From: Jan Hubicka <hubicka@ucw.cz>
To: Andreas Schwab <schwab@suse.de>
Cc: gcc-patches@gcc.gnu.org
Subject: Re: Add quality tracking for profile counter
Date: Thu, 22 Jun 2017 14:51:00 -0000 [thread overview]
Message-ID: <20170622145136.GB94431@kam.mff.cuni.cz> (raw)
In-Reply-To: <mvmk245bxg3.fsf@suse.de>
Hi,
I proofread the code and noticed that in some cases I may trigger division by 0 that may
get different outputs depending on optimization setting on Itanium.
This is what I comitted after profiledbootstrap and regtesting at x86_64.
* profile-count.h (apply_probability,
apply_scale, probability_in): Fix checks for zero.
Index: profile-count.h
===================================================================
--- profile-count.h (revision 249364)
+++ profile-count.h (working copy)
@@ -255,7 +255,7 @@ public:
profile_count apply_probability (int prob) const
{
gcc_checking_assert (prob >= 0 && prob <= REG_BR_PROB_BASE);
- if (*this == profile_count::zero ())
+ if (m_val == 0)
return *this;
if (!initialized_p ())
return profile_count::uninitialized ();
@@ -267,24 +267,25 @@ public:
/* Return *THIS * NUM / DEN. */
profile_count apply_scale (int64_t num, int64_t den) const
{
- if (*this == profile_count::zero ())
+ if (m_val == 0)
return *this;
if (!initialized_p ())
return profile_count::uninitialized ();
profile_count ret;
+ gcc_checking_assert (num >= 0 && den > 0);
/* FIXME: shrink wrapping violates this sanity check. */
- gcc_checking_assert ((num >= 0
- && (num <= REG_BR_PROB_BASE
- || den <= REG_BR_PROB_BASE)
- && den > 0) || 1);
+ gcc_checking_assert ((num <= REG_BR_PROB_BASE
+ || den <= REG_BR_PROB_BASE) || 1);
ret.m_val = RDIV (m_val * num, den);
ret.m_quality = MIN (m_quality, count_adjusted);
return ret;
}
profile_count apply_scale (profile_count num, profile_count den) const
{
- if (*this == profile_count::zero () || num == profile_count::zero ())
- return profile_count::zero ();
+ if (m_val == 0)
+ return *this;
+ if (num.m_val == 0)
+ return num;
if (!initialized_p () || !num.initialized_p () || !den.initialized_p ())
return profile_count::uninitialized ();
gcc_checking_assert (den > 0);
@@ -306,7 +307,7 @@ public:
OVERALL. */
int probability_in (profile_count overall)
{
- if (*this == profile_count::zero ())
+ if (!m_val)
return 0;
if (!initialized_p () || !overall.initialized_p ())
return REG_BR_PROB_BASE / 2;
prev parent reply other threads:[~2017-06-22 14:51 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-06-19 9:49 Jan Hubicka
2017-06-21 7:40 ` Andreas Schwab
2017-06-22 14:51 ` Jan Hubicka [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=20170622145136.GB94431@kam.mff.cuni.cz \
--to=hubicka@ucw.cz \
--cc=gcc-patches@gcc.gnu.org \
--cc=schwab@suse.de \
/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).