Re: [patch] Fix PR middle-end/68251

[Richard Biener <richard dot guenther at gmail dot com>]

On Mon, Nov 9, 2015 at 12:13 PM, Eric Botcazou <ebotcazou@adacore.com> wrote:
> Hi,
>
> I don't understand how this didn't show up sooner given the amount of testing,
> but there is a nasty overloading for the new REF_REVERSE_STORAGE_ORDER flag:
>
> #define REF_REVERSE_STORAGE_ORDER(NODE) \
>    (TREE_CHECK2 (NODE, BIT_FIELD_REF, MEM_REF)->base.u.bits.saturating_flag)
>
> saturating_flag is part of the 'bits' structure, which is itself part of the
> 'u' union, which also contains the dependence_info structure, and:
>
> #define MR_DEPENDENCE_CLIQUE(NODE) \
>   (TREE_CHECK2 (NODE, MEM_REF, TARGET_MEM_REF)->base.u.dependence_info.clique)
> #define MR_DEPENDENCE_BASE(NODE) \
>   (TREE_CHECK2 (NODE, MEM_REF, TARGET_MEM_REF)->base.u.dependence_info.base)
>
> So there is a conflict on MEM_REF nodes.  Therefore the attached patch changes
> REF_REVERSE_STORAGE_ORDER to use the default_def_flag instead.
>
> Tested on x86_64-suse-linux, OK for the mainline?  I'll install the Fortran
> testcase once it is reduced because it takes a while to compile ATM.

Looks good to me.  I wonder where you store the info on
TARGET_MEM_REFs though?  Or
is IVOPTs prohibited from messing with such refs (pessmizing them)?

Thanks,
Richard.

>
> 2015-11-09  Eric Botcazou  <ebotcazou@adacore.com>
>
>         PR middle-end/68251
>         * tree-core.h (REF_REVERSE_STORAGE_ORDER): Move around.
>         * tree.h (REF_REVERSE_STORAGE_ORDER): Change to default_def_flag.
>         * tree-streamer-in.c (unpack_ts_base_value_fields): Adjust.
>         * tree-streamer-out.c (pack_ts_base_value_fields): Likewise.
>
> --
> Eric Botcazou