alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - Logs https://freenode.irclog.whitequark.org/panfrost - <daniels> avoiding X is a huge feature
stikonas has quit [Remote host closed the connection]
remexre has joined #panfrost
rhyskidd has quit [Remote host closed the connection]
rhyskidd has joined #panfrost
<remexre> is EXT_shader_pixel_local_storage on the horizon? trying to learn how to do deferred shading on my pbp but that may have been a mistake, heh
<remexre> (including a version that's available on GLES2 in violation of the spec, I suppose)
<HdkR> That extension isn't even supported in core mesa, so it would be pretty low priority
<HdkR> Adreno might be the better initial target for implementing that extension
<remexre> rip, I don't have one of those; I guess I'll do a desktop GL backend for now and do a GLES3+pixel-local-storage backend when/if I get hardware that does it
vstehle has quit [Ping timeout: 240 seconds]
kaspter has quit [Remote host closed the connection]
kaspter has joined #panfrost
buzzmarshall has quit [Remote host closed the connection]
davidlt has joined #panfrost
nerdboy has quit [Ping timeout: 246 seconds]
kaspter has quit [Ping timeout: 258 seconds]
kaspter has joined #panfrost
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 260 seconds]
camus1 is now known as kaspter
<agrisis> HdkR: I managed to compile the kernel but odroid go2 won't boot off it
<HdkR> huh
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 246 seconds]
camus1 is now known as kaspter
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 256 seconds]
camus1 is now known as kaspter
kaspter has quit [Ping timeout: 256 seconds]
kaspter has joined #panfrost
bnieuwenhuizen has quit [Ping timeout: 246 seconds]
bnieuwenhuizen_ has joined #panfrost
bnieuwenhuizen_ is now known as bnieuwenhuizen
vstehle has joined #panfrost
agrisis has quit [Ping timeout: 272 seconds]
icecream95 has joined #panfrost
indy has joined #panfrost
<tomeu> mmind00: I think what needs to be upstreamed from that branch is mostly the meson bits (which rockchip shouldn't need), and avoiding GPU_LATEST_FLUSH_ID in the G31
guillaume_g has joined #panfrost
rhyskidd has quit [Remote host closed the connection]
rhyskidd has joined #panfrost
raster has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
<tomeu> mmind00: I'm pushing a panfrost-go-advance branch to https://gitlab.freedesktop.org/tomeu/linux that should be enough for that device
<tomeu> mmind00: I think the patches are fine to be mainlined, but haven't tested them and don't know when I will have time for that
raster has joined #panfrost
<tomeu> if you (or anybody else) will be able to test them, I can send to lkml and we can take it from there
<tomeu> it's based on top of linus/master because depends on a bunch of stuff going into 5.8
<mmind00> tomeu: woohoo ... I can probably do that testing - there are enough px30/rk3326 devices on my desk right now ;-)
rhyskidd has quit [Quit: rhyskidd]
<tomeu> nice!
rhyskidd has joined #panfrost
<tomeu> let me see if it builds :p
<tomeu> chewitt: thanks, fixed :)
<chewitt> also, any chance you can rebase the bifrost branch to collapse the commit/revert pairs
<tomeu> ok, let em clean it up
<chewitt> I need to x-ref with what I have in my branch
<chewitt> it's bugging me why Kodi works for Alyssa on a G31, but I see messed up text labels in the GUI
<chewitt> btw, I only have the Amlogic custom reset logic patch active for G12, not for GXM devices
<chewitt> ISTR some work was done on power things a while back that made it not required on GXM
<tomeu> haven't even built it yet
<chewitt> it's basically the same, but G12 only
<tomeu> yeah, I don't really know, would need to do some testing to figure out
<tomeu> what works and what doesn't
<chewitt> building..
