Tuning the Imaging Pipeline

To improve performance when rendering images or loading textures, follow these guidelines:

• Look at Example 4-1 to learn how to set the OpenGL state so that subsequent calls to glDrawPixels() or glCopyPixels() will be fast.

• Byte-sized components, particularly unsigned byte components, are fast. Use pixel formats where each of the components (red, green, blue, alpha, luminance, or intensity) is 8 bits long.

• Use fewer components, for example, use GL_LUMINANCE_ALPHA or GL_LUMINANCE.

• Know where the bottleneck is.

  Similar to polygon drawing, there can be a pixel-drawing bottleneck due to overload in host bandwidth, processing, or rasterizing. When all modes are off, the path is most likely limited by host bandwidth, and a wise choice of host pixel format and type pays off tremendously. This is also why byte components are sometimes faster. For example, use packed pixel format GL_RGB5_A1 to load texture with an GL_RGB5_A1 internal format.

  Zooming up pixels may create a raster bottleneck.

• For simple loading, turn all pixel modes off. Turn depth test and texture off when possible.

• A big pixel rectangle has a higher throughput (that is, pixels per second) than a small rectangle. Because the imaging pipeline is tuned to trade off a relatively large setup time with a high throughput pixel transfer efficiency, a large rectangle amortizes the setup cost over many pixels, resulting in higher throughput.

  This is an important consideration for hardware-accelerated machines where the image must be transferred from the host to the adaptor.