18 Jul
2014
18 Jul
'14
1:56 a.m.
On 7/17/2014 4:57 PM, Mark Linimon wrote:
Most of the DEC 36-bit machines use the normal integer ALU to calculate
the floating-point mantissa. They add a little 10-bit ALU on the side
called the "SCAD" to calculate the floating-point exponent.
For example, floating-point normalization requires shifting the mantissa
while incrementing (or decrementing) the exponent.
The VAX 11/780 (without the floating-point accelerator) is similar but
calls the 10-bit exponent ALU the "EALU".
There really isn't much additional circuitry to implement hardware
floating-point.
Of course if you want /fast/ floating-point, you can add as much
hardware as you can afford...
Rob.
On Wed, Jul 16, 2014 at 05:13:24PM -0600, Kevin Keith
wrote:
> There are
a few open FPU cores floating around, although adapting
> them to use VAX floating point format might be rather difficult.
In the early days of microprocessors, FPUs were a large enough chunk
of circuitry that sometimes it was only feasible to have it as a
coprocessor -- and fallback to software emulation if it wasn't
found.