[PATCH] reassoc: Fold some statements [PR108819]

[PATCH] reassoc: Fold some statements [PR108819]

[Jakub Jelinek]

Hi!

This spot in update_ops can replace one or both of the assign operands with
constants, creating 1 & 1 and similar expressions which can confuse later
passes until they are folded.  Rather than folding both constants by hand
and also handling swapping of operands for commutative ops if the first one
is constant and second one is not, the following patch just uses
fold_stmt_inplace to do that.  I think we shouldn't fold more than the
single statement because that could screw up the rest of the pass, we'd have
to mark all those with uids, visited and the like.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2023-02-18  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/108819
	* tree-ssa-reassoc.cc (update_ops): Fold new stmt in place.

	* gcc.dg/pr108819.c: New test.

--- gcc/tree-ssa-reassoc.cc.jj	2023-02-16 10:41:11.000000000 +0100
+++ gcc/tree-ssa-reassoc.cc	2023-02-17 22:43:18.949984553 +0100
@@ -4687,6 +4687,9 @@ update_ops (tree var, enum tree_code cod
       gimple_set_uid (g, gimple_uid (stmt));
       gimple_set_visited (g, true);
       gsi_insert_before (&gsi, g, GSI_SAME_STMT);
+      gimple_stmt_iterator gsi2 = gsi_for_stmt (g);
+      if (fold_stmt_inplace (&gsi2))
+	update_stmt (g);
     }
   return var;
 }
--- gcc/testsuite/gcc.dg/pr108819.c.jj	2023-02-17 17:50:10.084914418 +0100
+++ gcc/testsuite/gcc.dg/pr108819.c	2023-02-17 17:48:30.765369964 +0100
@@ -0,0 +1,19 @@
+/* PR tree-optimization/108819 */
+/* { dg-do compile } */
+/* { dg-options "-O1 -fno-tree-ccp -fno-tree-forwprop" } */
+
+int a, b;
+
+int
+main ()
+{
+  int d = 1;
+  for (; b; b++)
+    if (a < 1)
+      while (d <= a && a <= 0UL)
+	{
+	  int *e = &d;
+	  *e = 0;
+	}
+  return 0;
+}

	Jakub

Re: [PATCH] reassoc: Fold some statements [PR108819]

[Richard Biener]

> Am 18.02.2023 um 11:41 schrieb Jakub Jelinek via Gcc-patches <gcc-patches@gcc.gnu.org>:
> 
> Hi!
> 
> This spot in update_ops can replace one or both of the assign operands with
> constants, creating 1 & 1 and similar expressions which can confuse later
> passes until they are folded.  Rather than folding both constants by hand
> and also handling swapping of operands for commutative ops if the first one
> is constant and second one is not, the following patch just uses
> fold_stmt_inplace to do that.  I think we shouldn't fold more than the
> single statement because that could screw up the rest of the pass, we'd have
> to mark all those with uids, visited and the like.
> 
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

Ok

Richard 
> 2023-02-18  Jakub Jelinek  <jakub@redhat.com>
> 
>    PR tree-optimization/108819
>    * tree-ssa-reassoc.cc (update_ops): Fold new stmt in place.
> 
>    * gcc.dg/pr108819.c: New test.
> 
> --- gcc/tree-ssa-reassoc.cc.jj    2023-02-16 10:41:11.000000000 +0100
> +++ gcc/tree-ssa-reassoc.cc    2023-02-17 22:43:18.949984553 +0100
> @@ -4687,6 +4687,9 @@ update_ops (tree var, enum tree_code cod
>       gimple_set_uid (g, gimple_uid (stmt));
>       gimple_set_visited (g, true);
>       gsi_insert_before (&gsi, g, GSI_SAME_STMT);
> +      gimple_stmt_iterator gsi2 = gsi_for_stmt (g);
> +      if (fold_stmt_inplace (&gsi2))
> +    update_stmt (g);
>     }
>   return var;
> }
> --- gcc/testsuite/gcc.dg/pr108819.c.jj    2023-02-17 17:50:10.084914418 +0100
> +++ gcc/testsuite/gcc.dg/pr108819.c    2023-02-17 17:48:30.765369964 +0100
> @@ -0,0 +1,19 @@
> +/* PR tree-optimization/108819 */
> +/* { dg-do compile } */
> +/* { dg-options "-O1 -fno-tree-ccp -fno-tree-forwprop" } */
> +
> +int a, b;
> +
> +int
> +main ()
> +{
> +  int d = 1;
> +  for (; b; b++)
> +    if (a < 1)
> +      while (d <= a && a <= 0UL)
> +    {
> +      int *e = &d;
> +      *e = 0;
> +    }
> +  return 0;
> +}
> 
>    Jakub
>