More Floating Point Alternatives
2 days ago
- The text introduces five alternative ways to represent numbers in software, all slower than hardware floating point.
- First alternative: Decimal floating point uses base 10 and is standardized in IEEE 754, with examples like Python's decimal or Java's BigDecimal.
- Second alternative: Fractions allow exact rational arithmetic, e.g., Python's fractions module or Lisp's built-in support.
- Third alternative: Symbolic computation represents exact values like sqrt(2) instead of approximations, used in computer algebra systems.
- Fourth alternative: Interval arithmetic stores numbers as ranges to precisely track error bars, being the least mainstream.
- Fifth alternative: Binary-coded decimal was used on 1960s IBM computers and persists in financial formats like ISO 8583.