This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RS6000] pr65810, powerpc64 alignment of r2


On Thu, Apr 23, 2015 at 7:24 AM, Alan Modra <amodra@gmail.com> wrote:
> Revised patch, supporting linker that aligns the toc base.
>
> This fixes a thinko in offsettable_ok_by_alignment.  It's not the
> absolute placement that matters, but the toc-pointer relative offset.
> So alignment of r2 also needs to be taken into account.
>
> Changing offsettable_ok_by_alignment has a ripple effect into the 'm'
> constraint so we also need to ensure rs6000_legitimize_reload_address
> does not create invalid toc-relative addresses.  As found by
> gcc.dg/torture/builtin-math-2.c -Os.  That's the reason for the
> use_toc_relative_ref change.  I hope the size check along with
> reg_offset_p is sufficient here.  It seems so, but it's difficult to
> be certain due to how hard it is to get just the right combination of
> reload conditions to trigger.
>
> Bootstrapped and regression tested powerpc64-linux and
> powerpc64le-linux, both with a new and old linker.  OK for mainline?
>
>         PR target/65810
>         * config/rs6000/rs6000.c (POWERPC64_TOC_POINTER_ALIGNMENT): Define.
>         (offsettable_ok_by_alignment): Use minimum of decl and toc
>         pointer alignment.  Replace dead code with assertion.
>         (use_toc_relative_ref): Add mode arg.  Return false in -mcmodel=medium
>         case if size exceeds toc pointer alignment.
>         (rs6000_legitimize_reload_address): Update use_toc_relative_ref call.
>         (rs6000_emit_move): Likewise.
>         * configure.ac: Add linker toc pointer alignment check.
>         * configure: Regenerate.
>         * config.in: Regenerate.

Okay.

Thanks, David


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]