I just finished the STL rendering engine for the CAM software. It took a few iterations to get the performance and memory use to a reasonable level, and it could still use some optimizations. The current implementation uses a combination of raster rendering with associated metadata. The idea is that I can use the raster rendering to easily traverse the metadata rather than relying on complex geometric algorithms. The final g-code should be identical since the true coordinates are stored in the metadata, but it should give some additional performance and simplicity. The only downside is that it uses more memory than a purely geometric algorithm, but I already have some ideas on how to compress the rasterized representation. Here is an example of the rendering on a test model:
The shading represents depth (Z axis) and the green lines show the edges of each facet. This is just a visualization of the internal representation. I've posted the full size bitmap and the corresponding input files. The model started in Sketchup, then I converted it to an STL file using the program I wrote last week and then rendered using the new code. The next step is to profile/scan the rendering to generate the g-code.