Passing a million numbers from java to matlab?

I have a bit of code which runs in 10.919 s. Profiling it shows that 10.182 s are wasted in

opaque.double

Which is called when I use

jClass.GetArrays(jArray1,jArray2);

struct.prop1 = double(jArray1);
struct.prop2 = double(jArray1);

What can be done? I have to use Java to interact with external API.


EDIT: I used the following hack:

struct.prop1 = cell2mat( cell( jArray1) );

And got down to 1.5s / 2.2s

EDIT:

Making the java return long comma delimited string representation of the arrays and then using

data = strread(char(jString),'%f','delimiter',',' );

Produced almost bearable performance

From stackoverflow

• You might consider saving it to a file, then reading it in on the other end. It might also be similarly slow (especially if a network is involved), but it's worth a shot.

  You might also consider converting it to a blob of binary data, then passing it, and converting it back.

  I have a sneaking suspicion that with that bad performance it's passing each element of the array in its own transaction rather than all at once.

• The problem lays in the usage of boxed Java primitives - java.lang.Double in this case.

  Simply changing the signature of the Java from Double to double cause Matlab to seamlessly work with the array without the casting overhead.