marcan changed the topic of #asahi-gpu to: Asahi Linux: porting Linux to Apple Silicon macs | GPU / 3D graphics stack black-box RE and development (NO binary reversing) | Keep things on topic | GitHub: https://alx.sh/g | Wiki: https://alx.sh/w | Logs: https://alx.sh/l/asahi-gpu
TheLink has joined #asahi-gpu
aratuk has joined #asahi-gpu
aratuk has quit [Ping timeout: 264 seconds]
ransom has joined #asahi-gpu
ransom has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<bloom>
Alright, let's see what I can get done in an hour :)
<bloom>
(nothing is if I just read random github issues fo an hour..)
<bloom>
I guess wiring in the dwetile
<icecream95>
"Oddly NEON is making things slower". Because GCC's handling of ARM vector intrinsics sucks?
<bloom>
icecream95: Yeah, probably :p
<icecream95>
Memory access patterns are also important. Ideally you want to write entire cache lines at once, but I think Aarch64 dropped that instruction
<bloom>
Detiling should write linearly but read scattered, just as scalar code.
<jn__>
bloom: not sure if it's intended, but your deleted text comes through to the other side of IRC, e.g. "dw␡etile"
<bloom>
jn__: ?
<hthh_>
i just see "I guess wiring in the dwetile"
<jn__>
dw<character that represents backspace>tile
<jn__>
U+2421␡GRAPHIC FOR DELETE <- this character
<jn__>
ah wait, i mean "dw<backspace>etile"
<jn__>
no offence meant, it just seems like an obscure misconfiguration of something
<icecream95>
The matrix coprocessor probably has an instruction for tiling anyway
<bloom>
jn__: Contrary to what you might think based on my code, I am not good at using computers. ^^
<bloom>
icecream95: That's cheating :<
<bloom>
But you might be right :)
<jn__>
bloom: no worries :)
<bloom>
Ok, got the detiling wired in.
<bloom>
marcan: Turns out the framebuffer needs to be 128 byte aligned. I was previously aligning to 64 bytes only, and that's where the missing 64 bytes came from.
klaus_ has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<brentr123[m]>
So is metal going to be the graphics driver or will it be based off of moltenvk/molten gl?
<icecream95>
brentr123[m]: AFAIK the plan is to write a native OpenGL and/or Vulkan driver with no Metal involved
<bloom>
^^ that
<bloom>
I am still undecided about GL or Vulkan first.
<brentr123[m]>
Anyway to do both?
<bloom>
With 2x the work..
<brentr123[m]>
Get the best of both worlds
<brentr123[m]>
Ah
<bloom>
GL-on-VK is a thing but meh
ransom_ has joined #asahi-gpu
taowa has joined #asahi-gpu
ransom_ has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
aratuk has joined #asahi-gpu
aratuk has quit [Ping timeout: 265 seconds]
aratuk has joined #asahi-gpu
ransom has joined #asahi-gpu
aratuk has quit []
khronokernel[m] has joined #asahi-gpu
aratuk has joined #asahi-gpu
<davidrysk[m]>
the device probably maps better to VK
<ConeOfAttack[m]>
Quick question, I'm out of my league here but don't we already know Apple is running some custom instructions?
<davidrysk[m]>
it's a GPU, it's entirely custom instructions
<ConeOfAttack[m]>
... I figured they'd reuse as much between CPU and GPU as possible. So I was referring to ARM instructions.
<ConeOfAttack[m]>
Sorry, I was looking at an earlier discussion.
