[Bug tree-optimization/53385] New: "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[Bug tree-optimization/53385] New: "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[wschmidt at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

             Bug #: 53385
           Summary: "Error: operand out of range" after changes for
                    LSHIFT_EXPR in vrp.c
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: wschmidt@gcc.gnu.org
                CC: bergner@vnet.ibm.com, meissner@gcc.gnu.org,
                    pthaugen@us.ibm.com, rguenther@suse.de
              Host: powerpc64-linux
            Target: powerpc64-linux
             Build: powerpc64-linux


Starting with revision 187535, bootstrap is broken on
powerpc64-unknown-linux-gnu:

/home/wschmidt/gcc/build/gcc-mainline-base/./gcc/xgcc
-B/home/wschmidt/gcc/build/gcc-mainline-base/./gcc/
-B/home/wschmidt/gcc/install/gcc-mainline-base/powerpc64-linux/bin/
-B/home/wschmidt/gcc/install/gcc-mainline-base/powerpc64-linux/lib/ -isystem
/home/wschmidt/gcc/install/gcc-mainline-base/powerpc64-linux/include -isystem
/home/wschmidt/gcc/install/gcc-mainline-base/powerpc64-linux/sys-include    -g
-O2 -O2  -g -O2 -DIN_GCC   -W -Wall -Wno-narrowing -Wwrite-strings -Wcast-qual
-Wstrict-prototypes -Wmissing-prototypes -Wold-style-definition  -isystem
./include   -fPIC -mlong-double-128 -mno-minimal-toc -g -DIN_LIBGCC2
-fbuilding-libgcc -fno-stack-protector   -fPIC -mlong-double-128
-mno-minimal-toc -I. -I. -I../.././gcc
-I/home/wschmidt/gcc/gcc-mainline-base/libgcc
-I/home/wschmidt/gcc/gcc-mainline-base/libgcc/.
-I/home/wschmidt/gcc/gcc-mainline-base/libgcc/../gcc
-I/home/wschmidt/gcc/gcc-mainline-base/libgcc/../include
-I/home/wschmidt/gcc/gcc-mainline-base/libgcc/../libdecnumber/dpd
-I/home/wschmidt/gcc/gcc-mainline-base/libgcc/../libdecnumber -DHAVE_CC_TLS  -o
_gcov_interval_profiler.o -MT _gcov_interval_profiler.o -MD -MP -MF
_gcov_interval_profiler.dep -DL_gcov_interval_profiler -c
/home/wschmidt/gcc/gcc-mainline-base/libgcc/libgcov.c

/tmp/ccrZ0fWL.s: Assembler messages:
/tmp/ccrZ0fWL.s:49: Error: operand out of range (65 is not between 0 and 63)

*** [_gcov_interval_profiler.o] Error 1
make[3]: *** Waiting for unfinished jobs....

An example of the problem is:

long unsigned foo (long unsigned a) { return (a << 2) & 0x3fffffffc; }

which produces "rldic 3,3,2,65" instead of "rldic 3,3,2,30".

[Bug target/53385] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|tree-optimization           |target

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-05-17 02:08:05 UTC ---
The target should have done a better job on its constraints.

[Bug target/53385] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[dje at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

David Edelsohn <dje at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-05-17
     Ever Confirmed|0                           |1

--- Comment #2 from David Edelsohn <dje at gcc dot gnu.org> 2012-05-17 02:39:19 UTC ---
Definitely caused by this revision.

[Bug target/53385] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[amodra at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

Alan Modra <amodra at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |amodra at gmail dot com

--- Comment #3 from Alan Modra <amodra at gmail dot com> 2012-05-17 12:04:52 UTC ---
A data point.  My stage1 gcc compiles the testcase in comment #1 OK, stage2 gcc
gives the "operand out of range" error.

[Bug target/53385] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[amodra at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

--- Comment #4 from Alan Modra <amodra at gmail dot com> 2012-05-17 12:25:03 UTC ---
Here's the problem.  Compiled with -m64 -O2, this

int f (long val)
{
  int i;

  if (val < 0)
    i = -1;
  else
    for (i = 0; i < 64; i++)
      if ((val <<= 1) < 0)
        break;

  return i;
}

is miscompiled to

         sradi 3,3,63
        ori 3,3,64
        blr

[Bug tree-optimization/53385] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[amodra at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

Alan Modra <amodra at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |tree-optimization

--- Comment #5 from Alan Modra <amodra at gmail dot com> 2012-05-17 12:38:51 UTC ---
Changing back to tree-optimization. pinski, this time you don't know what
you're doing. ;-)

[Bug tree-optimization/53385] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[glisse at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

Marc Glisse <glisse at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |glisse at gcc dot gnu.org

--- Comment #6 from Marc Glisse <glisse at gcc dot gnu.org> 2012-05-17 13:00:34 UTC ---
(In reply to comment #4)
> Here's the problem.  Compiled with -m64 -O2, this
> 
> int f (long val)
> {
>   int i;
> 
>   if (val < 0)
>     i = -1;
>   else
>     for (i = 0; i < 64; i++)
>       if ((val <<= 1) < 0)
>         break;
> 
>   return i;
> }

Is that just a -fwrapv issue?

