#asahi-gpu · 2022-11-04 07:43
phireer10- 30808#asahi-gpu · 2022-11-04 07:41
phireer, what version?#asahi-gpu · 2022-11-04 07:40
phire14" macboo pro#asahi-gpu · 2022-11-04 07:40
phiret6001#asahi-gpu · 2022-11-04 07:40
phire0x0#asahi-gpu · 2022-11-04 07:40
phire0x30808#asahi-gpu · 2022-11-04 07:40
phire0x0#asahi-gpu · 2022-11-04 07:40
phire0x40a06#asahi-gpu · 2022-11-04 07:40
phire0x4042000#asahi-gpu · 2022-11-04 07:40
phirethese are 32bit regs?#asahi-gpu · 2022-11-04 07:35
phirelina: [FWLog[1]] [3 45.2119402] Build version: B0#asahi-gpu · 2022-11-04 07:31
linaphire, jannau: Any chance you could dump 0x404d04000 .. 20 on your machines? ^^ (might need to pmgr enable the GPU first)#asahi-gpu · 2022-11-01 01:23
linaI already know f-i is linear, so I left it overnight grabbing v-p curves @ two frequencies to get the dynamic v-i curves (core & sram). If we can solve that one, that's everything.#asahi-gpu · 2022-11-01 01:22
linaYeah, exactly#asahi-gpu · 2022-11-01 01:22
chadmedi guess the silver lining is that if apple can get it so obviously wrong and the gpu still works fine then we probably dont need to worry too much either :P #asahi-gpu · 2022-11-01 01:20
chadmedhow delightful!#asahi-gpu · 2022-11-01 01:19
linaProbably at the very least power state decisions#asahi-gpu · 2022-11-01 01:19
linachadmed: There's like 10 copies of the max-power number in initdata, so I'm pretty sure it's used for lots of things...#asahi-gpu · 2022-11-01 01:15
chadmedlina: if it only really matters up to where that break is then i dont see any issue with doing that. do we know what these derived values are actually used for? do they influence what state the firmware puts the GPU into?#asahi-gpu · 2022-11-01 00:01
DarkShadow4444is there already a way to play with the gpu driver on asahi linux (mac m1 mini)?#asahi-gpu · 2022-10-31 12:50
linaIf I just extrapolate the first piecewise component, that's already intersecting the discontinuity (closer to the top side)... maybe we should just go with that?#asahi-gpu · 2022-10-31 12:49
linaThe voltage range that matters is about 400-920mV, especially the high side... which conveniently is precisely dancing on the broken discontinuity.#asahi-gpu · 2022-10-31 12:48
linaI don't know what to do about this one. Obviously copying Apple's curve there is wrong and silly.#asahi-gpu · 2022-10-31 12:48
linaIt turns out their math is demonstrably wrong by several percentage points...#asahi-gpu · 2022-10-31 12:48
linachadmed: Meanwhile Apple: https://twitter.com/LinaAsahi/status/1587058352885882880#asahi-gpu · 2022-10-31 10:20
chadmedthere is a presposterous 10th order polynomial which gives (statistically) significantly less error (p ~ 1.1x10^-5) but it's absolutely ridiculous and materially the difference between means is 400 picowatts#asahi-gpu · 2022-10-31 10:16
chadmedabout the same accounting when rounding though#asahi-gpu · 2022-10-31 10:15
chadmedtheir average is wrong, its actually better than they claim#asahi-gpu · 2022-10-31 09:59
chadmedtighter error terms than what i got too#asahi-gpu · 2022-10-31 09:58
chadmedoh thats a pretty elegant solution#asahi-gpu · 2022-10-31 09:56
_jannau_chadmed: someone already fitted an exponential on twitter#asahi-gpu · 2022-10-31 09:53
chadmedlina: did you need f(V) to take any particular form?#asahi-gpu · 2022-10-31 08:34
linachadmed: New (hopefully less cursed) data fitting challenge: https://gist.github.com/asahilina/729b99c214ffcceadb396c5f4c4bd7ec ^^#asahi-gpu · 2022-10-31 07:59
phireActually no. I think they are just regular floats which have been chucked into some generic compression algorthm#asahi-gpu · 2022-10-31 02:14
phireinteresting. the TVB seems to have a variable length float format#asahi-gpu · 2022-10-30 04:25
phireoh right, using a retail kernel makes a diffrence#asahi-gpu · 2022-10-29 22:17
jannau'<@lina> jannau: mkdir initdata; python tools/run_guest.py -m hv/trace_agx_pwr.py 12.3/kernelcache.release.mac13j.macho -- "debug=0x14e serial=3 -enable-kprintf-spam wdt=-1"'#asahi-gpu · 2022-10-29 22:16
phirethis time: "assertion failed /AppleInternal/Library/BuildRoots/b19089b6-9308-11ec-ba99-4e67803e3fd9/Library/Caches/com.apple.xbs/Sources/AppleConvergedIPC/Core/ACIPCPort.cpp:105" @ACIPCPort.cpp:105#asahi-gpu · 2022-10-29 22:15
phireyeah#asahi-gpu · 2022-10-29 22:15
jannaumacos panics?#asahi-gpu · 2022-10-29 22:12
phirerunning both tracers just made it panic sooner (diffrent panic?), didn't get as far as agx#asahi-gpu · 2022-10-29 22:08
jannauin combination with trace_agx_pwr? the latter reboots when it is done#asahi-gpu · 2022-10-29 22:05
phiretrace_all_more times out and panics: panic(cpu 2 caller 0xfffffe0014d15fe4): cannot find IOAESAccelerator: timeout = 120 seconds#asahi-gpu · 2022-10-29 20:31
phireLina, just let me wake up#asahi-gpu · 2022-10-29 14:31
jannauphire: change line 76 to "for j in range(4):" for the m1 max#asahi-gpu · 2022-10-29 14:30
jannaulina: at the same place with the obvious s/8/4/ fix#asahi-gpu · 2022-10-29 14:23
linaAnd the f=0 data is proportional (to within the available significant digits) to the leakage table values.#asahi-gpu · 2022-10-29 14:22
linaSo that means there is no per-core shenanigans for dynamic power, and the static and dynamic power are just added together#asahi-gpu · 2022-10-29 14:22
linaThe core leakage/static power data is additive. If I subtract f=0 data from f=1000 data, the total power becomes constant#asahi-gpu · 2022-10-29 14:21
linaOK, so at least some things are adding up...