alyssa changed the topic of #panfrost to: Panfrost - FLOSS Mali Midgard & Bifrost - - Logs - <daniels> avoiding X is a huge feature
<HdkR> "All Chromebooks will also be Linux laptops going forward"
<HdkR> =O
<HdkR> Shut up and take my money
<alyssa> :)
<HdkR> alyssa: btw, how would you want to do hardware specific screen caps?
<HdkR> Currently I have it setup so a backend can register its own get_param/get_shader_param handler to the panfrost_screen, which additionally returns a boolean if it is actually returning a value
<alyssa> HdkR: For the compiler, you mean?
<alyssa> Having get_shader_param delegate to the compiler is reasonable
<alyssa> get_param should prolly stay where it is, don't be afraid to stick a little is_bifrost logic in there, it won't bite :)
<HdkR> alyssa: Yea, it's basically at the head of those functions so there can be common defines and then compiler specific variables
<HdkR> We also still need a gpu family identifier hanging out in panfrost_screen
<HdkR> or ctx
<HdkR> Not sure if screen can pull in ctx
<alyssa> screen
<alyssa> HdkR: 1 screen, many ctx, in general.
<HdkR> right, so the ctx is_t6xx needs to move
<alyssa> Probably
<HdkR> and...gpu_id I guess?
<HdkR> Basically need a place where I can set it to bifrost and do the screen checks :P
<alyssa> Why won't nmy new kernel boot
jeez_ has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<HdkR> Alright, with a bit more...constructive handling of opts in the standalone compiler I managed to remove all the wacky derefs that wouldn't even appear in a non-standalone environment
stikonas_ has quit [Remote host closed the connection]
<alyssa> Nice.
vstehle has quit [Ping timeout: 258 seconds]
<HdkR> Cleaning up a bit more stuff, made it so uniform loading wasn't faking itself to be UBO loading. Still need to do a fallback for when the uniform registers overflow their maximum capacity but w/e
<HdkR> Also a few other things, but it is becoming cleaner which is making me happy
jeez_ has joined #panfrost
vstehle has joined #panfrost
jeez_ has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<tomeu> o/
<HdkR> \o
pH5_ has quit [Quit: bye]
<HdkR> More Bifrost IR work pushed, started the basics of conditional branching. Not correct but it is showing intent. Need to do comparison + branch fusion eventually as well to make it as effective as it can be
<HdkR> Some other work as well
<tomeu> anarsoul: thanks for reminding, I will ask him directly if I decide to do any work in that direction
<HdkR> Getting in to a groove with the Bifrost work finally. Learning what nir wants from me
<tomeu> HdkR: yeah, looks like you are doing great progress!
<HdkR> That's the end of my ten hours this week. I'll have to do more next weekend :P
<tomeu> that starts in a few hours? :)
<HdkR> er, weekend after this one :)
<tomeu> no rush, the first rockchip socs with bifrost won't be out until next year :)
<hanetzer> hmm? then how are you developing for bifrost? different soc vendor with it?
<tomeu> well, I haven't done any bifrost work yet
<HdkR> tomeu: You can do all the amlogic things for me on the N2 :P
<tomeu> HdkR: what are those things?
<HdkR> Magical devices
<tomeu> HdkR: is there any work needed on the N2 so it's suitable for panfrost development?
stikonas has joined #panfrost
stikonas has quit [Remote host closed the connection]
<HdkR> Not sure actually. I just know I don't want to do the kernel things :D
<cwabbott> tomeu: how far along have we gotten with Bifrost in the upstream driver?
