[PATCH] [PR105455] Set edge probabilities when hardening conditionals

[PATCH] [PR105455] Set edge probabilities when hardening conditionals

[Alexandre Oliva]

When turning unconditional edges into conditional, as in
gimple-harden-conditionals.cc:insert_check_and_trap, the newly-created
edge's probability comes out uninitialized, while the previously
unconditional edge's probability is presumably
profile_probability::always.

Mixing initialized and uninitialized probabilities before expand
breaks predict.cc:force_edge_cold: the initialized probability may end
up copied to a REG_BR_PROB note in a conditional branch insn, but if
force_edge_cold is called on that edge, it will find another edge with
uninitialized probability and assume the note is absent.  Later on,
rtl_verify_edges complains that the note does not match the
probability modified by force_edge_cold in the edge.

This patch sets probabilities for edges affected by hardening of
conditionals, both the newly-created edges to trap blocks and the
previously-unconditional edges, so that the former are considered
never taken, while the latter are confirmed as always taken.

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


for  gcc/ChangeLog

	PR rtl-optimization/105455
	* gimple-harden-conditionals.cc (insert_check_and_trap): Set
	probabilities for newly-conditional edges.

for  gcc/testsuite/ChangeLog

	PR rtl-optimization/105455
	* gcc.dg/pr105455.c: New.
---
 gcc/gimple-harden-conditionals.cc |    2 ++
 gcc/testsuite/gcc.dg/pr105455.c   |   17 +++++++++++++++++
 2 files changed, 19 insertions(+)
 create mode 100644 gcc/testsuite/gcc.dg/pr105455.c

diff --git a/gcc/gimple-harden-conditionals.cc b/gcc/gimple-harden-conditionals.cc
index 28c4810f0a78e..240d0e5acd87a 100644
--- a/gcc/gimple-harden-conditionals.cc
+++ b/gcc/gimple-harden-conditionals.cc
@@ -254,8 +254,10 @@ insert_check_and_trap (location_t loc, gimple_stmt_iterator *gsip,
      equality.  */
   single_succ_edge (chk)->flags &= ~EDGE_FALLTHRU;
   single_succ_edge (chk)->flags |= neg_true_false_flag;
+  single_succ_edge (chk)->probability = profile_probability::always ();
   edge e = make_edge (chk, trp, true_false_flag);
   e->goto_locus = loc;
+  e->probability = profile_probability::never ();
 
   if (dom_info_available_p (CDI_DOMINATORS))
     set_immediate_dominator (CDI_DOMINATORS, trp, chk);
diff --git a/gcc/testsuite/gcc.dg/pr105455.c b/gcc/testsuite/gcc.dg/pr105455.c
new file mode 100644
index 0000000000000..81e9154baa1c3
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr105455.c
@@ -0,0 +1,17 @@
+/* { dg-do compile } */
+/* { dg-options "-O1 -fharden-conditional-branches -funroll-loops --param max-loop-header-insns=1" } */
+
+__attribute__ ((cold)) void
+bar (void);
+
+void
+foo (int x)
+{
+  if (x)
+    {
+      int i;
+
+      for (i = 0; i < 101; ++i)
+        bar ();
+    }
+}


-- 
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: [PATCH] [PR105455] Set edge probabilities when hardening conditionals

[Richard Biener]

On Thu, May 12, 2022 at 3:48 AM Alexandre Oliva via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
>
> When turning unconditional edges into conditional, as in
> gimple-harden-conditionals.cc:insert_check_and_trap, the newly-created
> edge's probability comes out uninitialized, while the previously
> unconditional edge's probability is presumably
> profile_probability::always.
>
> Mixing initialized and uninitialized probabilities before expand
> breaks predict.cc:force_edge_cold: the initialized probability may end
> up copied to a REG_BR_PROB note in a conditional branch insn, but if
> force_edge_cold is called on that edge, it will find another edge with
> uninitialized probability and assume the note is absent.  Later on,
> rtl_verify_edges complains that the note does not match the
> probability modified by force_edge_cold in the edge.
>
> This patch sets probabilities for edges affected by hardening of
> conditionals, both the newly-created edges to trap blocks and the
> previously-unconditional edges, so that the former are considered
> never taken, while the latter are confirmed as always taken.
>
> Regstrapped on x86_64-linux-gnu.  Ok to install?

