davidlt has quit [Remote host closed the connection]
davidlt has joined #panfrost
yann has joined #panfrost
davidlt has quit [Ping timeout: 265 seconds]
NeuroScr has joined #panfrost
karolherbst has quit [Ping timeout: 276 seconds]
karolherbst has joined #panfrost
karolherbst has quit [Client Quit]
karolherbst has joined #panfrost
NeuroScr has quit [Quit: NeuroScr]
NeuroScr has joined #panfrost
megi has joined #panfrost
urjaman has quit [Ping timeout: 265 seconds]
urjaman has joined #panfrost
kaspter has quit [Ping timeout: 250 seconds]
kaspter has joined #panfrost
NeuroScr has quit [Quit: NeuroScr]
raster has joined #panfrost
cwabbott has quit [Remote host closed the connection]
cwabbott has joined #panfrost
davidlt has joined #panfrost
<alyssa>
urjaman: compiler is extremely leaky, so nowhere.
<alyssa>
Patch away...? :)
<alyssa>
Usually it's not a huge deal but probably good to clear the valgrind noise
<urjaman>
yeah i'll send some free()s your way once i've reasoned my way through this (also, apparently after this nap i feel i've not quite woken up from yet...)
<alyssa>
:)
<urjaman>
(those two are simple, but there are 6 allocations and one free() in the whole file :P)
<alyssa>
Oops!
<HdkR>
Allocating memory is easy. Freeing it is as easy as killing the application allocating it ;)
<alyssa>
which the kernel usually takes cares of me ;)
<urjaman>
I've read some 4k+ lines of code in the last day-ish :P
<urjaman>
not enough yet to understand as much as i need to, but i do understand midgard assembler a bit better than i used to
<alyssa>
Though it hasn't been updated but nothing huge has changed
<urjaman>
that one i had read before (but also did re-read a few times over during this process)
<alyssa>
Cool
suihkulokki has quit [*.net *.split]
grw has quit [*.net *.split]
Marex has quit [*.net *.split]
hopetech has quit [*.net *.split]
samueldr has quit [*.net *.split]
suihkulokki has joined #panfrost
Marex has joined #panfrost
samueldr has joined #panfrost
grw has joined #panfrost
hopetech has joined #panfrost
suihkulokki has quit [*.net *.split]
grw has quit [*.net *.split]
Marex has quit [*.net *.split]
hopetech has quit [*.net *.split]
samueldr has quit [*.net *.split]
<urjaman>
btw, my little "old good" trace of xorg+xfce (just start it and quit it via the menu) had panfrost compile some 7067 shaders
<urjaman>
so based on that, that * 2 * nodes * blocks * 16 * size of a struct dynarray (i dunno how big) kinda starts to make even those last_read/last_write leaks kinda matter
hopetech has joined #panfrost
grw has joined #panfrost
Marex has joined #panfrost
samueldr has joined #panfrost
suihkulokki has joined #panfrost
<alyssa>
urjaman: 7067?!!
<urjaman>
atleast that's how many shaders the trace dumped (like yeah there are like a handful of unique ones but)
<urjaman>
is the last line of "grep shader: x_oldstable.log" :P
<urjaman>
in the comparison x_newfail.log it compiled 1211 shaders before the failing one
<alyssa>
urjaman: That doesn't mean it compiled 7067 shaders
<alyssa>
That means it had 7067 draws
<alyssa>
Most of those were probably the same shader ;)
<urjaman>
oh so it dumps it anyway for each draw?
<urjaman>
(makes for a very unnecessarily big traces if nothing else...)
<tomeu>
bbrezillon: so, the faults that I mentioned in #dri-devel happen only when deqp-gles2 gets restarted between batches
<tomeu>
so when a client closes, the BOs of other clients get thrashed
<tomeu>
bbrezillon: do you have any idea of how could that be possible?
<urjaman>
but yeah i think i've figured out that the failing one is a glamor fragment shader, most likely used for gradients? (though i guess you already figured this much)
<urjaman>
has stats like that when compiled by old panfrost
<urjaman>
(atleast based on comparison of my grep|cut|sort|uniq-filtering for the two runs i did...)
<urjaman>
but like, 8 registers doesnt sound like one that'd be super-difficult to compile when you have 16 (even tho 8 is way better than 9 for the threads :P) so i think there must be some sort of a gotcha there...
* urjaman
will get back to life for a bit
<bbrezillon>
tomeu: nope
<bbrezillon>
tomeu: I guess my fixes don't help here
* urjaman
has apparently been effectively confuzzled by there being two different debug outputs with "shader%d" in them
<alyssa>
urjaman: glamor indeed, yes.
<alyssa>
gradients.
<tomeu>
robher: robmur01: what about you: any ideas about why a client closing its drm fd could cause other clients' BOs to be released or overwritten?
<bbrezillon>
tomeu: are any of those BO exposed through FLINK?
<bbrezillon>
GEM_FLINK
<robmur01>
tomeu: sorry, not found the time to dig in yet, so I have nothing more than a vague hunch or two
<tomeu>
bbrezillon: I don't think any of those are shared, this is with surfaceless
<tomeu>
robmur01: np, I haven't started debugging myself yet
<tomeu>
will do tomorrow
karolherbst has quit [Ping timeout: 265 seconds]
<robher>
tomeu: are the clients sharing an fd? You have to do a dup() on the fd if so. Or to transfer who controls keeping the fd open.
<tomeu>
robher: they should be any sharing, it's a runner process that forks and execves into deqp-gles2, which then does all the GL stuff from zero
<tomeu>
*shouldn't
yann has quit [Ping timeout: 268 seconds]
karolherbst has joined #panfrost
<HdkR>
It's SFBD that splits up MRT fragment jobs in to multiple shader stages right?
yann has joined #panfrost
pH5 has quit [Quit: bye]
<alyssa>
HdkR: SFBD is the framebuffer descriptor for systems without MRT
<alyssa>
MFBD is for the descriptor for systems with
<alyssa>
The driver lowers for systems without
* alyssa
doesn't recall how the lowering works
<alyssa>
Often we get lazy and just talk about "SFBD systems" or "MFBD systems"
<HdkR>
alyssa: Right. But ES mandates MRT, so how does that work with SSBOs and Images? :D
<HdkR>
Compute fallback?
<alyssa>
I don't remember
<HdkR>
me either
<alyssa>
Don't think I ever tried since MRT is only for ES3+
<alyssa>
and we don't expose ES3 right now (and T860 has been our dev target for ES3)
<alyssa>
T760+ supports MRT
<HdkR>
Right. Was mostly curious if you had seen it at some point when messing around :D
guillaume_g has quit [Quit: Konversation terminated!]
stikonas has joined #panfrost
pak0st has joined #panfrost
pak0st has quit [Remote host closed the connection]
pak0st has joined #panfrost
Pak0st58 has joined #panfrost
pak0st has quit [Ping timeout: 260 seconds]
Pak0st58 has left #panfrost [#panfrost]
pak0st has joined #panfrost
<pak0st>
Hello, amazing people! I've been enjoying the Pinebook Pro for the past few weeks and Panfrost has played a big role to make the device even more fun! I wonder if someone would have an idea where to look for the cause of this odd mirroring on the Godot Editor with mesa-git? https://drive.google.com/open?id=1AvYMyqv_TgC1Pk4sZlpupzRqHQRzlmja