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
warpme_ has quit [Quit: Connection closed for inactivity]
<alyssa> icecream95: also I started looking at 16-bit int support for bf, my vec2 branch might have patches you want for cl
icecream95 has joined #panfrost
<icecream95> alyssa: Dropping the +1 in bi_get_index regressed a few tests where ssa_0 is an intrinsic load
<macc24> icecream95: there was something about you in backlog
<macc24> when you weren't online
raster has quit [Quit: Gettin' stinky!]
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
kaspter has joined #panfrost
vstehle has quit [Ping timeout: 246 seconds]
kaspter has quit [Remote host closed the connection]
kaspter has joined #panfrost
kaspter has quit [Ping timeout: 240 seconds]
kaspter has joined #panfrost
davidlt has joined #panfrost
chrisf has quit [Remote host closed the connection]
chrisf has joined #panfrost
vstehle has joined #panfrost
q4a has joined #panfrost
<kinkinkijkin> i have to wonder if anyone's played supertuxkart on a switch yet
camus has joined #panfrost
kaspter has quit [Ping timeout: 240 seconds]
camus is now known as kaspter
kaspter has quit [Ping timeout: 256 seconds]
kaspter has joined #panfrost
phh has quit [Ping timeout: 265 seconds]
phh has joined #panfrost
tomeu has quit [Excess Flood]
tomeu has joined #panfrost
<narmstrong> alyssa: is AFBC on bifrost only for internal buffers or also enabled for the display framebuffer ?
<icecream95> On mt8183 the CCI frequency makes a massive difference for tinymembench results - forcing it to maximum frequency makes memcpy go from 2GB/s to 5GB/s for big cores. Now to see if it makes compiling faster...
icecream95 has quit [Ping timeout: 264 seconds]
kaspter has quit [Ping timeout: 246 seconds]
kaspter has joined #panfrost
karolherbst has quit [Quit: duh 🐧]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
camus has joined #panfrost
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 246 seconds]
camus1 is now known as kaspter
camus has quit [Ping timeout: 246 seconds]
Ke has quit [Quit: Bridge terminating on SIGTERM]
wiizzard has quit [Quit: Bridge terminating on SIGTERM]
exit70[m] has quit [Quit: Bridge terminating on SIGTERM]
l-as has quit [Quit: Bridge terminating on SIGTERM]
clementp[m] has quit [Quit: Bridge terminating on SIGTERM]
kaspter has quit [Ping timeout: 256 seconds]
raster has joined #panfrost
clementp[m] has joined #panfrost
exit70[m] has joined #panfrost
wiizzard has joined #panfrost
l-as has joined #panfrost
Ke has joined #panfrost
karolherbst has joined #panfrost
kaspter has joined #panfrost
kaspter has quit [Remote host closed the connection]
kaspter has joined #panfrost
warpme_ has joined #panfrost
stepri01 has quit [Ping timeout: 268 seconds]
raster has quit [Quit: Gettin' stinky!]
raster has joined #panfrost
camus has joined #panfrost
kaspter has quit [Ping timeout: 260 seconds]
camus is now known as kaspter
<macc24> icecream95: CCI frequency is inter-locked(i think that's the word) with cores frequency
<macc24> so if any core wants 1.5ghz(opp 7), cci has to have 700mhz(opp 7)
<chewitt> after bumping mesa to HEAD to include the AFBC thing, I got this on XU4 https://pastebin.com/raw/a3vfKG3w
<chewitt> probably not enough to be helpful, but more than I normally see
<bbrezillon> chewitt: Bifrost or Midgard?
<chewitt> Midgard .. T628
kaspter has quit [Ping timeout: 272 seconds]
kaspter has joined #panfrost
<bbrezillon> chewitt: hm, this one has the MIDGARD_NO_AFBC flag set, right?
<chewitt> as a mesa env variable?
<bbrezillon> no, I mean in panfrost_get_quirks()
<bbrezillon> I don't see the 0x628 case, but it's probably mapping to a different ID
<chewitt> it's not in the supported list, I've patched the ID in
<bbrezillon> ok, you should probably add this flag then
<chewitt> and it was faulting before, this is not a regression :)
<bbrezillon> oh, ok
stepri01 has joined #panfrost
<chewitt> I'm just optimistic that one day more completeness and fixes will .. magically make it work
<chewitt> else I need to maintain mali stuff in the distro, which these days feels a bit unclean
<HdkR> Mali blob is unclean yes. :)
<chewitt> MIDGARD_NO_AFBC is already set for T620
<HdkR> Time to enable it for t628 and hope for the best?
