memorization function decorator

From ActiveState Recipes. “Probably the fastest memoization decorator in the world”

Python code:

#coding:utf-8

from time import time

def memofunc(f):
    """ Memoization decorator for a function taking one argument"""
    class MemoDict(dict):
        def __missing__(self, key):
            ret = self[key] = f(key)
            return ret
    return MemoDict().__getitem__

def memoize(f):
    """ Memoization decorator for a function taking several argument"""
    class MemoDict(dict):
        def __getitem__(self, *key):
            return dict.__getitem__(self, key)
        
        def __missing__(self, key):
            ret = self[key] = f(*key)
            return ret
    return MemoDict().__getitem__

if __name__ == '__main__':
    loops = 10**6
    
    def square(x):
        return x**2
    @memofunc
    def m_square(x):
        return x**2
    
    start_time = time()
    for i in xrange(loops):
        square(10)
    end_time = time()
    print 'Total run time of square with', loops, 'times is', \
            (end_time - start_time)
    start_time = time()
    for i in xrange(loops):
        m_square(10)
    end_time = time()
    print 'Total run time of m_square with', loops, 'times is', \
            (end_time - start_time)
    
    def add(x, y):
        return x + y
    @memoize
    def m_add(x, y):
        return x + y
        
    start_time = time()
    for i in xrange(loops):
        add(120, 12)
    end_time = time()
    print 'Total run time of add with', loops, 'times is', \
            (end_time - start_time)
    start_time = time()
    for i in xrange(loops):
        m_add(120, 12)
    end_time = time()
    print 'Total run time of m_add with', loops, 'times is', \
            (end_time - start_time)
    

Result:

Total run time of square with 1000000 times is 0.209908008575
Total run time of m_square with 1000000 times is 0.0916221141815
Total run time of add with 1000000 times is 0.20007109642
Total run time of m_add with 1000000 times is 0.522424936295

m_add居然要比add还慢！可能是tuple作为dict key时效率不高吧？有待验证。