<icecream95>
It's mainly boring stuff, except that I have to divide the stride by two in panfrost_setup_slices
<icecream95>
I'm still not sure what the difference between the 0x01 and 0x13 formats is...
leinax has quit [Ping timeout: 260 seconds]
kaspter has quit [Ping timeout: 268 seconds]
kaspter has joined #panfrost
<tomeu>
icecream95: nice!
icecream95 has quit [Ping timeout: 258 seconds]
troii has joined #panfrost
troii has left #panfrost [#panfrost]
kaspter has quit [Quit: kaspter]
kaspter has joined #panfrost
Xalius_Ph has joined #panfrost
<Xalius_Ph>
is there a way to disable glamor under xwayland?
<Xalius_Ph>
still trying to narrow down panfrost kernel crashes
megi has joined #panfrost
raster has joined #panfrost
NeuroScr has joined #panfrost
<tomeu>
narmstrong: what are my options for running mainline and debian on a nexbox a1?
alpernebbi has joined #panfrost
<narmstrong>
tomeu: in terms of ?
<tomeu>
narmstrong: I got it with android, so I wonder what's the easiest way of getting those instead
<tomeu>
googling returns lots of outdated information
<tomeu>
guess it has fastboot
<alyssa>
icecream95: neato!
<alyssa>
Xalius_Ph: I don't believe so, but don't quote me on that
<Xalius_Ph>
yeah it only seems to be a build-time option, I'm going to try that
<Xalius_Ph>
anyone coming to FOSDEM?
gtucker has joined #panfrost
<tomeu>
narmstrong: so far I tried booting from Armbian_19.11.3_Aml-g12_stretch_legacy_5.3.0_20191126.img , but after fixing the dtb_name property in boot.ini I still don't get any signs of live on either HDMI or DHCP
buzzmarshall has joined #panfrost
yann has joined #panfrost
robertfoss has quit [Ping timeout: 268 seconds]
<tomeu>
nm, had to do the same but in uEnv.ini
robertfoss has joined #panfrost
<narmstrong>
tomeu: yep boot.ini of for odroid-c2 only
<Xalius_Ph>
mmh reading up on some documentation, EGL can request either GL or GLES context, now I'm wondering what glamor uses and which GL version it requires
Xalius_Ph has quit [Remote host closed the connection]
leinax has joined #panfrost
NeuroScr has quit [Quit: NeuroScr]
<tomeu>
leinax: is memory consumption fine after those fixes?
<leinax>
tomeu: It looks like there might be a very slow leak while running webgl aquarium in cog. After running it overnight I only saw an increase in about 50MB of usage whereas before it would have run out by then. Annoyingly valgrind doesn't want to cooperate with me and kills the rendering subprocess due to an unhandled instruction, so I'm trying to
<leinax>
figure out a good way to analyze it.
<tomeu>
leinax: ok, there's also libasan and libmtrace
<leinax>
Cool. I'll give those a try.
alpernebbi has quit [Quit: alpernebbi]
<tomeu>
narmstrong: happen to know what's up with the internal emmc in the nexbox a1? here with 5.3 looks like it isn't partitioned, but android boots from it out of the box
<robmur01>
tomeu: is it like Rockchip Android where they just pass a 'partition map' on the kernel command-line?
<narmstrong>
tomeu: ah ah, there is no real partitionning, it's partitionned like a NAND raw devices with table in vendor DT and they made a pseudo MTD device that maps on the eMMC to have a single image for NAND or MMC devices
<tomeu>
don't know, but I'm going to just go ahead and partition it
<tomeu>
oh wow
<narmstrong>
tomeu: don't or you'll loose the bootloader
<narmstrong>
tomeu: or at least dump it on an SDCard before
alpernebbi has joined #panfrost
<narmstrong>
amlogic BSP is full of heavy crappy crap
<tomeu>
hmm, guess I could partition leaving the bootloader which I presume is at the start of the device?
<narmstrong>
it's in the first sector, but don't overwrite the first 512bytes... well it's not so simple
<narmstrong>
they have some data before the firsy 256 bytes and then the rest after the first sector
<tomeu>
guess I should be safe if I don't touch the first 1M or so?
<narmstrong>
maybe, be dump the emmc in case you break the bootloader
<tomeu>
how could I restore the contents if that happens?