<macc24> chewitt: haha in my distro i tell users to use software rendering(or compile mesa-git)
<chewitt> I'd choose blobs over llvmpipe
raster has quit [Quit: Gettin' stinky!]
zkrx has quit [Ping timeout: 240 seconds]
chewitt_ has joined #panfrost
raster has joined #panfrost
chewitt has quit [Ping timeout: 258 seconds]
<macc24> my mali boards will never have mali blobs if panfrost can render anything on them
<chewitt_> that's the goal
chewitt_ is now known as chewitt
<chewitt> but .. T628 :)
<macc24> well
<macc24> g72 support is not the best either
<macc24> does linux disable gpu cache memory on t628 too?
karolherbst has quit [Ping timeout: 246 seconds]
<chewitt> how to see this?
<macc24> you don't see anything at all when running kmscube if cache is disabled and there's suspicious line in dmesg "gpu_vsram: disabling"
<chewitt> kmscube runs fine
<HdkR> gpu_vsram?
<HdkR> oh, seems to be some mediatek thing. voltage controller for sram?
<chewitt> doesn't show
<HdkR> Makes sense it wouldn't show on exynos :P
kaspter has quit [Quit: kaspter]
tgall_foo has quit [Read error: Connection reset by peer]
<macc24> HdkR: yep
<macc24> had to slap a little regulator-always-on; onto it
<macc24> in mt6358.dtsi
tgall_foo has joined #panfrost
karolherbst has joined #panfrost
kaspter has joined #panfrost
agrisis has joined #panfrost
<agrisis> is gles3 supported on the g31?
<HdkR> No, but you could try a mesa driver override and things might work
zkrx has joined #panfrost
<agrisis> HdkR: thank you
zkrx has quit [Ping timeout: 272 seconds]
zkrx has joined #panfrost
<HdkR> Panfrost should hit GLES3 on Bifrost sometime this year
tomboy64 has quit [Ping timeout: 240 seconds]
<q4a> HdkR and what about vulkan on Midgard?
agrisis has quit [Ping timeout: 272 seconds]
<HdkR> No one working on it as far as I'm aware
<q4a> That's sad
<HdkR> There's a bunch of GLES3.1 things that need to be solved before Vulkan is even interesting
<HdkR> Since minspec for Vulkan is effectively GLES3.1
<alyssa> q4a: Patches welcome.
<HdkR> alyssa: plez to implement Vulkan so I can use vkd3d
<HdkR> :>
<macc24> HdkR: valhall working would be better than vulkan support
<macc24> iom
<macc24> imo*
<HdkR> macc24: gimme a valhall linux board and I'll do it :P
<macc24> HdkR: gimme a valhall board and i'll do linux on it :P
tomboy64 has joined #panfrost
<HdkR> Can't RE it then, no blob
<HdkR> well...can't RE easily
<macc24> cat /dev/urandom > panfrost_valhall.c
<macc24> in few years it might produce valhall code
<macc24> HdkR: there will be valhall blob\
<HdkR> Hm?
<macc24> there will probably be a chromebook with valhall
<HdkR> I'll buy it immediately
<HdkR> But until someone offers one, SOL since RE on Android is a pita
<macc24> SOL?
<HdkR> and as far as I can tell, since ARM changed how they license their blob out, and dropping X11 support, most SBC makers don't give a flip now
<HdkR> Shit out of luck
<macc24> ok
lukaszErecinski has joined #panfrost
<lukaszErecinski> alyssa, Lukasz from PINE64 here - may I PM you? :)
<HdkR> macc24: I mean, if a Valhall board with blobs launch. I'll dedicate at least a month in RE and then several half months afterwards :P
<macc24> HdkR: well, if valhall board with blob launches i'll do nothing about it aside from "oh i can't wait to get one"
<HdkR> hah
<tomeu> HdkR: mediatek patches mainlining support for the MT8192 are being sent with Project_Global_Chrome_Upstream_Group@mediatek.com in CC
<tomeu> that one has a G-57
<macc24> ^ that's why i got excited about it
<macc24> HdkR: mt8192 has armv8.2 cores ;)
<HdkR> perfect. I'll buy ten
* HdkR wiggles fingers
<HdkR> give me
<HdkR> bleh
<tomeu> guess that doesn't necessarily mean chromebooks in q2, but that press release is all about chromebooks, so maybe?
