ChanServ changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - Logs https://oftc.irclog.whitequark.org/panfrost - <macc24> i have been here before it was popular
rasterman has quit [Quit: Gettin' stinky!]
nlhowell is now known as Guest144
Guest144 has quit [Read error: Connection reset by peer]
jernej has quit [Read error: Connection reset by peer]
jernej_ has quit [Remote host closed the connection]
jernej has joined #panfrost
nlhowell is now known as Guest162
nlhowell has joined #panfrost
Guest162 has quit [Ping timeout: 480 seconds]
nlhowell is now known as Guest166
Guest166 has quit [Read error: Connection reset by peer]
nlhowell has joined #panfrost
camus1 has joined #panfrost
camus has quit [Ping timeout: 480 seconds]
vstehle has joined #panfrost
JulianGro has joined #panfrost
chewitt has quit [Ping timeout: 480 seconds]
tolszak has joined #panfrost
chewitt has joined #panfrost
nlhowell is now known as Guest174
nlhowell has joined #panfrost
Guest174 has quit [Ping timeout: 480 seconds]
chewitt has quit [Quit: Zzz..]
Major_Biscuit has joined #panfrost
tolszak has quit [Ping timeout: 480 seconds]
nlhowell is now known as Guest182
Guest182 has quit [Read error: Connection reset by peer]
nlhowell has joined #panfrost
atler has joined #panfrost
atler has quit [singleton.oftc.net charon.oftc.net]
ndufresne has quit [singleton.oftc.net charon.oftc.net]
ente` has quit [singleton.oftc.net charon.oftc.net]
jenneron has quit [singleton.oftc.net charon.oftc.net]
CalebFontenotHeHim[m] has quit [singleton.oftc.net charon.oftc.net]
Dylanger has quit [singleton.oftc.net charon.oftc.net]
strongtz[m] has quit [singleton.oftc.net charon.oftc.net]
psydroid has quit [singleton.oftc.net charon.oftc.net]
fahien has quit [singleton.oftc.net charon.oftc.net]
megi1 has quit [singleton.oftc.net charon.oftc.net]
Major_Biscuit has quit [singleton.oftc.net charon.oftc.net]
vstehle has quit [singleton.oftc.net charon.oftc.net]
JulianGro has quit [singleton.oftc.net charon.oftc.net]
jernej has quit [singleton.oftc.net charon.oftc.net]
erlehmann has quit [singleton.oftc.net charon.oftc.net]
DPA has quit [singleton.oftc.net charon.oftc.net]
Guest135 has quit [singleton.oftc.net charon.oftc.net]
indy has quit [singleton.oftc.net charon.oftc.net]
suihkulokki has quit [singleton.oftc.net charon.oftc.net]
wolfshappen has quit [singleton.oftc.net charon.oftc.net]
alarumbe has quit [singleton.oftc.net charon.oftc.net]
mav has quit [singleton.oftc.net charon.oftc.net]
robertfo1 has quit [singleton.oftc.net charon.oftc.net]
rtp has quit [singleton.oftc.net charon.oftc.net]
CounterPillow has quit [singleton.oftc.net charon.oftc.net]
Stary has quit [singleton.oftc.net charon.oftc.net]
sigmaris has quit [singleton.oftc.net charon.oftc.net]
mriesch has quit [singleton.oftc.net charon.oftc.net]
stebler[m] has quit [singleton.oftc.net charon.oftc.net]
FLHerne has quit [singleton.oftc.net charon.oftc.net]
samuelig has quit [singleton.oftc.net charon.oftc.net]
macc24_ has quit [singleton.oftc.net charon.oftc.net]
robmur01 has quit [singleton.oftc.net charon.oftc.net]
pjakobsson has quit [singleton.oftc.net charon.oftc.net]
stepri01 has quit [singleton.oftc.net charon.oftc.net]
