01:33 <italove> newton688: sure

03:31 <cphealy> With vendor provided Mali kernel driver, there are some sysfs entries that allow seeing the memory usage by the 3D GPU. Is there equivalent functionality with the Panfrost driver and if not is it due to the functionality not yet being developed or some other limitation?

03:38 <alyssa> cphealy: It should be available but not sure off hand ho

03:38 <HdkR> Looks like i915 and amdgpu create some /sys/kernel/debug/dri/0/*_gem_* entries, but panfrost doesn't have the same interface. Would actually be quite nice

03:43 <alyssa> ^ there you go then

03:43 <alyssa> Probably just needs someone to wire it up then

03:44 <cphealy> HdkR: Oh my, I see what you are talking about on i915, there's a lot of stuff in there!

03:46 <HdkR> yea, i915 gives yo ua total at the top in gem_objects, amdgpu gives you all the objects so you'd need to accumulate the total yourself

03:54 <alyssa> HdkR: TIL /sys/kernel/debug/dri/1/clients

03:54 <alyssa> looks like some stuff I poked at on macOS :p

03:54 <HdkR> :D

07:44 <macc24> alyssa: minetest is broken

07:45 <macc24> and it's different commit than one that broke minecraft bedrock

07:56 <tomeu> cphealy: I think shadeslayer had some patches related to this?

13:42 <alyssa> macc24: f'naaa

14:39 <shadeslayer> cphealy: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2426

14:49 <shadeslayer> cphealy: the idea I was pursuing was that userspace applications could label objects with glLabel and we could pass those labels all the way down to the gem object level in the kernel, which would lead to better visibility when debugging issues with applications and tracking how much memory was allocated for various assets

15:51 <alyssa> Is that still being worked on?

15:55 <alyssa> shadeslayer: ^

16:06 <cphealy> shadeslayer: Interesting. I took a quick look at the MR and understand a bit of it. Would this allow seeing how much texture memory is used by an individual application? I ask as this is an area that can result in large amounts of memory usage.

16:06 <cphealy> Ideally, there would be a way to show uncompressed and compressed texture memory usage per application.

16:06 <cphealy> I'm not sure if this is possible though.

16:13 <robmur01> cphealy: there have also been a few patchsets on dri-devel from Android folks attempting to implement that kind of accounting - IIRC, so far they all seem to have got snagged on subtleties of dma-buf sharing

16:14 <cphealy> robmur01: Can you point me to any of these attempts by the Android folks??

16:21 <robmur01> This is possibly where it started - https://lore.kernel.org/dri-devel/CAKT=dDk0sNAXxz-angd5WvQXXLF8p3sPLEzOt=wVSLhuaP8dkQ@mail.gmail.com/

16:22 <robmur01> I've not been keeping track, but this looks like it might be a more recent evolution - https://lore.kernel.org/dri-devel/20210126204240.418297-1-hridya@google.com/

16:36 <newton688> italove: Should I share all the details of the problem on here or is it too much noise?

16:44 <robmur01> I'd imagine https://gitlab.freedesktop.org/mesa/mesa/-/issues is probably the best place for anything serious (as opposed to the things that we like to trivially moan about... :D)

16:45 <alyssa> #9001 DATA_INVALID_FAULT when running on Valhall

16:45 <alyssa> :p

16:48 <macc24> robmur01: define 'trivial'

16:49 <robmur01> macc24: pretty much anything I bring up, for starters ;)

16:59 <macc24> robmur01: i guess 'minecraft broke' would fall into that category

17:02 <robmur01> and if I ever get the CTS to work on Juno, well... :D

17:03 <macc24> robmur01: i could do regression testing manually :p

17:05 <macc24> i wonder what's that for http://lists.infradead.org/pipermail/linux-rockchip/2020-August/021780.html

17:05 <robmur01> in practice, it's for making the screen not look like a psychedelic colour mosaic

17:06 <macc24> oh, was that with PAN_MESA_DEBUG=afbc?

17:06 <macc24> i think i saw some afbc fail messages in dmesg when my kevin booted 5.12

17:07 <robmur01> once, but isn't AFBC default-on now?

17:07 <italove> robmur01: can anyone open issues? or does it require reporter permission? if so, newton688 might not be able to do so

17:07 <macc24> uhhhh, robmur: https://gitlab.freedesktop.org/mesa/mesa/-/blob/master/src/gallium/drivers/panfrost/pan_screen.c#L559

17:08 <robmur01> italove: I could, so AFAIK anyone can. I think you might need to be a reporter to assign labels, though

17:08 <macc24> italove: "submit issue" button isn't blacked out or anything on my accounr

17:08 <macc24> account*

17:10 <italove> ok

17:14 <robmur01> macc24: oh, fair enough - now I'm wondering how I managed to reproduce what looked like that issue :/

17:15 <macc24> after trying to fit 2 laptops, 3 monitors and a full sized keyboard i can understand why some people have giant desks

17:15 <cphealy> robmur01: Thanks for the links. They do look to overlap with what I'm looking for.

17:52 <chrisf> mess expands to fill whatever desk you have macc24

17:53 <macc24> chrisf: there is only 1450cm² of mess on my desk

17:53 <macc24> yes i measured it

18:40 <amonakov> just apply Banach-Tarski when you run out of space

18:40 <macc24> amonakov: unfortunately my desk isn't a sphere in vacuum

18:43 <alyssa> Moo.

19:40 <macc24> robmur01: i run PAN_MESA_DEBUG=gl3,afbc on my kevin lol