Advancements in info table profiling (Lightning Talk)
summary
Info table profiling attributes heap allocations to specific info tables using flags like -finfo-table-map and -fdistinct-constructor-tables, and produces event logs consumable by eventlog2html and ghc-debug. Recent work reduces compile-time and memory overhead from roughly +130% time and +67% memory to about +33% time and +10% memory, adds filters such as -fno-info-table-map-with-stack to omit stack/defaulted entries, and narrows distinct-constructor expansion to selected constructors or per-module to keep binaries small (Agda case: +36% → +21%).
AI-written summary · about
chapters (8)
- 0:00Introduction
- 0:30What is an info table
- 1:06Info table profiling
- 2:34Why info table profiling
- 3:38Using info table profiling
- 4:11Lightning demo
- 5:48Advancements
- 8:55Outro
related talks
19mProfiling Your Code: 5 Tips to Significantly Boost Performance
26mProfiling Mastery: Decode Your Data and Elevate Observability
33mProfiling Code
16mMemory Allocation Profiling upcoming features
13mLightning Talk Slot #3: The Haskell Optimization Handbook
24mLow-overhead memory allocation tracking
16mTypeCraft: A Lightweight Data Type Profiler with High Resolution
05mHow To Profile Compilation Duration?