<bnieuwen1uizen> HdkR: what kind of buildbot craziness are you trying to do?
<HdkR> Using buildbot's latent workers to have a PC that wakes up from sleeping then puts itself to sleep when done. PC will be cross-compiling and then pushing binaries off to other devices
<bnieuwen1uizen> oh fancy
<bnieuwen1uizen> you don't have a pc pretty much permanent on at home?
<HdkR> This PC is a threadripper system that consumes 137w idling
<HdkR> Or 107w if I kick its GPU in to P8 state, but it boots in P0 because oops
<bnieuwen1uizen> roughly same here (though a bit more power because 4 GPUs). Honestly, over the cost of the threadripper system I considered that cost acceptable
<HdkR> Yea, it isn't terrible of a cost but I'm also curious about messing with latent workers
<HdkR> I'll also be sticking another GPU or two in to it in the future
<bnieuwen1uizen> has it been stable for and what mobo are you using?
<HdkR> I'm using an MSI X399 MEG Creation on it
<HdkR> There are two stability issues with it
* bnieuwen1uizen still gets a full system lockup every month or so
<HdkR> 1) The I211 gigabit ethernet controllers on it sometimes don't wake up correctly from sleeping
<alyssa> HdkR: bnieuwen1uizen: Y'all are insane. What is this "100 watts" you speak of?
<bnieuwen1uizen> alyssa: 16-32 core processors
<alyssa> energy-rate: 3.5644 W
<HdkR> 2) The Nvidia GPU sometimes freaks the heck out when waking up from sleep, forcing me to hard reset the system
<bnieuwen1uizen> with way too many GPUs
<HdkR> Mine is a 2990WX with the 32 cores :P
<bnieuwen1uizen> mine too
<HdkR> Definitely worth the investment
* bnieuwen1uizen wonders if I can get some easy power numbers for my qualcomm SoC, that should be closer to the 3W idling
<alyssa> Hah! Your puny cores pale in comparison to the Great and Powerful Kevin!
<HdkR> If I can figure out why the I211 device breaks and why nvidia-uvm hatest restoring from sleep then I think it is perfect.
<HdkR> s/hatest/hates
<bnieuwen1uizen> alyssa: it is very useful to bring L:LVM compile times down to < 5 minutes
<HdkR> ^
<HdkR> That is exactly what I'm using it for
<alyssa> Well, there's your problem ;)
<alyssa> LLVM? In _my_ channel? Blasphemy! Heresy!
<alyssa> :P
<urjaman> the C201 is using 2.51W. The entire system, including the display and backlight :P
<HdkR> hehe
<alyssa> urjaman: Yeah but the C201 is unusable for anything more intensive than IRC sooo ;)
<HdkR> That's also why I'm not too concerned about a million ARM boards in my storage once I get CI going on those
<bnieuwen1uizen> alyssa: now if you could convince the other AMD devs of that I'd be happy too ;)
<bnieuwen1uizen> are the mali boards pricey?
<HdkR> I can get ~50 ARM boards before it matches the idling cost of TR
<HdkR> $45-$100 for SBCs usually
<urjaman> now if you'd get me some 3D drivers i wouldnt need to have one of them running ChromeOS ... :P
<bnieuwen1uizen> oh that is doable. (my qualcom board was > $500 ... which is almost a full PC/laptop)
<HdkR> Although Utgard boards are suuuuper cheap
<HdkR> and there wil be some Bifrost G3x/G5x boards coming eventually getting to those prices
<bnieuwen1uizen> HdkR: you'll need a lot of them to get fast CTS times though ;)
<HdkR> Yea
<bnieuwen1uizen> thenb again, no vulkan yet, no clue how fast/slow GLES2 CTS is
<HdkR> 10-20 in the near term will be nice
<bnieuwen1uizen> a full vulkan CTS run on atom class devices that ~60 min
<HdkR> oof
<bnieuwen1uizen> well, at least I got it down from 24 hours+ ;)
<HdkR> Is this just splitting CTS tests across multiple devices?
