ci: Run the tests on macOS.
Merge request reports
Activity
added 2 commits
added 2 commits
added 2 commits
added 2 commits
added 58 commits
-
7cecedd5...45541dd9 - 56 commits from branch
wine:master
- d72c91ef - ci: Run the tests on macOS.
- 3113f167 - ci: Do not drop artifacts after 1 day.
-
7cecedd5...45541dd9 - 56 commits from branch
I don't think this is the most appropriate place to detect MoltenVK regressions, because we don't use MoltenVK from git, but a released version, which I think is installed through brew. I don't even think it is the latest released version. At any rate, even when there is a regression I don't have an automated way to know whether what's broken is vkd3d or MoltenVK. Also, notice that currently vkd3d is broken in many ways on top of MoltenVK (the CI job is marked as allowed to fail for this reason); test
d3d12
even segfaults (I am preparing some patches for that). Before this can give a useful signal vkd3d tests should be appropriately marked as todo on MoltenVK, so that actual breakages can be detected. Eventually I hope to do that, but it might take a while (for example, I've been planning to do the same for Intel and NVIDIA on Linux, which have similarly been broken for a long time if not forever; and it hasn't happened yet). Of course we'll be happy to take patches, if you have some bandwidth for that.To address my first concern, I'd suggest to avoid using this specific project, but rather set up another one that tracks both vkd3d and MoltenVK development branches, and run the tests each time any of the two projects is committed to: this might help to give an idea of which project broke it when there is a problem. Or maybe run the stable vkd3d on the development MoltenVK, I don't know. I can help to set this up, but marking the tests as todo must be done before.
BTW, the macOS runner available here is based on an Intel chip. In your experience, can testing on Intel be considered representative of the M1 behavior too, or should we come up with an M1 runner too?
BTW, the macOS runner available here is based on an Intel chip. In your experience, can testing on Intel be considered representative of the M1 behavior too, or should we come up with an M1 runner too?
I think you should probably have an Apple Silicon runner, too. For one thing, unlike the GPUs on Intel Macs, which are immediate-mode, Apple's GPUs are all TBDR. This has actually affected behavior in some cases--I believe someone at CW submitted a patch to SPIRV-Cross to add inline texture barriers because Apple's GPUs required them to work properly. Also, I'm pretty sure that Apple's Metal driver for Apple Silicon gets more attention than their Intel and AMD drivers...