正常
Volume 6: Performance Optimization
Performance is the one area of C++ engineering where it is easiest to be confidently wrong — the complexity of microarchitecture runs far ahead of human intuition. The spine of this volume is a single chain: correctness first (the correctness foundation) → measure first (the benchmark methodology anchor) → attribute and optimize by bottleneck site (the four TMA buckets) → land on the performance cost of C++ abstractions. Every topic walks the same loop: cut in with C++ code → drop down to the hardware or the methodology → come back to how to change the C++.
One thesis runs through the whole volume: efficiency (algorithmic complexity) ≠ performance (real behavior on hardware). Don't stare at big-O alone — watch how the data actually flows through the hardware.
This volume has settled into a stable eight-chapter structure (2026-07-03: the legacy standalone articles 02-inline / avx were deleted, 06-evaluating was moved out as embedded material, and the three sanitizer articles were relocated into ch00 as 03/04/05).
Chapter navigation
ch00 · Performance mindset and correctness firstch01 · Benchmark methodology (the volume's anchor)ch02 · CPU microarchitecture and the memory hierarchych03 · Attribution methodology: from measurement to bottleneckch04 · Tuning by bottleneck site (the technical core)ch05 · Multicore performance (picking up from vol5)ch06 · The performance cost of C++ abstractionsch07 · Compiler optimization boundaries and binary size