[Bug c/61068] New: -O3 generates bad code (when tree-vectorize or inline-functions is on)

[Bug c/61068] New: -O3 generates bad code (when tree-vectorize or inline-functions is on)

[wence at gmx dot li]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

            Bug ID: 61068
           Summary: -O3 generates bad code (when tree-vectorize or
                    inline-functions is on)
           Product: gcc
           Version: 4.8.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: wence at gmx dot li

Created attachment 32738
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=32738&action=edit
preprocessed source

The attached program segfaults at runtime when compiled with -O3.  If I compile
with either -fno-tree-vectorize or -fno-inline then the output is correct
(verified externally, this is a cut down example).

Note that if I split the increments into A[j][k] and A[(j+3)][(k+3)] into
separate loops, then the segfault also goes away.

I am currently building 4.9 to see if the problem still exists there.

Compilation command:

gcc-4.8 -v -std=c99 -O3  foo.c  -save-temps

-v output:

Using built-in specs.
COLLECT_GCC=gcc-4.8
COLLECT_LTO_WRAPPER=/usr/local/Cellar/gcc48/4.8.2/libexec/gcc/x86_64-apple-darwin13.0.0/4.8.2/lto-wrapper
Target: x86_64-apple-darwin13.0.0
Configured with: ../configure --build=x86_64-apple-darwin13.0.0
--prefix=/usr/local/Cellar/gcc48/4.8.2 --enable-languages=c,c++,objc,obj-c++
--program-suffix=-4.8 --with-gmp=/usr/local/opt/gmp4
--with-mpfr=/usr/local/opt/mpfr2 --with-mpc=/usr/local/opt/libmpc08
--with-cloog=/usr/local/opt/cloog018 --with-isl=/usr/local/opt/isl011
--with-system-zlib --enable-version-specific-runtime-libs
--enable-libstdcxx-time=yes --enable-stage1-checking --enable-checking=release
--enable-lto --disable-werror --enable-plugin --disable-nls --disable-multilib
Thread model: posix
gcc version 4.8.2 (GCC) 
COLLECT_GCC_OPTIONS='-mmacosx-version-min=10.9.1' '-v' '-std=c99' '-O3'
'-save-temps' '-mtune=core2'
 /usr/local/Cellar/gcc48/4.8.2/libexec/gcc/x86_64-apple-darwin13.0.0/4.8.2/cc1
-E -quiet -v -D__DYNAMIC__ foo.c -fPIC -mmacosx-version-min=10.9.1 -mtune=core2
-std=c99 -O3 -fpch-preprocess -o foo.i
ignoring nonexistent directory "/usr/local/Cellar/gcc48/4.8.2/include"
ignoring nonexistent directory
"/usr/local/Cellar/gcc48/4.8.2/lib/gcc/x86_64-apple-darwin13.0.0/4.8.2/../../../../x86_64-apple-darwin13.0.0/include"
#include "..." search starts here:
#include <...> search starts here:
 /usr/local/Cellar/gcc48/4.8.2/lib/gcc/x86_64-apple-darwin13.0.0/4.8.2/include
 /usr/local/include

