Itanium® Architecture for Programmers: Understanding 64-Bit Processors and EPIC Principles

8.8. Approximations for Reciprocals and Square Roots

The IEEE floating-point standard defines requirements for the operations of division and remainder and for the unary square root function, including cases where source arguments are IEEE special values (zeros, infinities, NaN, etc.).

Some RISC architectures include full hardware support for floating-point division and the square root function; these operations take more CPU cycles than other floating-point arithmetic operations and thus tend to cause pipeline stalls.

The designers of the Itanium architecture took a different approach, utilizing lookup tables that provide approximations for reciprocals and reciprocal square roots, with known bounds on accuracy. These lookups execute in the same ...