aratuk_ has joined #asahi-gpu
aratuk has quit [Ping timeout: 246 seconds]
wiloon has joined #asahi-gpu
aratuk_ has quit [Remote host closed the connection]
ransom has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
Tokamak_ has joined #asahi-gpu
<artemist>
Looking at benchmarks, Zink performans way better than I would expect
Tokamak has quit [Ping timeout: 256 seconds]
Tokamak_ has quit [Ping timeout: 265 seconds]
Tokamak has joined #asahi-gpu
Tokamak has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
ransom has joined #asahi-gpu
ransom has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
Tokamak_ has joined #asahi-gpu
<austriancoder>
bloom: Vulkan first :)
aratuk has joined #asahi-gpu
aratuk has quit [Ping timeout: 256 seconds]
Tokamak_ has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<marcan>
bloom: that makes sense
<marcan>
that reminds me of the wii NAND ECC calculation hardware... it ostensibly writes the calculated ECC buffer after the ECC spare buffer, but it actually writes it to the address XOR 0x40
<marcan>
so if you don't align it ends up *before* the buffer
<marcan>
(I figured out this behavior via experimentation, then many years later when the official docs leaked, it turns out they don't document it properly... my docs are better than the official ones :))
klaus_ has joined #asahi-gpu
solarkraft has joined #asahi-gpu
Necrosporus has quit [Ping timeout: 240 seconds]
icecream95 has quit [Ping timeout: 265 seconds]
klaus_ has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
klaus has joined #asahi-gpu
DarthCloud has quit [Ping timeout: 240 seconds]
DarthCloud has joined #asahi-gpu
DarthCloud has quit [Remote host closed the connection]
DarthCloud has joined #asahi-gpu
aratuk has joined #asahi-gpu
aratuk has quit [Ping timeout: 240 seconds]
klaus has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
klaus has joined #asahi-gpu
Necrosporus has joined #asahi-gpu
klaus has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
klaus has joined #asahi-gpu
klaus has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
klaus has joined #asahi-gpu
<bloom>
austriancoder: right, the problem with that is -- afaiu -- Vulkan requires a deep understanding of the hardware, whereas I'm at the point I could throw together a gallium driver passing most of deqp-gles2 by just bitbanging magic numbers
klaus has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<bloom>
(and more to the point - I am intimately familiar with Gallium, I do not know Vulkan)
<austriancoder>
bloom: passing most of deqp-gles2 .. I think this at least needs one more week for you. but yes I am in the same boat.. i know gallium quite well but never touched vulkan. also in linux userspace is more gles then vulkan atm
<bloom>
austriancoder: yeah, that's the other thing, there's no point in paying vulkan tax if the entire userspace is going through zink, and that's one more layer of abstraction to debug
<bloom>
especially on arm
aratuk has joined #asahi-gpu
aratuk has quit [Ping timeout: 246 seconds]
daniels has joined #asahi-gpu
Tokamak has joined #asahi-gpu
<bloom>
austriancoder: right, the problem with that is -- afaiu -- Vulkan requires a deep understanding of the hardware, whereas I'm at the point I could throw together a gallium driver passing most of deqp-gles2 by just bitbanging magic numbers
* j`ey
dejavu
<brentr123[m]>
what is gallian? ive never heard of that
klaus has joined #asahi-gpu
<modwizcode>
It's the state machine for mesa right?
<davidrysk[m]>
bloom: now I'm curious -- just how *are* vulkan drivers on Linux implemented
<davidrysk[m]>
they seem to be part of mesa
<davidrysk[m]>
> LunarG compared Vulkan to "Gallium3D 2.0." << heh
<neobrain>
I'm actually surprised to hear implementing a Gallium driver requires less understanding of hardware than a Vulkan one, but then again I've never done the former :)
<neobrain>
though the "bigbanging magic numbers" bit sounds like it wouldn't be a particularly nice or performant driver
rann has joined #asahi-gpu
Tokamak has quit [Ping timeout: 240 seconds]
Tokamak has joined #asahi-gpu
ransom has joined #asahi-gpu
<marcan>
that's not what that means
DarthCloud has quit [Remote host closed the connection]
DarthCloud has quit [Remote host closed the connection]
DarthCloud has joined #asahi-gpu
konrad__ has joined #asahi-gpu
<austriancoder>
davidrysk[m]: they are part of Mesa but the different drivers are not sharing much code as for OpenGL. For OpenGL (ES/CL) there is a common interface called gallium
DarthCloud has quit [Ping timeout: 240 seconds]
DarthCloud has joined #asahi-gpu
<bloom>
marcan: bwah?
<marcan>
bloom: I mean it doesn't mean it won't perform well :p
Tokamak has quit [Ping timeout: 240 seconds]
icecream95 has joined #asahi-gpu
Tokamak has joined #asahi-gpu
aratuk has joined #asahi-gpu
aratuk has quit [Ping timeout: 256 seconds]
modwizcode has quit [Ping timeout: 272 seconds]
ransom has joined #asahi-gpu
ransom has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
ransom has joined #asahi-gpu
TonyRedondo has quit [Quit: Connection closed for inactivity]
Tokamak has quit [Ping timeout: 256 seconds]
Tokamak has joined #asahi-gpu
m42uko has quit [Ping timeout: 260 seconds]
m42uko has joined #asahi-gpu
ransom has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]