How to Measure Elapsed Time if Longer Than System Clock Loop

coryco2

I am trying to set up a central "elapsed time" counter which methods in multiple cogs can all refer to. Easy enough, just write cnt to a global variable at the start and have the methods compare the current cnt at any given time to that value. But then how to handle when the system counter reaches its maximum value and loops around? I vaguely remember something to do with shifting bits to track a change like this, but the concept has escaped me. Can anyone point me in the right direction, please?

lonesock

I do this:

''  Timing
DAT
mstimer_ms    long      0
mstimer_last  long      0
PUB msUpdate : delta_ms | ms_clocks
  ms_clocks := (clkfreq + 500) / 1000
  delta_ms := (cnt - mstimer_last) / ms_clocks
  mstimer_ms += delta_ms
  mstimer_last += delta_ms * ms_clocks

Some main loop somewhere can call msUpdate repeatedly, and anyone can read the 'mstimer_ms' long. If you want the update function to be callable from multiple cogs, there is a lock version here.

Jonathan

coryco2

Ok, I think I figured out a way using the POSX constant:.

   IF cnt < time
      time := (POSX - time)+ cnt

Dave Hein

If your elapsed time is less than 2^31 cycles you can just do (cnt - start_cycles) to compute the elapsed time. This works even when CNT wraps. If your elapsed time can be more than 2^31 cycles, and msec resolution is OK then I would use lonesock's routine.

coryco2

@lonesock

Why are you adding 500 in this line?

ms_clocks := (clkfreq + 500) / 1000

lonesock

I call it pre-rounding. Integer division will just truncate the fractional part. If wanted to round to the nearest integer, you would typically say something like this (in C notation):

i = (int)( float_a / float_b + 0.5 )

Since 0.5 can be re-written as 0.5 * float_b / float_b, that means you can combine terms and say

float_a / float_b + 0.5 = (float_a + 0.5 * float_b) / float_b

So adding half of the denominator to the numerator before the division is the same as adding 0.5 after the division, and it can be done with only integers.

Does that make sense?
Jonathan