public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Andrew MacLeod <amacleod@redhat.com>
To: Richard Biener <richard.guenther@gmail.com>
Cc: gcc-patches <gcc-patches@gcc.gnu.org>,
	"hernandez, aldy" <aldyh@redhat.com>
Subject: Re: [PATCH] PR tree-optimization/107985 - Ensure arguments to range-op handler are supported.
Date: Thu, 8 Dec 2022 08:53:26 -0500	[thread overview]
Message-ID: <a0e35c7d-e7d8-966e-fd4c-a47c4fd1c2f2@redhat.com> (raw)
In-Reply-To: <CAFiYyc2pEOfVMuAV1POYekTCQXj3R2RE7HiPBYVO_xRe68=JXQ@mail.gmail.com>

[-- Attachment #1: Type: text/plain, Size: 1355 bytes --]

On 12/7/22 12:26, Richard Biener wrote:
> On Wed, Dec 7, 2022 at 5:45 PM Andrew MacLeod via Gcc-patches
> <gcc-patches@gcc.gnu.org> wrote:
>> THis patch invalidates a range-op handler object if an operand type in
>> the statement is not supported.
>>
>> This also triggered a check in stmt dependency resolution which assumed
>> there must be a valid handler for any stmt with an appropriate LHS
>> type... which is a false assumption.
>>
>> This should do for now, but long term I will rework the dispatch code to
>> ensure it matches the specifically supported patterns of operands. This
>> will make the handler creation a little slower, but speed up the actual
>> dispatch, especially as we add new range types next release.  Its also
>> much more invasive... too much for this release I think.
>>
>> bootstraps on x86_64-pc-linux-gnu with no regressions.  OK?
> +         if (!Value_Range::supports_type_p (TREE_TYPE (m_op1)) ||
> +             !Value_Range::supports_type_p (TREE_TYPE (m_op2)))
>
> The ||s go to the next line.  Since in a GIMPLE_COND both operand types
> are compatible it's enough to check one of them.
>
> Likewise for the GIMPLE_ASSIGN case I think - I don't know of any
> binary operator that has operands that would not be both compatible
> or not compatible (but it's less clear-cut here).
>
Doh.  Checked this in:

Andrew


[-- Attachment #2: 985.diff --]
[-- Type: text/x-patch, Size: 2892 bytes --]

commit e3251e14bccf3891b265293371c7b7f95e306271
Author: Andrew MacLeod <amacleod@redhat.com>
Date:   Tue Dec 6 10:41:29 2022 -0500

    Ensure arguments to range-op handler are supported.
    
            PR tree-optimization/107985
            gcc/
            * gimple-range-op.cc
            (gimple_range_op_handler::gimple_range_op_handler): Check if type
            of the operands is supported.
            * gimple-range.cc (gimple_ranger::prefill_stmt_dependencies): Do
            not assert if here is no range-op handler.
    
            gcc/testsuite/
            * g++.dg/pr107985.C: New.

diff --git a/gcc/gimple-range-op.cc b/gcc/gimple-range-op.cc
index 7764166d5fb..12068544bc5 100644
--- a/gcc/gimple-range-op.cc
+++ b/gcc/gimple-range-op.cc
@@ -148,6 +148,9 @@ gimple_range_op_handler::gimple_range_op_handler (gimple *s)
 	case GIMPLE_COND:
 	  m_op1 = gimple_cond_lhs (m_stmt);
 	  m_op2 = gimple_cond_rhs (m_stmt);
+	  // Check that operands are supported types.  One check is enough.
+	  if (!Value_Range::supports_type_p (TREE_TYPE (m_op1)))
+	    m_valid = false;
 	  return;
 	case GIMPLE_ASSIGN:
 	  m_op1 = gimple_range_base_of_assignment (m_stmt);
@@ -164,6 +167,9 @@ gimple_range_op_handler::gimple_range_op_handler (gimple *s)
 	    }
 	  if (gimple_num_ops (m_stmt) >= 3)
 	    m_op2 = gimple_assign_rhs2 (m_stmt);
+	  // Check that operands are supported types.  One check is enough.
+	  if ((m_op1 && !Value_Range::supports_type_p (TREE_TYPE (m_op1))))
+	    m_valid = false;
 	  return;
 	default:
 	  gcc_unreachable ();
diff --git a/gcc/gimple-range.cc b/gcc/gimple-range.cc
index ecd6039e0fd..8c055826e17 100644
--- a/gcc/gimple-range.cc
+++ b/gcc/gimple-range.cc
@@ -422,18 +422,20 @@ gimple_ranger::prefill_stmt_dependencies (tree ssa)
       else
 	{
 	  gimple_range_op_handler handler (stmt);
-	  gcc_checking_assert (handler);
-	  tree op = handler.operand2 ();
-	  if (op)
+	  if (handler)
 	    {
-	      Value_Range r (TREE_TYPE (op));
-	      prefill_name (r, op);
-	    }
-	  op = handler.operand1 ();
-	  if (op)
-	    {
-	      Value_Range r (TREE_TYPE (op));
-	      prefill_name (r, op);
+	      tree op = handler.operand2 ();
+	      if (op)
+		{
+		  Value_Range r (TREE_TYPE (op));
+		  prefill_name (r, op);
+		}
+	      op = handler.operand1 ();
+	      if (op)
+		{
+		  Value_Range r (TREE_TYPE (op));
+		  prefill_name (r, op);
+		}
 	    }
 	}
     }
diff --git a/gcc/testsuite/g++.dg/pr107985.C b/gcc/testsuite/g++.dg/pr107985.C
new file mode 100644
index 00000000000..8d244b54efb
--- /dev/null
+++ b/gcc/testsuite/g++.dg/pr107985.C
@@ -0,0 +1,18 @@
+/* { dg-do compile } */
+/* { dg-options "-O1 -ftree-vrp -fno-tree-ccp -fno-tree-forwprop -fno-tree-fre" } */
+
+struct B {
+  int f;
+};
+
+struct D : public B {
+};
+
+void foo() {
+  D d;
+  d.f = 7;
+
+  int B::* pfb = &B::f;
+  int D::* pfd = pfb;
+  int v = d.*pfd;
+}

      reply	other threads:[~2022-12-08 13:53 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-12-07 16:44 Andrew MacLeod
2022-12-07 17:26 ` Richard Biener
2022-12-08 13:53   ` Andrew MacLeod [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=a0e35c7d-e7d8-966e-fd4c-a47c4fd1c2f2@redhat.com \
    --to=amacleod@redhat.com \
    --cc=aldyh@redhat.com \
    --cc=gcc-patches@gcc.gnu.org \
    --cc=richard.guenther@gmail.com \
    /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).