ransom has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
ransom has joined #asahi-gpu
<bloom_>
The startup chime always surprises me after so many years away from Apple hardware.
<bloom_>
Also, I'm seeing some graphics artefacts, even on boot. Not sure which is worse, the possibility the GPU is so fragile that a few benign tests damaged something, the possibility the GPU is just unstable in general, or the Apple drivers are that unstable.
<marcan>
bloom_: I've seen graphics artifacts on boot on my macbook pro the other day (boot progress bar messed up), and also inside macos (glitches in system settings)
<marcan>
I think their drivers are just still buggy
browzing has quit [Ping timeout: 256 seconds]
<davidrysk[m]>
I've seen fewer graphical glitches than with nvidia or amd or even intel igpu drivers
<davidrysk[m]>
the nvidia ones were by far the worst
browzing has joined #asahi-gpu
browzing has quit [Ping timeout: 246 seconds]
ransom has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
aratuk has quit [Remote host closed the connection]
bear24rw has quit [Remote host closed the connection]
bear24rw has joined #asahi-gpu
bear24rw has quit [Ping timeout: 246 seconds]
aratuk has joined #asahi-gpu
aratuk has quit [Client Quit]
bear24rw has joined #asahi-gpu
aratuk has joined #asahi-gpu
_whitelogger has joined #asahi-gpu
browzing has joined #asahi-gpu
browzing has quit [Ping timeout: 240 seconds]
ts170[m] has joined #asahi-gpu
bear24rw has quit [Remote host closed the connection]
Mary_ has joined #asahi-gpu
bear24rw has joined #asahi-gpu
bear24rw has quit [Ping timeout: 264 seconds]
jinen[m] has joined #asahi-gpu
<Mary_>
Haven't thought of linking against libLLVM directly to test stuffs, I should maybe give it a try
<Mary_>
(Was actually building something around metallib files and altering AIR)
Jamie[m]2 has joined #asahi-gpu
sharpiro has quit [Ping timeout: 264 seconds]
<bloom_>
marcan: Cute :)
<bloom_>
I'm uncomfortable getting that close to proprietary code for obvious reasons and it was easy enough to get the actual GPU binaries out from memory
<bloom_>
But yeah, my sleepy morning brain confirms that's M1 assembly :)
<bloom_>
or at least could be
<bloom_>
bloom_: I should push my disasm
<bloom_>
er marcan
aratuk has joined #asahi-gpu
aratuk has quit [Remote host closed the connection]
<davidrysk[m]>
AGX2 is very likely M1 GPU
mah has joined #asahi-gpu
ceph3us has joined #asahi-gpu
ceph3us has quit [Remote host closed the connection]
<Mary_>
davidrysk[m], it is
mah is now known as mch
mch is now known as mah
mah has quit [Quit: Leaving]
<bloom_>
K
* bloom_
blogs
Axenntio has joined #asahi-gpu
Axenntio has quit [Client Quit]
Axenntio has joined #asahi-gpu
Axenntio_ has joined #asahi-gpu
Axenntio_ has quit [Remote host closed the connection]
<bloom_>
More code to be uploaded to that repo soon, need to clean some stuff up, maybe will push to a branch sooner
bloom_ is now known as bloom
<Mary_>
bloom, nice work!
<bloom>
Mary_: Thanks :)
<Mary_>
I think I will try to take a look at it a bit later today see if I can help a bit
<bloom>
the wrap library will be released shortly, need to disentangle it from unrelated experiment code
<Mary_>
bloom, is disassembling Metal dynamic library a no no for you? I noticed that you can ask Metal api to prebuild shaders for the current target you are running on and export it.
<bloom>
Mary_: Disassembly is absolutely a no-no.
<Mary_>
I mean you aren't disassembling any proprietary code here tho bloom
<bloom>
---Correct.
