non-local goto optimization

Kaz Kojima
Tue Sep 11 11:27:00 GMT 2007

Eric Botcazou <> wrote:
> It's a heads-up to remind you of the above change, now implemented after some
> annoying delay.  To recap, the Alpha, ARM, MMIX, s390 and SH back-ends should
> be audited because they directly test current_function_has_nonlocal_label.

I've confirmed that there is no problem with applying
s/_has_nonlocal_label/_saves_all_registers/ in config/sh/*
on sh-elf, sh64-elf and sh4-unknown-linux-gnu.  I'd like
to commit that change.
FYI, all uses of current_function_has_nonlocal_label in SH
back end are guarded with TARGET_SHCOMPACT which means that
the change affects only sh64 targets with -m5-compact* options.


More information about the Gcc-patches mailing list