Wizzup | freemangordon: so I'm running nomweather with LIBGL_ALWAYS_SOFTWARE=1 and the ants disappear | 01:49 |
---|---|---|
Wizzup | I have them running side by side, one native pvr gl, one with sw render | 01:50 |
Wizzup | and the ants/corruption is def. gl related | 01:50 |
freemangordon | Wizzup: Either driver issue or kernel (cache) issue | 07:22 |
freemangordon | in both cases not much I can do about it | 07:23 |
freemangordon | maybe send a mail to AD on pvr ML | 07:24 |
Wizzup | freemangordon: mhm | 11:55 |
arno11 | Wizzup: i made a PR with new changes into libcmtspeechdata (in calls-test-2 branch because it is already quite similar to my config) | 12:42 |
Wizzup | ty! | 12:44 |
arno11 | :) | 12:45 |
Wizzup | I see the email to pavel too | 12:49 |
arno11 | ah ok | 12:52 |
arno11 | Wizzup: sicelo: i made some progress with PA cpu usage | 19:28 |
arno11 | the cpu usage is still very high but sound desync decreases a lot if other apps are running | 19:31 |
arno11 | the trick is to disable high priority and realtime-scheduling in daemon.conf | 19:32 |
rafael2k | hey, is maemo-lest using the alsa-plugin maemo to talk to the DSP in the N900? | 19:33 |
rafael2k | I kind of need urgently write an alsa plugin, and bumped in the maemo plugin for alsa | 19:33 |
rafael2k | neat | 19:33 |
arno11 | hi, dsp is not working iirc | 19:34 |
Wizzup | arno11: interesting that disabling it helps | 19:37 |
arno11 | yes and the other interesting thing is that xorg burns 65% cpu when i use cmt_alsa for calls ! | 19:49 |
arno11 | instead of cmt_pulse | 19:49 |
arno11 | and only with sound in one way | 19:50 |
Wizzup | I really don't understand what X has to do with sound at all | 19:50 |
arno11 | mee too ! | 19:50 |
arno11 | i found nothing about that issue on internet | 19:51 |
arno11 | just one thing: alsa looking for wrong sound card per default | 19:51 |
Wizzup | alsa_cmt or something else? | 19:52 |
arno11 | sorry don't understand | 19:53 |
Wizzup | when you say alsa is looking at the wrong card per default, it's this dictated by ucm? | 19:54 |
arno11 | sorry misunderstanding. i was talking about a possible reason found on the net | 19:55 |
Wizzup | reason for what, X usage? | 19:56 |
Wizzup | or one way sound? | 19:56 |
arno11 | for high cpu usage for both PA and xorg | 19:56 |
Wizzup | hmm | 19:57 |
arno11 | but seems not possible with leste i think because asound shows only one card | 20:00 |
arno11 | 0 snd_soc_rx51 | 20:01 |
arno11 | and nothing else | 20:01 |
Wizzup | yeah | 20:01 |
Wizzup | no, that's not likely it. I would check the Xorg log files, or maybe figure out how we can profile it | 20:02 |
Wizzup | or attach with gdb, but eh... | 20:02 |
Wizzup | can I test this without actually calling? | 20:03 |
arno11 | hm not sure | 20:03 |
arno11 | maybe using pa_test | 20:03 |
arno11 | let me check | 20:04 |
sicelo | maybe it just seems to be X, but could be dbus? wild shot, but who knows ... | 20:06 |
arno11 | ah ok | 20:06 |
sicelo | it's just a guess ... don't take me too seriously on this one :-) | 20:07 |
sicelo | but i'll be 'happy' to be correct, heh | 20:07 |
arno11 | ok lol | 20:07 |
Wizzup | sicelo: it might be some interaction for sure yes | 20:07 |
Wizzup | like something repeatedly trying to do some action | 20:07 |
arno11 | so weird: no troubles using pa_test | 20:07 |
Wizzup | what does pa test do differently | 20:08 |
arno11 | excepting dbus and ofono links, nothing | 20:08 |
arno11 | ah sorry i think i'm wrong | 20:10 |
arno11 | let me check in cmt | 20:10 |
Wizzup | it might be worth checking dbus-monitor and dbus-monitor --system | 20:12 |
Wizzup | see if it is super spammy | 20:12 |
Wizzup | or maybe strace Xorg, see if it tries to do some system calls repeatedly that fail with say permission error | 20:12 |
arno11 | ok | 20:13 |
Wizzup | I can try to reproduce it on my n900, have to do some work first though | 20:13 |
arno11 | ok. i found the huge difference between pa_test and cmt | 20:13 |
arno11 | pa_test uses only one PA daemon instead of 3-4 for cmt_pulse or cmt_alsa | 20:14 |
Wizzup | I still can't believe we ever need more than one PA | 20:15 |
Wizzup | this is something I really don't get | 20:15 |
Wizzup | everything should go through one PA | 20:15 |
Wizzup | why are multiple servers started? | 20:15 |
arno11 | because cmt_pulse is split in 4 different files/progs | 20:16 |
Wizzup | ok, but those can just be four separate pulse clients? | 20:17 |
Wizzup | talking to the same server, right? | 20:17 |
arno11 | not sure, htop shows 4 different daemons. for me it is 4 different servers | 20:18 |
Wizzup | ok, that's very likely a problem regardless of X | 20:18 |
Wizzup | this might also be why you see that removing realtime helps | 20:19 |
Wizzup | if they all want to be realtime... | 20:19 |
arno11 | makes sense | 20:19 |
Wizzup | but, if the server is running, another one should never be spawned | 20:19 |
Wizzup | is this some env difference or something | 20:19 |
Wizzup | I'm really confused :D | 20:19 |
arno11 | me too | 20:19 |
sicelo | we may have to also look at cgroups ... at least fremantle used cgroups heavily for call handling | 20:20 |
arno11 | maybe we can ask Pavel why he has split cmtspeech_ofono_test... | 20:21 |
arno11 | because it was just one prog at the beginning | 20:21 |
arno11 | iirc | 20:22 |
Wizzup | cgroups exist on fremantle? | 20:22 |
Wizzup | I didn't know it did back in 2009 / 2.6 era | 20:22 |
rafael2k | https://github.com/alsa-project/alsa-plugins/tree/master/maemo | 20:30 |
rafael2k | interesting to read | 20:31 |
sicelo | yes cgroup support was already there in fremantle. see the bootlog here, https://wiki.maemo.org/N900_Hardware_Hacking/serial_dump | 20:51 |
Generated by irclog2html.py 2.17.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!