Future vector microprocessor extensions for data aggregations

Abstract

Abstract-As the rate of annual data generation grows exponentially, there is a demand to aggregate and summarise vast amounts of information quickly. In the past, frequency scaling was relied upon to push application throughput. Today, Dennard scaling has ceased and further performance must come from exploiting parallelism. Single instruction-multiple data (SIMD) instruction sets offer a highly efficient and scalable way of exploiting data-level parallelism (DLP). While microprocessors originally offered very simple SIMD support targeted at multimedia applications, these extensions have been growing both in width and functionality. Observing this trend, we use a simulation framework to model future SIMD support and then propose and evaluate five different ways of vectorising data aggregation. We find that although data aggregation is abundant in DLP, it is often too irregular to be expressed efficiently using typical SIMD instructions. Based on this observation, we propose a set of novel algorithms and SIMD instructions to better capture this irregular DLP. Furthermore, we discover that the best algorithm is highly dependent on the characteristics of the input. Our proposed solution can dynamically choose the optimal algorithm in the majority of cases and achieves speedups between 2.7× and 7.6× over a scalar baseline.

Online Monitoring of Spatio-Temporal Properties for Imprecise Signals

“…s.s. signal 𝑠 : L × T → 𝐷 𝑛 is a signal that can be represented as a sequence of pairs {(𝑡 𝑖 , V 𝑖 )} 𝑖 ∈N , where each pair (𝑡 𝑖 , V 𝑖 ) of the sequence represents a piece of the signal, such that it maps any time-instant between 𝑡 𝑖 and 𝑡 𝑖+1 in T, to the |L| × 𝑛 matrix V 𝑖 that represents the values of the 𝑛 dimensions of the signal at each location ℓ in L. The space-synchronization restriction might appear to be a severe limitation, but this shows one of the conceptual differences between online and offline monitoring: in an offline setting, the space-synchronization hypothesis would likely have detrimental effects on the performances, as it would force all the processing to happen at a temporal granularity that is the union of the temporal granularities of the signals at the different locations. In an online setting, on the other hand, the temporal granularity is determined by the time when new information is available, and the space-synchronization hypothesis makes it possible to exploit in future work the Single-Instruction Multiple-Data (SIMD) capabilities of modern processors (see [13,18]), resulting in execution times that are virtually independent from the number of locations, when appropriate hardware is available. In this context, we call signal update u the triplet (𝑡 𝑎 , 𝑡 𝑏 , V), representing a mapping to the value matrix V for any time instant between 𝑡 𝑎 (included) and 𝑡 𝑏 (excluded).…”

Online monitoring of spatio-temporal properties for imprecise signals

“…In an offline setting, the space-synchronization hypothesis would likely have detrimental effects on the performances, as it would force all the processing to happen at a temporal granularity that is the union of the temporal granularities of the signals at the different locations. In an online setting, on the other hand, the temporal granularity is determined by the time when new information is available, and the space-synchronization hypothesis makes it possible to exploit in future work the Single-Instruction Multiple-Data (SIMD) capabilities of modern processors (see [13,18]), resulting in execution times that are virtually independent from the number of locations, when appropriate hardware is available. In this context, we call signal update u the triplet (𝑡 𝑎 , 𝑡 𝑏 , V), representing a mapping to the value matrix V for any time instant between 𝑡 𝑎 (included) and 𝑡 𝑏 (excluded).…”