为何将 0.1f 改为 0 会使性能降低 10 倍？

Why doesn't GCC optimize a*a*a*a*a*a to (a*a*a)*(a*a*a)?