Using a global timer to time mutiple Events

dermot

Hi,

I want to have one central timer running and by doing calculations on this running timer I want to be able to time mutiple events (up to a max of 15 events). How can this be implemented most effectively using least resources. Accuracy or timer in hundreds of seconds is accurate enough. No event will be longer than 60seconds. Will I require a 3 seperate variable for each event to be timed??

Is starting a new cog for the timer and performing the following calculation for each event the best way to do it.

StartTime1:=Current Time
FinishTime1:=CurrentTime

FinishTime1-StartTime1:=Event1Time

Duane Degn

dermot wrote: »

Hi,

I want to have one central timer running and by doing calculations on this running timer I want to be able to time mutiple events (up to a max of 15 events). How can this be implemented most effectively using least resources. Accuracy or timer in hundreds of seconds is accurate enough. No event will be longer than 60seconds. Will I require a 3 seperate variable for each event to be timed??

Is starting a new cog for the timer and performing the following calculation for each event the best way to do it.

StartTime1:=Current Time
FinishTime1:=CurrentTime

FinishTime1-StartTime1:=Event1Time

You'll need a variable for each start time and some some way to indicate the interval you wish to time.


This is pretty much the way I do it.

  interval1 := [B]clkfreq[/B] / 4
  startTime1 := [B]cnt
[/B] [B] repeat
[/B]    if [B]cnt[/B] - startTime1 > interval1
      startTime1 += interval1 ' use this if you want somethink done [B]every[/B] quarter of a second
      ' do stuff when a quarter of a second if up
    ' do stuff all the time

This take two variables per timer. You could use a constant as the interval if you check prefer.

Make sure you make these variables longs.

As I read your post again I see you might not want to check when a set time passes but how much time has passed.

There's a trick to do this with one variable. I'll see if I can remember it.

timer1 := -[B]cnt[/B] ' this is the trick part
  ' do stuff that takes time
  timer1 += [B]cnt
[/B]

I don't use this trick myself, I generally use two variables.

startTime1 := [B]cnt
[/B]  'do stuff that takes time
  timeInterval1 := [B]cnt[/B] - startTime1

I think I do it this way out of habit.

I only recently saw Phil do the "-cnt" trick but I haven't started using it in my code yet.

jmg

dermot wrote: »

Accuracy or timer in hundreds of seconds is accurate enough.

I guess that was meant to be hundredths of a second (10ms) ?

dermot wrote: »

No event will be longer than 60seconds.

Will I require a 3 seperate variable for each event to be timed??

Is starting a new cog for the timer and performing the following calculation for each event the best way to do it.

StartTime1:=Current Time
FinishTime1:=CurrentTime

FinishTime1-StartTime1:=Event1Time

( I have assumed "each event to be timed" means capturing external pin changes, if you want to create time ticks, that is a different approach. )

Depends if the events are completely independent, and your desired precision. PASM + many cogs would likely get under 1us.

At 10ms, you may be able to poll all inputs in a tight loop, looking for a change, and then store at that location.

You will want to slow the Prop down a little, to avoid any chance of two overflows in your maximum time (+ margin).
eg 1/(80e6/2^32) = 53.6870912 seconds.

If two inputs can change inside one sample, then your extract-code needs to be able to find multiple changes.

What is the maximum edge rate ?
If it is slow, another solution would be to have one Cog doing a fast-poll and store of Time and Change tags only, into a FIFO, and then another cog can read out from there, at a slower average speed.
This would give you the best precisions, over many sampled channels.

JonnyMac

If you can spare a cog it would be easy to create a 1ms resolution timer that would get you past the limit of the cnt register. Without monkey motion, i.e., just using a 31-bit value your timer(s) can run up to 2,147,483 seconds (35,791 minutes; 596 hours -- you get the point).

Mike G

I second JonnyMac's 1ms timer. You could place the 1ms timer in an object along with an array of 15 variables exposed by getters and setters. A SetMark(id) method could mark the current timer value. A GetDelta(id) method could return the difference between the marked time and current timer value.

kwinn

Or you could use the variables as counters in an object and use start_timer, stop_timer, get_time.

dermot

I wish to both check when a set time has passed and check how much time has passed.

Timers will be triggered by serial data coming from external sensors. Up to 15 may run simultaneously.

10ms resolution is accurate enough for me.

Will I need 2 cogs for all this? One cog for timer (see code below) and another cog for taking all the time references.

repeat
 waitcnt(clkfreq/100+cnt)       'Hold 10ms
 timer:=timer+1                    'Add one to timer

If I set up a 31 bit timer will do that do the trick for just using one cog?? How do you set this up?

Mag748

If the below statement:

startTime1 += interval1

results in startTime1 being the value 4294967295, then it appears to me that the following statement:

if cnt - startTime1 > interval1

will never be true. This would cause the timer to stop functioning.
How come this is not a worry?

Unless these numbers are signed, in which case the maximum value for interval1 cannot exceed half of the value of a long, 2147483647?

Duane Degn

Mag748 wrote: »

Unless these numbers are signed, in which case the maximum value for interval1 cannot exceed half of the value of a long, 2147483647?

They are signed. I think your right about the 31-bit limit so this will only work for intervals up to 26.84 seconds @80MHz.

jmg

dermot wrote: »

Timers will be triggered by serial data coming from external sensors. Up to 15 may run simultaneously.
10ms resolution is accurate enough for me.

So this is some custom Sensor protocol, that encodes values in a very low baud rate way ?

dermot wrote: »

Will I need 2 cogs for all this? One cog for timer (see code below) and another cog for taking all the time references.

repeat
 waitcnt(clkfreq/100+cnt)       'Hold 10ms
 timer:=timer+1                    'Add one to timer

The waitcnt 'snaps' to the next 10ms, so if more code is added in the loop, _provided_ it never takes more than 10ms you still keep your timebase.
Fastest additional steps, as pseudo code, will be a simple write-on-change

 PinDiff = (MaskedPins XOR OldPins)
 IF  PinDiff <> 0 THEN 
   OldPins = MaskedPins
   StoreTime[WrPtr] = timer
   StoreDiff[WrPtr] = PinDiff
   StoreWrPtr = WrPtr       // WrPtr is local, StoreWrPtr is a copy visible to another cog, so you can make a fifo 
   WrPtr = (WrPtr+1) MOD Size(StoreTime)  // post INC, circular buffer
   IF WrPtr = StoreRdPtr THEN
      // oops, write has over-flowed, too many edges for read to keep up
   ENDIF
 ENDIF

and the other Cogs code can scan StoreTime, StoreDiff arrays, and StoreWrPtr which are all in main memory

You may be able to decode more inside this loop, but if you have two cogs available, often keeping the tight stuff deterministic if a good idea.
With simple code, you can easily decrease the sampling time, until it breaks, to get best resolution.

ie 10ms is coarse, if you can run this at 1ms, why not do so ?

Or code it in PASM, and I'd guess you would get close to 1us resolution. (just over 1 hour roll over)

kwinn

I'm not sure what you mean by " I wish to both check when a set time has passed and check how much time has passed.". Perhaps you could explain it with a diagram or example.

I am certain that a single cog could easily handle 15 interval timers and an elapsed time counter with 1mS resolution. The cog would spend most of that time waiting for the 1mS to elapse. For 10 or 100mS resolution the cog would probably be waiting over 99.9% of the time.