A Software Tool for Designing Fixed-Point Implementations of Computational Data Paths for Embedded and Recongurable Computational Environments
David M. Buehler, G.W. Donohoe, Pen-Shu Yeh, Nasa Gsfc · 2004
In computational environments where the cost of floating-point circuitry is prohibitive, such as deeply embedded computing and reconfigurable computing, computations requiring values from the mathematical field of reals can often be performed using fixed-point representations. However, designing fixed-point implementations of computational algorithms has a history of being a difficult, time-consuming and error-prone task which can result in sub-optimal implementations due to unnecessarily high computational error or even incorrect results. In conjunction with the Field Programmable Processor Array (FPPA) project, the software tool SIFOpt has been developed which helps algorithm designers create fixed-point implementations of computational algorithms. SIFOpt performs a static analysis of the computation, determining scaling factors for computed variables, alignment operations for additions and subtractions, rescaling operations for multiplications and determining optimal representations for the constants which are used in the computation.