alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - https://gitlab.freedesktop.org/panfrost - Logs https://freenode.irclog.whitequark.org/panfrost - Transientification is terminating. Memory reductions in progress.
stikonas has quit [Remote host closed the connection]
<alyssa> marcodiego: What SoC is it?
<alyssa> hanetzer: Presumably a RK3288 SBC (not a chromebook, just the singleboard -- say, a Tinkerboard or something) could be RYF, in theory
<alyssa> Ditto for RK3399 if DPTX were reversed etc
<marcodiego> alyssa, the pinebook pro? rk3399 I suppose
<alyssa> HdkR: Kinda none, they're mostly all questionable firmware, obvious exception of ath9k_htc :P
<alyssa> HdkR: Oh, huh, link to QCOM thing?
<alyssa> marcodiego: Er, you were askinga about renegade?
<marcodiego> alyssa, the renegade elite is a rk3399
<alyssa> marcodiego: Gotcha. So, in theory once Panfrost matures, there's still the DPTX (and wake-on-word but who cares) blobs
<alyssa> If you don't need that functionality (for a single-board computer, you probably don't), you can use it totally bloblessly
<hanetzer> yep
<alyssa> But I'm guessing it disqualifies from RYF
<marcodiego> nice
<marcodiego> is ryf that demanding
<alyssa> marcodiego: Only if the board is that blobby ;)
<marcodiego> what is this wake-on-word feature? Why does it need blobs?
<marcodiego> is a power management chip?
<alyssa> marcodiego: There's a DSP on-board so you can have the computer in sleep mode, but wake up when you say "OK Google"
<alyssa> And that DSP requires a blob firmware for that...
<alyssa> Obviously we don't care :P
<marcodiego> also, the renegade elite has no display port
<alyssa> (Which is why I suppose FSF could be lenient on it.)
<alyssa> marcodiego: USB-C ;)
<alyssa> (HDMI over DisplayPort over USB-C is a real thing... and absolutely necessary for video out on RK3399 chromebooks)
<marcodiego> is there any company financially supporting panfrost?
<alyssa> marcodiego: Define "supporting", but for many definitions, yes :)
<marcodiego> I mean, paying developers to work on it
<alyssa> Nobody is full-time
<alyssa> But there are some employees at Collabora and more recently Linaro that work on it part-time as part of $DAYJOB :)
<marcodiego> I saw something from Ezequiel Garcia I think
<marcodiego> and Tomeu Vizoso
<marcodiego> but didn't know linaro was involved
<marcodiego> where can I track it?
<alyssa> Track?
<marcodiego> follow... I mean when I take a look at gitlab to see what people have committed recently
<HdkR> alyssa: It's ath9k_htc
<alyssa> marcodiego: https://gitlab.freedesktop.org/panfrost/mesa as always
<marcodiego> alyssa, is this wake-on-word a feature of the SOC?
<Lyude> hm; i might not have a fast enough uart adapter for Bismuth :S
<alyssa> marcodiego: I'm not sure, actually
<Lyude> 1,500,000 baud is kinda impressive
<alyssa> IIRC, there's a DSP built into the SoC
<alyssa> I don't know if it does anything but wake-on-word, whether that's hw or purely firmware, if hypothetically we could do our own games with the DSP, etc
marcodiego has quit [Quit: Leaving]
<alyssa> HdkR: "Only supports SSA form" You'll probably regret that :p
<HdkR> I'll regret it later :D
<alyssa> HdkR: Since like
<alyssa> It's good to prefer SSA input but
<alyssa> You need to support registers, or otherwise you have to deal with phi nodes yourself
<alyssa> And you uh
<alyssa> Don't want that
<HdkR> Yea, not wanting to deal with phi nodes at all right now. It'll change soon enough :)
<alyssa> HdkR: ...or you could just accept NIR registers and avoid all of that :P
<HdkR> The lord and saviour, NIR registers :D
_whitelogger has joined #panfrost
_whitelogger has joined #panfrost
_whitelogger has joined #panfrost
rhyskidd has joined #panfrost
<alyssa> Question: is it closed source if the code is so illegible that nobody has any clue what's it's doing? :V
<anarsoul> :)
<alyssa> Never thought I'd have to reverse open-source code
<alyssa> Kind of an oxymoron tbh
<anarsoul> sometimes RE-ing is easier than understanding code from vendor
<alyssa> anarsoul: This isn't from a vendor.. :P
<anarsoul> oh, ok
<HdkR> alyssa: obfuscation flies in the face of open source but isn't necessarily a way to label something as not open source.
