Pavle Belanovic

Library of Parameterized Hardware Modules for Floating-Point Arithmetic with An Example Application

Wed, May 8 at 10am in 206 Egan

Abstract:

Due to inherent limitations of the fixed-point representation, it is sometimes desirable to perform arithmetic operations in the floating-point format. Although an established standard for floating-point arithmetic exists, optimal hardware implementations of algorithms require use of floating-point formats different from the ones specified in the standard. A library of fully parameterized hardware modules for floating-point format control, arithmetic operators and conversion to and from any fixed-point format are presented. Synthesis results for arithmetic operator modules in several floating-point formats, including the IEEE single precision format, are also shown. The library supports creation of custom format floating-point pipelines, as well as hybrid fixed and floating-point implementations. A hybrid implementation of the K-means clustering algorithm for multispectral and hyperspectral image processing is presented, illustrating the use of the library. Synthesis and processing results for both implementations are shown and compared.

Prof. Dana Brooks
Prof. Miriam Leeser (advisor)
Prof. Waleed Meleis