DirectX® 9.0 and SmartShader™ 2.0 White Paper

Microsoft® DirectX® 9.0

DirectX 9.0 is the latest major revision of Microsoft's Application Programming Interface (API) for Windows operating systems. It is designed to unlock the capabilities of PC multimedia hardware. The most significant new features of DirectX 9.0 are in the area of 3D graphics, where technology is moving forward at the fastest rate. These features are designed to bring movie-quality effects and new levels of photo-realism to 3D games and other interactive applications.

A long-standing goal of 3D graphics designers is to be able to display the same cutting-edge visual effects used in movies at interactive frame rates. One of the ways to facilitate this process is to give designers the ability to port movie effects directly to PC games and vice versa. DirectX 9.0 is designed to expose these capabilities on systems with the latest graphics hardware.

Today's graphics look pretty realistic, but there is always a little something lacking that makes them look computer-generated. Many new DirectX 9.0 features focus on reducing or eliminating these inconsistencies to bring us a big step closer to photo-realism. These same features provide the necessary flexibility to achieve a range of interesting non-photo-realistic special effects as well.

DirectX 9.0 is available as an update to Windows operating systems. To fully experience the benefits it provides, you will need one of the latest generation of high performance graphics accelerators that supports all the major new features, including 2.0 Vertex Shaders, 2.0 Pixel Shaders, and 128-bit floating point data formats. The first available products to support these features are the Radeon 9700 series and Radeon 9500 series GPUs from ATI. The demos described in this paper provide examples of how they can be used to achieve amazing new visual effects.

Shaders

Shaders are small programs that determine how a 3D object or image is rendered. They are designed to execute on a graphics processor rather than a CPU, which allows them to run at interactive frame rates while freeing up the CPU for other tasks. There are two types of shaders in DirectX - Vertex Shaders and Pixel Shaders. Vertex Shaders are used to modify the geometry or shape of 3D objects, and also to control lighting and shadows. Pixel Shaders manipulate the color and appearance of surfaces.

Starting with DirectX 8.0, each version of DirectX has included new versions of the vertex and pixel shader languages. DX9 introduces version 2.0 Vertex and Pixel Shaders. Whereas earlier versions of these shader languages were often incompatible with each other, the full floating point instruction set of version 2.0 shaders was designed to be flexible enough to be compatible with any future shading language.

2.0 Vertex Shaders

The version 2.0 Vertex Shader specification adds new flow control commands, including loops, jumps and subroutines. As standard features in higher-level languages like C, these commands make it easier to code vertex shaders and enhance processing efficiency.

For example, one common application for vertex shaders is to process lighting calculations in 3D scenes. With earlier versions of the vertex shader language, you would have to write a new shader to use in different environments with different lighting conditions. With 2.0 Vertex Shaders, developers can write a single general-purpose lighting shader that can handle any number of lights with customizable colors, shapes, and positions by calling different subroutines. Another example is real-time fur rendering, in which the fur is built up on the surface of a character by generating a series of concentric shells of textured and alpha-blended geometry. A looping vertex shader routine can generate these shells easily, repeating until the fur reaches the desired length.

Earlier versions of the DirectX vertex shader language limited shaders to a maximum of 128 instructions. With version 2.0, vertex shaders can now be thousands or even tens of thousands of instructions long - a massive increase in complexity and power. In fact, using multiple passes, 2.0 vertex shaders with practically unlimited instruction counts can be executed in hardware. New operations like sines, cosines, and power functions also simplify shader code, allowing real time processing of effects that previously took thousands of times longer.

2.0 Pixel Shaders

Powerful programmable pixel shaders are essential for creating photo-realistic, cinematic quality effects. The main things limiting the complexity of a pixel shader are the number of textures it can work with, and the maximum number of instructions it can contain. The previous version of the DirectX pixel shader language, version 1.4, was limited to a maximum of 6 textures and 28 instructions. With the 2.0 Pixel Shader language in DirectX 9.0, the maximums have been raised to 16 textures and 160 instructions, allowing far more complex shaders to be written than ever before.

In fact, these instruction counts are not really limitations, since new high precision floating point data formats make multi-pass pixel shader effects possible. This means 2.0 Pixel Shaders of any length can be used, allowing the rendering of arbitrarily complex visual effects - even the ones used in the latest computer generated films and television shows. For example, 2.0 Pixel Shaders can accurately simulate the scattering of light in semi-translucent materials like skin, marble, wax, or car paint.

Like 2.0 Vertex Shaders, 2.0 Pixel Shaders have also been enhanced with many new operations that make them more efficient and easier to write. For example, a new square root operation makes it feasible for 2.0 Pixel Shaders to make use of compressed bump maps that take up far less memory than standard bump maps - meaning high resolution bump maps can be used more liberally in games than ever before.

Floating Point Data Formats

Another limitation of previous pixel shader versions was that they could only work with integer values. For example, most colors are represented using 8-bit integers for Red, Green, and Blue, meaning each can have a value from 0 to 255. Since integers cannot have fractional values, they do not work well with operations like division and square roots. When they are run through the complex mathematical calculations in pixel shaders, errors can quickly build up that result in an obvious loss of precision.

DirectX 9.0 supports several new floating-point color formats that offer vastly greater range and precision than integer formats. For example, while a 16-bit integer value is limited to a range of +/-32,767, a 16-bit floating point value could represent numbers as small as 0.0000000000000001, or as large as 10,000,000,000,000,000! These levels of precision and range are standard when producing cinema-quality graphics, and are essential in order to bring the same quality to interactive PC applications.

Figure 1: DirectX® Feature Comparison