tomeu has quit [singleton.oftc.net charon.oftc.net]
mmind00 has quit [singleton.oftc.net charon.oftc.net]
enick_689 has quit [singleton.oftc.net charon.oftc.net]
wilkom has quit [singleton.oftc.net charon.oftc.net]
pendingchaos has quit [singleton.oftc.net charon.oftc.net]
karolherbst has quit [singleton.oftc.net charon.oftc.net]
rellla has quit [singleton.oftc.net charon.oftc.net]
minicom has quit [singleton.oftc.net charon.oftc.net]
AreaScout_ has quit [singleton.oftc.net charon.oftc.net]
urja has quit [singleton.oftc.net charon.oftc.net]
toggleton[m] has quit [singleton.oftc.net charon.oftc.net]
CME_ has quit [singleton.oftc.net charon.oftc.net]
remexre has quit [singleton.oftc.net charon.oftc.net]
wicastC has quit [singleton.oftc.net charon.oftc.net]
spawacz has quit [singleton.oftc.net charon.oftc.net]
ezequielg has quit [singleton.oftc.net charon.oftc.net]
jschwart has quit [singleton.oftc.net charon.oftc.net]
SolidHal8 has quit [singleton.oftc.net charon.oftc.net]
tlwoerner has quit [singleton.oftc.net charon.oftc.net]
kinkinkijkin has quit [singleton.oftc.net charon.oftc.net]
jolan has quit [singleton.oftc.net charon.oftc.net]
Bennett has quit [singleton.oftc.net charon.oftc.net]
tjcorley has quit [singleton.oftc.net charon.oftc.net]
cphealy has quit [singleton.oftc.net charon.oftc.net]
narmstrong has quit [singleton.oftc.net charon.oftc.net]
soreau has quit [singleton.oftc.net charon.oftc.net]
robclark has quit [singleton.oftc.net charon.oftc.net]
cyrozap has quit [singleton.oftc.net charon.oftc.net]
tchebb has quit [singleton.oftc.net charon.oftc.net]
join_subline has quit [singleton.oftc.net charon.oftc.net]
cwabbott has quit [singleton.oftc.net charon.oftc.net]
philpax_ has quit [singleton.oftc.net charon.oftc.net]
jstultz has quit [singleton.oftc.net charon.oftc.net]
jekstrand has quit [singleton.oftc.net charon.oftc.net]
nlhowell has quit [singleton.oftc.net charon.oftc.net]
rcf has quit [singleton.oftc.net charon.oftc.net]
skl131313 has quit [singleton.oftc.net charon.oftc.net]
jkl has quit [singleton.oftc.net charon.oftc.net]
MoeIcenowy has quit [singleton.oftc.net charon.oftc.net]
robher has quit [singleton.oftc.net charon.oftc.net]
ckeepax has quit [singleton.oftc.net charon.oftc.net]
tortoise has quit [singleton.oftc.net charon.oftc.net]
dschuermann has quit [singleton.oftc.net charon.oftc.net]
alyssa has quit [singleton.oftc.net charon.oftc.net]
robink has quit [singleton.oftc.net charon.oftc.net]
kenzie35 has quit [singleton.oftc.net charon.oftc.net]
anarsoul has quit [singleton.oftc.net charon.oftc.net]
Lyude has quit [singleton.oftc.net charon.oftc.net]
anholt has quit [singleton.oftc.net charon.oftc.net]
krh has quit [singleton.oftc.net charon.oftc.net]
rando25892 has quit [singleton.oftc.net charon.oftc.net]
trixie has quit [singleton.oftc.net charon.oftc.net]
steev has quit [singleton.oftc.net charon.oftc.net]
daniels has quit [singleton.oftc.net charon.oftc.net]
austriancoder has quit [singleton.oftc.net charon.oftc.net]
orkid has quit [singleton.oftc.net charon.oftc.net]
nlhowell has joined #panfrost
JulianGro has joined #panfrost
Guest135 has joined #panfrost
DPA has joined #panfrost
ndufresne has joined #panfrost
