2 releases
0.1.1 | Oct 21, 2024 |
---|---|
0.1.0 | Oct 21, 2024 |
#694 in Algorithms
331 downloads per month
15KB
259 lines
Squares RNG
Implementation of the Squares CBRNG
-
Counter-based RNGs are non-serial: you can jump to any index in the RNG sequence. This is especially useful in parallel / distributed contexts
-
Squares is the fastest known CBRNG, and has higher quality than Philox at 2x speed
-
Squares is still slower than serial RNGs
-
Provides
2^64
outputs per key -
This crate is
no_std
, and all functions areconst
Example
CBRNGs are stateless. A key
is like a seed
.
let r32 = squares::u32(key, idx);
let r64 = squares::u64(key, idx);
Admissible Keys
Squares keys are not arbitrary! Many bit patterns can lead to poor output quality.
The key
function makes an admissible Squares key from a "seed" index, which has no restrictions.
let key = squares::key(239482304);
For a manual key, see Key::checked
and Key::unchecked
.
rand
Compatibility
Enable the rand
feature to expose Squares
, an RNG struct compatible with the rand
crates.
Approximate Throughput
Results will vary. On my laptop (M1 Max):
fn | time | per core |
---|---|---|
u32 |
1.29ns |
3.10 GB/s |
u64 |
1.65ns |
4.85 GB/s |
key |
24.5ns |
40.8 M/s |
License
MIT OR Apache-2.0
Dependencies
~12KB