Project 2: Dineth Meegoda

[dinethmeegoda]

Repo Link

Features:

• CPU Scan & Compaction (with and without scan)
• GPU Based Naive Scan Algorithm
• GPU Based Work-Efficient Scan Algorithm
• Implementation of the Thrust Library's exclusive scan
• Performance Analysis & Benchmark

Feedback:

• This was an interesting project, but I struggled on the profiling section. Other than noting the running times, I tried to get more information through NSight Systems and Compute but I wasn't sure how to navigate them to get the information I needed. I wanted to particularly find instances of my kernels having memory-related stalls since I did not implement shared memory, but my NSight Systems did not give me much information. I might have some of my profiling settings messed up? Now looking at other readme's, I did not have the same NSight Systems rows available as they did for the profiles which confused me. As a result, I had to make my bottleneck predictions based on my thoughts on my own code and what we've learned in class.