[Bug tree-optimization/56145] New: [4.8.0 ICE] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[Bug tree-optimization/56145] New: [4.8.0 ICE] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[vchou79 at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

             Bug #: 56145
           Summary: [4.8.0 ICE] Use of too much optimizations -O2
                    -ffast-math -floop-parallelize-all
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: vchou79@gmail.com


Created attachment 29304
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29304
preprocessed source

When compiling the following source code from this url.
http://waix.dl.sourceforge.net/project/lame/lame/3.99/lame-3.99.5.tar.gz

gcc produced ice, when the particular combination of optimization flags are
used. -m32 -m64 made no difference. When adding -fwrapv to the command line it
stops gcc from produce the ice.

gcc version 4.8.0 20130129 (experimental)
[gcc trunk revision 195526]
(mingw-w64 trunk revision 5568) 

x86_64-w64-mingw32-gcc -m64 -DHAVE_CONFIG_H -I. -I.. -I../include
-I../libmp3lame -O1 -ffast-math -floop-parallelize-all -c layer3.c -o layer3.o

Compiled OK...

x86_64-w64-mingw32-gcc -m64 -DHAVE_CONFIG_H -I. -I.. -I../include
-I../libmp3lame -O2 -ffast-math -floop-parallelize-all -c layer3.c -o layer3.o
layer3.c: In function 'decode_layer3_sideinfo':
layer3.c:1613:1: internal compiler error: Segmentation fault
 decode_layer3_sideinfo(PMPSTR mp)
 ^
0x71f295 crash_signal
        ../../gcc-4_8-trunk/gcc/toplev.c:332
0xac6017 subtract_commutative_associative_deps
        ../../gcc-4_8-trunk/gcc/graphite-dependences.c:422
0xac6017 compute_deps(scop*, vec<poly_bb*, va_heap, vl_ptr>, isl_union_map**,
isl_union_map**, isl_union_map**, isl_union_map**, isl_union_map**,
isl_union_map**, isl_union_map**, isl_union_map**, isl_union_map**,
isl_union_map**, isl_union_map**, isl_union_map**)
        ../../gcc-4_8-trunk/gcc/graphite-dependences.c:494
0xac665a loop_level_carries_dependences
        ../../gcc-4_8-trunk/gcc/graphite-dependences.c:558
0xac665a loop_is_parallel_p(loop*, htab*, int)
        ../../gcc-4_8-trunk/gcc/graphite-dependences.c:591
0xac4eda translate_clast_for_loop
        ../../gcc-4_8-trunk/gcc/graphite-clast-to-gimple.c:1173
0xac4eda translate_clast_for
        ../../gcc-4_8-trunk/gcc/graphite-clast-to-gimple.c:1197
0xac4eda translate_clast
        ../../gcc-4_8-trunk/gcc/graphite-clast-to-gimple.c:1280
0xac54b4 gloog(scop*, htab*)
        ../../gcc-4_8-trunk/gcc/graphite-clast-to-gimple.c:1680
0xac29c4 graphite_transform_loops()
        ../../gcc-4_8-trunk/gcc/graphite.c:288
0x7d3726 graphite_transforms
        ../../gcc-4_8-trunk/gcc/tree-ssa-loop.c:258
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <http://gcc.gnu.org/bugs.html> for instructions.

Hopefully somebody could be able to narrow this down and produce a reduced test
case.

Thank you.

[Bug tree-optimization/56145] [4.8.0 ICE] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[vchou79 at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

--- Comment #1 from Vincent <vchou79 at gmail dot com> 2013-01-31 13:40:36 UTC ---
x86_64-w64-mingw32-gcc -Os -ffast-math -floop-parallelize-all -I. -std=gnu99 -c
-o common/predict.o common/predict.c
common/predict.c: In function 'x264_predict_16x16_dc_c':
common/predict.c:58:6: internal compiler error: Segmentation fault
 void x264_predict_16x16_dc_c( pixel *src )

Just another example same error.

[Bug tree-optimization/56145] [4.8 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-w64-mingw32
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2013-02-08
   Target Milestone|---                         |4.8.0
            Summary|[4.8.0 ICE] Use of too much |[4.8 Regression] Use of too
                   |optimizations -O2           |much optimizations -O2
                   |-ffast-math                 |-ffast-math
                   |-floop-parallelize-all      |-floop-parallelize-all
     Ever Confirmed|0                           |1

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> 2013-02-08 15:46:54 UTC ---
Is this with a cross-compiler or a native compiler?

[Bug tree-optimization/56145] [4.8 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[vchou79 at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

--- Comment #3 from Vincent <vchou79 at gmail dot com> 2013-02-08 18:29:28 UTC ---
(In reply to comment #2)
> Is this with a cross-compiler or a native compiler?

Cross compiler.

