00:36 <anarsoul> OK, I'm merging two ppir MRs

05:18 <anarsoul> rellla: re: https://gitlab.freedesktop.org/mesa/mesa/merge_requests/1415 -- do you have any plans to implement invalidate_resource() callback for lima?

07:35 <mardestan> I am very weak in understanding how GPU hw contexts really function. Also how does the command buffers know when to get back to submitting more work to the fifo i.e consuming more cpu guven hw commands?

07:45 <mardestan> i can see some ancient glxcopycontext or such stuff, which is deprecated for ages i think, well glx stuff or egl stuff from kernel, definitely are not taking in the video ram address only, i think they copy stuff to vram latern and submit the commands one by one?

08:21 <mardestan> haven't got a real clue, Only way it can ever work is if glFinish or fences are inserting some marker either to vram directly or ram, but i honestly do not think it's what glFinish does

08:33 <mardestan> well mesa source code proves otherwise indeed.

08:33 <mardestan> it inserts an obligatory fence with glFinish call

09:25 <mardestan> to be fair, if there are hw contexts possible (not sure) it does not seem like possible to have holes in the batchbuffer of commands, nothing to stitch it other than having just multiple contexts

09:27 <mardestan> it is not like the buffer objects allocated are as small as beginning from one byte, they start from 4096 or so, page sized minimums

09:31 <mardestan> switching context would probably work for hw commands, however to do that very often in user program like on every shader call like i need, this is going to be expensive imo

09:42 <mardestan> hangover today, i need a higher level command in GLX/EGL/GL whatever to append to current context without context switch

10:41 <rellla> anarsoul: i had a quick look at how the others do that but did no coding still. it's not my prio #1

10:42 <rellla> i'd like to clean up my undef MR a bit including some nir_opt_undef passes and push a new version before.

12:52 <rellla> hey, i'm doing some undef lowering trials and got that one: https://pastebin.com/raw/CYtwfs79

13:25 <mardestan> maybe wgl/glx/eglMakeCurrent family maybe is not going to be very slow even

17:11 <xdarklight> anarsoul: is that meant to be "boot_to_float" or "bool_to_float" here: https://gitlab.freedesktop.org/mesa/mesa/merge_requests/1838/diffs?commit_id=01db66fd35e5e01e1f07362f1c90284118878e0e#172f3784c3d58603f56ae6e5a28d250d3bd7a935_136_132 ?

17:43 <anarsoul> just a typo

17:43 <anarsoul> I'll fix it

20:40 <anarsoul> rellla: so what's wrong here? totally valid shader with unpredictable result :)

23:15 <rellla> anarsoul: my question was expected to be extended... i fixed it in the meantime, so just ignore it :)

23:15 <anarsoul> :)

23:18 <anarsoul> have you decided to go with replacing it with consts?