[Bug tree-optimization/63379] New: Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[Bug tree-optimization/63379] New: Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[jwyatt at feralinteractive dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63379

            Bug ID: 63379
           Summary: Incorrect vectorization when enabling SSE and O3,
                    initialises loop with wrong value
           Product: gcc
           Version: 4.8.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jwyatt at feralinteractive dot com

Created attachment 33579
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33579&action=edit
.i file for simple test case

Enabling SSE and 03 causes gcc to vectorize the following for loop:

Point min_pt = pt_array[0];
for (Point *ptr = pt_array+1, *ptr_end = pt_array+size; ptr != ptr_end; ++ptr)
{
    min_pt.x = (min_pt.x < ptr->x) ? min_pt.x : ptr->x;
    min_pt.y = (min_pt.y < ptr->y) ? min_pt.y : ptr->y;
}

where pt_array is an array of:

struct Point {
    int x;
    int y;
};

This sets the min_pt.y value to an x value if the y values are greater than the
x values.

Looking at the assembly, the vectorization splits this into two. It looks like
for one of these, it initialises the first min y value as the min x value.

Compiled with g++ -msse --std=c++11 -O3 -Wall -Wextra test.cpp, which gives no
warnings.

Version info:
Using built-in specs.
COLLECT_GCC=g++
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/x86_64-redhat-linux/4.8.3/lto-wrapper
Target: x86_64-redhat-linux
Configured with: ../configure --prefix=/usr --mandir=/usr/share/man
--infodir=/usr/share/info --with-bugurl=http://bugzilla.redhat.com/bugzilla
--enable-bootstrap --enable-shared --enable-threads=posix
--enable-checking=release --with-system-zlib --enable-__cxa_atexit
--disable-libunwind-exceptions --enable-gnu-unique-object
--enable-linker-build-id --with-linker-hash-style=gnu
--enable-languages=c,c++,objc,obj-c++,java,fortran,ada,go,lto --enable-plugin
--enable-initfini-array --enable-java-awt=gtk --disable-dssi
--with-java-home=/usr/lib/jvm/java-1.5.0-gcj-1.5.0.0/jre
--enable-libgcj-multifile --enable-java-maintainer-mode
--with-ecj-jar=/usr/share/java/eclipse-ecj.jar --disable-libjava-multilib
--with-isl=/builddir/build/BUILD/gcc-4.8.3-20140911/obj-x86_64-redhat-linux/isl-install
--with-cloog=/builddir/build/BUILD/gcc-4.8.3-20140911/obj-x86_64-redhat-linux/cloog-install
--with-tune=generic --with-arch_32=i686 --build=x86_64-redhat-linux
Thread model: posix
gcc version 4.8.3 20140911 (Red Hat 4.8.3-7) (GCC)

[Bug tree-optimization/63379] Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[jwyatt at feralinteractive dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63379

--- Comment #1 from Jason Wyatt <jwyatt at feralinteractive dot com> ---
Created attachment 33580
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33580&action=edit
Test case

Compiling this code and running it gives the following output:
Point 0,1000
Point 1,1001
Point 2,1002
Point 3,1003
Point 4,1004
Point 5,1005
Point 6,1006
Point 7,1007
Point 8,1008
Point 9,1009
Point 10,1010
Point 11,1011
Point 12,1012
Point 13,1013
Point 14,1014
Point 15,1015
Point 16,1016
Point 17,1017
Point 18,1018
Point 19,1019
Point 20,1020
Point 21,1021
Point 22,1022
Point 23,1023
Point 24,1024
Size: 25
25, 0, 0 <-------This value should be 1000 (minimum y value)
Point 0,1000
Point 1,1001
Point 2,1002
Point 3,1003
Point 4,1004
Point 5,1005
Point 6,1006
Point 7,1007
Point 8,1008
Point 9,1009
Point 10,1010
Point 11,1011
Point 12,1012
Point 13,1013
Point 14,1014
Point 15,1015
Point 16,1016
Point 17,1017
Point 18,1018
Point 19,1019
Point 20,1020
Point 21,1021
Point 22,1022
Point 23,1023
Point 24,1024

[Bug tree-optimization/63379] Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63379