<narmstrong>
you simply dump it on an sdcard and it will boot
<narmstrong>
using the Khadas VIM2 would have been simpler, they have a tool to generate images using mainline linux and mainline u-boot :-p
<tomeu>
oh, I'm already booting from a sd card, I just want to have the rootfs in emmc for speed
<tomeu>
yeah well, lvrp16 just sent me two of these boxes :)
<tomeu>
thought by overwriting the emmc I would lose the ability to boot from sd card
<tomeu>
but if that's not the case...
kaspter has quit [Quit: kaspter]
pH5 has quit [Quit: bye]
yann has quit [Ping timeout: 268 seconds]
guillaume_g has quit [Quit: Konversation terminated!]
guillaume_g has joined #panfrost
pH5 has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
guillaume_g has quit [Quit: Konversation terminated!]
karolherbst has joined #panfrost
cowsay_ has joined #panfrost
cowsay has quit [Ping timeout: 268 seconds]
cowsay has joined #panfrost
cowsay_ has quit [Ping timeout: 260 seconds]
robertfoss has quit [Ping timeout: 240 seconds]
robertfoss has joined #panfrost
icecream95 has joined #panfrost
buzzmarshall has quit [Remote host closed the connection]
alpernebbi has quit [Quit: alpernebbi]
icecream95 has quit [Ping timeout: 240 seconds]
robert_ancell has joined #panfrost
icecream95 has joined #panfrost
<alyssa>
leinax: "I only saw an increase in about 50MB of usage whereas before it would have run out by then"
<alyssa>
hey, that's still a really great improvement!
<narmstrong>
tomeu: nop if you erase the eMMC the bootROM will loop over the over boot modes (emmc, nand, sd, SPI, usb program)
icecream95 has quit [Read error: Connection reset by peer]
icecream95 has joined #panfrost
<icecream95>
Any ideas why the stride needs to be divided by two for ETC? I would have expected it to be /4.
<alyssa>
icecream95: *shrug*
<alyssa>
I've noticed strides have a lot of weird shifts
<alyssa>
As long as you're dividing by a power-of-two, it's not you, it's the hardware ;)
<alyssa>
Okay I need to fix non-32-bit handling of load stores ummm
<icecream95>
alyssa: Did you see my comment about LOD bias?
<alyssa>
icecream95: I did, it seems reasonable :)
<alyssa>
LOD bias support right now is... shall we say, an accident
<alyssa>
It's not exposed in GLES or anything, I just found the bits and hooked it up.
BenG83 has joined #panfrost
<alyssa>
I admit I'm not sure how a negative FIXED16 would look (I guess the uint16 fields should be regular int16 and then its two's complement and you do something funny with the decimal portion?)
<alyssa>
But if it's as simple as changing the clamp, that's great!
<BenG83>
I rebuilt xwayland with glamor disabled, and all the crashes are gone now...
<icecream95>
BenG83: Is there a way to check wheter glamor is enabled?
<BenG83>
not sure, there is no log like on the xfree86 backend I think
<BenG83>
at least I didn't find any logs for xwayland
<BenG83>
or a way to configure xwayland
<BenG83>
at least I can run kicad now and gimp ;)
<icecream95>
Sway prints the error `glamor: No eglstream capable devices found`, so maybe I do have it enabled?
<BenG83>
eglstream is some Nvidia specific thing for glamor, so I guess yet
<BenG83>
*yes
<urjaman>
for a negative FIXED16, i assume you just multiply the floating point value by the shift (as for an unsigned value) and store the result into int16 ...
<urjaman>
*by 2^shift ...
yann has joined #panfrost
<icecream95>
It works even if you keep the result as uint16. UB, maybe, but the compiler doesn't know that...
<alyssa>
Looking at STK -- we're choking hard on Z24S8 FBOs being broken, there are dEQP tests for that
<alyssa>
after that's fixed we should be on our way basically
davidlt has quit [Ping timeout: 265 seconds]
<alyssa>
Most of the effects can be enabled and visually appear to be fine?
<alyssa>
Shadows are a notable exception -- probably also choking on Z248S8, of course.
<alyssa>
(and indirect UBO access, I guess, but I have a patch somewhere for that)
<alyssa>
So getting this R11G11B10 stuff cleaned up and debugging Z24S8 are next for the STK checklist
<BenG83>
I don't know how to debug this glamor issue further, any ideas?
<leinax>
alyssa: Yeah for sure. What I'm seeing right now might not even be panfrost leaking for all I know. Web engines tend to like eating RAM over time.
<alyssa>
leinax: That's true, yes :)
<alyssa>
Although I do know panfrost has some leakiness.
<icecream95>
The patch doesn't work with tiling yet so you'll need to disable it for it to display correctly
stikonas has joined #panfrost
<icecream95>
alyssa: What do I need to do to fix tiling?
leinax has quit [Remote host closed the connection]
leinax has joined #panfrost
NeuroScr has joined #panfrost
Ntemis has joined #panfrost
<Ntemis>
hello
<Ntemis>
am about to enable panfrost on mesa 19.3.1 and 5.4 for xu4(odroid) and i was wondering if mali t628 is supported
<Ntemis>
ping @alyssa
<endrift>
Does panfrost have a feature matrix like nouveau and radeon on the freedesktop site?
<anarsoul>
nope
<endrift>
Didn't think so. Hopefully eventually!
