<robmur01>
shame we neglect to print the raw status value...
kaspter has quit [Quit: kaspter]
<robmur01>
nroberts: FWIW I'd suggest ditching any downstream vendor stuff, stick a mainline-based distro image (like Manjaro or Armbian) onto a dedicated SD card and just replace the kernel (and entire filesystem if you wish) directly
<robmur01>
the more fulfilling version involves building mainline TF-A/U-Boot, bootstrapping the partitions from there, then installing everything manually, but that's not quite so user-friendly :)
<nroberts>
robmur01, thanks, maybe I will start again with armbian then at some point. I have probably spent too much time on this already though so I might just put it off too :)
kaspter has joined #panfrost
<robmur01>
alyssa: did the offending job involve something funky? panfrost_exception_name() is not entirely exhaustive
<alyssa>
robmur01: I don't.. see anything funky
guillaume_g has quit [Quit: Konversation terminated!]
l-as has quit [Ping timeout: 258 seconds]
<robmur01>
if it's reproducible, first thing I'd do is see if the actual exception_code is enlightening (so hopefully not 0x7f...)
guillaume_g has joined #panfrost
l-as has joined #panfrost
<alyssa>
robmur01: 0x5b
<alyssa>
with no reported fault address
<alyssa>
this is horribly imprecise information
<robmur01>
indeed it is ;)
nlhowell has joined #panfrost
clementp[m] has quit [Ping timeout: 240 seconds]
<alyssa>
robmur01: Ok, this is interesting
<alyssa>
dealing with a buffer texture (4 bytes per element, 16384 width)
<alyssa>
and Gallium says it has an offset of 0x10
<alyssa>
so accessing the pointer 0xblahblah0010 is raising the fault
<alyssa>
if I force offset down to 0 so it's cache line aligned, it's fine
<alyssa>
this is despite rgba8 unorm supposedly not needing to be aligned on bifrost
<alyssa>
wonder how the blob handles buffer textures if not like this
<alyssa>
ok, gles3.1 fails list is coming down nicely
<daniels>
nroberts: if you look at Mesa CI, you can see the mainline-based kernel (with a couple of Panfrost patches I think?) and Debian rootfs we build for the Kevin Chromebook, which is also RK3399, and should work on the RockPi4 if you just change the DTB name
<alyssa>
I did a 1/10 run with two major sections of tests known to fail blocked, and that's down to 11 failures of 1745 that ran
jernej has quit [Remote host closed the connection]
jernej has joined #panfrost
robmur01_ has joined #panfrost
Net147_ has joined #panfrost
Depau has joined #panfrost
Moe_Icenowy has joined #panfrost
Net147 has quit [*.net *.split]
raster has quit [*.net *.split]
paulk-leonov has quit [*.net *.split]
robmur01 has quit [*.net *.split]
Depau_ has quit [*.net *.split]
urjaman has quit [*.net *.split]
rak-zero has quit [*.net *.split]
HdkR has quit [*.net *.split]
MoeIcenowy has quit [*.net *.split]
Moe_Icenowy is now known as MoeIcenowy
robmur01_ is now known as robmur01
robmur01 is now known as 17SABVDEO
urjaman has joined #panfrost
Depau_ has joined #panfrost
raster has joined #panfrost
robmur01 has joined #panfrost
raster has quit [Excess Flood]
raster has joined #panfrost
robmur01 has quit [Ping timeout: 254 seconds]
Depau_ has quit [Ping timeout: 254 seconds]
17SABVDEO is now known as robmur01
<nroberts>
daniels, thanks, maybe I give that kernel commit a go over the weekend
<daniels>
nroberts: kernelci does boot rk3399 on every commit from mainline + next + a million other trees, so it seems unlikely that it's totally broken, more likely just a misconfiguration somewhere, which is often a missing power regulator or clock config option
<nroberts>
yeah, that seems likely. the instructions I was trying to use don’t seem very helpful
<daniels>
any vendor BSP's instructions are a horribly misleading trap
<nroberts>
I’ve been spoiled by the raspberry pi website I guess :p
jernej has quit [Remote host closed the connection]
<daniels>
yeah, the fact that RPi has very well-documented and pretty universal support is great; the fact it has nothing at all to do with upstream (apart from Mesa!) is pretty sad
<daniels>
but this is certainly OT
<alyssa>
and then boards like rk3399 have great upstream support but are still absolutely painful to do bringup..
