🚀 go-pugleaf

>  For the 68030,  the utility is this : you present a paragraph address
>(bottom 7 bits zero).  Then you clock four times, loading four thirty-two
>bit words into cache.  You have just saved most of the work involved in
>your next three memory cycles.  As far as memory access time goes, lots of
>ripple mode devices can clock out ripple bits at better than 62.5 ns/bit.
>  Somebody who has a real live data sheet, is this right?

The 'nibble' mode devices (thats the buzz-word for the above type of
transfer) that I've seen *don't* need the bottom n-bits zero-ed. They
will accept any address as the first, and give you the next 3
addresses (modulo-4) next. This is even better, as it allows you to
read the address that you are really interested in first, and do the
prefetches last. 

-- 
Wolfgang Rupprecht	{harvard|decvax!cca|mit-eddie}!lmi-angel!wsr