Where to get older engine dev builds (for bisecting)

[abma]

hmm, got this error, too. replacing glew32.dll with an old version seems to fix it for me:

https://github.com/spring/mingwlibs/blo ... glew32.dll

(gimme a bit for the next build)

[abma]

689 and 1216

can you give the full tag please? commit number isn't really useful in git...

got it:

91.0.1-689-g16fb826
91.0.1-1216-g77e3dc3

awww.... wtf?

git checkout 91.0.1-689-g16fb826
Previous HEAD position was 54c69a4... Merge branch 'post_release'
HEAD is now at 16fb826... add missing glGetCompressedTexImage stub
buildbot@sv-furud /var/tmp/bisect $ git bisect good
You need to start by "git bisect start"
Do you want me to do it for you [Y/n]? y
buildbot@sv-furud /var/tmp/bisect $ git checkout 91.0.1-1216-g77e3dc3
Previous HEAD position was 16fb826... add missing glGetCompressedTexImage stub
HEAD is now at 77e3dc3... allow underwater units to build stuff on the waterplane fixes commander vs. shipfac building issue
buildbot@sv-furud /var/tmp/bisect $ git bisect bad
Bisecting: 263 revisions left to test after this (roughly 8 steps)
[b2829393a70625afd457d77d076a57a30def47af] Merge branch 'develop' into PFF
buildbot@sv-furud /var/tmp/bisect $ git describe
84.0.1-497-gb282939

looks like this from some merge conflicted version... will try to continue regardingless of that

edit: cause seems to be the merge of PFF, so output could be fine.

[abma]

next one:

http://springrts.com/dl/buildbot/defaul ... ortable.7z

it seems to be quiet difficult to get versions compile after the PFF merge, not sure if we get close to the bad commit

[Licho]

757 is still ok so its after that.

[abma]

next one:
http://springrts.com/dl/buildbot/defaul ... ortable.7z

(meh, had to skip many revisions because they don't compile with current setup: http://paste.springfiles.com/view/b8728906 )

i fear i have to setup a new build environment to compile between 1164 and 757

only good thing is, the build system seems to be not the cause

[Licho]

The 1164 fails to test.. it game overs in like 5s..

[abma]

how can i run the tests myself with a "selfcompiled" spring?

seems to be not possible in a acceptable time to compile / test it .

[Licho]

svn co http://zero-k.googlecode.com/svn/trunk/ ... Benchmarks ~/.spring/Benchmarks

Compile spring, put it to .spring/engine/name

Open benchmarker in ZKL (its on settings tab) and create a test case using fps_war, null ai and zkl defaults and the engine you want to test.

If you want to compare more versions just add more test cases with different engine versions.

[abma]

how to add "custom" versions of spring?

can "engine" be a full path to the executable of spring? (in other words, where is the "offline mode"?)

can i somehow create the .json files directly?

[Licho]

Just enter the folder name as engine.. like "94.1.1-623-gb376a3f" .. if it exists in .spring/engine it will use it and wont try to download..
it only checks for main executable there, can be a symlink too..

[abma]

ok, will try...

can you check if >=94.1.1-624-gdb6e9a7 fixes the performance problem?

http://springrts.com/dl/buildbot/defaul ... -gdb6e9a7/

[dansan]

I vote this as my favorite thread of the week!

It is exciting (seriously!) to see Homes and Watson clear one mystery after the other to find their way through the labyrinth in pursuit of the secret knowledge.

Everyone is holding their breath to see what will be the endboss... a rampaging loop? a wild semaphore? a recursion gone crazy? is it the code generously donated by someone with the nick "PRISM" that switches on spring players microphones? or is it the code that allows knorke to animate krogs to do a graceful pirouette on a single toe?

Only 540/2 builds to go!


PS: I really appreciate the [prolly rather numbing] work you guys are doing!

[Beherith]

You are chasing dragons. I think it has to do with the camera not being centered on the same spot and ground decals give the increasing load.



edit: tested with dsd, with camera on commander (to test sim performance only). springsettings.cfg was identical for all runs.

[abma]

@Beherith:

which game?

[Beherith]

fps_war from the ZK repo.

[Licho]

Beherith what do you mean?
For me camera is centered to exactly same spot and ground decals are in all runs..

Benchmarker also runs with identical engine and lua settings.. how did you run it?

I don't think its "dragon" also because decals woudlnt slow down sim that way and because even damage is lower in 94.x
(In 91 it manages to do ~= 120k damage and in 94.x only ~= 60k with 3x more sim load and 2-3x less FPS)

Also what value did you actually plot? You should use those outputted by the test itself.

And you should not exclude rendering by hiding camera in corner. I'm interested in overal game performance and what made it slower between 91 and 92 in this test. Does not matter if it is somhow rendering related in the end.

[Licho]

abma, I tested 624, it is significantly better than 94.1 !
Though not yet matching 91.0 power

http://i.imgur.com/PUEzqJl.png

[Beherith]

I modifed the widget:

Code: Select all

function widget:GetInfo()
  return {
    name      = "Widget_Fps_Log",
    desc      = "Some random logging",
    author    = "Licho",
    date      = "2013",
    layer     = 0,
    enabled   = true  --  loaded by default	
  }
end

local START_GRACE = 64

local file = io.open("dt.txt", "w")

function widget:Update(dt)
	if not gameStart then return end
	if (Spring.GetGameFrame() > START_GRACE) then
		--Spring.Echo("!transmitlobby w_update_dt: "..dt)
		file:write(dt)
		file:write("\n")
		--Spring.Echo("!transmitlobby w_fps: "..Spring.GetFPS())
	end
end


function widget:GameStart()
	gameStart = true
end

I have lots of spring versions. DL'ed zk 1.1.3.6 and fps war, put them into the shared folders, then launched spring.exe with the parameters you gave.
Then I collected the dt.txt files and plotted them in excel.

[Google_Frog]

Why don't you use benchmarker? It would be much easier than trying to diagnose the difference in your setup and settings.

Also could you both test with latest ZK?

[Beherith]

Because the benchmarker does not work for me, and Licho did not provide the exact settings used (screenshot pls).

I do love your lack of faith in my reproduction of the test. The difference in setup and settings is irrelevant, as we are comparing engine versions. Plus the test cant get much simpler than that.

Can test with latest zk, IF mod depends on latest version.

EDIT: also, licho if the number of units spawned per frame is different between engine versions, there isnt much point in the benchmark. Please revise the gadget to ensure consistency.

@dansan: log2(540)