18:54 <marex> austriancoder: https://gitlab.freedesktop.org/marex/mesa/-/commit/c007dc545d8377f789b3cecb9ab49db3971ffa07 does this look like a reasonably acceptable way to set W=1 if we replace pointcoord X,Y only ?

18:54 <marex> austriancoder: something similar will be needed for NIR

18:55 <marex> austriancoder: the idea is that if points are used, hardware only replaces X,Y inputs to PS with the point coord, but Z,W=0,0 ; and we need W=1 in PS

22:40 <marex> flto: is there any documentation for the etnaviv nir implementation at all ?

22:41 <marex> I am trying to implement the above ^ for nir, but the rewriting makes zero sense

22:41 <austriancoder> marex: will have a look tomorrow.. bed time here

22:43 <austriancoder> marex: maybe a simple nir pass is all you need to write. Look at nir_lower_xxx.c examples

22:43 <marex> austriancoder: sounds simple ... I spent last four or so hours looking for a simple example of just that, or documentation, or anything that would be helpful

22:44 <austriancoder> marex: the code is the documentation.. ;)

22:47 <marex> austriancoder: and a pretty bad one at that

22:48 <austriancoder> marex: I am not sure how I can help you. If can provide you with the final lowering but I think that's not what you want.

22:49 <marex> austriancoder: I am still even trying to validate this approach to setting W=1

22:49 <austriancoder> marex: the nir docs are not great, but that's a Mesa problem

22:50 <austriancoder> marex: you did piglit and deqp runs to validate that there are no regressions and that it even could fix some fails?

22:51 <austriancoder> marex: from a quick look your approach looks okay.. but yeah.. bed time .. really now

22:51 <marex> austriancoder: good night

23:11 <flto> marex: isn't gl_PointCoord supposed to be a vec2? so maybe this a problem with the gles1 (this is about neverball gles1 right?) lowering in mesa and not a etnaviv problem?

23:12 <marex> flto: it is vec4 and imirkin says that's correct, since the vec4 is st01