| greyback__ | racarr_: hey, install cmake-extras - it has the gmock module that qtmir uses | 00:37 |
|---|---|---|
| racarr_ | o | 00:37 |
| racarr_ | greyback__: Does that have the EnableCoverageReport stuff that I proposed in another mp too? | 00:37 |
| greyback__ | racarr_: yes | 00:37 |
| racarr_ | Ah :) | 00:38 |
| racarr_ | cool | 00:38 |
| racarr_ | greyback__: Thanks :) | 00:38 |
| greyback__ | no worries | 00:38 |
| === duflu_ is now known as duflu | ||
| duflu | Should it be possible that clang generates code that yields SIGILL??! | 10:02 |
| duflu | Oh, apparently clang does emit this invalid instruction intentionally sometimes. Wonder why gcc works | 10:04 |
| alan_g | targetting the wrong processor? | 10:07 |
| Kaleo | hey guys | 10:08 |
| duflu | alan_g: No, it's the right architecture. But apparently clang sometimes uses invalid instructions like an abort() | 10:08 |
| duflu | Kaleo: Hello | 10:08 |
| Kaleo | is there any difference in MIR's rendering or behaviour whether the phone is plugged on USB or not/charging or not? | 10:08 |
| duflu | Kaleo: Not explicitly in Mir but the driver could (and likely would) use a lower power mode on battery alone | 10:09 |
| Kaleo | right | 10:09 |
| duflu | Kaleo: You see such things? | 10:09 |
| Kaleo | duflu, one thing I notice for example | 10:09 |
| Kaleo | duflu, is that the lock screen's background sometimes takes time to be displayed | 10:09 |
| Kaleo | duflu, probably for the texture to be uploaded or rendered | 10:10 |
| Kaleo | duflu, only on battery | 10:10 |
| duflu | Kaleo: Yeah lower power mode is noticeably slow | 10:10 |
| duflu | It gets worse... if you're careful with your efficiency then the phone thinks it's in low power mode and can clock down too much, causing stuttering | 10:11 |
| Kaleo | duflu, interesting | 10:11 |
| duflu | It drives graphics people crazy | 10:11 |
| Kaleo | duflu, can you think of any reliable way to trigger that mode? | 10:11 |
| duflu | Kaleo: No, it's driven by how little you stress the GPU | 10:12 |
| Kaleo | duflu, does the low power mode impact CPU perf as well? | 10:13 |
| duflu | Kaleo: I think so..? | 10:13 |
| Kaleo | k | 10:13 |
| Kaleo | duflu, I believe this all relates to https://bugs.launchpad.net/camera-app/+bug/1398436 | 10:13 |
| ubot5 | Launchpad bug 1398436 in camera-app "Preview frozen after unlocking" [Critical,In progress] | 10:13 |
| duflu | Kaleo: There could be other bugs at play in our code, always. I'm reminded of this talk where I think John Carmack discusses the same issue driving him crazy: https://www.youtube.com/watch?v=nqzpAbK9qFk | 10:14 |
| Kaleo | duflu, ok | 10:15 |
| Kaleo | duflu, concretely what happens is that after locking and unlocking the phone the camera app's rendering is not happening for a few seconds | 10:15 |
| duflu | Kaleo: Oooh, you don't have framedropping (eglSwapInterval(0)) do you? | 10:15 |
| Kaleo | duflu, I don't have anything outside of the pristine RTM image | 10:15 |
| duflu | Kaleo: If a client has set eglSwapInterval(0) (which I think is unlikely) then we do have buffer scheduling bugs that will appear as a freeze | 10:16 |
| duflu | (fixed in Mir 0.10) | 10:16 |
| Kaleo | duflu, is MIR 0.10 in vivid? | 10:16 |
| duflu | Kaleo: No, it's future | 10:16 |
| Kaleo | duflu, ok | 10:16 |
| Kaleo | duflu, I don't recall the app setting eglSwapInterval(0) | 10:17 |
| duflu | Kaleo: There is one additional bug that will trigger such a freeze, but it's only known to happen in cases that the phone won't trigger | 10:17 |
| duflu | Still, I expect both to be fixed in 0.10 | 10:17 |
| duflu | Kaleo: I'll grep the camera-app source in case | 10:19 |
| Kaleo | duflu, done that, no match | 10:20 |
| Kaleo | duflu, but there is something more | 10:20 |
| duflu | Kaleo: Also, we only know that Android overlays are highly unstable with eglSwapInterval(0). This does not guarantee they're always going to be stable with the default eglSwapInterval(1), just moreso | 10:20 |
| Kaleo | duflu, for the duration of the freeze, the app is not notified it is focused/active (Qt.application.active is still false) | 10:21 |
| Kaleo | duflu, even though the process has been correctly resumed | 10:21 |
| duflu | Kaleo: To absolutely verify it's not another overlays bug, try running with env MIR_SERVER_DISABLE_OVERLAYS=true | 10:23 |
| Kaleo | duflu, ok, trying that | 10:23 |
| duflu | ... for USC | 10:23 |
| duflu | Kaleo: And then I must go for the week ... | 10:23 |
| Kaleo | USC? | 10:23 |
| duflu | Kaleo: Only unity-system-compositor will use the hardware overlays | 10:23 |
| duflu | So the environment needs setting for that | 10:24 |
| Kaleo | duflu, I can set it in /etc/environment no? | 10:24 |
| duflu | Kaleo: Yes, but make sure it sticks after rebooting (/proc/<pid>/env) | 10:24 |
| Kaleo | duflu, yep | 10:24 |
| duflu | Kaleo: Please also double-check the whole stack of projects to ensure nothing is triggering eglSwapInterval(0) | 10:26 |
| Kaleo | duflu, I grepped it all and only qtubuntu sets it | 10:26 |
| Kaleo | duflu, if an env var is set | 10:26 |
| duflu | ... or creating a GL context with zero | 10:26 |
| Kaleo | otherwise it defaults to 0 | 10:26 |
| Kaleo | duflu, bug still happens with MIR_SERVER_DISABLE_OVERLAYS=true | 10:27 |
| duflu | Interesting | 10:27 |
| duflu | Kaleo: I think the clock-down power management stuff is a big offender but that will only cause you to miss a frame or two in my experience | 10:27 |
| Kaleo | k | 10:28 |
| * duflu forgets what the kernel features are that stop it from happening. It's something Linux has inherited from Android recently-ish but is not yet widely used | 10:28 | |
| Kaleo | duflu, note the rest of unity8 behaves normally during the freeze | 10:28 |
| Kaleo | duflu, everything is fluid and responsive | 10:28 |
| duflu | Kaleo: I'm out of ideas | 10:28 |
| Kaleo | duflu, ok | 10:28 |
| duflu | and the work week | 10:28 |
| Kaleo | duflu, thank you | 10:28 |
| duflu | No worries... | 10:29 |
| === greyback__ is now known as greyback | ||
| === dandrader is now known as dandrader|afk | ||
| === dandrader|afk is now known as dandrader | ||
| === vesar_ is now known as vesar | ||
| === dandrader is now known as dandrader|lunch | ||
| === dandrader|lunch is now known as dandrader | ||
| === alan_g is now known as alan_g|EOW | ||
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!