martinus/random_bool_benchmark.cpp

## random_bool_benchmark.cpp
#include <algorithm>
#include <chrono>
#include <iostream>
#include <random>

#define LIKELY(x) __builtin_expect((x), 1)
#define UNLIKELY(x) __builtin_expect((x), 0)
#define NO_INLINE __attribute__((noinline))

// extremely fast random number generator that also produces very high quality random.
// see PractRand: http://pracrand.sourceforge.net/PractRand.txt
class sfc64 {
  public:
    using result_type = uint64_t;

    static constexpr uint64_t(min)() { return 0; }
    static constexpr uint64_t(max)() { return UINT64_C(-1); }

    sfc64() : sfc64(std::random_device{}()) {}

    explicit sfc64(uint64_t seed) : m_a(seed), m_b(seed), m_c(seed), m_counter(1) {
        for (int i = 0; i < 12; ++i) {
            operator()();
        }
    }

    uint64_t operator()() noexcept {
        auto const tmp = m_a + m_b + m_counter++;
        m_a = m_b ^ (m_b >> right_shift);
        m_b = m_c + (m_c << left_shift);
        m_c = rotl(m_c, rotation) + tmp;
        return tmp;
    }

  private:
    template <typename T> T rotl(T const x, int k) { return (x << k) | (x >> (8 * sizeof(T) - k)); }

    static constexpr int rotation = 24;
    static constexpr int right_shift = 11;
    static constexpr int left_shift = 3;
    uint64_t m_a;
    uint64_t m_b;
    uint64_t m_c;
    uint64_t m_counter;
};

// Probably the simplest correct implementation of random bool.
// Also one of the slowest variant.
class UniformDistribution {
  public:
    template <typename Rng> bool operator()(Rng &rng) {
        return std::uniform_int_distribution<>{0, 1}(rng);
    }
};

// bounded() is a highly optimized way to produce unbiased bounded random numbers.
// It's not really fast when just used for bool though.
class BoundedOperator {
  public:
    template <typename Rng> uint64_t bounded(Rng &rng, uint64_t boundExcluded) {
#ifdef __SIZEOF_INT128__
        // fastest solution I could find
        // 1.78746 ns/op total
        using u128 = unsigned __int128;

        u128 m = u128(std::uniform_int_distribution<uint64_t>{}(rng)) * u128(boundExcluded);
        uint64_t l = (uint64_t)m;
        if (l < boundExcluded) {
            uint64_t t = (0 - boundExcluded) % boundExcluded;
            while (l < t) {
                m = u128(rng()) * u128(boundExcluded);
                l = (uint64_t)m;
            }
        }
        return m >> 64;
#else
        // java's method usually needs only 1 modulus, faster than OpenBSD's method or pcg's.
        // also faster than std::shuffle's method. 9.73759 ns/op total
        // 7.02644 ns/op for std::random_shuffle which uses this.
        uint64_t x, r;
        do {
            x = rng();
            r = x % boundExcluded;
        } while (x - r > (0 - boundExcluded));
        return r;
#endif
    }

    template <typename Rng> bool operator()(Rng &rng) { return bounded(rng, 2); }
};

// Unbiased, but very slow. bernoulli_distribution is too generic for this.
class BernoulliDistribution {
  public:
    template <typename Rng> bool operator()(Rng &rng) { return m_dist(rng); }

  private:
    std::bernoulli_distribution m_dist;
};

// Biased if the rng does not give random numbers in [0, 2^x].
class BinaryAnd {
  public:
    template <typename Rng> bool operator()(Rng &rng) { return rng() & 1; }
};

// unbiased
template <typename U = uint64_t> class BinaryAndUnbiased {
  public:
    template <typename Rng> bool operator()(Rng &rng) {
        return (std::uniform_int_distribution<U>{}(rng)) & 1;
    }
};

