[Libre-soc-dev] daily kan-ban update 10nov2020
Luke Kenneth Casson Leighton
lkcl at lkcl.net
Fri Nov 13 21:11:07 GMT 2020
On 11/13/20, Jacob Lifshay <programmerjake at gmail.com> wrote:
> one issue: the fetch pipeline is more complex than you might think, it
> properly handles stopping fetching additional instructions when
> encountering a taken branch since fetch pipelines usually can only fetch
> consecutive instructions.
for a first iteration i believe it would be reasonable to assume that
this is a loop and it is in a single L1 cache line.
more complex effects can be added as later investigations, because,
yes, we need to know their effect on performance.
microwatt's strategy is to fire off extra 64bit WB consecutive reads,
filling L1 cache lines slightly ahead of actual execution.
jumps backwards for short loops have a high probability they are still
in L1 ICache.
More information about the Libre-soc-dev