<macc24> woah that's sooner than i thought
<macc24> and there's mt8195 :O
<macc24> and there's more mt8183 devices than i thought
chewitt_ has joined #panfrost
chewitt has quit [Ping timeout: 256 seconds]
karolherbst has quit [Remote host closed the connection]
<q4a> alyssa I'm ready to write pathes, but is there any plan to get gles3.1? I.e. list of missing features or failed Conformance Test Instructions?
<macc24> q4a: iirc icecream95 experimented with panfrost + lavapipe
<macc24> perhaps
<macc24> ask icecream95, idk
<q4a> ok, thanks
<alyssa> q4a: GLES3.1 is in the works, yeah.
<alyssa> GLES3.1 on v5 (Mali T760, T860) is missing just a few extensions right now, Bifrost is missing a bit more but we're making progress on it.
* macc24 waits for bifrost gles3.1
<alyssa> macc24: lavapipe is intended for software rasterizers only, I don't intend to merge those patches (and I don't believe icecream95 intends to send them)
<macc24> alyssa: ack
<q4a> I'm surrentltt on T760 - so hope, that it will be enought for vulkan
<alyssa> The proprietary Vulkan driver supports T760 so it's theoretically possible
<alyssa> AFAIU there are some ugly hacks needed for Midgard but we'll cross that bridge when we get there
<macc24> alyssa: proprietary drivers support multi rendering(or something) on t628
<alyssa> Yeah, that's a level of crazy I don't want to touch ;)
zkrx has quit [Ping timeout: 272 seconds]
<macc24> it still makes my brain short-circuit that midgard supports gl3.3
<alyssa> macc24: hmm?
<macc24> alyssa: like, all my life i thought "slow = gl2.1" and "arm = gl2.1" but then mali t760mp4 has gl3.3 and my brain just goes "??????" xD
<alyssa> mali t760 does not expose gl3.3 under either mesa or the ddk
<alyssa> although it theoretically could
<macc24> still, even if it's gl3.something
raster has quit [Quit: Gettin' stinky!]
kaspter has quit [Quit: kaspter]
raster has joined #panfrost
Ntemis has joined #panfrost
lukaszErecinski has quit [Quit: Leaving]
zkrx has joined #panfrost
karolherbst has joined #panfrost
zkrx has quit [Ping timeout: 260 seconds]
zkrx has joined #panfrost
Net147 has quit [Read error: Connection reset by peer]
Net147 has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
raster has joined #panfrost
davidlt has quit [Ping timeout: 240 seconds]
tomboy64 has quit [Ping timeout: 240 seconds]
tomboy64 has joined #panfrost
Ntemis has quit [Read error: Connection reset by peer]
<felipealmeida> hello, which compiler version should I use for panfrost?
<felipealmeida> I was able to get two hikey960, running on kernel 5.4 (without drm for now)
<felipealmeida> wanted to know if I should pursue a newer version or it might just be easier to cherry-pick some commits
<macc24> doesn't kirin 960 have g71?
<felipealmeida> yes
<macc24> if your compiler is not ancient, panfrost should compile
<felipealmeida> yes, I just haven't tried yet. Just wanted to know if I need a newer kernel version than 5.4 for drm and stuff to work correctly with panfrost
<felipealmeida> or if 5.4 is good enough
<macc24> i'd recommend something newer
<felipealmeida> I see, I think that will be challenging
<macc24> g72 is already kinda working
<macc24> fyi
<felipealmeida> thanks, I don't want to spend much in hardware yet. Is there much difference between 1st and 2nd generations?
<felipealmeida> of bifrost?
<macc24> i have no idea
<macc24> alyssa?
<alyssa> felipealmeida: macc24: G71 omits support for fp32 transcendentals so is totally broken on panfrost rn
<felipealmeida> oh damn
<alyssa> Could be fixed in an afternoon, just extremely low prio
<felipealmeida> then maybe I can do it
<alyssa> I took one look at the hikey960 install guide and noped the heck out.
<macc24> alyssa: >could be fixed in an afternoon ?