<alyssa> HdkR: Well, it took me hours to open that dang source
<alyssa> Only like 7 lines, too
<HdkR> Sounds like something that is a bit mean spirited
<alyssa> Mm
_whitelogger has joined #panfrost
NeuroScr has joined #panfrost
_whitelogger has joined #panfrost
belgin has joined #panfrost
belgin has quit [Client Quit]
belgin has joined #panfrost
belgin has quit [Client Quit]
NeuroScr has quit [Read error: Connection reset by peer]
NeuroScr has joined #panfrost
<sphalerite> alyssa: I'd say it depends on whether the unreadableness is intentional/malicious.
<sphalerite> mmind00: any chance of meeting up today? :)
<sphalerite> I'll be in the distributions devroom most of the time by default
<mmind00> sphalerite: I can try to come by there at some point :-) ... though without name tags I don't really know how to find you :-D ... for reference my backpack has a bright yellow cover
<mmind00> sphalerite: if you're still around, I'm sitting in the distri devroom right now for the grub talk ... second row, right in the middle
<sphalerite> mmind00: I'm sitting outside hoping to get in at some point
<mmind00> sphalerite: there are at least 2 chairs free in the first row
adjtm_ has quit [Ping timeout: 246 seconds]
<mmind00> sphalerite: and 2 behind me ... so there is definitly room :-D
<sphalerite> the sign says full though :/
<mmind00> sphalerite: just pretend you didn't see :-P
yann has joined #panfrost
belgin has joined #panfrost
stikonas has joined #panfrost
belgin has quit [Quit: leaving]
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
<mmind00> sphalerite: Just as a note the coreboot people in building aw are supposed to have a Chromebook with coreboot+Linuxboot on display
<sphalerite> mmind00: ooooooooooh! Need to go there then, thanks for the hint!
stikonas has quit [Remote host closed the connection]
<mmind00> sphalerite: Just talked to the Linuxboot Person up here :-)
<sphalerite> mmind00: sweet, I'll go check it out
adjtm_ has joined #panfrost
stikonas has joined #panfrost
<alyssa> sphalerite: I don't think it is... which might be worse ;P
<sphalerite> alyssa: something something never attribute something malice something incompetence
<alyssa> sphalerite: Oof
stikonas has quit [Remote host closed the connection]
TheCycoONE has quit [Quit: ZNC 1.7.1 - https://znc.in]
TheCycoONE has joined #panfrost
stikonas has joined #panfrost
rhyskidd has quit [Quit: rhyskidd]
<Lyude> cwabbott: btw, do we have an updated copy of the bifrost assembler that I wrote anywhere
<Lyude> I've got one of the original ones on my machine but that's from ye olde panfrost repo
<Lyude> and I don't see anything in spd
<Lyude> *ye olde panwrap repo
<alyssa> Lyude: I thought you guys were devving it in the panwrap repo?
<Lyude> maybe we are?
<Lyude> i haven't kept up to date with it for a while, at least until today
<alyssa> Lyude: I mean, I don't think anyone else has worked on it since you so
<Lyude> there was a bit of work from connor
<Lyude> i'm just wondering if it continued or not
<alyssa> Oh, I see
<Lyude> mhm, just wondering since I'm going through the asm to remember the order I ended up coming up with to emit a full clause
<Lyude> ah, alright
<Lyude> HdkR: by the time we're at the emitter in mesa for bifrost, have the previous stages done things like automatically insert NOPs for instructions that only use one of their stages?
<Lyude> huh, it looks like nir also is just able to handle assigning actual inputs/outputs to register ports without us having to do anything else?
empty_string has quit [Ping timeout: 252 seconds]
<Lyude> (excuse all of the questions, I'm quite new to nir)
<alyssa> Lyude: Hm? No, we have to do RA ourselves
<alyssa> NIR has a register abstraction, and we need to support inputting that
<Lyude> alright
<alyssa> But... don't rely on it for RA itself
empty_string has joined #panfrost
<Lyude> gotcha-that would have been boring if I had to anyway :p
<alyssa> The main purpose of NIR registers is resolving branches and such (without needing phi nodes)
<alyssa> There _is_ a helper for doing RA, of course, but that's not NIR-specific; you do it at the IR level and it's kind of a pain to use (I have it setup on Midgard tho and I think HdkR copied in the code for that)
<Lyude> do you remember what some of the functions for that were so I can do some grepping through the current bifrost code we've got?
<alyssa> Lyude: ra_add_node_interference, etc
<Lyude> ahhh, /* XXX: actually do RA */
<alyssa> Lyude: See midgard for how it's supposed to look
<alyssa> :P
<Lyude> alyssa: which file in the src would that be btw?
