BUG: -funroll-all-loops generates unresolved externals

[Manfred Hollstein]

On Thursday, 3 August 2000, 17:34:41 +0200, manfredh@redhat.com wrote:

> On Thursday, 3 August 2000, 11:26:53 -0400, ghazi@caip.rutgers.edu wrote:
> 
> >  > From: Manfred Hollstein <manfredh@redhat.com>
> >  > 
> >  > 2000-08-03  Manfred Hollstein  <manfredh@redhat.com>
> >  > 
> >  > 	* gcc.dg/20000803-1.c: New test.
> >  > 
> >  > --- /dev/null	Tue May  5 22:32:27 1998
> >  > +++ gcc/testsuite/gcc.dg/20000803-1.c	Thu Aug  3 12:44:19 2000
> >  > @@ -0,0 +1,57 @@
> >  > +/* Copyright (C) 2000  Free Software Foundation.
> >  > +
> >  > +   by Manfred Hollstein <manfredh@redhat.com>  */
> >  > +
> >  > +/* { dg-do link } */
> >  > +/* { dg-options "-O2 -funroll-all-loops" } */
> > 
> > Why not put this in gcc.c-torture/compile, so other opt levels get a
> > workout too?
> 
> That's fine. I just grabbed an arbitrary example which looked simple
> with regard to how to describe required flags; simply overlooked that
> TORTURE_OPTIONS also include `-funroll-all-loops' already.

No, it's not fine. The variant which includes `-funroll-all-loops'
also includes `-finline-functions' which makes the error go away;
and, it would have to be gcc.c-torture/execute anyway, since just
compiling wouldn't show the unresolved symbols.

Cheers, manfred.