hardcmp: split before dispatch edge

hardcmp: split before dispatch edge

[Alexandre Oliva]

If we harden a compare at the end of a block with an edge to the
abnormal dispatch block, it won't have a single successor.  Arrange to
split the block at its final stmt so as to have a single succ.

Regstrapped on x86_64-linux-gnu.  Ok to install?


for  gcc/ChangeLog

	PR middle-end/104975
	* gimple-harden-conditionals.cc
	(pass_harden_compares::execute): Force split in case of
	multiple edges.

for  gcc/testsuite/ChangeLog

	PR middle-end/104975
	* gcc.dg/pr104975.c: New.
---
 gcc/gimple-harden-conditionals.cc |   12 +++++++++---
 gcc/testsuite/gcc.dg/pr104975.c   |   20 ++++++++++++++++++++
 2 files changed, 29 insertions(+), 3 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/pr104975.c

diff --git a/gcc/gimple-harden-conditionals.cc b/gcc/gimple-harden-conditionals.cc
index 6a5fc3fb9e1a2..be01f3ea8c44a 100644
--- a/gcc/gimple-harden-conditionals.cc
+++ b/gcc/gimple-harden-conditionals.cc
@@ -509,10 +509,16 @@ pass_harden_compares::execute (function *fun)
 	gsi_insert_before (&gsi_split, asgnck, GSI_SAME_STMT);
 
 	/* We wish to insert a cond_expr after the compare, so arrange
-	   for it to be at the end of a block if it isn't.  */
-	if (!gsi_end_p (gsi_split))
+	   for it to be at the end of a block if it isn't, and for it
+	   to have a single successor in case there's more than
+	   one, as in PR104975.  */
+	if (!gsi_end_p (gsi_split)
+	    || !single_succ_p (gsi_bb (gsi_split)))
 	  {
-	    gsi_prev (&gsi_split);
+	    if (!gsi_end_p (gsi_split))
+	      gsi_prev (&gsi_split);
+	    else
+	      gsi_split = gsi_last_bb (gsi_bb (gsi_split));
 	    basic_block obb = gsi_bb (gsi_split);
 	    basic_block nbb = split_block (obb, gsi_stmt (gsi_split))->dest;
 	    gsi_next (&gsi_split);
diff --git a/gcc/testsuite/gcc.dg/pr104975.c b/gcc/testsuite/gcc.dg/pr104975.c
new file mode 100644
index 0000000000000..04532fc444340
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr104975.c
@@ -0,0 +1,20 @@
+/* { dg-do compile } */
+/* { dg-options "-O1 -fharden-compares -fno-inline -fno-ipa-pure-const" } */
+
+__attribute__ ((pure, returns_twice)) int
+bar (int);
+
+int
+quux (void)
+{
+  return 0;
+}
+
+int
+foo (short int x)
+{
+  x = !x;
+  bar (quux ());
+
+  return x;
+}


-- 
Alexandre Oliva, happy hacker                https://FSFLA.org/blogs/lxo/
   Free Software Activist                       GNU Toolchain Engineer
Disinformation flourishes because many people care deeply about injustice
but very few check the facts.  Ask me about <https://stallmansupport.org>

Re: hardcmp: split before dispatch edge

[Richard Biener]

On Thu, Mar 24, 2022 at 2:43 AM Alexandre Oliva via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
>
> If we harden a compare at the end of a block with an edge to the
> abnormal dispatch block, it won't have a single successor.  Arrange to
> split the block at its final stmt so as to have a single succ.
>
> Regstrapped on x86_64-linux-gnu.  Ok to install?

OK.

>
> for  gcc/ChangeLog
>
>         PR middle-end/104975
>         * gimple-harden-conditionals.cc
>         (pass_harden_compares::execute): Force split in case of
>         multiple edges.
>
> for  gcc/testsuite/ChangeLog
>
>         PR middle-end/104975
>         * gcc.dg/pr104975.c: New.
> ---
>  gcc/gimple-harden-conditionals.cc |   12 +++++++++---
>  gcc/testsuite/gcc.dg/pr104975.c   |   20 ++++++++++++++++++++
>  2 files changed, 29 insertions(+), 3 deletions(-)
>  create mode 100644 gcc/testsuite/gcc.dg/pr104975.c
>
> diff --git a/gcc/gimple-harden-conditionals.cc b/gcc/gimple-harden-conditionals.cc
> index 6a5fc3fb9e1a2..be01f3ea8c44a 100644
> --- a/gcc/gimple-harden-conditionals.cc
> +++ b/gcc/gimple-harden-conditionals.cc
> @@ -509,10 +509,16 @@ pass_harden_compares::execute (function *fun)
>         gsi_insert_before (&gsi_split, asgnck, GSI_SAME_STMT);
>
>         /* We wish to insert a cond_expr after the compare, so arrange
> -          for it to be at the end of a block if it isn't.  */
> -       if (!gsi_end_p (gsi_split))
> +          for it to be at the end of a block if it isn't, and for it
> +          to have a single successor in case there's more than
> +          one, as in PR104975.  */
> +       if (!gsi_end_p (gsi_split)
> +           || !single_succ_p (gsi_bb (gsi_split)))
>           {
> -           gsi_prev (&gsi_split);
> +           if (!gsi_end_p (gsi_split))
> +             gsi_prev (&gsi_split);
> +           else
> +             gsi_split = gsi_last_bb (gsi_bb (gsi_split));
>             basic_block obb = gsi_bb (gsi_split);
>             basic_block nbb = split_block (obb, gsi_stmt (gsi_split))->dest;
>             gsi_next (&gsi_split);
> diff --git a/gcc/testsuite/gcc.dg/pr104975.c b/gcc/testsuite/gcc.dg/pr104975.c
> new file mode 100644
> index 0000000000000..04532fc444340
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/pr104975.c
> @@ -0,0 +1,20 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O1 -fharden-compares -fno-inline -fno-ipa-pure-const" } */
> +
> +__attribute__ ((pure, returns_twice)) int
> +bar (int);
> +
> +int
> +quux (void)
> +{
> +  return 0;
> +}
> +
> +int
> +foo (short int x)
> +{
> +  x = !x;
> +  bar (quux ());
> +
> +  return x;
> +}
>
>
> --
> Alexandre Oliva, happy hacker                https://FSFLA.org/blogs/lxo/
>    Free Software Activist                       GNU Toolchain Engineer
> Disinformation flourishes because many people care deeply about injustice
> but very few check the facts.  Ask me about <https://stallmansupport.org>