Limitations of W buffers
Not transparent to an application
- must provide W ~Ze and/or projection matrix
Incomplete storage range [d/f, 1]
- large errors when d/f -> 1
- 1/W: float resolution is limited by Zs >= d/f; res(W)-exp size
Costly W buffer implementation, max 24 bits
- high-precision per-pixel divide, IEEE float input