[Bug rtl-optimization/23837] New: [4.0, 4.1 regression] Wrong code with -fschedule-insns

[Bug rtl-optimization/23837] New: [4.0, 4.1 regression] Wrong code with -fschedule-insns

[debian-gcc at lists dot debian dot org]

[forwarded from http://bugs.debian.org/326026]

falk@paer:~% cat test.c
void abort(void);

unsigned long long f(unsigned long long x) {
    return ((x >> 8) | (x << 56)) ^ ((x >> 48) | (x << 16)) ^ (x >> 1);
}

int main() {
    volatile unsigned long long v = 0x1122334455667788ULL;
    if (f(v) != 0xb3c46ef7196e4c91ULL)
        abort();
    return 0;
}

falk@paer:~% gcc-4.0 -O1 -fno-schedule-insns test.c && ./a.out
falk@paer:~% gcc-4.0 -O1 -fschedule-insns test.c && ./a.out   
zsh: abort (core dumped)  ./a.out

Reproduced with 4.0.2 20050821 on hppa-linux and with 4.1.0 20050705 on
i686-linux.

-- 
           Summary: [4.0, 4.1 regression] Wrong code with -fschedule-insns
           Product: gcc
           Version: 4.0.2
            Status: UNCONFIRMED
          Severity: critical
          Priority: P2
         Component: rtl-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: debian-gcc at lists dot debian dot org
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-09-12 19:05 -------
Confirmed but works with "4.0.0 20050225".

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
           Keywords|                            |wrong-code
   Last reconfirmed|0000-00-00 00:00:00         |2005-09-12 19:05:03
               date|                            |
            Summary|[4.0, 4.1 regression] Wrong |[4.0/4.1 regression] Wrong
                   |code with -fschedule-insns  |code with -fschedule-insns
   Target Milestone|---                         |4.0.2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[jakub at gcc dot gnu dot org]

------- Additional Comments From jakub at gcc dot gnu dot org  2005-09-13 09:04 -------
At least on i?86, this seems to be introduced by reload, .lreg dump is still
sane.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[janis187 at us dot ibm dot com]

------- Additional Comments From janis187 at us dot ibm dot com  2005-09-19 19:42 -------
A regression hunt on i686-linux showed the failure starting with this patch
from amylaar@gcc.gnu.org:

  http://gcc.gnu.org/ml/gcc-cvs/2005-05/msg00640.html

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |amylaar at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-09-19 21:04 -------
(In reply to comment #3)
> A regression hunt on i686-linux showed the failure starting with this patch
> from amylaar@gcc.gnu.org:
> 
>   http://gcc.gnu.org/ml/gcc-cvs/2005-05/msg00640.html

If the lreg dump is still sane that indicates that the problem was not
caused, by the patch, merely a latent bug was triggered.  emit_no_conflict_block
should not be called during reload.  to double-check, you can set a breakpoint
on emit_no_conflict_block when reload starts.

However, while looking for possible connections, I found that emit_libcall_block
has a similar bug as emit_no_conflict_block used to have, i.e. it ignores the
result of SETs if an insn has more than one.

FWIW, emit_libcall_block is used in i386.c in legitimize_tls_address, which in
turn is used in ix86_expand_move, which is used by the i386.md move expanders,
which are used by emit_move_insn_1, which is used by gen_move_insn, which
is used by gen_reload.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-09-20 19:06 -------
Confirmed with gcc version 4.1.0 20050919 (experimental) on i686-pc-linux-gnu.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2005-09-12 19:05:03         |2005-09-20 19:06:43
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[amylaar at gcc dot gnu dot org]

------- Additional Comments From amylaar at gcc dot gnu dot org  2005-09-20 20:52 -------
(In reply to comment #3)
> A regression hunt on i686-linux showed the failure starting with this patch
> from amylaar@gcc.gnu.org:
> 
>   http://gcc.gnu.org/ml/gcc-cvs/2005-05/msg00640.html

The i386.md x86_shrd_1 pattern is bogus, is should use lshiftrt,
not ashiftrt.  However, that is not the reason for this regression.

The reason is that we have a bogus REG_NO_CONFLICT block.  The generic
rotate expander expects emit_no_conflict_block to re-order its instructions,
but that is now suppresed by a match between clobbers of the flags register.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[joern dot rennecke at st dot com]

------- Additional Comments From joern dot rennecke at st dot com  2005-09-22 17:37 -------
Subject: RFA: fix PR 23837 (Re:  [4.0/4.1 regression] Wrong code with -fschedule-insns)

I have regtested the attached patch on i686-pc-linux-gnu in mainline
from 2005-09-19 18:00 UTC.
2005-09-22  J"orn Rennecke <joern.rennecke@st.com>

	PR rtl-optimization/23837
	*  optabs.c (no_conflict_move_test): Don't set must_stay for a
	clobber / clobber match between dest and p->first.