<chewitt> LGTM http://ix.io/2oO2
<chewitt> image was built from my kernel branch not yours, but with effectively the same patches
<chewitt> next up.. G52
<chewitt> G31 still doesn't show text in the Kodi GUI right, but that's not a regression
<chewitt> no difference on G52 either
<chewitt> dmesg http://ix.io/2oO4
<chewitt> G52 (VIM3) shows some faults that I don't see on G31 (VIM3L)
<chewitt> If you want to see what I mean about missing text https://chewitt.libreelec.tv/testing/9.80/LibreELEC-Amlogic.arm-9.80.0-odroid-n2.img.gz
<chewitt> there are images (with u-boot) for C4, VIM3/3L in the same folder
<chewitt> just write to SD card or emmc module
stikonas has joined #panfrost
robmur01 has joined #panfrost
stikonas_ has joined #panfrost
stikonas has quit [Ping timeout: 272 seconds]
tlwoerner has quit [Ping timeout: 256 seconds]
<mmind00> tomeu: patches work nicely on my px30-based board ... all glmark2 tests do run with mesa-master as well and I get a glmark2-score of 31 :-D
<robmur01> aww, that's too good... now we just have to rig it so that every Gxx-series GPU scores its own model number :D
<icecream95> robmur01: I get glmark2 scores of ~700 on T760, so almost there...
<mmind00> tomeu: out of curiosity, what is the "bunch of stuff going into 5.8" ... doing a git log on drivers/gpu/drm/panfrost doesn't show so many changes ?
<mmind00> (changes seem to be from march mainly)
rcf has quit [Ping timeout: 256 seconds]
rcf has joined #panfrost
nlhowell has quit [Ping timeout: 260 seconds]
rcf has quit [Remote host closed the connection]
rcf has joined #panfrost
rcf has quit [Remote host closed the connection]
rcf has joined #panfrost
rcf has quit [Remote host closed the connection]
rcf has joined #panfrost
icecream95 has quit [Quit: leaving]
rcf has quit [Ping timeout: 240 seconds]
rcf has joined #panfrost
rcf has quit [Ping timeout: 246 seconds]
rcf has joined #panfrost
tlwoerner has joined #panfrost
rcf has quit [Ping timeout: 256 seconds]
rcf has joined #panfrost
nlhowell has joined #panfrost
nlhowell has quit [Ping timeout: 256 seconds]
rcf has quit [Ping timeout: 265 seconds]
cwabbott_ has joined #panfrost
rcf has joined #panfrost
cwabbott has quit [Ping timeout: 272 seconds]
cwabbott_ is now known as cwabbott
nlhowell has joined #panfrost
agrisis has joined #panfrost
rcf has quit [Ping timeout: 246 seconds]
rcf has joined #panfrost
rcf has quit [Remote host closed the connection]
rcf has joined #panfrost
robmur01 has quit [Quit: Leaving]
robmur01 has joined #panfrost
warpme_ has joined #panfrost
bnieuwenhuizen has quit [*.net *.split]
milkii has quit [*.net *.split]
bnieuwenhuizen has joined #panfrost
milkii has joined #panfrost
<agrisis> HdkR: do you know how I can get in contact with the person that claims to have been able to boot an OGA using that kernel image?
rcf1 has joined #panfrost
rcf1 has quit [Client Quit]
<alyssa> remexre: PLS requires a lot of frontend mesa changes
<alyssa> esp at the GLSL level
<alyssa> tons of work for something no apps on linux (afaik) use
<alyssa> maybe fp16 is enough to get 760 on t760
<alyssa> + afbc
<alyssa> and if that's >760, disable one of them :p
<remexre> alyssa: oh, that's unfortunate... is supporting more color attachments closer?
<tomeu> mmind00: oh, was referring to the go2 additions
<tomeu> chewitt: have you disable runtime PM?
<tomeu> that seems to be needed on amlogic, as the GPU behaves erratically on power up
<mmind00> tomeu: ah ok ... thought that you were talking about other drm stuff ... but it sure is cool to have that running on px30 now :-D
<alyssa> remexre: yes, much
<alyssa> MRT in theory should be implemented now (with gles3 enabled) but in practice there are some bugs
<remexre> oh, right, I forgot there was an env var to do that
<remexre> I should probably install updates too, heh; glGetIntegerv(GL_MAX_COLOR_ATTACHMENTS_NV, ...) still gives me 1
<remexre> er, that = gles3
<alyssa> if you're going to try to use gles3 stuff definitely stick to master yes :)
<alyssa> (and that way if you find bugs you can fix them ;) )
<chewitt> tomeu: how would I disable runtime PM?
