#include <mpi.h>

#include "riemannsum.h"
#include "util.h"

double riemannsum(int num_intervals, int mpi_rank, int mpi_world_size,
                  int threads_per_process) {
  double pi = 0;
  double h = 1.0 / (double)num_intervals;

  // TODO: Parallelize the code using mpi_world_size processes (1 process per
  // node.
  // In total, (mpi_world_size * threads_per_process) threads will collaborate
  // to compute the Riemann sum.

  int from =
      1 + ((num_intervals + mpi_world_size - 1) / mpi_world_size) * mpi_rank;
  int to =
      ((num_intervals + mpi_world_size - 1) / mpi_world_size) * (mpi_rank + 1);
  if (num_intervals < to)
    to = num_intervals;

#pragma omp parallel for num_threads(threads_per_process) reduction(+ : pi)
  for (int i = from; i <= to; i++) {
    double x = h * ((double)i - 0.5);
    pi += h * f(x);
  }

  double pi_sum = 0.0;
  MPI_Reduce(&pi, &pi_sum, 1, MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
  pi = pi_sum;

  // Rank 0 should return the estimated PI value
  // Other processes can return any value (don't care)
  return pi;
}