Out-of-order execution significantly increases the performance of superscalar processors. The out-of-order execution mechanism is, however, energy-inefficient, which inhibits scaling superscalar processors to high issue widths and large instruction windows. In this paper, we build on the observation that between 19% and 36% of the instructions are immediately ready for execution, even before entering the issue queue. Yet, these instructions proceed to the energy-consuming steps of instruction wake-up and select and they needlessly occupy space in the issue queue. To save energy, we propose for these instructions to by-pass the out-of-order execution core. Instead, we execute them on an energy-efficient single-issue in-order by-pass pipeline. The by-pass pipeline executes a significant fraction of all instructions, allowing performance-energy trade-offs with respect to the issue width of the out-of-order pipeline and to the issue queue size. By making these trade-offs, we show energy reductions of 53% for the issue queue, 33% for the register file and 31% in the write-back and wake-up logic. Performance remains almost unaffected.

No abstract

Wireless sensor node platforms are very diversified and very constrained, particularly in power consumption. When choosing or sizing a platform for a given application, it is necessary to be able to evaluate in an early design stage the impact of those choices. Applied to the computing platform implemented on the sensor node, it requires a good understanding of the workload it must perform. Nevertheless, this workload is highly application-dependent. It depends on the data sampling frequency together with application-specific data processing and management. It is thus necessary to have a model that can represent the workload of applications with various needs and characteristics. In this paper, we propose a workload model for wireless sensor node computing platforms. This model is based on a synthetic application that models the different computational tasks that the computing platform will perform to process sensor data. It allows to model the workload of various different applications by tuning data sampling rate and processing. A case study is performed by modeling different applications and by showing how it can be used for workload characterization

The power efficiency of an HMCP heavily depends on the architecture of its processor cores. It is thus very important to choose it carefully. When comparing processing architectures for their use in a many-core platform, one must evaluate its IPC, but also its power and area. Precise power and area evaluations can only be done with real implementations. However, comparing processor implementations is a difficult task since the implementation specifities introduce interferences on the performances. This paper proposes a methodology that allows to realize precise comparisons of performance for different processor architectures. Using this methodology, it is possible to choose the best architecture for an HMCP targeting DSP applications. The methodology is based on the use of a common architural template to build the cores, and on the application of specific optimizations when relevant. In order to validate the methodology, three RISC cores are implemented: a single-issue core, and two VLIW processors with respectively 3 and 5 issues. The implemented cores are precisely compared on a set of DSP kernels.