Software Optimizations Become Simple with Top-down Microarchitecture Analysis

In this meeting we continue our quest for better performing code. Intel's Ahmad Yasin will present the top-down microarchitecture analysis method, which uses metrics provided by the CPU to get to the root of performance bottlenecks. Ahmad is a senior CPU architect and an expert in both hardware and software sides of performance evaluation.

Agenda:
12:00 - 12:30: Gathering and pizza, courtesy of Intel (Yes, this is a noon meeting, lunch included. Yey!)
12:30 - 13:30: Software Optimizations Become Simple with Top-down Microarchitecture Analysis through Linux perf and toplev Tools
13:30 - 13:45: Wrap-up and raffle - a license to a JetBrains product of choice will be given to one lucky participant

Abstract:
Top-down Microarchitecture Analysis (TMA) Method is an industry-proven systematic approach that identifies performance bottlenecks in out-of-order cores. Identifying true bottlenecks lets developers focus software tuning to remediate them and improve efficiently on same hardware. TMA simplifies cycle-accounting using microarchitecture independent-metrics organized in one single hierarchy which makes analysis simple. Using TMA, the high-learning curve associated with each microarchitecture generation is replaced by a structured drill-down that guides the user to true performance limiters.

This session will provide a short overview of the Intel Core™ microarchitecture, present the basic metrics of TMA (https://download.01.org/perfmon/TMA_Metrics.xlsx) and briefly demonstrate the interface through Intel’s VTune™ profiler. Finally, a demo on some C++ multithreaded code will be included using pmu-tools/toplev (a wrapper of the Linux perf tool).

Bio:
Ahmad Yasin is a CPU senior architect at Intel focusing on performance monitoring and analysis. He owned the Performance Monitoring Unit (PMU) of Core-based client and server products like SandyBridge and Skylake. He works closely with users and Intel customers on performance optimizations and software-debug capabilities. To date, Ahmad has 9 filed patents and 10 published conference/journal papers.

Note:
While this is a somewhat Intel-oriented talk, the insights it provides will be of great use for developers using other platforms as well. It is also a great chance to not only see what's under the hood but also hear from the people that build the engine.