This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Local update flow info assumes valid situations are illegal
On Jan 4, 2001, Richard Henderson <rth@redhat.com> wrote:
> On Thu, Jan 04, 2001 at 09:34:15PM -0200, Alexandre Oliva wrote:
>> Why not? The bb (and the function) ended with a call to a noreturn
>> function. The reg_set at the end of the bb was empty. There was no
>> use of the frame pointer remaining. So why isn't it dead?
> Because the frame pointer is live throughout the entire function
> before reload. This is because every pseudo is a potential
> reference to the frame pointer if it is spilled to the stack.
What if there are no remaining pseudos?
The remaining insns, after combine, are:
(set (reg:SI 0) (const_int 0))
(call (mem:SI (symbol_ref:SI "bar")) 0)
(use (const_int 0))
(barrier)
Now, this is a relatively old branch, so if forcing the frame pointer
to be live throughout the function is something new, maybe it's just
missing from the branch. Could this be the case?
--
Alexandre Oliva Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer aoliva@{cygnus.com, redhat.com}
CS PhD student at IC-Unicamp oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist *Please* write to mailing lists, not to me