FPGA Implementation of a Custom Floating-Point Library

This paper presents FPGA implementation of a floating-point library for high-performance video processing. The library comprises functions including addition, multiplication, division, square-root, exponentiation and logarithm, as well as floating-point to fixed-point and fixed-point to floating-point conversion. We implement a set of composite functions using this library to compute custom floating-point arithmetic using an Artix-7 FPGA. The synthesis of the library using the software toolkit Vivado from Xilinx is performed and compared with the open-source floating-point library Flopoco. The synthesized library has a maximum latency of 18 cycles to process 20 composite floating-point functions in parallel running at a clock frequency of 148.5 MHz.