// Unbiased & quite fast but requires 16byte on 64bit platforms.
template <typename U = uint64_t> class RandomizerWithCounterT {
  public:
    template <typename Rng> bool operator()(Rng &rng) {
        if (UNLIKELY(0 == m_counter)) {
            m_rand = std::uniform_int_distribution<U>{}(rng);
            m_counter = sizeof(m_rand) * 8;
        }
        return (m_rand >> --m_counter) & 1;
    }

  private:
    U m_rand = 0;
    int m_counter = 0;
};

// Unbiased & quite fast but requires 16byte on 64bit platforms.
template <typename U = uint64_t> class RandomizerWithCounter2 {
  public:
    template <typename Rng> bool operator()(Rng &rng) {
        auto b = m_counter & s_mask;
        if (UNLIKELY(0 == b)) {
            m_rand = std::uniform_int_distribution<U>{}(rng);
        }
        ++m_counter;
        return (m_rand >> b) & 1;
    }

  private:
    static constexpr U s_mask = sizeof(U) * 8 - 1;
    U m_rand = 0;
    uint_fast8_t m_counter = 0;
};

// Unbiased, fastest variant. Gets rid of the counter by sacrificing 1 bit of randomness.
// UNLIKELY macro seems important for clang++.
template <typename U = uint64_t> class RandomizerWithShiftT {
  public:
    template <typename Rng> bool operator()(Rng &rng) {
        if (UNLIKELY(1 == m_rand)) {
            m_rand = std::uniform_int_distribution<U>{}(rng) | s_mask_left1;
        }
        bool const ret = m_rand & 1;
        m_rand >>= 1;
        return ret;
    }

  private:
    static constexpr const U s_mask_left1 = U(1) << (sizeof(U) * 8 - 1);
    U m_rand = 1;
};

template <typename U = size_t> class RandomizerWithShiftMaxSize {
  public:
    template <typename Rng> bool operator()(Rng &rng) {
        if (UNLIKELY(1 == m_rand)) {
            m_rand = std::uniform_int_distribution<U>{0, s_mask_left1 - 1}(rng) | s_mask_left1;
        }
        bool const ret = m_rand & 1;
        m_rand >>= 1;
        return ret;
    }

  private:
    static constexpr const U s_mask_left1 = U(1) << (sizeof(U) * 8 - 1);
    U m_rand = 1;
};

template <typename U = uint64_t> class RandomizerWithShiftTNoLikely {
  public:
    template <typename Rng> bool operator()(Rng &rng) {
        if (1 == m_rand) {
            m_rand = std::uniform_int_distribution<U>{}(rng) | (U(1) << (sizeof(U) * 8 - 1));
        }
        bool const ret = m_rand & 1;
        m_rand >>= 1;
        return ret;
    }

  private:
    U m_rand = 1;
};

// fastest variant for 4x unrolled loop
template <typename U = uint64_t> class RandomizerWithCounterShiftT {
  public:
    template <typename Rng> bool operator()(Rng &rng) {
        if (UNLIKELY(0 == m_counter)) {
            m_rand = std::uniform_int_distribution<U>{}(rng);
            m_counter = sizeof(U) * 8;
        }
        --m_counter;
        bool ret = m_rand & 1;
        m_rand >>= 1;
        return ret;
    }

  private:
    U m_rand;
    int m_counter = 0;
};

template <typename U> class RandomizerWithMaskShift {
  public:
    template <typename Rng> bool operator()(Rng &rng) {
        if (UNLIKELY(0 == m_mask)) {
            m_rand = std::uniform_int_distribution<U>{}(rng);
            m_mask = 1;
        }
        bool ret = m_rand & m_mask;
        m_mask <<= 1;
        return ret;
    }

  private:
    U m_rand;
    U m_mask = 0;
};

template <typename U> class RandomizerWithRotl {
  public:
    template <typename Rng> bool operator()(Rng &rng) {
        if (UNLIKELY(1 == m_mask)) {
            m_rand = std::uniform_int_distribution<U>{}(rng);
        }
        m_mask = rotl(m_mask, 1);
        return m_rand & m_mask;
    }

  private:
    U m_rand;
    U m_mask = 1;
};