suihkulokki has joined #panfrost
indy has joined #panfrost
wolfshappen has joined #panfrost
Bennett has joined #panfrost
ente` has joined #panfrost
jenneron has joined #panfrost
alarumbe has joined #panfrost
CalebFontenotHeHim[m] has joined #panfrost
mav has joined #panfrost
robertfo1 has joined #panfrost
rcf has joined #panfrost
atler has joined #panfrost
Major_Biscuit has joined #panfrost
megi1 has joined #panfrost
strongtz[m] has joined #panfrost
Dylanger has joined #panfrost
psydroid has joined #panfrost
tjcorley has joined #panfrost
fahien has joined #panfrost
skl131313 has joined #panfrost
rtp has joined #panfrost
robmur01 has joined #panfrost
mriesch has joined #panfrost
CounterPillow has joined #panfrost
Stary has joined #panfrost
pjakobsson has joined #panfrost
narmstrong has joined #panfrost
stepri01 has joined #panfrost
sigmaris has joined #panfrost
mmind00 has joined #panfrost
tomeu has joined #panfrost
soreau has joined #panfrost
stebler[m] has joined #panfrost
jekstrand has joined #panfrost
FLHerne has joined #panfrost
enick_689 has joined #panfrost
wilkom has joined #panfrost
minicom has joined #panfrost
samuelig has joined #panfrost
Lyude has joined #panfrost
pendingchaos has joined #panfrost
cphealy has joined #panfrost
karolherbst has joined #panfrost
rellla has joined #panfrost
AreaScout_ has joined #panfrost
urja has joined #panfrost
anholt has joined #panfrost
toggleton[m] has joined #panfrost
jschwart has joined #panfrost
robclark has joined #panfrost
remexre has joined #panfrost
wicastC has joined #panfrost
ezequielg has joined #panfrost
spawacz has joined #panfrost
SolidHal8 has joined #panfrost
tlwoerner has joined #panfrost
cyrozap has joined #panfrost
robink has joined #panfrost
CME_ has joined #panfrost
ckeepax has joined #panfrost
alyssa has joined #panfrost
kenzie35 has joined #panfrost
daniels has joined #panfrost
orkid has joined #panfrost
trixie has joined #panfrost
steev has joined #panfrost
austriancoder has joined #panfrost
krh has joined #panfrost
philpax_ has joined #panfrost
dschuermann has joined #panfrost
rando25892 has joined #panfrost
jstultz has joined #panfrost
anarsoul has joined #panfrost
cwabbott has joined #panfrost
tortoise has joined #panfrost
jolan has joined #panfrost
robher has joined #panfrost
join_subline has joined #panfrost
MoeIcenowy has joined #panfrost
jkl has joined #panfrost
kinkinkijkin has joined #panfrost
tchebb has joined #panfrost
erlehmann has joined #panfrost
macc24_ has joined #panfrost
jernej has joined #panfrost
vstehle has joined #panfrost
camus1 has quit [Remote host closed the connection]
camus has joined #panfrost
tolszak has joined #panfrost
rasterman has joined #panfrost
atler has quit [Quit: atler]
macc24_ has quit []
macc24 has joined #panfrost
macc24 has quit []
Danct12 has quit [Quit: Quitting]
Danct12 has joined #panfrost
Daanct12 has joined #panfrost
Danct12 has quit [Ping timeout: 480 seconds]
Daanct12 is now known as Danct12
<alyssa>
robher: To the kernel, Bifrost and pre-CSF Valhall are identical; the drm/panfrost changes to support G57/G77/G78 are just errata handling.
<alyssa>
Is it acceptable to reuse the mali-bifrost compatible for pre-CSF Valhall?