<bloom>
Disassembly != disassembly
<bloom>
Writing out own disassembler to disassemble shaders we write ourselves, for ISA r/e -- ok
<bloom>
Using an arm64 disassembler to disassemble Apples's proprietary code -- noto ok
<Mary_>
ah okay I wasn't clear
<bloom>
Nor was I, sorry
<Mary_>
I was meaning using Metal APIs to generate a GPU compiled file and dissassembling the result of it
<Mary_>
(All of that is standard metal stuffs)
<bloom>
Oh, right. I didn't know a priori if that was actual GPU binaries instead of LLVM IR serialization
<Mary_>
there is the two variants
bear24rw has joined #asahi-gpu
bear24rw has quit [Ping timeout: 256 seconds]
bear24rw has joined #asahi-gpu
bear24rw has quit [Remote host closed the connection]
bear24rw has joined #asahi-gpu
Axenntio has joined #asahi-gpu
carlosn has joined #asahi-gpu
bear24rw has quit [Ping timeout: 264 seconds]
<bloom>
wrap code pushed but probably fairly broken rn, whoops
bear24rw has joined #asahi-gpu
j`ey has joined #asahi-gpu
Hexagon has joined #asahi-gpu
<marcan>
bloom: want a tweet from @AsahiLinux?
<marcan>
(sorry, was doing dinner and things)
tiago_ has joined #asahi-gpu
modwizcode has joined #asahi-gpu
browzing has joined #asahi-gpu
devinvs[m] has joined #asahi-gpu
Axenntio_ has joined #asahi-gpu
Axenntio has quit [Remote host closed the connection]
Axenntio_ has quit [Remote host closed the connection]
Axenntio has joined #asahi-gpu
Axenntio has quit [Remote host closed the connection]
<Shiz>
bloom: very nice going
<bloom>
thank you :)
neobrain has joined #asahi-gpu
bear24rw has quit [Remote host closed the connection]
bear24rw has joined #asahi-gpu
stormclad has joined #asahi-gpu
tiago_ is now known as tmartins
bear24rw has quit [Remote host closed the connection]
bear24rw has joined #asahi-gpu
bear24rw has quit [Remote host closed the connection]
bear24rw has joined #asahi-gpu
tmartins is now known as tiagom
tiagom has quit [Quit: tiagom]
tiagom has joined #asahi-gpu
tbodt has joined #asahi-gpu
<sven>
bloom: nice work :)
bear24rw has quit [Remote host closed the connection]
bear24rw has joined #asahi-gpu
maciejjo has joined #asahi-gpu
Baughn has joined #asahi-gpu
Axenntio has joined #asahi-gpu
Axenntio_ has joined #asahi-gpu
Axenntio_ has quit [Remote host closed the connection]
Axenntio has quit [Remote host closed the connection]
<dhewg>
yeah, seconded, just read your post ;)
<Yuzu>
so AGX is a fully dynamically scheduled machine? have you seen any sign of mandatory delays before dependent ops or anything like that?
<dhewg>
although i found "with the 8th bit from the 8th bit of the *second* byte" weird to parse
Axenntio has joined #asahi-gpu
Axenntio_ has joined #asahi-gpu
Axenntio_ has quit [Remote host closed the connection]
<dhewg>
you mean "(op[0] & 0x7f) | (op[1] & 0x80)", right?
Axenntio_ has joined #asahi-gpu
Axenntio_ has quit [Remote host closed the connection]
<dhewg>
nvm, that's like exactly line 264 further down :)
Axenntio has quit []
Axenntio_ has joined #asahi-gpu
Axenntio_ has quit [Remote host closed the connection]
ConeOfAttack[m] has joined #asahi-gpu
fl35[m] has joined #asahi-gpu
<bloom>
sven: thank you :)
<bloom>
Yuzu: Too early to tell and it would be speculation anyway, compiler can only see so much.
<bloom>
dhewg: I am embarassed by the code, yes
<Yuzu>
bloom: Fair - was just wondering if you had seen any patterns in generated code that would be obviously inefficient unless there's delays
<bloom>
Yuzu: No(t yet), which is why I'm concluding there's quite a bit of hw scheduling.
<bloom>
Contrast eg Adreno
brinly has joined #asahi-gpu
<q3k|m>
bloom: impressive work
<bloom>
thanks :)
Axenntio has joined #asahi-gpu
Axenntio has quit [Remote host closed the connection]
Axenntio has joined #asahi-gpu
<ConeOfAttack[m]>
With regards to 32 v 16 bit ALU performance, is there a way to see if 32-bit ALUs are repurposed as two 16-bit units? Iirc the last couple of generations from nvidia pushed this for ints and fps probably for neural network workloads.