[yocto] armhf support for meta-raspberrypi

Andrei Gherzan andrei at gherzan.ro
Sun Jan 25 14:17:58 PST 2015


Hello,

On Mon, Jan 19, 2015 at 2:22 AM, Philipp Wagner <lists at philipp-wagner.com>
wrote:

> Am 18.01.2015 um 23:07 schrieb Andrea Benini:
> > Thanks folks, and thank you Andrei.
> > As suggested by toolmmy and pleasant folks in this thread I'm using
> > DEFAULTTUNE option. I had already seen the arm1176jzfshf after a small
> > search into the meta-raspberry layer but I didn't know how to use it.
> > Now I have recompiled my toolchain with this option enabled and it
> > worked fine, I need to compile other tools (C, C++ libs, toolchain,
> > binutils) for my own test and I'll put it on a real Pi. It takes a while
> > but after this full recompile it seems everything is fine.
> > If someone is interested in some benchmarking or other tests please let
> > me know, my current image works. I'd like to add other packages useful
> > for a real distro (Python 2/3, mc, nano, gcc, ext utils, ...) as I did
> > before with the armel build but seems to be working without troubles so
> > I'll normally continue my job with yocto and the Pi. My armel image
> > works on Pi model A, B (256 and 512) and B+, as soon as I'll have a
> > complete armhf image I'll do complete tests on them as well.
>
> I did a comparison between softfp and hardfp a while back (i.e. using
> hard float both times, difference is just in the calling convention)
> using the standard web benchmarks Octane and sunspider with a full RPI
> image and Firefox built with the corresponding compiler options. The
> results showed no significant difference between the two, leading me to
> the conclusion that the hardfp trouble (breaking the ABI) is not worth
> the effort over softfp.
>

This is because the compile is already using the optimized instructions. So
this change is actually redundant and won't improve the performance in any
way. So, the current setup seems to be the one to use.

-- 
*Andrei Gherzan*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.yoctoproject.org/pipermail/yocto/attachments/20150126/3d73e8c4/attachment.html>


More information about the yocto mailing list