The Basics of Profiling
summary
Explains when sampling versus instrumentation actually finds root causes, demos the game profiler Optick live and compares it to vTune; shows a mutex that disappeared from the profile but left startup unchanged, a bug where a texture rebuilt every frame kills rendering, and A* pathfinder code where linear search and bad std::map use cost tens of percent of CPU. Covers timeline visualization, thread/core contention, and practical advice to add lightweight instrumentation to builds so the next regression is obvious.
AI-written summary · about
chapters (12)
- 0:00Introduction to profiling
- 2:12The theory of performance
- 4:29Profiling methodology
- 8:27Sampling vs instrumentation
- 13:26Practical implementation
- 16:23Live demo: Game performance
- 21:58Visualizing profiling data
- 28:54Interpreting profiler results
- 37:22Advanced metrics with vTune
- 43:06Threading and synchronization
- 49:56Bottlenecks and data patterns
- 56:02Conclusion and Q&A
