RAVIOLI - reconfigurable arithmetic variable-precision implementations of on-line instructions

In this paper, we present a library of floating-point arithmetic operators that can be dynamically reconfigured for either real-number or complex-number mode, and for which the precision is variable and is set prior to compilation. The library is compared to corresponding modules built using the Xilinx Alliance CORE library of floating-point arithmetic operators for the implementation of a unit that generates the inverse of a matrix. A significant lower cost and total cycle delay is demonstrated.