Project 2: Yunhao Qian

[yunhao-qian]

• Repo Link
• Features:
  • A CPU implementation of scan and stream compaction
  • GPU implementations of scan, using both naive and work-efficient methods
  • A GPU implementation of stream compaction based on the work-efficient scan
  • A carefully optimized work-efficient scan that leverages shared memory along with several additional optimizations
  • C++ and Python scripts used to automate performance measurement accurately and programmatically
  • Performance analysis comparing the different methods
  • More in-depth performance analysis conducted with Nsight
• Extra credits: all of Part 5 and Part 6 have been completed.
• Changes to CMakeLists.txt:
  • An additional executable,measure_time.exe, has been added to the project to support block size tuning, performance benchmarking, and profiling.
  • The files cpu_sort.h, cpu_sort.cu, radix_sort.h, radix_sort.h were introduced to implement Extra Credit 1.
  • The files efficient_plus.h and efficient_plus.cu were introduced to implement Extra Credit 2.
• Function Overloads: To make block size tuning easier, I added overloads of the following functions that accept an additional blockSize parameter. The original overloads remain unchanged and simply forward to the new versions with a tuned default value. These changes are only used by measure_time.exe and do not affect existing code paths.
  • Naive::scan(..., const int blockSize)
  • Efficient::scan(..., const int blockSize)
  • Efficient::compact(..., const int blockSize)

[yunhao-qian]

Please use the latest "Complete everything" (7b0719c) commit for grading. Thank you!