[Arm-netbook] [review] SoC proposal

[lkcl luke]

btw, i think it's also worthwhile pointing out that in most
"proprietary software" cases, from SoC manufacturers, they throw you a
proprietary library that operates with a proprietary hardware block
with proprietary instructions and don't give you jack shit.

tensilica on the other hand: yes they have proprietary libraries, but
a) we haven't even discussed the possibility of whether they would
like to release the source code as free software b) the software is
written on top of a RISC core which is entirely open and has a
publicly-available free software toolchain.

if they didn't publish the instruction set and have the gcc compiler i
sure as shit wouldn't have picked them!  i would have picked LEON or
some-such (which is even available under LGPL licenses!).  but i have
a good feeling about the XTensa RISC core and its numerous extensions,
whereas all the other RISC cores i've looked at i just don't think
they have the flexibility or the muscle when ramped up to over 1.2ghz
in 28nm.

i mean - the flexibility is just incredible.  i'm not sure how much
people have been reviewing the tensilica web site, but some time in
2004 or 2005 they were contacted about the results of a benchmark test
that an independent reviewer was doing.  he said that their
performance was about comparable to all the other RISC cores.
tensilica asked him to send them the benchmark applications, and that
they'd have a look at it.  within ONE HOUR they had the
"dial-a-RISC-core" options twiddled on their CPU generator application
(which the reviewer didn't have the experience to understand properly)
and they got i think it was something mad like a SEVENTEEN-fold
improvement.

what i don't quite understand is how the bloody hell i missed these
guys in my search last year for RISC cores.  they _just_ don't show up
on any radars.  but then, they're so successful at what they do, in
the markets they're in, that they don't _have_ to show up on anyone's
radars!

l.