Engine Testing - 3. Dec 2012 (91.0.1-566-g3bf8eb3)

[jK]

No, OpenAL is a left dependency same as SDL, GLU & GL.
There is no nice way to remove that.

[Beherith]

First the map was glowing like crazy (Tabula v4 I can post a screen-shot when I get home)
Secondly the small bushes were drown purple. The trees on the other head were ok.
I did not try it with other map yet, I'll check that first and then open a bug.

The biggest issue is with performance on the MT executable.
I play a 1v1 with a kAIK AI.
At first I got 200+ fps but after 2-3 minutes of play the frame rate started fluctuating from 200+ to 15-16 fps for half a second and then back to 200+ and then down again.

Please post a screenshot to mantis, as tabula v4 works fine for me. Does the fps drop happen in multiplayer too, not just vs AI?

[albator]

-> Speed increase is about x10 when you increase game speed in late game to x20: that is huge improvement for joining late game
-> I did not notice huge fps increase since I play with low graphics and I was always > 20 fps in all game already.
-> pathing updated faster solve the never ending problem about the reclaimed DT preventing the unit to go though part of the map: I notice an about 10 sec update. That works both way.

Would not is be possible to go even below 10sec ?


i5 750

[jK]

Updated version can be found on the download page. Should now run under Windows.

[xyz]

Please post a screenshot to mantis, as tabula v4 works fine for me. Does the fps drop happen in multiplayer too, not just vs AI?

here is how tabula looks for me:

DSD renders ok though.

The slowdowns only happen with the multi-thread build. the single threaded build never dropped below 200 fps. One thing I have noticed on Xonotic is that the fps fluctuates a lot if I do not enable the "Wait for GPU to finish every frame". Also in single threaded the CPU rarely jumps over 60%, I only managed to max it out when I set the game speed to 20 (see:
,
while in OMP 2 cores are 100% all the time (see
.
As for trying a multi-player game, I do not have a lobby yet.

Thanks.

[zerver]

The slowdowns only happen with the multi-thread build. the single threaded build never dropped below 200 fps.

I'm not 100% up to date wrt the status of MT in the develop branch due to lack of time.

Does this one have the same problem?

http://springrts.com/dl/buildbot/defaul ... ab8d75.exe

[xyz]

I'm not 100% up to date wrt the status of MT in the develop branch due to lack of time.

Does this one have the same problem?

http://springrts.com/dl/buildbot/defaul ... ab8d75.exe

Hmm .exe what a "strange" extension, is this coming from this git branch? I will have to build it for linux, or do you have a linux portable?

As an FYI I did a test with BOINC running 8 tasks and playing portal on top of it. The game ran smoothly.

[zerver]

Yes, from that branch. I think the branch has some fixes in it that are not in develop. Could also be some conflict between OMP and MT.

[xyz]

If my git skills are working "git checkout -b MTsim" then this branch has the same issues.

[hoijui]

if you do something in git, and you are not sure if it worked or where you are right now, use gitk!
gitk shows you where you currently are, and also all commits you additionally specify on the command line (could be tags, branches, SHA1s or other stuff).
for example, use this:

gitk MTsim 91.0.1-426-g8ab8d75

[Silentwings]

Will this: http://springrts.com/mantis/view.php?id=3356 end up being fixed or are the new larger automatically generated colvols here to stay?

[Deadnight Warrior]

Will this: http://springrts.com/mantis/view.php?id=3356 end up being fixed or are the new larger automatically generated colvols here to stay?

BA uses a gadget for CV scaling.

It seems that LUA unit defs have a bug/new feature in dev build, so that if you don't supply the model extension in objectname=mymodel tag you can't read it in UnitDefs[unitDefID].model.type, I know that there was some postprocessing about it in earlier version, and I did fix it in a test game via unitdefs_post.lua.
So what happens is that gadget doesn't recognize 3do models as being 3do and doesn't rescale thier autogenerated CV. Also the gadget tests for CV type 4 (BA7.72) and not >=3 (BA SVN)
Also dropping support for ellipsoids and eliptical cylinders replaces those CVs with box and circular cylinders, whose shape doesn't fit as good as it did before, so we need to fix them in unitDefs.
Finaly, elipsoids are still possible if you rescale a sphere via LUA gadget, did it via same CV rescaling gadget.

