Skip to content

Intel Decimal Floating Point – High-Precision IEEE 754-2008 Arithmetic in Modern C++

Project Goal This library brings full-featured IEEE 754-2008 compliant Decimal Floating Point (DFP) arithmetic to modern C++, backed by Intel’s Binary Integer Decimal (BID) implementation. It wraps Intel’s low-level C APIs in type-safe, user-friendly C++23 constructs, and introduces a rich decimal_t<T> abstraction for working with Decimal32, Decimal64, and Decimal128 numbers—with precision, correctness, and expressive syntax.

Features

  • IEEE 754-2008 DFP Arithmetic Full support for decimal base arithmetic over 32-, 64-, and 128-bit BID encodings, including proper rounding, quiet comparisons, and classification.

  • Native Decimal Types Typed wrappers:     • decimal_t<uint32_t> → 7-digit Decimal32     • decimal_t<uint64_t> → 16-digit Decimal64     • decimal_t<BID_UINT128> → 34-digit Decimal128

  • Operator Overloads & Formatting Arithmetic, comparison, and stream << operators, along with std::format support out of the box.

  • Bidirectional Conversions Convert between decimal ⇄ binary floating point (float, double, long double), strings, and integers.

  • Decimal Literals Support for user-defined literals:

using namespace math::literals;
auto pi = 3.14159_dec64;
  • Decomposition Utilities Extract and inspect raw mantissa, exponent, and sign bits:
auto [sign, mantissa, exponent] = decompose(value);

  • Constants by Precision Predefined constants for each precision class (𝜋, 𝑒, 𝜏, φ, √2, ln10, rad↔deg, etc.), matching the precision of the backing type.

  • Transcendental Math Support Built-in functions like log, exp, sqrt, sin, cosh, tgamma, cbrt, erf, asinh, and more—available for all three widths.

Example Usage

using namespace math;
decimal_t<uint32_t> a = "123.45";
decimal_t<uint32_t> b = 1.05_dec32;

auto c = a * b;                    // Multiplication with correct rounding
std::cout << "Result: " << c;     // Prints "129.6225"

auto [sign, mantissa, exp] = decompose(c);

decimal_t<uint64_t> pi = constants<uint64_t>::pi;
decimal_t<uint64_t> two = constants<uint64_t>::one + constants<uint64_t>::one;
auto circumference = two * pi * 10.0_dec64;

Supported Types

Type Precision Range Mantissa Digits
decimal_t<uint32_t> Decimal32 ±10⁻⁹⁵ to ±10⁹⁶ 7
decimal_t<uint64_t> Decimal64 ±10⁻³⁸³ to ±10³⁸⁴ 16
decimal_t<BID_UINT128> Decimal128 ±10⁻⁶¹⁵⁷ to ±10⁶¹⁶ 34

License: Custom / Proprietary Contact: info@vargaconsulting.ca