jrmuizel has quit [Remote host closed the connection]
<MoeIcenowy>
anarsoul: strange... the PP MMU fault happened on Android has no corresponding user-space error on Mesa
<anarsoul>
probably it didn't catch it?
<anarsoul>
sorry, no idea here
<MoeIcenowy>
anarsoul: could the PP MMU fault come from mipmap issue?
<anarsoul>
maybe
<anarsoul>
I'll look into mipmap issue tonight
<MoeIcenowy>
is there any instant hack, e.g., lower LIMA_MAX_MIP_LEVELS?
<anarsoul>
yeah, you can try it
<anarsoul>
well, not it
<MoeIcenowy>
?
<anarsoul>
see lima_texture.c
<anarsoul>
lima_update_tex_desc()
<anarsoul>
replace LIMA_MAX_MIP_LEVELS with lower value here
<anarsoul>
don't touch global define
yuq825 has joined #lima
<anarsoul>
hi yuq825
jrmuizel has joined #lima
<yuq825>
anarsoul: hi
<anarsoul>
yuq825: looks like mipmapping is broken, it assumes that max mip levels is 13 but it doesn't fit into texture descriptor which is 64 bytes
<anarsoul>
I have some work in progress to address it and refactor texture code while I'm at it
<anarsoul>
will publish MR tonight
<yuq825>
so it's not a regression, right? I don't remember we touch mipmap recently
<anarsoul>
it's not
<yuq825>
so mipmap level should be less than 13 to fit into texture desc
<anarsoul>
yeah
<yuq825>
do mesa have seperate PIPE_CAP for it? I remember mesa use the mipmap level CAP also for max texture GPU can support
<anarsoul>
it should be 11
<anarsoul>
yuq825: we can fake it for now
<anarsoul>
i.e. advertise 13 but support 11
<MoeIcenowy>
anarsoul: kill the over level mipmap doesn't help on pp fault
<anarsoul>
I'm also reworking texture descriptor to use a struct with bitfields, because currently it's a mess
<anarsoul>
MoeIcenowy: well, then you'll have to debug it
<MoeIcenowy>
and it seems that if I unlock, lock screen cannot be properly rendered any longer
<anarsoul>
dump shader and command stream?
<anarsoul>
s/shader/shaders
<yuq825>
Icenowy remind me if we fake two level, some fragment shader sample texture explicitly with level may fail with last two levels, but it should be rare to see such kind of shader
jrmuizel has quit [Remote host closed the connection]
megi has quit [Ping timeout: 248 seconds]
smaeul has joined #lima
Da_Coynul has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
dddddd has quit [Remote host closed the connection]
<anarsoul>
yuq825: yeah, that's true
jrmuizel has joined #lima
jrmuizel has quit [Remote host closed the connection]
<anarsoul>
MoeIcenowy: your pp mmu fault could also be related to broken reg spilling
<bshah>
anarsoul: so, since my apitrace was kinda useless, I did dump the log output using LIMA_DEBUG=gp,pp,dump and have uploaded it to : https://bshah.in/lima-gp-pp-dbg-vkb.log .. maybe this helps with pinpointing issue?
<anarsoul>
bshah: not really, it's 4k line log without any clue where to look at
<bshah>
:-(
<bshah>
Would it help if I upload dumped shaders somewhere?
<anarsoul>
bshah: what would be nice is to get a simple reproducer
<bshah>
hm :/ I can't even get it simple in terms of qt itself ...
<bshah>
okay, I'll keep trying
<MoeIcenowy>
anarsoul: how to do always writeback to workaround for buffer_age?
<anarsoul>
MoeIcenowy: are you using old mesa?
<anarsoul>
it's been fixed quite a while ago
<MoeIcenowy>
anarsoul: not so old, maybe 3 days ago
<MoeIcenowy>
so maybe I met another bug?
<anarsoul>
maybe
Barada has joined #lima
piggz_ has quit [Ping timeout: 248 seconds]
<MoeIcenowy>
anarsoul: I think I may met the vertex count bug of Lima
<anarsoul>
I see
<anarsoul>
however I wouldn't expect it in android shell
<MoeIcenowy>
at least some part of the interface is not rendered at Settings
<anarsoul>
I doubt it's vertices count bug
<anarsoul>
there can't be 64k vertices in regular ui
dri-logger has quit [Ping timeout: 245 seconds]
dri-logger has joined #lima
<MoeIcenowy>
and when I press an item to trigger the visual effect of it
<MoeIcenowy>
more items disappeared
Barada has quit [Quit: Barada]
<MoeIcenowy>
anarsoul: yes, I logged the count, and the maximum is 3072
<anarsoul>
MoeIcenowy: IIRC mali4x0 can do 65k
<MoeIcenowy>
anarsoul: can it do 65k in current code?
monstr has joined #lima
megi has joined #lima
anarsoul|c has joined #lima
Barada has joined #lima
<cwabbott>
anarsoul: I remember hearing about this a long time ago during the original lima project, iirc it actually supports all levels but the HW uses different offsets into the last pointer for the last few levels so they have to be packed into the same buffer
<cwabbott>
this was to reduce the size of the texture descriptor -- the last few levels are small so the overhead of copying them won't be too great
<libv>
so that the descriptor fits into the cache more easily
<libv>
the last few mip levels were implicit
anarsoul|c has quit [Quit: Connection closed for inactivity]
dddddd has joined #lima
<MoeIcenowy>
anarsoul: seems that the current code cannot do 65k
<MoeIcenowy>
the glmark2 high-poly bump model, which cannot be render fully, has 24002 vertices
<enunes>
MoeIcenowy: there are many problems with the driver, seems unlikely that a UI would have as many vertices to hit that particular problem
<MoeIcenowy>
but I cannot explain why some of UI items are rendered
<MoeIcenowy>
and when UI items become more, the number of failed UI items also increases
<enunes>
yeah it's hard to debug
<enunes>
are shaders being compiled correctly?
<enunes>
we also have issues with tests failing randomly which is reproduceable in piglit, and we are fixing other small issues one at a time
<MoeIcenowy>
because they think panfrost will break chromium
<MoeIcenowy>
lima is different -- no blacklist is needed because it breaks everything
<bshah>
(re: backlog, not sure if it matters, but with Linux/Plasma, after opening more then 2 apps, the whole thing stops rendering/renders black windows..
<wens>
is there a blacklist?
<wens>
for panfrost I mean
<wens>
ah, see it
Da_Coynul has joined #lima
yuq825 has quit [Remote host closed the connection]
Da_Coynul has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
Da_Coynul has joined #lima
Da_Coynul has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
Da_Coynul has joined #lima
Da_Coynul has quit [Quit: My MacBook Air has gone to sleep. ZZZzzz…]
<MoeIcenowy>
oh say fall_equal3 unsupported on Android Lima log
<MoeIcenowy>
I think I see fall_equal is solved on master
<MoeIcenowy>
should I git pull now?
<rellla>
master is always the best choice :)
jrmuizel has joined #lima
adjtm has quit [Ping timeout: 245 seconds]
<MoeIcenowy>
enunes: at least your fix commit didn't fix the Android misrender problem
jrmuizel has quit [Remote host closed the connection]
yuq825 has joined #lima
monstr has quit [Remote host closed the connection]
jrmuizel has joined #lima
afaerber has joined #lima
Barada has quit [Quit: Barada]
yuq825 has quit [Remote host closed the connection]
<anarsoul>
cwabbott: libv: I see thianks, I guess I'll implement it