5 releases (3 breaking)
0.4.0 | Nov 8, 2022 |
---|---|
0.3.0 | Mar 2, 2022 |
0.2.0 | Mar 1, 2022 |
0.1.1 | Feb 19, 2022 |
0.1.0 | Feb 15, 2022 |
#2514 in Algorithms
14KB
217 lines
fletcher-simd
A SIMD implementation of the Fletcher's checksum algorithm.
Note: This implementation uses a modulus of 2^k
where k
is the checksum block size in bits, as this is fast with wrapping math. Other implementations may use 2^k - 1
.
Features
- Uses
std::simd
, which currently requires nightly. - Supports all architectures supported by
std::simd
. - Both run-time and compile-time detection available via the
multiversion
crate. - Scalar fallback.
Cargo features
There is one cargo feature, runtime_dispatch
, enabled by default. When enabled, the crate will use CPU feature detection at runtime to dispatch to the appropriate SIMD implementation. Disabling this feature will result in static dispatch only.
Example
use byteorder::{ByteOrder, LittleEndian};
use fletcher_simd::Fletcher128;
fn main() {
const DATA: &str = "abcdefgh";
let mut fletcher = Fletcher128::new();
// Read bytes in little endian. Endianness matters!
fletcher.update_with_iter(
DATA.as_bytes()
.chunks(8)
.map(|chunk| LittleEndian::read_u64(chunk)),
);
assert_eq!(fletcher.value(), 0x68676665646362616867666564636261);
}
Dependencies
~2MB
~45K SLoC