Section: New Software and Platforms

FiPlib

Keywords: Compilation - Approximate computing - Fixed-point representation

Functional Description: FiPlib is a C++ library that provides type definition and conversion operations for computations in fixed-point representation. Basic arithmetic as well as logical operations are transparently supported thanks to operator overloading. FiPlib also provides optimized implementations of the transcendental math functions of libm. For convenient integration, FiPlib is released as C++ header files only. Optionally, FiPlib can detect overflows and compute errors compared to floating point representation.

• Participants: Pierre Le Meur, Imane Lasri and Erven Rohou

• Contact: Erven Rohou