[Libre-soc-dev] gigabit router design decisions

Luke Kenneth Casson Leighton lkcl at lkcl.net
Thu Nov 4 10:31:14 GMT 2021


On Thu, Nov 4, 2021 at 1:39 AM Jacob Lifshay <programmerjake at gmail.com> wrote:

> branch prediction doesn't require speculative execution:

of course it does!  you have to fetch the instructions ahead, and
you have to execute the instructions ahead... then cancel both.

that implies that cancellation infrastructure has to be added right
the way through the entire design.

> fetching and decoding instructions ahead *does not* require executing
> instructions -- branches can always stall execution until the branch
> condition is calculated, at which point we can either immediately execute
> the instructions that the branch prediction let us already have fetched and
> decoded, or we tell the fetch pipe it was wrong and wait for it to fetch
> and decode the correct instructions before executing them.

exactly!  and that is - itself - additional complexity that we DO NOT HAVE
TIME TO ADD.

yet you continue to argue for something that is additional complexity
when i have told you MULTIPLE TIMES that we are UNDER TIME PRESSURE

for god's sake please get a grip.

there are one hundred and fifty days including weekends in which to
complete the work and meet the deadline.

are you aware that of those one hundred and fifty days, you have
LITERALLY and already spent THREE PERCENT of those days wasting
time and arguing instead of accepting reality and saying, "shit, this is
huge time pressure, i should accept this and start focussing and
bloody well getting on with it"

does that shock you enough yet to appreciate how serious this is?

l.



More information about the Libre-soc-dev mailing list