Int256: Portable and Performant 256-bit integers

Boost.Int256 is a fully portable implementation of 256-bit integers, both unsigned (uint256) and signed (int256). The library supports a robust range of related functionality such as <bit>, <iostream>, <charconv>, and <format>. On all platforms both types are exactly 256 bits wide.

For comparison, the maximum value of a 64-bit unsigned integer is 18,446,744,073,709,551,615, and the maximum value of a 128-bit unsigned integer is 340,282,366,920,938,463,463,374,607,431,768,211,455. The maximum value of a 256-bit unsigned integer is 78 decimal digits long:

Values of this magnitude arise wherever a 256-bit hash, a curve scalar, or the exact product of two 128-bit numbers has to be treated as one integer rather than as a buffer of bytes.

The library is header-only, has no dependencies, and requires C++14 or later. When using C++20 or later, the library can optionally be imported as a module: import boost.int256;.

Boost.Int256 is also available as a single header, which is regenerated from include/ by CI on every change.

This release provides both the unsigned type uint256 and the signed type int256. int256 shares uint256’s layout (the same `std::uint64_t words[4], interpreted as two’s complement) and the same header set, so a value moves between the two types with a plain bit copy. See uint256 and int256.

C++ provides no 256-bit integer type, and no mainstream compiler offers one as an extension the way GCC and Clang offer __int128 on 64-bit targets. The usual substitute is a multiprecision integer, which solves the range problem at the cost of representation: Boost.Multiprecision uint256_t carries a limb count and a sign alongside the limbs, so it is larger than 32 bytes, is not trivially copyable, and cannot be handed to a GPU kernel or written straight into a packet.

Boost.Int256 provides a type that is exactly 32 bytes on every platform, holds four std::uint64_t words in a fixed order, is trivially copyable and standard layout, and is usable in constant expressions from C++14 onward. Operations use compiler intrinsics and the compiler’s own 128-bit type as an intermediate where those are available, and optimized software implementations everywhere else, so results are identical on every target while performance follows the hardware. The type and many of the functions also run unchanged in CUDA and SYCL device code.

Boost.Int256 is tested natively on Ubuntu (x86_64, x86_32, s390x, aarch64, ARM32v7), macOS (x86_64, and Apple Silicon), and Windows (x86_64, x86_32, and ARM64); as well as emulated PPC64LE using QEMU with the following compilers:

Tested on Github Actions and Drone. Coverage can be found on Codecov.