public inbox for gcc-bugs@sourceware.org
help / color / mirror / Atom feed
From: "hubicka at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug ipa/103246] [12 Regression] 416.gamess miscompare with -O2 -g -flto=auto since r12-5223-gecdf414bd89e6ba251f6b3f494407139b4dbae0e
Date: Wed, 17 Nov 2021 19:36:38 +0000 [thread overview]
Message-ID: <bug-103246-4-uPOvI1sUse@http.gcc.gnu.org/bugzilla/> (raw)
In-Reply-To: <bug-103246-4@http.gcc.gnu.org/bugzilla/>
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103246
--- Comment #20 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
linking with --dbg-cnt=ipa_cp_values:10 -flto-partition=one
--disable-tree-modref2 --dbg-cnt=ipa_mod_ref:836 -fdump-tree-all-details
-fdump-ipa-all-details is first that produces different results. Diff to 835
is:
diff -u 835/gamess.ltrans0.ltrans.113t.dse2 836/gamess.ltrans0.ltrans.113t.dse2
--- 835/gamess.ltrans0.ltrans.113t.dse2 2021-11-17 20:11:34.894715053 +0100
+++ 836/gamess.ltrans0.ltrans.113t.dse2 2021-11-17 20:12:40.387968675 +0100
@@ -14780,7 +14780,6 @@
ipa-modref: call to dco.isra/287 does not use ref: kz alias sets: 4->4
ipa-modref: call stmt dc_1560 = dco.isra (l_1973, m_1975, &kx, &ky, &kz,
la_1974, mu_1976, _513, _510, _507, _504, _501, _498);
ipa-modref: call to dco.isra/287 does not use ref: jndx alias sets: 4->4
-***dbgcnt: upper limit 835 reached for ipa_mod_ref.***
ipa-modref: call stmt dc_1560 = dco.isra (l_1973, m_1975, &kx, &ky, &kz,
la_1974, mu_1976, _513, _510, _507, _504, _501, _498);
ipa-modref: call to dco.isra/287 does not use ref: kz alias sets: 4->4
Deleted trivially dead stmt: _1548 = (long int) _1547;
@@ -14799,6 +14798,13 @@
Deleted trivially dead stmt: _1770 = indx_1668 + 1;
+***dbgcnt: upper limit 836 reached for ipa_mod_ref.***
+ipa-modref: call stmt dc_1710 = dco.isra (0, 0, &D.8524, &D.8523, &D.8522,
la_1977, mu_1978, _513, _510, _507, _504, _501, _498);
+ipa-modref: call to dco.isra/287 does not use ref: D.8522 alias sets: 4->4
+ Deleted dead store: D.8522 = _1709;
+
+ Deleted trivially dead stmt: _1709 = kz_1696 + kzp_1699;
+
At WPA time we first compute summary for original function and then update for
new signature:
Updating summary for dco.isra/287 from:
loads:
Limits: 32 bases, 16 refs
Base 0:int (alias set 4)
Ref 0:int (alias set 4)
access: Parm 0 param offset:0 offset:0 size:32 max_size:32
access: Parm 1 param offset:0 offset:0 size:32 max_size:32
access: Parm 5 param offset:0 offset:0 size:32 max_size:32
access: Parm 6 param offset:0 offset:0 size:32 max_size:32
access: Parm 2 param offset:0 offset:0 size:32 max_size:32
access: Parm 3 param offset:0 offset:0 size:32 max_size:32
access: Parm 4 param offset:0 offset:0 size:32 max_size:32
Base 1:int[0:] (alias set 4)
Ref 0:int (alias set 4)
access: Parm 9 param offset:0 offset:0 size:32 max_size:-1
access: Parm 10 param offset:0 offset:0 size:32 max_size:-1
access: Parm 11 param offset:0 offset:0 size:32 max_size:-1
access: Parm 12 param offset:0 offset:0 size:32 max_size:-1
Base 2:double[0:] (alias set 5)
Ref 0:double (alias set 5)
access: Parm 8 param offset:0 offset:0 size:64 max_size:-1
Base 3:double[15625] (alias set 5)
Ref 0:double (alias set 5)
access: Parm 7 param offset:0 offset:0 size:64 max_size:-1
stores:
Limits: 32 bases, 16 refs
parm 0 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 1 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 2 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 3 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 4 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 5 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 6 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 7 flags: not_returned_directly no_indirect_read
parm 8 flags: not_returned_directly no_indirect_read
parm 9 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 10 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 11 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 12 flags: not_returned_directly not_returned_indirectly no_indirect_read
to:
loads:
Limits: 32 bases, 16 refs
Base 0:int (alias set 4)
Ref 0:int (alias set 4)
access:
access:
access:
access:
access: Parm 2 param offset:0 offset:0 size:32 max_size:32
access: Parm 3 param offset:0 offset:0 size:32 max_size:32
access: Parm 4 param offset:0 offset:0 size:32 max_size:32
Base 1:int[0:] (alias set 4)
Ref 0:int (alias set 4)
access: Parm 9 param offset:0 offset:0 size:32 max_size:-1
access: Parm 10 param offset:0 offset:0 size:32 max_size:-1
access: Parm 11 param offset:0 offset:0 size:32 max_size:-1
access: Parm 12 param offset:0 offset:0 size:32 max_size:-1
Base 2:double[0:] (alias set 5)
Ref 0:double (alias set 5)
access: Parm 8 param offset:0 offset:0 size:64 max_size:-1
Base 3:double[15625] (alias set 5)
Ref 0:double (alias set 5)
access: Parm 7 param offset:0 offset:0 size:64 max_size:-1
stores:
Limits: 32 bases, 16 refs
parm 2 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 3 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 4 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 7 flags: not_returned_directly no_indirect_read
parm 8 flags: not_returned_directly no_indirect_read
parm 9 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 10 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 11 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 12 flags: not_returned_directly not_returned_indirectly no_indirect_read
This seems to correclty represent that some params was scalarised. Only one
oddity is that the list is not optimized and kept in redundant form but it
should be harmless.
