Re: int vs binary questions

[Hans Boldt <boldt@xxxxxxxxxx>]

Steve Richter wrote:
> I have a 170 with 460 cpw.  But that should not matter, all the results are
> relative to the same system.  Bin to bin is 15% faster than int to int.  Add
> to bin is off the charts, 20x slower than add to int.
>
> I would like to know if the rpg definition of "Binary" is unique to it and
> if it predates the s/38.   And was the as400 database handling of binary
> something forced on it by RPG?

Time again for a history lesson, eh?

Let's go back to RPG II, as implemented on machines like the S/34. RPG 
II only had zoned decimal numeric as the internal representation for 
numerics. (Externally, it supported a bunch of other formats, like 
packed and binary decimal, left/right signed, etc.)

RPG III on the S/38 added binary decimal and packed decimal as internal 
numeric formats, and made them work exactly the same as zoned decimal. 
(Call all three the "decimal" formats.) It did this by implicitly 
converting binary and zoned values to packed before all arithmetic.

As far as DDS is concerned, "binary" can hold 2, 4, or 8 byte two's 
complement values. RPG is the odd one out since it maps DDS's "B" format 
into RPG's "B" format. Clearly, that's not always the best match, and 
never was.

RPG IV added (eventually) three other numeric formats: float, integer, 
and unsigned, each with its own unique set of characteristics. As is 
true with most CPU's, the iSeries implements arithmetic fastest in 
binary, and so RPG's integer and unsigned formats are the fastest for 
arithmetic. For compatibility, RPG IV had to continue to map the "B" 
format in DDS to RPG's "B" format. Thus, the EXTBININT keyword is needed 
to tell the compiler to break compatibility, and do the right thing instead.

Cheers! Hans