[libre-riscv-dev] TLB

Luke Kenneth Casson Leighton lkcl at lkcl.net
Mon Apr 22 05:01:19 BST 2019


On Mon, Apr 22, 2019 at 3:50 AM Jacob Lifshay <programmerjake at gmail.com> wrote:

> I'd advocate using random selection as the replacement protocol,
> rather than LRU, since random doesn't require any per-entry state or
> complex state updates, and has essentially the same performance for
> more than a few ways per set. I'd advocate using a LFSR that advances
> continuously and picking the bottom few bits from it to select which
> cache line to replace.

 ... because the PLRU relies on top bits, and you need quite a lot of
them to stop overloading of one of the sets, i am guessing.  created
http://bugs.libre-riscv.org/show_bug.cgi?id=71 jacob can you quickly
write a LFSR module, drop it in that directory in soc.git?

l.



More information about the libre-riscv-dev mailing list