Graphics Reference
In-Depth Information
What lies between the vertex and fragment shaders is a mini-pipeline of sev-
eral modules,
5
shown in Figure 16.5 as a single box. They work together to com-
plete the transformation pipeline to convert to window coordinates, to perform
clipping/culling against the view frustum, and to perform rasterization to produce
a sequence of output fragments.
The
fragment shader
(whose equivalent in Direct3D is called the “pixel
shader”) is called once for each fragment and receives the fragment's location (in
window coordinates) and the already-interpolated varying outputs from the vertex
shader. In very simple cases, the fragment shader often does nothing but pass the
color unchanged through to the next stage, but the fragment shader is essential
for lighting algorithms for which simple linear interpolation between vertices is
insufficient, and is also instrumental in special effects such as blurring.
At the end of the pipeline lie the per-fragment operations, similar to those
found in the OpenGL FF pipeline.
The OpenGL Core API retains only a fraction of the entry points that were present
to support the FF pipeline. The techniques for specifying and transmitting tex-
tures and meshes have not changed significantly, and end-of-pipeline activities
like blending and double-buffering are similar. However, all other information and
operations now lie inside in the uniforms, attributes, and shader code. The leaner
OpenGL Core API is mostly concerned with activities of the following types:
• Buffer object management—control over all data stored on the GPU,
including allocation/deallocation and data transmission
• Drawing commands—sending meshes down the pipeline
• Shader-program management—downloading, compiling, activating, and
setting up uniforms and attributes
• Texturing management—installation and management of texture data
structures for use by vertex and fragment shaders
• Per-fragment operations—control over per-fragment operations at the end
of the pipeline, such as blending and dithering
• And framebuffer direct access—pixel-level read/write access
We now discuss the general structure of a typical graphics application, some
approaches to speed up certain parts of this structure, and various kinds of soft-
ware designed to offload that work from the typical designer.
A typical 3D application includes an
application model (AM)
—a collection
of data, resident in a database or in data structures, whose application-domain
5. On some hardware platforms, this part of the pipeline is also programmable, through
use of a third type of shader, known as a “geometry shader.”
