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

#define ORDER_FORWARDS 1
#define ORDER_BACKWARDS 2
#define CTL_SIZE 4

#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 init_ctl(int *ctl, int len);
void local_quicksort(int *arr, int lo, int hi);
char *string_of_list(int *arr, int len);
void recursive_quicksort(int *integers, int n, int root, MPI_Comm comm);

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];

  // Minor 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]);
  }

  recursive_quicksort(integers, n, 0, MPI_COMM_WORLD);

  // sleep(1);
  // printf("[%d] after: %s\n", rank, string_of_list(integers, n_over_p));

  // 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);
  int recvbuf[n];
  MPI_Gather(integers, n_over_p, MPI_INT, recvbuf, n_over_p, MPI_INT, 0,
             MPI_COMM_WORLD);

  if (rank == 0) {
    FILE *f = fopen(argv[2], "w");
    // printf("integers: %s\n", string_of_list(recvbuf, n));
    printf("[%d] ==== FINAL ====\n", rank);
    for (int i = 0; i < p; i += 1) {
      printf("[%d]  %s\n", rank,
             string_of_list(&recvbuf[i * n_over_p], n_over_p));
    }
    fclose(f);
  }

  MPI_Finalize();
  printf("Done.\n");
  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 = calloc(sizeof(char), 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;
}

void recursive_quicksort(int *integers, int n, int root, MPI_Comm comm) {
  int rank, p;
  MPI_Comm_size(comm, &p);
  MPI_Comm_rank(comm, &rank);

  if (p == 1) {
    // Recursion base case: just sort it serially
    local_quicksort(integers, 0, n);
    printf("Quicksorted: %s\n", string_of_list(integers, n));
    return;
  }

  sleep(1);
  printf("\n\n");
  int n_over_p_max = (n + p - 1) / p;
  int n_over_p = n / p;
  if (rank == root)
    n_over_p += n - p * n_over_p;
  // printf(
  //     "[%d] :::::::::::::::::::::::::::: RECURSIVE QUICKSORT (n=%d,
  //     n/p=%d)\n", rank, n, n_over_p);

  // 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;
  {
    // First, select a random element
    int rand_el = integers[rand() % n_over_p];

    // Gather it
    int rand_els[p];
    MPI_Gather(&rand_el, 1, MPI_INT, rand_els, 1, MPI_INT, root, comm);

    // Get the median
    if (rank == root) {
      // Sort
      local_quicksort(rand_els, 0, p);

      // Get the middle element
      pivot = rand_els[p / 2];
    }

    MPI_Bcast(&pivot, 1, MPI_INT, root, comm);
  }
  printf("[%d] Broadcasted pivot: %d\n", rank, 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;
  int L_lo = boundary, L_hi = n_over_p;
  int S_size = S_hi - S_lo, L_size = L_hi - L_lo;
  // 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, S_global_max_end;
  MPI_Scan(&S_size, &S_global_end, 1, MPI_INT, MPI_SUM, comm);
  MPI_Scan(&L_size, &L_reverse_end, 1, MPI_INT, MPI_SUM, comm);

  // printf("[%d] bruh %d\n", rank, S_global_end);
  // Get the boundary element between S and L
  MPI_Allreduce(&S_global_end, &S_global_max_end, 1, MPI_INT, MPI_MAX, comm);

  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] Prefixed S: [%d - %d], Prefixed L: [%d - %d]\n", rank,
  //        S_global_start, S_global_end - 1, L_global_start, L_global_end - 1);

  int S_starting_process = S_global_start / n_over_p,
      L_starting_process = L_global_start / n_over_p;
  int S_offset = S_global_start % n_over_p,
      L_offset = L_global_start % n_over_p;

  int *integers_recv_buf = calloc(sizeof(int), n);
  int S_ctl[p * CTL_SIZE];
  int L_ctl[p * CTL_SIZE];
  int S_send_ctl[p * CTL_SIZE];
  int L_send_ctl[p * CTL_SIZE];
  int recvpart[n_over_p];
  int ctl_send_counts[p];
  int ctl_send_displs[p];

  int send_counts[p];
  int send_displs[p];
  int recv_counts[p];
  int recv_displs[p];

  init_ctl(S_ctl, p);
  init_ctl(L_ctl, p);
  init_ctl(S_send_ctl, p);
  init_ctl(L_send_ctl, p);

  for (int i = 0; i < p; ++i) {
    send_counts[i] = n_over_p;
    send_displs[i] = i * n_over_p;

    ctl_send_counts[i] = CTL_SIZE;
    ctl_send_displs[i] = i * CTL_SIZE;
    recv_counts[i] = CTL_SIZE;
    recv_displs[i] = i * CTL_SIZE;
  }

