https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115325
Robin Dapp <rdapp at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |kito.cheng at gmail dot com
--- Comment #7 from Robin Dapp <rdapp at gcc dot gnu.org> ---
(In reply to Jan Wassenberg from comment #6)
> hm, we are a library that allows users to either affix HWY_ATTR to each of
> their functions (which expands to __attribute__((target("arch=+v"))), or
> instead to use a pragma to annotate all functions within a code region,
> typically most of the TU.
>
> I'm surprised there is a difference between pragma and per-function
> attribute, doesn't the pragma basically attach an attribute to every
> function?
>
> Clang seems to do it like that:
> `#pragma clang attribute push(__attribute__((target("+v"))), apply_to =
> function)`
>
> Unfortunately I don't think it will fly to require the HWY_ATTR on every
> function, there could be many of them, and I doubt many users will make that
> change in order to support RVV.
It's not that big of a difference, conceptually, yeah. It's just that (I
think) nobody has bothered so far to implement it.
Kito are you aware of any work regarding the pragma? Should we still support
it in this cycle? I don't think it's too much work. After all we have a bug
filed here, it's not a regression but... :)
