Re: [avr-gcc-list] Detecting ++/-- overflow

[David Brown]

> Speaking of overflows... what's an efficient way to detect when an
> increment or decrement results in an over- or underflow?
>
> Here's what I'm doing now:
>
>      signed int counter, overflow;
>
>      // Increment counter
>      if (counter == INT_MAX) {
>          counter = INT_MIN;
>          overflow++;
>      } else {
>          counter++;
>      }
>
> But you'd like to do something more like this: (I say this not having
> written assembly in a long time)
>
>      counter++;
>      if (bit_is_set(SREG, V)) {
>          overflow++;
>      }
>
> I don't know what you can count on the status register being preserved,
> though, and I don't really know how multiple byte and signed/unsigned
> types are handled. Looks like the INC instruction sets the overflow (V)
> flag for signed overflow only...
>

>From your first example, it looks like you were looking for singed types.
For unsigned types, I think it might be a bit easier.  Here's some ideas - I
haven't tried compiling them to see the code generated, but they might be of
interest.

1) Use a long/struct typedef:
    union { unsigned long big; struct { unsigned int counter, overflow; }; }
bigCounter;

    Then use "bigCounter.big++" for your increment, and read
bigCounter.counter or bigCounter.overflow as needed.  You could make "big"
and "overflow" signed, but not counter (your overflows would occur on the -1
to 0 transition).

2) For unsigned, you can use:
    if (!(++counter)) overflow++;
I don't know whether gcc will generate a seperate test for zero, or whether
it can use the flags from the ++counter directly.

3) Your "bit_is_set(SREG, V)" trick should work fine.  Use "V" for signed
counters, "C" for unsigned counters.


mvh.

David