public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Richard Biener <rguenther@suse.de>
To: Martin Jambor <mjambor@suse.cz>,GCC Patches <gcc-patches@gcc.gnu.org>
Subject: Re: [PATCH] sra: Do not create zero sized accesses  (PR93776)
Date: Tue, 18 Feb 2020 16:45:00 -0000	[thread overview]
Message-ID: <093AE8BF-98E3-4EB8-817A-1757407970A1@suse.de> (raw)
In-Reply-To: <ri6k14jj3d8.fsf@suse.cz>

On February 18, 2020 4:56:35 PM GMT+01:00, Martin Jambor <mjambor@suse.cz> wrote:
>Hi,
>
>SRA can get a bit confused with zero-sized accesses like the one in
>the testcase.  Since there is nothing in the access, nothing is
>scalarized, but we can get order of the structures wrong, which the
>verifier is not happy about.
>
>Fixed by simply ignoring such accesses.  Bootstrapped and tested on an
>x86_64-linux.  OK for trunk?

Ok. 

Richard. 

>Thanks,
>
>Martin
>
>2020-02-18  Martin Jambor  <mjambor@suse.cz>
>
>	PR tree-optimization/93776
>	* tree-sra.c (create_access): Do not create zero size accesses.
>	(get_access_for_expr): Do not search for zero sized accesses.
>
>	testsuite/
>	* gcc.dg/tree-ssa/pr93776.c: New test.
>---
> gcc/ChangeLog                           |  6 ++++++
> gcc/testsuite/ChangeLog                 |  5 +++++
> gcc/testsuite/gcc.dg/tree-ssa/pr93776.c | 27 +++++++++++++++++++++++++
> gcc/tree-sra.c                          |  5 ++++-
> 4 files changed, 42 insertions(+), 1 deletion(-)
> create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr93776.c
>
>diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr93776.c
>b/gcc/testsuite/gcc.dg/tree-ssa/pr93776.c
>new file mode 100644
>index 00000000000..c407a627718
>--- /dev/null
>+++ b/gcc/testsuite/gcc.dg/tree-ssa/pr93776.c
>@@ -0,0 +1,27 @@
>+/* { dg-do compile } */
>+/* { dg-options "-O1" } */
>+
>+struct empty {};
>+struct s { int i; };
>+struct z
>+{
>+  int j;
>+  struct empty e;
>+  struct s s;
>+  int k;
>+};
>+
>+void bar (struct z);
>+void baz (int);
>+
>+void foo (void)
>+{
>+  struct z z, z2;
>+
>+  z.k = 8;
>+  z2.s.i = 1;
>+  z = z2;
>+  bar (z);
>+  z.e = (struct empty) {};
>+  baz (z.k);
>+}
>diff --git a/gcc/tree-sra.c b/gcc/tree-sra.c
>index 0cfac0a8192..2c717805b68 100644
>--- a/gcc/tree-sra.c
>+++ b/gcc/tree-sra.c
>@@ -926,6 +926,8 @@ create_access (tree expr, gimple *stmt, bool write)
>       size = max_size;
>       unscalarizable_region = true;
>     }
>+  if (size == 0)
>+    return NULL;
>   if (size < 0)
>     {
>   disqualify_candidate (base, "Encountered an unconstrained access.");
>@@ -3629,7 +3631,8 @@ get_access_for_expr (tree expr)
> 	return NULL;
>     }
> 
>-  if (!bitmap_bit_p (candidate_bitmap, DECL_UID (base)))
>+  if (max_size == 0
>+      || !bitmap_bit_p (candidate_bitmap, DECL_UID (base)))
>     return NULL;
> 
>   return get_var_base_offset_size_access (base, offset, max_size);

      reply	other threads:[~2020-02-18 16:45 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-02-18 15:56 Martin Jambor
2020-02-18 16:45 ` Richard Biener [this message]

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=093AE8BF-98E3-4EB8-817A-1757407970A1@suse.de \
    --to=rguenther@suse.de \
    --cc=gcc-patches@gcc.gnu.org \
    --cc=mjambor@suse.cz \
    /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).