Project 2: Ruoyu Fan

[WindyDarian]

• Repo Link
• ruof
• Features
  • Implemented CPU scan and compaction, compaction, GPU naive scan, GPU work-efficient scan, GPU work-efficient compaction, GPU radix sort (extra), and compared my scan algorithms with thrust implemention
  • I optimized my work efficient scan, and speed is increased to 270% of my original implementation.
  • I also wrote an invlusive version of work-efficient scan - because i misunderstood the requirement at first! The difference of the inclusive method is that it creates a buffer that is 1 element larger and swap the last(0) and and second last elements before downsweeping. Although I corrected my implemention to exclusive scan, the inclusive scan can still be called by passing ScanType::inclusive to scan_implenmention method in efficient.cu.
  • Radix sort assumes inputs are between [0, a_given_maximum) . I compared my radix sort with std::sort and thrust's unstable and stable sort.
  • I added a helper class PerformanceTimer in common.h which is used to do performance measurement.