Why use it

CRC32C detects accidental corruption in buffers and records. The instruction updates a running remainder with up to eight bytes at once, avoiding a lookup table for a short checksum loop. It is a checksum primitive, with no cryptographic authentication property.

Operations

import native.x86.crc32c; provides native::crc32c<Arch>(accumulator, value). It belongs to native::minimal and is re-exported by native.x86 and native.

The accumulator and result are std::uint32_t. The value can be std::uint8_t, std::uint16_t, std::uint32_t or, on x86-64, std::uint64_t. An update consumes the numeric operand from its least significant byte to its most significant byte. Thus uint16_t{0x3231} consumes 0x31, then 0x32.

std::uint32_t crc = 0xffffffffu;
for (auto byte : bytes)
  crc = native::crc32c<arch>(crc, byte);
crc ^= 0xffffffffu;

The conventional initial seed and final complement are explicit here. The operation itself applies neither. All 32 seed bits participate.

Caveats

CRC32C uses the Castagnoli polynomial 0x1edc6f41, reflected as 0x82f63b78. It differs from the CRC-32 polynomial used by ZIP and Ethernet. Wider updates follow numeric byte order, so the caller chooses buffer loads and byte-order conversion.

Runtime calls require x86_feature::crc32, a "crc32" compiler target and CPU admission. This scalar instruction needs no SIMD or vector OS state. The broader sse42 compiler bundle includes it. Calls are noexcept. The module default Arch is the provider's NATIVE_BASELINE; caller target attributes do not change it. With the feature, overloads are constexpr and use native runtime instructions. Without it, only consteval calls exist.

See Intel's CRC32 instruction reference and Clang's CRC intrinsics.