"使用自动重载"

even if you don't use auto reload, you can still make the timing incredibly accurate, if you code carefully.

see the example I posted earlier. Its long-term accuracy depends only on the crystal.

我想，对于单片机，只要有一个定时器，可以赋初值，无论它可不可以重载，做RTC都只取决于晶振的精度，与其它无关，因为指令的执行时间和中断时间都是可测的，采用汇编就可以精确的控制定时器的时间。

回复【49楼】telefunken  德律MM
老子.......
各种放屁啊....
呃..我错了...粗口了....
单片机自己的内部的时钟究竟有多精密,诸君看过datasheet没有.很多片子内部时钟都是1%误差起步啊,居然还在这里谈论精密,有个啥意义哦.
-----------------------------------------------------------------------

我从来不用单片机内部的RC，为了省钱也不会这样。

回复【53楼】millwood0  
" th0=(65536-50000)/256;  
tl0=(65536-50000)%256;//这样的定时数据，是不是正好是50000*1微秒=50毫秒？  "
this way of setting timers is the biggest mistake that contributes to bad timing.
the reason is fairly simple: because of interrupt latency / context saving associated with entering an interrupt, by the time your code executes to this point, it is likely that the timer has advanced to a point where tl0 has some va......
-----------------------------------------------------------------------

我想不是。先记录了50000计数，溢出、进入中断程序，重新赋值，再计数50000……

the following hopefully will help you all understand the issue here.

this is a shot of single stepping through an AVR timer interrupt service routine. I stop'd the code right after it entered the isr, before executing the first statement.

on the right, I was showing the value of the timer / counter, TCNT0. The timer here works similar to the C51 timer: once TCNT0 hits a pre-set value (100 here and 0x100 in the case of a C51 8-bit timer), it rolls over and fires an interrupt.

so TCNT0 was just reset and that triggered the interrupt.

because of latency and context saving, by the time we are to execute the very 1st user code in the isr, the timer has advanced from 0x00 to 0x19.

if you ignore the value of the timer / counter at this point, by writing an explicit value to it, you would have created timing errors.

instead, you should simply increment or decrement the timer, as suggested earlier.



(原文件名:avr tmr isr.PNG)