template <typename B, typename Rng>
NO_INLINE void benchU4(size_t iters, const char *rng_name, const char *method_name) {
    size_t loops = iters / 4;
    size_t s = 0;

    std::vector<double> times;
    for (size_t i = 0; i < 7; ++i) {
        Rng rng{123};
        B b;
        auto t_start = std::chrono::high_resolution_clock::now();
        for (size_t t = 0; t < loops; ++t) {
            if (b(rng)) {
                ++s;
            }
            if (b(rng)) {
                ++s;
            }
            if (b(rng)) {
                ++s;
            }
            if (b(rng)) {
                ++s;
            }
        }
        auto t_end = std::chrono::high_resolution_clock::now();
        auto t_sec = std::chrono::duration<double>(t_end - t_start).count();
        times.push_back(t_sec / iters * 1e9);
    }
    std::sort(times.begin(), times.end());
    for (auto t : times) {
        std::cout << t << "; ";
    }
    std::cout << sizeof(B) << "; \"" << method_name << "\"; \"" << rng_name << "\"; 4; " << s
              << std::endl;
}

template <typename B, typename Rng>
NO_INLINE void benchU1(size_t iters, const char *rng_name, const char *method_name) {
    size_t s = 0;
    std::vector<double> times;
    for (size_t i = 0; i < 7; ++i) {
        Rng rng{123};
        B b;
        auto t_start = std::chrono::high_resolution_clock::now();
        for (size_t t = 0; t < iters; ++t) {
            if (b(rng)) {
                ++s;
            }
        }
        auto t_end = std::chrono::high_resolution_clock::now();
        auto t_sec = std::chrono::duration<double>(t_end - t_start).count();
        times.push_back(t_sec / iters * 1e9);
    }
    std::sort(times.begin(), times.end());
    for (auto t : times) {
        std::cout << t << "; ";
    }
    std::cout << sizeof(B) << "; \"" << method_name << "\"; \"" << rng_name << "\"; 1; " << s
              << std::endl;
}

template <typename Rng> void bench(size_t iters, const char *rng_name) {
    benchU1<RandomizerWithCounterT<uint64_t>, Rng>(iters, rng_name,
                                                   "RandomizerWithCounterT<uint64_t>");
    benchU4<RandomizerWithCounterT<uint64_t>, Rng>(iters, rng_name,
                                                   "RandomizerWithCounterT<uint64_t>");
    benchU1<BinaryAndUnbiased<uint64_t>, Rng>(iters, rng_name, "BinaryAndUnbiased<uint64_t>");
    benchU4<BinaryAndUnbiased<uint64_t>, Rng>(iters, rng_name, "BinaryAndUnbiased<uint64_t>");

    benchU1<RandomizerWithShiftT<uint64_t>, Rng>(iters, rng_name, "RandomizerWithShiftT<uint64_t>");
    benchU4<RandomizerWithShiftT<uint64_t>, Rng>(iters, rng_name, "RandomizerWithShiftT<uint64_t>");

    benchU1<RandomizerWithShiftMaxSize<uint64_t>, Rng>(iters, rng_name,
                                                       "RandomizerWithShiftMaxSize<uint64_t>");
    benchU4<RandomizerWithShiftMaxSize<uint64_t>, Rng>(iters, rng_name,
                                                       "RandomizerWithShiftMaxSize<uint64_t>");

    benchU1<RandomizerWithMaskShift<uint64_t>, Rng>(iters, rng_name,
                                                    "RandomizerWithMaskShift<uint64_t>");
    benchU4<RandomizerWithMaskShift<uint64_t>, Rng>(iters, rng_name,
                                                    "RandomizerWithMaskShift<uint64_t>");

    benchU1<RandomizerWithRotl<uint64_t>, Rng>(iters, rng_name, "RandomizerWithRotl<uint64_t>");
    benchU4<RandomizerWithRotl<uint64_t>, Rng>(iters, rng_name, "RandomizerWithRotl<uint64_t>");

