This is the mail archive of the
mailing list for the GCC project.
Re: Fix PR 46297
On Sat, Nov 06, 2010 at 09:20:19PM +0100, Richard Guenther wrote:
> On Sat, Nov 6, 2010 at 4:53 PM, Jeff Law <email@example.com> wrote:
> > ?The fundamental problem is some ports have embedded auto-inc addressing
> > modes without REG_INC nodes. ?The improved reload optimizations were not
> > accounting for the possibility of embedded auto-inc addressing, particularly
> > for argument pushing.
> > As a result the new optimization applied, but the offset we generated in the
> > MEM was wrong due to the intervening autoinc addressing mode.
> > While we could track the auto-incs, it's far easier to treat them as an
> > unrecordable use/set which disables the optimization across the argument
> > pushes. ? While we lose a few optimization opportunities, I suspect they
> > definitely fall into the unmeasurable category.
> > Bootstrapped & regression tested on i686-pc-linux-gnu.
> Can we detect this in RTL checking and fixup the targets? Obviously not
> for stage3 though.
Currently REG_INC notes are only used on AUTO_INC_DEC targets, and
i386/x86_64 is not such a target.