<bnieuwen1uizen> HdkR: now that I think about it what are you doing with hard reset?
<bnieuwen1uizen> if you want to do it via power switches, networked power switches can be pretty expensive per outlet
<HdkR> I have a USB controlled relay connected to the reset switch on the device
<HdkR> Which are something like $5
<bnieuwen1uizen> connected how?
<bnieuwen1uizen> you soldered the button off?
<HdkR> Nah, I have some cables connected to the reset header on the motherboard, run them outside of the case to my relay board
<bnieuwen1uizen> oh a mobo with actual connections instead of a button on the mobo
<bnieuwen1uizen> that works ;)
<HdkR> Right
<HdkR> Standard pinout for the front panel reset button
<HdkR> The relay board can handle something like 200w per relay, so in the future if I need to connect a board's power supply to it then it'll be easy
<HdkR> The company I bought it from also supports...16? relay versions that can be switched over ethernet
<HdkR> Although the ethernet one is a bit more expensive
<bnieuwen1uizen> well, divided over 16 boards, quite reasonable
<HdkR> right
<HdkR> Ethernet would ensure no shenanigans like multiple USB ones connected to a master, then them never appearing in the same order in /dev/ttyUSB*
<HdkR> If we require power cycling on ARM boards then I'm going to get the ethernet one
<bnieuwen1uizen> right, lets hope that GPU reset works for now
<HdkR> I'm also not accomplishing any CI things today like I thought I would since I managed to capture a fever
<bnieuwen1uizen> HdkR: btw did you look at the intel mesa_ci_results project for showing test results?
<HdkR> Looked at it. Haven't tried setting it up yet
<HdkR> Would be nice to use
<bnieuwen1uizen> okay, I have a config to put it entirely in docker including the mysql database if that would be useful
<HdkR> There is a good chance that will be useful
<bnieuwen1uizen> https://gitlab.freedesktop.org/bnieuwenhuizen/radv-ci , main contribution is setting up the SQL users so that you can have mysql in docker
<HdkR> nice
<HdkR> I'll make sure to pull that when I am setting that up
chewitt has quit [Excess Flood]
chewitt has joined #panfrost
<chewitt> @alyssa https://freenode.irclog.whitequark.org/panfrost is active now - if you could add the link to the channel topic
<chewitt> @HdkR are you aware of the "lab in a box" stuff that Baylibre did for kernelci.org ?
<HdkR> I'm not aware of that at all
<chewitt> I'll try to find the videos .. it's been presented at ELC a couple of times
<chewitt> ^ is probably talking generally about kernelci
<chewitt> but the mess of boards/psu's/cables in the pic half-way down that page got rationalised into something more compact
<chewitt> and it's all ARM (Amlogic) boards
<chewitt> khilman in #linux-amlogic is worth talking to about what you need to build
<chewitt> before you reinvent the wheel :)
<HdkR> Neat, is this specific to kernel testing since I'm not currently planning on kernel testing :P
<chewitt> theirs is specific to kernel testing .. but whatever testing you do, if it's with ARM boards you'll have the same power/connectivity type stuff to solve
<HdkR> Nice
<HdkR> I'll make sure to watch
<chewitt> so, dumb question .. what is it you want to test?
<HdkR> Mesa piglit mostly, but will want to run through CTS later
_whitelogger has joined #panfrost
embed-3d has quit [Ping timeout: 252 seconds]
<chewitt> problems with the nvidia driver post-wake from suspend are long (long) running
<chewitt> (recalling an earlier comment)
<chewitt> each time a new driver comes out with a "fixed suspend issues" comment in release notes there's a flurry of enthusiastic testing by users in our forum
<chewitt> followed by a bunch of pessimistic "nah, it's still broken" comments
<HdkR> Sounds about right
anarsoul has quit [Read error: Connection reset by peer]
anarsoul has joined #panfrost
alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - https://gitlab.freedesktop.org/panfrost - Logs https://freenode.irclog.whitequark.org/panfrost - Discord Discard
<alyssa> chewitt: Thank you!
