From Top-down Microarchitecture Analysis to Structured Performance Optimizations

Author:

Yasin Ahmad1

Affiliation:

1. Intel Corporation

Abstract

This session re-caps on the Top-down µarch Analysis (TMA) method - that is widely adopted in performance profiling tools, Microarchitecture challenges faced in out-of-order cores, and the abstraction that helped the method to be universally supported across CPU vendors (Intel as well as AMD ARM).Then, we show how the primary TMA metrics of Frontend Bound, Bad Speculation, Core Bound, Memory Bound and Retiring can be used to classify and direct exploitation of popular software optimizations.The session closes with a use-case that got deployed in code generation of modern compilers. The use-case demonstrates how to mitigate Instruction Fetch Bandwidth issue through tuning of loop unrolling to speedup tight loops in recent wide-issue out-of-order cores.

Publisher

Cassyni

Reference5 articles.

1. A Top-Down method for performance analysis and counters architecture

2. https://dyninst.github.io/scalable_tools_workshop/petascale2018/assets/slides/TMA%20addressing%20challenges%20in%20Icelake%20-%20Ahmad%20Yasin.pdf

3. aayasin/perf-tools: (https://github.com/aayasin/perf-tools)

4. Tuning Performance via Metrics with Expectations

5. A Metric-Guided Method for Discovering Impactful Features and Architectural Insights for Skylake-Based Processors

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3