Abstract

Gedae is a programming language and multiprocessor compiler that targets a wide variety of multiprocessor and multicore systems. The language has been used successfully on deployed programs to program multiboard/multiprocessor systems of DSPs. The tool is being enhanced to help transfer its proven benefits in productivity gains and application performance in the DSP arena to the new breed of multicore processors. One of the most exciting chips in this new breed of multicores is the IBM/Sony/Toshiba Cell Broadband Engine™ (Cell/B.E.). However its high degree of parallelism and tight memory restrictions on the lightweight cores make it difficult to program. Gedae aims to mitigate the difficulty of using this powerful hardware without loss in performance. One application for these new processors like the Cell/B.E. is medical imaging. The use of computing is now an essential part of both medical research and diagnostic medicine, using software to enhance and classify images from MRIs, CT scans, ultrasounds, and radioactive sources. Image processing functions such as filtering, feature extraction, image segmentation, and visualization provide a powerful method for viewing affected areas of the body which were difficult or even impossible to view in previous years. Gedae enables medical imaging software to more effectively utilize high performance computers, allowing physicians and biologists to process and manipulate these images in real time, shortening the gap between test and diagnosis. This speech provides an overview of model driven development on the Cell/B.E. with Gedae to enable software portability, efficient algorithmic upgrades and obsolescence-proof expansion with specific emphasis on medical imaging algorithms.