<chewitt> np
<HdkR> h, I didn't even look at what that link was :P
<HdkR> ah*
chewitt has quit [Quit: Zzz..]
chewitt has joined #panfrost
merry has quit [Read error: Connection reset by peer]
merry has joined #panfrost
chrisf has quit [Ping timeout: 244 seconds]
wookey has quit [Ping timeout: 244 seconds]
chrisf has joined #panfrost
chewitt has quit [Quit: Zzz..]
chewitt has joined #panfrost
chewitt has quit [Client Quit]
chewitt has joined #panfrost
chewitt has quit [Ping timeout: 252 seconds]
chewitt has joined #panfrost
TheCycoONE has quit [Quit: ZNC 1.7.1 - https://znc.in]
TheCycoONE has joined #panfrost
cwabbott has quit [Ping timeout: 272 seconds]
cwabbott has joined #panfrost
<alyssa> Iiiit's time to do that thing I vowed to be responsible enough not to!
ric96 has joined #panfrost
jernej has joined #panfrost
<urjaman> huh?
<alyssa> urjaman: Blending!
<alyssa> alyssa: Just wrote documentation for the fixed-function blending hardware (possibly the most complex part of the driver) and commented the accompanying code extensively.
<alyssa> alyssa: This is all your fault :^)
<alyssa> alyssa: https://rosenzweig.io/blending.c <--- docs
<alyssa> Admittedly the fixed-function stuff is buggy as it is, should probably put more love into that before diving off into programmable
* alyssa writes a tracer for blend modes
<alyssa> Cool, we're now able to "disassemble" blend modes
<alyssa> It's not an instruction set but it sure feels like one >_>
<alyssa> How these aren't identical is beyond me
<davidlt> IBM bought Red Hat
<davidlt> "Red Hat will join IBM’s Hybrid Cloud team as a distinct unit"
<alyssa> wha?
<alyssa> Is that a good thing or a bad thing?
<davidlt> some don't like this
<urjaman> "hybrid cloud team" ...... okay, that's a mental image
<alyssa> urjaman: Half of them are people
<alyssa> The other half are clouds
<davidlt> IBM didn't do well with other acquisitions in the past
<davidlt> PR states that management, direction, philosophy, etc doesn't change for Red Hat
<urjaman> alyssa: i was thinking technically augmented clouds
<davidlt> With this acquisition, IBM will remain committed to Red Hat’s open governance, open source contributions, participation in the open source community and development model, and fostering its widespread developer ecosystem. In addition, IBM and Red Hat will remain committed to the continued freedom of open source, via such efforts as Patent Promise, GPL Cooperation Commitment, the Open Invention Network and the LOT Network.
<ente> I'm curious what the firmware situation of the PinePhone will be. Any idea which GPU they use?
<ente> PineBook seems to be Mali 400, so maybe if they base this on a newer SoC it might be Midgard?
<alyssa> I recall reading it'd be Mali 400 again but maybe I'm mixing stuff u
<ente> hm..
<alyssa> Yeah
<alyssa> Allwinner A64
<alyssa> ARM Mali400 MP2
<ente> Maybe Panfrost is the hammer I'm trying to apply to all nails I find. :P
<ente> Even if they're actually screws.
<ente> :/
<alyssa> Oh, me too ;)
<alyssa> "You sure you want to buy this laptop? Not sure it'll be enough for your needs" "Huh? There aren't any free GPU drivers for it, that's all I need!"
<alyssa> :P
<ente> I remember a couple of weeks ago when someone gave me a Pine64 that's sitting in his drawer unused and I found out it's Mali400 :D
<ente> "Finally something I can try panfrost....oh."
<anarsoul> ente: try lima?
<ente> But I like it here. :D
<ente> that's how I ended up using NetBSD
<ente> But I guess I should give the lima IRC channel a chance
<anarsoul> ente: it's mostly silent :)
<anarsoul> yuq rarely joins it
<ente> Is it stable?
<ente> maybe I should ask over there.
