Computer Graphics · Cluster
Rasterization
Coverage, depth, and antialiasing.
- 01 Half-Space Coverage and the Edge Function Test How a rasterizer decides which pixels a triangle owns, why the test is written as three signed line equations, and what the top-left rule does for shared edges.
- 02 Triangle Setup: Subpixel Snapping and the Bounding Box What a rasterizer computes before it touches a single pixel, how vertex coordinates are quantized to a fixed-point grid, and why the bounding box is the cheapest large win in the pipeline.
- 03 Why Depth Interpolates as 1/w and Screen-Space Attributes Do Not Perspective projection destroys linearity in screen space, so barycentric interpolation of depth requires the reciprocal of w, and the same reciprocal distributes every varying correctly.
- 04 Guard-Band Clipping and Why Vertices Are Cut Before the Divide Clipping happens in homogeneous space because the perspective divide is undefined behind the eye, and a guard band lets most partially-visible triangles skip the clipper entirely.
- 05 Multisampling: Coverage Samples Without Shading Every Sample MSAA asks the coverage question at several positions per pixel but shades only one, and the gap between the two is where its bandwidth advantage and its characteristic failures live.
- 06 Exact Area Coverage and the A-Buffer Computing a pixel's covered fraction analytically instead of sampling it, the data structure that keeps fragment lists per pixel, and the memory cost that keeps the approach off most GPUs.
- 07 Conservative Rasterization and the Guarantee of Non-Empty Coverage Standard rasterization may miss a pixel that a triangle touches, which breaks shadow volumes and voxelization; conservative rasterization over-covers by a bounded margin to make touching sufficient.
- 08 Early Depth Tests and the Ordering Constraints They Impose Rejecting hidden fragments before shading is the largest single saving in a raster pipeline, and it is legal only when the fragment shader is known not to change depth or discard.
- 09 Reversed-Z: Using Floating-Point Depth Without Wasting It Standard depth mapping spends almost all of a floating-point buffer near the near plane; reversing it to 1 at the near plane and 0 at the far plane redistributes precision where it is needed.
- 10 Screen-Space Binning in Tiled Rasterizers Sorting primitives into per-tile bins before shading gives locality of depth and colour access, makes deferred shading affordable, and costs a variable amount of memory per tile.