[Bug middle-end/41365] New: gcc.dg/vect/vect-cond-[123].c abort due to bad code generation at -O1 and above

[Bug middle-end/41365] New: gcc.dg/vect/vect-cond-[123].c abort due to bad code generation at -O1 and above

[sje at cup dot hp dot com]

The gcc.dg/vect/vect-cond-[123].c tests started failing on IA64 Linux and HP-UX
starting with version 5150588, Bernd Schmidt's auto-increment changes.  The
tests in question don't actually get vectorized and the bug can be reproduced
using just -O1 or -O2 optimization.  (-mlp64 is also needed on HP-UX)  I will
attach a slightly cutdown version of vect-cond-1.c that will abort when
compiled with -O1 or -O2 but not with -O0.


-- 
           Summary: gcc.dg/vect/vect-cond-[123].c abort due to bad code
                    generation at -O1 and above
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: sje at cup dot hp dot com
 GCC build triplet: ia64-*-*
  GCC host triplet: ia64-*-*
GCC target triplet: ia64-*-*


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41365

[Bug middle-end/41365] gcc.dg/vect/vect-cond-[123].c abort due to bad code generation at -O1 and above

[sje at cup dot hp dot com]

------- Comment #1 from sje at cup dot hp dot com  2009-09-15 23:40 -------
Created an attachment (id=18591)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=18591&action=view)
test case that aborts when compiled with -O1 or -O2


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41365

[Bug middle-end/41365] [4.5 Regression] gcc.dg/vect/vect-cond-[123].c abort due to bad code generation at -O1 and above

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|gcc.dg/vect/vect-cond-      |[4.5 Regression]
                   |[123].c abort due to bad    |gcc.dg/vect/vect-cond-
                   |code generation at -O1 and  |[123].c abort due to bad
                   |above                       |code generation at -O1 and
                   |                            |above
   Target Milestone|---                         |4.5.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41365

[Bug middle-end/41365] [4.5 Regression] gcc.dg/vect/vect-cond-[123].c abort due to bad code generation at -O1 and above

[bernds_cb1 at t-online dot de]

------- Comment #2 from bernds_cb1 at t-online dot de  2009-09-17 14:24 -------
Created an attachment (id=18601)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=18601&action=view)
A patch that fixes the immediate problem

This is a bug in the ia64 backend, which puts autoinc addressing modes into
COND_EXECs, failing to notice that they have side effects.

This part of the problem can be fixed with the patch I'm attaching, but it
doesn't work: the rest of the ia64 backend expects that this kind of insn can
always be split.

At this point I'm leaving it for an ia64 maintainer to fix.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41365

[Bug target/41365] [4.5 Regression] gcc.dg/vect/vect-cond-[123].c abort due to bad code generation at -O1 and above

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |target
           Priority|P3                          |P2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41365

[Bug target/41365] [4.5 Regression] gcc.dg/vect/vect-cond-[123].c abort due to bad code generation at -O1 and above

[sje at cup dot hp dot com]

-- 

sje at cup dot hp dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |sje at cup dot hp dot com
                   |dot org                     |
             Status|UNCONFIRMED                 |ASSIGNED
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2009-09-22 23:13:08
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41365

[Bug target/41365] [4.5 Regression] gcc.dg/vect/vect-cond-[123].c abort due to bad code generation at -O1 and above

[sje at gcc dot gnu dot org]

------- Comment #3 from sje at gcc dot gnu dot org  2009-09-29 22:21 -------
Subject: Bug 41365

Author: sje
Date: Tue Sep 29 22:21:27 2009
New Revision: 152306

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=152306
Log:
2009-09-29  Steve Ellcey  <sje@cup.hp.com>
            Alexander Monakov  <amonakov@ispras.ru>

        PR target/41365 
        * config/ia64/predicates.md (not_postinc_destination_operand): New.
        (not_postinc_memory_operand): New.
        (not_postinc_move_operand): New.
        * config/ia64/ia64.md (*cmovdi_internal): Disallow autoincrement.
        (*cmovsi_internal): Ditto.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/ia64/ia64.md
    trunk/gcc/config/ia64/predicates.md


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41365

[Bug target/41365] [4.5 Regression] gcc.dg/vect/vect-cond-[123].c abort due to bad code generation at -O1 and above

[sje at cup dot hp dot com]

------- Comment #4 from sje at cup dot hp dot com  2009-09-29 22:36 -------
Fixed.


-- 

sje at cup dot hp dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41365