Tiled Rendering

I implemented multiple techniques of rendering geometry to test what their benefits and drawbacks are.
I started with implementing the easiest technique which is forward rendering. With forward rendering you directly do the lighting calculations while drawing an object, however, if there is much overdraw there are a lot of resources wasted in performing lighting calculations for pixels that get discarded. This is where deferred rendering aims to improve. Deferred rendering first draws the base color and other data such as normals into a collection of textures; this is called the G-Buffer. Once these textures are rendered, their information will be used to do lighting calculations which guarantees only performing lighting calculations once per pixel. Even though we have to perform lighting once per pixel with deferred, having many lights in the scene will still perform slowly because a lot of pixels have to perform lighting calculations even if that light does not influence that pixel. By splitting up the screen into tiles (usually 32*32 or 64*64 pixels) and calculating which lights affect that tile we can only have to do lighting calculations for lights that most likely affect those pixels, greatly improving performance.

GPU particles

Aurai also features both CPU and GPU-based particle simulations. The GPU particles are simulated using compute shaders and then the correct amount of active particles get drawn using DrawIndirect allowing for no requirement to synchronize the GPU to the CPU to retrieve the amount of active particles we need to draw.

Cascaded Shadow Maps

While implementing shadow mapping, I wasn’t satisfied with the low detail or high memory cost of a large shadow map, therefore I looked into techniques where these problems can be (partially) resolved. I chose to implement cascaded shadow mapping because it seemed to improve on the issues I’ve had and allows for higher detail over a large area. Cascaded shadow mapping splits the view frustum into multiple slices (cascades) and assigns a shadow map for each slice; this way the slice closest to the camera allows for higher texel density in the shadow map, whereas areas further away from the camera cover a larger area but it requires less detail so the reduced texel density is insignificant.