On 11 Aug 2023, at 12:51, Stuart Henderson <[email protected]> wrote:
> 
> yes, for amd64 it's intel Tiger Lake (11th gen) and newer, models of the
> format iX-11XXX, iX-12XXX, iX-13XXX. no AMD CPUs. On ark.intel.com, look
> for "Control-Flow Enforcement Technology" (the intel term which covers
> both shadow stacks and indirect branch tracking).
> 
> for arm64 currently only on Apple M2.

Thanks for all the pointers! Every machine for 100 miles around me seems
to be an AMD these days, so I've ordered myself a Raptor Lake NUC
to be my new OpenBSD desktop. As soon as that arrives I'll take a look
at the native code compiler output (unless someone else beats me to it).

In the meanwhile I think we should go for updating the OCaml ports
to 5.1.0 so that a patch against trunk will be manageable...

The only roadbump is from Daniel Dickman, who wrote:

> Fine with me so long as compcert continues to work on i386. I think the
> only impact is potentially running slower? That’s fine with me.
> 
> That being said I think there were some challenges with newer coq on
> i386. So maybe the end is close for compcert on that platform?

Indeed, I think it's time to move to a 64-bit license after so many years.
I strongly suspect you're the only user of OpenBSD/i386 compcert
these days, so it's worth just asking for an upgrade of the license
to shift to 64-bit.  OCaml 5 will definitely be bytecode-only for 32-bit
architectures for the foreseeable future. This in turn pushes up memory
requirements (due to the extra boxing of values), which makes it more
likely that Compcert/Coq will hit the 4GB process limit.

Anil

Reply via email to