Re: information you aren't addressing is that in the 16 way cache, 8 times as many addresses map into a particular cache index. Most of the remaining 14/16 of the cache set will be filled with addresses that are also resident in the 2 way cache (in different indexes.)
Hi Scumbria,
Thanks for the post. I'm just about there. I'm still having trouble because I understood (perhaps incorrectly) that you have said that a victim cache behind a 16-way is of little use, while a victim cache behind a 2-way can be very useful (that's OK, so far), but also that an independent 16-way behind a 2-way is not useful. I'm afraid I don't get that, since it seems to me that the two should be mutually exclusive: either the 16-way maintains most of the useful addresses that would have been picked up in the victim cache, in which case it can substitute for a victim cache, or it doesn't, in which case a victim cache behind a 16-way would be nearly as useful as one behind a 2-way.
If the code you mentioned is conveniently accessible, I'd be interested in seeing it - otherwise, please don't put yourself to any trouble - there's a good chance I wouldn't understand its implications, anyway.
Also, if this is getting to be too much, we can just drop it - I'm just curious, and thinking back to the competition between motherboards with different caches in the old 386 days. (Remember the Everex step?)
Regards,
Dan |