    benchU1<RandomizerWithCounter2<uint64_t>, Rng>(iters, rng_name,
                                                   "RandomizerWithCounter2<uint64_t>");
    benchU4<RandomizerWithCounter2<uint64_t>, Rng>(iters, rng_name,
                                                   "RandomizerWithCounter2<uint64_t>");

    benchU1<RandomizerWithShiftTNoLikely<uint64_t>, Rng>(iters, rng_name,
                                                         "RandomizerWithShiftTNoLikely<uint64_t>");
    benchU4<RandomizerWithShiftTNoLikely<uint64_t>, Rng>(iters, rng_name,
                                                         "RandomizerWithShiftTNoLikely<uint64_t>");

    benchU1<RandomizerWithCounterShiftT<uint64_t>, Rng>(iters, rng_name,
                                                        "RandomizerWithCounterShiftT<uint64_t>");
    benchU4<RandomizerWithCounterShiftT<uint64_t>, Rng>(iters, rng_name,
                                                        "RandomizerWithCounterShiftT<uint64_t>");
    benchU1<BinaryAnd, Rng>(iters, rng_name, "BinaryAnd");
    benchU4<BinaryAnd, Rng>(iters, rng_name, "BinaryAnd");

    benchU1<BoundedOperator, Rng>(iters, rng_name, "BoundedOperator");
    benchU4<BoundedOperator, Rng>(iters, rng_name, "BoundedOperator");

    benchU1<BernoulliDistribution, Rng>(iters, rng_name, "BernoulliDistribution");
    benchU4<BernoulliDistribution, Rng>(iters, rng_name, "BernoulliDistribution");

    benchU1<UniformDistribution, Rng>(iters, rng_name, "UniformDistribution");
    benchU4<UniformDistribution, Rng>(iters, rng_name, "UniformDistribution");
}

int main(int, char **) {
    size_t iters = UINT64_C(1'000'000'000);
    bench<sfc64>(iters, "sfc64");
    bench<std::mt19937>(iters, "std::mt19937");
    bench<std::mt19937_64>(iters, "std::mt19937_64");
}
	#include <algorithm>
	#include <chrono>
	#include <iostream>
	#include <random>

	#define LIKELY(x) __builtin_expect((x), 1)
	#define UNLIKELY(x) __builtin_expect((x), 0)
	#define NO_INLINE __attribute__((noinline))

	// extremely fast random number generator that also produces very high quality random.
	// see PractRand: http://pracrand.sourceforge.net/PractRand.txt
	class sfc64 {
	public:
	using result_type = uint64_t;

	static constexpr uint64_t(min)() { return 0; }
	static constexpr uint64_t(max)() { return UINT64_C(-1); }

	sfc64() : sfc64(std::random_device{}()) {}

	explicit sfc64(uint64_t seed) : m_a(seed), m_b(seed), m_c(seed), m_counter(1) {
	for (int i = 0; i < 12; ++i) {
	operator()();
	}
	}

	uint64_t operator()() noexcept {
	auto const tmp = m_a + m_b + m_counter++;
	m_a = m_b ^ (m_b >> right_shift);
	m_b = m_c + (m_c << left_shift);
	m_c = rotl(m_c, rotation) + tmp;
	return tmp;
	}

	private:
	template <typename T> T rotl(T const x, int k) { return (x << k) \| (x >> (8 * sizeof(T) - k)); }

	static constexpr int rotation = 24;
	static constexpr int right_shift = 11;
	static constexpr int left_shift = 3;
	uint64_t m_a;
	uint64_t m_b;
	uint64_t m_c;
	uint64_t m_counter;
	};

	// Probably the simplest correct implementation of random bool.
	// Also one of the slowest variant.
	class UniformDistribution {
	public:
	template <typename Rng> bool operator()(Rng &rng) {
	return std::uniform_int_distribution<>{0, 1}(rng);
	}
	};

