Problem with default case in select {}

[abk]

Hi everyone!

I'm trying to count the bursts I read from a pin and the number of cycles between those bursts.
Without a default case I can count the pulses of a burst but if I add a default case that increases a variable my program doesn't recognize the pulses anymore. I don't understand why The channel end still works though..

The code below is the thread containing the select statement.

Code: Select all

void receiver(chanend c){
	unsigned dataIn = 0;
	int channelData = 0;
	unsigned pulseCounter = 0;
	unsigned counter = 0;

	while(1){
		select {
			/* Pin for bursts */
			case inP when pinsneq(dataIn) :> dataIn:
				pulseCounter++;
				break;
			/* Channel end for frequent reset */
			case c :> channelData:
				if (channelData == 0){
					printf("ctr: %u; plsCtr: %u\n",counter,pulseCounter);
					counter = 0;
					pulseCounter = 0;
				}
				break;
			default:
				counter++;
				break;
		}
		//counter++;
	}
}

The output generated

with default:
ctr: 7142857; plsCtr: 0
ctr: 7088387; plsCtr: 0
ctr: 7109786; plsCtr: 0

without default:
ctr: 0; plsCtr: 744
ctr: 0; plsCtr: 672
ctr: 0; plsCtr: 672

counter outside switch:
ctr: 744; plsCtr: 744
ctr: 669; plsCtr: 668
ctr: 673; plsCtr: 672

What is the reason for that behavior?

[richard]

Is dataIn declared as an unbuffered port (i.e. port dataIn)?

A default is case implemented by enabling events and then disabling events immediately afterwards. If any cases are ready it will cause an event when events are enabled, otherwise execution will fall-through into the code for the default case.

An unbuffered port will only event when it samples data from the pins meets the condition. By default the pins are only sampled on the rising edges the port clock. If no rising edge occours between events being enabled for the select and then disabled again then the case with the port will not fire.

[abk]

Richard, thank you for the quick and helpful reply!!
Buffering the port fixed the problem. I think I'll read the chapter about port buffering again :)

[rhind]

A default is case implemented by enabling events and then disabling events immediately afterwards. If any cases are ready it will cause an event when events are enabled, otherwise execution will fall-through into the code for the default case.

An unbuffered port will only event when it samples data from the pins meets the condition. By default the pins are only sampled on the rising edges the port clock. If no rising edge occours between events being enabled for the select and then disabled again then the case with the port will not fire.

If you don't have a 'default' in a select, and you are waiting on 2 unbuffered ports, and one of them fires an interrupt and while you'll handling that (in the case statement), the other port fires, does this mean you'll miss the second port firing because you're handling the first one? Or is this only an issue if you have a 'default' case?

If it is an issue, then this sounds like ports should always be declared as buffered so you don't miss any edges, doesn't it?

Thanks

Russell

[segher]

You'll get the event from that second port as soon as you enable events again, i.e.,
as soon as the first event is handled and you enter the select statement again.

If it fires twice before you get there, you get only one event; if both ports have an
event ready, you "randomly" get one of the two.

[rhind]

You'll get the event from that second port as soon as you enable events again, i.e.,
as soon as the first event is handled and you enter the select statement again.

If it fires twice before you get there, you get only one event; if both ports have an
event ready, you "randomly" get one of the two.

Thanks. Why does having a 'default' case cause you to miss if the pin is unbuffered then? From the sounds of this, once you've re-enabled the events after re-enterting the select you'd see the event on pin.