#ifdef __CIVL_MPI__
#else
#define __CIVL_MPI__

#include <concurrency.cvh>
#include <bundle.cvh>
#include <mpi.h>
#include <civlmpi.cvh>
#include <string.h>

/* Completed definition for mpi-common.h */
struct MPI_Request{
  int id;
};

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;

/************************** MPI LIB Implementations *******************************/
double MPI_Wtime() {
  double result = CMPI_time(CMPI_time_count);
  
  CMPI_time_count++;
  return result;
}

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 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 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;
}
/******************************** Collective ***********************************/
/* 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);
  return 0;
}

/* 1. If comm is an intracommunicator, each process (includes root process) sends the content
   of its send buffer to the root process. Root process receives the messages and stores 
   them in rank order 
   2. TODO: If comm is an intercommunicator, it's not supported yet */
int MPI_Gather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, 
	       void* recvbuf, int recvcount, MPI_Datatype recvtype, 
	       int root, MPI_Comm comm){
  int rank, nprocs;
  _Bool rootInPlace = $false;
  MPI_Status status;

  $assert comm.status == __INIT :
  "MPI_Gather() cannot be invoked without "
    "MPI_Init() being called before.\n";
  rank = $comm_place(comm.col);
  nprocs = $comm_size(comm.col);
  if(sendbuf == MPI_IN_PLACE){// Sendbuf is ignored at root
    $assert root == rank:
    "Only root can replace 'sendbuf' with 'MPI_IN_PLACE'.";
    rootInPlace = $true;
  }
  else
    MPI_Send(sendbuf, sendcount, sendtype, root, GATHER_TAG, comm);
  // sendbuf cannot be ignored
  if(rank == root){
    int real_recvcount;
    int offset;

    // For root process, check if sendtype is equal to 
    // recvtype which is required by MPI standard.
    $assert sendtype == recvtype :
    "MPI_Gather() asks for equality "
      "between 'sendtype' and 'recvtype'.";    
    for(int i=0; i<nprocs; i++){
      // Since currently we don't support pointer addition 
      // on non-array type obejcts, we don't need 
      // to care about datatype extent.
      offset = i * recvcount; 
      //If optional ignorance not used or it's not at root iteration, then we do MPI_Send()
      if(!rootInPlace || i != root) {
	MPI_Recv(recvbuf + offset, recvcount, recvtype, i, GATHER_TAG, comm, &status);
	MPI_Get_count(&status, recvtype, &real_recvcount);
	$assert real_recvcount == recvcount :
	"MPI_Gather() asks for equality between"
	  " the amount of data sent and the "
	  "amount of data received.";
      }
    }
  }
  return 0;
}

/* The inverse operation of MPI_Gather() */
int MPI_Scatter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, 
		void* recvbuf, int recvcount, MPI_Datatype recvtype, int root,
		MPI_Comm comm){
  int rank, nprocs;
  _Bool rootInPlace = $false;

  $assert comm.status == __INIT :
  "MPI_Scatter() cannot be invoked without "
    "MPI_Init() being called before.\n";
  rank = $comm_place(comm.col);
  nprocs = $comm_size(comm.col);
  if(recvbuf == MPI_IN_PLACE){// Sendbuf is ignored at root
    $assert root == rank:
    "Only root can replace 'recvbuf' with 'MPI_IN_PLACE'.";
    rootInPlace = $true;
  }
  if(rank == root){
    int offset;

    // For root process, check if sendtype is equal to  
    // recvtype which is required by MPI standard.
    $assert sendtype == recvtype :
    "MPI_Scatter() asks for equality "
      "between 'sendtype' and 'recvtype'.";
    for(int i=0; i<nprocs; i++){
      offset = i * sendcount;
      if(!rootInPlace || i != root)
	MPI_Send(sendbuf + offset, sendcount, sendtype, i, SCATTER_TAG, comm);
    }
  }
  if(!rootInPlace){//Already checked that (rootInPlace == true)==>(root==rank)
    int real_recvcount;
    MPI_Status status;
    
    MPI_Recv(recvbuf, recvcount, recvtype, root, SCATTER_TAG, comm, &status);
    MPI_Get_count(&status, recvtype, &real_recvcount);
    $assert real_recvcount == recvcount :
    "MPI_Gather() asks for equality between"
      " the amount of data sent and the "
      "amount of data received.";
  }
}


/* MPI_Gatherv extends the functionality of MPI_Gather by allowing a varying count of data to be sent to root process, since recvcounts is now an array.*/
int MPI_Gatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype,
		void* recvbuf, const int recvcounts[], const int displs[],
		MPI_Datatype recvtype, int root, MPI_Comm comm){
  int rank, nprocs;
  _Bool rootInPlace = $false;

  $assert comm.status == __INIT :
  "MPI_Gatherv() cannot be invoked without "
    "MPI_Init() being called before.\n";
  rank = $comm_place(comm.col);
  nprocs = $comm_size(comm.col);
  if(sendbuf == MPI_IN_PLACE){
    $assert root == rank:
    "Only root can replace 'sendbuf' with 'MPI_IN_PLACE'.";
    rootInPlace = $true;
  }else{
    MPI_Send(sendbuf, sendcount, sendtype, root, GATHERV_TAG, comm);
  }
  //Root receive
  if(rank == root){
    int real_recvcount;
    MPI_Status status;

    $assert sendtype == recvtype :
    "MPI_Gatherv() asks for equality "
      "between 'sendtype' and 'recvtype'.";
    for(int i=0; i<nprocs; i++){
      if(!rootInPlace || i != root){
	MPI_Recv(recvbuf + displs[i], recvcounts[i], recvtype, i, GATHERV_TAG, comm, &status);
	MPI_Get_count(&status, recvtype, &real_recvcount);
	$assert real_recvcount == recvcounts[i] :
	"MPI_Gatherv() asks for equality between"
	  " the amount of data sent and the "
	  "amount of data received.";
      }
    }
  }
}

/* MPI_Scatterv extends the functionality of MPI_Scatter by allowing a varying count of data to be sent to each process, since sendcounts is now an array.*/
int MPI_Scatterv(const void* sendbuf, const int sendcounts[], const
		 int displs[], MPI_Datatype sendtype, void* recvbuf,
		 int recvcount, MPI_Datatype recvtype, int root, MPI_Comm comm){
  int rank, nprocs;
  _Bool rootInPlace = $false;

  $assert comm.status == __INIT :
  "MPI_Scatterv() cannot be invoked without "
    "MPI_Init() being called before.\n";
  rank = $comm_place(comm.col);
  nprocs = $comm_size(comm.col);
  if(recvbuf == MPI_IN_PLACE){
    $assert root == rank:
    "Only root can replace 'recvbuf' with 'MPI_IN_PLACE'.";
    rootInPlace = $true;
  }
  if(rank == root){
    //For process root, check sendtype and recvtype
    $assert sendtype == recvtype :
    "MPI_Scatterv() asks for equality "
      "between 'sendtype' and 'recvtype'.";

    for(int i=0; i<nprocs; i++){
      if(!rootInPlace || i != root)
	MPI_Send(sendbuf + displs[i], sendcounts[i], sendtype, i, 
		 SCATTERV_TAG, comm);
    }
  }
  if(!rootInPlace){
    MPI_Status status;
    int real_recvcount;

    MPI_Recv(recvbuf, recvcount, recvtype, root, SCATTERV_TAG, comm, &status);
    MPI_Get_count(&status, recvtype, &real_recvcount);
    $assert real_recvcount == recvcount :
    "MPI_Gather() asks for equality between"
      " the amount of data sent and the "
      "amount of data received.";
  }
}

#endif