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
<icecream95> alyssa: With SuperTuxKart: bi_register_allocate: Assertion `success' failed.
<urjaman> :D
<alyssa> icecream95: Oof.
<icecream95> (this is with !8135, it seems to work fine on master)
<alyssa> ah
<alyssa> regressions will be the bane of me
<alyssa> `midgard_instruction` who has time to type that
<alyssa> `bi_instr` much better
<icecream95> `panfrost_bifrost_internal_representation_instruction_struct`
<alyssa> 😇
<alyssa> Someone's been writing C++ :p
<alyssa> IOGraphicsLLVMBifrostCompilerImmediateRepresentationInstruction
<alyssa> look I can write mac drivers too
<alyssa> So how is a single pair of tests failing with this..
<alyssa> valgrind clean
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
<icecream95> alyssa: C++ would be mesa::drivers::panfrost::bifrost::compiler::bifrost_ir::Instruction<int Arch, typename Ref=std::variant<std::shared_ptr<mesa::drivers::panfrost::bifrost::compiler::bifrost_ir::Instruction_base>, std::weak_ptr<mesa::drivers::panfrost::bifrost::compiler::bifrost_ir::Instruction_base>>, typename
<icecream95> It=std::vector<mesa::drivers::panfrost::bifrost::compiler::bifrost_ir::Instruction_base>::iterator>
<alyssa> | c++filt
* alyssa trying to debug midgard compiler bug erg
<alyssa> something extremely subtle is broken in this patch..
<urjaman> aww
<alyssa> also dearie the midgard compiler horrifies me what was past alyssa thinking
<alyssa> If I force in-order scheduling it's ok
<alyssa> WTF that should've fixed it
mixfix41_ has joined #panfrost
<alyssa> WTF that should not have fixed it
<alyssa> Is it possible there's some unrelated bug being unmasked elsewhere?
* alyssa needs a magic oracle
<icecream95> alyssa: Maybe you should ask Santa for one
<alyssa> icecream95: I'm Jewish :(
<alyssa> 🎅
<icecream95> So is it already too late for presents this year?
<alyssa> Hmm, I guess there's always Purim, anything goes~
mixfix41 has quit [Ping timeout: 256 seconds]
<icecream95> PIGLIT: {"result": "pass" }. That's cl-custom-run-simple-kernel on Midgard done, now to fix it on Bifrost...
<alyssa> sounds like someone's having fun :)
<alyssa> Meanwhile, I found my bug. Just.. huge ugh.
<alyssa> Oh no......
<icecream95> alyssa: You already said that it sounded like I was having fun three months ago
<alyssa> Your memory is better than mine, I only have room for random trivia and multple instruction sets
* alyssa sighs
<alyssa> Another idea down the drain. Meh.
kaspter has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
vstehle has quit [Ping timeout: 260 seconds]
bbrezillon has quit [Ping timeout: 258 seconds]
stikonas has quit [Remote host closed the connection]
dstzd has quit [Quit: ZNC - https://znc.in]
dstzd has joined #panfrost
jernej has quit [Quit: Free ZNC ~ Powered by LunarBNC: https://LunarBNC.net]
dstzd has quit [Client Quit]
jernej has joined #panfrost
dstzd has joined #panfrost
kaspter has quit [Ping timeout: 246 seconds]
jernej has quit [Client Quit]
kaspter has joined #panfrost
dstzd has quit [Client Quit]
dstzd has joined #panfrost
dstzd has quit [Client Quit]
dstzd has joined #panfrost
dstzd has quit [Client Quit]
kaspter has quit [Ping timeout: 246 seconds]
kaspter has joined #panfrost
jernej has joined #panfrost
jernej has quit [Client Quit]
jernej has joined #panfrost
jernej has quit [Client Quit]
jernej has joined #panfrost
jernej has quit [Client Quit]
jernej has joined #panfrost
jernej has quit [Client Quit]
dstzd has joined #panfrost
dstzd has quit [Client Quit]
jernej has joined #panfrost
<icecream95> I've got panwrap starting on Chrome OS now: panwrap: Detected GPU ID 6221
dstzd has joined #panfrost
rando25892 has joined #panfrost
archetech has joined #panfrost
vstehle has joined #panfrost
davidlt has joined #panfrost
archetech has quit [Quit: Konversation terminated!]
