Reconfigurable Computing

Chapter 31. The Implications of Floating Point for FPGAs

Keith D. Underwood, K. Scott HemmertSandia National Laboratories

FPGA-based computing has a long history of accelerating assorted types of computations in integer and fixed-point arithmetic. Until recently, however, applications based on floating-point arithmetic have been a relative rarity. This stems from early work [6, 12, 13] that indicated that IEEE-754 standard [11] floating point was a poor match for field-programmable gate array (FPGA) technology. This led directly to numerous efforts that created libraries using specialized floating-point formats [1, 3, 7], where the width of the exponent and the width of the mantissa could be specified. Unfortunately, many scientific applications ...