ChanServ changed the topic of #lima to: Development channel for open source lima driver for ARM Mali4** GPUs - Kernel has landed in mainline, userspace driver is part of mesa - Logs at https://people.freedesktop.org/~cbrill/dri-log/index.php?channel=lima and https://freenode.irclog.whitequark.org/lima - Contact ARM for binary driver support!
paulk-leonov has quit [Ping timeout: 252 seconds]
paulk-leonov has joined #lima
paulk-leonov has quit [Ping timeout: 240 seconds]
paulk-leonov has joined #lima
kaspter has quit [Ping timeout: 240 seconds]
kaspter has joined #lima
<anarsoul> enunes: I wonder if it's a good idea to do the same: https://gitlab.freedesktop.org/mesa/mesa/-/commit/c8620005d24ea01dd237dbef452aa9ea401beb54
<anarsoul> that should improve hit rate for sure
kaspter has quit [Ping timeout: 240 seconds]
kaspter has joined #lima
kaspter has quit [Ping timeout: 240 seconds]
kaspter has joined #lima
Barada has joined #lima
chewitt has quit [Quit: Zzz..]
Barada has quit [Quit: Barada]
kaspter has quit [Ping timeout: 240 seconds]
kaspter has joined #lima
dev1990 has quit [Quit: Konversation terminated!]
dev1990 has joined #lima
<enunes> anarsoul: seems like a good idea, with the small ones already being rounded up and the bigger ones are already fairly big anyway
<bshah> anarsoul: I am not sure what fixed it, but I have not seen that issue in very very long time.
<bshah> So can be closed yes.
<bshah> I will do so
monstr has joined #lima
ente has quit [Ping timeout: 240 seconds]
monstr has quit [Ping timeout: 260 seconds]
camus has joined #lima
kaspter has quit [Ping timeout: 240 seconds]
camus is now known as kaspter
monstr has joined #lima
ente has joined #lima
camus has joined #lima
kaspter has quit [Ping timeout: 268 seconds]
camus is now known as kaspter
ente has quit [Ping timeout: 240 seconds]
<enunes> anarsoul: wait, but we already align it in lima_bo_create, so maybe we dont need to do anything
kaspter has quit [Ping timeout: 240 seconds]
kaspter has joined #lima
monstr has quit [Remote host closed the connection]
kaspter has quit [Ping timeout: 240 seconds]
kaspter has joined #lima
<anarsoul> enunes: panfrost now aligns it to power of two
<anarsoul> glmark2 refract generates mipmaps each draw :\
<anarsoul> moving it out of loop yields +5 fps (9fps -> 14fps)
camus has joined #lima
kaspter has quit [Remote host closed the connection]
camus is now known as kaspter
anarsoul|c has quit [Quit: Connection closed for inactivity]
<enunes> anarsoul: check this out, I think I finally figured this one out https://i.imgur.com/yXOjeae.png
<enunes> not an easy one
<anarsoul> enunes: what's that?
<anarsoul> gallium hud?
<enunes> GALLIUM_HUD has been broken since forever
<linkmauve> Oh, great!
<anarsoul> cool
<enunes> this is still a hack but I finally figured out what seems to be the root cause so now I can work on a fix
<anarsoul> enunes: I don't see an MR :)
<anarsoul> so what was the issue?
<enunes> the shaders in gallium hud are imported from tgsi and it seems to assume a different varying setup so we mess up the pp varying setup
<enunes> this hacked version is using a modified tgsi shader to render the text that matches the varying setup that we assume
<anarsoul> so it's a bug in lima after all?
<enunes> I dont know, the tgsi shader is weird to me since when it is translated to nir the vec2 varyings are listed as vec4
<enunes> so maybe I still need to investigate
<enunes> but it took a lot of time to eliminate all the other possible stuff from what the gallium hud code does
<enunes> maybe we are the only driver to hit it since I think the only other vec4 drivers are old intel ones that were not gallium drivers
<anarsoul> oh
<anarsoul> right
<linkmauve> Would it make sense to eventually convert those to NIR, so that no translation would have to happen?
<enunes> I think it would make sense and it might be possible that the are just still tgsi because nobody bothered to convert yet
<enunes> but I just got it to work moments ago after hours staring at that code, now need to bo back and evaluate the options
<enunes> another thing to consider is that bypasses the opengl layer so some things are skipped, in particular it reuses a vertex shader with more varyings than the fragment shader has
<enunes> I think normally that would be resolved by link time optimization (?) but this seems to be skipped at this level
dev1990 has quit [Quit: Konversation terminated!]