为什么使用局部常量比访问全局常量更快

Question

In [8]: %paste
PRIME = 16777619
UINT32_MAX = 4294967295  # (2 ** 32) - 1
def fnv1a_32_global(string):
    hashval = 2166136261
    for character in string:
        hashval ^= ord(character)
        hashval = (hashval * PRIME) & UINT32_MAX
    return hashval

In [9]: %timeit fnv1a_32_global(mystr)
1000000 loops, best of 3: 990 ns per loop

In [6]: %paste
def fnv1a_32_local(string):
    hashval = 2166136261
    PRIME = 16777619
    UINT32_MAX = 4294967295  # (2 ** 32) - 1
    for character in string:
        hashval ^= ord(character)
        hashval = (hashval * PRIME) & UINT32_MAX
    return hashval
In [7]: %timeit fnv1a_32_local(mystr)
1000000 loops, best of 3: 967 ns per loop

我正在尝试确认（并理解为什么）在每个函数调用上初始化本地var并且访问它比一次初始化全局var并访问它更快。

此结果在多次测试尝试中保持一致。

Additinal问题：这个代码可以更快地制作（我尝试使用c-library pyfasthash。它比上面的实现更慢）