OK.

>
> for  gcc/ChangeLog
>
>         PR rtl-optimization/105455
>         * gimple-harden-conditionals.cc (insert_check_and_trap): Set
>         probabilities for newly-conditional edges.
>
> for  gcc/testsuite/ChangeLog
>
>         PR rtl-optimization/105455
>         * gcc.dg/pr105455.c: New.
> ---
>  gcc/gimple-harden-conditionals.cc |    2 ++
>  gcc/testsuite/gcc.dg/pr105455.c   |   17 +++++++++++++++++
>  2 files changed, 19 insertions(+)
>  create mode 100644 gcc/testsuite/gcc.dg/pr105455.c
>
> diff --git a/gcc/gimple-harden-conditionals.cc b/gcc/gimple-harden-conditionals.cc
> index 28c4810f0a78e..240d0e5acd87a 100644
> --- a/gcc/gimple-harden-conditionals.cc
> +++ b/gcc/gimple-harden-conditionals.cc
> @@ -254,8 +254,10 @@ insert_check_and_trap (location_t loc, gimple_stmt_iterator *gsip,
>       equality.  */
>    single_succ_edge (chk)->flags &= ~EDGE_FALLTHRU;
>    single_succ_edge (chk)->flags |= neg_true_false_flag;
> +  single_succ_edge (chk)->probability = profile_probability::always ();
>    edge e = make_edge (chk, trp, true_false_flag);
>    e->goto_locus = loc;
> +  e->probability = profile_probability::never ();
>
>    if (dom_info_available_p (CDI_DOMINATORS))
>      set_immediate_dominator (CDI_DOMINATORS, trp, chk);
> diff --git a/gcc/testsuite/gcc.dg/pr105455.c b/gcc/testsuite/gcc.dg/pr105455.c
> new file mode 100644
> index 0000000000000..81e9154baa1c3
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/pr105455.c
> @@ -0,0 +1,17 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O1 -fharden-conditional-branches -funroll-loops --param max-loop-header-insns=1" } */
> +
> +__attribute__ ((cold)) void
> +bar (void);
> +
> +void
> +foo (int x)
> +{
> +  if (x)
> +    {
> +      int i;
> +
> +      for (i = 0; i < 101; ++i)
> +        bar ();
> +    }
> +}
>
>
> --
> 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: [PATCH] [PR105455] Set edge probabilities when hardening conditionals

[Alexandre Oliva]

On May 12, 2022, Richard Biener <richard.guenther@gmail.com> wrote:

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

> OK.

Thanks.  I failed to ask, but I'm confident this is also ok for gcc-12,
so I'm putting it in there as well.  Please let me know otherwise.

>> PR rtl-optimization/105455
>> * gimple-harden-conditionals.cc (insert_check_and_trap): Set
>> probabilities for newly-conditional edges.

-- 
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: [PATCH] [PR105455] Set edge probabilities when hardening conditionals

[Richard Biener]

On Fri, May 13, 2022 at 8:03 AM Alexandre Oliva <oliva@adacore.com> wrote:
>
> On May 12, 2022, Richard Biener <richard.guenther@gmail.com> wrote:
>
> >> Regstrapped on x86_64-linux-gnu.  Ok to install?
>
> > OK.
>
> Thanks.  I failed to ask, but I'm confident this is also ok for gcc-12,
> so I'm putting it in there as well.  Please let me know otherwise.

Yes, sure.

> >> PR rtl-optimization/105455
> >> * gimple-harden-conditionals.cc (insert_check_and_trap): Set
> >> probabilities for newly-conditional edges.
>
> --
> 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>