[MamboJambo]

Hello to all,


How can one print the information in a processor sequentially?

Let us assume:

const volVectorField& C = mesh.C();

I would like to have all information in processor 1, 2, 3 ...,N, sequentially.


If I do:

Pout << C << endl;

It will get mixed up.

[dlahaye]

Wrap below in loop over processor id?

// retrieve processor ID 
label n=Pstream::nProcs();
printf(' my processor id = %d\n', n); 

// print data on a given processor ID only as a first test 
if (n==1) {
	  // print statement goes here
	  }
}

[olesen]

If you want it completely sorted, you will need to serialize this yourself. That means that on the master process first print its values, then loop over the remaining number of processes, receiving and writing their values. On the non-master processes, they send their contents to the master.
For example, https://develop.openfoam.com/Develop...mplates.C#L186

[olesen]

Originally Posted by dlahaye

Wrap below in loop over processor id?

// retrieve processor ID 
label n=Pstream::nProcs();
printf(' my processor id = %d\n', n); 

// print data on a given processor ID only as a first test 
if (n==1) {
	  // print statement goes here
	  }
}

This will not work.

[HPE]

Unless I misunderstood your question, I sometimes do the following to group few things (not tested, and it should not resolve the sequential issue - see Mark's comment - yet still might be useful to print a specific processor field):

// a geometric field from upstream, say U (velocity field)
// or some other parallelised object

if (Pstream::parRun())
{
    for (label i = 0; i < Pstream::nProcs(); ++i)
    {
        if (i == Pstream::myProcNo()) // if ( i == 2)
        {
            for (const auto& u : U)
            {
                Sout<< u << nl;    // or Pout << ...
            }
            Sout<< endl;
            // or Sout << U << endl;
        }
    }
}
else
{
    for (const auto& u : U)
    {
        Info<< u << nl;
    }
    Info<< endl;
}

[MamboJambo]

Hello to all and thank you for your input.



For my application this sufficed:

for (label i = 0; i < Pstream::nProcs(); ++i)

if (i == Pstream::myProcNo()) 
{
// Loop through field/boundar, etc..
         for (const auto& u : U)
            {
                Pout<< u << nl;
            }
        Pout<< “Finish reading data in processor ” << i << endl;
 }

}

[MamboJambo]

Originally Posted by olesen

If you want it completely sorted, you will need to serialize this yourself. That means that on the master process first print its values, then loop over the remaining number of processes, receiving and writing their values. On the non-master processes, they send their contents to the master.
For example, https://develop.openfoam.com/Develop...mplates.C#L186

This also seems interesting:
I am using OF2106 and the same function reads:

template<class Type>
void Foam::vtk::writeListParallel
(
    vtk::formatter& fmt,
    const UList<Type>& values
)
{
    if (Pstream::master())
    {
        vtk::writeList(fmt, values);

        List<Type> recv;

        // Receive and write
        for (const int slave : Pstream::subProcs())
        {
            IPstream fromSlave(Pstream::commsTypes::blocking, slave);

            fromSlave >> recv;

            vtk::writeList(fmt, recv);
        }
    }
    else
    {
        // Send to master
        OPstream toMaster
        (
            Pstream::commsTypes::blocking,
            Pstream::masterNo()
        );

        toMaster << values;
    }
}

So it would stay, something like this?

    if (Pstream::master())
    {
        Info << myField << endl;

        List<Type> recv;

        // Receive and write
        for (const int slave : Pstream::subProcs())
        {
            IPstream fromSlave(Pstream::commsTypes::blocking, slave);

            fromSlave >> recv;
 

            Info << "Data in processor " << slave << endl;
             Info << recv << endl;
        }
    }
    else
    {
        // Send to master
        OPstream toMaster
        (
            Pstream::commsTypes::blocking,
            Pstream::masterNo()
        );

        toMaster << myField;
    }
}

Is it possible to get the patch name from the above code?
So that the output is "Data from patch: somePatchName"

[olesen]

The code in VTK writing is optimized for contiguous data, but shows the general idea of collecting in values into the master process. If you want a more general gathering (non-contiguous) then globalIndexTemplates.C would also provide some reasonable snippets.


BTW: you may also want non-blocking modes, but then you will have to wait for the requests yourself.