<alyssa>
I see we've split mali-midgard and mali-bifrost. Obviously too late to undo that but I'm not sure why those were split
<alyssa>
arm,mali-bifrost.yaml matches on
<alyssa>
- const: arm,mali-bifrost # Mali Bifrost GPU model/revision is fully discoverable
<alyssa>
Valhall GPU model/revision is also fully discoverable by the same mechanism, so we don't need to invent compatibles for every impl under the sun
<alyssa>
The three options are
<robmur01>
yup, if you can read the same registers and tell exactly what you have then it is compatible
<alyssa>
1. Use arm,mali-bifrost as a valhall compatible too, just update the dt-binding documentation for that
<alyssa>
2. Add arm,mali-valhall to arm,mali-bifrost.yaml
<alyssa>
Apparently MediaTek sent a patch a year ago proposing #3, but it seems wholly unnecessary to me
<alyssa>
My preference is #2 but I'm not sure if that would generate too much confusion
<alyssa>
Er
<alyssa>
preference is #1, sorry
<robmur01>
#1 is justifiable in its own right, but I think it's also what downstream has done too ;)
<alyssa>
*nod*
<alyssa>
After spending enough time with Malis you learn the marketing names are all nonsense and all that matters is the arch major version..
<alyssa>
In Panfrost and Arm stuff alike, "Midgard" is short for "Midgard Bifrost or Valhall", and "Bifrost" is short for "Bifrost or Valhall", except where otherwise indicated......
camus1 has joined #panfrost
<robher>
The real question is whether there are differences which matter before you can discover what you have. IOW, you have to turn on clocks, power-domains, de-assert resets, etc.
<alyssa>
Right.
<alyssa>
I'm not in a position to answer that in general,
tolszak has quit [Ping timeout: 480 seconds]
<alyssa>
the Valhall device I'm mainlining is a weird MediaTek which needs extra special handling anyway so it's a bit theoretical
<stepri01>
pre-CSF there aren't any differences (well except perhaps for specific SOCs)
<robmur01>
we still want SoC-specific compatibles, though, which should be able to cover anything in that respect as necessary, right?
<robher>
It gets messy too because vendors change those (even though the underlying h/w doesn't).
<alyssa>
(We already have a `mediatek,mt8183-mali` compatible, we need `mediatek,mt8192-mali` too because MediaTek changed/broke things)
<alyssa>
(*Different* things than mt8183, even)
<robmur01>
essentially, we still have to describe the integration, but the integration doesn't really depend on the exact model
<alyssa>
stepri01: good, that's what I thought
<robher>
Splitting schema docs or not really comes down to a judgement call of how many if/then schemas a single doc has.
camus has quit [Ping timeout: 480 seconds]
<robher>
I would lean towards option 2. More compatible strings is not a big deal.
mmind00 has quit [Quit: No Ping reply in 180 seconds.]
<alyssa>
OK
<alyssa>
so compatible = "mediatek,mt8192-mali", "arm,mali-valhall"?
<alyssa>
or compatible = "mediatek,mt8192-mali", "arm,mali-valhall", "arm,mali-bifrost"? because b/w compat?
<alyssa>
or just plain compatible = "mediatek,mt8192-mali" because the integration is screwed up so regular arm,mali-* won't probe?
mmind00 has joined #panfrost
<robher>
If someone wants to confirm in the trm's that the clocks, interrupts, register space are all the same, then I'd be okay with #1.
<alyssa>
robmur01: Assigning you as the dedicated TRM reader ;-P
<robher>
And what's post CSF going to be? arm,mali-valhall-csf?
<alyssa>
I don't have any information about CSF, and the kbase code is completely opaque.
<stepri01>
currently it's a different kbase for CSF (compile time switch)
<alyssa>
Yeah that's not going to fly upstream ..
<stepri01>
so I suspect it's going to be a case of needing a different compatible but I'm not 100% sure whether it's probe-able or not
<alyssa>
"drivers/gpu/arm/midgard/csf"
<alyssa>
so good
<stepri01>
yeah kbase is still stuck with the "midgard" name ;)
<robher>
No point in "arm,mali-bifrost" fallback unless you want it to work on unmodified s/w that only understands "arm,mali-bifrost". If you have to add errata handling anyways, you're already changing the s/w.
