CAREER: Slice-Processing: Highly-Accurate Prediction for Future High-Performance Processors

日期：2000年4月19日PROPOSAL编号：C-CR 9984371INSTITUTION：西北大学PI：Andreas MohovosTITLE：Career：Slice-Processing：对未来高性能处理器的高精度预测摘要几乎所有现代高性能处理器都使用预测驱动的推测技术。这些技术为进一步提高性能提供了一个很有前途的方向。然而，这样获得的好处只与基于预测的底层方法在猜测程序需要什么方面一样好。本项目研究切片预测，这是一种旨在提供比现有方法所能提供的高精度预测的范例。现有方法主要基于结果，因为它们依赖于程序结果流中的重复模式(例如，地址或分支方向)。然而，这样的模式并不总是存在的。更重要的是，随着当前基于结果的方法的完善，不规则模式越来越多地主导性能。切片预测通过预测产生这种不规则但对性能至关重要的结果的计算切片来攻击这种情况。然后，预测切片被视为微型、自主的程序，可以预先计算出性能关键的结果。直觉是，虽然程序的结果可能是不规则的，但所使用的方法通常相对简单和相当稳定。本项目专注于切片预测的体系结构不可见的硬件实现，并针对以下应用：(1)分支预测和(2)数据预取。通过一个详细的软件模拟器，在一组广泛使用的基准上评估了切片预测所获得的性能。

DATE: April 19, 2000PROPOSAL NUMBER: C-CR 9984371INSTITUTION: Northwestern UniversityPI: Andreas MoshovosTITLE: CAREER: Slice-Processing: Highly-Accurate Prediction for Future High-Performance ProcessorsABSTRACTPrediction-driven speculative techniques are used in virtually all modern, high-performance processors. Such techniques offer a promising direction for furtherboosting performance. However, the benefits so obtained are only as good as theunderlying prediction-based methods are at guessing what the program needs.This project investigates slice-prediction, a paradigm that aims at providinghighly accurate prediction beyond what is possible with the existing methods.Existing methods are primarily outcome-based since they rely on repeatingpatterns in a program's outcome stream (e.g., addresses or branch directions).However, such patterns do not always exist. More importantly, as currentoutcome-based methods are perfected, irregular patterns increasingly dominateperformance. Slice-prediction attacks such cases by predicting the computation-slice that produces such irregular yet performance-critical outcomes. Predictedslices are then treated as miniature, autonomous programs, which can precomputeperformance critical outcomes. The intuition is that while the outcome of aprogram can be irregular the method used is typically relatively simple andfairly stable.This project focuses on architecturally invisible, hardware implementations ofslice-prediction, and targets the following applications: (1) Branch Predictionand (2) Data Prefetching. Performance achieved by slice-prediction is evaluatedthrough a detailed software simulator over a set of widely used benchmarks.