[Bug target/49069] New: ICE in gen_cstoredi4, at config/arm/arm.md:7554

[Bug target/49069] New: ICE in gen_cstoredi4, at config/arm/arm.md:7554

[rmansfield at qnx dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

           Summary: ICE in gen_cstoredi4, at config/arm/arm.md:7554
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: rmansfield@qnx.com
              Host: x86_64-linux-gnu
            Target: arm-unknown-linux-gnueabi
             Build: x86_64-linux-gnu


Created attachment 24294
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=24294
preprocessed source

$ ./xgcc -v
Using built-in specs.
COLLECT_GCC=./xgcc
Target: arm-unknown-linux-gnueabi
Configured with: ../configure --target=arm-unknown-linux-gnueabi
--prefix=/home/ryan/x-tools/arm-unknown-linux-gnueabi
--with-sysroot=/home/ryan/x-tools/arm-unknown-linux-gnueabi/arm-unknown-linux-gnueabi//sys-root
--disable-multilib
--with-local-prefix=/home/ryan/x-tools/arm-unknown-linux-gnueabi/arm-unknown-linux-gnueabi/sys-root
--disable-nls --enable-threads=posix --enable-symvers=gnu --enable-c99
--enable-long-long --enable-target-optspace
target_alias=arm-unknown-linux-gnueabi --enable-languages=c++ --disable-shared
--disable-libmudflap --disable-libssp
Thread model: posix
gcc version 4.7.0 20110519 (experimental) [trunk revision 173917] (GCC) 
$ ./xgcc -B. /home/ryan/ice.i -Os
/home/ryan/ice.i: In function 'foo':
/home/ryan/ice.i:23:11: internal compiler error: in gen_cstoredi4, at
config/arm/arm.md:7554
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

[Bug target/49069] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[ibolton at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

Ian Bolton <ibolton at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011.05.20 10:05:11
                 CC|                            |ibolton at gcc dot gnu.org
     Ever Confirmed|0                           |1
      Known to fail|                            |4.7.0

--- Comment #1 from Ian Bolton <ibolton at gcc dot gnu.org> 2011-05-20 10:05:11 UTC ---
Confirmed.  Fails for ARM and Thumb.

Here's my compiler command-line:

cc1 -fpreprocessed 49069.i -quiet -dumpbase 49069.i -marm -mcpu=cortex-a9
-mfloat-abi=softfp -mfpu=vfpv3-d16 -auxbase 49069 -Os -version -o 49069.s

This has actually been broken since at least r164700 (I had some old toolchains
available to try), so I expect 4.6.0 is broken too, but I don't have time to
confirm at the moment.

[Bug target/49069] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[rmansfield at qnx dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

--- Comment #2 from Ryan Mansfield <rmansfield at qnx dot com> 2011-05-20 13:08:05 UTC ---
This ICE happens for me on the 4.6 branch.

[Bug target/49069] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[mikpe at it dot uu.se]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

Mikael Pettersson <mikpe at it dot uu.se> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mikpe at it dot uu.se

--- Comment #3 from Mikael Pettersson <mikpe at it dot uu.se> 2011-09-24 13:17:36 UTC ---
This still ICEs gcc-4.6-20110923, but stopped ICEing trunk with the
tree-ssa-forwprop optimization improvement in r174428:

http://gcc.gnu.org/ml/gcc-cvs/2011-05/msg01209.html

However the underlying issue remains.  The ARM cstoredi4 expander has a very
loose operand match spec which allows both operands to be constants, but that
case doesn't work(*), so it has a gcc_assert () to cause an error in that case.
But asserting is wrong: it should either use stricter operand specs to avoid
undesirable operand combinations, or FAIL the expander for such operands.

Replacing the assert with a FAIL on the negated condition seem appropriate and
fixes this test case for me.

[(*) Simply removing the gcc_assert in 4.6.1 leads to

pr49069.i: In function 'foo':
pr49069.i:26:1: error: unrecognizable insn:
(insn 8 7 9 3 (set (reg:CC 24 cc)
        (compare:CC (const_int 0 [0])
            (const_int 1 [0x1]))) pr49069.i:23 -1
     (nil))
pr49069.i:26:1: internal compiler error: in extract_insn, at recog.c:2109]

[Bug target/49069] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[rearnsha at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

--- Comment #4 from Richard Earnshaw <rearnsha at gcc dot gnu.org> 2012-03-01 16:08:35 UTC ---
The real question is why is a cstore operation being called with constants for
the two operands in the comparison.  If it knows the constants, then it knows
the result of the comparison and can collapse the entire result.

[Bug tree-optimization/49069] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

Steven Bosscher <steven at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |steven at gcc dot gnu.org
          Component|target                      |tree-optimization
      Known to work|                            |4.7.0
      Known to fail|4.7.0                       |4.6.3

--- Comment #5 from Steven Bosscher <steven at gcc dot gnu.org> 2012-03-01 18:00:55 UTC ---
(In reply to comment #4)
> The real question is why is a cstore operation being called with constants for
> the two operands in the comparison.  If it knows the constants, then it knows
> the result of the comparison and can collapse the entire result.

That is because "GCC: (GNU) 4.6.3" fails to fold a comparison in GIMPLE:

$ cat ice.i.143t.optimized 

;; Function func1 (func1)

func1 ()
{
<bb 2>:
  return;

}



;; Function foo (foo)

foo ()
{
  int D.1291;
  long long unsigned int D.1290;

<bb 2>:
  array1[0] = 0;
  D.1290_6 = 0;
  D.1291_7 = D.1290_6 == 1;
  bar (D.1291_7); [tail call]
  return;

}

TER propagates D.1290_6 into the comparison:

Replacing Expressions
D.1290_6 replace with --> D.1290_6 = 0;
D.1291_7 replace with --> D.1291_7 = D.1290_6 == 1;

The ICE goes away with "-fno-tree-ter". This is not a target bug IMHO, but a
tree-optimization bug.

