01:54 <alyssa> urjaman: Nice!

09:24 <urjaman> O.O

09:25 <urjaman> anyways the case of the missing inside faces of the cube that was cut out from the bigger cube

09:26 <urjaman> i moved the cut cube like halfway outside the bigger cube, so it only cuts out a smaller cube and ....

09:27 <urjaman> the area of the inside faces of the cut-out small cube that is outside the big cube (and thus should not exist) is rendered

09:27 <urjaman> but the area tat is inside the big cube (that should be rendered) is not

09:40 <urjaman> https://urja.dev/ya_cube_wut.png https://youtu.be/3ZrWVWLymi8 (and yeah video form to also show that there is some clipping to the area in which the bigger cube in shown, that is the cut-out faces are not rendered when not in front of the bigger cube)

11:52 <alyssa> As if face culling is backwards?

11:54 <urjaman> nah i think it is one of the tests working funky (depth? stencil? alpha? my brain melted reading that code and i need a break) :P

11:55 <urjaman> wait a sec i'll try to make sense in a moment :P

11:55 <alyssa> Quite possible

11:56 <urjaman> like the algorithm works by first doing draws (with weird colors that identify the object being drawn) to fill in the depth/stencil buffers, and then using that in the actual draw to well.. draw the correct cubes in the correct areas

11:57 <alyssa> fragment_ops.* fails indeed.

12:02 <urjaman> (fun note: the opencsg docs say they have ran this on a Riva TNT2 ... old stuff)

12:19 <urjaman> like, during the actual draw it uses glDepthFunc(func = GL_EQUAL) which sounds fun

14:19 <TheCycoONE> I had a TNT2

22:25 <urjaman> alyssa: is my grepping correct, and panfrost never even looks at the depth write mask?

22:28 <urjaman> ah sorry, it is looked at in an another file (should've actually grepped :P)

22:34 <alyssa> urjaman: Hm

22:38 <urjaman> so yeah it is used, but i'm still not 100% if it's used enough :P

22:39 <urjaman> the one thing common with the dEQP failure images to me is that the depth channel (that it says it rendered out as blue, so the amount of blue) comes out wrong

22:41 <urjaman> usually less blue (darker) than the reference ...

23:05 <alyssa> urjaman: Used too much? ;P

23:09 <urjaman> ... i doubt that ;P

23:11 <urjaman> btw, '/* What?! */' (line 139 of panfrost-job.h) got an actual lol out of me

23:45 <alyssa> urjaman: Haven't seen that bit in ages!