Monday, June 15, 2009

Choosing a DirectX Platform In C#

Apparently my game AI War: Fleet Command is the first selling game to use the SlimDX development framework. Does it reflect poorly on me that I didn't know that until today, a month after my game was actually released? Well, that knowledge probably wouldn't have scared me off from using their library, anyway -- the proof is in the pudding, and I liked their pudding the best out of all the ones I tried.

At the time I remember seeing a "projects that use SlimDX" page, and I even looked a few of those, but I had no idea that they were all freeware, hobbyist, or yet to be released. Color me surprised. In this article, I'm going to talk a bit about why I wound up choosing SlimDX, my experiences with it, and what other frameworks I used before it.

Early Alpha: Managed DirectX (MDX)
When I first started prototyping AI War, I did so in MDX simply because that's what my other games in the past have been coded in. However, as is widely known, MDX simply isn't supported anymore. There are some bugs and glitches in there, a few things that don't perform as well as they should, and a lot of the latest stuff (and anything beyond DX9) is not well supported to my knowledge. Overall I was pretty happy with MDX, though, and for my purposes it worked well enough to use it for the first four or so months of work on AI War.

After a certain point, however, I was having performance issues that I suspected were due to MDX. Also, it was coming up to where I needed to start adding music to the game, and I didn't want to just use DirectShow, which I had used in the past and found very limiting. Since I knew MDX was out of date, I figured it would be a good time to start looking for a more robust, modern platform.

Evaluating the XNA Framework
The first place that I turned was to the XNA framework, since it was Microsoft's spiritual successor to MDX. I actually converted my game Alden Ridge to be XNA-based for a while (never did try with AI War), and in general the conversion was pretty painless and quick. Some things are done differently, but overall the transition was smooth and the XNA way of naming things is a little more sensible sometimes. I skipped a lot of their higher-level Game classes and such, instead just embedding the XNA content in a panel in my window, same as I had been doing with MDX previously.

Very quickly I found out that having Shader Model 2.0 as a minimum hardware requirement for Alden Ridge was not going to be a good thing. It meant that my and my wife's laptops, both IBM Thinkpads (R42 and T40, respectively), could no longer run the game even though they had been able to run it fine with MDX.

That was a dealbreaker for me with XNA, but the other big issue for me was how resources, particularly audio, were handled in the 2.0 version of XNA (they may have changed this since, I am not certain). Basically I would have to run all of my audio through a special conversion tool that they provided, and then it would sit on the disk in a not-compressed format taking way more room than nice little ogg or MP3 files do. This would, in effect, make Alden Ridge an 800MB download for users instead of a 150MB download -- this would have been a dealbreaker on its own, too.

In the future I have some interest in porting some of my games to the XBox, and XNA is the obvious way to go with that. Conversion from MDX to XNA was easy given that I had kept most of the library-touching code pretty isolated, so that is something that I can approach when the time comes without having to build in XNA support all along.

Other Libraries
I looked at a number of other libraries as well, such as the Tao Framework and others based on Mono and OpenGL. Those were interesting, and in the future I might want to work with them to get some better cross-platform support, but converting all my code to Mono and OpenGL instead of some form of DirectX is going to be a lot more time consuming. I also had worries about what performance might be like on two totally new platforms, so I didn't want to sink days into this and then still find out it was unworkable for some reason or another.

Switching To SlimDX
I found SlimDX basically through google searches, and once I had exhausted all of the more official paths to modern DirectX in C# I decided to give them a try. To be clear, I wasn't wary because of their status as "hobbyists" or because I was worried the project would die, I was wary simply because it was small, new, and had relatively few tutorials. Basically all the things they say on their home page. All of those points are simply The Way Things Are(tm), and are not at all a reason not to choose SlimDX -- it's quite a good product (as I'll explain below).

Conversion to their library from MDX was crazy easy, even moreso than when I converted to XNA, and the first thing I noticed was how much smaller their installer and runtime dlls were than the MDX equivalents. And how many fewer of them there were (everything is in one convenient SlimDX.dll, rather than being spread out through nearly a dozen MDX dlls).

When I got everything working in SlimDX, there was a noticeable performance jump from MDX. Often on the order of 10% or so, and occasionally even more than that (30% during some key spikes). This was on the August 2008 version of SlimDX, which apparently had some performance issues that are not present in the newer March 2009 version. Those issues were primarily with matrix math, which I use a fair bit of for all the transforms in AI War, but I saw little difference between the two versions of SlimDX. March 2009 was perhaps a bit faster, but it was not nearly so notable as the original jump from MDX had been.

The conversion to SlimDX took part of one afternoon, and then suddenly things were running faster, so I decided to stay with the library given that I knew it was also under active maintenance (unlike MDX). And it works just fine on older graphics cards, same as MDX does.

The only problems that I really encountered with SlimDX were:

1) At first, they did not support Force Feedback, which is an issue for Alden Ridge. They have added support for that in the latest June 2009 release, but I've been so tied up with post-release work for AI War that I have not had a chance to try it yet.

2) The performance of the Line class in SlimDX was terrible, but it was still at least on par with MDX so I suspect this was an issue with the underlying DirectX code and not SlimDX itself. A simple solution was to just have a 1px square sprite that I transform and scale into drawing very nice lines (angled or straight). That's a fair bit of matrix transformation when there are a lot of lines on the screen, and SlimDX handled those with no issue whatsoever.

3) When I was originally trying to set up music in the game, I ran into a lot of trouble trying to choose between XAudio2 and DirectSound for playback. I had already settled on using an excellent C#-based ogg vorbis decoder by Atachiants Roman (which I also highly recommend as the best solution for decoding ogg files in C#), but I was having some issues with both tools. With some advice from the SlimDX team, I managed to cobble together a home-grown streaming approach using DirectSound, but I know it is not the ideal way to handle music streaming. I will probably update that in a future release of AI War, but for now it is reliable and works well enough, the main issue is that it wastes around 70MB of RAM that could be put to some other use. I'm told that the SlimDX team is planning to later add a sample that shows how to properly handle streaming in DirectSound in SlimDX, and that will be the trigger for me to update AI War in that capacity.

In general, I suppose the low volume of tutorials was my only complaint with SlimDX, and that's something that I can't really fault them for given their newness and small size. There were enough tutorials in their SDK that I was able to get everything going except for the optimal DirectSound streaming, and that's pretty darn good as far as I'm concerned.

I should also point out that my needs as far as graphics go are pretty limited, I was mostly using their matrixes, DirectX9 sprites and text classes, and things like that. I can't really comment on the 3D aspects, or the DX10 or DX11 support, though I've seen the classes there and they are reputed to be very good. My experience with their DirectInput, DirectSound, XAudio2, and Direct3D9 components have left me feeling like this is a very solid library throughout, and the clear choice for C# developers who want to do DirectX development but don't need XBox 360 support.

Update: I now have a streaming solution for SlimDX.

4 comments:

Unknown said...

MSDN lists SM 1.1 as the minimum requirement for XNA.

Christopher M. Park said...

Evidently that is changed from XNA 2, I believe it said it required SM 2.0 back then. But, either way, the laptops I have don't have SM 1.1, so they didn't work in either case. Cutting out backwards compatibility like that was a dealbreaker either way for me, although I do really like a lot of what XNA does.

Unknown said...

After becoming fairly competent with GML (My first coding experience) I have decide dto switch to XNA.

Good move?

Great work on AI War btw. Keep it up... And keep the indie style ;)

Christopher M. Park said...

XNA is not my personal favorite, but it is easier to learn, I would say. Definitely a good entry point into PC development, and you can do some powerful things with it.