  // Send S to the correct target
  {
    for (int i = S_lo, dest_pos = S_global_start,
             processor = S_starting_process;
         i < S_hi;) {
      int next_break = MIN(int, S_global_end,
                           MIN(int, dest_pos + (S_hi - S_lo),
                               (dest_pos / n_over_p) * n_over_p + n_over_p));
      int count = next_break - dest_pos;

      int from_local_start = i, from_local_end = i + count;
      int from_global_start = rank * n_over_p + from_local_start,
          from_global_end = from_global_start + count;

      int to_global_start = dest_pos, to_global_end = dest_pos + count;
      int to_local_start = to_global_start - processor * n_over_p,
          to_local_end = to_global_end - processor * n_over_p;

      // printf("[%d] S ->> (count=%d), from local [%d..%d] {%d..%d} -to-> "
      //        "p#%d [%d..%d] {%d..%d}\n",
      //        rank, count, from_local_start, from_local_end,
      //        from_global_start, from_global_end, processor, to_local_start,
      //        to_local_end, to_global_start, to_global_end);
      S_send_ctl[processor * CTL_SIZE] = count;
      S_send_ctl[processor * CTL_SIZE + 1] = from_global_start;
      S_send_ctl[processor * CTL_SIZE + 2] = to_local_start;
      S_send_ctl[processor * CTL_SIZE + 3] = from_local_start;

      i += count;
      dest_pos += count;
      processor += 1;
    }

    MPI_Alltoallv(S_send_ctl, ctl_send_counts, ctl_send_displs, MPI_INT, S_ctl,
                  recv_counts, recv_displs, MPI_INT, comm);
  }

  // Send L to the correct target
  {
    for (int i = L_lo, dest_pos = L_global_start,
             processor = L_starting_process;
         i < L_hi;) {
      int next_break = MIN(int, L_global_end,
                           MIN(int, dest_pos + (L_hi - L_lo),
                               (dest_pos / n_over_p) * n_over_p + n_over_p));
      int count = next_break - dest_pos;

      int from_local_start = i, from_local_end = i + count;
      int from_global_start = rank * n_over_p + from_local_start,
          from_global_end = from_global_start + count;

      int to_global_start = dest_pos, to_global_end = dest_pos + count;
      int to_local_start = to_global_start - processor * n_over_p,
          to_local_end = to_global_end - processor * n_over_p;

      // printf("[%d] L ->> (count=%d), from local [%d..%d] {%d..%d} -to-> "
      //        "p#%d [%d..%d] {%d..%d}\n",
      //        rank, count, from_local_start, from_local_end,
      //        from_global_start, from_global_end, processor, to_local_start,
      //        to_local_end, to_global_start, to_global_end);
      L_send_ctl[processor * CTL_SIZE] = count;
      L_send_ctl[processor * CTL_SIZE + 1] = from_global_start;
      L_send_ctl[processor * CTL_SIZE + 2] = to_local_start;
      L_send_ctl[processor * CTL_SIZE + 3] = from_local_start;

      i += count;
      dest_pos += count;
      processor += 1;
    }

    MPI_Alltoallv(L_send_ctl, ctl_send_counts, ctl_send_displs, MPI_INT, L_ctl,
                  recv_counts, recv_displs, MPI_INT, comm);
  }

  // After sending S and L information

  for (int i = 0; i < p; ++i) {
    recv_counts[i] = n_over_p;
    recv_displs[i] = i * n_over_p;
  }

  // MPI_Alltoallv(integers, send_counts, send_displs, MPI_INT,
  // integers_recv_buf,
  //               recv_counts, recv_displs, MPI_INT, MPI_COMM_WORLD);
  // MPI_Allgather(integers, n_over_p, MPI_INT, integers_recv_buf, n_over_p,
  //               MPI_INT, comm);
  // printf("[%d] ints: %s\n", rank, string_of_list(integers_recv_buf, n));

  // Scheme for all send
  int integers_recv_2[n_over_p];
  int integers_recv_3[n_over_p];
  for (int i = 0; i < n_over_p; ++i) {
    integers_recv_2[i] = -1;
    integers_recv_3[i] = integers[i];
  }

  for (int host_p = 0; host_p < p; ++host_p) {
    if (rank == host_p) {
      // Your {S,L}_ctl is a mapping from source_processor -> ctl
      // Everyone already knows who needs to send to who now
      for (int sender_p = 0; sender_p < p; ++sender_p) {
        int S_count = S_ctl[sender_p * CTL_SIZE];
        if (S_count > 0) {
          int to_local_start = S_ctl[sender_p * CTL_SIZE + 2];
          int from_local_start = S_ctl[sender_p * CTL_SIZE + 3];

          if (sender_p == host_p) {
            for (int k = 0; k < S_count; ++k) {
              integers_recv_3[to_local_start + k] =
                  integers[from_local_start + k];
            }
            continue;
          }

