Now ready !
After a short while and few positive feedbacks from you I’m proud to announce an important upgrade for Moebius: ARM HF Support is back again. In previous 1.x versions HF support was an easy addon because I derived my toolchain and initial build from a Debian base (Raspbian), now with new 2.x versions I have a full distro rebuilt from scratch.
I’m still using .deb files because dpkg is a strong package manager but I’m not based on Debian (nor Raspbian) anymore and this is a totally new distro based on package sources (and patches as well…). I have always thought that ArmHF should be just an easy compile flag to put somewhere. Unfortunately it’s not just an easy flag on complex packages (python, gcc, kernel-modules, …) so I dedicated some time to it. Now packages are still terminating with suffix: “*armel.deb” but are armhf compliant and the github repository fully is updated. Available github repository packages have been rebuilt with the new support and I’m adding previous files again, in a short while I’ll restore original situation.
How do I know if the distro/package/lib is armhf ready ?
If you use readelf command (gcc/binutils stuff) to get binary information from an elf you’ll see:
root@moebius:~# readelf -A /proc/self/exe |grep Tag_ABI_VFP_args
Tag_ABI_VFP_args: VFP registers
If the above line is present then ARMHF support is enabled, if you don’t see any output then you don’t have armhf (hard float) and you’re using armel (soft float), I have read wrong things on few web pages about the .deb package name. I can assign my own favorite name to a .deb file so having a package with the trailing “*armel.deb” part doesn’t necessary mean it’s compiled with soft float. readelf is what you really need to check an executable, this is a general suggestion and it’s valid for whatever distro you use.
Why Hard Float is so important for Moebius ?
Performance is the key, using the internal floating point unit doesn’t affect speed on common/daily operations but you may see huge benefits if you’re using the raspi camera, a FFT algorithm or whatever software that requires numbers crunching. Take a look at here
, and here
, and here
to understand details.
Now with locales and hard float support workload will be focused on package deployment and containers
Obviously you can get the new image from the download section