Many modern applications exhibit a dynamic and nonstationary behavior, with certain characteristics in one phase of their execution, which change as the application enters new phases, in a manner unpredictable at design-time. In order to meet the demands of such applications, it is important to have adaptive and self-reconfiguring hardware platforms, coupled with intelligent on-line optimization algorithms, that together can adjust to the run-time requirements. Partially dynamically reconfigurable field programmable gate array architectures offer both high performance and flexibility. Despite these potential advantages, the challenges faced by designers trying to set-up a functioning system are still significant, mainly because of the still immature design tools and limited device drivers. We propose a complete framework, based on Xilinx's commercial design suite, that enables an application designer to leverage the advantages of partial dynamic reconfiguration with minimal effort. Our IP-based architecture, together with the comprehensive application programming interface, can be employed to accelerate an application by dynamically scheduling hardware prefetches. Moreover, a piecewise linear predictor is used to capture correlations and predict the hardware modules that will generate the highest performance improvement. Our evaluation comprises of extensive simulations, as well as a complete implementation of the smallest univalue segment assimilating nucleus image processing application on the ML605 board from Xilinx. The measurements show a significant reduction of the expected execution time compared to previous state-of-the-art prefetching algorithms, with only a minor energy overhead.

Modern systems demand high performance, as well as high degrees of flexibility and adaptability. Many current applications exhibit a dynamic and nonstationary behavior, having certain characteristics in one phase of their execution, that will change as the applications enter new phases, in a manner unpredictable at design-time. In order to meet the performance requirements of such systems, it is important to have on-line optimization algorithms, coupled with adaptive hardware platforms, that together can adjust to the run-time conditions. We propose an optimization technique that minimizes the expected execution time of an application by dynamically scheduling hardware prefetches. We use a piecewise linear predictor in order to capture correlations and predict the hardware modules to be reached. Experiments show that the proposed algorithm outperforms the previous state-of-art in reducing the expected execution time by up to 27% on average.

Dynamically reconfigurable computing devices have the ability to adapt their hardware to application demands, providing the performance of hardware acceleration, as well as high flexibility, at competitive costs. For these reasons, FPGA-based reconfigurable systems are becoming popular in many application domains, including soft real-time computing. Unfortunately, one of their biggest limitations is the high reconfiguration overhead. One method to overcome this problem is configuration prefetching, which tries to reduce the reconfiguration penalty by preloading modules on the FPGA before they are needed, and overlapping the reconfiguration with useful computation. In this paper we present a speculative approach to context-aware inter-procedural configuration prefetching that provides statistical guarantees by minimizing the alpha-percentile of the execution time distribution of a soft real-time application. Our method uses profile information and takes into account the calling context of a procedure in order to generate better prefetch solutions. We also propose a middleware needed to apply the context-dependent prefetches at run-time. Our experiments show that the developed algorithm outperforms the previous state-of-art.

This paper presents an approach to minimize the average program execution time by optimizing the hardware/software implementation of error detection. We leverage the advantages of partial dynamic reconfiguration of FPGAs in order to speculatively place in hardware those error detection components that will provide the highest reduction of execution time. Our optimization algorithm uses frequency information from a counter-based execution profile of the program. Starting from a control flow graph representation, we build the interval structure and the control dependence graph, which we then use to guide our error detection optimization algorithm.