[Bug tree-optimization/49069] [4.6/4.7 Regression] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

Steven Bosscher <steven at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-on-valid-code
      Known to work|4.7.0                       |4.4.2, 4.5.3
            Summary|ICE in gen_cstoredi4, at    |[4.6/4.7 Regression] ICE in
                   |config/arm/arm.md:7554      |gen_cstoredi4, at
                   |                            |config/arm/arm.md:7554
      Known to fail|                            |4.7.0

--- Comment #6 from Steven Bosscher <steven at gcc dot gnu.org> 2012-03-01 18:18:14 UTC ---
Latent on trunk, but tree-ssa-ter.c has no changes (wow!) between GCC 4.6.3 and
GCC 4.7.0 trunk of today, so the bug is still present on trunk also (but just
harder to trigger).

The test case compiles fine with GCC 4.4.2 and GCC 4.5.3, so this is a
regression.

[Bug tree-optimization/49069] [4.6/4.7 Regression] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[steven at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

--- Comment #7 from Steven Bosscher <steven at gcc dot gnu.org> 2012-03-01 19:27:56 UTC ---
Created attachment 26804
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=26804
Tighten predicates

[Bug target/49069] [4.6/4.7 Regression] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|tree-optimization           |target
   Target Milestone|---                         |4.6.4

--- Comment #8 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-03-02 10:06:17 UTC ---
I don't think what the tree level does is wrong.  Target patch suggests target
bug.

[Bug target/49069] [4.6/4.7/4.8 Regression] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

[Bug target/49069] [4.6/4.7/4.8 Regression] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

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

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

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-17 15:55:26 UTC ---
The issue isn't latent, can be still reproduced with:
/* PR target/49069 */
/* { dg-do compile } */
/* { dg-options "-Os -fno-tree-forwprop -Wno-div-by-zero" } */

int a;
const unsigned long long b[1] = { 1ULL };
extern void bar (int);

void
foo (void)
{
  for (a = 0; a == 1; a = 2)
    ;
  bar (b[0] == (a == 0 ? a : a / 0));
}

Which also shows that the middle-end usually optimizes that well (cf. #c4), but
the arm backend relying on that always happening is just a very bad assumption.

[Bug target/49069] [4.6/4.7/4.8 Regression] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-17 16:08:15 UTC ---
Created attachment 29193
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29193
gcc48-pr49069-1.patch

One possible fix.

[Bug target/49069] [4.6/4.7/4.8 Regression] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

--- Comment #11 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-17 16:10:45 UTC ---
Created attachment 29194
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29194
gcc48-pr49069-2.patch

Another possible fix.  Or Steven's fix (+ testcase) if it works.  It will be
optimized by RTL optimizers anyway.  Just asserting that an unlikely, but
possible thing, won't happen, is wrong.  Can we get this fixed for 4.8?

[Bug target/49069] [4.6/4.7/4.8 Regression] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

--- Comment #12 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-23 08:37:39 UTC ---
Author: jakub
Date: Wed Jan 23 08:37:16 2013
New Revision: 195398

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=195398
Log:
    PR target/49069
    * config/arm/arm.md (cbranchdi4, cstoredi4): Use s_register_operand
    instead of cmpdi_operand for first comparison operand.
    Don't assert that comparison operands aren't both constants.

    * gcc.dg/pr49069.c: New test.

Added:
    trunk/gcc/testsuite/gcc.dg/pr49069.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/arm/arm.md
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/testsuite/ChangeLog

[Bug target/49069] [4.6/4.7 Regression] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |4.8.0
            Summary|[4.6/4.7/4.8 Regression]    |[4.6/4.7 Regression] ICE in
                   |ICE in gen_cstoredi4, at    |gen_cstoredi4, at
                   |config/arm/arm.md:7554      |config/arm/arm.md:7554

--- Comment #13 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-23 08:45:45 UTC ---
Fixed on the trunk so far.

[Bug target/49069] [4.6/4.7 Regression] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

--- Comment #14 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-02-01 14:14:19 UTC ---
Author: jakub
Date: Fri Feb  1 14:14:11 2013
New Revision: 195661

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=195661
Log:
    Backported from mainline
    2013-01-23  Jakub Jelinek  <jakub@redhat.com>

    PR target/49069
    * config/arm/arm.md (cbranchdi4, cstoredi4): Use s_register_operand
    instead of cmpdi_operand for first comparison operand.
    Don't assert that comparison operands aren't both constants.

    * gcc.dg/pr49069.c: New test.

Added:
    branches/gcc-4_7-branch/gcc/testsuite/gcc.dg/pr49069.c
Modified:
    branches/gcc-4_7-branch/gcc/ChangeLog
    branches/gcc-4_7-branch/gcc/config/arm/arm.md
    branches/gcc-4_7-branch/gcc/testsuite/ChangeLog

[Bug target/49069] [4.6 Regression] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |4.7.3
            Summary|[4.6/4.7 Regression] ICE in |[4.6 Regression] ICE in
                   |gen_cstoredi4, at           |gen_cstoredi4, at
                   |config/arm/arm.md:7554      |config/arm/arm.md:7554

--- Comment #15 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-02-01 14:36:49 UTC ---
Fixed for 4.7.3+ too.

[Bug target/49069] [4.6 Regression] ICE in gen_cstoredi4, at config/arm/arm.md:7554

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49069

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|4.6.4                       |4.7.3

--- Comment #16 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-04-12 16:29:08 UTC ---
The 4.6 branch has been closed, fixed in GCC 4.7.3.