Not logged in.
Quick Search - Contribution
Contribution Details
| Type | Bachelor's Thesis |
| Scope | Discipline-based scholarship |
| Title | A GPU-enabled Single-Point Incremental Fourier Transform |
| Organization Unit | |
| Authors |
|
| Supervisors |
|
| Language |
|
| Institution | University of Zurich |
| Faculty | Faculty of Business, Economics and Informatics |
| Date | 2020 |
| Abstract Text | Although Fourier transforms are widely used, special implementations are still being developed for high performance applications. For use in a streaming environment, the Single Point Incremental Fourier Transform (SPIFT) was recently proposed (Saad et al. 2020). In SPIFT the main computational bottleneck is the incremental addition, where the Single Point Fourier Transform of each new datapoint is integrated into the previous result. Two key optimizations to speed up SPIFT were introduced and tested. Firstly, GPUs are used to efficiently sum the Single Point Fourier Transforms of the individual datapoints to the final result. Secondly, Single Point Fourier Transform of the different datapoints with the same shift are combined using an aggregation matrix instead of individually being integrated into the resultant image. Five different implementations combining these two optimizations were evaluated. Both optimizations are crucial and are together able to increase the throughput on tested matrix dimensions by three orders of magnitude. |
| PDF File |
Download
|
| Export |
BibTeX
|