Drawcall batching, a no-brainer?

Ever since the release of FSX SP2 enabling drawcall batching has been a popular method of increasing the performance of scenery. The basic idea behind drawcall batching is that the scenery engine will gather all polygons with the same material, even if they belong to different objects, and then render them in one go. This eliminates the overhead of switching material and texture settings multiple times and therefore improves the performance.

But is it a good idea to always enable drawcall batching on objects? For a while we already know that drawcalls with too many polygons in them give trouble. They can result in disappearing objects for example. So in that case you better turn off the drawcall batching.

Last week while debugging an issue reported on the forum I found out that drawcall batching can have one more negative side effect. It turns out that when objects are used at higher altitude, the drawcall batching results in an offset of the position of the object. I never noticed this myself, since I’m from the Netherlands and always test my objects at sea level. But when you make scenery in a mountainous area you will for sure encounter this. So also in that case it might be better to turn off drawcall batching.

By default ModelConverterX has drawcall batching enabled, but you can easily disable it in the exporter options. If you encounter any of the issues described above, it might be a good try.

So it turns out that drawcall batching is not the magic performance solution we have thought for a long time, but if you are aware of the limitations you can still make good use of it.

Oops, wrong position

wrong-way-signA little while ago a bug was reported in the position accuracy of the scenProc placed autogen library objects. After some debugging it indeed turned out that the location of these objects was wrong. And the bigger the size of the library object, the bigger the position offset.

I have now fixed this issue, so from tomorrow’s development release you should see that your library objects are placed more accurate.

A bug was already reported in the heading of the library objects. I still have to investigate that, but there seems to be a small offset in there as well. More news about that soon hopefully.

Simplify polygons

The features that are generated using the detect features step of scenProc are not simplified. Before these polygons would be a bit blocky because of the raster image that has been converted into the polygon. But with the simplification a more smooth polygon results. In the image below you can see an example without (left) and with (right) the simplification applied.

The simplification makes the polygons less complex and therefore also reduces the size of the AGN files. And with this simplification it also becomes possible to use the feature detect polygons for buildings. For the blocky polygons it was not possible to calculate the best fitting rectangle correctly, but for the simplified ones that no longer applies.

simplify

My tools and Windows XP

wpid-windowsxp.pngAs you have most likely heard by now Microsoft has stopped supporting Windows XP as operating system. I know there are still many users using my tools on Windows XP as well and the decision of Microsoft doesn’t mean of course that  my tools will stop working on Windows XP.

However given that official support has stopped now, I have decided that I won’t fix bugs that occur only on Windows XP from now. So if a problem doesn’t occur in Windows 7 or 8, but only happens on XP, I will not spend time to fix it anymore.

For the rest I expect that my tools will keep working on XP as well for the foreseeable future.

Full of new inspiration

20140401_145228I have returned from my vacation yesterday. As usual I don’t bring a computer on vacation, so I haven’t done anything on FSDeveloper or my tools in the last 3 weeks. But of course that doesn’t stop my brain from coming up with new ideas. Luckily I had brought a notebook (you know, a paper one) to write those ideas down. So now I have returned full of interesting ideas to try again.

At the moment the things I want to work on first are:

  • Finish the Nantucket autogen I have been working on as a test project for scenProc for a while now.
  • Get the scenProc 1.0 release ready. This mainly involves updating and writing the new manual, but there are also a few bugs to fix.
  • Start experimenting with the other scenProc ideas I have written down in the last weeks. I have some cool ideas to make even more realistic autogen that I can hopefully tell more about later on.

The photo on the right is a photo I took in Shanghai (China) from the hotel we stayed. I guess it would be a real challenge to represent just a city realistically with autogen. And also during our train travels through China I have seen a nice variation of buildings that would be interesting as autogen objects. You see, even when I am on vacation and just enjoying the landscape, my mind keeps busy with how to present something like that in a scenery.

As you might see ModelConverterX is not so high on the current prio list (but those lists tend to be quite fluid, as I quickly can get distracted by other interesting ideas to try). But I hope to spend a day in the near future to look at some of the recently reported bugs.

Replace building by rectangles updated

I have finished updating the algorithm of scenProc to replace building polygons by multiple rectangles. From the tests I have done for the Nantucket area I am quite pleased with the results.

I just spend most of tonight to type in the manual how the algorithm works, so I’m not going to repeat that here now. I would say just download the new development release tomorrow and have a good read in the manual.

Upcoming updates

The last weeks updates of my tools have been a bit slower than usual. On one hand that was because I have been busy with non-FS activities, but also because I have been experimenting a lot with creating autogen for Nantucket with scenProc. It’s useful to sometimes do some actual work with my own tools, that gives a lot of ideas for future improvements.

For ModelConverterX the highest priority new feature is to improve support for aircraft MDL files. But since I’m quite busy with scenProc now, it might take a few months before I will concentrate on ModelConverterX again. In the mean time I do try to fix bugs that are reported, but that also goes a bit slower than usual.

In the last week I have been working on improving the scenProc algorithm to split complex building footprints in multiple rectangles. I’m almost done with this update, I only need to implement one last improvement that I have in my mind now. After that I will have to update the documentation so that it is clear how to use this improved step.

I had hoped to finish this scenProc improvement by this weekend, but last week I had less time than expected. Next week we’ll go on vacation, so it might be I will only be able to finish it after that. So in that case you would have to wait a few more weeks…

Reading complex OpenStreetMap files

I got some error reports about the scenProc IMPORTOGR step. Sometimes it would throw an error that interleaved reading mode should be used. This mainly happened on bigger, more complex files. But when working on my Nantucket test area I also noticed that reading with certain required attributes stalled and got in some kind of endless loop.

I have now fixed both of these problems. So in the next development release it should be easier to use OpenStreetMap data in scenProc.

Where did my options go?

In the next development release of scenProc some options are gone from the global options form of the tool. But I have not removed them, I have just moved them to the configuration file. That way it is easier to enter different values for these options for different projects (especially when using batch mode).

So which options are affected by this?

  • KeepXML is now an option in the EXPORTBGL step
  • ProcessHoles is now an option in IMPORTOGR and DETECTFEATURES. Be careful though, the new option is called DONTPROCESSHOLES. The default behaviour is now to process the holes.

Detect features and holes

I have just fixed some bugs in the scenProc detect feature step. With these fixes the holes in the detected features should now be processed correctly. Before the holes where not always represented correctly.

Be aware that processing holes in features does take more processing time. So you might see that detecting the features takes longer if you have processing of holes enabled.