Re: Toshiba AC100
On Saturday 09 October 2010 00:21:04 Wookey wrote:
> > My guess is that in the case of the FP variants
> > there are only relatively few packages that are really fp-intensive, so
> > a scheme that allows them to have optimised versions distributed
> > separately might be best, and have the core architecture just cover the
> > "ARMv7" aspect.
>
> Absolutely - and this is on my list of things to get linaro/ubuntu to
> spend some time on, although I've been told nothing much is going to
> happen on that front until multiarch is actually done, so it'll be a
> while yet before there is much in the way of practical improvements.
> It should start to become clear over the next few months how many
> variants are actually worth supporting. That'll be something at least.
In the meantime he could try the armhf port on debian-ports.org, which targets
exactly those base requirements: armv7-a, hardfloat ABI (vs softfp), vfpv3-d16
and thumb2. No NEON is required by default so it should run quite fine on the
tegra2. Instead of providing just a bunch of optimized packages, every binary
was rebuilt using the new abi and base specs. Performance has been found
10-40% faster than equivalent softfp code, in some rare cases 200% gain has
been observed.
There is no installer yet but I'll upload an updated tarball on
http://freevec.org/repository/images/
(There are already a couple there, but they're a bit outdated and I intend to
provide a new version during the weekend). I'll post the url here on Sunday.
IMHO, starting a new port is a much faster process than waiting for multiarch
-it took me just a month to build 60% of the debian archive, even with lots of
problems and it keeps on compiling.
My 2c.
Regards
Konstantinos
Reply to: