Perl is considering using PCG64 as the PRNG for it's next version

[scottchiefbaker]

There has been a discussion started about using PCG64 as the PRNG for the next version of the Perl interpreter. We need a simple C function to process PRNG requests so we based our implementation off of the SmokeRand PCG64_64 generator. I poked around the full library code available in this repo and there are a lot of macros and function wrappers that obscure the algorithm. Below is the generator code we landed on:

typedef struct { uint64_t state;  uint64_t inc; } pcg64_random_t;

uint64_t pcg64_rand64()
{
    const uint64_t word = ((prng.state >> ((prng.state >> 59) + 5)) ^ prng.state) * 12605985483714917081ull;
    prng.state = prng.state * 6364136223846793005ull + prng.inc;

    return (word >> 43) ^ word;
}

Can you please sanity check our code to make sure we deconstructed the macros and functions appropriately? We want to triple confirm the algorithm is correct before we launch.

[Ivoz]

I haven't seen the author communicate on the library in a long time, but checking the output aligns with the existing tests should help a lot...

Since it seems you want a simple generator, that doesn't make use of the ability to "create disparate output streams within the same alogrithm & seed", I think your output would want to align with that of check-pcg64si.c (si means a 'one sequence' (not multiple) algorithm). Since you probably don't want to exhaustively check the entire sequence, if you want to be prudent then might want extra tests around max / min limits... check-base.c seems to start with an initial seed of 42u although ofc you could expand that code carefully