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
             Target|                            |x86_64-*-*
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-10-09
     Ever confirmed|0                           |1
      Known to fail|                            |4.7.4, 4.8.3, 4.9.1, 5.0

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  We're choosing bogus "neutral" operands for MIN/MAX in the case of
SLP.  I am testing

Index: gcc/tree-vect-slp.c
===================================================================
--- gcc/tree-vect-slp.c (revision 215917)
+++ gcc/tree-vect-slp.c (working copy)
@@ -2395,13 +2395,8 @@ vect_get_constant_vectors (tree op, slp_
             neutral_op = build_int_cst (TREE_TYPE (op), -1);
             break;

-          case MAX_EXPR:
-          case MIN_EXPR:
-            def_stmt = SSA_NAME_DEF_STMT (op);
-            loop = (gimple_bb (stmt))->loop_father;
-            neutral_op = PHI_ARG_DEF_FROM_EDGE (def_stmt,
-                                                loop_preheader_edge (loop));
-            break;
+         /* For MIN/MAX we don't have an easy neutral operand but
+            the initial values can be used fine here.  */

           default:
             neutral_op = NULL;

[Bug tree-optimization/63379] Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63379

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

[Bug tree-optimization/63379] Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63379

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Blocks|                            |50414

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
The simple fix makes gcc.dg/vect/no-scevccp-noreassoc-slp-reduc-7.c ICE as
there
we have a MAX<MAX<x, red_1>, y> reduction.  Not sure why this goes the SLP
way here - still need to investigate properly.  Ah, that testcase was added
when that broken MIN/MAX_EXPR handling was added:

r178941 | irar | 2011-09-18 13:36:15 +0200 (Sun, 18 Sep 2011) | 6 lines


        PR tree-optimization/50414
        * tree-vect-slp.c (vect_get_constant_vectors): Handle MAX_EXPR
        and MIN_EXPR.

[Bug tree-optimization/63379] [4.8/4.9/5 Regression] Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63379

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |4.6.3
   Target Milestone|---                         |4.8.4
            Summary|Incorrect vectorization     |[4.8/4.9/5 Regression]
                   |when enabling SSE and O3,   |Incorrect vectorization
                   |initialises loop with wrong |when enabling SSE and O3,
                   |value                       |initialises loop with wrong
                   |                            |value

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
Makes it a regression.

[Bug tree-optimization/63379] [4.8/4.9/5 Regression] Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63379

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
Author: rguenth
Date: Fri Oct 10 11:17:13 2014
New Revision: 216070

URL: https://gcc.gnu.org/viewcvs?rev=216070&root=gcc&view=rev
Log:
2014-10-10  Richard Biener  <rguenther@suse.de>

    PR tree-optimization/63379
    * tree-vect-slp.c (vect_get_constant_vectors): Do not compute
    a neutral operand for min/max when it is not a reduction chain.

    * gcc.dg/vect/pr63379.c: New testcase.

Added:
    trunk/gcc/testsuite/gcc.dg/vect/pr63379.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-vect-slp.c

[Bug tree-optimization/63379] [4.8 Regression] Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63379

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
Author: rguenth
Date: Fri Oct 10 14:23:35 2014
New Revision: 216086

URL: https://gcc.gnu.org/viewcvs?rev=216086&root=gcc&view=rev
Log:
2014-10-10  Richard Biener  <rguenther@suse.de>

    PR tree-optimization/63379
    * tree-vect-slp.c (vect_get_constant_vectors): Do not compute
    a neutral operand for min/max when it is not a reduction chain.

    * gcc.dg/vect/pr63379.c: New testcase.

Added:
    branches/gcc-4_9-branch/gcc/testsuite/gcc.dg/vect/pr63379.c
Modified:
    branches/gcc-4_9-branch/gcc/ChangeLog
    branches/gcc-4_9-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_9-branch/gcc/tree-vect-slp.c

[Bug tree-optimization/63379] [4.8 Regression] Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63379

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |4.9.2, 5.0
            Summary|[4.8/4.9/5 Regression]      |[4.8 Regression] Incorrect
                   |Incorrect vectorization     |vectorization when enabling
                   |when enabling SSE and O3,   |SSE and O3, initialises
                   |initialises loop with wrong |loop with wrong value
                   |value                       |
      Known to fail|5.0                         |

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed on trunk and 4.9 branch sofar.