The much bigger issue is with noexplode weapons as they get stuck inside a CV and even move around with the unit they've hit until it dies, and its wreck dies (if its CV was also big enough to capture the weapon)

[jK]

It seems that LUA unit defs have a bug/new feature in dev build, so that if you don't supply the model extension in objectname=mymodel tag you can't read it in UnitDefs[unitDefID].model.type, I know that there was some postprocessing about it in earlier version, and I did fix it in a test game via unitdefs_post.lua.

seems you didn't tested the new builds, that should be fixed already.

[zerver]

is this new version the one that uses mt by default?

no, it will be default disabled, as sadly its still unstable, but it can be easily enabled by a config setting. see the fixme in changelog.txt.

"Unstable" may not be the right term, I would say "incompatible" instead. And if you ask me, this situation is not going to change. The separate sim thread unfortunately does not come for free, game devs will have to tweak the game a bit to make it work.

However, one goal with this release is to remove the single threaded Spring executable, since the MT executable can also run in single threaded mode.

[PepeAmpere]

I was reading the changelog carefully and im not sure if the the problematic unloading of spring 91 is solved.

The solution for mods, that have small cargo/soliders and big airtransporters, is: lower the model sphere, but its problematic in two ways:

1) the model sphere (radius) affect the distance from which i see the transport (if small enough for unloading, i dont see transporter most of the time). For example, unit corvalkii in NOTA is big airbus transport, but needs sphere ~ as big as units it transports... so very small for (because NOTA scale). Is some way how to set the distance from which the model can be seen (big, for big unit) and still let transport unload small units? If i set globaly better LOD (like described here), does it affect performance of game much?

2) small radius for unloading small units means big unit sits strange on/in the ground when unloading.

Can moving "middle position of unit" with Spring.SetUnitMidAndAimPos() somehow help situation. I tried, not much success.

Can saving the model "moved" a bit up on Y axis help this?

[xyz]

Great news, looks like the slowdowns have been fixed in master, I suspect the not too long ago commit to separate OMP from MT. And the purple stuff is also gone.

The only thing I have noticed now are the AIs. A big AI vs AI game is no longer possible no matter what AI one chooses.
The game runs fine at the beginning but then the console starts getting flooded with messages like:

Code: Select all

[f=0040515] Warning: Bandwidth limit was reached for Player AI #2 [packets delayed]
[f=0040575] Warning: Bandwidth limit was reached for Player AI #2 [packets delayed]
[f=0040620] Warning: Bandwidth limit was reached for Player AI #2 [packets delayed]
[f=0041955] Warning: Waiting packet limit was reached for Player AI #2 [packets dropped]
[f=0041958] Warning: Waiting packet limit was reached for Player AI #2 [packets dropped]
[f=0042078] Warning: Waiting packet limit was reached for Player AI #2 [packets dropped]

And then the AI starts acting weird, having its units go in circles and not really do anything.

[gajop]

Could you please also give exact detail of the following (post your infolog.txt):
- Which Map?
- Which Game/Mod?
- How many and which AI(s)/Players?

Also note that you can reduce those limits, i.e increase maximum steady and peak bandwidth by modifying the settings.

[xyz]

no matter what AI one chooses.

The only mod I tried was BA 7.2, and the map does not matter.
The Ai number also does not matter (a 1 vs 1 will have this problem as well as a 8 vs .

Also note that you can reduce those limits, i.e increase maximum steady and peak bandwidth by modifying the settings.

May I ask where this option is?

Thanks.

[gajop]

search for "Bandwidth" on http://springrts.com/wiki/Springrc

[xyz]

That looks like network bandwidth, does that mean that the AI is sending all its instructions using sockets?