#ifndef __MPI_CVL__
#define __MPI_CVL__

#include<mpi.h>
#include<civlc.cvh>
#include<comm.cvh>
#include<concurrency.cvh>
#include<string.h>

//TODO make a Datatype struct, which has a field "int size;" Define one of these objects for MPI_INT, MPI_DOUBLE, etc. 
//TODO Then provide methods like MPI provides for creating new ones.
//TODO then support MPI_Type_contig(datatype, int n).

#define BCAST_TAG   999
#define REDUCE_TAG  998

/* Completed definition for mpi-common.h */
struct MPI_Status{
  int MPI_SOURCE;
  int MPI_TAG;
  int MPI_ERROR;
  int size;
};

/* Definition of CIVL-MPI */
typedef enum {
  __UNINIT,
  __INIT,
  __FINALIZED
}__MPI_Sys_status__;

struct MPI_Request{
  int id;
};

/* Definition of CMPI_Gcomm and MPI_Comm */
typedef struct CMPI_Gcomm {
  $gcomm p2p; // point-to-point communication
  $gcomm col; // collective communication
  $gbarrier gbarrier;
} CMPI_Gcomm;

struct MPI_Comm {
  $comm p2p; // point-to-point communication
  $comm col; // collective communication
  $barrier barrier;
  __MPI_Sys_status__ status;
};

/********************************** State **************************************/

/* The number of times the MPI_Wtime function has been called */
int CMPI_time_count = 0;

/****************************** Helper Functions **********************************/
int sizeofDatatype(MPI_Datatype datatype) {
  switch (datatype) {
  case MPI_INT:
    return sizeof(int);
  case MPI_FLOAT:
    return sizeof(float);
  case MPI_DOUBLE:
    return sizeof(double);
  case MPI_CHAR:
    return sizeof(char);
  default:
    $assert(0, "Unreachable");
  }
}

/************************** MPI LIB Implementations *******************************/

$abstract double CMPI_time(int count);

double MPI_Wtime() {
  double result = CMPI_time(CMPI_time_count);
  
  CMPI_time_count++;
  return result;
}

CMPI_Gcomm CMPI_Gcomm_create($scope scope, int size) {
  CMPI_Gcomm result;

  result.p2p = $gcomm_create(scope, size);
  result.col = $gcomm_create(scope, size);
  result.gbarrier = $gbarrier_create(scope, size);
  return result;
}

void CMPI_Gcomm_destroy(CMPI_Gcomm gc) {
  $gcomm_destroy(gc.p2p);
  $gcomm_destroy(gc.col);
  $gbarrier_destroy(gc.gbarrier);
}

MPI_Comm CMPI_Comm_create($scope scope, CMPI_Gcomm gc, int rank) {
  MPI_Comm result;
  
  result.p2p = $comm_create(scope, gc.p2p, rank);
  result.col = $comm_create(scope, gc.col, rank);
  result.barrier = $barrier_create(scope, gc.gbarrier, rank);
  result.status = __UNINIT;
  return result;
}

void CMPI_Comm_destroy(MPI_Comm comm) {
  $comm_destroy(comm.p2p);
  $comm_destroy(comm.col);
  $barrier_destroy(comm.barrier);
}

int __MPI_Init(MPI_Comm *comm) {
  comm->status = __INIT;
  return 0;
}

int __MPI_Finalize(MPI_Comm *comm) {
  comm->status = __FINALIZED;
  return 0;
}

int MPI_Comm_size(MPI_Comm comm, int *size) {
  $assert(comm.status == __INIT, "MPI_Comm_size() cannot be invoked without MPI_Init() being called before.\n");
  *size = $comm_size(comm.p2p);
  return 0;
}

int MPI_Comm_rank(MPI_Comm comm, int *rank) {
  $assert(comm.status == __INIT, "MPI_Comm_rank() cannot be invoked without MPI_Init() being called before.\n");
  *rank = $comm_place(comm.p2p);
  return 0;
}


int CMPI_Send(void *buf, int count, MPI_Datatype datatype, int dest,
	      int tag, $comm comm) {
  if (dest >= 0) {
    int size = count*sizeofDatatype(datatype);
    int place = $comm_place(comm);
    $message out = $message_pack(place, dest, tag, buf, size);
    $comm_enqueue(comm, out);
  }
  return 0;
}

int MPI_Send(void *buf, int count, MPI_Datatype datatype, int dest,
	     int tag, MPI_Comm comm) {
  $assert(comm.status == __INIT, "MPI_Send() cannot be invoked without MPI_Init() being called before.\n");
  return CMPI_Send(buf, count, datatype, dest, tag, comm.p2p);
}


int CMPI_Recv(void *buf, int count, MPI_Datatype datatype, int source,
	      int tag, $comm comm, MPI_Status *status) {
  if (source >= 0 || source == MPI_ANY_SOURCE) {
    $message in = $comm_dequeue(comm, source, tag);
    int size = count*sizeofDatatype(datatype);

