Instead of using sigma in that -args list, use
sigma_type = coder.typeof(0,[2,2],[true,true])
This means the type is double and real (that is the type of the first argument "0"). The second argument says that the array is up to 2-by-2, and the third argument says that both dimensions are variable in size.
Good luck. The conversion looks doable, but there may be other snags, and it's the type of function where I'd probably do a lot of re-writing to make it efficient for code generation (e.g., I'd write a little loop to calculate sum(D<0) instead of creating a logical array and then summing it). -- Mike
