[J-core] Linux on J1?
Rob Landley
rob at landley.net
Tue Aug 2 16:14:02 EDT 2016
On 08/02/2016 01:59 PM, D. Jeff Dionne wrote:
> On Aug 2, 2016, at 11:06, Rich Felker <dalias at libc.org> wrote:
>>
>> Disabling dcache is a huge hit to performance. It might be more practical to do a
>> shared-dcache implementation for SMP, where both cores are accessing
the same cache
>> (even if subject to stalls when both access it at the same time) that
would allow
>> all the snooping logic (and half of the cache memory size) to be omitted.
>
> Also, combined I and D cache. There is an instruction prefetcher in the repo that
> shows just how far you can go without cache, and it's surprisingly far...
But I thought Niishii-san implemented cmpxchg using some of the existing
cache line invalidation logic?
> Which reminds me, something to keep on the list: J2 should prefetch the next cache
> line assuming straight line execution, it doesn't right now and takes the cache miss
> each line on long runs not already in cache...
If we could get that into the third open source release with the 62.5
mhz speed increase, that would be really nice. :)
Actually something we could propose for an LCA talk (cfp is open through
the 5th) is "Optimizing a new processor architecture". Talking about all
the things we've done to speed up linux, gcc, musl, and the VHDL itself
since you first got Linux booted on the thing ~3 years ago.
(And also why we selected this architecture instead of i386/sparc/m68k
which are just as old and thus out of patent, which we've touched on a
bit and you explained to me on the phone once, but comparative
architecture analysis from a performance perspective would be nice to
get out in front of an audience. And we could talk about scaling the
design up and down, prefetch vs cache from a footprint perspective, why
we're scaling in BOTH directions down into j1 and up into j64.)
I mean _yes_ we chose an architecture on which the patents had expired,
but there's a dozen of those. Why _this_ one, and then how did we build
on that base? I think there's a good talk there...
Rob
More information about the J-core
mailing list