#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>

#define ORDER_FORWARDS 1
#define ORDER_BACKWARDS 2

#define GENERIC_MAX(x, y) ((x) > (y) ? (x) : (y))
#define GENERIC_MIN(x, y) ((x) < (y) ? (x) : (y))

#define ENSURE_int(i) _Generic((i), int : (i))
#define ENSURE_float(f) _Generic((f), float : (f))

#define MAX(type, x, y) (type) GENERIC_MAX(ENSURE_##type(x), ENSURE_##type(y))
#define MIN(type, x, y) (type) GENERIC_MIN(ENSURE_##type(x), ENSURE_##type(y))

void local_quicksort(int *arr, int lo, int hi);
char *string_of_list(int *arr, int len);

int main(int argc, char **argv) {
  int rank, p;
  MPI_Init(&argc, &argv);

  int n = atoi(argv[1]);

  MPI_Comm_rank(MPI_COMM_WORLD, &rank);
  MPI_Comm_size(MPI_COMM_WORLD, &p);

  // Generate integers
  int n_over_p = n / p;
  int integers[n_over_p];

  // Important implementation detail: srand(0) is specially handled by glibc to
  // behave as if it was called with srand(1). To get around this, I'm seeding
  // with rank + 1
  //
  // See more: https://stackoverflow.com/a/27386563
  srand(rank + 1);

  for (int i = 0; i < n_over_p; ++i) {
    // TODO: For readability during debugging, I'm capping this
    integers[i] = rand() % 101;
    // printf(" - %d\n", integers[i]);
  }

  int group_root = 0;

  // Locally sort
  printf("[%d] Numbers before:           %s\n", rank,
         string_of_list(integers, n_over_p));
  local_quicksort(integers, 0, n_over_p);
  printf("[%d] Numbers after first sort: %s\n", rank,
         string_of_list(integers, n_over_p));

  // Select a pivot.
  // This pivot is broadcasted to all nodes
  int pivot;

  // The pivot is selected as the median (see chp. 9.4.4)
  // Not the real median though, need an existing element of the array
  pivot = integers[n_over_p / 2];
  MPI_Bcast(&pivot, 1, MPI_INT, 0, MPI_COMM_WORLD);

  printf("Median: %d\n", pivot);

  // Determine where the boundary between S (lower) and L (higher) lies
  int boundary;
  for (int i = 0; i < n_over_p; ++i) {
    if (integers[i] >= pivot) {
      boundary = i;
      break;
    }
  }
  int S_lo = 0, S_hi = boundary - 1;
  int L_lo = boundary, L_hi = n_over_p - 1;
  int S_size = S_hi - S_lo + 1, L_size = L_hi - L_lo + 1;
  printf("[%d] S: [%d - %d] (%d), L: [%d - %d] (%d)\n", rank, S_lo, S_hi,
         S_size, L_lo, L_hi, L_size);

  // Perform global arrangement
  int S_global_end, L_reverse_end;
  MPI_Scan(&S_size, &S_global_end, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);
  MPI_Scan(&L_size, &L_reverse_end, 1, MPI_INT, MPI_SUM, MPI_COMM_WORLD);

  int S_global_start = S_global_end - S_size,
      L_reverse_start = L_reverse_end - L_size,
      L_global_start = n - L_reverse_end, L_global_end = n - L_reverse_start;
  printf("[%d] S: [%d - %d], L: [%d - %d]\n", rank, S_global_start,
         S_global_end - 1, L_global_start, L_global_end - 1);

  // Send it to the correct target
  int S_starting_process = S_global_start / n_over_p;
  int S_offset = S_global_start % n_over_p,
      L_offset = L_global_start % n_over_p;
  int i, local_start;

  for (i = S_starting_process, local_start = 0; i * n_over_p < S_global_end;
       ++i) {
    int processor_start = S_offset;
    int mod = S_global_end % n_over_p;
    int processor_end = MIN(int, n_over_p, mod == 0 ? n_over_p : mod);

    int total_start = i * n_over_p + processor_start;
    int total_end = i * n_over_p + processor_end;

    int n_bytes = total_end - total_start;
    int local_end = local_start + n_bytes;

    printf("[%d] - sending %d elements from local S [%d..%d] to destination S "
           "[%d..%d]\n",
           rank, n_bytes, local_start, local_end, processor_start,
           processor_end);
    // MPI_Sendrecv(const void *sendbuf, int sendcount, MPI_Datatype sendtype,
    //              int dest, int sendtag, void *recvbuf, int recvcount,
    //              MPI_Datatype recvtype, int source, int recvtag, MPI_Comm
    //              comm, MPI_Status *status)

    local_start += n_bytes;
  }

  int L_starting_process = L_global_start / n_over_p;
  for (i = L_starting_process; i * n_over_p < L_global_end; ++i) {
    int processor_start = L_offset;
    int mod = L_global_end % n_over_p;
    int processor_end = MIN(int, n_over_p, mod == 0 ? n_over_p : mod);

    int total_start = i * n_over_p + processor_start;
    int total_end = i * n_over_p + processor_end;

    int n_bytes = total_end - total_start;
    int local_end = local_start + n_bytes;

    printf("[%d] - sending %d elements from local L [%d..%d] to destination L "
           "[%d..%d]\n",
           rank, n_bytes, local_start, local_end, processor_start,
           processor_end);
  }

  // The first node is responsible for collecting all the data and then printing
  // it out to the file
  // MPI_Gather(const void *sendbuf, int sendcount, MPI_INT, void *recvbuf,
  //            int recvcount, MPI_INT, 0, MPI_COMM_WORLD);
  if (rank == 0) {
    FILE *f = fopen(argv[2], "w");
    fclose(f);
  }

  MPI_Finalize();
  return 0;
}

// hi not inclusive
void local_quicksort(int *arr, int lo, int hi) {
  int temp;

  if (lo >= hi || lo < 0)
    return;

  int pivot = arr[hi - 1];
  int pivot_idx = lo - 1;
  for (int j = lo; j < hi; ++j) {
    if (arr[j] < pivot) {
      pivot_idx += 1;

      temp = arr[j];
      arr[j] = arr[pivot_idx];
      arr[pivot_idx] = temp;
    }
  }

  pivot_idx += 1;
  temp = arr[hi - 1];
  arr[hi - 1] = arr[pivot_idx];
  arr[pivot_idx] = temp;

  // Recursive call
  local_quicksort(arr, lo, pivot_idx);
  local_quicksort(arr, pivot_idx + 1, hi);
}

char *string_of_list(int *arr, int len) {
  char *buffer = malloc(1000);
  int offset = 0; // Keep track of the current position in the buffer
  for (int i = 0; i < len; i++) {
    offset += sprintf(buffer + offset, "%d", arr[i]);
    if (i < len - 1) {
      // Add a separator (e.g., comma or space) if it's not the last element
      offset += sprintf(buffer + offset, " ");
    }
  }

  return buffer;
}