> This is interesting. How fast is the I2O controller? Would it be
> possible to do some useful things on the i960 other than controlling
> hardware? From memory the i960 supports time-slicing in the middle of
> instructions so that a long instruction such as the floating point modulo
> operation (which may take up to 75,000 clock cycles to complete) can
75000 sounds excessive to me.
The I960 family has different models, some of which don't have a FP
unit. For an IO processor I'd say I don't need an FP unit. I'd choose
things like "embedded, low interrupt latency, super scalar". The
"I960CA" fits the bill nicely, but its too old to have a built-in PCI
bus. Moreover they now have these beasts at a higher clock rate. (CA:
Max 33MHz, nowadays at least 75Mhz....)
Low interrupt latency? Yes! The I960CA can be executing instructions
from your interrupt routine (NMI) about 100ns after the interrupt line
goes active.
longest instruction on the I960 CA are the divide with about 35 cycles,
and the multiply at 4 cylces. Everything else is 1 or 2. No floating point
though.
> complete even in the presense of rapid interrupt activity. This should
> allow a system to run a RT Linux kernel on the i960 which would do fun
> things like plotting the Mandelbrot set at background priority while
> controlling IO.
Your pentium is faster at doing that sort of stuff. Did you have a
Comodore 64? You could do similar things on the processor in the
disk drive. Calculating Mandelbrots was an ideal application because
the link between the central CPU and the drive was kind of slow
(9600 baud as supplied from Comodore).
Fun for hackers, but not worth putting serious work into....
Roger.