Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

C++: How to generate array to function pointers in a loop

Given

// from an external C api
void f(int i, void (*g)());

const int n = ...
void a0(); void a1(); ...
void (*a[n])();

int main()
{
  a[0] = a0; a[1] = a1; ...
  for (int i = 0; i != n; ++i)
    f(i, a[i]);
  ...
}

I don't want to generate each function a0, a1, ... and assign it to a separately. Instead I want to generate the functions and assign them to a in a loop, something like that (sorry for the hideous code, it won't compile):

for (int i = 0; i != n; ++i)
{
    void b() { cout << i; };
    a[i] = b;
}

Is this possible? How can I do it?

like image 283
Johannes Flügel Avatar asked Nov 27 '18 07:11

Johannes Flügel


2 Answers

Try like this:

#include <iostream>
#include <vector>
using namespace std;

// from an external C api
void f(int i, void(*g)())
{
    //g();
}

const int n = 100;
using fnType = void(*)();
vector<fnType> a(n);

template <int N>
void func()
{
    cout << N << endl;
}

template <int N>
void init()
{
    a[N - 1] = func<N - 1>;
    init<N - 1>();
}

template <>
void init<0>()
{
}


int main()
{
    init<n>();

    for(int i = 0; i < n; ++i)
        a[i]();
    for(int i = 0; i < n; ++i)
        f(i, a[i]);   
}

Also, you can define

vector<std::function<void()>> funcs(n);

and call it from f:

template <int N>
void func()
{
    //cout << N << endl;
    funcs[N]();
}

So, you can simply define it:

for(int k = 0;k < n;k++)
    funcs[k] = [k](){cout << k << endl;};
like image 61
snake_style Avatar answered Sep 25 '22 04:09

snake_style


No really sure it fits your use case, but I encountered a quite similar problem to create a templated C++ functions C wrapper. The demo shows how to create (at compile time) and reuse a std::array of function pointers.

Let imagine that you have a basic C++ function that computes sum of squares

template <std::size_t N>
double sum2_stat(const double* p)
{
  double s = 0;
  for (size_t i = 0; i < N; i++)
  {
    s += p[i] * p[i];
  }
  return s;
}

for efficiency reason N is a static size (known at compile time) that should allow compiler to do tricky optimizations (vectorize the loop...).

Now we also have a dynamic fallback, when N is too big or unknown at compile time

double sum2_dyn(const double* p, const std::size_t n)
{
  double s = 0;
  for (size_t i = 0; i < n; i++)
  {
    s += p[i] * p[i];
  }
  return s;
}

Now you want to create a C API. A naive approach would be to define something like:

extern "C" {

double sum2_naive(const double* p, const std::size_t n)
{
  assert(n >= 0);

  switch (n)
  {
    case 0:
      return sum2_stat<0>(p);
    case 1:
      return sum2_stat<1>(p);
    case 2:
      return sum2_stat<2>(p);
    case 3:
      return sum2_stat<3>(p);
    case 4:
      return sum2_stat<4>(p);
    case 5:
      return sum2_stat<5>(p);
    case 6:
      return sum2_stat<6>(p);
    case 7:
      return sum2_stat<7>(p);
    case 8:
      return sum2_stat<8>(p);
    default:
      return sum2_dyn(p, n);
  }
}
}

However this approach is tedious because you have to repeat yourself a lot and you can not automatically change Nmax=8 value.

Now a more elegant solution I suggest. First define some helpers to automatically create, at compile time, a static array of function pointers:

template <std::size_t... I>
constexpr auto sum2_call_helper(std::index_sequence<I...>)
{
  return std::array<double (*)(const double* p), sizeof...(I)>({&sum2_stat<I>...});
}

template <std::size_t N, typename Indices = std::make_index_sequence<N>>
constexpr auto sum2_call_helper()
{
  return sum2_call_helper(Indices());
}

Then define your C API:

extern "C" {

double sum2(const double* p, const std::size_t n)
{
  constexpr auto N_Max = 8;
  constexpr auto indirections = sum2_call_helper<N_Max + 1>();

  assert(N_Max >= 0);

  if (n <= N_Max)
  {
    return indirections[n](p);
  }

  return sum2_dyn(p, n);
}
}

There are clear advantages, you have a clean code and you can easily change Nmax value without further modifications of the code. Also note that you use a std::array and do not use std::function which minimize the risk of performance penalties.


I hope this partially answer your question. To adapt it to your problem you must have indexed a() functions (a<0>(), a<1>(), ...) as follows:

template <std::size_t INDEX>
... a(...)

and not (your example)

... a0(...)
... a1(...)
... a2(...)

If it is not the case I fear that you will have to write glue code as you mentioned in your question:

 a[0] = a0; a[1] = a1; 

Complete working example:

#include <array>
#include <cassert>
#include <iostream>
#include <utility>
#include <vector>

template <std::size_t N>
double sum2_stat(const double* p)
{
  double s = 0;
  for (size_t i = 0; i < N; i++)
  {
    s += p[i] * p[i];
  }
  return s;
}

template double sum2_stat<10>(const double*);

double sum2_dyn(const double* p, const std::size_t n)
{
  double s = 0;
  for (size_t i = 0; i < n; i++)
  {
    s += p[i] * p[i];
  }
  return s;
}

template <std::size_t... I>
constexpr auto sum2_call_helper(std::index_sequence<I...>)
{
  return std::array<double (*)(const double* p), sizeof...(I)>({&sum2_stat<I>...});
}

template <std::size_t N, typename Indices = std::make_index_sequence<N>>
constexpr auto sum2_call_helper()
{
  return sum2_call_helper(Indices());
}

extern "C" {

double sum2(const double* p, const std::size_t n)
{
  constexpr auto N_Max = 8;
  constexpr auto indirections = sum2_call_helper<N_Max + 1>();

  assert(N_Max >= 0);

  if (n <= N_Max)
  {
    return indirections[n](p);
  }

  return sum2_dyn(p, n);
}

double sum2_naive(const double* p, const std::size_t n)
{
  assert(n >= 0);

  switch (n)
  {
    case 0:
      return sum2_stat<0>(p);
    case 1:
      return sum2_stat<1>(p);
    case 2:
      return sum2_stat<2>(p);
    case 3:
      return sum2_stat<3>(p);
    case 4:
      return sum2_stat<4>(p);
    case 5:
      return sum2_stat<5>(p);
    case 6:
      return sum2_stat<6>(p);
    case 7:
      return sum2_stat<7>(p);
    case 8:
      return sum2_stat<8>(p);
    default:
      return sum2_dyn(p, n);
  }
}
}

int main()
{
  std::vector<double> buffer(100, 2);

  std::cout << "\n" << sum2(buffer.data(), 5);
  std::cout << "\n" << sum2(buffer.data(), 10);

  std::cout << "\n" << sum2_naive(buffer.data(), 5);
  std::cout << "\n" << sum2_naive(buffer.data(), 10);
}
like image 23
Picaud Vincent Avatar answered Sep 25 '22 04:09

Picaud Vincent