SH sibcalls

[Alexandre Oliva]

On Nov 23, 2000, Alexandre Oliva <aoliva@redhat.com> wrote:

> I'm a bit unsure about whether it's more appropriate for `sibcalli'
> and `sibcalli_pcrel' and `sibcall_pcrel' to have type `jump_ind' or
> `call', but the way it is now (the latter is `call', the others are
> `jump_ind') appears to be working fine (i.e., no regressions in the
> testsuite).

> Ok to install?

Nope.  There were actually 2 failures I had failed to notice before.
We must not assign a call-saved register to hold the address of the
function to be sibcalled, because the epilogue will overwrite it.

I'll probably have to create a new constraint to match only
call-clobbered registers.  Meanwhile `z' (r0) will do.  Ok to install
with this modification, assuming a new round of testing fixes the
problems without introducing new ones?

-- 
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