#asahi-dev · 2023-02-10 14:35
poviki started implementing epic in kernel's dcp driver to get to the audio interface#asahi-dev · 2023-02-09 12:17
povikyeah, <= makes sense#asahi-dev · 2023-02-09 12:11
povikshould that assert be there?#asahi-dev · 2023-02-09 12:11
povikthis doesn't look right#asahi-dev · 2023-02-09 12:11
povikhttps://github.com/AsahiLinux/m1n1/blob/main/src/afk.c#L342#asahi-dev · 2023-02-08 16:19
povikhow come a trace script accesses the dart at all?#asahi-dev · 2023-02-08 12:16
poviki remember there could be offsets applied in bypass mode but you probably checked for those#asahi-dev · 2023-02-08 11:18
povik(i'm just guessing here, no hard numbers at hand)#asahi-dev · 2023-02-08 11:17
poviki assume the number of events is much lower and we spend many irqs per event#asahi-dev · 2023-02-08 11:17
povikwouldn't there be significant overhead in handling the 10k/s irqs?#asahi-dev · 2023-02-08 11:00
povikother than that... :P#asahi-dev · 2023-02-08 11:00
povikor is there something else we can do to push the irq number down?#asahi-dev · 2023-02-08 10:59
povikdo we want to bring up sio to address that?#asahi-dev · 2023-02-08 10:59
povikhah#asahi-dev · 2023-02-07 10:21
poviklet's call the yesterday's form an approximation for tc_vc/tc_mag << 1 :p#asahi-dev · 2023-02-07 09:04
povik^-- this is the fixed eta#asahi-dev · 2023-02-07 09:03
povik eta = 1 / (1 - tc_vc/tc_mag)#asahi-dev · 2023-02-07 09:03
povikmade a mistake in the catchup function yesterday#asahi-dev · 2023-02-06 14:23
povikmaybe we can just close the sense pcm from the kernel side, if the rate no longer matches?#asahi-dev · 2023-02-06 14:23
povikmaybe we can relax the rate symmetry condition, but then the safety daemon needs to know the rate changed#asahi-dev · 2023-02-06 14:21
povikbut if sound servers indeed do reopen pcms to change rates, then that race is an issue#asahi-dev · 2023-02-06 14:21
povikyeah, i was wondering if the current solution isn't good enough already#asahi-dev · 2023-02-06 14:19
poviksince that would constrain the rates on the playback pcm#asahi-dev · 2023-02-06 14:19
povikwe still wouldn't want the safety daemon to keep the sense pcm open unnecessarily#asahi-dev · 2023-02-06 14:18
povikbut even if this weren't the case#asahi-dev · 2023-02-06 14:18
povikso that's an artifact of the way it's handled with the separate frontends#asahi-dev · 2023-02-06 14:17
povikif you open the pcm device without it, it's dead, there's no samples#asahi-dev · 2023-02-06 14:17
povikthe tx path needs to be up for the sense rx path to work#asahi-dev · 2023-02-06 14:17
povikand to my surprise you haven't ran into it#asahi-dev · 2023-02-06 14:17
povikyeah, there's one thing i haven't pointed out#asahi-dev · 2023-02-06 14:15
poviki hope i made my calculus professor proud :p#asahi-dev · 2023-02-06 14:14
povik return (A + B*eta, B)#asahi-dev · 2023-02-06 14:14
povik B = np.exp(-t/tc_mag) * T_mag#asahi-dev · 2023-02-06 14:14
povik A = np.exp(-t/tc_vc) * (T_vc - eta*T_mag)#asahi-dev · 2023-02-06 14:14
povik eta = 1 + tc_vc/tc_mag#asahi-dev · 2023-02-06 14:14
povik # All temperatures relative to ambient#asahi-dev · 2023-02-06 14:14
povikdef catchup(T_vc, tc_vc, T_mag, tc_mag, t):#asahi-dev · 2023-02-06 14:14
povikhm, still doesn't do newlines#asahi-dev · 2023-02-06 14:14
povikdef catchup(T_vc, tc_vc, T_mag, tc_mag, t): # All temperatures relative to ambient#asahi-dev · 2023-02-06 14:13
povikhuh, that didn't work#asahi-dev · 2023-02-06 14:13
povikdef catchup(T_vc, tc_vc, T_mag, tc_mag, t): # All temperatures relative to ambient eta = 1 + tc_vc/tc_mag A = np.exp(-t/tc_vc) * (T_vc - eta*T_mag) B = np.exp(-t/tc_mag) * T_mag return (A + B*eta, B)#asahi-dev · 2023-02-06 14:13
povikokay, got it#asahi-dev · 2023-02-06 13:57
povikyeah, that would be my guess too#asahi-dev · 2023-02-06 13:55
poviknot sure how large the introduced error is though#asahi-dev · 2023-02-06 13:54
povikthe tc_vc relaxation should be raising t_mag a bit#asahi-dev · 2023-02-06 13:54
povikhm, maybe the update step isn't exactly right since if t_mag=t_ambient and p=0, t_mag never raises even if t_vc is initially high#asahi-dev · 2023-02-06 13:29
povikyeah#asahi-dev · 2023-02-06 13:27
povikor however it's spelled...#asahi-dev · 2023-02-06 13:27
povikmarcan: bon apetite!#asahi-dev · 2023-02-06 13:26
povikright