00:50 <alyssa> Huh?

03:40 <alyssa> Just FWIW -- I'm on break from Panfrost from today to maybe a week from now

03:40 <alyssa> Ton of IRL stuff happening :)

03:40 <alyssa> I'll still be on IRC and whatever if people need to reach me (and I'll keep up on scrollback -- it's not, like, a break from the project, just a break from coding)

03:41 <alyssa> [I can maybe still do code review. But no new patches? I dunno]

03:50 <HdkR> alyssa: Sounds good :D

03:51 <alyssa> Guess I'll let my viewport transform patches sit on the list

03:51 <alyssa> I don't have time to deal with style bikeshedding.

03:52 <alyssa> (...If someone else would like to bikeshed, feel free)

05:03 <kszaq_> alyssa: since you wrote the code, is my assumption on LP64 ifdefs correct, i.e. only if LP64 is defined the code works with T8xx?

05:16 <HdkR> kszaq_: There's a high chance that there are some bad assumptions in some cases. Expectation was T8xx was going to be 64bit and all the testing was done with 64bit apps. There's probably a mismatch between 32bit userspace pointers and the hardware which operates on 64bit pointers

05:19 <kszaq_> Thanks. Are there any pointers passed to hardware other than in ioctls?

05:22 <HdkR> Depends on what you mean by that I guess. the job submission works through linked lists of objects. So they carry pointers as well

06:42 <tomeu> alyssa: ack, hope everything goes great IRL :)

12:30 <raster> ha! it boots finally

12:30 <raster> frigging dt is broken in mainline

12:36 <daniels> rastit is? how?

12:36 <daniels> *facepalm*

14:58 <alyssa> kszaq_: In theory, LP64 is for 64-bit and ->is_t8xx is for T8xx. In practice, we probably/definitely have bad assumptions in places where we mix those up, since I don't have traces of 64-bit T7xx or 32-bit T8xx

14:58 <alyssa> It might be possible to force 64-bit and have that work ok

14:59 <alyssa> I.e. change all the LP64's to something that is defined and maybe hope for the best

14:59 <alyssa> Mayhaps that's a clumsy workaround but hey

15:00 <alyssa> tomeu: Thank you!

15:30 <kszaq_> alyssa: Thank you. I'll see if I can implement some kind of workaround for my use case.

19:23 <alyssa> kszaq_: Best of luck :)

19:25 <MoeIcenowy> alyssa: Allwinner H6 is 64-bit T720

19:26 <alyssa> MoeIcenowy: OK?

19:26 <MoeIcenowy> alyssa: ?

20:15 <alyssa> MoeIcenowy: I don't have _traces_ from a 64-bit T720 target; I am aware there are boards I could get, but I don't have the resources to expand hw yet :)

21:08 <kszaq> Happy times: I replaced LP64 with something defined, then sed -i 's/uintptr_t/uint64_t/g' include/panfrost-job.h pandecode/decode.c and I have Kodi running on T820 in arm userspace. :)

21:10 <kszaq> yuv2rgb shader does not work correctly (colors are off) so I need to figure out whether all of the replacement were actually required

21:11 <kszaq> and when it's figured I will wrap it into something pretty - I guess a build-time parameter would be OK to override 64/32-bit GPU choice?

21:12 <kszaq> Thank you alyssa and HdkR for pointers and explanations, that helped me a lot. :)

21:14 <kszaq> I would appreciate if someone can take a look at https://gitlab.freedesktop.org/mesa/mesa/merge_requests/595 as it's required for ioctls to pass data to kernel correctly in mixed arch environment.

21:47 <alyssa> kszaq: Oh, awesome!!