<endrift>
I find those very helpful, though they often seem to be out of date
raster has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
<Ntemis>
empty handed i come out
<Ntemis>
oh well i shall try my luck
<anarsoul>
Ntemis: it may or may not work but try to whitelist it and use the same quirks as for T7xx?
<Ntemis>
isnt whitelisted already?
<anarsoul>
no
<anarsoul>
see src/gallium/drivers/panfrost/pan_screen.c
<anarsoul>
panfrost_create_screen()
<anarsoul>
you need to add your gpu_id to the switch/case there
<anarsoul>
and then check wherever gpu_id is referenced and apply the same quirks as for T720
<Ntemis>
ow ok thanks
<Ntemis>
since i am building the image over and over i should patch mesa for this
<anarsoul>
I doubt T6xx can do hierarchical tiling
<anarsoul>
so you'll likely have to disable that
<Ntemis>
not good for my rockchip socs :(
<Ntemis>
i can move the mesa package to the project dir so it wont be applied system wide
<Ntemis>
so is this gpu id "case 0x620" correct?
<anarsoul>
no idea
<anarsoul>
you'll have to look it up
<Ntemis>
thanks now i know where to look for supported gpu's
<Ntemis>
so tinkerboard S912 and rk3399 are all supported
<Ntemis>
nice
<Ntemis>
dont know why t620 isnt
<anarsoul>
because it's different generation of midgard
<anarsoul>
and devs didn't get to it yet
<anarsoul>
also likely boards with t6xx are not as popular as those with t7xx and t8xx
<Ntemis>
they are but getting old
<Ntemis>
cant find the t720 quirks any helping hand?
<anarsoul>
grep is your friend
<Ntemis>
let me clone the repo then :p
<alyssa>
icecream95: `can_tile` in pan_resource.c should check that it's !compressed
<alyssa>
Ntemis: T628 is not officially supported. Some people have reported it sort of works, others have reported it doesn't at all. We don't have any boards earlier than <T720 in our test farm so regressions are possible or even maybe likely.
<alyssa>
But in short if you're very adventurous:
<alyssa>
whitelist in pan_screen
<Ntemis>
i did
<alyssa>
panfrost/midgard/midgard_quirks.h
<alyssa>
panfrost/include/panfrost-quirks.h
<alyssa>
Quirks should already be there as far as I know
<Ntemis>
lovely then
<alyssa>
But that's the other issue with T6xx... they're early hardware with a *lot* of bugs.
<alyssa>
So a lot of workarounds in the driver are needed for stuff, and those aren't implemented since we don't support them yet.
<Ntemis>
do you have an xu4?
<alyssa>
No, I'm sorry.
<Ntemis>
can i ask odroid for one for your team?
<Ntemis>
or more if needed ofc
<anarsoul>
one is definitely not enough for a lab
<alyssa>
I'm afraid it's a time issue to support T6xx.
<Ntemis>
no hurry as you know
<Ntemis>
you can take your time and when is done is done
<Ntemis>
if ever
<alyssa>
:)
<HdkR>
I have stacks of XU4s here for testing, just needs time :P
<Ntemis>
sure i can wait then
<Ntemis>
no 19.3.x for xu4 and thats ok
<Ntemis>
i will port whats available to lakka and batocera
<Ntemis>
thanks for the support
<Ntemis>
i hope lima kicks in for our other board like allwinner H5 and amlogic S905
<Ntemis>
*board's
<anarsoul>
you'll need 20.0 for lima
<Ntemis>
oh
<anarsoul>
or backport all the fixes to 19.3 if you want it
<icecream95>
alyssa: That would work, but there's no reason we can't tile compressed textures.
<Ntemis>
nah it goes to the back burner too
<icecream95>
I wonder how tiling for ASTC formats with e.g. 5x5 blocks works...
<HdkR>
Everyone loves ASTC 5x5
<Ntemis>
thank you anarsoul you saved me precious time so you know
<anarsoul>
np
<Ntemis>
i can wait until summer :)
<Ntemis>
*it
<anarsoul>
I hope 20.0 will be released sooner
<Ntemis>
btw you guys need to setup some 3d cores to squash the serious bugs e.g ppsspp,n64,dreamcast etc this way you can push the gpus to the limits
<icecream95>
There are already enough open bugs reported...
<anarsoul>
Ntemis: well, I can't talk for panfrost folks, but for lima we're focusing on fixing deqp failures
<Ntemis>
lol i guess :)
<anarsoul>
there's just not enough manpower to address all the app issues. Hopefully most of them will be fixed when we get to corresponding deqp failure
<HdkR>
Piglit, CTS, deqp should capture most of the bugs
<Ntemis>
what am trying to say is, even if you focus onto one core only lets say a stable recent ppsspp version you could immediately catch a lot of issues with the driver and those can be eliminated in due time and ever spot regressions asap with this and 1-2 games