	// bounded() is a highly optimized way to produce unbiased bounded random numbers.
	// It's not really fast when just used for bool though.
	class BoundedOperator {
	public:
	template <typename Rng> uint64_t bounded(Rng &rng, uint64_t boundExcluded) {
	#ifdef __SIZEOF_INT128__
	// fastest solution I could find
	// 1.78746 ns/op total
	using u128 = unsigned __int128;

	u128 m = u128(std::uniform_int_distribution<uint64_t>{}(rng)) * u128(boundExcluded);
	uint64_t l = (uint64_t)m;
	if (l < boundExcluded) {
	uint64_t t = (0 - boundExcluded) % boundExcluded;
	while (l < t) {
	m = u128(rng()) * u128(boundExcluded);
	l = (uint64_t)m;
	}
	}
	return m >> 64;
	#else
	// java's method usually needs only 1 modulus, faster than OpenBSD's method or pcg's.
	// also faster than std::shuffle's method. 9.73759 ns/op total
	// 7.02644 ns/op for std::random_shuffle which uses this.
	uint64_t x, r;
	do {
	x = rng();
	r = x % boundExcluded;
	} while (x - r > (0 - boundExcluded));
	return r;
	#endif
	}

	template <typename Rng> bool operator()(Rng &rng) { return bounded(rng, 2); }
	};

	// Unbiased, but very slow. bernoulli_distribution is too generic for this.
	class BernoulliDistribution {
	public:
	template <typename Rng> bool operator()(Rng &rng) { return m_dist(rng); }

	private:
	std::bernoulli_distribution m_dist;
	};

	// Biased if the rng does not give random numbers in [0, 2^x].
	class BinaryAnd {
	public:
	template <typename Rng> bool operator()(Rng &rng) { return rng() & 1; }
	};

	// unbiased
	template <typename U = uint64_t> class BinaryAndUnbiased {
	public:
	template <typename Rng> bool operator()(Rng &rng) {
	return (std::uniform_int_distribution<U>{}(rng)) & 1;
	}
	};

	// Unbiased & quite fast but requires 16byte on 64bit platforms.
	template <typename U = uint64_t> class RandomizerWithCounterT {
	public:
	template <typename Rng> bool operator()(Rng &rng) {
	if (UNLIKELY(0 == m_counter)) {
	m_rand = std::uniform_int_distribution<U>{}(rng);
	m_counter = sizeof(m_rand) * 8;
	}
	return (m_rand >> --m_counter) & 1;
	}

	private:
	U m_rand = 0;
	int m_counter = 0;
	};

	// Unbiased & quite fast but requires 16byte on 64bit platforms.
	template <typename U = uint64_t> class RandomizerWithCounter2 {
	public:
	template <typename Rng> bool operator()(Rng &rng) {
	auto b = m_counter & s_mask;
	if (UNLIKELY(0 == b)) {
	m_rand = std::uniform_int_distribution<U>{}(rng);
	}
	++m_counter;
	return (m_rand >> b) & 1;
	}

	private:
	static constexpr U s_mask = sizeof(U) * 8 - 1;
	U m_rand = 0;
	uint_fast8_t m_counter = 0;
	};

	// Unbiased, fastest variant. Gets rid of the counter by sacrificing 1 bit of randomness.
	// UNLIKELY macro seems important for clang++.
	template <typename U = uint64_t> class RandomizerWithShiftT {
	public:
	template <typename Rng> bool operator()(Rng &rng) {
	if (UNLIKELY(1 == m_rand)) {
	m_rand = std::uniform_int_distribution<U>{}(rng) \| s_mask_left1;
	}
	bool const ret = m_rand & 1;
	m_rand >>= 1;
	return ret;
	}

	private:
	static constexpr const U s_mask_left1 = U(1) << (sizeof(U) * 8 - 1);
	U m_rand = 1;
	};

	template <typename U = size_t> class RandomizerWithShiftMaxSize {
	public:
	template <typename Rng> bool operator()(Rng &rng) {
	if (UNLIKELY(1 == m_rand)) {
	m_rand = std::uniform_int_distribution<U>{0, s_mask_left1 - 1}(rng) \| s_mask_left1;
	}
	bool const ret = m_rand & 1;
	m_rand >>= 1;
	return ret;
	}

