On 2013-10-24 09:35, David Cole wrote:
>
> The reason that this useful is that this quirk can be exploited to
> create, for example, a MAX or MIN function that is evaluated at
> assembly time. (Maybe this has changed, but the last time I looked,
> the Assembler did not offer native MAX and MIN functions.)
>
Are you confident that was the motivation for the behavior?
I suppose it's possible.
> ...
> I'll translate this into something more readable:
> (A/B)/(A/B)*A
>
> Here's a better layout (better when your font is courier):
> A/B
> --- * A
> A/B
>
I recognize from the comments that you had a specialized
requirement and were not trying to provide a general purpose
MAX function. Regardless, I believe I spot the restrictions
(by inspection, not tested):
o I believe this gives MAX(-5,+3)=-5, not the +3 I'd expect.
o I believe MAX(-2147483648,-1) results in an integer overflow
error, ASMA057E.
(Should I see the source of your macro somewhere, or is it
proprietary?)
Here's my attempt at it. It has a more symmetric, although
(much) more restricted but still useful argument domain. It
does not depend on division by zero -- tested by translating
to Rexx which treats division by zero as a error. Test code
attached:
* MAX function through the marvel of integer arithmetic. Valid for
* arguments in the range [-357913940,357913940]. Arguments outside
* that range will either give correct results or report overflow
* with ASMA075E. Division by zero does not occur.
*
MACRO
&L MAX &X,&Y
L&SYSNDX DC A(((((&X)-(&Y))*3)/(((&X)-(&Y))*3+1))*(&Y)+(-3+((&Y)-(&X,
))*3)/(-2+((&Y)-(&X))*3)*(&X))
MEND
*
TINY CSECT
MAX +5,+3
MAX -5,+3
MAX +5,-3
MAX -5,-3
MAX 08,11
MAX 09,11
MAX 10,11
MAX 11,11
MAX 12,11
MAX 13,11
MAX 14,11
MAX 357913940,-357913940
MAX -357913940,357913940
MAX 999999999,999999999
MAX 357913941,-357913941
MAX 357913942,-357913942
END
Coefficients chosen empirically. I believe that with much more
Diophantine gymnastics I could make it work for all integer values.
I'm not much motivated.
-- gil
