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.
<HdkR> robclark: Don't worry about the `#ifdef MIDGARD<XXX>` bits, those will go away as the bifrost stuff moves in :)
marcodiego has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
<alyssa> HdkR: Plus I think a lot of the stuff marked as T6XX vs T8XX is actually 32-bit vs 64-bit
<alyssa> Though I guess without more devices to look at, we can't know that quite yet
<HdkR> :D
<alyssa> (64-bit T6xx/T7xx and/or 32-bit T8xx would be interesting to play with)
marcodiego has quit [Quit: Leaving]
<Lyude> alyssa: :)
<Lyude> wait, hm
<Lyude> i did not notice the 32/64 bit parts of that sentence when I first read it
<tomeu> alyssa: yep, understood
<alyssa> :+1: :+1:
<tomeu> robher has already added support for gpu addresses in the new driver :)
<alyssa> Huzzah!
<tomeu> congrats on the reviews, btw :)
<tomeu> it's going to feel great once it's in mainline
<HdkR> Grape soda, root beer, and other flavoured carbonated waters all around once merged
<alyssa> :)
<HdkR> I personally enjoy La Croix
<Lyude> narmstrong: bifrost is fun :)
<HdkR> Lyude: Whatja working on? :)
<Lyude> HdkR: nothing ATM :/, been relaxing the last few weekends but I'm hoping to start playing with setting up the systems I've got in the mail over the weekend
<HdkR> Ah. I'm poking again today to see if there is an update on more boards as well
<Lyude> I'd really like to play around with getting code emitting working btw, if you'd like me to help out with that!
<HdkR> I'm currently working on clause packing :P
<HdkR> worked on it for a smidge last weekend while on a plane
<Lyude> Ah, anything I could help out with?
<HdkR> It's so early that I've basically got a NOP clause emitting for testing
<HdkR> Super hacky. I don't have a smart way of scheduling instructions in to clauses atm
<alyssa> :D
<HdkR> Lyude: Theoretically we could split the task in to scheduling and emitting
<HdkR> Which would allow some parallel work
<HdkR> and it would just have us stepping on each other's toes when adding opcodes
<alyssa> I split Midgard that way
<HdkR> Yea, we could be more explicit and keep it in different files even :P
<Lyude> I'm fine with that-I'll be learning nir stuff in the process as well I'm guessing? :) (has not gotten a chance to look at any of the bifrost code yet)
<HdkR> Scheduling side would be pretty tightly coupled with nir
<HdkR> Emitting side not so much
<Lyude> I know bifrost's isa pretty alright from writing the assembler, so I could probably do the emitting pretty easily
<HdkR> Alright. I can split scheduling and emitting more cleanly and get a working branch set up and we can work in parallel if you want to do the emitting side
<HdkR> The disassembler is super useful for verifying correctness :P
<Lyude> I'd imagine :p, that's how I wrote the assembler
<alyssa> :+1: for midgard
<HdkR> Lyude: Alright. I'll split the two...tonight, tomorrow? maybe this weekend. Will see what happens :P
pH5 has quit [Quit: 🥦]
<sphalerit> Morning! Anyone coming to FOSDEM? :)
<HdkR> I'm not
<Lyude> HdkR: hehe, alright
raster has joined #panfrost
* mmind00 waves to sphalerit
* mmind00 is sitting in Frankfurt waiting for his flight to Brussels right now
<sphalerite> mmind00: cool! I'm in the train to Frankfurt, where I'll take another train to Brussels :p
* mmind00 is curious where you're coming from ... but it's of course none of my business :-D
<sphalerite> Munich
<mmind00> ah, so you get some resemblance if high-speed train service ... Dresden -> Frankfurt takes at least 5 hours ... through the eastern dessert (in terms of rail infrastructure) ;-)
<sphalerite> mmmm dessert!
<mmind00> sphalerite: :-P ... desert
<mmind00> But I feel similar ... a tiny voice in my head chants the whole day already: waffles waffles waffles ....
* daniels is in a very very crowded departure lounge at St Pancras awaiting a delayed Eurostar
<rtp> alyssa: about t6xx vs t8xx. I do hope that it's a 64bit vs 32bit issue too but I wouldn't bet on that.
<hopetech> daniels: Please don't scare me. My train is in few hours
<daniels> hopetech: only half an hour delayed it seems
<robertfoss_> robher: are you using your HiKey960 for panfrost development?
raster has quit [Remote host closed the connection]
<robher> robertfoss_: No, I don't have one plus it has bifrost (G71?). So I have a rock960.
<alyssa> rtp: Here's hoping, anyway
<alyssa> I know there are some differences that are architectural, but others I suspect are just bit-ness, and it's all mixed up I think
<rtp> alyssa: and some issues may even disappear in the long run when switching to our own kernel module, like the .job_descriptor_size thing
<alyssa> rtp: This is true, yeah :)
<alyssa> Well
<alyssa> Not that one
<alyssa> That's in the hw :p
<alyssa> But others, yes
<rtp> oh :(
* rtp found one place were it was always set to 1, even on 32bit
<alyssa> SET_VALUE jobs, yeah
<alyssa> I've only ever seen them 64-bit so
<alyssa> I think you _can_ set it to 1 on 32-bit systems (on any job), it's just.. not particularly helpful, since your pointers fit in 32-bit anyway
<rtp> tbh, I dont care that much about the value, as long as we manage to set the proper value for everyone eg, panfrost working on t6xx and t8xx :)
<alyssa> ;)
<alyssa> Aside: what are SET_VALUE jobs anyway? :P
<daniels> based purely on speculation, are they maybe for hardware counters that you'd use for e.g. synchronisation?
<alyssa> Mayhaps
<cwabbott> alyssa: btw, what happens if you have t6xx and a 64-bit userspace, or vice versa, t8xx and a 32-bit userspace? Does it switch to the other format?
<rtp> alyssa: btw, is there any chance that panwrap works for the mali blob too ? I'm wondering if I could compare the logs and maybe find the issue(s), assuming I can get my test app working on my "old" rootfs
BenG83 has joined #panfrost
cwabbott has quit [Quit: cwabbott]
cwabbott has joined #panfrost
cwabbott has quit [Client Quit]
cwabbott has joined #panfrost
BenG83 has quit [Quit: Leaving]
<Lyude> robertfoss_: I will probably be
<Lyude> robertfoss_: cwabbott and I have been using a hikey960 I leave hooked up to my server w/ serial console + adb
<Lyude> although I'm going to try to put it and all my other arm systems on mainline linux this weekend
stikonas has joined #panfrost
jolan has quit [Quit: leaving]
jolan has joined #panfrost
<alyssa> cwabbott: That's what I want to find out :p
<alyssa> rtp: Yeah, most of the cmdstream stuff was figured out with panwrap on the blob
<alyssa> Though whether any particular version of panwrap works on any particular version of the blob is a bit of a crapshoot
<alyssa> (Biggest thing that needs to match is kernel version.. new kernels/new blobs are needed for new [since December] panwraps built with mesa... old panwrap [in the panloader repo] is, well, really old and not terribly useful anymore..)
<anarsoul> alyssa: what's the status of kernel driver? is anyone working on it?
<alyssa> anarsoul: Yes, robher and tomeu
<anarsoul> what about 1st Q? :)
<alyssa> Very early stages
* hanetzer mumbles something about getting off his ass
<alyssa> I hear sitting is the new smoking~
<hanetzer> prollem is my entire job and school is on my arse :P
<alyssa> Relatable, yes..