sfcnn_knng is a specialized version of the sfcnn algorithm
designed to quickly generate K-nearest neighbor graphs.
If compiled using openmp, it will use multiple processors
to compute the solution.  To include it in a program,
use sfcnn_knng.hpp

sfcnn_knng< MyPointType,
                        Dimension,
                        CoordType >
                NN(&FirstDataPoint, Size, k, ThreadCount);

• MyPointType
  
• This can be any user defined point type, as long as the
  coordinates can be referenced by the bracket operator.
  For example, vector<double> or float[], are both valid.
• Dimension  
  
• This is the dimension of the points in the data set.
• CoordType
• This is the type of coordinate the points use.
  Currently supported types are:
  (unsigned) short, int, long
  short, int, long
  float, double, long double
  
• IMPORTANT! Searches on integer types operate
  about 3x faster than floating point types.  If you
  can use them, it is recommended.
• FirstDataPoint
• A pointer to the first point in the input data set
• Size
• The size of the input data set
• k
• The number of nearest neighbors to return. 
  Ie. k=5 will return the 5 nearest points to Q in
  the data set
• ThreadCount
• This is an optional parameter whose default
  value is 1. If compiled using openmp, this will
  set the number of processors to use during
  construction.
  

Note: All calculation for the k nearest neighbor graph is
           done at initialization.

NN[i][j]     This returns the index of the j^th nearest
                 neighbor to the i^th data point.  The indexes
                 are all relative to the first data point
                 given at initialization.  If the data points
                 have moved, the answer will be incorrect.