Function random_real

crawl-ref/source/random.cc:408–418

range [0, 1.0) This uses a technique described by Saito and Matsumoto at MCQMC'08. Given that the IEEE floating point numbers are uniformly distributed over [1,2), we generate a number in this range and then offset it onto the range [0,1). The choice of bits (masking v. shifting) is arbitrary and should be immaterial for high quality generators.

// choice of bits (masking v. shifting) is arbitrary and
// should be immaterial for high quality generators.
double random_real()
{
    static const uint64_t UPPER_BITS = 0x3FF0000000000000ULL;
    static const uint64_t LOWER_MASK = 0x000FFFFFFFFFFFFFULL;
    const uint64_t value = UPPER_BITS | (rng::get_uint64() & LOWER_MASK);
    double result;
    // Portable memory transmutation. The union trick almost always
    // works, but this is safer.
    memcpy(&result, &value, sizeof(value));
    return result - 1.0;
}

// Roll n_trials, return true if at least one succeeded. n_trials might be
// not integer.