XXHashNative.jl

For a wrapper package, see XXhash.jl.

We referred to the reference spec from the official repository.

This package provides native Julia implementations for some of the XXHash algorithms. Currently, only one-shot hashing is supported. And see the following table for the supported algorithms:

• XXH32
• XXH64
• XXH3_64
• XXH3_128

Example

julia> using XXHashNative: xxh3_64

# xxh3_64(input::AbstractVector{UInt8}, seed=UInt64(0), secret=XXHashNative._ksecret)
# xxh3_64(input::AbstractString, seed=UInt64(0), secret=XXHashNative._ksecret)

julia> xxh3_64(codeunits("Hello, world!"))
0xf3c34bf11915e869
julia> xxh3_64("Hello, world!")
0xf3c34bf11915e869

Benchmark

Note: not particularly fast but fast enough

julia> using XXHashNative: xxh3_64

julia> using BenchmarkTools

julia> @benchmark xxh3_64(x) setup=(x=rand(UInt8, 2^20))
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
 Range (min … max):  126.968 μs … 371.351 μs  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     132.413 μs               ┊ GC (median):    0.00%
 Time  (mean ± σ):   138.741 μs ±  21.616 μs  ┊ GC (mean ± σ):  0.00% ± 0.00%

  ▅█▇▆▆▅▅▄▃▃▃▂▁▁▁▁▁ ▁                                           ▂
  █████████████████████▇▇▇▇▆▆▆▆▆▆▆▅▅▆▄▆▄▄▅▄▄▅▃▄▃▄▁▃▃▅▁▆▅▆▇███▇▇ █
  127 μs        Histogram: log(frequency) by time        247 μs <

 Memory estimate: 80 bytes, allocs estimate: 1.

julia> 1/126*10^6
7936.50#MB/s

# for comparison, the wrapper XXhash.jl is 3 times faster
julia> 1/40.725*10^6
24554.94#MB/s