Re: do people think openjdk17 builds on aarch64 without heroic resources?

On Mon, Sep 23, 2024 at 02:06:17AM +0200, Tobias Nygren wrote:
> On Sun, 22 Sep 2024 08:47:29 -0400
> Greg Troxel <[email protected]> wrote:
> 
> > I do see
> > 
> >   => Broken FPU detected: applying floating point workarounds
> > 
> > which I don't understand, but that seems not troubling so far.
> 
> It comes from hacks.mk.
> Some parts of the java runtime need full IEEE754 semantics for
> denormal numbers in order to generate correct code. NetBSD by
> default uses flush-to-zero mode, and disabling that comes with
> performance pentalty. See the comment in hack.mk for details.

Can we add an ATF test to demonstrate this issue?

Does it happen on all neon FPUs? I see a lot of cpu feature flags related
to that, is it depending on some of them?

Martin