Re: 'Big Integers' -re a Fall06 thread ADDLC

Bruce Wilson wrote:
<snip>

-- the POWER6 overview presented recently noted that this 'technology'
includes native 'decimal arithmatic'  .. the 'puff' piece said that it
is a response to federal requirements that 3 x 5 = will never be
14.999999999...

<snip>


Here's a link to a 'puff' piece about POWER6:

http://news.zdnet.com/2100-9584_22-6124451.html

The 14.999999 they mention would only occur if you used binary
floating-point.  If you use all fixed-point decimal operands for
calculating either 3 x 5, or as the article mentions, .1 x $1.50, you
would get exactly 15 (or 15 cents).

Currently, POWER chips can do fixed-point arithmetic only in binary,
which requires some elaborate machine code to do decimal arithmetic.
Here's an example of the iSeries machine code generated in an MI pgm for
adding two 9-digit packed fields (42 machine instructions!):

http://groups.google.com/group/comp.arch/msg/f7b8a64c7e85591e

The wikipedia page:

http://en.wikipedia.org/wiki/POWER6

links to some descriptions of the new DFPU (Decimal Floating Point Unit)
in POWER6.

Sounds like it will do floating point operations directly on decimal
data, to eliminate the 14.999999 approximations caused by conversions to
and from binary.  It should also greatly reduce the number of machine
instructions generated.

--Dave