[Bug tree-optimization/53385] [4.8 Regression] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[dje at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

David Edelsohn <dje at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
   Target Milestone|---                         |4.8.0
            Summary|"Error: operand out of      |[4.8 Regression] "Error:
                   |range" after changes for    |operand out of range" after
                   |LSHIFT_EXPR in vrp.c        |changes for LSHIFT_EXPR in
                   |                            |vrp.c

[Bug tree-optimization/53385] [4.8 Regression] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[glisse at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

--- Comment #7 from Marc Glisse <glisse at gcc dot gnu.org> 2012-05-17 21:34:23 UTC ---
Created attachment 27429
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27429
patch

I believe you need something like this completely untested patch to avoid
undefined behavior without -fwrapv.

[Bug tree-optimization/53385] [4.8 Regression] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[wschmidt at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

--- Comment #8 from William J. Schmidt <wschmidt at gcc dot gnu.org> 2012-05-18 03:42:46 UTC ---
I suspect you're right.  I'll have a look at it tomorrow.

[Bug tree-optimization/53385] [4.8 Regression] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-05-18 08:17:42 UTC ---
Isn't this just a fancy way of saying
  i = val ? clz_hwi (val) - 1 : HOST_BITS_PER_WIDE_INT;
?  Note that the predicate ensures that the whole operand isn't 0,
so one might as well use just i = clzi_hwi (val) - 1; and just adjust the
else if (GET_CODE (x) == CONST_DOUBLE && i == 32)
test to compare i == 31 instead (and tweak the body not to i-- if the low hwi
is negative and do another clz_hwi otherwise).

[Bug tree-optimization/53385] [4.8 Regression] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-05-18 08:24:40 UTC ---
So I think you could do:
    case 'W':
      /* MB value for a PowerPC64 rldic operand.  */
      i = clz_hwi (GET_CODE (x) == CONST_INT
                   ? INTVAL (x) : CONST_DOUBLE_HIGH (x));

#if HOST_BITS_PER_WIDE_INT == 32
      if (GET_CODE (x) == CONST_INT && i)
        i += 32;  /* zero-extend high-part was all 0's */
      else if (GET_CODE (x) == CONST_DOUBLE && i == 32)
        i = clz_hwi (CONST_DOUBLE_LOW (x)) + 32;
#endif

      fprintf (file, "%d", i);
      return;

[Bug target/53385] [4.8 Regression] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

Richard Guenther <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|tree-optimization           |target

--- Comment #11 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-05-18 09:45:31 UTC ---
Target issue.

[Bug target/53385] [4.8 Regression] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[wschmidt at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

--- Comment #12 from William J. Schmidt <wschmidt at gcc dot gnu.org> 2012-05-18 14:05:06 UTC ---
Created attachment 27433
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27433
Tentative patch

Tentative patch following Jakub's suggestion.  This fixes bootstrap.  Pending
approval.

[Bug target/53385] [4.8 Regression] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[wschmidt at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

William J. Schmidt <wschmidt at gcc dot gnu.org> changed:

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

--- Comment #13 from William J. Schmidt <wschmidt at gcc dot gnu.org> 2012-05-18 16:06:16 UTC ---
Fixed by r187657.

[Bug target/53385] [4.8 Regression] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[wschmidt at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

--- Comment #14 from William J. Schmidt <wschmidt at gcc dot gnu.org> 2012-05-24 13:53:03 UTC ---
Author: wschmidt
Date: Thu May 24 13:52:56 2012
New Revision: 187834

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=187834
Log:
2012-05-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

    Backport from mainline
        2012-05-18  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

    PR target/53385
    * config/rs6000/rs6000.c (print_operand): Revise code that unsafely
    relied on signed overflow behavior.

Modified:
    branches/gcc-4_6-branch/gcc/ChangeLog
    branches/gcc-4_6-branch/gcc/config/rs6000/rs6000.c

[Bug target/53385] [4.8 Regression] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[wschmidt at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

--- Comment #14 from William J. Schmidt <wschmidt at gcc dot gnu.org> 2012-05-24 13:53:03 UTC ---
Author: wschmidt
Date: Thu May 24 13:52:56 2012
New Revision: 187834

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=187834
Log:
2012-05-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

    Backport from mainline
        2012-05-18  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

    PR target/53385
    * config/rs6000/rs6000.c (print_operand): Revise code that unsafely
    relied on signed overflow behavior.

Modified:
    branches/gcc-4_6-branch/gcc/ChangeLog
    branches/gcc-4_6-branch/gcc/config/rs6000/rs6000.c

--- Comment #15 from William J. Schmidt <wschmidt at gcc dot gnu.org> 2012-05-24 13:54:24 UTC ---
Author: wschmidt
Date: Thu May 24 13:54:16 2012
New Revision: 187835

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=187835
Log:
2012-05-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

    Backport from mainline
        2012-05-18  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

    PR target/53385
    * config/rs6000/rs6000.c (print_operand): Revise code that unsafely
    relied on signed overflow behavior.

Modified:
    branches/gcc-4_7-branch/gcc/ChangeLog
    branches/gcc-4_7-branch/gcc/config/rs6000/rs6000.c

[Bug target/53385] [4.8 Regression] "Error: operand out of range" after changes for LSHIFT_EXPR in vrp.c

[wschmidt at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53385

--- Comment #15 from William J. Schmidt <wschmidt at gcc dot gnu.org> 2012-05-24 13:54:24 UTC ---
Author: wschmidt
Date: Thu May 24 13:54:16 2012
New Revision: 187835

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=187835
Log:
2012-05-24  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

    Backport from mainline
        2012-05-18  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

    PR target/53385
    * config/rs6000/rs6000.c (print_operand): Revise code that unsafely
    relied on signed overflow behavior.

Modified:
    branches/gcc-4_7-branch/gcc/ChangeLog
    branches/gcc-4_7-branch/gcc/config/rs6000/rs6000.c