felipealmeida has quit [Ping timeout: 240 seconds]
felipealmeida has joined #panfrost
uis has quit [Client Quit]
uis has joined #panfrost
uis has quit [Client Quit]
guillaume_g has joined #panfrost
<bbrezillon>
amonakov: thank you
<amonakov>
bbrezillon: missed an essential tweak there, please apply this patch: http://sprunge.us/wi3kSx
<amonakov>
#ProgrammerHumour
<bbrezillon>
unfortunately that will disappear when converted to NIR :P
kaspter has quit [Quit: kaspter]
kaspter has joined #panfrost
alpernebbi has joined #panfrost
<macc24>
what's up with G71?
* macc24
spotted kirin 960 laptop that hasn't been mainlind
<HdkR>
G71 is first gen so it's missing transcendentals, so it's missing compiler bits
karolherbst has joined #panfrost
<HdkR>
Also Kirin 960 dev board is available for testing
<macc24>
hmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmm
<HdkR>
Or more like, s/testing/development
<macc24>
i am not spending $200 for a sbc xD
nlhowell has joined #panfrost
stikonas has joined #panfrost
uis has joined #panfrost
uis has quit [Client Quit]
uis has joined #panfrost
leper` has quit [Ping timeout: 260 seconds]
uis has quit [Client Quit]
leper` has joined #panfrost
raster has joined #panfrost
gcl_ is now known as gcl
karolherbst has quit [Quit: duh 🐧]
karolherbst has joined #panfrost
karolherbst has quit [Quit: duh 🐧]
karolherbst has joined #panfrost
uis has joined #panfrost
uis has quit [Client Quit]
Net147 has joined #panfrost
<alyssa>
HdkR: I have patches for G71 transcendentals here
<HdkR>
:>
<alyssa>
amonakov: I don't get it.
<amonakov>
alyssa: the humour in the diff about the "magic wand" operator?
<alyssa>
Yeah
macc24 has quit [Ping timeout: 264 seconds]
<amonakov>
this is a riff on that, playing with how a '*-' can be seen as a magic wand (a stick with a star on end)
<amonakov>
well, there was this famous stackoverflow question about "what does the --> operator do in C"
<alyssa>
OK
<amonakov>
this was funny in my head
<HdkR>
You could use *-- to make it better
<amonakov>
that would be the "long magic wand" operator ;)
<alyssa>
i have too many branches, let's add another ig
macc24 has joined #panfrost
<robmur01>
ah yes, the good old "tends toward" operator :D
<robmur01>
BTW I did try setting up G71, but despite being the exact same bitfile that worked a couple of months ago, the GPU now doesn't reset properly :(
<robmur01>
please don't make me have to set up the Hikey...
<yann>
on rk3399 modetest gets EOPNOTSUPP on DRM_IOCTL_SET_CLIENT_CAP and DRM_IOCTL_MODE_GETRESOURCES, any idea why it would ?
<yann>
(on a custom board with kenel 5.10.8)
uis has joined #panfrost
uis has quit [Client Quit]
yann has quit [Quit: Quitte]
yann has joined #panfrost
<yann>
/dev/dri/card0 is platform-ff9a0000.gpu-card - I'm also suprised that card1 (which is platform-display-subsystem-card, no even sure how it relates to the GPU) is not properly selected by "modeset -D", strace still shows that card0 is used, whatever I pass as argument to -D
<macc24>
yann: how did you get this error?
camus has joined #panfrost
kaspter has quit [Ping timeout: 246 seconds]
camus is now known as kaspter
<robmur01>
oh, you probably do need to use the display device, since the GPU itself is just a render-only device. If tools expect card0 to be the display, the cheap hack is just to make sure the rockchipdrm driver probes before panfrost does
<macc24>
yann: FYI mali gpu *only* does 3d acceleration, and for displaying results is other IP core/module/whatever-i-dont-know-naming, in your case rockchip's own thing
<urjaman>
and the simplest hack to make rockchipdrm to probe first is rockchipdrm built-in, panfrost module
karolherbst has quit [Remote host closed the connection]
karolherbst has joined #panfrost
<yann>
ok - but "modetest -D /dev/dri/card1" should normally open that device instead of card0, shouldn't it ?
<yann>
macc24: not sure of the scope of your "how did you get this error?" - the immediate context is running "modetest" (with or without "-M rockchip")
<yann>
it seems from dmesg output that panfrost is indeed the first one loaded, "rockchip-drm display-subsystem" and "dwhdmi-rockchip" appear only afterwards
<KungFuJesus>
is it expected that glamour performance is lackluster right now? Drawing terminal windows seems to be a bit on the slow slide, slower than it was with the 'fbturbo' with llvmpipe
<alyssa>
glamor perf is known to be suboptimal, yeah :(