Cache memories are an essential component of modern processors and consume a large percentage of their power consumption. Its efficacy depends heavily on the memory demands of the software. Thus, finding the optimal cache for a particular program is not a trivial task and usually involves exhaustive simulation. In this article, we propose a machine learning–based methodology that predicts the optimal cache reconfiguration for any given application, based on its dynamic instructions. Our evaluation shows that our methodology reaches 91.1% accuracy. Moreover, an additional experiment shows that only a small portion of the dynamic instructions (10%) suffices to reach 89.71% accuracy.
Graphics Processing Units (GPUs) have found application in a broad range of domains due to their ability to massively exploit data parallelism and thereby achieve high energy efficiency, being now referred to as General-Purpose GPUs (GPGPUs). However, with a market dominated by two or three industrial players and a lack of open-source hardware realizations and standardized GPU Instruction-Set Architectures (ISAs), the full potential of these architectures for academic research and use by small hardware development enterprises remains far from being unlocked. In this paper, we start by discussing a few exciting research directions in GPGPU architectures and the benefits of open-source Hardware realizations. Furthermore, we show how our open-source FGPU core can be used to enable overcoming these challenges and contribute to the agile development of Domain-Specific Architectures (DSAs) featuring System-on-Chip (SoC) platforms extended with adapted GPGPU cores. FGPU is described in VHDL, can be programmed in OpenCL by use of our LLVM-based compiler, and has been validated against an ARM Cortex A9 with NEON support, achieving 4 performance improvement in fixed and floating-point benchmarks.