Data Dependent Energy Modeling for Worst Case Energy Consumption Analysis

Abstract

Safely meeting Worst Case Energy Consumption (WCEC) criteria requires accurate energy modeling of software. We investigate the impact of instruction operand values upon energy consumption in cacheless embedded processors. Existing instruction-level energy models typically use measurements from random input data, providing estimates unsuitable for safe WCEC analysis.We examine probabilistic energy distributions of instructions and propose a model for composing instruction sequences using distributions, enabling WCEC analysis on program basic blocks. The worst case is predicted with statistical analysis. Further, we verify that the energy of embedded benchmarks can be characterised as a distribution, and compare our proposed technique with other methods of estimating energy consumption.

Energy-Aware Software Engineering

“…In Ref. [18], 15% data-induced variation has been reported for the 8-bit AVR processor, while up to 1.7x data-dependent variation was observed for the 32-bit XMOS XCore in [19]. Variation of as much as 50% is reported in [19] for an ST20 32-bit microprocessor.…”

“…[18], energy modelling for worst-case energy consumption analysis has been explored. The most promising approach uses probabilistic energy distributions to characterise individual instruction pairs and proposes techniques to compose these to block-level instruction sequences.…”

ENTRA: Whole-systems energy transparency

“…The profiling framework executes tightly coupled threads through the xCORE pipeline, with random, valid operand values to produce an average power estimate for each instruction. Random input data is shown to cause higher power dissipation than more constrained data that would be found in real-world programs [19], e.g. due to data dependencies, thus creating a modest over-estimate in most cases.…”

“…We demonstrated that the variation due to data can range from 5 to 25% [8]. In [19] we examine the impact of operand values on instruction level energy consumption and propose a probabilistic approach to developing worstcase energy models suitable for safe worst-case energy consumption analysis.…”

Measuring Energy

“…In research, the wands have been used to collect energy data for processor and communication modelling [7], data-dependent energy modelling [8,9] and exploration of compiler optimizations for energy efficiency. Finally, the MAGEEC wand was of course pivotal in the research output of the MAGEEC energy efficient compiler optimization research.…”