<chewitt> the three changes for bifrost I have in my branch are:
<chewitt> amlogic runs the gpu at fixed speed (not dvfs) if that's what you meant?
<alyssa> iirc tomeu's referring to powering on/off the gpu at runtime in response to workload
<alyssa> which is important to save power but for a set-top box is pretty eh
<robmur01> chewitt: `echo on > /sys/devices/<blah>/power/control`
<remexre> oh, weird, even before updating, PAN_MESA_DEBUG=gles3 increases the number of color attachments to 4; are multiple color attachments experimental on gles2 w/ the nvidia extension as well, or is this a misbehavior, or will this change once newer mesa finishes compiling?
<tomeu> chewitt: what Rob said, that way the GPU will stay on all the time
<tomeu> please tell us how well that works
<alyssa> remexre: I can't remember what made it past which branch point tbh
<remexre> lol, fair enough; I guess I'll see in 10 min if it changes
<robmur01> I guess we could probably quirk enabling RPM in the kernel driver if it proves reasonable
Elpaulo has joined #panfrost
<chewitt> I added http://ix.io/2oPg to a script that runs at the start of userspace boot
<chewitt> but I still see this kind of thing
<chewitt> [ 8.208082] panfrost ffe40000.gpu: js fault, js=0, status=DATA_INVALID_FAULT, head=0x5c21800, tail=0x5c21800
<chewitt> [ 8.208120] panfrost ffe40000.gpu: gpu sched timeout, js=0, config=0x7300, status=0x58, head=0x5c21800, tail=0x5c21800, sched_job=00000000eed13461
<chewitt> full dmesg http://ix.io/2oPh
<chewitt> I don't see any of these errors on a G31 board, only G52
<chewitt> and the text is still missing in the Kodi GUI, but I think that's some other issue
<chewitt> with TODO error http://ix.io/2oPk
<alyssa> it's entirely possible G52 has some difference in the { command stream, shaders } from G31 that I haven't noticed
<alyssa> my G52 board is meson so I'm pending on the errata being figured out so..
<chewitt> I don't visually see any difference between them in the Kodi GUI
<chewitt> only that G52 has those faults logged
<alyssa> yeah, that's really strange
<chewitt> are you running Kodi in wayland?
<alyssa> I checked with an apitrace
<alyssa> since I didn't want to figure out kodi's build system and the debian package is x11-only
<remexre> hm, do I need to reboot after updating mesa, or should restarting X be sufficient? I'm still getting max color attachments = 1 for gles2 after restarting X
<tomeu> hmm, hard to tell if those errors could be related to the meson errata or not
<alyssa> remexre: you still need =gles3
<alyssa> I haven't flipped on gles3 by default yet
<remexre> ah, okay
<alyssa> mostly because there are bugs. a lot of bugs.
<alyssa> and for apps that can do both gles2 and gles3, atm it's usually better for end users to use the gles2 one
<remexre> I was thinking it'd make sense (if color attachments by themselves are non-buggy) to increase the limit for gles2 via the one nvidia extension
<chewitt> I don't feel the faults no G52 are due to it being 'amlogic'
<chewitt> because G31 is stable, and when using blob there's no difference in the power-up hack we (narmstrong) added
<alyssa> blob is with kbase but panfrost is with mainline panfrost, no?
<chewitt> i.e. both G31 and G52 are using the same hack
<alyssa> so that doesn't rule out something in kbase that mainline is missing
<alyssa> (there could *also* be mesa issues. but I would expect to see that from the apitrace.)
<chewitt> also, despite the faults in dmesg I can navigate around the GUI fine (using memory.. as no text)
<alyssa> chewitt: I wonder if you could do PAN_MESA_DEBUG=trace and send some resulting pandecode.* files? (warning: large)
<alyssa> I don't have much confidence it will help but maybe something obvious
<chewitt> where are they logged to?
