>
> Measured with VICE (empty disk):
>
> - $F4ED-$F4F0 (GCR -> binary): 22.890 cycles
> - $F586-$F589 (binary -> GCR): 22.693 cycles
>
> Of course, it might change according to the actual data, but you have
> rough estimates of the typically needed time.
>
> And: Yes, the GCR routines in the 1541 *are* slow.
>
> The 1571 has different routines, cf. $9606-$970C (read) or $976e-$97f6
> (write). The do the GCR <-> binary conversion "on the fly" with the  
> help
> of the (big) tables at $970D - $A40C.

Yes, exactly. That reminds me that I started also with "procedural"  
approach, which I initially optimised but then I switched to the  
"lookup table" approach derived from the 1571. I had the comfort of  
having enough RAM in the 64 to hold buffers, code and lookup tables.  
The "procedural" approach was substantially slower. Now - reverse  
engineering my revision hitory: 21.3 + 1.3 + 3 + 5 = 30.6s using the  
routines like those from 1541, which is probably somewhat more than  
double the decoding time of the latest version.

-- 
SD!


       Message was sent through the cbm-hackers mailing list