No abstract

In multimedia and other streaming applications a significant portion of energy is spent on data transfers. Exploiting data reuse opportunities in the application, we can reduce this energy by making copies of frequently used data in a small local memory and replacing speed and power inefficient transfers from main off-chip memory by more efficient local data transfers. In this paper we present an automated approach for analyzing these opportunities in a program that allows modification of the program to use custom scratch pad memory configurations comprising a hierarchical set of buffers for local storage of frequently reused data. Using our approach we are able to reduce energy consumption of the memory subsystem when using a scratch pad memory by a factor of two on average compared to a cache of the same size.

The memory subsystem of a complex multiprocessor systemson-chip (MPSoC) is an important contributor to the chip power consumption. The selection of memory architecture, as well as of communication architecture, both affect the power efficiency of the design. In this paper we propose a novel approach that enables energy-aware co-synthesis of both memory and communication architecture for streaming applications. As opposed to earlier techniques, we employ a powerful compile-time analysis of memory access behavior that adds flexibility in selecting memory architectures. Additionally, we target TDMA bus-based communication architectures, which not only guarantee performance, but also greatly reduce the design time and allow us to find the energy optimal system configuration. We propose and compare three techniques: an optimal mixed ILPbased co-synthesis technique, a mixed ILP-based traditional twostep synthesis approach where memory and communication synthesis is performed sequentially, and a co-synthesis heuristic that synthesizes energy-efficient hierarchical bus-based communication architectures with guaranteed throughput. Our experimental results on a number of streaming applications show that both the traditional two-step synthesis approach and heuristic result in up to 50% worse power consumption in comparison with proposed co-synthesis approach. However, on some of the streaming benchmarks, our co-synthesis heuristic approach was able to find optimal or near-optimal results in a much shorter time than the MILP co-synthesis approach.294 traditional two-step synthesis approach (first memory, then communication synthesis) and also with a simple co-synthesis heuristic. We show that MILP-based optimal co-synthesis approach provides results that outperform two-step or heuristic approach in a reasonable amount of time for the benchmarks we used. However, our heuristic, being much less computationally expensive than MILP co-synthesis approach, achieves near-optimal results on some of the benchmarks, which makes it a good candidate for solving the problems for which the MILP co-synthesis technique is not able to produce the results in a reasonable amount of time.

No abstract

Abstract