> On Jun 23, 2016, at 11:17 AM, Chuck Guzis <ccl...@sydex.com> wrote:
>
> ...
> Of course, there were also machines that used the floating point
> facility for all arithmetic. Integer computations is performed as a
> subset of floating-point. This has the ramification that an integer
> does not occupy an entire word, but only part of it.
The CDC 6000 did that in part. It has full 60 bit integer add/subtract, but
multiply and divide are done using the floating point operations so they work
only for numbers up to 47 bits.
The Electrologica X8 is yet another take on this. There, the mantissa is
viewed as an integer, and the normalization rule is to make the exponent as
close to zero as possible without losing bits. The consequence is that all
integral values under 2**40 are represented as exponent zero and the mantissa
equal to the number, which amounts to simply the integer representation of that
number. This makes conversion from float to integer rather easy (and of
course, conversion in the other direction takes no code at all).
paul
