Re: GCC 3.0 Release Criteria

[Geoff Keating <geoffk at cygnus dot com>]

Michael Meissner <meissner@cygnus.com> writes:

> On Thu, Apr 27, 2000 at 12:37:27PM +0100, Bernd Schmidt wrote:
> > >  Could the 10% increase in code size be due to changed alignment policy?
> > 
> > Try again with
> >   -malign-jumps=0 -malign-loops=0 -malign-functions=0 -mpreferred-stack-boundary=2
> > in the CFLAGS.  I don't remember whether 2.7.2 already had the alignment
> > flags, if so, pass them to that compiler as well.
> 
> Yes 2.7.2 had the -m alignment flags (but not the -f ones we now
> have).  I believe I added them in the 2.6.x time frame (they were
> added over 5 1/2 years ago when I was working at OSF).

If this is on ia32, note that I think the ia32 backend has still not
been updated to handle the new -falign-* options, so they don't work;
they are overriden by the -m options, even by the defaults of the -m
options.  The problem is that the ia32 backend defines LOOP_ALIGN and
LABEL_ALIGN_AFTER_BARRIER, overriding the defaults.

ia32 also misuses FUNCTION_BOUNDARY, unless -malign-functions is
supposed to cause an ABI change.

-- 
- Geoffrey Keating <geoffk@cygnus.com>