Re: [PATCH] ARM: ep93xx: remove MaverickCrunch support
From: Martin Guy
Date: Mon Aug 02 2021 - 11:45:23 EST
On 02/08/2021, Arnd Bergmann <arnd@xxxxxxxxxx> wrote:
> The MaverickCrunch support for ep93xx never made it into glibc
This is true, but is not needed except for preserving FP registers in
catch/throw exeptions or longjmps.
> was removed from gcc in its 4.8 release in 2012.
I was sad about that, as I had managed to get it working correctly in
4.2 4.3 and 4.4.
Unfortunately the GCC ARM maintainer at the time was paid by ARM, and
they had no interest in it, as I gather the ARM-Cirrus partnership
ended up disappointingly.
> It is now one of
> the last parts of arch/arm/ that fails to build with the clang
> integrated assembler
Even if crunch support is not configured?
> According to Hartley Sweeten:
>
> "Martin Guy did a lot of work trying to get the maverick crunch working
> but I was never able to successfully use it for anything.
> It "kind" of works but depending on the EP93xx silicon revision
It works fine on all production revisions except D0 (the first one)
which is so rare as to be unfindable. All the others from D1 on have
the exact same HW bugs.
> there are still a number of hardware bugs that either give imprecise or garbage results.
Not my experience, See http://martinwguy.net/crunch/#CorrectnessTests
If "imprecise" means "doesn't handle unnormalized FP values correctly"
then yes, that's a feature if you need the 2.5 to 4gimes speedup it
provides.
But I would be interested (Hartley) if you have a concrete example.
> This touches mostly the ep93xx platform,
The ep93xx is the only chip series it was ever included in.
> If there are remaining users of MaverickCrunch, they can use LTS
> kernels for at least another five years before kernel support ends.
The only user ! am aware of is the HEAT deep space telescope in the Antarctic
http://martinwguy.blogspot.com/2015/10/maverickcrunch-code-runs-on-heat-deep.html
Personally I have no problem with removing kernel support if that
makes people's lives easier in some way.
M
M