How to speed up this graph filtering algorithm?

Hi,

I'm using the PMFG algorithm to filter out the insignificant links of a graph.

The function pmfg() is fast as long as the graph is small (<= 100 nodes). However, I need to apply it to graphs with thousands of nodes and in this case the function runs for several hours. In order to improve the function's computation time I was thinking about converting the function to C. But according to the profiler the bottleneck is actually a child function, which is already implemented in C ('planar_test_mex', part of the MatlabBGL library).

What else could be done here to speed things up?

p.s.: The function pmfg() shows warnings about sparse indexing expressions, which might be slow. However, the algorithm is too involved for me to resolve these warnings on my own.