An ActivePivot post-processor is a normal Java class. There is nothing special about it. So you can use any existing technique so call a C++ DLL function while inside a Java program.
This can be achieve for instance with JNA and BridJ. I don't consider JNI as in most cases, you don't need to use such a low-level API.
For instance, with BridJ:
Given a C++ header looking like:
__declspec(dllexport) int multiply(double multiplier, int size, double* const vector);
I made the following class:
import org.bridj.BridJ;
import org.bridj.Pointer;
import org.bridj.ann.Library;
import org.bridj.ann.Runtime;
import org.bridj.cpp.CPPRuntime;
// Generated with http://code.google.com/p/jnaerator/
@Library(CPP_Collateral.JNA_LIBRARY_NAME)
@Runtime(CPPRuntime.class)
public class CPP_Collateral {
public static final String JNA_LIBRARY_NAME = "dummy";
static {
// In eclipse, the DLL will be loaded from a resource folder
// Else, one should add a program property: -Djava.library.path
BridJ.addLibraryPath("src/main/resources/DLL");
BridJ.register();
}
/**
* My dummy.dll has one method:
* int multiply(double multiplier, int size, double* const vector)
*/
public static native int multiply(double multiplier, int size, Pointer<Double> vector);
}
and my IPostProcessor is simply
@Override
protected Object doEvaluation(ILocation location, Object[] underlyingMeasures) throws QuartetException {
double[] asArray = (double[]) underlyingMeasures[0];
if (asArray == null) {
return null;
} else {
// Size of the array
int size = asArray.length;
// Allocate a Pointer to provide the double[] to the C++ DLL
Pointer<Double> pCount = allocateDoubles(size);
pCount.setDoubles(asArray);
CPP_Collateral.multiply(2D, size, pCount);
// Read again: the double[] is copied back in the heap
return pCount.getDoubles();
}
}
In term of performance, here I worked on 2.000 double[] of size 10000 and the impact is of about 100ms