[Bug tree-optimization/63379] [4.8 Regression] Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[clyon at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63379

clyon at gcc dot gnu.org changed:

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

--- Comment #8 from clyon at gcc dot gnu.org ---
After this commit, I've noticed that this new test appears as FAIL on target
aarch64_be-none-linux-gnu.

My gcc.log shows:
PASS: gcc.dg/vect/pr63379.c (test for excess errors)

*** EXIT code 4242
emu: host signal 6
FAIL: gcc.dg/vect/pr63379.c execution test

I'm using qemu as execution engine.

[Bug tree-optimization/63379] [4.8 Regression] Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63379

--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to clyon from comment #8)
> After this commit, I've noticed that this new test appears as FAIL on target
> aarch64_be-none-linux-gnu.
> 
> My gcc.log shows:
> PASS: gcc.dg/vect/pr63379.c (test for excess errors)
> 
> *** EXIT code 4242
> emu: host signal 6
> FAIL: gcc.dg/vect/pr63379.c execution test
> 
> I'm using qemu as execution engine.

Maybe you run into known reduction endianess dependence issues here (as you
are on big-endian).  Who knows...

[Bug tree-optimization/63379] [4.8 Regression] Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63379

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

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

--- Comment #11 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.

[Bug tree-optimization/63379] [4.8 Regression] Incorrect vectorization when enabling SSE and O3, initialises loop with wrong value

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63379

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
Author: rguenth
Date: Wed Nov 26 13:15:16 2014
New Revision: 218079

URL: https://gcc.gnu.org/viewcvs?rev=218079&root=gcc&view=rev
Log:
2014-11-26  Richard Biener  <rguenther@suse.de>

    Backport from mainline
    2014-10-08  Richard Biener  <rguenther@suse.de>

    PR tree-optimization/61969
    * tree-nrv.c (pass_nrv::execute): Properly test for automatic
    variables.

    2014-08-15  Richard Biener  <rguenther@suse.de>

    PR tree-optimization/62031
    * tree-data-ref.c (dr_analyze_indices): Do not set
    DR_UNCONSTRAINED_BASE.
    (dr_may_alias_p): All indirect accesses have to go the
    formerly DR_UNCONSTRAINED_BASE path.
    * tree-data-ref.h (struct indices): Remove
    unconstrained_base member.
    (DR_UNCONSTRAINED_BASE): Remove.

    * gcc.dg/torture/pr62031.c: New testcase.

    2014-10-10  Richard Biener  <rguenther@suse.de>

    PR tree-optimization/63379
    * tree-vect-slp.c (vect_get_constant_vectors): Do not compute
    a neutral operand for min/max when it is not a reduction chain.

    * gcc.dg/vect/pr63379.c: New testcase.

    2014-11-07  Richard Biener  <rguenther@suse.de>

    PR tree-optimization/63605
    * fold-const.c (fold_binary_loc): Properly use element_precision
    for types that may not be scalar.

    * gcc.dg/vect/pr63605.c: New testcase.

    2014-10-28  Richard Biener  <rguenther@suse.de>

    PR middle-end/63665
    * fold-const.c (fold_comparison): Properly guard simplifying
    against INT_MAX/INT_MIN with !TYPE_OVERFLOW_WRAPS.

    * gcc.dg/pr63665.c: New testcase.

Added:
    branches/gcc-4_8-branch/gcc/testsuite/gcc.dg/pr63665.c
    branches/gcc-4_8-branch/gcc/testsuite/gcc.dg/torture/pr62031.c
    branches/gcc-4_8-branch/gcc/testsuite/gcc.dg/vect/pr63379.c
    branches/gcc-4_8-branch/gcc/testsuite/gcc.dg/vect/pr63605.c
Modified:
    branches/gcc-4_8-branch/gcc/ChangeLog
    branches/gcc-4_8-branch/gcc/fold-const.c
    branches/gcc-4_8-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_8-branch/gcc/tree-data-ref.c
    branches/gcc-4_8-branch/gcc/tree-data-ref.h
    branches/gcc-4_8-branch/gcc/tree-nrv.c
    branches/gcc-4_8-branch/gcc/tree-vect-slp.c