial
14:55dabrain34[m]: I can imagine but definitvely need to be on the table
14:55mupuf: Marcan would like to have M1s too, but for Linux testing
14:55mupuf: Asahi*
15:10ylatuya[m]: dabrain34: You don't need and M1/M2 hardware to create aarch64 builds, you can cross-compile. GStreamer's x86_64 builds on aarch64 M1/M2 work perfectly fine, though. I don't have metrics but the overhead is barely noticeable, especially nowadays where a multimedia pipeline is fully hardware accelerated. LongoMatch users have been using these builds since the launch of M1's. I remember having to fix a bug in which we were connecting to a
15:10ylatuya[m]: signal a line after changing states instead of right before. This bug was only reproducible in M1's because they were a lot faster, all this with x86_64 build through Rosetta. Obviously, an aarch64 build will be faster (how much will depend on the workload of the app) but your GStreamer Pipeline Studio x86_64 macOS builds will run perfectly on an M1.
15:12ylatuya[m]: * dabrain34: You don't need M1/M2 hardware to create aarch64 builds, you can cross-compile. GStreamer's x86\_64 builds on aarch64 M1/M2 work perfectly fine, though. I don't have metrics but the overhead is barely noticeable, especially nowadays where a multimedia pipeline is fully hardware accelerated. LongoMatch users have been using these builds since the launch of M1's. I remember having to fix a bug in which we were connecting to a
15:12ylatuya[m]: signal a line after changing states instead of right before. This bug was only reproducible in M1's because they were a lot faster, all this with x86\_64 build through Rosetta. Obviously, an aarch64 build will be faster (how much will depend on the workload of the app) but your GStreamer Pipeline Studio x86\_64 macOS builds will run perfectly on an M1.
15:14dabrain34[m]: except that the gstreamer libs are also buit for x86_64, so I might need to cross compile for aarch64.
16:35mupuf: Oh yeah, that's a waste of exotic hardware to use a M1 to compile code
16:35mupuf: I thought it was to test hardware acceleration
16:36__tim:personally thinks it's not quite appropriate for a big open source project to use a hackintosh setup which violates the macOS license for its official CI when the project has the financial means for a proper setup, since we expect people to adhere to our licenses as well
16:45mupuf: __tim: too true, too true!
16:45mupuf: Can you run macOS in a VM legally though?
16:47__tim: aiui it's fine as long as the VM runs on Apple hardware, but I have to admit I did not investigate that in too much detail. Parallels etc build entire business models on it, so I assume it's fine ;)
16:48__tim: at least that's adherence in spirit then :D
16:55eric_engestrom: yeah my understanding was also that macos is only allowed to run on apple hardware, with whatever you want in between
17:29kxkamil: could someone look into mailing list problems for igt-dev ? some of our users cannot reach to ML
17:29kxkamil: I want to confirm if problem is on ML or on our server
17:31kxkamil: what is strange that some mails are reaching ML, few are not
18:39__tim: getting lots of 500 errors all of a sudden on image pull/prep, anything going on? :)
23:16ystreet00: for reference, other people have already figured out the gitlab-runner VM on M! mac problem and I also know how to solve that so that problem atm is only lack of hardware
23:17ystreet00: see https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1595 for some baackground
23:28ystreet00: yes, the Apple EULA allows 2 VM instances running on apple hardware and that is all