> On 6 Jan 2026, at 17:43, Alice Carlotti <[email protected]> wrote:
>
> On Tue, Jan 06, 2026 at 03:12:51PM +0000, Iain Sandoe wrote:
>> Hi Alice,
>>
>>> On 5 Jan 2026, at 15:39, Alice Carlotti <[email protected]> wrote:
>>>
>>> This series has three parts. The first three patches fix issues arising
>>> from
>>> the original assumption that SME implied SVE2. Patch 4 can then remove the
>>> "sorry" we put in place last year for "+sme+nosve" configurations. The
>>> remaining patches adapt existing tests (either in place or by creating
>>> duplicate copies) to provide good test coverage for +sme+nosve
>>> configurations.
>>>
>>> Apart from the addition of __arm_get_current_vg to libgcc, this series
>>> should
>>> have no functional change for any feature combinations that are currently
>>> supported on trunk.
>>
>> thanks for adding me to the cc.
>>
>> I merged this with my current Darwin branch and built / tested it - the
>> tests are
>> somewhat inconclusive at the moment, since there are hundreds of function
>> body check fails - which (probably) just mean I need to tweak the tests to
>> cater
>> for the code-gen differences (but of course could be more sinister).
>> Tweaking
>> is very time-consuming so I don’t think it will get done any time soon.
>
> Is this just a rebase on top of https://github.com/iains/gcc-darwin-arm64?
> If I can test this on a cross-compiler then I could investigate there
For my testing I just merged in a branch that Jonathan pushed*** which had the
series
on it (but I’d think the patch series would just apply on top of the current
iains/gcc-darwin-arm64)
> - I know
> already that many of these tests could be sensitive to tuning.
function body tests are tricky to make portable indeed (I have a bunch of x86
fixes to do).
There are some differences in ABI that tend to affect prologues and integer
promotions.
>> Is there some generic execution smoke test for SME (without SVE)?
>
> I don't think we have any "generic" execution tests, but we do have a couple
> in
> g++.target/aarch64/sme/sme_throw_*. If those tests execute correctly, then
> that indicates that various parts of the toolchain are doing the right thing
> (at least some of the time).
will take a look at these.
thanks
Iain
*** forge / redi / aarch64-support-sme-without-sve
> Thanks,
> Alice
>
>>
>> thanks
>> Iain
>>
>>>
>>> Is this ok for master?
>>>
>>> Thanks,
>>> Alice
