Wesnoth performance and optimization (trunk)

[Sapient]

[Version 1.1.4+svn]

Lag times (controls unresponsive)
test scenario - DotM v1.1.9, upgrades purchased during turn 1

= = = = = = my own unoptimized windows build

- map appears
24 seconds

- p1(local) gains control of leader for turn 1
19 seconds

- p2(remote) gains control of leader for turn 1
23 seconds

- p1(local) gains control of leader for turn 2
37 seconds

- p2(remote) gains control of leader for turn 2
26 seconds

- p1(local) gains control of leader for turn 3
33 seconds

- p2(remote) gains control of leader for turn 3
36 seconds

= = = = = xan's optimized windows build

- map appears
11 seconds

- p1(local) gains control of leader for turn 1
1 second

- p2(remote) gains control of leader for turn 1
5 seconds

- p1(local) gains control of leader for turn 2
6 seconds

- p2(remote) gains control of leader for turn 2
7 seconds

- p1(local) gains control of leader for turn 3
5 seconds

- p2(remote) gains control of leader for turn 3
8 seconds

- - - - -
side note: generating lots of OOS for "unfound source of movement/attack"

[JW]

Holy crap? Is that to brag?

That's pretty incredible.

[rusty]

Along similar lines, I'm working on Wesbench. The idea is to create a set of scenarios, each of which tests a seperate thing. The current one tests scrolling, but I imagine scenarios which test:

• fog reveal: moving units through fog
  shroud reveal: moving units through shroud
  complex WML parsing
  scroll tests over different domains (esp. owned villages with waving flags)
  lots of units for AI to analyze

Contributions gratefully accepted!
Rusty.

[Xan]

Wow.. I certainly didn't notice that much improvement.

Just for the record, that build was made with GCC 3.4.5 with -O3, architecture set to Pentium 4 with MMX and SSE enabled.

Perhaps you should try an optimized build with VC++, to see how it compares.

[Boucman]

for the release of 1.0, some campaigns which were in SVN didn't make it into the final package...

this means that isaac has a script to easily remove a campaign from SVN when packing...


I would really like to have wesbench added to mainline, and removed when releasing, this would allow ieasy editing by all testers and provide an easy way to have WML testcases around...

I would also love to have a custom unit for that campaign which would be a simple gemoetric frame to easily test animations, a unit with huge mvt and all special attacks...

[Xan]

Recently, I've been trying out Wesnoth (1.1.8 offical windows build) on a Pentium 2, 400Mhz, 128MB RAM, running Win98.

In general, the performance isn't bad, but there's several things which cause massive performance hits.

1) Floating damage/healing labels are somewhat bad on the smoothness of animations.
2) Map labels and Multiplayer chat messages are really terrible killers of a smooth game.
3) Scrolling the lobby chat is nice and fast, but scrolling the game list is somewhat slow, and the user list is even worse.

Unfortunately, I don't have the time to look at these myself, but maybe someone else will have an opportunity to do so.