Project 2: Zhanbo Lin

[skszb]

Repo Link

I'm using a late day for this project.

Implemented features:

• CPU sequential scan and stream compact algorithm
• Naive parallel scan
• Work-efficient parallel scan & GPU stream compaction
• Thrust's implementation wrapper

Additional Features:

• (Extra Credit in Part 5) Work-efficient scan outperforms both CPU and naïve GPU implementations
• (Extra Credit 2) Enhanced work-efficient scan with shared memory and support for arrays of arbitrary sizes by splitting work into blocks; uses iterative kernel launches to simulate recursion, allowing pre-allocation of buffers and more accurate timing of kernel execution
• Explored variadic function templates and implemented a helper function (testForIterations() in testing_helper.hpp) to repeatedly execute a target function and compute the average runtime, reducing manual effort in performance testing

Some thoughts:

• Really enjoyed this project, but I wish we could have a crash course about interpreting Nvidia Nsight Compute profiling reports before it.