Have you measured the performance of this form of tessellation vs. using GL_INVERT on the stencil buffer to implement the even-odd fill rule? I've always been curious about the relative performance of the two approaches when you have to tessellate at runtime.
I haven't benchmarked but I suspect the stencil buffer would be better. But the time is so small that for static meshes (as in this demo) it is negligible.
Unfortunately you lose MSAA with the stencil buffer approach.
Have you measured the performance of this form of tessellation vs. using GL_INVERT on the stencil buffer to implement the even-odd fill rule? I've always been curious about the relative performance of the two approaches when you have to tessellate at runtime.