Explore a groundbreaking profiler called MemPerf, designed to identify performance slowdowns caused by memory allocators in this 19-minute conference talk from OOPSLA2 2023. Learn about the innovative life-cycle based detection approach that separates slow memory management operations from slow accesses. Discover how MemPerf utilizes thread-aware and type-aware performance modeling to identify slow management operations, and employs a top-down approach to pinpoint reasons for slow memory accesses. Gain insights into the profiler's effectiveness in correctly reporting allocator-induced slowdowns and its ability to uncover both known and unknown design issues in widely-used allocators. Understand the significance of this tool in helping programmers focus their optimization efforts more effectively, potentially saving time and resources in application development.
Overview
Syllabus
[OOPSLA23] MemPerf: Profiling Allocator-Induced Performance Slowdowns
Taught by
ACM SIGPLAN