Thanks to @vejbomar work on unit tests grouping, we identified SVE/SVE2 and SME/SME2 as significant contributors to failing tests, totaling 212334. Based on the investigation, SVE is an optional extension for armv8.2-a, and SVE2/SME/SME2 require armv9-a. This means they are not priorities for the GCC 15 release for now.
The architecture version target is "armv8-a" for the GCC 15 release. It will help to focus on the right issues. SVE is an optional extension and is not needed for armv8-a.
Once SVE becomes a priority, the target architecture version will be changed to "armv8 + sve".
To enable SVE support in GCC it should be enough to enable "armv8-a + sve" architecture.
I have not tried that, and enabled SVE for armv8-a.
It emits .variant_pcs (Symbol may follow different call convention) which is not supported by COFF.
It was enough to add an empty handler for .variant_pcs to get bits for tests.
Tests were failing and this is where investigation stopped.
Thanks to @vejbomar work on unit tests grouping, we identified SVE/SVE2 and SME/SME2 as significant contributors to failing tests, totaling 212334. Based on the investigation, SVE is an optional extension for armv8.2-a, and SVE2/SME/SME2 require armv9-a. This means they are not priorities for the GCC 15 release for now.
The architecture version target is "armv8-a" for the GCC 15 release. It will help to focus on the right issues. SVE is an optional extension and is not needed for armv8-a. Once SVE becomes a priority, the target architecture version will be changed to "armv8 + sve".
To enable SVE support in GCC it should be enough to enable "armv8-a + sve" architecture. I have not tried that, and enabled SVE for armv8-a. It emits .variant_pcs (Symbol may follow different call convention) which is not supported by COFF. It was enough to add an empty handler for .variant_pcs to get bits for tests. Tests were failing and this is where investigation stopped.