07:53 <agx_> marex: yes, with older firmware bt broke wifi. Recent ones seem to have that fixed.

13:40 <marex> agx_: which are the recent ones ? the one in linux-firmware is recent enough ?

13:41 <marex> agx_: in fact, is there a version tag in the firmware blob ?

15:13 <dos1> yes, there is, the downstream driver has some logic for reading it out

15:14 <dos1> the 2.0RC6 repo has the latest firmware blob I'm aware of right now

15:20 <marex> dos1: why does RSI not update the firmware in linux-firmware ?

15:20 <marex> dos1: the logic in the downstream driver is for reading it out from a running firmware (it has to be running on the WiFi SoC)

20:05 <JohnnyonFlame> hmm... somehow my code is falling into some slow path that memcpies attributes, seems to happen here https://github.com/mesa3d/mesa/blob/master/src/gallium/auxiliary/util/u_vbuf.c#L840

20:06 <JohnnyonFlame> anyone knows what triggers this? I only have one GL_ARRAY_BUFFER in my entire program

20:07 <marex> JohnnyonFlame: using some weird pixelformat ?

20:07 <JohnnyonFlame> only 32bit floats

20:08 <JohnnyonFlame> seems like the worst case scenario is 14 floats per vertex, most are 7 to 12

20:08 <marex> isnt there something hitting ... err ... wait

20:08 <JohnnyonFlame> 12% of my entire execution time is sitting waiting memcpies :/

20:09 <marex> there was something like if the tiling/detiling isn't supported, it is done by software helper

20:10 <austriancoder> More vertex buffers are used than the hardware supports

20:10 <JohnnyonFlame> what are these "vertex buffers"?

20:11 <JohnnyonFlame> is it caused by glVertexAttribPointer or glBindBuffer?

20:12 <austriancoder> Did you used any search engine? https://en.m.wikipedia.org/wiki/Vertex_buffer_object

20:13 <JohnnyonFlame> :/

20:13 <JohnnyonFlame> I mean, I said a bit ago, only one array buffer was created

20:14 <JohnnyonFlame> I only have a single glGenBuffers and one bindBuffer call, since I'm streaming data to orphaned buffers

20:14 <austriancoder> JohnnyonFlame: what GPU?

20:14 <JohnnyonFlame> GC860

20:14 <JohnnyonFlame> testing on JZ4770 - RG350M and GCW0

20:19 <JohnnyonFlame> https://pastebin.com/xm6kXxyX

20:20 <austriancoder> one vertex buffer supported by that gpu

20:24 <austriancoder> count=2

20:26 <austriancoder> my crystal ball has no idea.. without the application (sources) it is quite hard to guess

20:29 <JohnnyonFlame> It's a modified version of this: https://github.com/sm64-port/sm64-port/blob/master/src/pc/gfx/gfx_opengl.c#L481

20:29 <JohnnyonFlame> I'm only pushing the changeset towards the end for reasons

20:31 <JohnnyonFlame> the added code just does texture atlasing (streaming textures to a binpacker) and encodes the necessary information on a vertex attribute, no added bindbuffers, anything

20:31 <marex> JohnnyonFlame: I think pcercuei also has this gcw zero

20:32 <marex> austriancoder: could it be some unsupported tiling format ?

20:32 <austriancoder> marex: I don't think so

20:32 <austriancoder> marex: I have one too :)

20:34 <marex> austriancoder: you are great

20:34 <austriancoder> not really..

21:30 <JohnnyonFlame> brb

22:02 <austriancoder> bed time

23:36 <JohnnyonFlame> I checked the available extensions and noticed that we have VAOs, so I implemented these in the code and I'm not hitting the slowpath anymore

23:36 <JohnnyonFlame> funnily enough, there are more VBOs now than before

23:36 <JohnnyonFlame> but I can't get a single memcpy hit from inside gallium

23:48 <JohnnyonFlame> spoke too early I guess, I just dont hit that specific breakpoint