On 01/12/2018 09:44 AM, Jan Hubicka wrote:
Hi.
Following patch adds new sanitization checks for profile_quality.
Problem is that zero initialization of a struct with profile_count will
lead to an invalid counter. This can help to catch them.
Patch can bootstrap on ppc64le-redhat-linux and survives regression tests.
Ready to be installed?
OK,
thanks!
Honza
Martin
>From edec114cf1dd29bb571855a80e1b45ae040da200 Mon Sep 17 00:00:00 2001
From: marxin <mli...@suse.cz>
Date: Wed, 10 Jan 2018 14:46:08 +0100
Subject: [PATCH] Add new verification for profile-count.h.
gcc/ChangeLog:
2018-01-12 Martin Liska <mli...@suse.cz>
* profile-count.h (enum profile_quality): Use 0 as invalid
enum value of profile_quality.
---
gcc/profile-count.h | 16 ++++++++++------
1 file changed, 10 insertions(+), 6 deletions(-)
diff --git a/gcc/profile-count.h b/gcc/profile-count.h
index 3c5f720ee81..7a43917ebbc 100644
--- a/gcc/profile-count.h
+++ b/gcc/profile-count.h
@@ -30,27 +30,27 @@ enum profile_quality {
or may not match reality. It is local to function and can not be
compared
inter-procedurally. Never used by probabilities (they are always local).
*/
- profile_guessed_local = 0,
+ profile_guessed_local = 1,
/* Profile was read by feedback and was 0, we used local heuristics to guess
better. This is the case of functions not run in profile fedback.
Never used by probabilities. */
- profile_guessed_global0 = 1,
+ profile_guessed_global0 = 2,
/* Same as profile_guessed_global0 but global count is adjusted 0. */
- profile_guessed_global0adjusted = 2,
+ profile_guessed_global0adjusted = 3,
/* Profile is based on static branch prediction heuristics. It may or may
not reflect the reality but it can be compared interprocedurally
(for example, we inlined function w/o profile feedback into function
with feedback and propagated from that).
Never used by probablities. */
- profile_guessed = 3,
+ profile_guessed = 4,
/* Profile was determined by autofdo. */
- profile_afdo = 4,
+ profile_afdo = 5,
/* Profile was originally based on feedback but it was adjusted
by code duplicating optimization. It may not precisely reflect the
particular code path. */
- profile_adjusted = 5,
+ profile_adjusted = 6,
/* Profile was read from profile feedback or determined by accurate static
method. */
profile_precise = 7
@@ -505,6 +505,8 @@ public:
/* Return false if profile_probability is bogus. */
bool verify () const
{
+ gcc_checking_assert (profile_guessed_local <= m_quality
+ && m_quality <= profile_precise);
Hi,
FYI, in a no-bootstrap build, I'm seeing a lot of new warnings like this:
...
../../src/gcc/profile-count.h: In member function ‘bool
profile_probability::verify() const’:
../../src/gcc/profile-count.h:509:20: warning: comparison is always true
due to limited range of data type [-Wtype-limits]
&& m_quality <= profile_precise);
^
../../src/gcc/system.h:742:14: note: in definition of macro ‘gcc_assert’
((void)(!(EXPR) ? fancy_abort (__FILE__, __LINE__, __FUNCTION__), 0
: 0))
^
../../src/gcc/profile-count.h:508:7: note: in expansion of macro
‘gcc_checking_assert’
gcc_checking_assert (profile_guessed_local <= m_quality
...
Indeed, profile_precise is 7 and m_quality is a 3 bits wide bitfield.
Thanks,
- Tom
if (m_val == uninitialized_probability)
return m_quality == profile_guessed;
else if (m_quality < profile_guessed)
@@ -784,6 +786,8 @@ public:
/* Return false if profile_count is bogus. */
bool verify () const
{
+ gcc_checking_assert (profile_guessed_local <= m_quality
+ && m_quality <= profile_precise);
return m_val != uninitialized_count || m_quality ==
profile_guessed_local;
}
--
2.14.3
