[2.10] Dmap optimizations

[stgatilov 1314]

The latest dev build (dev16215-9215) includes many changes in dmap: 5562, 5488, 5486

The main goal of these changes is to make dmap faster.
In most cases it meant writing a faster version of some small piece of code.
There is one major exception: I had to write a brand new algorithm for optimizing polygon triangulations (~1200 LOC).

I used two well-known FMs to track progress. Here is the total time improvement:

• Painter's Wife (city): 8:10 -> 1:28
• No Honor Among Thieves (politics): 4:35 -> 0:30

You should expect less improvement on other maps, since these two are rather outstanding cases. The nhat/politics map was very slow to compile due to a bunch of stalagmites: they generated hundreds of thousands of triangles, which caused slowdowns in several places all over the code. And the painterswife/city is... just too big although I guess it has some local bottlenecks too.

Of course, with such massive changes there is always a chance that something breaks. The worst part of it is that dmap was never perfectly reliable to begin with: it just used to fail rarely enough, and hopefully this trend continues. While I did test new dmap on some existing FMs, I didn't have much time to check the outcome. So more widespread testing would be appreciated!

In case something does not work, there are cvars which can disable most of the changes and switch back to the old code. There are a lot of them, and all of them start with "dmap_". Toggling them individually is not fun, so I suggest using the new meta-cvar dmap_compatibility. You can set it to the version of TDM that you want to emulate in dmap:

• dmap_compatibility 207: legacy mode
• dmap_compatibility 208: enable major changes related to visportals and opacity (shipped in TDM 2.08)
• dmap_compatibility 210: enable dmap optimizations too (to be shipped in TDM 2.10, also default mode)

Note: not everything can be switched back via cvar! Just setting the cvar won't give you exactly the same result as the actual old version does. But I hope the biggest differences should be removed this way.

 

P.S. A general tip: if you could check the latest dev build from time to time (e.g. once a month) with the map you work with, that would help a lot in detecting issues early

[Hooded Lantern 10]

You've got my love stgatilov.

[joebarnin 256]

I just tried my most recent FM (Now and Then). On 2.09, dmap took 60 seconds. On 2.10 (dev16215-9215), it took 23 seconds. Nice! The build succeeded, and the mission seem to run fine (I played it for a few minutes). Good stuff!

[OGDA 71]

The amount of features TDM 2.10 will provide for mappers is insane. I'm constantly finding things I need for mapping in the bug tracker and forum with target version 2.10.

Is there an estimated release date? I'm asking for a friend (the .script-file).

Neeeed: https://bugs.thedarkmod.com/view.php?id=5543

[stgatilov 1314]

19 minutes ago, OGDA said:

The amount of features TDM 2.10 will provide for mappers is insane. I'm constantly finding things I need for mapping in the bug tracker and forum with target version 2.10.

Be careful! Having target version 2.10 does not mean it will be in 2.10
On rare occasions even seeing issue as "resolved" does not mean it will be in the next release

Quote

Is there an estimated release date?

February 2022.

 

[OGDA 71]

I'm looking forward to it.

Edited April 9 by OGDA

[stgatilov 1314]

The next dev build will include more dmap changes, but this time about improving precision.
Corresponding issues are: 5486 and 5580

Unfortunately, I did not put a cvar for disabling changes of 5580 (it would need duplicating too much code).
So dmap in the next dev build and in 2.10 will differ a lot in its numeric behavior, and there would be no way back.

Please keep this in mind when using new builds.