          // printf("[%d] - S inbound from host %d to [%d..%d] (%d)\n", rank,
          //        sender_p, to_local_start, to_local_start + S_count,
          //        S_count);
          MPI_Recv(&integers_recv_2[to_local_start], S_count, MPI_INT, sender_p,
                   124, comm, MPI_STATUS_IGNORE);
          for (int k = 0; k < S_count; ++k) {
            integers_recv_3[to_local_start + k] =
                integers_recv_2[to_local_start + k];
          }
        }
      }
    } else {
      // Your {S,L}_send_ctl contains a mapping from dest_processor -> ctl
      for (int dest_p = 0; dest_p < p; ++dest_p) {
        int S_count = S_send_ctl[dest_p * CTL_SIZE];
        if (S_count > 0 && dest_p == host_p) {
          int from_local_start = S_send_ctl[dest_p * CTL_SIZE + 3];
          // printf("[%d] - S outbound to host %d from [%d..%d] (%d)\n", rank,
          //        dest_p, from_local_start, from_local_start + S_count,
          //        S_count);
          MPI_Send(&integers[from_local_start], S_count, MPI_INT, dest_p, 124,
                   comm);
        }
      }
    }
  }

  for (int host_p = 0; host_p < p; ++host_p) {
    if (rank == host_p) {
      // Your {S,L}_ctl is a mapping from source_processor -> ctl
      // Everyone already knows who needs to send to who now
      for (int sender_p = 0; sender_p < p; ++sender_p) {
        int L_count = L_ctl[sender_p * CTL_SIZE];
        if (L_count > 0) {
          int to_local_start = L_ctl[sender_p * CTL_SIZE + 2];
          int from_local_start = L_ctl[sender_p * CTL_SIZE + 3];

          if (sender_p == host_p) {
            for (int k = 0; k < L_count; ++k) {
              integers_recv_3[to_local_start + k] =
                  integers[from_local_start + k];
            }
            continue;
          }

          // printf("[%d] - L inbound from host %d to [%d..%d] (%d)\n", rank,
          //        sender_p, to_local_start, to_local_start + L_count,
          //        L_count);
          MPI_Recv(&integers_recv_2[to_local_start], L_count, MPI_INT, sender_p,
                   125, comm, MPI_STATUS_IGNORE);
          for (int k = 0; k < L_count; ++k) {
            integers_recv_3[to_local_start + k] =
                integers_recv_2[to_local_start + k];
          }
        }
      }
    } else {
      // Your {S,L}_send_ctl contains a mapping from dest_processor -> ctl
      for (int dest_p = 0; dest_p < p; ++dest_p) {
        int L_count = L_send_ctl[dest_p * CTL_SIZE];
        if (L_count > 0 && dest_p == host_p) {
          int from_local_start = L_send_ctl[dest_p * CTL_SIZE + 3];
          // printf("[%d] - L outbound to host %d from [%d..%d] (%d)\n", rank,
          //        dest_p, from_local_start, from_local_start + L_count,
          //        L_count);
          MPI_Send(&integers[from_local_start], L_count, MPI_INT, dest_p, 125,
                   comm);
        }
      }
    }
  }

  printf("[%d] after: %s\n", rank, string_of_list(integers_recv_3, n_over_p));
  for (int i = 0; i < n_over_p; ++i) {
    integers[i] = integers_recv_3[i];
  }

  // Now, determine which processes should be responsible for taking the S and L
  // arrays

  // Specifically, the part where it's split, break the tie to see if it goes
  // down or up
  int colors[p];
  int p_of_split = S_global_max_end / n_over_p;
  int split_point = S_global_max_end % n_over_p;
  // printf("[%d] p_of_split = %d / %d = %d\n", rank, S_global_max_end,
  // n_over_p,
  //        p_of_split);
  int S_split_add = split_point, L_split_sub = n_over_p - split_point;

  int lo_start = 0, lo_end;
  int hi_start, hi_end = p;
  if (split_point > n_over_p / 2) {
    // Belongs to the lower group
    lo_end = hi_start = p_of_split + 1;
  } else {
    // Belongs to the higher group
    lo_end = hi_start = p_of_split;
  }

  int child_root = -1;
  for (int i = 0; i < p; ++i) {
    if (i < lo_end)
      colors[i] = 100;
    else {
      colors[i] = 200;
      if (child_root == -1)
        child_root = i;
    }
  }

  // MPI_Comm child;
  // MPI_Comm_split(comm, colors[rank], rank, &child);
  // printf("[%d] Recursing...\n", rank);

  // int child_size;
  // MPI_Comm_size(child, &child_size);

  // int start_at = 0, new_n = child_size * n_over_p;
  // if (colors[rank] == 100) {
  //   new_n += S_split_add;
  // } else {
  //   new_n -= L_split_sub;
  //   if (rank == p_of_split)
  //     start_at = split_point;
  // }
  // recursive_quicksort(integers, n, child_root, child);

  // printf("[%d] Done recursing.\n", rank);
  // MPI_Comm_free(&child);
}

void init_ctl(int *ctl, int len) {
  for (int i = 0; i < len; ++i) {
    for (int j = 0; j < CTL_SIZE; ++j) {
      ctl[i * CTL_SIZE + j] = -1;
    }
  }
}