01:25 <marex> https://gitlab.freedesktop.org/marex/mesa/-/commit/2a47cc2d8faf57bc33af0c342b8fa25941054f85 so this seems to work with NIR

01:25 <marex> austriancoder: ^

01:26 <marex> but apparently this needs to be enabled only in case rasterizer did set sprite_coord_enable

03:49 <marex> one more update https://gitlab.freedesktop.org/marex/mesa/-/commit/ad0ebd8f3299cab9d8612a19d8a6fc92865e3f88 here

03:49 <marex> that should fix the texcoord for any TEXx

03:54 <marex> but I have no idea where to put similar stuff in TGSI

03:55 <marex> also maybe the nir_imm_float is not necessary, maybe it is possible to somehow get 1.0f without constants

05:40 <austriancoder> marex: nice.. nir_imm_float(..) is the way to go.. the backend compiler will handle uniform allocations better in the near future. I am working on something...

07:08 <austriancoder> and nir is quite nice with its opts passes to clean up stuff

11:57 <marex> austriancoder: still, some xnor opcode would be probably more efficient

11:57 <marex> austriancoder: also, any idea how to implement the same thing for TGSI ?

13:06 <marex> I guess with tgsi, we need to add instruction in case src is texcoord

16:10 <marex> austriancoder: https://gitlab.freedesktop.org/marex/mesa/-/commit/3e889fa38673a0986867e532ef8781b65821e5aa I think this might be it

16:14 <marex> mntmn: can you test ^ that with neverball and stuff ?

16:17 <mntmn> marex: sure

16:17 <mntmn> marex: quite the patch!!

16:34 <marex> my head explodes

16:35 <marex> and it wouldnt be possible without the help of imirkin, jekstrand and austriancoder

16:36 <marex> https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8618 MR is here

16:36 <marex> austriancoder: ^

16:53 <mntmn> wondering if this has vivante https://github.com/T-head-Semi/aosp-riscv

16:56 <marex> mntmn: notice the rendering corruption in the Clock application

16:57 <marex> lynxeye: did you ever try DSI on mx8mn ? it seems to work on mm, but on mn not so much

16:57 <marex> the pipeline should be almost identical, except for a few clock/blkctl bits

17:25 <mntmn> marex: ha, well spotted @ clock

17:26 <marex> mntmn: did you work with 8mn before ?

17:32 <lynxeye> marex: Blocked on DT binding, but to be honest I didn't really have time to follow up. Maybe I should ping a bit more.

17:32 <marex> lynxeye: would be nice to unblock it, really

17:33 <marex> lynxeye: what's the problem with the bindings, the composite GPU domain ?

17:33 <lynxeye> Nope, simply the fact that I can't name the resets...

17:33 <lynxeye> Same as clocks really, but they are already in the bindings. resets are new :/

17:33 <marex> lynxeye: I should probably read the discussion

17:34 <marex> why cant you name the resets ?

17:49 <lynxeye> marex: because the power domains have a generic binding and resets (as well as clocks) belong to the devices _in_ the domain

21:22 <marex> austriancoder: I updated the patch to use SET in TGSI and updated the commit message , se https://gitlab.freedesktop.org/marex/mesa/-/commit/3ad03b2ab102752b4d9ae6576fc85cf7bd27c8e4

21:23 <marex> mntmn: ^

21:23 <marex> so the next problem is the logo alpha :)

21:24 <mntmn> marex: mmm alpha :3 your patch works with NIR yeah?

21:26 <marex> mntmn: yeah

21:26 <marex> mntmn: at least on mx6q

21:40 <mntmn> marex: ok i will try it tomorrow, need to drive home now