Jose @ ShanghaiTech University 2019.05.31
Simple proof of concepts of parallel GPU hashing.
Under Serial-vs-NaiveCUDA
is comparison between serial cuckoo hashing v.s. naive CUDA implementation of parallel cuckoo hashing. Under NaiveCUDA-vs-Multilevel
is comparison between naive CUDA implementation v.s. multi-level shared-memory optimized parallel cuckoo hashing.
References:
- SIGGRAPH'09 paper: Dan A. Alcantara, Andrei Sharf, Fatemeh Abbasinejad, Shubhabrata Sengupta, Michael Mitzenmacher, John D. Owens, and Nina Amenta. 2009. Real-time parallel hashing on the GPU. ACM Trans. Graph. 28, 5, Article 154 (December 2009), 9 pages. DOI: https://doi.org/10.1145/1618452.1618500. URL: https://www.cs.bgu.ac.il/~asharf/Projects/RealTimeParallelHashingontheGPU.pdf
- Poster of this paper: https://www.nvidia.com/content/GTC/posters/82_Alcantara_Real_Time_Parallel_Hashing.pdf
Environment prerequisites:
- CUDA >= 10.1
- C++ STL >= C++11 standard
Run the experiments:
- In each sub-directory, build by
make
- Execute the built binaries (each described in its comments)
- Clean the build by
make clean