The type std::tuple can bundle any number of values, potentially including values of different types, into a single return object:

std::tuple<int, int, int, int> foo(int a, int b) { // or auto (C++14)
   return std::make_tuple(a + b, a - b, a * b, a / b);
}

In C++17, a braced initializer list can be used:

std::tuple<int, int, int, int> foo(int a, int b)    {
    return {a + b, a - b, a * b, a / b};
}

Retrieving values from the returned tuple can be cumbersome, requiring the use of the std::get template function:

auto mrvs = foo(5, 12);
auto add = std::get<0>(mrvs);
auto sub = std::get<1>(mrvs);
auto mul = std::get<2>(mrvs);
auto div = std::get<3>(mrvs);

If the types can be declared before the function returns, then std::tie can be employed to unpack a tuple into existing variables:

int add, sub, mul, div;
std::tie(add, sub, mul, div) = foo(5, 12);

If one of the returned values is not needed, std::ignore can be used:

std::tie(add, sub, std::ignore, div) = foo(5, 12);

auto [add, sub, mul, div] = foo(5,12);

If you want to return a tuple of lvalue references instead of a tuple of values, use std::tie in place of std::make_tuple.

std::tuple<int&, int&> minmax( int& a, int& b ) {
  if (b<a)
    return std::tie(b,a);
  else
    return std::tie(a,b);
}

which permits

void increase_least(int& a, int& b) {
  std::get<0>(minmax(a,b))++;
}

In some rare cases you'll use std::forward_as_tuple instead of std::tie; be careful if you do so, as temporaries may not last long enough to be consumed.

The container std::array can bundle together a fixed number of return values. This number has to be known at compile-time and all return values have to be of the same type:

std::array<int, 4> bar(int a, int b) {
    return { a + b, a - b, a * b, a / b };
}

This replaces c style arrays of the form int bar[4]. The advantage being that various c++ std functions can now be used on it. It also provides useful member functions like at which is a safe member access function with bound checking, and size which allows you to return the size of the array without calculation.

#include <utility>
std::pair<int, int> foo(int a, int b) {
    return {a+b, a-b};
}

struct foo_return_type {
    int add;
    int sub;
    int mul;
    int div;
};

foo_return_type foo(int a, int b) {
    return {a + b, a - b, a * b, a / b};
}

auto calc = foo(5, 12);

Instead of assignment to individual fields, a constructor can be used to simplify the constructing of returned values:

struct foo_return_type {
    int add;
    int sub;
    int mul;
    int div;
    foo_return_type(int add, int sub, int mul, int div)
    : add(add), sub(sub), mul(mul), div(div) {}
};

foo_return_type foo(int a, int b) {
     return foo_return_type(a + b, a - b, a * b, a / b);
}

foo_return_type calc = foo(5, 12);

In order to unpack a struct returned from a function, structured bindings can be used. This places the out-parameters on an even footing with the in-parameters:

int a=5, b=12;
auto[add, sub, mul, div] = foo(a, b);
std::cout << add << ' ' << sub << ' ' << mul << ' ' << div << '\n';

The output of this code is identical to that above. The struct is still used to return the values from the function. This permits you do deal with the fields individually.

C++17 introduces structured bindings, which makes it even easier to deal with multiple return types, as you do not need to rely upon std::tie() or do any manual tuple unpacking:

std::map<std::string, int> m;

// insert an element into the map and check if insertion succeeded
auto [iterator, success] = m.insert({"Hello", 42});

if (success) {
    // your code goes here
}

// iterate over all elements without having to use the cryptic 'first' and 'second' names
for (auto const& [key, value] : m) {
    std::cout << "The value for " << key << " is " << value << '\n';
}

Structured bindings can be used by default with std::pair, std::tuple, and any type whose non-static data members are all either public direct members or members of an unambiguous base class:

struct A { int x; };
struct B : A { int y; };
B foo();

// with structured bindings
const auto [x, y] = foo();

// equivalent code without structured bindings
const auto result = foo();
auto& x = result.x;
auto& y = result.y;

If you make your type "tuple-like" it will also automatically work with your type. A tuple-like is a type with appropriate tuple_size, tuple_element and get written:

namespace my_ns {
    struct my_type {
        int x;
        double d;
        std::string s;
    };
    struct my_type_view {
        my_type* ptr;
    };
}

namespace std {
    template<>
    struct tuple_size<my_ns::my_type_view> : std::integral_constant<std::size_t, 3>
    {};

    template<> struct tuple_element<my_ns::my_type_view, 0>{ using type = int; };
    template<> struct tuple_element<my_ns::my_type_view, 1>{ using type = double; };
    template<> struct tuple_element<my_ns::my_type_view, 2>{ using type = std::string; };
}

namespace my_ns {
    template<std::size_t I>
    decltype(auto) get(my_type_view const& v) {
        if constexpr (I == 0)
            return v.ptr->x;
        else if constexpr (I == 1)
            return v.ptr->d;
        else if constexpr (I == 2)
            return v.ptr->s;
        static_assert(I < 3, "Only 3 elements");
    }
}

now this works:

my_ns::my_type t{1, 3.14, "hello world"};

my_ns::my_type_view foo() {
    return {&t};
}

int main() {
    auto[x, d, s] = foo();
    std::cout << x << ',' << d << ',' << s << '\n';
}

template <class F>
void foo(int a, int b, F consumer) {
    consumer(a + b, a - b, a * b, a / b);
}

// use is simple... ignoring some results is possible as well
foo(5, 12, [](int sum, int , int , int ){
    std::cout << "sum is " << sum << '\n';
});

This is known as "continuation passing style".

You can adapt a function returning a tuple into a continuation passing style function via:

template<class Tuple>
struct continuation {
  Tuple t;
  template<class F>
  decltype(auto) operator->*(F&& f)&&{
    return std::apply( std::forward<F>(f), std::move(t) );
  }
};
std::tuple<int,int,int,int> foo(int a, int b);

continuation(foo(5,12))->*[](int sum, auto&&...) {
  std::cout << "sum is " << sum << '\n';
};

with more complex versions being writable in C++14 or C++11.