This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: GCC-3.4 reorders asm() with -O2
> On Fri, Jan 23, 2004 at 05:15:24PM -0800, Mike Eager wrote:
> > GCC-3.4 orders asm() outside of functions differently depending on
> > whether code is compiled -O1 or -O2. This behavior has changed from
> > gcc-3.3.1.
> >
> > Here's a small piece of C code (extracted from glibc):
> >
> > initfini.c:
> > asm ("\n/*@HEADER_ENDS*/");
> >
> > asm ("\n/*@TESTS_BEGIN*/");
> > void dummy (void) { }
> > asm ("\n/*@TESTS_END*/");
> >
> > asm ("\n/*@_init_PROLOG_BEGINS*/");
>
> > So, this is different behavior between gcc-3.3.1 and gcc-3.4.
> > Is there any guarantee that asm() outside of a function will
> > not be reordered with respect to functions?
> > Is there any way to suppress this reordering?
>
> No, and -fno-unit-at-a-time (for now, at least). This should probably
> be in changes.html if it isn't already.
>
> Take a look at the glibc CVS; this is handled now. The Linux kernel
> has or had a number of similar problems.
>
> linuxthreads/Makefile:CFLAGS-pt-initfini.s = -g0 -fPIC -fno-inline-functions $(fno-unit-at-a-time)
I think Linux kernel compiles completely with unit-at-a-time now. I
would much prefer glibc to avoid -fno-unit-at-a-time, so it can be
dropped in the future.
Honza
>
> --
> Daniel Jacobowitz
> MontaVista Software Debian GNU/Linux Developer