[Libre-soc-dev] daily kan-ban update 14jul2022

Luke Kenneth Casson Leighton lkcl at lkcl.net
Thu Jul 14 12:26:30 BST 2022

On Thu, Jul 14, 2022 at 10:33 AM Jacob Lifshay via Libre-soc-dev
<libre-soc-dev at lists.libre-soc.org> wrote:
> While I was thinking about how isync isn't suitable for a memory fence, I
> noticed we have no tests that it operates correctly to allow executing
> modified instructions (e.g. JIT compilers), so I wrote one. Feel free to
> move to a more appropriate location.

no it's great.

i cross-linked to bug 236.  basically from what paul was saying yesterday,
there's been one hell of a lot of research here (by IBM) surrounding massive
SMP multi-core systems and distributed systems

basically this one is going to take a *lot* of justification and careful
discussion, and some in-depth exploration of what it is precisely
and exactly IBM has done, so that it can be clearly explained to them -
in terms of what *they* have successfully done - why what they have
done may not be appropriate for our use-case.

that is going to take time.

i am inclined to suggest that all of these things be documented,
preparing for such a discussion, and then move on.

demonstrations like this, especially if it can be shown how performance
completely sucks (nowhere near like the number of atomic operations
needed per second is achieved) and/or degrades significantly with
the number of cores, would be perfect.

aside from anything, it shows a starting point where IBM experts
may be able to point out, quite simply, "you're doing it wrong".

in looking at this:

and notes from paul yesterday:
3.1 p1077 eh hint

i added those to the wiki page past night.


More information about the Libre-soc-dev mailing list