Hi,

I see that the LM32 can have an 8kiB i-cache and an 8kiB d-cache.

I have not started looking at the specifics of the LM32 implementation
for MilkyMist, but if cache was implemented then a careful choice of
MMU and cache interaction will be needed, it seems.

The reason I say this is that Marvell implemented a VIVT cache design
on their 78x00 chips, and it turned out to be problematic. The
performance of that chip was no where as good as it could have been
with a better cache design, say, perhaps a VIPT cache, or perhaps even
a PI indexed cache (but that means you cannot begin memory accesses in
the cache until you have done virtual to physical translation, it
seems to me).

-- 
Regards,
Richard Sharpe
_______________________________________________
http://lists.milkymist.org/listinfo.cgi/devel-milkymist.org
IRC: #milkymist@Freenode
Twitter: www.twitter.com/milkymistvj
Ideas? http://milkymist.uservoice.com

Reply via email to