From: Jan Hubicka <hubicka@kam.mff.cuni.cz>
To: Jeff Law <jeffreyalaw@gmail.com>
Cc: "H.J. Lu" <hjl.tools@gmail.com>,
GCC Patches <gcc-patches@gcc.gnu.org>,
Richard Biener <rguenther@suse.de>
Subject: Re: Basic kill analysis for modref
Date: Mon, 15 Nov 2021 21:56:17 +0100 [thread overview]
Message-ID: <20211115205617.GA72251@kam.mff.cuni.cz> (raw)
In-Reply-To: <4eb689bc-e5b7-0ef9-72b0-1084d8700dcd@gmail.com>
> > > + if (always_executed
> > > + && callee_summary->kills.length ()
> > > + && (!cfun->can_throw_non_call_exceptions
> > > + || !stmt_could_throw_p (cfun, stmt)))
> > > + {
> > > + /* Watch for self recursive updates. */
> > > + auto_vec<modref_access_node, 32> saved_kills;
> > > +
> > > + saved_kills.reserve_exact (callee_summary->kills.length ());
> > > + saved_kills.splice (callee_summary->kills);
> > > + for (auto kill : saved_kills)
> > > + {
> > > + if (kill.parm_index >= (int)parm_map.length ())
> > > + continue;
> > > + modref_parm_map &m
> > > + = kill.parm_index == MODREF_STATIC_CHAIN_PARM
> > > + ? chain_map
> > ^^^^^^^^^^^^^^^^ chain_map isn't initialized.
> >
> > This caused:
> >
> > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103262
> Yup. It's causing heartburn in various ways in the tester. I was just
> tracking it down with valgrind...
> jeff
Oops, either me or patch much have mislocated the change within the
function when updating to new tree. I am testing the following fix and
will cook up a testcase verifying that merging of kills works as
expected.
Thanks!
Honza
diff --git a/gcc/ipa-modref.c b/gcc/ipa-modref.c
index df4612bbff9..4784f68f585 100644
--- a/gcc/ipa-modref.c
+++ b/gcc/ipa-modref.c
@@ -964,38 +980,6 @@ merge_call_side_effects (modref_summary *cur_summary,
if (flags & (ECF_CONST | ECF_NOVOPS))
return changed;
- if (always_executed
- && callee_summary->kills.length ()
- && (!cfun->can_throw_non_call_exceptions
- || !stmt_could_throw_p (cfun, stmt)))
- {
- /* Watch for self recursive updates. */
- auto_vec<modref_access_node, 32> saved_kills;
-
- saved_kills.reserve_exact (callee_summary->kills.length ());
- saved_kills.splice (callee_summary->kills);
- for (auto kill : saved_kills)
- {
- if (kill.parm_index >= (int)parm_map.length ())
- continue;
- modref_parm_map &m
- = kill.parm_index == MODREF_STATIC_CHAIN_PARM
- ? chain_map
- : parm_map[kill.parm_index];
- if (m.parm_index == MODREF_LOCAL_MEMORY_PARM
- || m.parm_index == MODREF_UNKNOWN_PARM
- || m.parm_index == MODREF_RETSLOT_PARM
- || !m.parm_offset_known)
- continue;
- modref_access_node n = kill;
- n.parm_index = m.parm_index;
- n.parm_offset += m.parm_offset;
- if (modref_access_node::insert_kill (cur_summary->kills, n,
- record_adjustments))
- changed = true;
- }
- }
-
/* We can not safely optimize based on summary of callee if it does
not always bind to current def: it is possible that memory load
was optimized out earlier which may not happen in the interposed
@@ -1043,6 +1027,38 @@ merge_call_side_effects (modref_summary *cur_summary,
if (dump_file)
fprintf (dump_file, "\n");
+ if (always_executed
+ && callee_summary->kills.length ()
+ && (!cfun->can_throw_non_call_exceptions
+ || !stmt_could_throw_p (cfun, stmt)))
+ {
+ /* Watch for self recursive updates. */
+ auto_vec<modref_access_node, 32> saved_kills;
+
+ saved_kills.reserve_exact (callee_summary->kills.length ());
+ saved_kills.splice (callee_summary->kills);
+ for (auto kill : saved_kills)
+ {
+ if (kill.parm_index >= (int)parm_map.length ())
+ continue;
+ modref_parm_map &m
+ = kill.parm_index == MODREF_STATIC_CHAIN_PARM
+ ? chain_map
+ : parm_map[kill.parm_index];
+ if (m.parm_index == MODREF_LOCAL_MEMORY_PARM
+ || m.parm_index == MODREF_UNKNOWN_PARM
+ || m.parm_index == MODREF_RETSLOT_PARM
+ || !m.parm_offset_known)
+ continue;
+ modref_access_node n = kill;
+ n.parm_index = m.parm_index;
+ n.parm_offset += m.parm_offset;
+ if (modref_access_node::insert_kill (cur_summary->kills, n,
+ record_adjustments))
+ changed = true;
+ }
+ }
+
/* Merge with callee's summary. */
changed |= cur_summary->loads->merge (callee_summary->loads, &parm_map,
&chain_map, record_adjustments);
next prev parent reply other threads:[~2021-11-15 20:56 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-11-11 12:58 Jan Hubicka
2021-11-12 10:38 ` Richard Biener
2021-11-12 10:47 ` Jan Hubicka
2021-11-12 11:00 ` Richard Biener
2021-11-14 18:53 ` Jan Hubicka
2021-11-15 18:51 ` H.J. Lu
2021-11-15 19:00 ` Jeff Law
2021-11-15 20:56 ` Jan Hubicka [this message]
2021-11-16 8:19 ` Jan Hubicka
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=20211115205617.GA72251@kam.mff.cuni.cz \
--to=hubicka@kam.mff.cuni.cz \
--cc=gcc-patches@gcc.gnu.org \
--cc=hjl.tools@gmail.com \
--cc=jeffreyalaw@gmail.com \
--cc=rguenther@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).