Assignment 3 4210 COMP SCI_X_0020
Weighting and Due Date
Copyright By cscodehelp代写 加微信 cscodehelp
• Marks: This assignment contributes 15% of the overall course mark.
• Due date: 23:59 Monday 13 Jun
• Late penalties: For each part, the maximum mark awarded will be reduced by 25% per day or part-day late. If your mark is
greater than the maximum, it will be reduced to the maximum.
For this course, all assignment work will be submitted using the School Web Submission system.
You need to use the Subversion source control system (svn) in developing your project.
I have made a separate post about how to use these systems for Assignment 1 – and only a slight adaptation will be needed for
Assignment 3. This advises on how to set up directories for your project, and how to submit your work. Make sure to read this
before you start work on the assignment, so that you can set up your directories correctly.
The Web Submission system will not automatically fetch your commits from the SVN, so please make sure you submit your latest
commit to the system before the deadline.
Using the AWS RONIN server set up for this assignment (description for access is TO BE ANNOUNCED), provide a comparison of
the median performance of the Odd-Even sorting algorithm implemented both as a parallel algorithm using OpenCL, and as a serial
algorithm using C. Effects of choosing either a serial or parallel implementation, of the size of the array to be sorted, and the
number of workitems per workgroup used in parallel sort are to be investigated.
Note: For a good performance comparison, it is important that you make both implementation of the serial C code and the parallel
OpenCL kernel code as close to the same as possible. (Hint: This is a rare case where it may actually help you to first write kernel.cl,
and then adapt that to make serial.c as similar as possible)
SUPPLEMENTARY MATERIAL IS YET TO BE RELEASED THAT WILL HELP YOU FURTHER UNDERSTAND THIS ASSIGNMENT
BUT THE CORE OF THE ASSIGNMENT IS HERE AND WILL NOT BE CHANGED
For this assignment, you may also use any code presented in the lectures or for previous assignments as a basis for your code.
• report.pdf (your report document)
• serial.c and parallel.c (C code for the two implementations)
• kernel.cl (OpenCL kernel code for the parallel implementation)
• any job scripts used with batch to generate your “slurm” data files
• the “slum” data files used to create the table and plot for report.pdf
• A very brief description of the odd even algorithm (not the code!)
• A table of median times for each algorithm and number of workitems, for each array size
• A log/log plot of median times using data from your table
• An analysis of your results (description and explanation of what your data indicates)
60% for the report (15% for each of the 4 points listed above)
30% for all aspects of the code (consistent style: clarity: comments for all code: etc)
10% for including your “slurm” data files
程序代写 CS代考 加微信: cscodehelp QQ: 2235208643 Email: firstname.lastname@example.org