src/kudu/util/random.h

// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#ifndef KUDU_UTIL_RANDOM_H_
#define KUDU_UTIL_RANDOM_H_

#include <cmath>
#include <cstdint>
#include <mutex>
#include <random>
#include <vector>

#include "kudu/gutil/casts.h"
#include "kudu/gutil/map-util.h"
#include "kudu/util/int128.h"
#include "kudu/util/locks.h"

namespace kudu {

namespace random_internal {

static const uint32_t M = 2147483647L;   // 2^31-1

} // namespace random_internal

template<class R>
class StdUniformRNG;

// A very simple random number generator.  Not especially good at
// generating truly random bits, but good enough for our needs in this
// package. This implementation is not thread-safe.
class Random {
 private:
  uint32_t seed_;
 public:
  explicit Random(uint32_t s) {
    Reset(s);
  }

  // Reset the RNG to the given seed value.
  void Reset(uint32_t s) {
    seed_ = s & 0x7fffffffu;
    // Avoid bad seeds.
    if (seed_ == 0 || seed_ == random_internal::M) {
      seed_ = 1;
    }
  }

  // Next pseudo-random 32-bit unsigned integer.
  // FIXME: This currently only generates 31 bits of randomness.
  // The MSB will always be zero.
  uint32_t Next() {
    static const uint64_t A = 16807;  // bits 14, 8, 7, 5, 2, 1, 0
    // We are computing
    //       seed_ = (seed_ * A) % M,    where M = 2^31-1
    //
    // seed_ must not be zero or M, or else all subsequent computed values
    // will be zero or M respectively.  For all other values, seed_ will end
    // up cycling through every number in [1,M-1]
    uint64_t product = seed_ * A;

    // Compute (product % M) using the fact that ((x << 31) % M) == x.
    seed_ = static_cast<uint32_t>((product >> 31) + (product & random_internal::M));
    // The first reduction may overflow by 1 bit, so we may need to
    // repeat.  mod == M is not possible; using > allows the faster
    // sign-bit-based test.
    if (seed_ > random_internal::M) {
      seed_ -= random_internal::M;
    }
    return seed_;
  }

  // Alias for consistency with Next64
  uint32_t Next32() { return Next(); }

  // Next pseudo-random 64-bit unsigned integer.
  uint64_t Next64() {
    uint64_t large = Next();
    large <<= 31;
    large |= Next();
    // Fill in the highest two MSBs.
    large |= implicit_cast<uint64_t>(Next32()) << 62;
    return large;
  }

  // Next pseudo-random 128-bit unsigned integer.
  uint128_t Next128() {
    uint128_t large = Next64();
    large <<= 64U;
    large |= Next64();
    // Next64() provides entire range of numbers up to 64th bit.
    // So unlike Next64(), no need to fill MSB bit(s).
    return large;
  }

  // Returns a uniformly distributed value in the range [0..n-1]
  // REQUIRES: n > 0
  uint32_t Uniform(uint32_t n) { return Next() % n; }

  // Alias for consistency with Uniform64
  uint32_t Uniform32(uint32_t n) { return Uniform(n); }

  // Returns a uniformly distributed 64-bit value in the range [0..n-1]
  // REQUIRES: n > 0
  uint64_t Uniform64(uint64_t n) { return Next64() % n; }

  // Returns a uniformly distributed 128-bit value in the range [0..n-1]
  uint128_t Uniform128(uint128_t n) { return Next128() % n; }

  // Randomly returns true ~"1/n" of the time, and false otherwise.
  // REQUIRES: n > 0
  bool OneIn(int n) { return (Next() % n) == 0; }

  // Skewed: pick "base" uniformly from range [0,max_log] and then
  // return "base" random bits.  The effect is to pick a number in the
  // range [0,2^max_log-1] with exponential bias towards smaller numbers.
  uint32_t Skewed(int max_log) {
    return Uniform(1 << Uniform(max_log + 1));
  }

  // Samples a random number from the given normal distribution.
  double Normal(double mean, double std_dev);

  // Return a random number between 0.0 and 1.0 inclusive.
  double NextDoubleFraction() {
    return Next() / static_cast<double>(random_internal::M + 1.0);
  }
};

// Thread-safe wrapper around Random.
class ThreadSafeRandom {
 public:
  explicit ThreadSafeRandom(uint32_t s)
      : random_(s) {
  }

  void Reset(uint32_t s) {
    std::lock_guard<simple_spinlock> l(lock_);
    random_.Reset(s);
  }

  uint32_t Next() {
    std::lock_guard<simple_spinlock> l(lock_);
    return random_.Next();
  }

  uint32_t Next32() {
    std::lock_guard<simple_spinlock> l(lock_);
    return random_.Next32();
  }

  uint64_t Next64() {
    std::lock_guard<simple_spinlock> l(lock_);
    return random_.Next64();
  }

  uint128_t Next128() {
    std::lock_guard<simple_spinlock> l(lock_);
    return random_.Next128();
  }

  uint32_t Uniform(uint32_t n) {
    std::lock_guard<simple_spinlock> l(lock_);
    return random_.Uniform(n);
  }

  uint32_t Uniform32(uint32_t n) {
    std::lock_guard<simple_spinlock> l(lock_);
    return random_.Uniform32(n);
  }

  uint64_t Uniform64(uint64_t n) {
    std::lock_guard<simple_spinlock> l(lock_);
    return random_.Uniform64(n);
  }

  uint128_t Uniform128(uint128_t n) {
    std::lock_guard<simple_spinlock> l(lock_);
    return random_.Uniform128(n);
  }

  bool OneIn(int n) {
    std::lock_guard<simple_spinlock> l(lock_);
    return random_.OneIn(n);
  }

  uint32_t Skewed(int max_log) {
    std::lock_guard<simple_spinlock> l(lock_);
    return random_.Skewed(max_log);
  }

  double Normal(double mean, double std_dev) {
    std::lock_guard<simple_spinlock> l(lock_);
    return random_.Normal(mean, std_dev);
  }

  double NextDoubleFraction() {
    std::lock_guard<simple_spinlock> l(lock_);
    return random_.NextDoubleFraction();
  }

 private:
  simple_spinlock lock_;
  Random random_;
};

// Wraps either Random or ThreadSafeRandom as a C++ standard library
// compliant UniformRandomNumberGenerator:
//   http://en.cppreference.com/w/cpp/concept/UniformRandomNumberGenerator
template<class R>
class StdUniformRNG {
 public:
  typedef uint32_t result_type;

  explicit StdUniformRNG(R* r) : r_(r) {}
  uint32_t operator()() {
    return r_->Next32();
  }
  constexpr static uint32_t min() { return 0; }
  constexpr static uint32_t max() { return (1L << 31) - 1; }

 private:
  R* r_;
};

// Defined outside the class to make use of StdUniformRNG above.
inline double Random::Normal(double mean, double std_dev) {
  std::normal_distribution<> nd(mean, std_dev);
  StdUniformRNG<Random> gen(this);
  return nd(gen);
}

// Generate next random integer with data-type specified by IntType template
// parameter which must be a 32-bit, 64-bit, or 128-bit unsigned integer. This constexpr function
// is useful when generating random numbers in a loop with template parameter
// and avoids the run-time cost of determining Next32() v/s Next64() v/s Next128() in RNG.
//
// Note: This constexpr function can't be a member function of
// Random/ThreadSafeRandom class because it invokes non-const member function.
template<typename IntType, class RNG>
constexpr IntType GetNextRandom(RNG* rand) {
  // type_traits not defined for 128-bit int data types, hence not checking for integral value.
  static_assert(sizeof(IntType) == 4 || sizeof(IntType) == 8 || sizeof(IntType) == 16,
                "Only 32-bit, 64-bit, or 128-bit integers supported");

  // if/switch statement disallowed in constexpr function in C++11.
  return sizeof(IntType) == 4 ? rand->Next32() :
                                sizeof(IntType) == 8 ? rand->Next64() : rand->Next128();
}

// Generate next random integer in the [0, n-1] range with data-type specified by IntType template
// parameter which must be a 32-bit, 64-bit, or 128-bit unsigned integer.
// Note: Same comments apply to this constexpr function as GetNextRandom() function above.
template<typename IntType, class RNG>
constexpr IntType GetNextUniformRandom(IntType n, RNG* rand) {
  // type_traits not defined for 128-bit int data types, hence not checking for integral value.
  static_assert(sizeof(n) == 4 || sizeof(n) == 8 || sizeof(n) == 16,
                "Only 32-bit, 64-bit, or 128-bit integers generated");

  // if/switch statement disallowed in constexpr function in C++11.
  return sizeof(n) == 4 ? rand->Uniform32(n) :
                          sizeof(n) == 8 ? rand->Uniform64(n) : rand->Uniform128(n);
}

}  // namespace kudu

#endif  // KUDU_UTIL_RANDOM_H_