Hello,

The current lrand's algorithm is quite suboptimal.
After benchmarking, we find:
Current algorithm:
repeated srand with different values: 820.740000
repeated lrand: 643.880000

Proposed algorithm (PCG32-XSH-RR[1]):
repeated srand with different valudes: 1.920000
repeated lrand: 600.590000

I've attached the benchmark code for reproducibility.
Though the lrand calls performance aren't themselves
all that bad, the design of the algorithm is not very
good. It has 4856 bytes of state (vs 4 bytes for PCG),
is very expensive to seed, as the benchmarks show
(because it actually uses an other PRNG to generate a
somewhat random state) and has bad statistical properties.
See [2] for a more detailed rationale and supplementary
materials on why this specific algorithm is suboptimal.

Regards,
Elie Le Vaillant

[1]: https://pcg-random.org
[2]: https://leviathansecurity.com/blog/attacking-gos-lagged-fibonacci-generator