45
edits
Changes
N/A
,→Assignment 2
As for merge, I had difficulty doing recursion on the kernel but was solved using “-arch=sm_35 -rdc=true” command line switch. Merge can be optimized and improved greatly unlike the odd-even bubble sort since I found out there to be multiple solutions to creating it. I also happened to find a merge sort implementation in CUDA 6.0 Samples which were quite complex to understand but overall much faster than my implementation.
[[File:A2_merge.cu.txt]]
'''Woosle Park – Insertion Sort & Heap Sort'''
'''Jordan Pitters – Selection & Quick Sort'''
----
'''Final thoughts'''
[[File:A2_comparison.xlxs]]
Our parallel sorting algorithms were slower on the GPU. However, after reprofiling on an average computer with the following specs: