Althornin said:
last time this came up, i got all jumped over for saying that 3dfx's AA was just RG. Everyone said "no its JITTERED!!"
Now what gives?
Let's say it like that:
The hardware is able to use a configurable pattern of N subsamples on a 8x8 grid. This pattern can be changed as Colourless' tool proves.
Theoretically it can be changed per primitive, but this was never exposed by the driver.
The preset for 4x is the optimal 4-sample sparse grid. It is commonly referred to as "rotated grid", because it is similar to, but not equal to a 4-sample ordered grid rotated through its midpoint.
The preset for 2x is the optimal 2-sample sparse grid. It is commonly referred to as "rotated grid", because it is similar to, but not equal to a 2-sample ordered grid rotated through its midpoint.
The preset for 8x is not optimal.
Now for the jittered part:
Every triangle goes through triangle setup N times. For each sample you add the sample's x and y offset from the pixel center to the triangle's position. This could be called "triangle jittering". Perceptionally, instead of taking samples at different positions, you move the triangle around and always sample the pixel center. But mathematically this is of course equal.
However, it is not jitterered in the sense that sampling patterns vary, neither per pixel (spatial) nor per frame (temporal).
The first post is wayyy too messy IMO.
