Portable Framework for Real-Time Parallel Image Processing on High Performance Embedded Platforms

Abstract

The trend to efficient, however more complex, multicore designs has also reached the world of Digital Signal Processors (DSP), a field where typically low-level programming has been prevalent. To overcome the additional complexity of programming multi-core and multi-chip DSP systems, we present an object-oriented framework for task-based parallel programming on the highly power efficient Texas Instruments TSMC320C6678 platform. Our framework incorporates hardware architectural details of this platform such as DMA units in a high-level manner, while maintaining portability - guiding the path for algorithmic designers from PCs to embedded DSP platforms. The whole framework has been designed and implemented with real-time requirements and low overhead in mind, which is crucial for the acceptance of higher-level solutions on embedded systems.