From: Richard Biener <richard.guenther@gmail.com>
To: Richard Earnshaw <Richard.Earnshaw@arm.com>
Cc: "gcc-patches@gcc.gnu.org" <gcc-patches@gcc.gnu.org>
Subject: Re: cselib: add function to check if SET is redundant [PR106187]
Date: Fri, 29 Jul 2022 09:06:15 +0200 [thread overview]
Message-ID: <CAFiYyc2ALG986ZW=7L0OSTnCYgpEYEmHfozbwW9QYLFdKMj+qA@mail.gmail.com> (raw)
In-Reply-To: <e0748603-3540-d067-c61c-97f60115a588@foss.arm.com>
On Thu, Jul 28, 2022 at 6:46 PM Richard Earnshaw
<Richard.Earnshaw@foss.arm.com> wrote:
>
> [resend with correct subject line]
>
> A SET operation that writes memory may have the same value as an earlier
> store but if the alias sets of the new and earlier store do not conflict
> then the set is not truly redundant. This can happen, for example, if
> objects of different types share a stack slot.
>
> To fix this we define a new function in cselib that first checks for
> equality and if that is successful then finds the earlier store in the
> value history and checks the alias sets.
>
> The routine is used in two places elsewhere in the compiler. Firstly
> in cfgcleanup and secondly in postreload.
I can't comment on the stripping on SUBREGs and friends but it seems
to be conservative apart from
+ if (!flag_strict_aliasing || !MEM_P (dest))
+ return true;
where if dest is not a MEM but were to contain one we'd miss it.
Double-checking
from more RTL literate people appreciated.
+ /* Lookup the equivalents to the dest. This is more likely to succeed
+ than looking up the equivalents to the source (for example, when the
+ src is some form of constant). */
I think the comment is misleading - we _do_ have to lookup the MEM,
looking up equivalences of a reg or an expression on the RHS isn't
what we are interested in.
+ return alias_sets_conflict_p (MEM_ALIAS_SET (dest),
+ MEM_ALIAS_SET (src_equiv));
that's not conservative enough - dse.cc has correct boilerplate, we have
to check both MEM_ALIAS_SET and MEM_EXPR here (the latter only
if the former load/store has a MEM_EXPR). Note in particular
using alias_set_subset_of instead of alias_sets_conflict_p.
/* We can only remove the later store if the earlier aliases
at least all accesses the later one. */
&& ((MEM_ALIAS_SET (mem) == MEM_ALIAS_SET (s_info->mem)
|| alias_set_subset_of (MEM_ALIAS_SET (mem),
MEM_ALIAS_SET (s_info->mem)))
&& (!MEM_EXPR (s_info->mem)
|| refs_same_for_tbaa_p (MEM_EXPR (s_info->mem),
MEM_EXPR (mem)))))
+ /* We failed to find a recorded value in the cselib history, so try the
+ source of this set. */
+ rtx src = SET_SRC (set);
+ while (GET_CODE (src) == SUBREG)
+ src = XEXP (src, 0);
+
+ if (MEM_P (src) && rtx_equal_for_cselib_1 (dest_addr, XEXP (src, 0),
+ GET_MODE (dest), 0))
+ return alias_sets_conflict_p (MEM_ALIAS_SET (dest),
+ MEM_ALIAS_SET (src));
this looks like an odd case to me - wouldn't that only catch things
like self-assignments, aka *p = *p? So I'd simply drop this fallback.
Otherwise it looks OK to me.
Thanks,
Richard.
> gcc/ChangeLog:
> * cselib.h (cselib_redundant_set_p): Declare.
> * cselib.cc: Include alias.h
> (cselib_redundant_set_p): New function.
> * cfgcleanup.cc: (mark_effect): Use cselib_redundant_set_p instead
> of rtx_equal_for_cselib_p.
> * postreload.c (reload_cse_simplify): Use cselib_redundant_set_p.
> (reload_cse_noop_set_p): Delete.
next prev parent reply other threads:[~2022-07-29 7:06 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-07-28 16:46 Richard Earnshaw
2022-07-29 7:06 ` Richard Biener [this message]
2022-07-29 9:52 ` Richard Earnshaw
2022-07-29 9:56 ` Richard Biener
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CAFiYyc2ALG986ZW=7L0OSTnCYgpEYEmHfozbwW9QYLFdKMj+qA@mail.gmail.com' \
--to=richard.guenther@gmail.com \
--cc=Richard.Earnshaw@arm.com \
--cc=gcc-patches@gcc.gnu.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).