Overview
Dive into a 43-minute conference talk from GOTO Chicago 2024 where Matt Godbolt, a Low-level Latency Geek, explores the essential knowledge programmers should have about CPU operations. Gain insights into the performance features of modern CPUs and understand how your code already benefits from them. The presentation covers crucial topics including modern CPU pipelines, branch prediction, execution processes, compiler optimizations, and memory management. Through multiple demonstrations, learn to diagnose performance issues and implement solutions when things go wrong. The talk provides a deeper appreciation for chip designers and compiler authors while equipping you with practical tools for performance analysis, including top-down analysis and perf tools. Perfect for developers wanting to understand the layers of abstraction between written code and actual CPU execution.
Syllabus
00:00 Intro
00:25 What to get out of this talk
01:13 A modern CPU pipeline
11:47 Branch prediction
21:27 Demo
26:15 Execution
28:56 Demo
34:53 Can the compiler save us?
35:08 Demo
36:46 Memory
37:52 Results
40:11 Top-down analysis
41:12 Perf
42:06 Conclusion
43:11 Outro
Taught by
GOTO Conferences