Looks very cool, but why? Unlike many other applications of SDFs, for this particular one a traditional rasterization seems more appropriate.
The polyhedron being rendered only has 32 vertices and 60 triangles. The indexed mesh gonna take 744 bytes of VRAM with FP32 coordinates and uint16_t indices, which can be reduced to 552 bytes by using FP16 coordinates. That’s less data than the GLSL source code to compute and render the SDF. Also, if MSAA is enabled rasterization will deliver better visual quality.
That's true for _this_ polyhedron, but it's useful and more clear to demonstrate the technique with a simple example and allow the mind to then explore what else could be modelled this way than to blow someone away with complexity that they miss the point.
This is far more about SDFs than it is about the polyhedron.
wow, you made something complex understandable to me. I'd dabbled in shaders but Raymarching (I now know why it's called that!) and SDF had eluded me, thank you.
The polyhedron being rendered only has 32 vertices and 60 triangles. The indexed mesh gonna take 744 bytes of VRAM with FP32 coordinates and uint16_t indices, which can be reduced to 552 bytes by using FP16 coordinates. That’s less data than the GLSL source code to compute and render the SDF. Also, if MSAA is enabled rasterization will deliver better visual quality.