<icecream95> It turns out the faults on Bifrost were just because I had set uniform_count but push_uniforms were still zero
<icecream95> Oh well, at least I have panwrap set up now if I need it
<icecream95> Now I just need to find out how to stop piglit running every test twice...
Prf_Jakob has quit [Ping timeout: 256 seconds]
Prf_Jakob has joined #panfrost
rando25892 has quit [Remote host closed the connection]
bbrezillon has joined #panfrost
dstzd has quit [Quit: ZNC - https://znc.in]
dstzd has joined #panfrost
icecream95 has quit [Ping timeout: 256 seconds]
Ashleee has joined #panfrost
camus has joined #panfrost
kaspter has quit [Ping timeout: 264 seconds]
camus is now known as kaspter
yann has quit [Ping timeout: 260 seconds]
yann has joined #panfrost
showliu has quit [Quit: ZNC - http://znc.in]
steev_ has joined #panfrost
narmstrong has joined #panfrost
cwabbott has joined #panfrost
kinkinkijkin has joined #panfrost
raster has joined #panfrost
stikonas has joined #panfrost
Net147 has quit [Ping timeout: 240 seconds]
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
Net147 has joined #panfrost
empty_string has quit [Ping timeout: 240 seconds]
empty_string has joined #panfrost
<macc24> HdkR: cool
<macc24> but won't increase of display resolution kill g31?
<robmur01> IKR, just scanning out 854x480 takes literally hundreds of megabytes per second of memory bandwidth! :P
<robmur01> (specifically, 0.98 hundreds)
<macc24> that's a lot of bytes per second
<alyssa> macc24: I guess for the intended use case it's ok, but yeah don't do any 3D gaming maybe
<macc24> is g32 that slow?
<alyssa> g31 but... it's low power ;)
<macc24> finally low resolution is a plus
<alyssa> Yeah
<alyssa> "Pixel/Texturing Throughput 1.3 Gpix/s"
<alyssa> So you're not going to be fillrate limited, but..
<macc24> would mpv work in full 60fps on that device?
<alyssa> Ought to
<alyssa> Panfrost is slow on Bifrost rn
<alyssa> "make it work, _then_ make it fast" ;)
<macc24> oh, talking about "slow", minetest on g72 is slower than i think it should be
<macc24> when looking at trees the performance tanks
<alyssa> Yeah, I am not taking performance issues on Bifrost :p
<alyssa> Optimization is a Big™ project that's already in progress, just be patient and maybe play less video games if it bugs you 😇
<alyssa> :p
<macc24> the slowness is also seen in firefox
<macc24> when i don't force minimum clock to be 500something
kaspter has quit [Quit: kaspter]
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
tomboy64 has quit [Remote host closed the connection]
tomboy64 has joined #panfrost
<alyssa> ok
<macc24> wait a second
<macc24> robmur01: >just scanning out 854x480 takes literally hundreds of megabytes per second of memory bandwidth
<macc24> isn't that what afbc is for?
kaspter has joined #panfrost
<alyssa> yeah
<macc24> so it isn't as bad
<robmur01> but also: sarcasm ;)
<robmur01> G31 is optimised for low power, but that's still in the context of running modern-day Android on low-end phones - that's still a relatively tiny display, even if it's bigger than the even tinier one
<macc24> i wonder if i can run portal on odroid go2, it surely won't be gpu'd on a 480x320 display xD
<robmur01> I'd be pretty sure that OGS is still going to be CPU-limited for the things it's intended for
<macc24> i gotta get sound working on duet before getting a new toy, i mean device
<alyssa> g31 = toys :p
<macc24> it's a quadcore 1.3ghz arm chip that may be faster than my old phone
<robmur01> A35 is distinctly less beefy than even A53, though
<macc24> distrinctly as in?
unoccupied has joined #panfrost
<robmur01> as in they're not even competing in the same class - A35's goal was basically to be a 64-bit A7: https://community.arm.com/developer/ip-products/processors/b/processors-ip-blog/posts/introducing-cortex-a35-arm-s-most-efficient-application-processor
unoccupied is now known as leah
<robmur01> A35/G31 is really the direct successor to the ubiquitous A7/Mali-400 combo
<chewitt> s/simple_ondemand/performance makes a noticeable difference to Kodi GUI performance with bifrost
<chewitt> it never seems to scale up from the min GPU freq for me
<chewitt> (very unscientific testing tho)
<alyssa> It astonishes me I ever got the Midgard compiler working
<macc24> chewitt: does it not scale frequency with vblank_mode=0 glxgears too?