	private:
	static constexpr const U s_mask_left1 = U(1) << (sizeof(U) * 8 - 1);
	U m_rand = 1;
	};

	template <typename U = uint64_t> class RandomizerWithShiftTNoLikely {
	public:
	template <typename Rng> bool operator()(Rng &rng) {
	if (1 == m_rand) {
	m_rand = std::uniform_int_distribution<U>{}(rng) \| (U(1) << (sizeof(U) * 8 - 1));
	}
	bool const ret = m_rand & 1;
	m_rand >>= 1;
	return ret;
	}

	private:
	U m_rand = 1;
	};

	// fastest variant for 4x unrolled loop
	template <typename U = uint64_t> class RandomizerWithCounterShiftT {
	public:
	template <typename Rng> bool operator()(Rng &rng) {
	if (UNLIKELY(0 == m_counter)) {
	m_rand = std::uniform_int_distribution<U>{}(rng);
	m_counter = sizeof(U) * 8;
	}
	--m_counter;
	bool ret = m_rand & 1;
	m_rand >>= 1;
	return ret;
	}

	private:
	U m_rand;
	int m_counter = 0;
	};

	template <typename U> class RandomizerWithMaskShift {
	public:
	template <typename Rng> bool operator()(Rng &rng) {
	if (UNLIKELY(0 == m_mask)) {
	m_rand = std::uniform_int_distribution<U>{}(rng);
	m_mask = 1;
	}
	bool ret = m_rand & m_mask;
	m_mask <<= 1;
	return ret;
	}

	private:
	U m_rand;
	U m_mask = 0;
	};

	template <typename U> class RandomizerWithRotl {
	public:
	template <typename Rng> bool operator()(Rng &rng) {
	if (UNLIKELY(1 == m_mask)) {
	m_rand = std::uniform_int_distribution<U>{}(rng);
	}
	m_mask = rotl(m_mask, 1);
	return m_rand & m_mask;
	}

	private:
	U m_rand;
	U m_mask = 1;
	};

	template <typename B, typename Rng>
	NO_INLINE void benchU4(size_t iters, const char rng_name, const char method_name) {
	size_t loops = iters / 4;
	size_t s = 0;

	std::vector<double> times;
	for (size_t i = 0; i < 7; ++i) {
	Rng rng{123};
	B b;
	auto t_start = std::chrono::high_resolution_clock::now();
	for (size_t t = 0; t < loops; ++t) {
	if (b(rng)) {
	++s;
	}
	if (b(rng)) {
	++s;
	}
	if (b(rng)) {
	++s;
	}
	if (b(rng)) {
	++s;
	}
	}
	auto t_end = std::chrono::high_resolution_clock::now();
	auto t_sec = std::chrono::duration<double>(t_end - t_start).count();
	times.push_back(t_sec / iters * 1e9);
	}
	std::sort(times.begin(), times.end());
	for (auto t : times) {
	std::cout << t << "; ";
	}
	std::cout << sizeof(B) << "; \"" << method_name << "\"; \"" << rng_name << "\"; 4; " << s
	<< std::endl;
	}

	template <typename B, typename Rng>
	NO_INLINE void benchU1(size_t iters, const char rng_name, const char method_name) {
	size_t s = 0;
	std::vector<double> times;
	for (size_t i = 0; i < 7; ++i) {
	Rng rng{123};
	B b;
	auto t_start = std::chrono::high_resolution_clock::now();
	for (size_t t = 0; t < iters; ++t) {
	if (b(rng)) {
	++s;
	}
	}
	auto t_end = std::chrono::high_resolution_clock::now();
	auto t_sec = std::chrono::duration<double>(t_end - t_start).count();
	times.push_back(t_sec / iters * 1e9);
	}
	std::sort(times.begin(), times.end());
	for (auto t : times) {
	std::cout << t << "; ";
	}
	std::cout << sizeof(B) << "; \"" << method_name << "\"; \"" << rng_name << "\"; 1; " << s
	<< std::endl;
	}

