MPI: added an additional line at the upper MPI-bound for correct material averaging

This also fixes the current send and receive issue
--> Engine_MPI::SendReceiveCurrents()

FDTD/engine_mpi.cpp

@@ -57,11 +57,11 @@ void Engine_MPI::Init()
 
 			if (m_Op_MPI->m_NeighborDown[n]>=0)
 			{
-				m_BufferDown[n] = new float[m_BufferSize[n]*3];
+				m_BufferDown[n] = new float[m_BufferSize[n]*2];
 			}
 			if (m_Op_MPI->m_NeighborUp[n]>=0)
 			{
-				m_BufferUp[n] = new float[m_BufferSize[n]*3];
+				m_BufferUp[n] = new float[m_BufferSize[n]*2];
 			}
 		}
 	}
@@ -100,7 +100,7 @@ void Engine_MPI::SendReceiveVoltages()
 
 		//send voltages
 		unsigned int iPos=0;
-		pos[n]=numLines[n]-1;
+		pos[n]=numLines[n]-2;
 		if (m_Op_MPI->m_NeighborUp[n]>=0)
 		{
 			for (pos[nP]=0; pos[nP]<numLines[nP]; ++pos[nP])
@@ -136,14 +136,6 @@ void Engine_MPI::SendReceiveVoltages()
 
 void Engine_MPI::SendReceiveCurrents()
 {
-	/*
-	TODO:
-	  the FDTD engine could update the normal currents (e.g. i_z at z_max) (magnetic fields) on the last line, but the default engine does not need to...
-	  workaround: transfer all three current components
-	  drawback: data transfer is larger than necessary
-	  future fix: update normal currents
-	  */
-
 	if (!m_Op_MPI->GetMPIEnabled())
 		return;
 
@@ -152,7 +144,7 @@ void Engine_MPI::SendReceiveCurrents()
 	//non-blocking prepare for receive...
 	for (int n=0;n<3;++n)
 		if (m_Op_MPI->m_NeighborUp[n]>=0)
-			MPI_Irecv( m_BufferUp[n] , m_BufferSize[n]*3, MPI_FLOAT, m_Op_MPI->m_NeighborUp[n], m_Op_MPI->m_MyTag, MPI_COMM_WORLD, &Recv_Request[n]);
+			MPI_Irecv( m_BufferUp[n] , m_BufferSize[n]*2, MPI_FLOAT, m_Op_MPI->m_NeighborUp[n], m_Op_MPI->m_MyTag, MPI_COMM_WORLD, &Recv_Request[n]);
 
 	for (int n=0;n<3;++n)
 	{
@@ -168,16 +160,15 @@ void Engine_MPI::SendReceiveCurrents()
 			{
 				for (pos[nPP]=0; pos[nPP]<numLines[nPP]; ++pos[nPP])
 				{
-					m_BufferDown[n][iPos++] = Engine_SSE_Compressed::GetCurr(n ,pos);
 					m_BufferDown[n][iPos++] = Engine_SSE_Compressed::GetCurr(nP ,pos);
 					m_BufferDown[n][iPos++] = Engine_SSE_Compressed::GetCurr(nPP,pos);
 				}
 			}
-			MPI_Isend( m_BufferDown[n] , m_BufferSize[n]*3, MPI_FLOAT, m_Op_MPI->m_NeighborDown[n], m_Op_MPI->m_MyTag, MPI_COMM_WORLD, &Send_Request[n]);
+			MPI_Isend( m_BufferDown[n] , m_BufferSize[n]*2, MPI_FLOAT, m_Op_MPI->m_NeighborDown[n], m_Op_MPI->m_MyTag, MPI_COMM_WORLD, &Send_Request[n]);
 		}
 
 		//receive currents
-		pos[n]=numLines[n]-1;
+		pos[n]=numLines[n]-2;
 		iPos=0;
 		if (m_Op_MPI->m_NeighborUp[n]>=0)
 		{
@@ -187,7 +178,6 @@ void Engine_MPI::SendReceiveCurrents()
 			{
 				for (pos[nPP]=0; pos[nPP]<numLines[nPP]; ++pos[nPP])
 				{
-					Engine_SSE_Compressed::SetCurr(n  ,pos,m_BufferUp[n][iPos++]);
 					Engine_SSE_Compressed::SetCurr(nP ,pos,m_BufferUp[n][iPos++]);
 					Engine_SSE_Compressed::SetCurr(nPP,pos,m_BufferUp[n][iPos++]);
 				}

