Am Montag, dem 15.09.2025 um 16:04 -0700 schrieb Andrew Pinski: > On Mon, Sep 15, 2025 at 3:59 PM Alexandre Oliva <ol...@adacore.com> wrote: > > > > On Sep 13, 2025, Alexandre Oliva <ol...@adacore.com> wrote: > > > > > gcc.target/aarch64/sme/nonlocal_goto_[123].c fail on aarch64 targets > > > configured with --enable-default-pie. > > > > FTR, Martin's commit afa74d37e8170d696f97424da7ab0f71883aac70 regressed > > nonlocal_goto_2.c in a slightly different way, even in a standard > > aarch64-elf build. I was surprised that the warning patch caused any > > codegen changes, but it seems to prevent collapsing the nonlocal label > > with the return block it falls through to, and the expected code pattern > > no longer matches because the separation survives through to the end of > > compilation, with an extra smstart and padding. > > Yes the problem with non-local lable/goto's is recorded already as > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121933 . > It caused gcs-nonlocal-1-track-speculation.c to fail too.
Sorry, for that. The fix shown there is bootstrapped and regression tested on aarch64. I am travelling now but plan to submit it at the end of the week. But it would certainly be nice if we *could* use the DECL_NONLOCAL bit in the FE to track this information for decls. So I am do not know what consequences this might have, but if it could be made to work I would prefer this. Martin
