Floating-Point Numerical Function Generators Using EVMDDs for Monotone Elementary Functions
Abstract
This paper proposes a design method for floating-point numerical function generators (NFGs) using multi-valued decision diagrams (MDDs). Our method applies to monotone elementary functions in which real values are converted into integer values that are represented by edge-valued MDDs (EVMDDs). We show that EVMDDs use fewer nodes by one or two orders of magnitude than two other types of decision diagrams, MTBDDs and BMDs. EVMDDs produce fast and compact floating-point NFGs for real-valued elementary functions, with a speed improvement of 86% over a recently proposed floating-point implementation.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2009
- Accession Number
- ADA548224
Entities
People
- Jon T. Butler
- Shinobu Nagayama
- Tsutomu Sasao
Organizations
- Naval Postgraduate School