FAQs
General Questions
Q: What is the purpose of this website?
A: The purpose is to create a high-quality open-source GPGPU library code to promote a wide adoption of GPUs as an effective accelerator for applications.
We are currently focusing on performance-critical functions in MATLAB Image Processing Toolbox. Plans on other applications are underway.
Q: Who are the developers?
A: We have developers from UCF, NCSU and helps from AMD/ATI experts with tremendous skills.
Q: What are the performance advantage of using your library code other than the original MATLAB code?
A: As listed on the front page, great speedups (up to 135x) are achieved compared to the original MATLAB functions, which are executed on CPU.
Q: How do I contribute if I'd like to?
A: You are more than welcome to contribute to the library. We welcome questions, success stories, comments, feature requests from public also.
Please contact Jingfei Kong at this time.
Technical Questions
Q: What are the system requirements for running your code?
A: Currently, in addition to Windows XP and MATLAB, AMD/ATI Stream SDK 2.0 and AMD/ATI Raedon HD5870 are recommended.
Q: How do I run your code in MATLAB?
A: Please refer to our instructions to run OpenCL in MATLAB.
Q: How do I develop my own code?
A: A simple way is to take one of our function code, which is a good template, and make changes.
For OpenCL code interface to MATLAB, please refer to the MEX documentation provided by MATLAB. For developing optimized OpenCL code, please
refer to AMD/ATI Strean SDK page, which contains numerous resources on helping developers on this issue.
Q. In the MATLAB command line example, why execute GPU-accelerated function once to warm up before collecting performance results?
A: The warm-up is to setup necessary GPU resources, especially kernels which currently have to be compiled before usage. After the one-time warm-up,
those resources do not have to be setup again, improving the overall GPU performance. The warm-up may be removed with pre-compilation support from Stream
SDK in the future.