Re: Fast GCR decoding?
Date: 2005-11-09 00:46:06

On 2005-11-08, at 10:33, MagerValp wrote:

>>>>>> "s" == silverdr  <> writes:
> s> Do you mean you were able to split the GCR nybbles in RT? What do
> s> you mean by "partially" though? Could you elaborate a bit on the
> s> method and results, please?
> No, there aren't enough cycles to decode to nybbles in realtime,

I don't remember how many cycles there were between the GCR bytes in  
the worst case but that's what I expected and would be surprised if  
you could do it as that would be already close to a speeder nirvana ;-)

> thus
> only partially decoded.

OK. Got it.

> I don't have my notes here, but it was
> something like this:

I see. Thank you.
> 3 nybbles just require ANDing, 3 just shifting, and only 2 nybbles
> require several operators.

And you end up using the regular GCR nybble decoding tables as found  
in the 1541's ROM but you also wrote it was faster than the 1571  
decoding routines (I think I found them already in the 1571 ROM but  
didn't yet have time to analyse)? Or was it so that the 1571 alike  
routines would be faster but wouldn't fit in the RAM and the best  
[out of what could still fit] was to do it this way?

> I implemented the full decode as a 2nd pass
> that merges the nybbles into bytes, but for the next version I'll see
> if I can merge the decoding with the transfer loop.

Actually, while gathering the info during yesterday's research I  
heard that it's been done (or something very similar, judging by the  
info) and published in one of our old paper magazines here. May try  
to find a copy of that one.

Heck is a place for people who don't believe in gosh!

       Message was sent through the cbm-hackers mailing list

Archive generated by hypermail pre-2.1.8.