University of Pennsylvania, CIS 565: GPU Programming and Architecture, Project 4
- Yu Sun
- Tested on: Tested on: Windows 10 , i7-6700HQ CPU @ 2.60GHz × 8 , GeForce GTX 960M/PCIe/SSE2, 7.7GB Memory (Personal Laptop)
In this project, a basic rasterized graphic pipeline is implemented using cuda. The pipeline includes vertex shading, primitive assembly, rasterization, fragment shading, and a framebuffer. The special features I implement for this project includes
* Rasterization for lines and points
* Back-face culling
* UV texture mapping with bilinear texture filtering and perspective correct texture coordinates.
* Super-sample Anti-Aliasing
Forgive my briefness for the README, I'll add more things if I get more time later.
Rasterization for Points, Lines and Triangles
Rasterization for points is straightforward by computing the corresponding pixel index and color.
Rasterization for lines is approximated using the Bresenham Algorithm since a naive approach would lead to artifact coming from fixed grid resolution, and rounding float pixel locations to integer pixel locations
Rasterization for traingles is achieved by using barycentric coordinates
| Milk Car | Duck | Flower |
|---|---|---|
 |
 |
 |
Texture Mapping and Prospective Correction with Bilinear Interpolation
The texture mapping can be achieved by using uv coordinates that warp 2D textures onto 3D mesh. Techniques used to make the texture look better include bilinear interpolation and perspective correction using the depth information.
A comparision of texture mapping with and without perspective correction is shown below.
| Scene without Perspective Correction | Scene with Perspective Correction |
|---|---|
 |
 |
Bilinear interpolation is basically a techique used to prevent aliasing effect and make the resulting image looks more natural and smooth by taking the color of surrounding pixels into account while generating the final color for a specific pixel. Since the effect isn't that obvious, I picked one image that I found through Google Image to demonstrate the effect.
Back-face Culling
Back-face culling is intended to reduce the amount of computation by eliminating the pixels that cannot be captured from the camera. However, while implementing it I found that I didn't see a significant speed up, and it actually creates some funny effect.
| Scene without Back-face Culling | Scene with Back-face Culling |
|---|---|
 |
 |
The things that are culled out can be seen below
Super-Sampled Antialiasing
By super-sampling, one is essentially making more grids and creating higher resolution. This is a sacrifice on memory to give more details to the display. The difference can be seen below.
| SSAA Factor = 1 | SSAA Factor= 4 |
|---|---|
 |
 |
Below is a graph demonstrating the amount of time spent on each stage of the graphics pipeline for different type of scenes. Without much surprise, it can be seen that most time of the computation is spent on primitive rasterization.
Notice how computation increase while we move closer to the rendered object due to the increase amount of checks we need to perform.
