On 4/28/20 10:38 AM, Jakub Jelinek wrote:
On Tue, Apr 28, 2020 at 04:23:24PM +0200, Andreas Krebbel wrote:
Our ABI doesn't specify anything regarding C++ so there is nothing in there
which really conflicts
with that. I assume these things will be part of a cross-platform C++ ABI
instead? I don't see a way
to add this to our platform ABI without introducing C++ in general there.
Since "no_unique_address" is a new feature we want to pick whatever is most
efficient and matches
what other archs do. Passing F, G, J, K in FPRs looks reasonable to me.
Ok, will tweak the patch then once the powerpc+generic one is finalized.
Given that this is something which hasn't been covered by the ABI so far I'm
not sure we really need
a -Wpsabi warning for that.
But [[no_unique_address]] has been introduced already in GCC 9, so e.g. in
the powerpc64le patch I want to warn users about the ABI incompatibility
when padding those.
Your call though.
Though C++20 support is still highly experimental, we're not trying to
maintain compatibility.
Btw. is the no_unique_address flag visible also in Dwarf? Probably it is enough
for GDB to just
observe the effect of it to decide how to pass arguments?!
I agree that's probably enough.
Jason
