00:47 <marex> [ 1.052834] etnaviv-gpu 38000000.gpu: model: GC7000, revision: 6204

02:13 <marex> austriancoder: I have mx8mp gpu status if you need it for gpu comparison

17:44 <marex> lynxeye: you mentioned the 8mp gpu doesnt contain blt engine ? seems the minorfeatures5 has blt bit set, but maybe it means something else ?

17:46 <lynxeye> marex: Not sure what you are looking at. The entry in the HWDB is GC7000 r6204 for the 8mp GPU, which has no BLT bit set.

17:46 <lynxeye> And I've validated that the GPU has the RS engine over the last few days ;)

17:47 <marex> lynxeye: so you plan to roll out patches soon then ?

17:47 <marex> lynxeye: btw re point sprite stuff, see https://gitlab.freedesktop.org/marex/mesa/-/commit/3e889fa38673a0986867e532ef8781b65821e5aa

17:48 <lynxeye> Yes, I haven't figured out the new RS supertiling yet, but things work with some small adjustments and supertiling disabled for now.

17:49 <lynxeye> Do you have a 8MP-EVK?

17:50 <marex> lynxeye: yep, so I obviously wanted to try etnaviv, I got the GPU to bind to etnaviv and started poking around

17:50 <marex> but since you're already on it, maybe neverball logo transparency is another thing I can look at

17:55 <lynxeye> marex: What's wrong with the logo? TBH all those neverball issues did mostly just scroll by me...

17:55 <marex> lynxeye: it is supposed to be transparent, but it is not transparent

17:55 <marex> lynxeye: seems gl1/gles1 is a source of fun

17:56 <marex> lynxeye: some alpha/depth test voodoo again

17:56 <lynxeye> marex: If it's alpha testing then GC7000 (at least most of them) is currently broken.

17:57 <marex> lynxeye: its broken on mx6q too, so no

17:57 <marex> lynxeye: I need to look into how its implemented and then figure out whats wrong with it

17:58 <lynxeye> Seems Vivante removed the fixed-function alpha-test unit and replaced it with a shader variant. There's a feature bit for called PE_NO_ALPHA_TEST

17:59 <lynxeye> Oh, okay. GC2000/3000 still has the alpha test unit.

18:04 <marex> lynxeye: the shader based gles2 depth/alpha stuff works fine, the deqp tests pass and all

18:04 <marex> this is again some weird ff stuff

18:09 <flto> marex: alpha-test isn't in gles2, so deqp isn't testing it

18:09 <marex> flto: depth stencil is

18:10 <marex> flto: alpha is only in gles3 ?

18:15 <lynxeye> marex: alpha testing is a desktop GL feature

18:32 <lynxeye> marex: since you are testing neverball, could you check if https://gitlab.freedesktop.org/mesa/mesa/-/issues/3422 is fixed with recent mesa?

18:42 <marex> lynxeye: gc7000l is tested by mntmn ^

18:53 <mntmn> mmmm 8mp

18:54 <mntmn> marex: fwiw i "fixed" alphatest in minetest by submitting a patch to them that replaces it with discard on etnaviv...

18:56 <lynxeye> mntmn: this replacement should be done by the driver on GPUs without the alpha test unit, instead of requiring application level patches

19:40 <marex> mntmn: link ?

19:41 <marex> mntmn: also I have the unit on 6q, so that's likely unrelated

20:22 <marex> gl-1.0-drawpixels-depth-test seems to be the piglit which fails

20:23 <marex> also, thats likely why the neverball logo fails

20:43 <mntmn> i know, but that was the faster way to get it done ;)

20:43 <mntmn> marex: https://github.com/minetest/minetest/pull/10036

20:43 <mntmn> marex: the discussion there is fun too ;)