    $message_unpack(in, buf, size);
    if (status != MPI_STATUS_IGNORE) {
      status->size = $message_size(in);
      status->MPI_SOURCE = $message_source(in);
      status->MPI_TAG = $message_tag(in);
      status->MPI_ERROR = 0;
    }
  }
  return 0;
}

int MPI_Recv(void *buf, int count, MPI_Datatype datatype, int source,
	     int tag, MPI_Comm comm, MPI_Status *status) {
  $assert(comm.status == __INIT,
	  "MPI_Recv() cannot be invoked without "
	  "MPI_Init() being called before.\n");
  return CMPI_Recv(buf, count, datatype, source, tag, comm.p2p, status);
}

int MPI_Get_count(MPI_Status *status, MPI_Datatype datatype, int *count) {
  *count = status->size/sizeofDatatype(datatype);
  return 0;
}

int MPI_Sendrecv(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) {
  $assert(comm.status == __INIT,
	  "MPI_Sendrecv() cannot be invoked "
	  "without MPI_Init() being called before.\n");
  // not correct for checking potential deadlock...rewrite:
  MPI_Send(sendbuf, sendcount, sendtype, dest, sendtag, comm);
  MPI_Recv(recvbuf, recvcount, recvtype, source, recvtag, comm, status);
  return 0;
}

/* Broadcasts a message from root to everyone else.
 * Need to use a differnt comm.
 */
int MPI_Bcast(void *buf, int count, MPI_Datatype datatype, int root,
	      MPI_Comm comm) {
  $assert(comm.status == __INIT,
	  "MPI_Bcast() cannot be invoked without MPI_Init() "
	  "being called before.\n");
  if ($comm_place(comm.col) == root) {
    int nprocs = $comm_size(comm.col);

    for (int i=0; i<nprocs; i++)
      if (i != root)
	CMPI_Send(buf, count, datatype, i, BCAST_TAG, comm.col);
  } else {
    CMPI_Recv(buf, count, datatype, root, BCAST_TAG, comm.col,
	      MPI_STATUS_IGNORE);
  }
  return 0;
}

/* Reduces values on all processes to a single value */
int MPI_Reduce(void* sendbuf, void* recvbuf, int count,
	       MPI_Datatype datatype, MPI_Op op, int root,
	       MPI_Comm comm) {
  int rank;

  $assert(comm.status == __INIT,
	  "MPI_Reduce() cannot be invoked without "
	  "MPI_Init() being called before.\n");
  rank = $comm_place(comm.col);
  if (rank != root)
    CMPI_Send(sendbuf, count, datatype, root, REDUCE_TAG, comm.col);
  else {
    int nprocs = $comm_size(comm.col);
    int size;

    for (int i = 0; i<nprocs; i++) {
      if(i == root) continue;
      else{
	$message in = $comm_dequeue(comm.col, i, REDUCE_TAG);
	size = count * sizeofDatatype(datatype);

	/* the third argument "count" indicates the number of cells needs doing the
	   operation. */
	$bundle_unpack_apply(in.data, sendbuf, count, op);
	$assert(in.size <= size, 
		"Message of size %d exceeds the specified size %d.",
		in.size, size);
      }
    }
    size = count * sizeofDatatype(datatype);
    memcpy(recvbuf, sendbuf, size);
  }
  return 0;
}

/* Combines values from all processes and distributes the result back to all processes */
/* default root is 0 */
int MPI_Allreduce(void* sendbuf, void* recvbuf, int count,
		  MPI_Datatype datatype,
		  MPI_Op op, MPI_Comm comm) {
  int root = 0;
  MPI_Status status;
  
  $assert(comm.status == __INIT,
	  "MPI_Allreduce() cannot be invoked without "
	  "MPI_Init() being called before.\n");
  MPI_Reduce(sendbuf, recvbuf, count, datatype, op, root, comm);
  MPI_Bcast(recvbuf, count, datatype, root, comm);
  return 0;
}

int MPI_Barrier(MPI_Comm comm){

  $assert(comm.status == __INIT, "MPI_Allreduce() cannot be invoked without MPI_Init() being called before.\n");
  $barrier_call(comm.barrier);
}
#endif