<cwabbott> since we don't have a mesa driver, you could try to port the old shader_runner to it
<cwabbott> I made it to test the ISA, so you have to give it a raw shader blob, but you can extract that from the offline compiler or use the assembler
<cwabbott> that way, we can make sure job submission works
<tomeu> HdkR: oh, that, I don't expect much, tbh
kszaq has quit [Quit: Connection closed for inactivity]
<tomeu> from what I have seen of Arm's kernel driver
<tomeu> I think it should just work, barring some errata if you have a GPU revision with some really unfortunate hw bugs
<tomeu> cwabbott: it's untested on hw with bifrost
<tomeu> narmstrong: are you planning to test the first bitfrost bits on the N2 (or other amlogics with bifrost) once they reach mesa?
<narmstrong> tomeu: sure, what's the actual status ? what can I test ?
<narmstrong> I have another bifrost SoC with G31
<tomeu> I think it's too early, but hanetzer was hoping not to do any kernel work
<tomeu> I don't expect much work to be needed there
<hanetzer> tomeu: hmm?
<tomeu> hanetzer: sorry, I meant HdkR
<hanetzer> heh
<hanetzer> even though most of my code was cribbed and factored out by smarter people than I, I did help get the ball rolling :P
<tomeu> narmstrong: actually, iirc the MMU can/needs a different page table format
<tomeu> hanetzer: we just dumped more code on top of it :)
<narmstrong> tomeu: what are the other differences you are aware of ?
<tomeu> nothing else, robher might know more
<narmstrong> seems we will go by trial an error :-p
<narmstrong> HdkR: you are currently testing with mali_kbase and the non-drm mesa driver ?
<tomeu> HdkR: do you know how far we are from being able to submit such a simple clear job on bifrost?
<tomeu> guess I could get some bifrost hw, but given that baylibre is taking care of amlogic in mainline, it may not be the best use of our time
<tomeu> I should get the igt tests in order and get them merged
<tomeu> there's apparently lots of cheap car radios in aliexpress with the px30, but only the evaluation board is supported in mainline so I'm a bit afraid that testing bifrost on rockchip could be a bit of a sink hole atm
<HdkR> tomeu: Take your time, it's going to be a couple months away still probably
<HdkR> Haven't looked at the job submission side of things yet and we still need to actually emit code
<tomeu> I think job submission should be the same in regards to kernel UABI
<tomeu> and I think the cmdstream structs involved in that function should be also quite similar
<tomeu> well, the second is more of a guess/hope
raster has joined #panfrost
<cwabbott> tomeu: iirc, mali_rt_format has to be a little different
<cwabbott> they moved some of the render target format stuff to the shader descriptor, and changed mali_rt_format to be more suitable to clearing since that's what it's mostly used for in bifrost
<cwabbott> that's the only cmdstream difference relevant to a simple job like that that I can think of
<tomeu> cool
<tomeu> narmstrong: so if you want to validate the kernel UABI with bifrost ^ :)
rhyskidd has quit [Ping timeout: 246 seconds]
rhyskidd has joined #panfrost
metaverse has joined #panfrost
MoeIcenowy has quit [Quit: ZNC 1.6.5+deb1+deb9u1 -]
MoeIcenowy has joined #panfrost
MoeIcenowy has quit [Quit: ZNC 1.6.5+deb1+deb9u1 -]
MoeIcenowy has joined #panfrost
TheKit has quit [Ping timeout: 255 seconds]
TheKit has joined #panfrost
pH5 has joined #panfrost
metaverse has quit [Remote host closed the connection]
MoeIcenowy has quit [Quit: ZNC 1.6.5+deb1+deb9u1 -]
MoeIcenowy has joined #panfrost
rhyskidd has quit [Ping timeout: 246 seconds]
rhyskidd has joined #panfrost
fysa has quit [Read error: Connection reset by peer]
MoeIcenowy has quit [Quit: ZNC 1.6.5+deb1+deb9u1 -]
MoeIcenowy has joined #panfrost
stikonas has joined #panfrost
raster has quit [Remote host closed the connection]
mknawabi has joined #panfrost
pH5 has quit [Quit: -_-]
_whitelogger has joined #panfrost
stikonas has quit [Remote host closed the connection]