RE: floating point and Sun T2000

  • From: "Christian Antognini" <Christian.Antognini@xxxxxxxxxxxx>
  • To: <rjamya@xxxxxxxxx>
  • Date: Fri, 5 May 2006 19:33:55 +0200

Ray

>and I thought CBO computations included fractions ... isn't that
>floating point arithmatic?

Of course there are lot of floating point computations. But, as I understand, 
the OP was interested in floating point computations that are performed 
directly by the CPU (otherwise, IMHO, the question makes no sense...). 

Since the internal format of the NUMBER datatype (and all others based on it) 
is oracle specific, either:
1) they implemented it without taking advantage of the capabilities of the CPU, 
i.e. they have a library that is platform independent doing such computations
or
2) they convert the NUMBER values to CPU-dependent floating point numbers 
(probably IEEE 754 conform) before sending them to the CPU

As of 10g through BINARY_FLOAT and BINARY_DOUBLE they support IEEE 754 numbers. 
With them is highly probable that the floating point computations are directly 
performed by the CPU. For this reason, in 10g, computations done on 
BINARY_FLOAT and BINARY_DOUBLE values are much faster than the same 
computations done on NUMBER values.


Regards,
Chris
--
//www.freelists.org/webpage/oracle-l


Other related posts: