130 likes | 254 Views
This paper presents an efficient method for Layered Depth Images (LDI) generation using CUDA programming. We outline our algorithm implementation, focusing on rasterization and depth sorting techniques with thread optimization. Our correctness validation compares results with the SOFA Framework, demonstrating consistent outputs. Performance evaluation highlights the efficiency of our approach while discussing useful CUDA programming tricks for maximizing resource utilization. The conclusions summarize the benefits and future applications of this technique in computer graphics.
E N D
CUDA Based LDI Generation Rui Hu, Xitong Liu
Outline • Algorithm Implementation • Correctness Validation • Performance Evaluation • CUDA Programming Tricks • Conclusion
Depth Sorting Thread Each thread processes one fragment __shared__ float sortBufS[MaxFragPerThread][LDImaximumLayer];
Correctness Validation • Compare our results of Simple Cube with the result generated by SOFA Framework • Both results match
CUDA Programming Tricks • Register variable reuse – Save register space • Check CUDA runtime errors • cutilCheckMsg • cudaGetLastError • cutilCheckError • cutilSafeCall • Select the appropriate compiling options