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!
ecloud has joined #lima
ente has quit [Ping timeout: 240 seconds]
ecloud_wfh has quit [Ping timeout: 240 seconds]
ente has joined #lima
_whitelogger has joined #lima
mripard has quit [Ping timeout: 255 seconds]
mripard has joined #lima
minicom8 has joined #lima
chewitt has quit [*.net *.split]
tlwoerner has quit [*.net *.split]
enunes has quit [*.net *.split]
minicom has quit [*.net *.split]
tlwoerner has joined #lima
dddddd has quit [Ping timeout: 256 seconds]
<anarsoul> MoeIcenowy: I reworked your zsbuf reload implementation, see https://gitlab.freedesktop.org/anarsoul/mesa/-/commit/de39094a7940a20a61c13fdc1c47733c42920a44
<anarsoul> testing it now
_whitelogger has joined #lima
<anarsoul> z16 still fails
<anarsoul> hm
<MoeIcenowy> anarsoul: mask Z16?
<MoeIcenowy> Z16 may need another shader
<anarsoul> I doubt it
<anarsoul> I think I understand why we need separate format for Z24S8
<anarsoul> basically we can't sample stencil in GLES2
<anarsoul> so old format is actually Z24X8
<anarsoul> and new is Z24S8
<MoeIcenowy> BTW
<MoeIcenowy> I think when Z16, it samples Z to the X part of the vector
<MoeIcenowy> but Z24S8 samples Z to Z part
<anarsoul> actually it samples to xyz
<anarsoul> both should do that
<anarsoul> since Z16 is actually L16
<anarsoul> so it's (L,L,L,1.0)
<anarsoul> for Z it should be (Z,Z,Z,1.0)
<MoeIcenowy> anarsoul: is it a repeat
<anarsoul> yes
<MoeIcenowy> or is more variables used for higher accuracy?
<anarsoul> it's broadcast to multiple channels
<MoeIcenowy> anarsoul: on 0x32 format it's not broadcast
<MoeIcenowy> xyz has different values
<anarsoul> hm
<anarsoul> we also have a lot of failures with depth16 in deqp-lima-fails
chewitt has joined #lima
buzzmarshall has quit [Remote host closed the connection]
<anarsoul> rellla: why do we dump only one texture descriptor?
<anarsoul> we can have up to 16
<rellla> anarsoul: because it's not implemented?
<anarsoul> having up to 16 textures is implemented :)
<anarsoul> MoeIcenowy: tests seem to fail because we're doing unnecessary zs reload at the very beginning
<anarsoul> I'm debugging why we're doing it...
<anarsoul> (I already fixed few bugs that I introduced)
<anarsoul> yeah, works now :)
<rellla> anarsoul: iirc i already knew, that i missed sth while introducing texture dumps :p i can do that the next days...
_whitelogger has joined #lima
<anarsoul> rellla: that's not something urgent :)
<anarsoul> MoeIcenowy: so recreate_colorbuffer tests are still failing :(
<anarsoul> but looks like regressions are fixed
<anarsoul> btw these tests also fail on blob
<anarsoul> hm, hold on
<anarsoul> they look like lima fails
<rellla> yes they are. also the skips.
<anarsoul> where I can find list of mali fails?
<rellla> they are mali fails iirc
<anarsoul> oh
<anarsoul> weird
<anarsoul> why do they match lima fails? :)
<rellla> not sure if i'm right atm :)
<anarsoul> yeah, that looks better
<anarsoul> thanks
_whitelogger has joined #lima
kaspter has quit [Remote host closed the connection]
kaspter has joined #lima
<MoeIcenowy> anarsoul: maybe we hit the zone of HW bug
minicom8 is now known as minicom
rak-zero has joined #lima
chewitt has quit [Ping timeout: 265 seconds]
<MoeIcenowy> anarsoul: do you know how can let blob use Z16?
buzzmarshall has joined #lima
chewitt has joined #lima
marcodiego has joined #lima
dddddd has joined #lima
<anarsoul> MoeIcenowy: no idea
<anarsoul> but the point is that we don't even attempt to reload zsbuf in dEQP-GLES2.functional.fbo.render.recreate_colorbuffer.no_rebind_rbo_rgb565_depth_component16
<anarsoul> oh, we do, my traces are wrong
<anarsoul> wtf
<anarsoul> key.zsbuf->texture->format is argb
<MoeIcenowy> anarsoul: what the hell
<anarsoul> ah, sorry, I'm wrong, it's Z16
<anarsoul> but why is it Z16
<anarsoul> :\
<anarsoul> I commented it out
<anarsoul> ah
<anarsoul> one more place to comment it out
<anarsoul> well, I promoted it to Z24
<anarsoul> but it still doesn't work
rak-zero has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
<anarsoul> MoeIcenowy: so yeah, maybe we get Z24 split into r,g,b components with 0x32 format
<anarsoul> btw blob uses z24s8 for z16
<anarsoul> I guess we should do that too
<anarsoul> OK, depth reload now works fine with z24s8
<anarsoul> stencil is broken atm
<anarsoul> shared_depthbuffer failures are not related to missing depth/stencil reload since they don't need it
<anarsoul> looking into stencil
<anarsoul> looks like we need some extra magic in rsw :)