<felipealmeida> if I get it to work first
<felipealmeida> to show something on the screen
<alyssa> macc24: with appropriate reference code anyway
<macc24> felipealmeida: try building 5.10 or 5.11-rc kernel and load panfrost module after adding mali-bifrost binding in dts
<felipealmeida> macc24: I'll try, but I don't have much hope. This 5.4 has lots of HACKs
<felipealmeida> it is actually an android linux kernelk
<macc24> oh my god
<alyssa> I rest my case.
<alyssa> macc24: Now do you see why nobody's fixed G71? :p
<alyssa> ^ I've been busy
<alyssa> HdkR: 😺
<macc24> felipealmeida: there is dts in mainline linux so that suggests some level of support
<felipealmeida> I'm updating the kernel recipe, let's see
<HdkR> alyssa: That's a lot of changes!
<alyssa> HdkR: =D
<alyssa> unfortunately need more changes :<
<macc24> felipealmeida: oh yeah and it looks like mainline linux doesn't support drm on hi3660(kirin 960)
<alyssa> HdkR: that's also entirely new code, not just shuffling around like the IR rework :<
<macc24> alyssa: how to test this code?
<alyssa> macc24: grab my branch alyssa/mesa:scheduler
<macc24> if it renders anything then it's good?
<alyssa> hehe
<felipealmeida> no space left on device :D
<HdkR> Always needs more changes
<alyssa> macc24: It passes GLES2 CI on g52 so
<macc24> can't wait to see some bifrost perf workl
<alyssa> That branch theoretically could improve perf slightly but mostly it's ground work
<macc24> testing is just fancy talk for playing video games
<HdkR> testing, for when you /really/ hope that you didn't break kernel
<macc24> wow
<macc24> xonotic runs on g72
<macc24> much better than my x201 could ever run it
<macc24> alyssa: my duet didn't catch on fire after installing mesa with your patch
<macc24> i count that as a success
<macc24> alyssa: btw that patch has made supertuxkart a bit more broken with PAN_MESA_DEBUG=deqp
<macc24> lemme post a pic
<dstzd> what does deqp do
<macc24> level of weirdness is changing with rotation of kart
<macc24> present in kart-selecting menu too
<macc24> (that's with deqp debug option)
<alyssa> :|
<macc24> dmenu colors are broken
<alyssa> :||
<macc24> it's better when i find it than other people with duet on linux who don't know how to panfrost
<macc24> oh wait a minute
raster has quit [Quit: Gettin' stinky!]
<alyssa> => CRASHED <= while processing these shaders:
<alyssa> ugh
<macc24> alyssa: with gl2 shaders, xonotic walls are yellow
<macc24> without, characters and weapons are just black
<alyssa> with the scheduler branch or master?
<HdkR> Time to put the panfrost shader compiler in a different process and communicate in IPC fashion. So when it crashes you can just continue happily
<macc24> scheduler
<macc24> master renders fine
<alyssa> groan
<macc24> (at least with shaders master is fine)
_whitelogger has joined #panfrost
<alyssa> bug 0 fixed (a crash)
q4a has quit [Remote host closed the connection]
<macc24> agrisis: i PMed you
q4a has joined #panfrost
* alyssa looks how hard constants will be to sort
<macc24> std::sort(constants);
<macc24> hope that helps alyssa
<alyssa> it's C code..
<macc24> std_sort?
<alyssa> qsort
<anarsoul> alyssa: likely you also need to merge them
<anarsoul> however it can be done in nir if it's not done yet :)
<alyssa> anarsoul: Ye, but you just want to make a mergesort joke don't you
<anarsoul> nope
<felipealmeida> well, it boots 5.11
<macc24> felipealmeida: drm works?
<felipealmeida> no drm message on dmesg
<felipealmeida> so, at least not yet
<felipealmeida> nothing from HDMI too, ofc
<felipealmeida> dts doesn't have the gpu bits. But the code seems to be there
<macc24> alyssa: ping me when i can test your patch on duet
<alyssa> ok, I have constant packing sketched in my notebook, not enough time to implement tonight tho
<alyssa> tomorrow then
icecream95 has joined #panfrost
<icecream95> macc24: "CCI frequency is inter-locked". Maybe, but that only seems to apply for the LITTLE cores-if I restrict tinymembench to LITTLE cores, the CCI frequency increases to maximum, but not for the big cores
<macc24> icecream95: little and big core opp table has required-opps
<macc24> it's a bug hmm
<icecream95> Forcing the frequency to maximum speeds up linking Mesa by 25%