Re: [hatari-devel] 030 cache & prefetch

[ Thread Index | Date Index | More Archives ]

Yes there may not be any such prefetch - the behaviour I saw in Hatari looks close to what I would expect *if* instructions must be fetched from memory ahead of the PC (since they must also then move via the cache and cause invalidations).

However I can't say / don't know that the 68030 actually does this at all (68000 and 68040/60 do fetch beyond execution in different forms). The 'bug' case actually looks more likely at this point.


I might be (and possibly I am) terribly wrong here but about what kind of prefetch are we talking about? Because to my knowledge, there's only one kind of "prefetch" happening and that is in case of:

$1002: dbra d0,loop
$1004: add.l d1,d1   ; <-- this is part of the i-cache as soon as first 'dbra' reaches PC because 'dbra' is not aligned on long boundary

I'm not aware of any othe 'ahead' prefetch at all.

MiKRO / Mystic Bytes

Mail converted by MHonArc 2.6.19+