Re: [patch] Do not directly change TARGET_MEM_REF in maybe_canonicalize_mem_ref_addr

[Richard Biener <richard.guenther@gmail.com>]

On Thu, Jun 15, 2017 at 3:51 PM, Eric Botcazou <ebotcazou@adacore.com> wrote:
> Hi,
>
> the transformation done to TARGET_MEM_REF in maybe_canonicalize_mem_ref_addr
> is exactly the same as one of those done in maybe_fold_tmr, the latter is
> better written and the former function calls the latter, so this patch changes
> maybe_canonicalize_mem_ref_addr to avoid touching TARGET_MEM_REF directly.
>
> Tested on x86-64/Linux, OK for the mainline?

I don't think so.  get_address_description assumes TMR_BASE is in
canonical form,
that is, when it is an ADDR_EXPR we have a symbol and when not we have
a pointer.
TMR[&p->a] violates this and the gimple-fold.c part first canonicalizes this to
TMR[p + offsetof(a)].

Richard.

>
> 2017-06-15  Eric Botcazou  <ebotcazou@adacore.com>
>
>         PR bootstrap/80897
>         * gimple-fold.c (maybe_canonicalize_mem_ref_addr): Do not change
>         TARGET_MEM_REF expressions directly.
>
> --
> Eric Botcazou