Last Updated:
02
/
22
/
2005
Floating-Point Specials
Floating-point math sometimes results in the generation of floating-point specials (such as NaNs and Infs) on CPUs as well as on GPUs. These special numbers often result in unintuitive rendering results that are hard to debug. This paper describes when floating-point specials are generated in various stages of a modern GPU pipeline, and discusses methods for handling and protecting against these situations.