At stream in time we see:
Read modref for dco.isra/287
loads:
Limits: 32 bases, 16 refs
Base 0: alias set 4
Ref 0: alias set 4
access: Parm 9 param offset:0 offset:0 size:32 max_size:-1
access: Parm 10 param offset:0 offset:0 size:32 max_size:-1
access: Parm 11 param offset:0 offset:0 size:32 max_size:-1
access: Parm 12 param offset:0 offset:0 size:32 max_size:-1
Base 1: alias set 3
Ref 0: alias set 3
access: Parm 8 param offset:0 offset:0 size:64 max_size:-1
access: Parm 7 param offset:0 offset:0 size:64 max_size:-1
stores:
Limits: 32 bases, 16 refs
Try dse
parm 2 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 3 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 4 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 7 flags: not_returned_directly no_indirect_read
parm 8 flags: not_returned_directly no_indirect_read
parm 9 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 10 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 11 flags: not_returned_directly not_returned_indirectly no_indirect_read
parm 12 flags: not_returned_directly not_returned_indirectly no_indirect_read
So types are translated to alias sets however the stream in is wrong since
there was accesses without known params and they are gone. So it is streaming
bug (and was present already in the initial modref version)
next prev parent reply other threads:[~2021-11-17 19:36 UTC|newest]
Thread overview: 27+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-11-15 10:22 [Bug ipa/103246] New: " marxin at gcc dot gnu.org
2021-11-15 10:24 ` [Bug ipa/103246] " marxin at gcc dot gnu.org
2021-11-15 10:39 ` marxin at gcc dot gnu.org
2021-11-15 10:39 ` marxin at gcc dot gnu.org
2021-11-15 10:40 ` marxin at gcc dot gnu.org
2021-11-15 10:40 ` marxin at gcc dot gnu.org
2021-11-15 11:13 ` hubicka at gcc dot gnu.org
2021-11-16 16:09 ` jamborm at gcc dot gnu.org
2021-11-16 17:09 ` jamborm at gcc dot gnu.org
2021-11-16 19:24 ` marxin at gcc dot gnu.org
2021-11-16 19:34 ` marxin at gcc dot gnu.org
2021-11-16 20:53 ` marxin at gcc dot gnu.org
2021-11-16 22:42 ` hubicka at gcc dot gnu.org
2021-11-17 0:44 ` cvs-commit at gcc dot gnu.org
2021-11-17 9:11 ` marxin at gcc dot gnu.org
2021-11-17 9:38 ` hubicka at kam dot mff.cuni.cz
2021-11-17 12:27 ` hubicka at gcc dot gnu.org
2021-11-17 13:03 ` hubicka at gcc dot gnu.org
2021-11-17 13:15 ` hubicka at gcc dot gnu.org
2021-11-17 13:31 ` hubicka at gcc dot gnu.org
2021-11-17 18:58 ` hubicka at gcc dot gnu.org
2021-11-17 19:36 ` hubicka at gcc dot gnu.org [this message]
2021-11-17 21:06 ` cvs-commit at gcc dot gnu.org
2021-11-18 8:27 ` hubicka at gcc dot gnu.org
2021-11-18 8:28 ` hubicka at gcc dot gnu.org
2021-11-18 12:22 ` marxin at gcc dot gnu.org
2021-11-18 19:10 ` cvs-commit at gcc dot gnu.org
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=bug-103246-4-uPOvI1sUse@http.gcc.gnu.org/bugzilla/ \
--to=gcc-bugzilla@gcc.gnu.org \
--cc=gcc-bugs@gcc.gnu.org \
/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).