[peyman.havaej]

Hello Foamers,

I have written a new boundary condition in which I have to solve a semi-infinite integral to obtain a face center value, e.g. if there are 1000 faces on a patch, an integral should be solved 1000 times. Hence, it is really a time-consuming calculation, and I do require a parallel calculation for it.
The code works perfectly in the serial mode, but the parallel computation is not the same as serial calculation. I have found that it is because of a nested loop used in the code, a loop within another loop. For a loop, the "Reduce" operate should be used to collect data from all processors to a single processor.

I assumed that I have to use the reduce operator after each loop, but in this case, the solver goes blank and the calculation is not continued after the first loop. However, suppose the reduce operator is applied for the second loop only, in that case, the calculated values are different to the single processors' computation.

I am looking to know how to use the MPI Reduce operator in the parallel calculation of a nested loop?
Any comments or suggestions will be greatly appreciated.


Best regards,
Peyman

void Foam::CarslawFvPatchScalarField::SijCoeffientsCalculationMatrix(
    scalarField& dTNew,
    scalarField& SIJ_
)
{
vectorField	BMeshV = patch().Cf();
const scalarField& magS = patch().magSf();
scalar dij;
scalar cij;

    forAll(SIJ_, i) 
 {
    	SIJ_ = 0;
		
	 forAll(SIJ_, j) 
	     {
			cij = magS[j] /thickness_;//mag(deltaPatch[j].x());
			
			dij =  BMeshV[i].x() - BMeshV[j].x();
			SIJ_[j] = (dij>0)? cij /mag(sqrt(dij)):0 ;

			SIJ_[j] = SIJ_[j] * 1.0/sqrt(mathematicalConstant::pi * kappaS_ *rhoS_ *CpS_ * Us_);
		
		}
	// reduce(SIJ_,  sumOp<scalarField>()); //  *** First reduce operator ***


	dTNew[i] = sum(SIJ_ * heatFlux_);  
 }
	reduce(dTNew ,  sumOp<scalarField>()); //  *** Second reduce operator***

 	Info<<"Maximum dTNew is \n"<<max(dTNew)<<endl;
 	Info<<"Minimum dTNew is \n"<<min(dTNew)<<endl;

}