Making the Most Out of Your Compiler
summary
Claims: switching compilers and build flags gives measurable wins. Tools: clang with thin LTO, PGO, and -O3 -flto or GCC equivalents; CPU targets like -march=haswell/sse4.2 and inline-threshold tuning. Numbers: Mozilla sees ~5–7% with clang+LTO+PGO, ClickHouse reports 7–10% faster queries and ~12% smaller binaries, Google reports typical 3–5% platform wins and 2–5% binary-size drops from trivial-ABI/unstable-API. What’s built and works: databases and search engines rebuilt with these flags and yielded stable, repeatable speed and size improvements.
AI-written summary · about
chapters (18)
- 0:00Intro
- 1:04Trust your compiler
- 5:43The truth
- 8:37Visibility
- 15:54Optimizations
- 21:37Slope Vectorization
- 24:37Dash og
- 26:02bisect
- 27:43llvm
- 29:21Win forgotten story
- 33:37Architecture
- 36:17Native Architecture
- 37:37Micro Architecture
- 39:57Functional Lining
- 41:45Inline Threshold
- 45:32Always Inline
- 47:57Standard Library
- 50:51Charity
related talks
58mOptimization Tips - Mo' Hustle Mo' Problems
53mThinLTO Whole Program Optimization
37mC++ as an Optimizing Assembler - a Performance Talk
01h 37mAn Overview of Program Optimization Techniques
01h 05mPerformance Optimization in Software Development - Being Friendly to Your Hardware
56mWhy Your Go Benchmarks Are Lying
38mLow effort, High impact Android performance wins
33mReaching Ludicrous Speed