On Wed, 21 Jul 1999, David A. Gatwood wrote:

> On Wed, 21 Jul 1999, Bill Studenmund wrote:
> 
> > I don't think so, unfortunatly. As I remember it, the IOP's don't do DMA,
> > they off-load the I/O. The difference is that the CPU writes a whole
> > buffer to the IOP, and then the IOP does its thing. Or the IOP says, "I
> > have data," and the CPU reads a whole block at a time. It's not DMA, but
> > it's much more efficient. :-)
> 
> Sounds an awful lot like DMA, just using a different physical address for
> the buffer and maybe having to program the IOP before you send it a start
> command.  Is it mostly just an issue of terminology or is there a
> substantial difference there that I'm missing?

As I understand it (which might be wrong), it's like there is either a
FIFO or a shared memory space (like an Ehternet card's memory) into which
the CPU copies the data, from which the IOP reads it.

If it were DMA, the CPU'd just pass the IOP the in-host-memory address of
the data and go from there. There'd be no CPU_copying step. :-)

Take care,

Bill