<alyssa>
fair enough
* robher
needs to write a manifesto on how compatible works...
<alyssa>
the L2 MMU config register got shuffled I think
<alyssa>
technically that's an implementation detail and yet ..
<alyssa>
so that in itself justifies dropping the bifrost compatible
<alyssa>
though I think sticking to the same yaml is reasonable
<alyssa>
so that'd be option #2 I guess
<robmur01>
Looks like 1st-gen Valhall still has the same 1 clock, 1 reset and 4 IRQs as Bifrost; after that things start to vary
<alyssa>
"1st-gen" being G57/G77 but not G78?
* robmur01
has to look up what G78 is...
<alyssa>
Borr, I think
<alyssa>
Norr is Bifrost and Borr is Valhall, definitely not confusing
<robmur01>
yup, G78 starts growing extra clocks
<robher>
the extra clocks (from vendors) tend to be bus clocks. So if the bus interfaces change, we can assume those clocks will too.
<robmur01>
There shouldn't be any significant change in interfaces
<stepri01>
AFAIK G78 is "normal", it's G78-AE that has grown extra stuff for virtualization
<stepri01>
but I believe the expectation is that if you're not automotive then all the clock domains would be driven the same
<robmur01>
ah, OK
<stepri01>
the point of the "slices" is you can split the GPU up if you're crazy enough to try to run safety critical things on it
<stepri01>
(or at least that was the theory back when I wrote the patent application - I've no idea what they actual did since I left GPU...)
<robmur01>
heh, G78 has 3 clocks, G78AE has 18
<robmur01>
yeah let's not support automotive
<stepri01>
:)
<robher>
+1000
<stepri01>
the original theory was that the automotive SoC could also be sold into other markets - so it is possible that we could see an interesting platform with it in though
JulianGro has quit [Remote host closed the connection]
<alyssa>
stepri01: Admittedly I do wish Mali supported sane preemption
<stepri01>
I don't think you're going to get 'sane' - that's not the Mali way ;)
<alyssa>
Not for safety-critical stuff, just for not having random freezes
<alyssa>
No, I suppose not :v
<stepri01>
the "soft-stop" command was an attempt, and there was lots of talk about a "suspend" command - but in the end it's hard to do in hardware and really slow if you actually need to save/restore the GPU state
<alyssa>
yeah, I understand why it didn't happen
<alyssa>
One of those hidden tradeoffs of tilers they don't tell you about in school.
<stepri01>
Panfrost could do better though - kbase will soft-stop a job if it hangs around for a while to give something else a chance
<stepri01>
panfrost only soft-stops when recovering before a reset
<alyssa>
that does seem better than the timeout handler which is both too high for smooth desktop with misbehaved apps and too low for well-behaved massive compute kernels
<rasterman>
yeah - this is one of those quality things still floating about
vstehle has quit [Quit: WeeChat 3.3]
vstehle has joined #panfrost
macc24 has joined #panfrost
<daniels>
stepri01: I don't think we have to worry too hard about upstreaming G78AE ...
<stepri01>
No I agree. And I'm hoping that no interesting platforms come about. But many years ago when I helped write up the patent that was the primary motivation (that the same SoC could be sold into different markets)
atler has joined #panfrost
<stepri01>
I'm not at all familiar with how the commercial landscape has changed - so it's quite possible nobody is interested in that anymore
<robmur01>
free software die-hards will want to run upstream software on their car, but frankly I'd rather go the other way back to points ignition and carbs...
<alyssa>
choo choo
<rasterman>
robmur01: i suspect that'll be legislated into a crime soon enough :)
<macc24>
robmur01: smh just bike
<robmur01>
macc24: even my 33-year-old carb'ed bike has CDI ignition :(
<robmur01>
it's a pretty dang primitive computer but still a computer, and computers must be defeated!
