Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Exploring CAN CSA ISO IEC TR 24732-12 (2016): Decimal Floating-Point Extensions for the C Programming Language
Technical insights into the Canadian adoption of ISO/IEC TR 24732:2016 for decimal floating-point arithmetic in C
Scope and Purpose
The standard CAN CSA ISO IEC TR 24732-12 (2016) is the Canadian adoption of ISO/IEC TR 24732:2016, a Technical Report that specifies extensions to the C programming language to support decimal floating-point arithmetic. It defines data types, operations, and library interfaces for decimal floating-point computation, aligning with IEEE 754-2008 (IEEE Standard for Floating-Point Arithmetic). This Technical Report serves as the foundation for portable, high-performance decimal arithmetic in C, targeting financial, commercial, and scientific applications that require exact decimal representation and rounding.
Key Benefit: Decimal floating-point arithmetic eliminates binary rounding errors common in monetary calculations, providing the exactness required by accounting and regulatory frameworks.
Technical Requirements
Decimal Floating-Point Types
The standard introduces three primary decimal floating-point types:
_Decimal32– 32-bit decimal floating-point (7-digit precision)_Decimal64– 64-bit decimal floating-point (16-digit precision)_Decimal128– 128-bit decimal floating-point (34-digit precision)
These types follow the decimal encoding scheme defined in IEEE 754-2008, using a densely packed decimal (DPD) or binary integer decimal (BID) encoding depending on the implementation. The following table summarizes their characteristics:
| Type | Storage Width | Significant Digits | Exponent Bias | Round-Trip Guarantee |
|---|---|---|---|---|
_Decimal32 | 32 bits | 7 | 101 | 6 decimal digits |
_Decimal64 | 64 bits | 16 | 398 | 15 decimal digits |
_Decimal128 | 128 bits | 34 | 6176 | 33 decimal digits |
Tip: Use
_Decimal64 as the default choice for most financial applications—it offers a good balance between precision and performance. Literals and Constants
Decimal floating-point literals are suffixed with df (for _Decimal32), dd (for _Decimal64), and dl (for _Decimal128). Example:
_Decimal64 price = 1234.56dd; _Decimal64 tax = 0.08dd; _Decimal64 total = price + tax;Library Functions
The Technical Report specifies a set of functions declared in <decimal.h>. These cover arithmetic operations, conversion to/from strings, and formatting according to locale. Core functions include:
dec32add,dec64sub,dec128mul, etc.dec32fromString,dec64toStringdec128SameQuantum,dec64Quantize
Implementation Highlights
Implementing CAN CSA ISO IEC TR 24732-12 (2016) requires a C compiler that recognizes the decimal floating-point types as built-in or via a conforming library. Several production compilers (e.g., GCC, Clang, Oracle Developer Studio) provide partial or full support. The standard does not mandate a specific encoding, leaving compilers free to choose BID or DPD as long as external representation follows the IEEE interchange format.
⚠ Implementation Status: As of 2025, not all compilers fully implement the TR. Developers should verify with
__STDC_IEC_559__ and __STDC_DECIMAL_FP__ macros. The standard was later largely absorbed into ISO/IEC 9899:2024 (C23), which now defines std::decimal in <stdfloat.h> with syntax using std::decimal32 etc. Macros for Feature Testing
Conforming implementations define the macro __STDC_DECIMAL_FP__ to the value 201601L (mirroring the TR’s publication year). Additionally, the <float.h> header may provide attributes such as DEC32_MIN, DEC64_MAX, and DEC128_EPSILON.
Compliance and Adoption Notes
The CSA adoption as a National Standard of Canada implies that any product or service claiming conformance to this standard must satisfy all normative requirements of ISO/IEC TR 24732:2016.
Critical: This Technical Report is not a full International Standard; therefore, compliance is voluntary unless cited by regulation. However, sectors such as banking, insurance, and supply chain auditing increasingly mandate decimal arithmetic to meet audit trails (e.g., ISO 8583 or PCI DSS).
Key compliance points include:
- All decimal types must support the five standard rounding modes (round-half-even, round-half-up, round-toward-zero, round-down, round-up).
- Exceptions (inexact, underflow, overflow, division by zero, invalid operation) must be signaled according to IEEE 754-2008 status flags.
- Conversions between decimal and binary floating-point shall be performed using the algorithms defined in Annex A of the TR.
Good News: Since C23 largely incorporated this TR, writing code today against C23 ensures forward compatibility with the Canadian standard.
Frequently Asked Questions
Q: What is the status of this Technical Report in the 2025 programming landscape?
A: The TR is still valid, but its normative content has been largely subsumed by the C23 standard (ISO/IEC 9899:2024). Implementations that support C23 automatically conform to the technical requirements of TR 24732.
A: The TR is still valid, but its normative content has been largely subsumed by the C23 standard (ISO/IEC 9899:2024). Implementations that support C23 automatically conform to the technical requirements of TR 24732.
Q: Are _Decimal32, _Decimal64, and _Decimal128 always hardware-accelerated?
A: No. While some processors provide native decimal floating-point instructions (e.g., IBM POWER6 and later), many platforms emulate them in software. Performance varies, so it is advisable to profile when using large decimal workloads.
A: No. While some processors provide native decimal floating-point instructions (e.g., IBM POWER6 and later), many platforms emulate them in software. Performance varies, so it is advisable to profile when using large decimal workloads.
Q: Does the standard specify a particular encoding (BID vs DPD)?
A: The standard allows either BID or DPD as internal representation, as long as the external (interchange) format follows IEEE 754-2008. This flexibility was provided to allow hardware vendors to use their native format.
A: The standard allows either BID or DPD as internal representation, as long as the external (interchange) format follows IEEE 754-2008. This flexibility was provided to allow hardware vendors to use their native format.
— Published 2026 —