Skip to content

Latest commit

 

History

History
27 lines (16 loc) · 1.08 KB

File metadata and controls

27 lines (16 loc) · 1.08 KB

CUDA Rasterizer

Renders

Performance Analysis

###Execution Time Distribution

Since the rasterizer is parallelized over polygons, it takes majority of the computation time in rasterizing a single triangle (right). On the other hand, the fragment shader starts to take more time to compute with more smaller polygons (left). This matches what was was mentioned in class about how the fragment shader usually becomes the performance bottleneck in shaders.

Execution Time Distribution Pie Chart

Note: Fragment shader implements only Lambertian reflectance with single point light.

###Effectiveness of Backface Culling

Below shows execution time improvement after implementing backface culling (tested with cow model). Performance benefit is greatest at rasterization since there are fewer polygons to process.

Backface Culling Effectiveness Graph

Note: This was for a class at UPenn: CIS565 Project 4 Fall 2013 (Due 10/31/2013)