@Asgeir_Mortensen the heart of any slicer is an extremely robust polygon offset routine that handles degenerate cases gracefully. This task, by its nature does not lend itself readily to GPU processing because it’s highly serialised and so SIMD doesn’t bring a significant performance increase.
note that a number of slic3r bugs are directly related to its polygon offset routine (in the C/C++ Clipper library) being less than stellar.
If you’re interested in tackling this routine, see Stefan Huber’s work on straight skeletons for some prerequisite reading. He has created a method that can construct a framework for arbitrary offsetting in O(n log n) time which I strongly suspect is significantly faster than either Clipper or skeinforge.
Once such a routine has been found or created, the rest of the slicing process is significantly easier. You’ll also need routines for doing boolean ops on polygons with holes, then you can intersect planes with the STL to get layer outlines, propagate the outlines up and down the stack to generate regions such as solid surfaces, bridge, support candidates, etc, boolean the overlapping regions according to a list of priorities, then offset the borders to generate perimeters, then offset again to get a clipping border for your infill pattern.