Writing Efficient C++ Code
2 days ago
- C++ offers a unique compromise between high-level features and low-level hardware access, making it ideal for performance-critical applications like games.
- Object-oriented programming can lead to cache misses and parallelization difficulties due to scattered small objects and pointer-heavy data structures.
- Data-Oriented Design (DOD) focuses on data layout in memory first, using contiguous arrays and minimizing cache misses.
- The memory hierarchy (registers, cache, RAM, disk, internet) shows that accessing RAM is hundreds of cycles slower than cache, emphasizing the importance of cache-friendly patterns.
- Good practices include using arrays (like std::vector) over linked lists, preferring Structure of Arrays (SOA) over Array of Structures (AOS) for hot/cold data separation, and reusing objects to avoid repeated allocations.
- Compiler optimizations can be hindered by pointer aliasing; using local copies or __restrict helps, and declaring variables outside loops can save significant time (e.g., 25% for string reuse).
- Disabling unnecessary features (exceptions, RTTI) and using custom allocators can improve performance. The STL is efficient but requires awareness of its internal allocations (e.g., list/set vs vector).
- Optimization should be considered throughout development, not just after profiling, to avoid costly redesigns.