Slashdot Mirror
What is Happening with OpenGL?
Trapped In Windows Hell asks: "I was just at the local game store looking for a new game, and I noticed the absolute lack of ANYTHING other than DirectX games. Where has OpenGL gone, and what does this mean for games on GNU/Linux? If DirectX is so hard to program in, so clunky to use, and limits the game to being sold on only one OS, WHY do so many programmers use it? It seems logical (to me, at least) that programming as portably as possible, as simply as possible, and using standards where possible, leaves a lot more sales options open for the future... and DirectX seems to close all options *but one*." OpenGL use in Windows gaming has decreased dramatically in favor of the use of DirectX which is improving with each release. Will OpenGL continue to mature on the Windows platform (which arguably is the platform that drives most of the mainstream demand for graphics) or will it continue to stagnate as game and driver developers concentrate on the offerings from Microsoft?
I see this "If you code in DirectX, at least it will work on all cards" canard repeated so often, I can only marvel at Microsoft's propaganda machine and the essential stupidity of mankind. Why do you think so many new DirectX games fail to work under a number of cards until they release patches?
As someone who's coded in both (and Direct X for a game that must support a wide range of cards), an OpenGL program is far more likely to work on a given range of cards if you haven't coded in explicit support for those cards. The equivalent DX program will require far more setup and test code.
People don't seem to understand this about Direct X: It supplies a feature set, but different cards implement different parts of that feature set. You have to *explicitly test* almost any feature you can name to see if the card can support it, via "capability bits". Try getting hold of the DX caps viewer, and you'll be able to see just how many of them there are.
In fact, it's worse than that. Because old drivers often lie about their caps. (Hello Virge.) Also, the caps, especially the texture caps, often don't map nicely to a card's capabilities. So the only real way of seeing whether a particular texture stage setup will fly is to try it, and see if the driver rejects it. Plus there are the weird-arse ones like the NVIDIA 8-stage setup where they jump through the hoops of the DX API to expose their register combiner functionality. (Functionality that is directly exposed in OpenGL, BTW.) Cards also have different capabilities depending on which release of direct X their drivers support. So the same card can report completely different caps depending on driver version. It's a support and programming nightmare.
The only real way to deal with all this is know what each card is capable of, read the manufacturer's release materials, and spend time on the DX mailing list. You spend a lot of time programming a particular effect in a number of different ways, and hope like hell all your combinations cover all the cards out there. You cannot guarantee a particular card will work with your app until you test it.
You may assume that DX provides fallback paths for some features, but you'd be wrong. Or where it does, it does it in a braindead way; because your card lacks fabby next-generation feature X, and you requested it, wham, it emulates the *entire* pipe in software instead of just feature X. You lose T&L, and your framerate slows to a crawl. Then you either code your own pipe to do feature X on the CPU and then hand off properly to the card, or you just go without. Either way involves a lot of testing code. OpenGL is much, much better about falling back gracefully.
I'm also not clear on why people think DirectX has a technical edge. Microsoft do a fine job of going to the current hot hardware vendor, incorporating their upcoming features in their API, and then making a lot of noise about it. But you could access vertex shaders on the GeForce3 from OpenGL on the Mac before you could ever use them from DX. The OpenGL extension mechanism means that when a part comes out, you almost immediately have access to its new features, rather than having to wait for the next DX rollout. (Remember, NVIDIA had to install a "back door" to provide access to the full register combiner functionality of geforce1-2 cards from DX. Many people don't even know it exists.) Go to NVIDIA's web site and ponder how many of their tech demos these days are in OpenGL.
The only thing DirectX has going for it, its so-called unified API, makes no difference at all in the end. You end up doing exactly what you'd do in OpenGL -- testing for certain cards, and using their exposed abilities if they exist. Writing a lot of fallback code. In the end, you're better off going with the card manufacturer's APIs, IMHO.
Think of it this way: rather than exposing a unified API, DX exposes *every possible API*. So many wonderful standards to choose from! If you're a unix guy: DX8 is the X11 of the 3D graphics API world.
A.
P.S. Sheesh, I haven't even touched on the headache that is resource management in DX.
P.P.S As someone who has to deal with all this shit (card compatibility), I'm a mite touchy on the subject =)