On 6/19/26 5:30 PM, Jeffrey Law wrote:
> Either works for me.  If we go with the backport, then I'd suggest 
> letting it simmer on the trunk for a week or so.  While I don't expect 
> problems, the soak time never hurts.  Duplication is safer, but 
> obviously would require more maintenance if we find something needs to 
> change in the ascalon bits.
> 
> Ultimately, your call.  I'll support either approach.

The duplicated entries both use the same tt_ascalon_d8 and 
tt_ascalon_d8_tune_info
data, which is where the bulk of the Ascalon code lives, so the only extra
maintenance would be if we needed to change the extensions in the RISCV_CORE()
entries and I think those are fairly solid.  Therefore, I think I'm leaning
towards just using the duplicated entry patch to be super safe.  I'm currently
testing that now.

I didn't change any of the tt_ascalon_d8 usage in the tt-ascalin-d8.md file,
since that would be a huge mechanical patch for little gain.

That said, I think we are missing some of the Ascalon changes that went into
GCC 16 in the GCC 15 release branch, so I have some homework to see what else
we may want/need to backport to GCC 15.

Peter

Reply via email to