Matlab cuda
From SuperMe
Contents |
[edit] GPU-Acceleration in Matlab
[edit] Project Members
Project by: Rob King (University of Maine)
Advisors: Dr. Yifeng Zhu (Department of Electrical and Computer Engineering, UMaine)
[edit] Abstract
Matlab is a very powerful computer program used to solve complicated mathematical problems. Scripts can be written to perform long calculations many times very easily. This is how many buil-in functions are implemented in Matlab. These scripts are the target of the optimizations. Also, MEX-files can be written in the C programming language and run through the Matlab command window just like the built-in functions. In order to speed up Matlab the toolboxes will be studied and profiled to find the main bottlenecks in the sequential code. Then a MEX-file will be written including CUDA code to decrease the completion time of the Matlab functions. The parallel code will run in many threads on an NVIDIA graphical processing unit. By decreasing the completion time of the longest part of the code the largest performance increase can be achieved.
[edit] Schedule
- Week 1 (06/1-06/07):
- SuperMe program orientation
- Week 2 (06/08-06/014):
- Read background information
- Create project abstract
- Week 3 (06/15-06/21):
- Continue to read background information
- Week 4 (06/22-06/28):
- Begin working with the Matlab PDE tool.
- Determine the main bottleneck in the PDE tool solver.
- Week 5 (06/29-07/05):
- Begin working on the LU matrix factorization routine.
- Week 6 (07/6-07/12):
- Midterm report due.
- Week 7 (07/13-07/19):
- Week 8 (07/20-07/26):
- Begin to prepare report and presentation
- Week 9 (07/27-08/2):
- Complete project report
- Prepare presentation
- Week 10 (08/3-08/07):
- Finish preparing poster and presentation
- Written report due - Wednesday, August 5th
- Poster due - Thursday, August 6th
- Final symposium - Friday, August 7th
[edit] Mid-Term Report
[edit] Final Report
[edit] Poster
Final poster (pdf)
[[Media: | Final Poster (odp)]]