/usr/local/Cellar/gcc48/4.8.2/lib/gcc/x86_64-apple-darwin13.0.0/4.8.2/include-fixed
 /usr/include
 /System/Library/Frameworks
 /Library/Frameworks
End of search list.
COLLECT_GCC_OPTIONS='-mmacosx-version-min=10.9.1' '-v' '-std=c99' '-O3'
'-save-temps' '-mtune=core2'
 /usr/local/Cellar/gcc48/4.8.2/libexec/gcc/x86_64-apple-darwin13.0.0/4.8.2/cc1
-fpreprocessed foo.i -fPIC -quiet -dumpbase foo.c -mmacosx-version-min=10.9.1
-mtune=core2 -auxbase foo -O3 -std=c99 -version -o foo.s
GNU C (GCC) version 4.8.2 (x86_64-apple-darwin13.0.0)
    compiled by GNU C version 4.8.2, GMP version 4.3.2, MPFR version 2.4.2, MPC
version 0.8.1
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
GNU C (GCC) version 4.8.2 (x86_64-apple-darwin13.0.0)
    compiled by GNU C version 4.8.2, GMP version 4.3.2, MPFR version 2.4.2, MPC
version 0.8.1
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
Compiler executable checksum: 077b529d6cfe035508c45a2f30143407
COLLECT_GCC_OPTIONS='-mmacosx-version-min=10.9.1' '-v' '-std=c99' '-O3'
'-save-temps' '-mtune=core2'
 as -arch x86_64 -force_cpusubtype_ALL -o foo.o foo.s
COMPILER_PATH=/usr/local/Cellar/gcc48/4.8.2/libexec/gcc/x86_64-apple-darwin13.0.0/4.8.2/:/usr/local/Cellar/gcc48/4.8.2/libexec/gcc/x86_64-apple-darwin13.0.0/4.8.2/:/usr/local/Cellar/gcc48/4.8.2/libexec/gcc/x86_64-apple-darwin13.0.0/:/usr/local/Cellar/gcc48/4.8.2/lib/gcc/x86_64-apple-darwin13.0.0/4.8.2/:/usr/local/Cellar/gcc48/4.8.2/lib/gcc/x86_64-apple-darwin13.0.0/
LIBRARY_PATH=/usr/local/Cellar/gcc48/4.8.2/lib/gcc/x86_64-apple-darwin13.0.0/4.8.2/:/usr/local/Cellar/gcc48/4.8.2/lib/gcc/x86_64-apple-darwin13.0.0/4.8.2/../../../:/usr/lib/
COLLECT_GCC_OPTIONS='-mmacosx-version-min=10.9.1' '-v' '-std=c99' '-O3'
'-save-temps' '-mtune=core2'

/usr/local/Cellar/gcc48/4.8.2/libexec/gcc/x86_64-apple-darwin13.0.0/4.8.2/collect2
-dynamic -arch x86_64 -macosx_version_min 10.9.1 -weak_reference_mismatches
non-weak -o a.out
-L/usr/local/Cellar/gcc48/4.8.2/lib/gcc/x86_64-apple-darwin13.0.0/4.8.2
-L/usr/local/Cellar/gcc48/4.8.2/lib/gcc/x86_64-apple-darwin13.0.0/4.8.2/../../..
foo.o -no_compact_unwind -lSystem -lgcc_ext.10.5 -lgcc -lSystem -v
collect2 version 4.8.2
/usr/bin/ld -dynamic -arch x86_64 -macosx_version_min 10.9.1
-weak_reference_mismatches non-weak -o a.out
-L/usr/local/Cellar/gcc48/4.8.2/lib/gcc/x86_64-apple-darwin13.0.0/4.8.2
-L/usr/local/Cellar/gcc48/4.8.2/lib/gcc/x86_64-apple-darwin13.0.0/4.8.2/../../..
foo.o -no_compact_unwind -lSystem -lgcc_ext.10.5 -lgcc -lSystem -v
@(#)PROGRAM:ld  PROJECT:ld64-236.4
configured to support archs: armv6 armv7 armv7s arm64 i386 x86_64 armv6m armv7m
armv7em
Library search paths:
    /usr/local/Cellar/gcc48/4.8.2/lib/gcc/x86_64-apple-darwin13.0.0/4.8.2
    /usr/local/Cellar/gcc48/4.8.2/lib
    /usr/lib
    /usr/local/lib
Framework search paths:
    /Library/Frameworks/
    /System/Library/Frameworks/

[Bug c/61068] -O3 generates bad code (when tree-vectorize or inline-functions is on)

[wence at gmx dot li]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

--- Comment #1 from Lawrence Mitchell <wence at gmx dot li> ---
4.9 does not fail.

[Bug target/61068] -O3 generates bad code (when tree-vectorize or inline-functions is on)

[mpolacek at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

Marek Polacek <mpolacek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mpolacek at gcc dot gnu.org
          Component|c                           |target

--- Comment #2 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
Can't reproduce on Linux.

