Why Compilers Miscalculate Floating Point Numbers: Piles of Sand, Redux

[magazine]
Volume-Issue: 
2007-08
Article Summary: 

In this continuation of his June 2007 article on floating-point numbers, Chuck Allison explains why certain compilers miscalculate sin(x) for large arguments and why some get it right. He also divulges that floating-point spacing is the key to getting the most from numeric computations.