more analysis on update_flow_info() crashes on sparc/rs6k
Jeffrey A Law
law@cygnus.com
Mon Oct 19 09:29:00 GMT 1998
In message < 199810191327.GAA01771@dm.cobaltmicro.com >you write:
> Jeff, consider the code you removed here in update_flow_info() in both
> instruction scheduler implementations the other day, both of which
> looked something like:
>
> if (insn == first)
> {
> if (reload_completed && this_is_reg_dead_note)
> {
> forget_the_note();
> break;
> }
> abort();
> }
>
> And as I said my splits only run after reload.
>
> This is why it always worked before, it isn't rocket science :-)
Sure isn't rocket science. The comment in that sched.c code is rather
misleading. For some reason it didn't trigger that I'd just changed that
code as part of the post-reload flow patch.
I'll bring it back with a better comment. :-)
jeff
More information about the Gcc-bugs
mailing list