Your backtester stores a matrix of 1000 stocks across 100,000 timestamps. You wrote a loop to calculate the daily cross-sectional return, but it's terribly slow. Memory profiling shows a 90% cache miss rate. What is the root cause and how do you redesign the memory layout?

你的回测框架用矩阵存了1000只股票在10万个时间点的数据。你写了个循环算每天的横截面收益，但跑得极慢，性能分析显示缓存未命中率高达90%。根本原因是什么？你会怎么重新设计内存布局？