Re: whole number constants

"Richard Maine" <nospam@[EMAIL PROTECTED]
> wrote in message
  news:1ikghk4.1qenezmuqlo5sN%nospam@[EMAIL PROTECTED]
>>robin <robin_v@[EMAIL PROTECTED]
> wrote:

>>A single precision value in storage is converted to double precision
>>by appending (extending with) binary zeros.

>Nothing in the Fortran standard specifies that.

The IBM 7030:   http://en.wikipedia.org/wiki/IBM_7030

had the ability to ****ft in either zeros or ones during post
normalization.  (I don't know if it applied in single to double
precision conversion.)

As with conversion, it is usual to ****ft in zeros.
(and as usual, the standard doesn't require it.)

The idea for the 7030 was that you would run a program twice,
once with zeros once with ones, and compare the results.
That would give you some idea on how accurate the results
were, as a common cause of significance loss is cancellation
in subtraction.  I haven't heard of any later machines
doing something like this, though.

-- glen