Skip to content

Python Decimal BID Wrapper – High-Precision IEEE 754-2008 in Pure Python

Project Goal The Python implementation of the Intel Binary Integer Decimal (BID) encoding offers an approachable, high-level way to work with IEEE 754-2008 decimal floating point numbers. Unlike binary floating point, decimal types can represent financial and scientific values exactly—without introducing rounding errors from binary fractions.

This library encodes and decodes Decimal64 BID values into Python’s Decimal or float types and provides conversion routines back into BID integers. It is meant as a reference, teaching, and prototyping tool that mirrors the functionality of the C++ wrapper over Intel’s libBID.

Features

  • IEEE 754-2008 Compliance Works with Decimal64 (16 digits, 64-bit BID encoding) and supports proper handling of NaN, ±∞, and signed zero.

  • Conversions

  • to_decimal(int) → decode a 64-bit BID integer into a Decimal

  • to_float(int) → decode into a native float (with possible precision loss)
  • from_decimal(Decimal) → encode a Python Decimal into BID format

  • from_float(float) → encode a Python float into BID format

  • Correct Exponent/Mantissa Decomposition Supports both 53-bit “normal” mantissas and special 100m encodings with extended precision.

  • Context-Aware Uses Python’s decimal.Context with precision = 16 digits, exponent range [−383, +384], matching IEEE Decimal64 constraints.

  • Special Values Encodes/decodes NaN, +∞, −∞ in both float and Decimal domains.

Example Usage

from decimal import Decimal
import bid64   # assume module name

# Encode from Python float → BID64
raw = bid64.from_float(-12.2345678)
print(raw)  
# → 12745186945580849358

# Decode BID64 → Python Decimal
val = bid64.to_decimal(raw)
print(val)  
# → -12.2345678

# Round-trip with special values
print(bid64.from_float(float('nan')))
print(bid64.to_float(0x7c << 56))   # +inf

Supported Values

Value Type Python Input BID Encoding (hex)
Normal numbers -12.2345678 0xb0c0000c 62e76b8e
Max coefficient 9999999999999999 0x6c…ffff
Min coefficient -9999999999999999 0xf0…ffff
NaN float('nan') 0x7c00000000000000
+∞ float('inf') 0x7800000000000000
−∞ float('-inf') 0xf800000000000000

Implementation Notes

  • Mantissa & Exponent: Mantissas up to 16 digits (≤ 9,999,999,999,999,999) are supported. Exponents range from −383 to +384, encoded with a bias of 398.

  • Masks & Constants: The implementation follows Intel’s BID encoding rules, using discriminant, combination, and sign masks to classify values.

  • Decimal Context: All conversions use a fixed context:

Context(prec=16, Emin=-383, Emax=384, clamp=1)
Use Cases

  • Trading Systems – exact money values, interest rates, tax computations
  • Encode/Decode over wire decimal exponents without binary roundoff
  • Interop – with other languages with confidence
  • Reference Implementation – verify correctness of C++/libBID bindings
  • Teaching Tool – explore IEEE 754 decimal encoding in Python

License: Proprietary © 2018–2024 Varga Consulting