#
✕ nick:chadmed_
#asahi-dev · 2023-01-12 08:52
chadmed_its basically dhrystone but better at measuring "real" performance (still not great)
#asahi-dev · 2023-01-12 08:46
chadmed_kh: as above, i used coremark
#asahi-dev · 2023-01-11 10:28
chadmed_i can do the runs again if youd like
#asahi-dev · 2023-01-11 10:27
chadmed_then took the average coremark score from these runs to get coremarks/mhz then normalised that for the ecores based on the dmips/mhz already in the DT
#asahi-dev · 2023-01-11 10:27
chadmed_i did 100 runs of coremark on each type of core for each pstate in a singleuser environment with no superfluous services started and everything that isnt coremark masked off the core running coremark
#asahi-dev · 2023-01-11 10:26
chadmed_which are like, the worst things you could possibly stall on
#asahi-dev · 2023-01-11 10:26
chadmed_kh: they were slightly wrong based on my testing, and the new values fix some seriously bad stalling on rendering and pipewire threads
#asahi-dev · 2023-01-10 13:31
chadmed_j`ey: thanks, ill sort that tomorrow. 707 is the correct value
#asahi-dev · 2023-01-10 12:53
chadmed_(specifically testing on t8103 and t6002 would be nice)
#asahi-dev · 2023-01-10 12:53
chadmed_https://github.com/chadmed/linux/tree/eas-v2 <-- EAS patches to asahi-wip if anyone's interested in testing. it actually works for real this time :P
#asahi-dev · 2023-01-09 12:13
chadmed_finally a win with EAS
#asahi-dev · 2023-01-09 12:13
chadmed_playing a 16/44.1 flac resampled to 96kHz with the pipewire dsp: 3.78 W/7.30 W
#asahi-dev · 2023-01-09 12:12
chadmed_idle core power: 3.57 W, idle total system power: 6.59 W
#asahi-dev · 2022-12-06 11:07
chadmed_also your name makes me hungry :P
#asahi-dev · 2022-12-06 11:07
chadmed_it might be worth investigating at some point just as an exercise in curiosity
#asahi-dev · 2022-12-06 10:31
chadmed_e.g. some work has been done on AVD, AVE and APR but no ones actively looking at it right now
#asahi-dev · 2022-12-06 10:30
chadmed_yeah. for some things there is some prior art, but it means work has either stalled or not begun
#asahi-dev · 2022-12-05 14:09
chadmed_could you coax the jpeg block to do mjpeg :P
#asahi-dev · 2022-12-04 07:35
chadmed_i have sorted the last of the over-excursion on big bass hits
#asahi-dev · 2022-12-04 07:34
chadmed_im reasonably certain that the pipewire DSP graph is 100% safe at 18.5 dBV on j314 now
#asahi-dev · 2022-12-01 08:06
chadmed_i had an idea for a multi-level failsafe and part of that was being able to lock/unlock elems at will
#asahi-dev · 2022-12-01 08:05
chadmed_huh so i can read and write values to the elems but still cant lock or unlock which sucks
#asahi-dev · 2022-11-30 11:01
chadmed_the alsa-lib rust bindings dont have the methods to change or get current values out of mixer controls hooked up though :/
#asahi-dev · 2022-11-30 11:00
chadmed_progress!
#asahi-dev · 2022-11-30 11:00
chadmed_Right Tweeter Speaker Volume range: (MilliBel(-9999999), MilliBel(0))
#asahi-dev · 2022-11-27 13:17
chadmed_povik: works a treat
#asahi-dev · 2022-11-27 12:44
chadmed_so have like a tas2764_hw_{suspend,resume} that deals with the chip and then use the asoc suspend/resume just to save and restore the regs?
#asahi-dev · 2022-11-27 12:40
chadmed_another 1UP over macos perhaps :D
#asahi-dev · 2022-11-27 12:39
chadmed_do we know if macos puts them into hw shutdown?
#asahi-dev · 2022-11-27 12:38
chadmed_we have the benefit of not officially supporting speakers yet (and probably for a while given we havent even started working on userspace safety) so theres time to work it out
#asahi-dev · 2022-11-27 12:33
chadmed_but sw shutdown quiescent current is 10x hw shutdown on two voltage rails so we really should get it working properly
#asahi-dev · 2022-11-27 12:33
chadmed_aiui asoc spawns a tas2764 driver for each codec, so option 1 (the good option) would be quite hard
#asahi-dev · 2022-11-27 12:28
chadmed_so i take it that means pulling SDZ low will never work on suspend for us?
#asahi-dev · 2022-11-27 11:31
chadmed_resetting them before trying to restore the registers doesnt help
#asahi-dev · 2022-11-27 11:30
chadmed_it looks like the chips just disappear off the bus. the regcache sync fails with -ENXIO, and tas2764->sdz_gpio doesnt exist upon resume...
#asahi-dev · 2022-11-22 14:09
chadmed_hotplug/switching works in the default stereo audio profile
#asahi-dev · 2022-11-22 14:09
chadmed_ive been dogfooding pipewire + DSP et al for ages and it all works with no issues
#asahi-dev · 2022-10-30 13:24
chadmed_given that we dont actually populate the regs under any error condition, we should just check for all three arrays and free any that were allocated on all error paths to guarantee we dont leave anything behind
#asahi-dev · 2022-10-30 13:23
chadmed_my thinking is that since theres no longer a hierarchy of props, any of them could be allocated out of order and having individual ops to error out of parsing them could leave some non-freed memory behind depending on the order they were parsed/allocated
#asahi-dev · 2022-10-30 13:22
chadmed_instead of having free_*, we condense them into one operation that just frees any of the arrays that were allocated
#asahi-dev · 2022-10-30 13:16
chadmed_in that case would it be better to have one error: label which checks all three and cleans up ones that have been allocated instead of three discrete ones
#asahi-dev · 2022-10-30 13:14
chadmed_oh of course that makes sense
#asahi-dev · 2022-10-30 13:10
chadmed_the previous code just returned straight out of the branch like i had it before
#asahi-dev · 2022-10-30 13:10
chadmed_yeah so why do we need to jump to free_* in branches where theyre not allocated
#asahi-dev · 2022-10-30 13:09
chadmed_and they should not be allocated
#asahi-dev · 2022-10-30 13:09
chadmed_if !vcount, !icount or !pcount then theres no data to populate so we dont need those arrays
#asahi-dev · 2022-10-30 13:08
chadmed_theyre only there to be indexed into for populating the regulator struct
#asahi-dev · 2022-10-30 13:08
chadmed_if ive written the code correctly they should never be allocated under such conditions
#asahi-dev · 2022-10-30 13:03
chadmed_same goes for microamp and microvolt, theyre only allocated iff icount/vcount are valid, so if we make it to free_* then we know that i/vcount arent 0
#asahi-dev · 2022-10-30 13:01
chadmed_yeah

plus ›