University of Pennsylvania, CIS 565: GPU Programming and Architecture, Project 5
- Jonathan Lee
- Tested on: Google Chrome 61.0.3163.100 on Windows 10, Xeon ES-1630 @ 3.70GHz 32GB, GTX 1070 24GB (SigLab)
In this project, I was able to explore Clustered Forward+ and Deferred Shading.
Both of these implementations of Forward+ and Deferred Shading were done using clustering. All of the operations are computed in camera space and the camera frustum gets split into clusters.
Forward+ is an extension of Forward Shading except we use the light's z position determines how to find the cluster it's in. Once we find the cluster that the light's in, the shader only computes for the lights that lie within the cluster.
Deferred Shading utilizes the same logic of dividing the camera frustum based on the light's position. The shading, however, requires two passes. Fragments get written the the g-buffer and store position, color, and normal values.
| 100 Shininess | 200 Shininess | 500 Shininess |
|---|---|---|
 |
 |
 |
| Without Blinn | With Blinn |
|---|---|
 |
 |
Clearly, Deferred Shading performs much better than Forward+ and Forward Shading. The divide between the clustered shaders and Foward begins rapidly around 200 lights. Deferred becomes nearly impossible to see at around 15000 lights.
To do compression, I stored normal.x and normal.y at the w position of the vec4 that also stores position and color. Rather than reading in the value from a third g-buffer that only stores the normal, normal.z can be calculated in the shader. Surpsingly, I found a huge dropoff between uncompressed and compressed normals after 500 lights.
- Three.js by @mrdoob and contributors
- stats.js by @mrdoob and contributors
- webgl-debug by Khronos Group Inc.
- glMatrix by @toji and contributors
- minimal-gltf-loader by @shrekshao
- Blinn Shading
- Various Camera Stuff
- OpenGL-Tutorail
- Foward+ vs. Deferred
- CIS560 and CIS565 Slides