首先我们看下面的程序,它在循环中翻转LED,然后通过运行的时间和翻转次数,计算出每秒翻转的频率。
代码: 全选
from machine import Pin
import time
led = Pin('A13')
N = 200000
t0 = time.ticks_us()
for i in range(N):
led.on()
led.off()
t1 = time.ticks_us()
dt = time.ticks_diff(t1, t0)
fmt = '{:5.3f} sec, {:6.3f} usec/blink : {:8.2f} kblink/sec'
print(fmt.format(dt * 1e-6, dt / N, N / dt * 1e3))
3.381 sec, 16.905 usec/blink : 59.16 kblink/sec
在 MicroPython程序优化原则 中,提到尽量在程序中执行功能,不要在主程序中运行,因此可以将LED翻转放在函数中执行。
代码: 全选
from machine import Pin
import time
led = Pin('A13')
N = 200000
def blink_simple(n):
for i in range(n):
led.on()
led.off()
def time_it(f, n):
t0 = time.ticks_us()
f(n)
t1 = time.ticks_us()
dt = time.ticks_diff(t1, t0)
fmt = '{:5.3f} sec, {:6.3f} usec/blink : {:8.2f} kblink/sec'
print(fmt.format(dt * 1e-6, dt / n, n / dt * 1e3))
time_it(blink_simple, N)
2.902 sec, 14.509 usec/blink : 68.92 kblink/sec
可以看到,我们没有做什么实质的改到,就明显提高了速度。
循环是最消耗运行时间的,我们对循环中led.on()和led.off()两个动作进行优化,将它们预先载入内存,而无需循环中每次载入。
代码: 全选
from machine import Pin
import time
led = Pin('A13')
N = 200000
def blink_simple(n):
on = led.on
off = led.off
for i in range(n):
on()
off()
def time_it(f, n):
t0 = time.ticks_us()
f(n)
t1 = time.ticks_us()
dt = time.ticks_diff(t1, t0)
fmt = '{:5.3f} sec, {:6.3f} usec/blink : {:8.2f} kblink/sec'
print(fmt.format(dt * 1e-6, dt / n, n / dt * 1e3))
time_it(blink_simple, N)
1.617 sec, 8.086 usec/blink : 123.68 kblink/sec
速度提高了将近一倍。
进一步将循环中对 range(n) 也进行优化
代码: 全选
from machine import Pin
import time
led = Pin('A13')
N = 200000
def blink_simple(n):
on = led.on
off = led.off
r = range(n)
for i in r:
on()
off()
def time_it(f, n):
t0 = time.ticks_us()
f(n)
t1 = time.ticks_us()
dt = time.ticks_diff(t1, t0)
fmt = '{:5.3f} sec, {:6.3f} usec/blink : {:8.2f} kblink/sec'
print(fmt.format(dt * 1e-6, dt / n, n / dt * 1e3))
time_it(blink_simple, N)
1.121 sec, 5.607 usec/blink : 178.35 kblink/sec
效果非常明显。
进一步对循环中的操作优化,减少循环次数
代码: 全选
from machine import Pin
import time
led = Pin('A13')
N = 200000
def blink_simple(n):
n //= 8
on = led.on
off = led.off
r = range(n)
for i in r:
on()
off()
on()
off()
on()
off()
on()
off()
on()
off()
on()
off()
on()
off()
on()
off()
def time_it(f, n):
t0 = time.ticks_us()
f(n)
t1 = time.ticks_us()
dt = time.ticks_diff(t1, t0)
fmt = '{:5.3f} sec, {:6.3f} usec/blink : {:8.2f} kblink/sec'
print(fmt.format(dt * 1e-6, dt / n, n / dt * 1e3))
time_it(blink_simple, N)
0.913 sec, 4.563 usec/blink : 219.16 kblink/sec
根据MicroPython的优化功能,可以将程序声明为native code(本地代码),它使用CPU的操作码(opcode),而不是字节码(bytecode)
代码: 全选
from machine import Pin
import time
led = Pin('A13')
N = 200000
@micropython.native
def blink_simple(n):
n //= 8
on = led.on
off = led.off
r = range(n)
for i in r:
on()
off()
on()
off()
on()
off()
on()
off()
on()
off()
on()
off()
on()
off()
on()
off()
def time_it(f, n):
t0 = time.ticks_us()
f(n)
t1 = time.ticks_us()
dt = time.ticks_diff(t1, t0)
fmt = '{:5.3f} sec, {:6.3f} usec/blink : {:8.2f} kblink/sec'
print(fmt.format(dt * 1e-6, dt / n, n / dt * 1e3))
time_it(blink_simple, N)
0.704 sec, 3.521 usec/blink : 284.00 kblink/sec
除了native,还可以使用viper code模式,它进一步提升了整数计算和位操作性能
代码: 全选
from machine import Pin
import time, stm
led = Pin('A13')
N = 200000
@micropython.viper
def blink_simple(n:int):
n //= 8
p = ptr16(stm.GPIOB + stm.GPIO_BSRR)
for i in range(n):
p[0] = 1 << 4
p[1] = 1 << 4
p[0] = 1 << 4
p[1] = 1 << 4
p[0] = 1 << 4
p[1] = 1 << 4
p[0] = 1 << 4
p[1] = 1 << 4
p[0] = 1 << 4
p[1] = 1 << 4
p[0] = 1 << 4
p[1] = 1 << 4
p[0] = 1 << 4
p[1] = 1 << 4
p[0] = 1 << 4
p[1] = 1 << 4
def time_it(f, n):
t0 = time.ticks_us()
f(n)
t1 = time.ticks_us()
dt = time.ticks_diff(t1, t0)
fmt = '{:5.3f} sec, {:6.3f} usec/blink : {:8.2f} kblink/sec'
print(fmt.format(dt * 1e-6, dt / n, n / dt * 1e3))
time_it(blink_simple, N)
0.016 sec, 0.078 usec/blink : 12879.13 kblink/sec
最终我们还可以通过嵌入汇编方式,最大限度提升性能
代码: 全选
from machine import Pin
import time, stm
led = Pin('A13')
N = 200000
@micropython.asm_thumb
def blink_simple(r0):
lsr(r0, r0, 3)
movwt(r1, stm.GPIOB + stm.GPIO_BSRR)
mov(r2, 1 << 4)
label(loop)
strh(r2, [r1, 0])
strh(r2, [r1, 2])
strh(r2, [r1, 0])
strh(r2, [r1, 2])
strh(r2, [r1, 0])
strh(r2, [r1, 2])
strh(r2, [r1, 0])
strh(r2, [r1, 2])
strh(r2, [r1, 0])
strh(r2, [r1, 2])
strh(r2, [r1, 0])
strh(r2, [r1, 2])
strh(r2, [r1, 0])
strh(r2, [r1, 2])
strh(r2, [r1, 0])
strh(r2, [r1, 2])
sub(r0, 1)
bne(loop)
def time_it(f, n):
t0 = time.ticks_us()
f(n)
t1 = time.ticks_us()
dt = time.ticks_diff(t1, t0)
fmt = '{:5.3f} sec, {:6.3f} usec/blink : {:8.2f} kblink/sec'
print(fmt.format(dt * 1e-6, dt / n, n / dt * 1e3))
time_it(blink_simple, N)
0.007 sec, 0.037 usec/blink : 27322.40 kblink/sec
这个结果已经非常接近极限了。
从前面的优化顺序,可以看到我们并没有大幅修改程序,就可以极高程序的性能。实际使用中,大家可以灵活选择,提高程序的性能。