<alyssa> You think panfrost is stable? "D
<alyssa> :D
<anarsoul> ente: it's not ready yet
<ente> With panfrost I know who to bother if I can't fix it myself (and send a patch)
<anarsoul> ente: with lima just file a bug, just check for existing bugs first before filing a new one
<alyssa> out of curiousity, which is closer to "practical reality for users"? lima or panfrost?
<anarsoul> alyssa: no idea, panfrost is more active so I'd bet on it
<ente> My issue is that I'd rather join the active project where people talk to each other outside of the bug tracker
<alyssa> anarsoul: so "lima but panfrost is moving faster and will surpass sooner"?
<ente> that's the feeling I get
<alyssa> ente: flip side, for all my activity only a quarter of it is actually useful
<anarsoul> ente: you may be in different timezone with other people
<alyssa> since I am VERY EASILY DISTRACTABLE :p
<anarsoul> ente: we also have mailing list!
<ente> How different is Mali-400 from Midgard? Can some things be done together or is it a completely different architecture?
<anarsoul> completely different, both command stream and ISA
<anarsoul> I think only tiling format for textures is common
<ente> having a mailing list is good news though. I'll go subscribe.
<sphalerite> ooh, mailing list??
<ente> (but that's lima, not panfrost, not sure how much you've read)
rellla has joined #panfrost
<sphalerite> oh right
<sphalerite> yeah I only read the last few lines
<alyssa> ^^ what anarsoul said
<ente> I'm also very distractable, you actually seem less distractable than me
<ente> you've been at this for more than 3 months without just throwing it away and doing something else
<ente> but I guess if you actually buy hardware that means you're somewhat invested now
<ente> (or you could put it on a heap and never do anything with it *whistle*)
<alyssa> where was I over the summer again? :D
<anarsoul> alyssa: you were taking some classes
<anarsoul> e.g. "Linear Algebra and Differential Equations"
<alyssa> ah yes this is true
<alyssa> well that and GroupMe :^)
<alyssa> :P
* anarsoul barely remembers a thing from linear algebra
<ente> university getting in the way of writing FOSS drivers again
<alyssa> Tut tut
* ente remembers *some* level of linear algebra but then again I studied math for a while
<ente> and before that cS
<ente> and in the end I discovered that neither math nor CS is my problem but studying at university is
<alyssa> Ooo! There -is- limited support for constant colors in blending
<alyssa> Specifically if all the constants are the same, it's happy
<alyssa> I.e. we can blend with an immediate float, but not an immediate vector
<alyssa> ...Am I the only one excited about that? :P
<alyssa> Can I just say
<alyssa> This fixed-function blending hardware is _awesome_
<alyssa> Like it's a major pain to program for and was easily the hardest thing in the cmdstream to RE but
<alyssa> Oh my gosh it's so adorable
jernej has quit [Ping timeout: 246 seconds]
<ente> blog post please? :>
<alyssa> ente: on...?
<ente> the adorable blending hardware? :D
<urjaman> ...
<alyssa> ente: I mean, I kind of already did
<ente> nevermind then
<ente> I thought there would be something new because you were excited
<alyssa> I dunno it's just adorable hw
<alyssa> :p
<alyssa> The only new is a really small point that some constant modes are allowed after all
<alyssa> It's just
<alyssa> cute okay
<alyssa> no me juzgues
<ente> I see!
<ente> I vaguely remember that blog post
<ente> I've too much other things in my head to keep it all sorted
<ente> many*
<ente> which is probably a mistake, I know
* alyssa shrugs
<alyssa> I just had a funny visual of a blob writer reading my blog post and just thinking "no! you got it all backwards! the hw--" but not being allowed to say anything :p
<ente> rest assured that I'm not a blob writer
<HdkR> Spooky blob writers
<alyssa> HdkR: Hey, I know what I'm going as for Halloween!
<HdkR> The horror
<ente> but I'd say the blob writers don't get to complain. If they demand perfect code, they should not have made a blob.
