01:19 <anarsoul|2> alyssa: hey, do you set any upper limit for tile heap size for panfrost?

01:32 <alyssa> anarsoul|2: Not particularly

01:32 <alyssa> Something very big(tm) and then just use growable memory in the kernel since it doesn't really matter

01:32 <alyssa> I don't know much about the tiler, at least on midgard+ the driver doesn't do very much and it mostly just works

01:33 <anarsoul|2> heh

01:33 <alyssa> (I don't think that logic changed at all since r/e'ing days)

01:34 <anarsoul|2> lima sets upper limit at 16M and some app is able to hit it: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4469

01:36 <alyssa> 16M might be low, not sure

01:37 <alyssa> I guess we set it to 16MB too

01:38 <alyssa> although midgard tiler might be more space efficient

01:38 <anarsoul|2> 16M max or 16M initial size?

01:39 <alyssa> max

01:39 <anarsoul|2> I see, thanks

01:40 <alyssa> but that should probably be raised tbh

01:40 <icecream95> IIRC Bifrost also compresses tiler data

01:42 <alyssa> wild

01:44 <anarsoul|2> sounds reasonable - it saves some memory bandwidth in addition to smaller memory footprint

01:50 <icecream95> Or maybe that's only on Valhall

07:47 <icecream95> Some Qt applications spend ~50% of CPU time on blend shader recompilation, because they alternate between two (more without another of my patches) sets of constants and `panfrost_compile_blend_shader` only skips compilation if the constants match from the previous time

07:53 <bbrezillon> icecream95: might be worth considering reserving a push constant for the constant color to avoid recompiling those shaders

08:19 <icecream95> bbrezillon: Fragment shaders rarely get close to the push constant limit so reserving some push constants shouldn't affect them too much

08:24 <bbrezillon> yes, and if that ever becomes a problem, we can always add FS variants for blend_shader+constant case

13:25 <alyssa> fwiw the blob does not, but it almost certainlydoes cache compiles

16:58 <macc24> alyssa: remember the !8698?

19:48 <alyssa> what about it

19:58 <macc24> alyssa: my comment

19:58 <macc24> those messages appear when device is under high memory pressure

19:58 <alyssa> yes, and?

19:58 <macc24> look at what fancy words i can use :D

19:59 <macc24> anyway, i saw those messages on 5.11-rcsomething on my odroid go advance

19:59 <macc24> alyssa: btw minetest is broken

20:02 <icecream95> macc24: Minetest was fixed recently, did it get broken again?

20:03 <macc24> icecream95: idk got a ping from cadmium user that it's broken

20:03 <macc24> and my duet is not with me now :(

20:05 <macc24> icecream95: they say that they cloned the repo hour ago

20:05 <macc24> so it probably broke

20:06 <macc24> aaand this was on mali t860........