[Coco] Tcp off-loading on CoCo

[L. Curtis Boyle]

The cart interrupt might be a better option (it can be IRQ, not just FIRQ, right?), as you can shut that on or off at will. So, if a routine knew that polling would be better for it’s situation, it could shut the IRQ off, and poll. If it preferred an IRQ (to block read a bunch of data after completed packets are in - read it like a disk drive sector), it could do that as well.

L. Curtis Boyle
curtisboyle at sasktel.net



> On Apr 13, 2017, at 10:25 AM, Zippster <zippster278 at gmail.com> wrote:
> 
> Yep.  I’m referring to the cycles consumed saving the entire machine state.
> If you’re getting a lot of these interrupts, that can add up to a lot of cycles.
> 
> Obviously, if you aren’t pushing the limit with clock cycles, it’s not really an issue,
> but it can be if cycles are tight, where polling might be a better option.  Streaming
> audio for instance.
> 
> - Ed
> 
> 
>> On Apr 13, 2017, at 11:17 AM, William Astle <lost at l-w.ca> wrote:
>> 
>> On 2017-04-13 09:57 AM, Lee Patterson wrote:
>>> NMI = non maskable interrupt? Why is it slow?
>> 
>> I'm wondering the same thing.
>> 
>> The NMI is serviced like the IRQ so the full processor state is saved. That will, obviously, make the minimum service time higher than for FIRQ. However, in the case of FIRQ, you're going to have to explicitly save any registers you clobber on the way in and out of the routine which will affect how much of a benefit you get from using it.
>> 
>> Basically, the NMI isn't any slower than the IRQ once it triggers. As I understand it, NMI is edge triggered while  FIRQ and IRQ are level triggered.
>> 
>> 
>> -- 
>> Coco mailing list
>> Coco at maltedmedia.com
>> https://pairlist5.pair.net/mailman/listinfo/coco
> 
> 
> -- 
> Coco mailing list
> Coco at maltedmedia.com
> https://pairlist5.pair.net/mailman/listinfo/coco
>