<daniels>
probably worth poking eballetbo and getting him to write a companion which will push the image through to eMMC
<daniels>
we definitely have scripts from demos of yore which detect that they've been booted from micro-SD rather than eMMC, copy the appropriate things to the appropriate places, set the appropriate flags, display the appropriate 'remove the SD card and reboot' message, and boot into the appropriate OS when you do so
<alyssa>
Oh, this is new to me :o
<daniels>
!
<alyssa>
hm.
enty is now known as ente
BenG83 has quit [Quit: Leaving]
archetech has joined #panfrost
<macc24>
alyssa: hey, i like depthcharge!
<icecream95>
alyssa: I'm sure I tested texbos on Bifrost. Maybe the alignment requirements are different for v7 compared to v6?
<alyssa>
^^ might be worth seeing what they set on v6
<icecream95>
I can run anything on the G72 blob as long as it's dEQP
<alyssa>
lol
<alyssa>
icecream95: Alternatively just see if unpatched Panfrost passes dEQP-GLES31.functional.texture.texture_buffer.render.as_vertex_texture.offset_1_alignments
<alyssa>
and if it doesn't whether the posted patch fixes it
<icecream95>
alyssa: That passes (with alignment at 16)
<alyssa>
Interesting.
<alyssa>
So it's possible they bumped it on v7, yeah
<alyssa>
Or they just have a silicon bug :>
jolan has quit [Quit: leaving]
jolan has joined #panfrost
hexdump0815 has joined #panfrost
kaspter has quit [Quit: kaspter]
guillaume_g has quit [Quit: Konversation terminated!]
tomboy64 has quit [*.net *.split]
remexre has quit [*.net *.split]
alyssa has quit [*.net *.split]
ids1024 has quit [*.net *.split]
klaxa has quit [*.net *.split]
griffinp- has quit [*.net *.split]
samueldr has quit [*.net *.split]
alpernebbi has quit [Quit: alpernebbi]
alyssa has joined #panfrost
tomboy64 has joined #panfrost
remexre has joined #panfrost
klaxa has joined #panfrost
ids1024 has joined #panfrost
samueldr has joined #panfrost
griffinp- has joined #panfrost
samueldr has quit [Max SendQ exceeded]
tomboy64 has quit [Max SendQ exceeded]
samueldr has joined #panfrost
tomboy64 has joined #panfrost
<alyssa>
netspit
<icecream95>
The blob uses an alignment of 64 on G72
<alyssa>
Interesting
kherbst has joined #panfrost
karolherbst has quit [Ping timeout: 272 seconds]
davidlt has quit [Ping timeout: 240 seconds]
nlhowell has quit [Ping timeout: 240 seconds]
kherbst has quit [Ping timeout: 256 seconds]
jernej has joined #panfrost
robmur01_ has joined #panfrost
robmur01 has quit [Ping timeout: 240 seconds]
<icecream95>
alyssa: Minetest does fault with older Mesa versions. It uses ARB shaders, so maybe there's a bug with uniform handling there
<alyssa>
Hrm.
<alyssa>
Not sure what the correct behaviour is then
<icecream95>
Looks like it isn't an ARB shader after all, it's a generated fixed-function shader reading the gl_AlphaRefMESA uniform
<alyssa>
I have so many questions ;-;
<icecream95>
alyssa: Such as?
<alyssa>
Why does that have its own UBO/
<alyssa>
Or is that UBO #0 and there's nothing bound to UBO #0?
<alyssa>
Why wouldn't UBO #0 be bound that's absurd
<icecream95>
alyssa: Generated fixed-function shaders usually don't use any uniforms, but nir_lower_alpha_test adds one
* alyssa
did some preliminary work on modifier fusing
<alyssa>
and now I shall poof for the weekend \o
hexdump0815 has quit [Ping timeout: 240 seconds]
archetech has quit [Quit: Konversation terminated!]
warpme_ has quit [Quit: Connection closed for inactivity]