<alyssa> Lyude: It's all one file :p
<Lyude> oh! ok
urjaman has quit [Ping timeout: 250 seconds]
<Lyude> HdkR: btw, I'm pretty sure I'll need to implement the RA in the scheduler (or some placeholder for it at the very least), since we need to have an idea of what the port assignments per-instruction are going to look like before we can actually emit clauses
urjaman has joined #panfrost
<HdkR> Lyude: Yea, Scheduler will need to be RA aware in order to bundle instructions effectively
<Lyude> HdkR: that might be a good place for me to start then
<Lyude> if that's alright with yall I mean
<HdkR> That is 100% fine with me
<alyssa> Yay communication!
<HdkR> Ah, mention of lima and panfrost in the Zink talk :D
<alyssa> HdkR: Oo, link?
marcodiego has joined #panfrost
<HdkR> Will take a few weeks for the recording to come up :P
<alyssa> HdkR: Oh, hi
<alyssa> HdkR: I'm confused about the slide mentioning us?
<alyssa> "No automatic forwarding" from what to what?
<alyssa> I implemented gl_PointSize a few weeks ago, it works fine :p
<HdkR> alyssa: glPointSize the API facing function versus gl_PointSize the shader variable
<alyssa> HdkR: There's a glPointSize function?
<HdkR> aye
<HdkR> It's older than the GLSL facing construct
<HdkR> Sets the pointsize in the global state
<alyssa> Ah
<alyssa> Well, we have hw for that too :p
<HdkR> If you use both at the same time then I think the glsl side overrides the global state
<alyssa> HdkR: Reminds me, I need a dev environment for GL 2.0
<alyssa> Something like freedreno tests but.. 2.0
<alyssa> Or uh
<HdkR> Oh, you have to glEnable GL_PROGRAM_POINT_SIZE :D
<alyssa> Something like es2tri but desktop
<HdkR> Oh you just need a GL 2.0 application for banging on?
<alyssa> Mm
<HdkR> nehe tutorials may work there
<HdkR> The older ones
<alyssa> glxdemo it is
<alyssa> :p
<HdkR> That too
<marcodiego> neon helium... I remember those from the 90's... cool tutorials, not very portable though
rhyskidd has joined #panfrost
<hanetzer> so, when panfrost is daily-use ready, is there a need for a xf86-video-panfrost or no?
<alyssa> hanetzer: There's not, that's for old stuff
<hanetzer> schweet.
<marcodiego> how far is panfrost from been daily-use rady?
<alyssa> Define daily-use
<alyssa> * Not Panfrost related
<alyssa> :p
<alyssa> (Graphics and GPUs, but not 'frost, but I figure folks here will still be interested)
<alyssa> Plus *spooky noises* face reveal!
<alyssa> ("Didn't you, um, 'face reveal' in the XDC talk recording?" "Shush, you")
<ente> you looked more colorful in the XDC recording :P
<alyssa> ente: I was probably blushing then ;)
<ente> I didn't realize
<alyssa> (Kidding)
<ente> actually I wonder if blushing would have been visible on the recording
<marcodiego> by daily-use I mean... being able to run X or wayland an entire day without crashes, running a browser Opengl 2 or 3... mostly that
<marcodiego> your latex notation on the blog post is not working on my browser (chromium 65.0.3325.181 (Official Build) Built on Raspbian , running on Raspbian 9.6 (32-bit))
<ente> works for me
<ente> (on firefox)
<urjaman> hmm yeah, seems it doesnt work for any of my chromes (71 on the x86 desktop, 73 on the chromebook) but does for firefox
<marcodiego> works for me in firefox too
<alyssa> marcodiego: Hmm, that sounds like a challenge :)
<marcodiego> you mean the daily-use or making to blog post work on chromium?
<alyssa> Daily use
<alyssa> Taking a look at Chromium now, wonder what's different
<alyssa> Oh right, MathML isn't standard
<alyssa> brb switching to mathjax
<marcodiego> just one more thing... will panfrost (the kernel or mesa part) be merged only when "complete" or as soon as it gets "daily-use ready"?
<alyssa> Dunno about kernel
<alyssa> Mesa is being merged in increments
<alyssa> (First bits of mesa code will hit upstream this week, it sounds like)
<marcodiego> once merged, will all work happen upstream?
<alyssa> More or less
<Lyude> I don't think anyone has intentions of anything that gets done being left downstream
<Lyude> forever at least
<marcodiego> good to know that
<Lyude> i'd hope that when we've got more merged we can go straight to upstreaming WIP hw support though
<alyssa> Math rendering fixed
<urjaman> confirmed
<urjaman> that it works on chrome now that is :)
<marcodiego> works here too
<alyssa> Cool