Why is std::vector iteration faster than std::array iteration despite initial benchmarks suggesting otherwise?

Why std::vector Iteration Surpasses std::array Iteration

Original Conclusion Misconception

Initially, a flawed benchmark suggested that std::array iteration was faster than std::vector iteration. However, upon correcting the benchmark, it emerged that std::vector was significantly faster.

Benchmark Implementation

To ensure accuracy, the benchmark employed several improvements:

• Result utilization to prevent loop optimization
• -O3 optimization flag for enhanced speed
• std::chrono for isolated loop measurement without static variables interference

Results and Explanation

The benchmark results revealed that std::vector iteration completed in approximately 30 milliseconds, while std::array iteration took about 99 milliseconds.

The disparity stems from the memory page behavior. In the benchmark, the std::array was in the .bss section of the executable (with zero initialization), so its memory pages were not loaded into the process address space. Conversely, the std::vector had been allocated and zero-filled, resulting in page presence.

Solution

Pre-faulting the std::array's pages by zero-filling or using mlock() on Linux brings its pages into the address space, equating its iteration speed with that of std::vector.

The above is the detailed content of Why is std::vector iteration faster than std::array iteration despite initial benchmarks suggesting otherwise?. For more information, please follow other related articles on the PHP Chinese website!