#include<config.h>
CPS_START_NAMESPACE

//====================================================================
//*  SUI 3/27/97
//*  slice_sum.C
//*  last modified 7/15/04
//====================================================================

CPS_END_NAMESPACE
#include<comms/glb.h>
#include<util/smalloc.h>
#include<util/gjp.h>
#include <comms/sysfunc_cps.h>
#include <comms/scu.h>
#include <comms/bgl_net.h>
#include <sys/bgl/bgl_sys_all.h>
CPS_START_NAMESPACE


//-------------------------------------------------------------------
//* sum over a slice(hyperplane) which is orthogonal to the direction
//* "dir". There are "blcklength" summations to be done:
//* float_p[i] = sum_over_nodes_of_this_slice(float_p[i])
//-------------------------------------------------------------------

//-------------------------------------------------------------------
/*!
  The vector pointed to by \a float_p is summed over each 3-dimensional
  hyperplane of nodes which is perpendiculat to the \a dir direction

  \param float_p The number to be summed.
  \param blcklength The number of floating point numbers in the vector.
  \param dir The normal direction defining the hyperplane; one of {0, 1, 2, 3} 
  corresponding to {x, y, z, t}.
  \post The vector sum is written back to \a float_p, which is identical on
  all nodes in this hyperplane.

  \ingroup comms
*/
//-------------------------------------------------------------------

void slice_sum(Float * float_p, int blcklength, int dir)
{
  char *cname = "slice_sum";
  char *fname = "slice_sum(*float_p, int, int)";

  int NP[4] = { GJP.Xnodes(), GJP.Ynodes(), GJP.Znodes(), GJP.Tnodes() };
  const int MAX=30;

  if (blcklength > MAX)
    ERR.General(cname, fname, "blcklength (%d) too big > MAX (%d) \n",blcklength, MAX);
		    
  IFloat *transmit_buf = (IFloat *)smalloc(blcklength*sizeof(IFloat));

  // added by manke
  if(transmit_buf == 0)
    ERR.Pointer(cname,fname, "transmit_buf");
  VRB.Smalloc(cname,fname, "transmit_buf", transmit_buf, blcklength*sizeof(IFloat));
  // end added

  IFloat *receive_buf  = (IFloat *)smalloc(blcklength*sizeof(IFloat));

  // added by manke
  if(receive_buf == 0)
    ERR.Pointer(cname,fname, "receive_buf");
  VRB.Smalloc(cname,fname, "receive_buf", receive_buf, blcklength*sizeof(IFloat));
  // end added

  IFloat *transmit_buf_p;
    IFloat *receive_buf_p;
  IFloat *free_buffer_p;  // buffer to be used as next receive 

  int count;		  
	// loop index where 0 <= count < blcklength 
  int itmp;		  
	// loop index with 1<= itmp < NP[i] 
  int i;

  // Sum.If torus the result is in all nodes. 
  // If mesh the result is in the node with highest coordinates

  transmit_buf_p = transmit_buf;
  receive_buf_p = receive_buf;
  free_buffer_p = transmit_buf_p;

  for(i = 0; i < 4; ++i) {

    if(i == dir) continue;

    //--------------------------------------------------------------
    // address of buffer of to be sent (data on this node) 
    //--------------------------------------------------------------
    transmit_buf_p = (IFloat *)float_p; 
    
    //--------------------------------------------------------------
    // tranmit & receive NP[i] - 1 times in snd_dir[i] direction
    //--------------------------------------------------------------
    for ( itmp = 1; itmp < NP[i]; itmp++) {
      
      //-----------------------------------------------------------
      // do SCU transfers
      //-----------------------------------------------------------
      getMinusData(receive_buf_p, transmit_buf_p, blcklength, i);
      
      //-----------------------------------------------------------
      // accumulate received data	
      //-----------------------------------------------------------
      for (count = 0; count < blcklength; ++count) 
	float_p[count] += receive_buf_p[count];
      
      //-----------------------------------------------------------
      // the received data will be sent out     
      // the free buffer will be used to receive      
      //-----------------------------------------------------------
      transmit_buf_p = receive_buf_p;
      receive_buf_p = free_buffer_p;
      
      //-----------------------------------------------------------
      // transmit_buf WILL be free buffer NEXT round of transmit
      //-----------------------------------------------------------
      free_buffer_p = transmit_buf_p;
    }
  }


  // Set the local node contribution to 0 unless the node has the highest 
  // coordinate within the subvolume.
  if(       dir == 0){
    if( !( (GJP.Ynodes() == GJP.YnodeCoor()+1) &&
	   (GJP.Znodes() == GJP.ZnodeCoor()+1) &&
	   (GJP.Tnodes() == GJP.TnodeCoor()+1) )) {
      for (count = 0; count < blcklength; ++count) {
	float_p[count] = 0;
      }
    }
  } else if(dir == 1){
    if( !( (GJP.Xnodes() == GJP.XnodeCoor()+1) &&
	   (GJP.Znodes() == GJP.ZnodeCoor()+1) &&
	   (GJP.Tnodes() == GJP.TnodeCoor()+1) )) {
      for (count = 0; count < blcklength; ++count) {
	float_p[count] = 0;
      }
    }
  } else if(dir == 2){
    if( !( (GJP.Xnodes() == GJP.XnodeCoor()+1) &&
	   (GJP.Ynodes() == GJP.YnodeCoor()+1) &&
	   (GJP.Tnodes() == GJP.TnodeCoor()+1) )) {
      for (count = 0; count < blcklength; ++count) {
	float_p[count] = 0;
      }
    }
  } else if(dir == 3){
    if( !( (GJP.Xnodes() == GJP.XnodeCoor()+1) &&
	   (GJP.Ynodes() == GJP.YnodeCoor()+1) &&
	   (GJP.Znodes() == GJP.ZnodeCoor()+1) )) {
      for (count = 0; count < blcklength; ++count) {
	float_p[count] = 0;
      }
    }

  } 



  // Now resum back

  transmit_buf_p = transmit_buf;
  receive_buf_p = receive_buf;
  free_buffer_p = transmit_buf_p;

  for(i = 0; i < 4; ++i) {
    
    if(i == dir) continue;
    
    //--------------------------------------------------------------
    // address of buffer of to be sent (data on this node) 
    //--------------------------------------------------------------
    transmit_buf_p = (IFloat *)float_p; 
    
    //--------------------------------------------------------------
    // tranmit & receive NP[i] - 1 times in snd_dir[i] direction
    //--------------------------------------------------------------
    for ( itmp = 1; itmp < NP[i]; itmp++) {
      
      //-----------------------------------------------------------
      // do SCU transfers
      //-----------------------------------------------------------
      getPlusData(receive_buf_p, transmit_buf_p, blcklength, i);
      
      //-----------------------------------------------------------
      // accumulate received data	
      //-----------------------------------------------------------
      for (count = 0; count < blcklength; ++count) 
	float_p[count] += receive_buf_p[count];
      
      //-----------------------------------------------------------
      // the received data will be sent out     
      // the free buffer will be used to receive      
      //-----------------------------------------------------------
      transmit_buf_p = receive_buf_p;
      receive_buf_p = free_buffer_p;
      
      //-----------------------------------------------------------
      // transmit_buf WILL be free buffer NEXT round of transmit
      //-----------------------------------------------------------
      free_buffer_p = transmit_buf_p;
    }
  }



  sfree(transmit_buf);
  sfree(receive_buf);
}


CPS_END_NAMESPACE