On Round-Off Error of Floating-Point Addition with Guard Digits,
Abstract
The assumption of double precision binary arithmetic operation is often made in the analysis of the finite word effect on digital signal processors. Some recent computers, such as those in the IBM 360 series, use radix 16 and single precision with guard digit in floating-point addition. In this paper, a bound on the round-off error for floating-point addition in single precision with guard digits is derived. Comparison with double precision addition is made. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1971
- Accession Number
- AD0736519
Entities
People
- Bede Liu
- Toyahisa Kaneko
Organizations
- Princeton University