Rendering engines that implement Variable Rate Shading (VRS) are capable of changing at run-time the rate at which regions of pixels are shaded by the fragment shader. Regions of a given frame may be shaded at full rate, with one fragment shader invocation per fragment, as usual, while others may share the same output. When implemented properly, rendering time per frame can be reduced substantially, while maintaining the same perceived quality.

This is a work in progress. I haven’t structured it properly along a logically-sequenced thread of ideas.

Sampling is the process of taking discrete values of a continuous function. Reconstruction is the process of converting the discrete samples back to a continuous function, which tries to approximate the original one as much as possible. In rendering, the scene’s incident radiance is the continuous function defined over the film plane that we try to approximate by sampling the incident radiance value of discrete pixels. Sampling is done by tracing rays and the result is a digital pixel image; we can say that the image is a discretization of the incident radiance function. Reconstruction is then performed by the display, typically via some form of interpolation, and the resulting function is defined over the surface of the display.

This is a supplementary post to a series of posts I’m writing about the CPBRT renderer, an implementation of Pharr, Jakob and Humphrey’s PBRT.

Lightmass is the system in Unreal that bakes the lighting in the scene. When you place a static light object in the scene, it will illuminate the scene dynamically until you rebuild the lighting, that is, the lighting won’t be baked until lightmass runs and bakes it. Before you bake it, the light is temporarily dynamic, because it shows you a preview of what the baked lighting would look like.

I first heard of WebGPU a few days ago. In essence, WebGPU is a Javascript API that interfaces with Vulkan, D3D12 and Metal, all of which were designed for modern GPUs. WebGPU will one day supersede WebGL, so we’d better check it out. WebGL was designed with the GPU computation model of the last decade. Since then, GPU technology has advanced. WebGPU intends to use the modern computation model.

Unlike static lighting, dynamic lighting happens in the GBuffer in deferred rendering mode.