python - 制作无分支计数器的最佳方法是什么？

Question

我正在为现场设备使用 Python3 中的计数器。它尝试联系设备并返回在线和离线的总数。

def count_online(mac_list):
    online = 0
    offline = 0

    for mac in mac_list:

        is_online = poll_device(mac) # ask the device if it is online, return bool

        if is_online:
            online += 1
        else:
            offline += 1

    return online, offline

我见过“ Branchless method to convert false/true to -1/+1?”以及“ Why "If" is Sloowww”，但是我需要对在线和离线设备进行计数。我也很欣赏在 Python 中，尝试无分支可能有点毫无意义，因为它不是像 C 这样的低级语言。另一方面， Hopper Necklace提醒我们不要浪费我们的微秒，那么有没有更有效的方法来写这个？无分支方法会减少运行时间吗？

Answer 1

您可以简单地将 True 相加并从长度中减去它们 - 这将为您提供无分支代码:

def poll_device(_):
    poll_device.state = not poll_device.state
    return poll_device.state

poll_device.state = False

def count_online(mac_list):
    online = 0
    offline = 0
    for mac in mac_list:
        online += poll_device(mac) # ask the device if it is online, return bool

    return online, len(mac_list)-online

on,off = count_online(range(10))

print(on,off)

import dis

dis.dis(count_online)

输出:

5 5

# disassembly
  9           0 LOAD_CONST               1 (0)
              2 STORE_FAST               1 (online)

 10           4 LOAD_CONST               1 (0)
              6 STORE_FAST               2 (offline)

 11           8 SETUP_LOOP              24 (to 34)
             10 LOAD_FAST                0 (mac_list)
             12 GET_ITER
        >>   14 FOR_ITER                16 (to 32)
             16 STORE_FAST               3 (mac)

 12          18 LOAD_FAST                1 (online)
             20 LOAD_GLOBAL              0 (poll_device)
             22 LOAD_FAST                3 (mac)
             24 CALL_FUNCTION            1
             26 INPLACE_ADD
             28 STORE_FAST               1 (online)
             30 JUMP_ABSOLUTE           14
        >>   32 POP_BLOCK

 14     >>   34 LOAD_FAST                1 (online)
             36 LOAD_GLOBAL              1 (len)
             38 LOAD_FAST                0 (mac_list)
             40 CALL_FUNCTION            1
             42 LOAD_FAST                1 (online)
             44 BINARY_SUBTRACT
             46 BUILD_TUPLE              2
             48 RETURN_VALUE

与您的代码反汇编相比:

 21           0 LOAD_CONST               1 (0)
              2 STORE_FAST               1 (online)

 22           4 LOAD_CONST               1 (0)
              6 STORE_FAST               2 (offline)

 24           8 SETUP_LOOP              42 (to 52)
             10 LOAD_FAST                0 (mac_list)
             12 GET_ITER
        >>   14 FOR_ITER                34 (to 50)
             16 STORE_FAST               3 (mac)

 26          18 LOAD_GLOBAL              0 (poll_device)
             20 LOAD_FAST                3 (mac)
             22 CALL_FUNCTION            1
             24 STORE_FAST               4 (is_online)

 28          26 LOAD_FAST                4 (is_online)
             28 POP_JUMP_IF_FALSE       40

 29          30 LOAD_FAST                1 (online)
             32 LOAD_CONST               2 (1)
             34 INPLACE_ADD
             36 STORE_FAST               1 (online)
             38 JUMP_ABSOLUTE           14

 31     >>   40 LOAD_FAST                2 (offline)
             42 LOAD_CONST               2 (1)
             44 INPLACE_ADD
             46 STORE_FAST               2 (offline)
             48 JUMP_ABSOLUTE           14
        >>   50 POP_BLOCK

 33     >>   52 LOAD_FAST                1 (online)
             54 LOAD_FAST                2 (offline)
             56 BUILD_TUPLE              2
             58 RETURN_VALUE

但我不太确定这种事先进行的“优化”是否明智 - 我只会选择 if ... else ...方法，直到此代码部分成为瓶颈(测量它)...对 poll_device(mac) 的调用比这个无分支结构慢 100-1000 倍，这会影响你的表现。