<ente> alyssa: what does a spooky blob writer look like?
<ente> I mean, a blob writer is probably a more difficult costume than a blob
<alyssa> You tell me
<ente> given that a blob writer looks just like a normal programmer with some standards
<alyssa> :v
<ente> maybe that's the creepy part, we could all be blob writers. we have it inside us, this animalistic thing. all of us humans.
<alyssa> I wouldn't write blobs
<ente> you say that now.
<ente> *thunder*
<alyssa> ...I mean I guess if money got really tight T_T
<ente> I'd like to see more FOSS funding in the EU
<HdkR> ente: I'm supposed to have standards?!
* HdkR has been doing it wrong
<ente> You're on one of the best channels on IRC.
<alyssa> PSA: The type-c charger on Kevin isn't well-shielded; combined with an electronics-heavy headphones that are also not well-sheilded (noise cancellation box kind of thing), and the close proximity of the charger if you charge on the left side [rather than the right], audible EMF interference can occur.
<ente> you can use any usb-c port?
<HdkR> PSA: My laptop also has the same problem D:
<alyssa> Solutions include "charge on the other side", "use passive headphones", "move the headphone box away from the charger", et
<ente> my current laptop only charges from the left
<alyssa> ente: kevin charges on both but I've only ever used left-charging laptops so it's force of habit
<alyssa> charge on the left, wifi on the right
<ente> I'm really happy with HP except for the lack of finished libreboot support. And the fact that it's x86.
<alyssa> Join the ARM side; we have cookies
<HdkR> It's really hard to find an ARM device that has everything that I want D:
<ente> I probably will, this thing is heavy.
<HdkR> x86 devices have spoiled me
<ente> .oO("join the ARMy")\
<alyssa> HdkR: For me that's Kevin
<HdkR> :D
<alyssa> I.... literally can't imagine using any other computer
<HdkR> The main thing for me now is high resolution. I can't go back to a low resolution terminal now
<alyssa> Just hoping it lasts
<alyssa> (In terms of both physical wear/tear, as well as "will they still exist when I need to replace some number of years down the line?"... or maybe there'll be something genuinely better then)
<bnieuwen1uizen> alyssa: what makes kevin so well-liked for you?
<bnieuwen1uizen> HdkR: also non-x86 cpus are kind of annoying when running your average desktop game (assuming you want to make sure that games work on your driver of course :P )
<HdkR> bnieuwen1uizen: Also really annoying when almost nothing supports ES in the desktop space :P
<bnieuwen1uizen> come to the dark side, implement vulkan
<bnieuwen1uizen> a lot of games run pretty well with just the ES equivalent of vulkan features
<HdkR> True
<HdkR> One of the boards has Vulkan on it. Could trace from it directly
<HdkR> Although it's on Android only
<bnieuwen1uizen> I think everyone in the mobile space has that limitation right now of vulkan being android only?
<alyssa> bnieuwen1uizen: In no particular order, HiDPI display, embedded pressure sensitive tablet that isn't total garbage, aarch64 (fast enough for everything I've tried, except for running Discord's proprietary client. Seriously, it's like they decided jQuery was too fast so they went with React), 4th gen Midgard (dev platform anywhere I go), free boot firmware (runs bloblessly... except for the Mali ;) ), 2-in-1 (nice for ebooks and TV, not
<alyssa> that I get around to either much these days), keyboard isn't terrible, type-c is starting to grow on me...
<bnieuwen1uizen> oh hm, the bloblessly does that include firmware?
<HdkR> Pretty sure Mali supports Vulkan on Linux, just need a board shipping those binaries
<bnieuwen1uizen> right, but does anyone do that?
<HdkR> Not yet
<alyssa> bnieuwen1uizen: prety much
<alyssa> There's a firmware blob needed for display out but honestly the internalscreen is higher res than any monitor in my posession so w/e :P
<alyssa> Ships with blobless coreboot
<anarsoul> why do they use coreboot on arm64?
<alyssa> Embedded controller is free sw as well
<bnieuwen1uizen> right, isn't that pretty common on chromebook (blobless coreboot)?
