An Analysis of Latency in Data

Recent evidence indicates that the exploitation of locality in dataaow programs could have a dramatic impact on performance. The current trend in the design of dataaow processors suggest a synthesis of traditional non-strict ne grain instruction execution and a strict coarse grain execution in order to exploit locality. While an increase in instruction granularity will favor the exploitation of locality within a single execution thread, the resulting grain size may increase latency among execution threads. In this paper, the resulting la-tency incurred through the partitioning of ne grain instructions into coarser grain threads is evaluated. We deene the concept of a cluster of ne grain instructions to quantify coarse grain input and output latencies using a set of numeric//oating point benchmarks. The results ooer compelling evidence that the inner loops of a signiicant number of numeric codes would beneet from coarse grain execution. Based on cluster execution times, more than 60% of the measured benchmarks favor a non-ne grain execution. In 45% of the benchmarks, the overall execution time of clusters is not eeected by the execution mode. These results suggest that the eeects of increased instruction granularity on latency is minimal for a high percentage of the measured codes, and in large part is ooset by available intra-thread locality.