alpernebbi has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
<dschuermann> can someone help out on https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6666 ? I get 2 unexpected passes on CI which block me from merging :)
<alyssa> dschuermann: oh no, what a tragedy :)
<dschuermann> yeah, that mr kills me with all the ci fails all the time :(
* alyssa trying to get lower_vars_to_scratch working
stikonas has quit [Remote host closed the connection]
* alyssa can't read
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
<alyssa> oh, it's uncovered a bug in the spill ocde -.-
<alyssa> --and now it's failing to spill at all
<alyssa> *whistles*
stikonas has quit [Remote host closed the connection]
stikonas has joined #panfrost
<br_> I think I ported the DRM scheduler to FreeBSD :). But I see that some jobs are failed with error 0x58 (DATA_INVALID_FAULT)
<alyssa> Oh boy.
<br_> 4th job received from mesa, sometimes 5th or 6th job
<alyssa> which GPU?
<br_> T860
<alyssa> Hm.
stikonas has quit [Remote host closed the connection]
<br_> I've just rebuilt mesa with GCC9 (it was clang). But no change
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
<alyssa> :<
stikonas has joined #panfrost
tgall_foo has quit [Quit: My iMac has gone to sleep. ZZZzzz…]
<alyssa> br_: cache issues? race conditions?
felipealmeida has joined #panfrost
<felipealmeida> hi, which unexpensive board is available to test panfrost on bifrost?
<macc24> inexpensive*
<macc24> and define "inexpensive"
<felipealmeida> thanks
<felipealmeida> well, less than 200usd preferably
<felipealmeida> the lower, the better
<br_> alyssa: possibly. or maybe syncobj issues
<alyssa> br_: also possible
<alyssa> maybe worth dumping GPU memory before/after jobs, etc
<alyssa> (PAN_MESA_DEBUG=trace may help)
<macc24> felipealmeida: my guess is that an sbc with mainline linux support would be good enough
<robmur01> cheapest would probably be S905X3-based ones
<felipealmeida> banana pi is quite inexpensive. M5 is G31
<felipealmeida> that's a good tip, thanks robmur01 and macc24
<robmur01> VIM3L, ODROID-C4, or hacking cheap TV boxes if you're braver
<macc24> some rk3326 board as a handheld works too with panfrost
<robmur01> chewitt is probably the authority there ;)
<robmur01> Allwinner H616 might become another possibility in future - initial upstream support is just starting to happen now
* robmur01 is staunchly holding out for RK356x as first Bifrost
<robmur01> *my
rando25892 has joined #panfrost
<felipealmeida> problem with odroid for me is that shipping from hardkernel is usually expensive.
<felipealmeida> maybe aliexpress has it
<macc24> or a local reseller
<felipealmeida> yeah, I'm brazil. Couldn't find any
<felipealmeida> I may try with salveti, he worked at Linaro. He might lend me one board
<br_> alyssa: thanks, will try !
<sfan5> shipping US->Brazil is also expensive I assume?
<felipealmeida> sfan5: depends, usps not so much
<felipealmeida> DHL is really expensive though
<felipealmeida> aliexpress has really inexpensive shipping fees though
<sfan5> you could look into US resellers that have an odroid-c4 then
<felipealmeida> didn't show any in my google search, I'll try that
<felipealmeida> robmur01: oh! that makes it easier, thanks
tgall_foo has joined #panfrost
alpernebbi has quit [Quit: alpernebbi]
steev_ is now known as steev
kaspter has quit [Quit: kaspter]
davidlt has quit [Ping timeout: 268 seconds]
raster has quit [Quit: Gettin' stinky!]
<alyssa> Hmm, shaders that are exactly 4096 bytes seem to confuse Bifrost (leading to execution permission faults)
<alyssa> Not sure who's bug this is.
<alyssa> *whose
<macc24> definitely i didn't introduce it
<urjaman> add a nop on the end? :P
cphealy has quit [Quit: Leaving]
raster has joined #panfrost
<dschuermann> alyssa: I would be interested to know which patch of !6666 resolves the 2 crashes and if they are also resolved on t760 so that I can remove them from the CI list
<dschuermann> (I assume I know the answers, but would like to confirm)
uis has joined #panfrost
uis has quit [Client Quit]
uis has joined #panfrost
raster has quit [Quit: Gettin' stinky!]