https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92692
--- Comment #14 from Jakub Jelinek <jakub at gcc dot gnu.org> --- (In reply to Wilco from comment #10) > (In reply to Jakub Jelinek from comment #9) > > Any -march= or similar? Can't reproduce with current trunk, nor > > with even Oct 10 GCC snapshot (crosses in both cases). > > grep -B1 stxr pr92692.s > > doesn't show any stores before stxr. > > It's going to be extremely sensitive to register allocation, so it's not > clear it's worth trying to reproduce. > > The easiest option is to see whether replacing "reload_completed" with > "epilogue_completed" in aarch64/atomics.md works fine. For the shrink wrapping issue that is really needed on splitters that split into multiple basic blocks, atomic splitters that don't create any branches nor labels should be fine with reload_completed.