<alyssa> particular files that contain the string `fault_pointer` or `exception_status`
<alyssa> uhhh you tell me
<alyssa> current directory, whatever that is for libreeelc
<chewitt> one way too find out..
<alyssa> pff.
nlhowell has quit [Ping timeout: 256 seconds]
<chewitt> how to have multiple values in `PAN_MESA_DEBUG=trace` .. I need `PAN_MESA_DEBUG=bifrost` to start Kodi
<la-s> chewitt: comma
<la-s> i.e., trace,bifrost
<chewitt> kodi.sh[1621]: pandecode: failed to open command stream log file pandecode.dump.0000
<chewitt> and it fails to start
<robmur01> I would tend to assume that faults and stuff not rendering are related - it's not entirely impossible that whatever goes wrong just happens to go wrong on G31 in a way that manages not to fault
<chewitt> MESA_LOG_FILE allows to redirect output, but I don't think that's applicable
nlhowell has joined #panfrost
Werner has quit [Quit: ZNC 1.7.2+deb3 - https://znc.in]
Werner has joined #panfrost
<alyssa> robmur01: cheaping out on error detection? :P
<robmur01> heh, I'm thinking more like uarch-specific undefined behaviour, or subtle second-order effects - e.g. loading a pointer from the wrong place, which due to descriptor differences might contain a mapped (but still bogus) address in one case and NULL in another
<alyssa> fair
<alyssa> ---Oh gosh
<alyssa> really mali?
guillaume_g has quit [Quit: Konversation terminated!]
buzzmarshall has joined #panfrost
nerdboy has joined #panfrost
nerdboy has quit [Ping timeout: 265 seconds]
<HdkR> alyssa: I know of only one linux application that uses PLS :P
<daniels> let me guess ... Dolphin
<HdkR> Nah, Dolphin doesn't need it
<HdkR> Redream the Dreamcast emulator uses it
raster has quit [Quit: Gettin' stinky!]
<alyssa> HdkR: ...Dreamcast?
<urjaman> the 90s game console
<HdkR> Dreamcast has a PowerVR GPU in it that can do order independent transparency in hardware. So it's a bit tricky :P
<alyssa> yeah but
<alyssa> PLS? :p
<HdkR> Pretty sure Anthony uses it for optimizing the OIT emulation
<HdkR> Makes it faster on mobile chips
<alyssa> I guess that also works on literal PowerVR devices
<HdkR> I don't know if they ever exposed a GL extension for it
<alyssa> yeah, PLS :)
<alyssa> mali and pvr
<HdkR> Oh, I thought you meant the OIT bit :P
<HdkR> Yes, PLS can work on Mali, PVR, Adreno, Intel, and ....latest AMD I think?
<alyssa> it's only exposed on the first two afaik
<alyssa> at any rate, PLS lives on as vulkan subpasses
<HdkR> ew :P
raster has joined #panfrost
unoccupied has quit [Quit: WeeChat 2.5]
raster has quit [Quit: Gettin' stinky!]
raster has joined #panfrost
macc24 has quit [Quit: WeeChat 2.8]
macc24 has joined #panfrost
warpme_ has quit [Quit: Connection closed for inactivity]
tgall_foo has quit [Ping timeout: 260 seconds]
raster has quit [Quit: Gettin' stinky!]
tgall_foo has joined #panfrost
raster has joined #panfrost
davidlt has quit [Ping timeout: 246 seconds]
nerdboy has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
rcf has quit [Quit: WeeChat 2.7.1]
rcf has joined #panfrost
yann has quit [Ping timeout: 246 seconds]
nerdboy has quit [Ping timeout: 264 seconds]
nerdboy has joined #panfrost
yann has joined #panfrost
robink has quit [Ping timeout: 260 seconds]
robink has joined #panfrost
yann has quit [Ping timeout: 256 seconds]
nerdboy has quit [Ping timeout: 258 seconds]
tlwoerner has quit [Quit: Leaving]
macc24 has quit [Ping timeout: 246 seconds]