| gnarface | (this would in theory accrue a slightly higher CPU load but as we have seen elsewhere, not as high as the CPU load of having to constantly calculate transparency of multiple layers without the aid of a dedicated hardware feature in the GPU) | 00:00 |
|---|---|---|
| arno11 | yeah, at least now, things stay smooth even with 4 or 5 qt apps loaded | 00:02 |
| arno11 | included a heavy app like telegram-desktop | 00:02 |
| arno11 | not too bad for n900 :D | 00:03 |
| gnarface | yea, for sure | 00:03 |
| arno11 | i'll open an issue tomorrow to keep a trace | 00:08 |
| arno11 | thx for details and clarifications btw | 00:11 |
| gnarface | no problem... i wonder if qt5 stuff can still be built against gles2 | 00:11 |
| gnarface | google suggests the n900 can do gles2 in hardware, and also on the pinephones we've seen that stuff that migrated away from gles2 support took a severe performance hit | 00:12 |
| gnarface | in theory that might get your performance back without losing rendering features | 00:12 |
| gnarface | i think it depends though on whether the fundamental qt code has been changed too much to be able to use it | 00:13 |
| gnarface | (specifics of this are over my head) | 00:13 |
| gnarface | in theory also, you might just be able to force an earlier opengl version with a mesa environment variable, but so far i don't think we've seen anything coded in such a way that actually works | 00:15 |
| arno11 | ok, thx again. time to zzz | 00:25 |
| arno11 | Wizzup: ok, so qt5 issues have been completely solved on my device with daedalus: no more segfault, slowness or weird warnings msgs. conversations display is fine after reboot. | 09:21 |
| arno11 | for the record: we need appmenu-gtk2-module and qt5ct pkgs, few env var and particular settings in qt5ct. | 09:23 |
| arno11 | env vars: QT_STYLE_OVERRIDE=qt5ct-style and QT_QPA_PLATFORMTHEME=qt5ct | 09:26 |
| arno11 | for qt5ct: style: maemo5, Standard dialogs: GTK2, Troubleshooing: Force raster surface | 09:29 |
| arno11 | *Troubleshooting | 09:30 |
| arno11 | ah...conversations main window is ok but chat windows show the main window. weird. all other apps seem ok | 09:43 |
| arno11 | hmm even if i revert my changes, conversations chat windows are buggy | 09:59 |
| arno11 | dsc_: ^^^ | 10:00 |
| arno11 | in fact i didn t test conversations after very last update | 10:01 |
| dsc_ | arno11: im silent in this channel but been actually quite busy with a large refactor of conversations | 10:05 |
| dsc_ | as for the complaints, kindly state the version :) `apt policy conversations` | 10:05 |
| arno11 | ok NP | 10:07 |
| arno11 | 0.7.27.1+4m7 | 10:07 |
| arno11 | under daedalus | 10:07 |
| dsc_ | arno11: im inclined to think this is related to rendering as you mention the wrong content is being shown | 10:12 |
| dsc_ | but just speculation | 10:13 |
| arno11 | ok | 10:14 |
| Wizzup | dsc_: 7.27.1 is a revert to an earlier version | 10:28 |
| Wizzup | arno11^ | 10:28 |
| Wizzup | I reverted it to the pre-freeze version for stable | 10:29 |
| arno11 | conversations is ok after reboot and no qt5ct changes. everythings works fine if i do the changes after boot. | 10:42 |
| Wizzup | so what is this qt5ct change for? | 10:42 |
| arno11 | see logs: it speeds up launching time a lot | 10:43 |
| Wizzup | interesting, do you know why? | 10:43 |
| arno11 | not exactly | 10:43 |
| arno11 | the change which makes conversations not happy on boot is the force raster surface option | 10:46 |
| arno11 | this option is the 'game changer' on n900 lol | 10:49 |
| dsc_ | arno11: so this qt5ct, you used it to override the maemo Qt5 integration, and now it loads faster? | 10:49 |
| arno11 | a lot faster yeah | 10:50 |
| Wizzup | well, but he sets it back to the regular integration | 10:51 |
| Wizzup | the maemo one | 10:51 |
| arno11 | yes | 10:51 |
| arno11 | as i said, the raster surface option is the main trick | 10:52 |
| arno11 | and it is the cause of chat window rendering issue. if i activate the option after boot, no issue so far | 10:55 |
| arno11 | *with conversations | 10:55 |
| Wizzup | does it come at some performance cost? | 10:59 |
| Wizzup | nothing visible I guess? | 11:04 |
| arno11 | no performance cost, memory usage is divised by 2 | 11:08 |
| Wizzup | interesting | 11:09 |
