#asahi-dev · 2022-06-04 08:10
chadmedits not actually the run time in this case, i have it set to few enough loops that it comes back successfully after running#asahi-dev · 2022-06-04 07:13
chadmedbut when i try to run too many iterations of it (trying to make a power virus) the system irrecoverably hangs and i have to hard reboot it #asahi-dev · 2022-06-04 07:13
chadmedis there something deeper going on in m1n1 that causes that little bench: piece of assembly to hang the system? i cant see why it would #asahi-dev · 2022-06-03 01:25
chadmedthe "correct" way to do it upstream is still to build the energy model manually in the cpufreq driver or let the em subsystem build its own with a dodgy little formula that tries to guess how much power each core uses based on its operating points, which seems so dirty and gross#asahi-dev · 2022-06-03 01:24
chadmedargh TIL that the dt bindings that make doing energy aware scheduling easier are a downstream android-only feature... and arm's documentation on energy aware scheduling only covers using the android specific bindings#asahi-dev · 2022-05-26 01:01
chadmedthe arduino method still works#asahi-dev · 2022-05-26 01:01
chadmedi think theres been bigger fish to fry and since its not really needed anymore no ones worked on it#asahi-dev · 2022-04-07 05:40
chadmedm1n1 experiments?#asahi-dev · 2022-04-05 12:40
chadmedthere will be no m2, there will only be M1/Pro/Max/Ultra Gen 2 SuperSpeed#asahi-dev · 2022-04-05 07:07
chadmedmore generally, m1n1 handles all the low level init required for linux to see a "normal" arm64 machine in a state it would expect to find one in#asahi-dev · 2022-04-03 06:52
chadmedi did mean the encoders#asahi-dev · 2022-04-03 06:52
chadmedidk why i said decoders in my initial message#asahi-dev · 2022-04-03 06:52
chadmedso am i, it's sunday :-P #asahi-dev · 2022-04-03 06:49
chadmedyeah which is why i think it's the encoders that are slow#asahi-dev · 2022-04-03 05:21
chadmedso its just the encoders that seem limited#asahi-dev · 2022-04-03 05:21
chadmedbut if i then try to encode that to H.264 or HEVC the framerate tanks #asahi-dev · 2022-04-03 05:20
chadmedusing hardware decode #asahi-dev · 2022-04-03 05:20
chadmedthe gist i posted above has the figures, but i can transcode an HDR10 4K HEVC stream to ProRes at 300+ fps #asahi-dev · 2022-04-03 05:20
chadmedyeah the decoders are fast #asahi-dev · 2022-04-03 05:19
chadmedi mean we could also just apply occams razor and conclude that the encoders are not that fast#asahi-dev · 2022-04-03 05:15
chadmedbut then why would the 1080p video i used decode at like 200fps#asahi-dev · 2022-04-03 05:14
chadmedit could be possible that it limits itself to some number above realtime so that video plays back and the buffer fills up at a decent speed, but no unnecessary power is used#asahi-dev · 2022-04-03 05:13
chadmedi might see if i have any high framerate video lying around i can feed it. since all the framerates are still above realtime theres still one theory id like to eliminate#asahi-dev · 2022-04-03 05:11
chadmedJamie[m]1: ffmpeg just passes the data through VideoToolbox so if its not fully utilising the hardware that would be an apple thing, right? #asahi-dev · 2022-04-03 05:09
chadmedhuh odd, brga8888, brga10101010 and brga12121212 all worked for me on the prores block #asahi-dev · 2022-04-03 04:59
chadmedi find it odd that theyre slower than 60fps, im not entirely convinced#asahi-dev · 2022-04-03 04:59
chadmedlooks to me like the AVCHD and HEVC decoding blocks are indeed limited#asahi-dev · 2022-04-03 04:59
chadmedhttps://gist.github.com/chadmed/f2ab8370056e84a6a9a4b9e226a93302#asahi-dev · 2022-04-03 02:25
chadmedi really just wanted to confirm my theory about the pixel format and got super carried away (spectrum moment) #asahi-dev · 2022-04-03 02:24
chadmedit could just be a macos thing too, once we have access to these blocks from linux we might be able to shed some light on the situation#asahi-dev · 2022-04-03 02:24
chadmedwould kind of ruin my plans to buy an m2 mac mini and use it as a server if that's the case :P #asahi-dev · 2022-04-03 02:23
chadmedyeah its possible#asahi-dev · 2022-04-03 02:23
chadmedquestions to answer after lunch maybe #asahi-dev · 2022-04-03 02:22
chadmedbut then why was software decoding of that same stream like 500fps? #asahi-dev · 2022-04-03 02:22
chadmedthey are capable of decoding fast enough, since encoding into prores is fast#asahi-dev · 2022-04-03 02:21
chadmedthe hevc and h264 blocks are definitely behaving like theyre limited in some way#asahi-dev · 2022-04-03 02:21
chadmedyeah #asahi-dev · 2022-04-03 02:18
chadmednever goes above 53fps on that 4K HDR10 stream #asahi-dev · 2022-04-03 02:17
chadmedthe bloomer in me wants to say they just "forgot" to take off the throttle in the firmware or something :P#asahi-dev · 2022-04-03 02:16
chadmedthe cynic in me says apple is pushing people into using prores #asahi-dev · 2022-04-03 02:16
chadmeddecoding the HEVC stream, encoding in prores to null is about 300fps #asahi-dev · 2022-04-03 02:16
chadmedthe prores block, which is brand new, is much much much faster #asahi-dev · 2022-04-03 02:15
chadmedlike i said its something youd definitely do on a mobile device and if they just plonked the blocks down into these chips with no changes...#asahi-dev · 2022-04-03 02:15
chadmedit would make sense if the hardware blocks are just throttled imo#asahi-dev · 2022-04-03 02:15
chadmed~60% on two firestorm cores#asahi-dev · 2022-04-03 02:15
chadmednah its behaving threaded#asahi-dev · 2022-04-03 02:13
chadmedencoding to null does not make it faster either#asahi-dev · 2022-04-03 02:13
chadmedyeah it sort of ramps up to that speed and then stays there#asahi-dev · 2022-04-03 02:10
chadmedspeed doesnt seem to change when forcing lower or higher bitrates too #asahi-dev · 2022-04-03 02:10
chadmed50fps on a 4K HDR10 stream