search

overte-org / overte

Overte open source virtual worlds platform.
https://overte.org/
Other
130 stars 47 forks source link

Low AppImage performance #874

Closed ksuprynowicz closed 3 months ago

ksuprynowicz commented 3 months ago

It looks like AppImage performance is very poor again. On local RelWithDebInfo build I'm getting 7 times the performance of AppImage, which means that most possibly AppImage is built with -O0 again?

JulianGro commented 3 months ago

I cannot reproduce this on latest master. (Left is newest release; right is master) Auswahl_162

Here is the AppImage I created a couple of hours ago: https://data.moto9000.moe/Overte-2023.11.1-173-g3d5afbdd31-nodebug-x86_64.AppImage (The filename presumably comes from the AppStream data. This is freshly cloned master.)

ksuprynowicz commented 3 months ago

For me in Overte Hub on master build engine time is about 10 ms and on Appimage I built it goes up to 130 ms, and sometimes 250 ms, resulting in framerates as low as 3 FPS.

JulianGro commented 3 months ago

I cannot actually join Overte Hub with a master AppImage, so I didn't test the Hub. It crashes on joining.

ksuprynowicz commented 3 months ago

Thanks! I'll investigate this today

JulianGro commented 3 months ago

Thanks! I'll investigate this today

You can download the debug version as well: https://data.moto9000.moe/Overte-2023.11.1-173-g3d5afbdd31-debug-x86_64.AppImage It looks like a deadlock, or rather humongous slowdown, related to MultiSphereShape::computeSpheres, though I thought that this was already fixed.

ksuprynowicz commented 3 months ago

It seems that the performance drop is caused by an unrelated bug in OpenGL drivers. The appimage itself is fine, and restrting system fixes this temporarily. I'll close this issue and open a new one if I'm able to find the cause of performance drop.

You can download the debug version as well: https://data.moto9000.moe/Overte-2023.11.1-173-g3d5afbdd31-debug-x86_64.AppImage It looks like a deadlock, or rather humongous slowdown, related to MultiSphereShape::computeSpheres, though I thought that this was already fixed.

I tested it on newest master appimage and can't replicate the issue. Maybe it's avatar-related?

JulianGro commented 3 months ago

I tested it on newest master appimage and can't replicate the issue. Maybe it's avatar-related?

I forgot about that. We meant to debug this last time, but I ended up being busy with other stuff.