<alyssa> anarsoul: Chromebooks... questions we do not ask :)
<alyssa> Yeah
<bnieuwen1uizen> I was more thinking of e.g. blobs for wifi HW
<alyssa> Oh
<anarsoul> alyssa: that's weird, but OK
<bnieuwen1uizen> unless you use ethernet over USB-C? :P
<alyssa> Yeah, that's a blob, but I've been using a dongle since forever so
<anarsoul> alyssa: what about ATF?
<anarsoul> is it a blob?
<alyssa> ATF?
<anarsoul> arm trusted firmware
<alyssa> No idea
<anarsoul> OK
<alyssa> Someone should probably research that :p
<anarsoul> IIRC it's a blob on rock64pro
<anarsoul> so I suspect same is true for all rk3399 boards
* bnieuwen1uizen is guessing blobless + containing your dev platform are pretty hard things to match
<alyssa> anarsoul: ..disappointing
<alyssa> bnieuwen1uizen: I mean, I chose Midgard since there were boards that were essentially blobless -except- for the Mali
<alyssa> I guess that's cheating tho :)
<anarsoul> alyssa: slightly
<bnieuwen1uizen> alyssa: right, but is going to exclude anything in the future we are making with x86 or non-mali arm or power or whatever
<alyssa> this is true
<alyssa> ...assuming I'm parsing that right
<alyssa> Anyways, I got the constant float blend factor stuff working
<alyssa> It was an insane amount of code, but it works
<alyssa> 5 files changed, 143 insertions(+), 21 deletions(-)
<mmind00> anarsoul: nope ... rk3399 is pretty well supported in upstream ATF ... especially as Google made sure the relevant stuff got upstreamed first
<Lyude> alyssa: so I am considering making my current gameplan for panfrost being trying to get things working on the new kernel api
<alyssa> mmind00: ah?
<anarsoul> mmind00: OK, I'm happy to be wrong here :)
<alyssa> Lyude :D
<Lyude> i mean
<mmind00> I remember reading, that they don't do CrOS-specific ATF patches but instead expect _everything_ to be coming from upstream
<Lyude> new mali kernel api, the one they currently have
<alyssa> ah, yeah
<Lyude> i' m thinking it might be a good place to at least start
<alyssa> Might be a good exercise to dip your toes in
<Lyude> yeah
<Lyude> it's also like, documented now
<cyrozap> anarsoul: To add to what mmind00 said, the only blob part of RK3399 ATF is the HDCP key deobfuscation code, but if you don't care about HDCP you could stub that out.
<alyssa> lol
<Lyude> i might give it a shot without using traces it too much
<alyssa> cyrozap: which is ironic because iirc it's like _one_ function they blobulate and it'd be trivial for someone to just grab IDA and figure it out? :P
<Lyude> *traces too much
<Lyude> also unrelated, but played with trying to use the emmc for stuff on my board and lordy it's so much faster
<alyssa> yup
<anarsoul> Lyude: what did you use prior to emmc?
<cyrozap> alyssa: And it's BSD-licensed, so it's totally cool to RE :D
<alyssa> cyrozap: LOL
<bnieuwen1uizen> wait, BSD on a binary blob?
<Lyude> anarsoul: the sd card, but I have all of the stuff I'm accessing the most on a usb flash drive because it's slightly faster
urjaman has quit [Ping timeout: 264 seconds]
<alyssa> lul
<Lyude> the emmc did not work originally because of hs400 being enabled in the vim2 devicetree, despite the emmc driver apparently not being able to handle it right now (it throws i/o errors trying to read from it), but someone suggested I turn that off and now it finally works
<bnieuwen1uizen> hmm, technically hard to say what the license is since separate files and file headers, but I'll buy that
<cyrozap> bnieuwen1uizen: Yeah, it's ambiguous, but it wasn't uploaded with a license that forbit RE, at least :P
<cyrozap> s/forbit/forbid/
urjaman has joined #panfrost