delay() duration

[fulda1]

Hi,

yesterday I did some easy application, which needs to generate 2 millisecond pulse. Very easy, so I used:

digitalWrite(MyPin, HIGH);
delay(2);
digitalWrite(MyPin, LOW);

Nothing special, but I was surprised, that pulse is not 2ms, but 1ms > pulse > 2ms. So I go to delay() implementation and found, that it count number of milliseconds from SysTick (System Counter). But the first millisecond is not properly rounded.

To demonstrate issue, let us have time scale like following. It present 2 milliseconds of time, each millisecond is split to "J" tenths.

0                             1                             2
|                             |                             |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
|  |  |  |  |  |  |  |  |  |  |  |  |  |  |  |  |  |  |  |  |
A  B  C  D  E  F  G  H  I  J  A  B  C  D  E  F  G  H  I  J  A

situation - when delay(2) is called in time 0.A, it will take almost 2ms. in case delay(2) is called in time 0.J, it will take 1.1 ms only.

so, looking how to "fix" it. It seems, it will be easy fix like:

  uint64_t m0 = GetTick();
  __IO uint32_t u0 = SysTick->CNT;
  uint64_t m1 = GetTick();
  __IO uint32_t u1 = SysTick->CNT;   //may be a interruption

  if (m1 != m0) {
    do {              // full millisecond
      yield();
    } while (getCurrentMillis() - m1 < ms);
    do {              // remaining partial millisecond
      yield();
    } while ((SysTick->CNT < u1 ) && (getCurrentMillis() - m1 < (ms + 1)));
  } else {
    do {              // full millisecond
      yield();
    } while (getCurrentMillis() - m0 < ms);
    do {              // remaining partial millisecond
      yield();
    } while ((SysTick->CNT < u0 ) && (getCurrentMillis() - m0 < (ms + 1)));
  }

I need to check and spent some time on it.

[maxint-rd]

That's interesting. In my projects I've not encountered this as a problem yet, but I have not yet done any precise measurements to verify such timings either. Unfortunately I won't have much spare time this month to reproduce your findings.

The delay() function simply uses getCurrentMillis() which calls GetTick() to get the value of the msTick global. As you found that value isn't rounded; it's just truncated to the next millisecond (by definition). I now understand better why delay(1) is not the same as a delayMicrosecond(1000). I haven't looked into it much further to see what the best approach would be to make delay more precise.

BTW. to reduce flash usage in the V003/V002 I made msTick global a 32 bit value. This avoids using lengthy div64 code.
Also make sure to check the fix mentioned here in issue #65.

[fulda1]

Yes, you are right. I did not spent enough time reading micros().
And yes, that surprise with different length od delay() is more visible with "short times", like 1,2 or 3.

I'm also overloaded for this week, but next week will look for it.