Support parallel runs (intra-solver parallelism)

[BenjaminRodenberg]

Currently, the bindings do not support parallel runs. We should add this feature.

[uekerman]

What exactly do you mean by "parallel runs"? Shared memory parallelization of Matlab? Or does Matlab also offer distributed parallelization? If shared memory: do we want to treat the Matlab solver as a serial or as a parallel solver from preCICE perspective? If we treat it as a serial solver this should always be possible or am I overlooking sth?

[BenjaminRodenberg]

What exactly do you mean by "parallel runs"? Shared memory parallelization of Matlab? Or does Matlab also offer distributed parallelization? If shared memory: do we want to treat the Matlab solver as a serial or as a parallel solver from preCICE perspective? If we treat it as a serial solver this should always be possible or am I overlooking sth?

To be more clear: I mean intra-solver parallelism. I.e. the MATLAB solver running on several processes. Currently, we require the user to provide solverProcessIndex = 0 and solverProcessSize = 1, otherwise an error is thrown:

matlab-bindings/+precice/@SolverInterface/SolverInterface.m

Lines 23 to 27 in 199fddc

	function obj = SolverInterface(SolverName,configFileName,solverProcessIndex,solverProcessSize)
	%SOLVERINTERFACE Construct an instance of this class
	if (solverProcessIndex > 0 || solverProcessSize > 1)
	error('Parallel runs are currently not supported with the MATLAB bindings.')
	end

From preCICE perspective nothing speaks against supporting intra-solver parallelism, but since the feature is untested with the MATLAB bindings, we currently do not support it and rather throw an error on unexpected user input.

[uekerman]

But which type of parallelization would you want to support? Shared memory parallelization? Using then more threads in preCICE has never been tested afaik. I am not sure if preCICE has enough thread-safety for this. Only letting one thread interact with preCICE might be just the right approach.