Engine Testing - 97.0.1-20-gcd4602b (10. Jun 2014)

[Silentwings]

You are right, in that I care for performance in the long run, but for the particular case of 91<->97: its that 91 is old enough that I don't remember it well (I was barely even involved in development at that time) and there is no chance of BA regressing so, without wanting to sound unhelpful, I am not very concerned by it. The API has changed so much since then that it would be some effort to even make my perf tests compatible with 91.

For the recent releases where I have been involved (->94->95->96->97->current) I have kept a close watch on it, but it's only really within the last 9 months that my perf testing has got good enough to feel that I'm getting an accurate idea of what goes on.

As for searching for the difference, if there ever was one single step that made a substantial part of the difference (although I am not sure there ever was), I'd guess the place to start would be the dev builds of 91 through to 94; but there have been so many changes to the libs and build setup that I think doing it on the buildbot or on windows (which is me) would be unfeasible and probably whats needed is some moderately capable linuxer. At least 92 & 93 must still exist in the master branch, so that cuts down the work before you have to build anything considerably.

[Google_Frog]

As far as I know all the old builds are around somewhere. Lua has changed very little since 91.0 and ZK is compatible with 91.0 so if you need a game to do performance testing with it is available. Although I don't know how far past 91.0 the compatibility runs. We have the reverse compatibility because the 94, 95 and 96 releases had significant performance problems with many ZK players so 91.0 would be a good place to start. Judging from the the level of complaining from the playerbase there has been a performance improvement through the versions 94->95->96->96 just not since 91.

[Jools]

As far as I know all the old builds are around somewhere. Lua has changed very little since 91.0 and ZK is compatible with 91.0 so if you need a game to do performance testing with it is available.

There was a revision to the gadgethandler between 91 and 96 somewhere, so that has changed at least.

I agree that perfornance should be an important consideration, not just features. Players most of the time don't see how spring versions differ in terms in features, only modders do that. But it is the players who are the customers/end users if you say so.

Wouldn't it be better to test with some game that has less heavy lua?

[jK]

jk: if memory usage doubles with no apparent reason, it is called bloat. From your response i deduct there was some significant improvements under the hood that justifies double memory consumption?

You don't understand lua ... google "garbage collector" ...

[jK]

I did a proper test this time with only one spring instance running at a time, 1000 moving flash, and let it run until first profiler reached 100s.

http://i.imgur.com/aTHQKnJ.png
http://i.imgur.com/FeDK7Xl.png

In 96 you have shadows disabled and enabled in 97.0.1 ...

[varikonniemi]

Same cfg file was used in both (+the nonexistent file. whats up with that?)

But you are correct. I deleted the cfg file and let springlobby create a new one, and now the performance is equal with both versions.

[malric]

my perf testing has got good enough to feel that I'm getting an accurate idea of what goes on.

'
Is this available somewhere - as an example of how you do it ? I tried to look in the BA repo but could not find it easily.

[Silentwings]

I use some gadgets to produce something close-ish to conditions of an intense game. For the moment it's not available, sorry. (Partly because its still wip and partly because many perf discussions I read are people misusing what tools they do have & without meaning anything personal to you, I don't have time/motivation to argue with the inevitable misuse of it.)

[raaar]

tested "97.0.1-20-gcd4602b (10. Jun 2014)" with metal factions v0.71 vs AI and it worked fine (40 min game).

then i spawned all units with "/give all" and gave some orders and found that this bug is still alive:

0004228: hovering aircraft get "deactivated" mid air while not idle

it can be reproduced with XTA 9.74. Spawn an arm_construction_aircraft on "the cold place remake v3" and order it to reclaim some icy shards. It'll open and close its wings while being busy in the air with its reclaiming job.

[Jools]

That's true, i managed to reproduce it in coldplace remake v2. Haven't noticed this issue before.

Edit: it only happens if the reclaim queue contains only 1 item. If the queue contains > 1 item them they do not keep turning on and off.

[Jools]

What controls which screen the game launches on?

I have 97-20 launching on my second screen even if I alt-return to make it windowed and use win+shift+left arrow to move it to screen 1, and then fullscreen it again with alt+return.

Is there a setting somewhere that defines what screen to launch on? I think this may also be a win 7 bug, I remember I read somewhere about this, something that you must minimise application and then restore it for screen to be correct.

[jK]

What controls which screen the game launches on?

I have 97-20 launching on my second screen even if I alt-return to make it windowed and use win+shift+left arrow to move it to screen 1, and then fullscreen it again with alt+return.

Is there a setting somewhere that defines what screen to launch on? I think this may also be a win 7 bug, I remember I read somewhere about this, something that you must minimise application and then restore it for screen to be correct.

No, Spring 100% trust SDL2 in that case, and it doesn't give any option where to fullscreen nor what resolution should be selected.

[raaar]

about performance

i think it always uses 100% of one CPU core.

1 min after i start a spring game (idle commander on the cold place) on xta 9.74 (happens with my game as well) my cpu fan starts spinning fast and making noise.

this happens with this test version as well as spring 96.0

when i use the debug view (alt-b), the "Gamecontroller :: Draw" is near 100%. And the windows task manager cpu usage history shows a near constant line at about 25% usage (i have a quad core).

[Jools]

No, Spring 100% trust SDL2 in that case, and it doesn't give any option where to fullscreen nor what resolution should be selected.

Would it be possible to make such an option. My 97-63 starts always in the wrong screen and I can't get it to start in my main one.

If I start it in and run the game I can move it back to my main window with shift+win, but it will have the configuration of the second screen, meaning I have to adjust all game windows. And next time it will again start in that window.

Maybe it's an issue with SDL2 then, but any idea how to troubleshoot the issue?

[abma]

next test candidate: RC3: http://springrts.com/phpbb/viewtopic.php?f=12&t=32424