02:05 <icecream95> It turns out the first four bytes of the AFBC header are not unused, but a kernel bug trigger: https://gitlab.freedesktop.org/-/snippets/1190/raw

07:48 <icecream95> macc24: I bisected STK advanced rendering pipeline problems to dc7fbe114ba ("panfrost: Drop implicit blend pooling")

07:48 <macc24> huh

07:51 <macc24> icecream95: and what would be next step in fixing this?

08:12 <icecream95> macc24: I think the commit should be reverted upstream

08:24 <icecream95> It seems blend shaders need the upper bits of PC to be equal between them, so need to be allocated near each other. After that commit, they are allocated separately and could have completely different addresses

09:21 <icecream95> macc24: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6785

09:24 <uis> alyssa: will arm provide datasheets?

10:16 <icecream95> LZDoom with the GL3.3 render crashes now because it manages to create a UBO that is far bigger than what is supported by hardware

10:22 <HdkR> 65k is supposed to be the GL minimum maximum so if it is that size it makes sense

10:34 <icecream95> HdkR: Looks like a panfrost bug - the ARB_uniform_buffer_object spec says that "The uniform buffer could be larger than the amount of uniform block(s) data inside it", and the shader only uses 32768 bytes, which is fine

10:37 <kaichi> Huge congrats to the panfrost team for the progress and the 'partnership' with ARM!

10:38 <icecream95> macc24: GL 3.3 just needs ARB_texture_buffer_object, ARB_timer_query and geometry shaders. Work on geometry shaders hasn't started yet, so you may need to be very patient...

10:38 <macc24> icecream95: i want to help but i don't know how so i just wait for now

10:54 <HdkR> icecream95: Yea, you just bind a base view of it

11:07 <icecream95> LZDoom also crashes in panfrost_emit_viewport after setting a viewport of (0, 0, 0, 0)

16:56 <alyssa> icecream95: Unfortunately the issue is more subtle.

16:57 <alyssa> I believe not only do blend shaders in a given draw need to be in the same 16MB block (i.e.: top 40-bits of GPU VA are the same) but so does the fragmen tshader

16:57 <alyssa> I meant to fix this a few weeks back but then Bifrost stuff happened.

16:58 <alyssa> AFAICT, it working before that commit is just as much of an accident as working after