jfcg/rng: compact, fast, sponge-based random number generator

[jfcg...@gmail.com]

Hi,

jfcg/rng v0.7.0 is just released. It has the following advantages (std=math/rand, alt=golang.org/x/exp/rand):

Effective Entropy (hidden information, in bits):

rng: 128

std: 31

alt: 64

Used memory (in bytes):

rng: 24

std: 4920

alt: 48

Benchmarks:

goos: linux

goarch: amd64

pkg: github.com/jfcg/rng/bench

cpu: Intel(R) Xeon(R) Platinum 8272CL CPU @ 2.60GHz

BenchmarkGet-2       227183912  5.259 ns/op

BenchmarkStdGet-2    267202182  4.497 ns/op

BenchmarkAltGet-2    310959830  3.859 ns/op

BenchmarkExp-2       50470122  23.13  ns/op

BenchmarkStdExp-2    170507682  7.035 ns/op

BenchmarkAltExp-2    162541906  7.387 ns/op

BenchmarkOne-2       228389506  5.248 ns/op

BenchmarkStdOne-2    295455202  4.070 ns/op

BenchmarkAltOne-2    182677640  6.516 ns/op

BenchmarkNormal-2    25363636  46.07  ns/op

BenchmarkStdNormal-2 153550454  7.854 ns/op

BenchmarkAltNormal-2 145823773  8.228 ns/op

BenchmarkTwo-2       228640303  5.243 ns/op

Let me know what you think,

Thanks..

Notes:

- rng tries to be very small & fast

- rng ignores locking by design & compared with lockless std/alt

- rng does not have pseudo-rng yet, just the global one

- rng does not have all API like Perm yet

- rng does not have advanced table-based Exp/Normal like std/alt, hence quite slower for those

- rng.Normal returns two samples, so it is as fast as rng.Exp per sample

- rng is not throughly analized

Questions:

- Can you estimate the length of shortest cycle for rng state?

- Does rng mix bits good enough?

[jfcg...@gmail.com]

Hi,

jfcg/rng v1.0 has just been released. Let me know what you think.

Cheers..