Reading 16bit registers on 8bit mcu

[Pret]

I was curious on how can you read an 16bit timer (while it runs), store it in a 16bit variable and having consistent high and low byte. As I read into a random datasheet (say 18F26k22), I've seen that they made a trick with a hidden buffer on the high byte of the 16bit timer register. That's neat and it works fine.

But what if I'm using CCP on capture mode, being set on falling edge, without any interrupts. Just want to know how much time occurred since last falling edge. But reading the CCP register (say CCP_2) and store it in a local var, exposes the same issue.

For instance:

[asmboy]

you just did it the right way as far as i'm concerned

with or without INTS on - this will never fail

IMHO the optimum solution as you never have to kill global or an individual INTs

i regard killing global ints as a serious crime ,
when more than one actual, individual int is enabled.

[Mike Walne]

Could you consider this?

When a capture occurs, an int flag is set.
Clear the int flag before reading the timer.
If the int flag is set after reading the timer, another capture has occured.
You don't have to be actually using the int routines to use of the flags.

Mike

[Ttelmah]

Killing GLOBAL INTs, is perfectly acceptable, for reading a 16bit value updated in an interrupt. Remember it is only for a couple of instructions.
Microchip in their own examples and application notes use both methods for accessing values updated inside interrupts.

However it is not the way to handle this, since in this case it a hardware value, which will still update when interrupts are disabled. The read and test, _is_ the only way to go. Depending on the circumstances, you might want to do the update the other way round, so:

[Pret]

Yes, in this case it has nothing to do with interrupts.

Ttelmah, in your last example, it will definitely will create heart attacks on Lint or a second viewer for that uninitiated value, even it makes sense not to initialize it.

Though, I think your last example is not fully proof. You have to read it twice. For instance, if your random value for capt matches with a single read of a faulty (inconsistent) read of CCP_2? It's possible, right? Basically you might simply guess the wrong value of CCP_2.

[Ttelmah]

That's why I slightly prefer the first way.
It does depend on circumstances. In a loop that reads the value, the second is the more efficient, and since the value then should never be uninitialised, it avoids the extra read if it hasn't changed.

Best Wishes

[Pret]

Yes, but I still believe your very last version is not fully proof. Because with your uninitialized value you can simply guess a faulty read of CCP_2, and it moves on as the right value...

[Mike Walne]

[Pret]

We are not worrying about the meaning of CCP_x in application, we are worrying about the read of the 16bit register in a consistent manner, regarding its low and high byte. Using the interrupt flag seems to be a solution as well, but I find reading the register twice a more elegant manner.