Inside Macintosh: PowerPC Numerics / Part 3 - Appendixes

Appendix B - Porting Programs to PowerPC Numerics

Floating-Point Precision

Floating-point precision may differ from the original machine to the target machine.

Some computers have floating-point formats that have a wider range than the current PowerPC Numerics formats. Wider formats include the VAX\x89 H format, the IBM Q format, and the HP quad format. Programs use these wide formats for computation involving input data from a narrower format to minimize the occurrence of overflow and underflow and to preserve accuracy. The double-double data format provides enough precision to preserve accuracy; but it offers no greater range than the double format, so it will not protect against overflow and underflow. Keep in mind that problems may arise when a program uses formats wider than double-double.

CDC and Cray computers have a single format that is wider than IEEE single and a double format that is wider than IEEE double format. When porting code from those machines, you should consider changing type declarations from single to double format.