Re: earmv4 packages coming soon

[David Brownlee <abs%absd.org@localhost>]

On Sat, 14 May 2022 at 15:55, John Klos <john%ziaspace.com@localhost> wrote:
>
> Hi, all,
>
> I've never been quite sure about the differences between earm (earmv5) and
> earmv4, but since earmv5 packages won't run on earmv4, I've started a bulk
> build of earmv4 packages.
>
> Which makes me wonder: is there any downside to running earmv4 packages on
> earmv5? Does earmv5 make use of features that significantly improve
> performance over the same earmv4 code?
>
> If there is no significant improvement to running earmv5 specific
> packages, then I might just build earmv4 instead.
>
> Also, not that this matters much, but build.sh with -m evbarm -a earmv4
> doesn't work, although -m shark does (and I'm sure -m hpcarm and friends
> work, too).

IIRC it was StrongARM (v4) vs XScale (v5), I think there were a fair
number of Thumb additions... but I think the main difference is v5
added VFPv2 which is handy for floating point.

evbarm already has an ld.so.conf:
libc.so.12  machdep.fpu_present     1:libc_vfp.so.0,libc.so.12

I wonder if it makes sense to have some form of libc_vfp2 to allow v5t
CPUs to run a v4 build without (most of the) floating point penalty...

David


Now... I don't know if it's possible to build an armv4 userland with
an optional libm