[Bug target/61068] -O3 generates bad code (when tree-vectorize or inline-functions is on)

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

Zdenek Sojka <zsojka at seznam dot cz> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |zsojka at seznam dot cz

--- Comment #3 from Zdenek Sojka <zsojka at seznam dot cz> ---
You are casting "double arg1_0[12][3]" (and others) to "(double *)arg1_0" and
then using it as a single-dimensional array. The code has undefined behaviour.

[Bug target/61068] -O3 generates bad code (when tree-vectorize or inline-functions is on)

[wence at gmx dot li]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

--- Comment #5 from Lawrence Mitchell <wence at gmx dot li> ---
(In reply to Zdenek Sojka from comment #3)
> You are casting "double arg1_0[12][3]" (and others) to "(double *)arg1_0"
> and then using it as a single-dimensional array. The code has undefined
> behaviour.

Apologies, I I've updated the test case to use 1D arrays everywhere and the
problem persists.  I believe this latter does not invoke undefined behaviour.

Thanks,
Lawrence

[Bug tree-optimization/61068] [4.8 Regression] -O3 generates bad code (when tree-vectorize or inline-functions is on)

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
So - would be interesting to know what fixed it on trunk.

[Bug tree-optimization/61068] [4.8 Regression] -O3 generates bad code (when tree-vectorize or inline-functions is on)

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-*-*
             Status|UNCONFIRMED                 |NEW
      Known to work|                            |4.7.3, 4.9.0
   Last reconfirmed|                            |2014-05-06
          Component|target                      |tree-optimization
                 CC|                            |rguenth at gcc dot gnu.org
     Ever confirmed|0                           |1
            Summary|-O3 generates bad code      |[4.8 Regression] -O3
                   |(when tree-vectorize or     |generates bad code (when
                   |inline-functions is on)     |tree-vectorize or
                   |                            |inline-functions is on)
   Target Milestone|---                         |4.8.3
      Known to fail|                            |4.8.2

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed on x86_64-linux with -O3 -fPIC (default on darwin) and with
-O2 -ftree-vectorize -funroll-loops -fPIC.

There are a few vectorizer wrong-code bugs on the 4.8 branch that I haven't
managed to backport yet from 4.9.  4.7 seems to work.

[Bug tree-optimization/61068] [4.8 Regression] -O3 generates bad code (when tree-vectorize or inline-functions is on)

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

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

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

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Seems it has been introduced by r195609 and fixed by r196782.

[Bug tree-optimization/61068] [4.8 Regression] -O3 generates bad code (when tree-vectorize or inline-functions is on)

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org

--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> ---
Ok, I have a look.

[Bug tree-optimization/61068] [4.8 Regression] -O3 generates bad code (when tree-vectorize or inline-functions is on)

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
-fno-ivopts fixes it.  Reverting r195609 on top of the branch doesn't fix it,
backporting r196782 does.

IVOPTs ends up choosing different IVs, but I don't see anything obviously
wrong ...

[Bug tree-optimization/61068] [4.8 Regression] -O3 generates bad code (when tree-vectorize or inline-functions is on)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.8.3                       |4.8.4

--- Comment #11 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 4.8.3 is being released, adjusting target milestone.

[Bug tree-optimization/61068] [4.8 Regression] -O3 generates bad code (when tree-vectorize or inline-functions is on)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
           Priority|P3                          |P2

[Bug tree-optimization/61068] [4.8 Regression] -O3 generates bad code (when tree-vectorize or inline-functions is on)

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.8.4                       |4.8.5

--- Comment #12 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 4.8.4 has been released.

[Bug tree-optimization/61068] [4.8 Regression] -O3 generates bad code (when tree-vectorize or inline-functions is on)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61068

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED
   Target Milestone|4.8.5                       |4.9.0
      Known to fail|                            |4.8.5

--- Comment #13 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed for 4.9.0.