<macc24>
robmur01: i was talking about the manually operated bikes, you know, bicycles
<robmur01>
yup, I have at least 6 of those too, but they're hardly practical for the tasks for which I own motor vehicles
<macc24>
:/
<macc24>
i usually go to school by train, 10 minute walk onto train station, 10 minute train ride and 20 minutes of walking
<alyssa>
Pro's of living in Toronto: don't need a car
<alyssa>
Con's of living in Toronto: can't afford to park a car
<rasterman>
but that pretty much traps you in toronto
<robmur01>
macc24: and in a later phase of your life, you may well have need to move 100kg+ of stuff 150km or more... GLHF :)
<rasterman>
getting out eiher means a plane or a car (sure trains exist but are going to not be very convenient/optimal and only go to a few places)
<rasterman>
europe is much better off with that
<alyssa>
that's what the go train is for ... and just magically have a car at the destination, um
<rasterman>
but north maeirca, australia etc. prety uch need a car unless you want to never leave a city center :)
tjcorley has quit [Ping timeout: 480 seconds]
<macc24>
robmur01: buy three train tickets and sit the stuff next to u :v
<rasterman>
well sure - you need to rent a car then :)
<rasterman>
still need a car :)
<alyssa>
(IMO - Trains in Ontario/Quebec are better than the States but not yet up to European standards)
<macc24>
alyssa: "european standards" uhhh, i usually ride a en57 straight from 60's
<rasterman>
macc24: ebtter than crawling on has and knees :)
* robmur01
LOLs/cries at the thought of trains being any more viable than unicorns in rural England
tjcorley has joined #panfrost
<rasterman>
robmur01: well .. you have a point. i want to get from here to i think it was edinburgh - i had to get a train down to london first
<rasterman>
then back up...
<rasterman>
or like gettign to cotswolds - train to london - futz about, then back out
<rasterman>
and then only to some towns there.
<macc24>
robmur01: and when it gets less than 0 celsius the train switch things freeze :v
<rasterman>
but still better than e.g. aus
<rasterman>
:)
<rasterman>
there at leas is some way to get there by train at all.
<rasterman>
my parents live about a 2hr drive west of sydney
<daniels>
protip: living in London short-circuits the 'get to London' bit
* alyssa
taps forehead
<rasterman>
train takes more like 3hrs...
<alyssa>
macc24: Okay but have you considered Ottawa's light rail system which has failed due to cold weather
<rasterman>
and THEN my parents have to drive for 45 mins to pick me up at bell station
<alyssa>
("Isn't Ottawa the coldest national capital in the world?")
<rasterman>
also bekll station - you have to find the conductor and ask them to specially stop for you as the train wont stop unless you ask them to
<macc24>
alyssa: at least it's better than buses...
<alyssa>
macc24: ...Is it?
<macc24>
alyssa: traffic jams affect buses too
<rasterman>
and thats the MAIN train line heading west from sydney... good luck if you have to go a bit norht or south of mt wilson where my parents are :)
<macc24>
and i'd rather take an electric vehicle than burn more gasoline...
<rasterman>
daniels: hahahaha
<rasterman>
daniels: i know... there is some method to my madness of looking to buy a home in london :)
<rasterman>
but hell its still better than trains in oz.
<rasterman>
we did fun things like dismantled our tram systems and are now busy re-building light rail in the city and surroudning burbs ... smart move fellas
<rasterman>
in the end though it boils down to there being many places esp in "the new world" (north maerica, oz etc.) that you basiclaly have to have a car to have a sane life. cant depend on public transport.
rasterman has quit [Quit: Gettin' stinky!]
camus has joined #panfrost
camus1 has quit [Read error: Connection reset by peer]
Major_Biscuit has quit [Ping timeout: 480 seconds]