	template <typename Rng> void bench(size_t iters, const char *rng_name) {
	benchU1<RandomizerWithCounterT<uint64_t>, Rng>(iters, rng_name,
	"RandomizerWithCounterT<uint64_t>");
	benchU4<RandomizerWithCounterT<uint64_t>, Rng>(iters, rng_name,
	"RandomizerWithCounterT<uint64_t>");
	benchU1<BinaryAndUnbiased<uint64_t>, Rng>(iters, rng_name, "BinaryAndUnbiased<uint64_t>");
	benchU4<BinaryAndUnbiased<uint64_t>, Rng>(iters, rng_name, "BinaryAndUnbiased<uint64_t>");

	benchU1<RandomizerWithShiftT<uint64_t>, Rng>(iters, rng_name, "RandomizerWithShiftT<uint64_t>");
	benchU4<RandomizerWithShiftT<uint64_t>, Rng>(iters, rng_name, "RandomizerWithShiftT<uint64_t>");

	benchU1<RandomizerWithShiftMaxSize<uint64_t>, Rng>(iters, rng_name,
	"RandomizerWithShiftMaxSize<uint64_t>");
	benchU4<RandomizerWithShiftMaxSize<uint64_t>, Rng>(iters, rng_name,
	"RandomizerWithShiftMaxSize<uint64_t>");

	benchU1<RandomizerWithMaskShift<uint64_t>, Rng>(iters, rng_name,
	"RandomizerWithMaskShift<uint64_t>");
	benchU4<RandomizerWithMaskShift<uint64_t>, Rng>(iters, rng_name,
	"RandomizerWithMaskShift<uint64_t>");

	benchU1<RandomizerWithRotl<uint64_t>, Rng>(iters, rng_name, "RandomizerWithRotl<uint64_t>");
	benchU4<RandomizerWithRotl<uint64_t>, Rng>(iters, rng_name, "RandomizerWithRotl<uint64_t>");

	benchU1<RandomizerWithCounter2<uint64_t>, Rng>(iters, rng_name,
	"RandomizerWithCounter2<uint64_t>");
	benchU4<RandomizerWithCounter2<uint64_t>, Rng>(iters, rng_name,
	"RandomizerWithCounter2<uint64_t>");

	benchU1<RandomizerWithShiftTNoLikely<uint64_t>, Rng>(iters, rng_name,
	"RandomizerWithShiftTNoLikely<uint64_t>");
	benchU4<RandomizerWithShiftTNoLikely<uint64_t>, Rng>(iters, rng_name,
	"RandomizerWithShiftTNoLikely<uint64_t>");

	benchU1<RandomizerWithCounterShiftT<uint64_t>, Rng>(iters, rng_name,
	"RandomizerWithCounterShiftT<uint64_t>");
	benchU4<RandomizerWithCounterShiftT<uint64_t>, Rng>(iters, rng_name,
	"RandomizerWithCounterShiftT<uint64_t>");
	benchU1<BinaryAnd, Rng>(iters, rng_name, "BinaryAnd");
	benchU4<BinaryAnd, Rng>(iters, rng_name, "BinaryAnd");

	benchU1<BoundedOperator, Rng>(iters, rng_name, "BoundedOperator");
	benchU4<BoundedOperator, Rng>(iters, rng_name, "BoundedOperator");

	benchU1<BernoulliDistribution, Rng>(iters, rng_name, "BernoulliDistribution");
	benchU4<BernoulliDistribution, Rng>(iters, rng_name, "BernoulliDistribution");

	benchU1<UniformDistribution, Rng>(iters, rng_name, "UniformDistribution");
	benchU4<UniformDistribution, Rng>(iters, rng_name, "UniformDistribution");
	}

	int main(int, char **) {
	size_t iters = UINT64_C(1'000'000'000);
	bench<sfc64>(iters, "sfc64");
	bench<std::mt19937>(iters, "std::mt19937");
	bench<std::mt19937_64>(iters, "std::mt19937_64");
	}