Index: optabs.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/optabs.c,v
retrieving revision 1.291
diff -p -r1.291 optabs.c
*** optabs.c	15 Sep 2005 21:51:13 -0000	1.291
--- optabs.c	22 Sep 2005 17:31:13 -0000
*************** no_conflict_move_test (rtx dest, rtx set
*** 3021,3027 ****
      return;
    /* If this insn sets / clobbers a register that feeds one of the insns
       already in the list, this insn has to stay too.  */
!   else if (reg_mentioned_p (dest, PATTERN (p->first))
  	   || reg_used_between_p (dest, p->first, p->insn)
  	   /* Likewise if this insn depends on a register set by a previous
  	      insn in the list.  */
--- 3021,3028 ----
      return;
    /* If this insn sets / clobbers a register that feeds one of the insns
       already in the list, this insn has to stay too.  */
!   else if (reg_overlap_mentioned_p (dest, PATTERN (p->first))
! 	   || (CALL_P (p->first) && (find_reg_fusage (p->first, USE, dest)))
  	   || reg_used_between_p (dest, p->first, p->insn)
  	   /* Likewise if this insn depends on a register set by a previous
  	      insn in the list.  */


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-09-23 12:06 -------
Subject: Bug 23837

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	amylaar@gcc.gnu.org	2005-09-23 12:05:58

Modified files:
	gcc            : ChangeLog optabs.c 

Log message:
	PR rtl-optimization/23837
	*  optabs.c (no_conflict_move_test): Don't set must_stay for a
	clobber / clobber match between dest and p->first.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.10007&r2=2.10008
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/optabs.c.diff?cvsroot=gcc&r1=1.291&r2=1.292



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[mmitchel at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.0.2                       |4.0.3


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[pinskia at gcc dot gnu dot org]

------- Comment #14 from pinskia at gcc dot gnu dot org  2005-11-19 01:28 -------
This also effects hppa2.0w-hp-hpux11.11 which means this effects a primary
target.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mmitchel at gcc dot gnu dot
                   |                            |org
 GCC target triplet|hppa-linux                  |hppa2.0w-hp-hpux11.11, hppa-
                   |                            |linux
           Priority|P5                          |P3


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[steven at gcc dot gnu dot org]

-- 

steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW
   Last reconfirmed|2005-09-20 19:06:43         |2005-11-06 17:47:35
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[danglin at gcc dot gnu dot org]

------- Comment #13 from danglin at gcc dot gnu dot org  2005-11-06 17:32 -------
Fails on hppa2.0w-hp-hpux11.11 with 4.0.0, 4.0.1 and 4.0.2.  However, it
doesn't fail using 3.4.4.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[steven at gcc dot gnu dot org]

------- Comment #12 from steven at gcc dot gnu dot org  2005-11-06 11:50 -------
Folks, what's up with this bug?  Is it HPPA-only now?  Can someone comment on
the status of this bug please?


-- 

steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |danglin at gcc dot gnu dot
                   |                            |org, steven at gcc dot gnu
                   |                            |dot org
             Status|NEW                         |WAITING


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[pinskia at gcc dot gnu dot org]

-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P2                          |P5
   Target Milestone|4.2.0                       |4.1.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[pinskia at gcc dot gnu dot org]

-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |critical


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[pinskia at gcc dot gnu dot org]

-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.1.0                       |4.2.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[steven at gcc dot gnu dot org]

------- Comment #11 from steven at gcc dot gnu dot org  2005-10-21 11:21 -------
How do you know this is hppa-linux only now?  There's a bit if information
missing about how you got to that conclusion.

Is there some simple way to test this bug on HPPA? (Maybe add a HPPA 
maintainer to the CC: list? ;-)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837

[Bug rtl-optimization/23837] [4.0/4.1 regression] Wrong code with -fschedule-insns

[pinskia at gcc dot gnu dot org]

------- Comment #10 from pinskia at gcc dot gnu dot org  2005-10-10 04:29 -------
Only a hppa-linux bug now, which means this should be targetted at 4.2.0.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |normal
 GCC target triplet|                            |hppa-linux
   Target Milestone|4.0.3                       |4.1.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23837