Enable Linux Feral Interactive's GameMode for systems that have it installed

[dslul]

Recently a Linux gamemode daemon/library has been announced by Feral Interactive. Upcoming games (such as Tomb Raider) will use it. It is quite straightforward to use, just an include in the project. Systems that have the library installed will use it to improve performance by adjusting the gpu governor.

[elexis]

​https://github.com/FeralInteractive/gamemode

You have it installed? If so, it would be useful to show a performance difference. Since it's for the CPU and not GPU, the benefits would be mostly in the simulation side, i.e. the pathfinder and can be measured well with the non-visual replay graphs wiki:EngineProfiling#Simulationreplaymode

[dslul]

profiling without gamemode

[dslul]

profiling with gamemode

[Stan]

Sounds good. Platform specific improvements are always nice especially if it's doesn't clutter too much the code.

Someone will have to test it on OSX too. Maybe Vlad can help with that.

Patches should be uploaded to Phabricator see SubmittingPatches

I believe we will need some preprocessor defines for windows.

[elexis]

The replay is too short and there is not enough stuff going on to be conclusive. It should look something like the one here #5099. I can upload a replay after the next svn lobby game.

[dslul]

profiling without gamemode on a longer replay

[dslul]

profiling with gamemode on a longer replay

[elexis]

I see some considerable differences in the two profiling.txt, but the replay still meaningful enough, the graphs look identical. It should be a serious lag replay, 8 players * 200 units or just the average Jebel Barkal or Danubius game with gaia spam. As mentioned I might upload such a game in the next 24-48h if the lobby players are with me.

[elexis]

In 21694:

Ease comparison of two replay profile graphs by typing the filename in the browser instead of renaming the data.json files, refs #5116.

[elexis]

laggy replay for r21690

[dslul]

Profiling without gamemode on the "laggy replay"

[dslul]

profiling with gamemode on the "laggy replay".I'll try with a lower end system, where cpu can be a bottleneck (I tried on a i7 3630qm)

[elexis]

10ms sounds very strange to me, I would rather expect 200ms to 500ms. Did you replay it with the correct revision? Otherwise it might be OOS and be equal to nothing happening there. It should say hash mismatch in the commandline output and the visual replay should show a message box error equally.

This might also explain why there is no measurable difference. Another hypothesis why we can't observe a measurable difference is because the patch doesn't actually work, or because all CPU resources are already allocated.