| arno11 | as i said even with tg-desktop running, i m able to launch and use several other qt apps at the same time with no troubles | 11:10 |
| Wizzup | so it's the raster surface thing | 11:13 |
| Wizzup | I wonder if that's also an env var | 11:13 |
| Wizzup | I think it should be | 11:13 |
| arno11 | yeah, i have to go but will try different configs this evening and let you know | 11:15 |
| arno11 | (at least we need qt5ct env vars to make qt5ct overriding settings working) | 11:16 |
| dsc_ | arno11: so you disabled GPU hardware accel | 11:18 |
| dsc_ | and now it works (c)(tm) | 11:18 |
| Wizzup | I don't think that's necessarily true | 11:19 |
| arno11 | i didn t do anything @gpu | 11:19 |
| dsc_ | just trying to understand what changed | 11:19 |
| arno11 | bbl | 11:21 |
| Wizzup | force raster does sound like it might disable opengl | 11:21 |
| Wizzup | yeah bbiab | 11:21 |
| arno11 | Wizzup: yeah, it uses raster graphics system instead of opengl (according to qt.io, it is indeed a way faster because of shared memory management) | 15:12 |
| arno11 | but there is not too much doc on the web about that functionality | 15:15 |
| dsc_ | cpu rasterization should not be faster | 15:16 |
| dsc_ | gpu goes brrr | 15:16 |
| freemangordon1 | I wonder if this is related to compiling shaders in $HOME and launching the tracker | 15:18 |
| arno11 | on my device, tracker is blocked btw | 15:19 |
| arno11 | and even with no tracker it is really slow without qt5ct stuff | 15:20 |
| freemangordon | because it compiles shaders on startup | 15:20 |
| freemangordon | (perhaps) | 15:20 |
| freemangordon | how is tracker blocked? | 15:21 |
| arno11 | i chmod'd -x tracker3 | 15:24 |
| dsc_ | I think init.d (or whatever manages processes) will go into a loop that way | 15:25 |
| dsc_ | not sure about timeout | 15:26 |
| arno11 | i don t see any troubles | 15:28 |
| arno11 | there is no init.d or xssession for it btw | 15:29 |
| arno11 | anyway there is a specific command to block it iirc | 15:30 |
| arno11 | freemangordon: the 'main' surprise with raster stuff is memory usage (x1.5 or x2 less) | 15:37 |
| freemangordon | how do you measure memory usage? | 15:40 |
| arno11 | just with top/htop (ok that's not ideal but gives an idea) | 15:41 |
| arno11 | anyway i can launch lot of qt stuff at the same time now without a huge swaping | 15:42 |
| arno11 | i m currently running telegram-desktop, hamsterfiler, qalendar and playing a mp4 on smplayer at the same time | 15:51 |
| arno11 | and still able to use irssi to write stuff | 15:51 |
| arno11 | no overclock | 15:51 |
| arno11 | htop returns 125MB of ram usage and 290 of swap (but 150 only for tg-desktop btw) | 15:53 |
| arno11 | funny that the last tg-desktop version works fine on n900 and daedalus btw. this app is incredibly well optimized | 16:09 |
| gnarface | arno11: hey, i just remembered something that might be important about that hardware opengl feature i mentioned yesterday; it might not actually be a good idea because the n900's GPU might generate too much heat doing it, so thereby might stress the hardware and shorten its life | 22:25 |
| gnarface | (i don't remember for sure it's the case, i just suddenly remembered vaguely someone might have said that, or i might be conflating it with a separate issue) | 22:25 |
| gnarface | so maybe you're better off just disabling it like you have already | 22:25 |
| gnarface | rather than seeking to get everything using Opengl ES 2.0 | 22:25 |
| gnarface | (i could be wrong) | 22:25 |
| gnarface | someone else around here would know for sure | 22:27 |
| gnarface | i haven't read the scrollback yet today so maybe that's already been covered... | 22:28 |
| arno11 | ok thx for info. atm nobody knows exactly what's going on but the raster option seems to disable opengl with qt5 | 22:30 |
| arno11 | and use raster instead | 22:30 |
| gnarface | so there's still some question about whether the blame really lies with QT, Mesa, or the GPU driver itself, or some combination of all of the above | 22:34 |
| gnarface | might not be trivial to actually diagnose fully | 22:35 |
| arno11 | yes indeed | 22:35 |
| gnarface | the thing is, late last night i somehow remembered talking to someone in this channel who already knew exactly what was going on, i just can't remember who | 23:08 |
Generated by irclog2html.py 2.17.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!