A prototype of an invasive tightly-coupled processor array

Invasive computing is a novel paradigm for the exploitation of runtime parallelism of future MPSoC architectures through resource-aware programming and dynamic reconfiguration of the underlying architectures. Based on the state and availability of resources, an invasive algorithm organizes its computation itself. A highly-parameterizable weakly programmable VLIW processor array called Tightly-Coupled Processor Array (TCPA) that is augmented with decentralized units for hardware-accelerated resource exploration is well suited for invasive computing, especially for accelerating computationally intensive loop programs. We present a prototype of an invasive TCPA, which demonstrates the benefits of invasive computing using the example of video processing. More specifically, the targeted applications in this prototype are 2D convolution filters (such as edge detection and Gaussian filtering) that are applied on a real-time input video stream. For different workload scenarios and resource competition between applications, a varying quality of the output video streams is shown.