Re: Definition of integer division

Geoffrey.Coram wrote:

>Div by zero should be an error, whether integer or real.
>  
>
For real you can return NaN (not a number) which will propagate through 
arithmetic like x, so it can be considered a recoverable error - i.e. 
the program doesn't need to terminate. From a performance standpoint 
it's probably cheaper to evaluate a large piece of arithmetic without 
checking for zeros and just check at the end for NaN.

>What happens when I write this:
>
>integer a, res;
>a = 0;
>res = 5 + (1/a);
>
>1/a is now "x" but then is converted/coerced to integer(?)
>for addition with 5.  What is the result of that "conversion"?
>  
>
x - it's a 4-state '+' because it is a 4-state '/'.

>How do you make "x" an integer?  Do you get an error because
>the simulator can't convert it to an integer?  That's sort of
>indirect.
>  
>
I think SV just maps 4-state x/z to 2-state 0 (silently), so if  "res" 
was 2-state it would become 0.

This would be simpler if you could do C++ style try/catch around suspect 
code and have the language throw exceptions for division by zero or bad 
mappings, but I don't think you'll be seeing that in AMS or SV (ever).

Kev.

>-Geoffrey
>
>
>
>Kevin Cameron wrote:
>  
>
>>Not much you can do about integer division by zero. Presumably this will
>>have to be addressed in SV which supports 2-state types (where x cannot
>>be the result) - I can't remember it coming up.
>>
>>Kev.
>>    
>>
>
>  
>