FDTD/openems_fdtd_mpi.cpp

@@ -239,17 +239,29 @@ bool openEMS_FDTD_MPI::SetupMPI(TiXmlElement* FDTD_Opts)
 					if (i>0)
 						m_MPI_Op->SetNeighborDown(0,procTable[i-1][j][k]);
 					if (i<SplitNumber[0].size()-2)
+					{
+						//add one additional line
+						grid->AddDiscLine(0, m_Original_Grid->GetLine(0,SplitNumber[0].at(i+1)+1 ));
 						m_MPI_Op->SetNeighborUp(0,procTable[i+1][j][k]);
+					}
 
 					if (j>0)
 						m_MPI_Op->SetNeighborDown(1,procTable[i][j-1][k]);
 					if (j<SplitNumber[1].size()-2)
+					{
+						//add one additional line
+						grid->AddDiscLine(1, m_Original_Grid->GetLine(1,SplitNumber[1].at(j+1)+1 ));
 						m_MPI_Op->SetNeighborUp(1,procTable[i][j+1][k]);
+					}
 
 					if (k>0)
 						m_MPI_Op->SetNeighborDown(2,procTable[i][j][k-1]);
 					if (k<SplitNumber[2].size()-2)
+					{
+						//add one additional line
+						grid->AddDiscLine(2, m_Original_Grid->GetLine(2,SplitNumber[2].at(k+1)+1 ));
 						m_MPI_Op->SetNeighborUp(2,procTable[i][j][k+1]);
+					}
 
 				}
 			}

FDTD/operator_cylinder.cpp

@@ -61,9 +61,9 @@ inline unsigned int Operator_Cylinder::GetNumberOfLines(int ny) const
 {
 	//this is necessary for a correct field processing... cylindrical engine has to reset this by adding +1
 	if (CC_closedAlpha && ny==1)
-		return numLines[1]-1;
+		return Operator_Multithread::GetNumberOfLines(ny)-1;
 
-	return numLines[ny];
+	return Operator_Multithread::GetNumberOfLines(ny);
 }
 
 string Operator_Cylinder::GetDirName(int ny) const

FDTD/operator_mpi.cpp

@@ -77,20 +77,6 @@ void Operator_MPI::SetBoundaryCondition(int* BCs)
 	Operator_SSE_Compressed::SetBoundaryCondition(BCs);
 }
 
-void Operator_MPI::ApplyElectricBC(bool* dirs)
-{
-	if (!m_MPI_Enabled)
-		return Operator_SSE_Compressed::ApplyElectricBC(dirs);
-
-	for (int n=0;n<3;++n)
-	{
-		//do not delete operator at upper inteface
-		if (m_NeighborUp[n]>=0)
-			dirs[2*n+1] = false;
-	}
-	Operator_SSE_Compressed::ApplyElectricBC(dirs);
-}
-
 Engine* Operator_MPI::CreateEngine() const
 {
 	if (m_MPI_Enabled)
@@ -171,6 +157,14 @@ void Operator_MPI::SetOriginalMesh(CSRectGrid* orig_Mesh)
 	}
 }
 
+unsigned int Operator_MPI::GetNumberOfLines(int ny) const
+{
+	if ((!m_MPI_Enabled) || (m_NeighborUp[ny]<0))
+		return Operator_SSE_Compressed::GetNumberOfLines(ny);
+
+	return Operator_SSE_Compressed::GetNumberOfLines(ny)-1;
+}
+
 string Operator_MPI::PrependRank(string name)
 {
 	stringstream out_name;

FDTD/operator_mpi.h

@@ -31,7 +31,6 @@ public:
 	bool GetMPIEnabled() const {return m_MPI_Enabled;}
 
 	virtual void SetBoundaryCondition(int* BCs);
-	virtual void ApplyElectricBC(bool* dirs);
 
 	virtual Engine* CreateEngine() const;
 
@@ -45,6 +44,8 @@ public:
 	virtual void SetSplitPos(int ny, unsigned int pos) {m_SplitPos[ny]=pos;}
 	virtual void SetOriginalMesh(CSRectGrid* orig_Mesh);
 
+	virtual unsigned int GetNumberOfLines(int ny) const;
+
 protected:
 	Operator_MPI();
 	bool m_MPI_Enabled;