[Bug tree-optimization/56145] [4.8 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[vchou79 at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

--- Comment #4 from Vincent <vchou79 at gmail dot com> 2013-02-09 00:36:41 UTC ---
Hi Richard,

The same also with native win-64bits build.
host=x86_64-w64-mingw32
build=x86_64-slackware-linux
target=x86_64-w64-mingw32


gcc version 4.8.0 20130208 (experimental)
[gcc trunk revision 195894]
(mingw-w64 trunk revision 5583)

x86_64-w64-mingw32-gcc -m64 -O2 -floop-parallelize-all -ffast-math  -c layer3.i
layer3.c: In function 'decode_layer3_sideinfo':
layer3.c:1613:1: internal compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

[Bug tree-optimization/56145] [4.8 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[vchou79 at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

--- Comment #5 from Vincent <vchou79 at gmail dot com> 2013-03-03 05:00:12 UTC ---
The problem started with (ppl changeover to isl) this revision
http://gcc.gnu.org/viewcvs?view=revision&revision=189156

[Bug tree-optimization/56145] [4.8/4.9 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.8.0                       |4.8.1

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-03-22 14:44:04 UTC ---
GCC 4.8.0 is being released, adjusting target milestone.

[Bug tree-optimization/56145] [4.8/4.9 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.8.1                       |4.8.2

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 4.8.1 has been released.

[Bug tree-optimization/56145] [4.8/4.9 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.8.2                       |4.8.3

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 4.8.2 has been released.

[Bug tree-optimization/56145] [4.8/4.9 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW

--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed on x86_64 with some 4.8 version, reducing.

[Bug tree-optimization/56145] [4.8/4.9 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|x86_64-w64-mingw32          |
           Priority|P3                          |P2
                 CC|                            |grosser at gcc dot gnu.org

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
Reduced testcase, fails on trunk and the 4.8 branch:

struct frame {
    int stereo;
};
struct gr_info_s {
    unsigned part2_3_length;
};
struct III_sideinfo {
    unsigned main_data_begin;
    struct {
        struct gr_info_s gr[2];
    } ch[2];
};
typedef struct mpstr_tag {
    struct frame fr;
    struct III_sideinfo sideinfo;
} MPSTR, *PMPSTR;
int decode_layer3_sideinfo(PMPSTR mp) 
{
  struct frame *fr = &mp->fr;
  int stereo = fr->stereo;
  int granules;
  int ch, gr, databits = 0;
  for (gr = 0; gr < granules; ++gr) {
      for (ch = 0; ch < stereo; ++ch) {
          struct gr_info_s *gr_infos = &(mp->sideinfo.ch[ch].gr[gr]);
          databits += gr_infos->part2_3_length;
      }
  }
  return databits - 8 * mp->sideinfo.main_data_begin;
}


Program received signal SIGSEGV, Segmentation fault.
0x000000000120c3b5 in subtract_commutative_associative_deps (scop=0x1de6ec0, 
    pbbs=..., original=0x1e22200, must_raw=0x7fffffffd5f8, 
    may_raw=0x7fffffffd5f0, must_raw_no_source=0x0, may_raw_no_source=0x0, 
    must_war=0x7fffffffd5e8, may_war=0x7fffffffd5e0, must_war_no_source=0x0, 
    may_war_no_source=0x0, must_waw=0x7fffffffd5d8, may_waw=0x7fffffffd5d0, 
    must_waw_no_source=0x0, may_waw_no_source=0x0)
    at /space/rguenther/src/svn/trunk/gcc/graphite-dependences.c:424
424                                                          
x_must_raw_no_source);
(gdb) l
419             gcc_assert (res == 0);
420
421             *must_raw = isl_union_map_subtract (*must_raw, x_must_raw);
422             *may_raw = isl_union_map_subtract (*may_raw, x_may_raw);
423             *must_raw_no_source = isl_union_map_subtract
(*must_raw_no_source,
424                                                          
x_must_raw_no_source);
425             *may_raw_no_source = isl_union_map_subtract
(*may_raw_no_source,
426                                                         
x_may_raw_no_source);
427             *must_war = isl_union_map_subtract (*must_war, x_must_war);
428             *may_war = isl_union_map_subtract (*may_war, x_may_war);
(gdb) p must_raw_no_source
$2 = (isl_union_map **) 0x0

which means that

        res = isl_union_map_compute_flow (isl_union_map_copy (r),
                                          isl_union_map_copy (must_w),
                                          isl_union_map_copy (may_w),
                                          isl_union_map_copy (original),
                                          &x_must_raw, &x_may_raw,
                                          &x_must_raw_no_source,
                                          &x_may_raw_no_source);
        gcc_assert (res == 0);

doesn't compute it but returns no error either.  Tobias?  maybe
that's valid and we simply should not do anything with that set then?

[Bug tree-optimization/56145] [4.8/4.9/4.10 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

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/56145] [4.8/4.9/5 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

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/56145] [4.8/4.9/5 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

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

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

--- Comment #13 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Doesn't ICE for me on the trunk with ISL 0.14.  Richard, can you still
reproduce this?

[Bug tree-optimization/56145] [4.8/4.9/5 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[mircea.namolaru at inria dot fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

Mircea Namolaru <mircea.namolaru at inria dot fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mircea.namolaru at inria dot fr

--- Comment #14 from Mircea Namolaru <mircea.namolaru at inria dot fr> ---
Didn't succeed to reproduce the problem on Linux.

The flag loop->can_be_parallel set by Graphite for parallelization is computed
in a different manner. This is made possible by the new ISL based code
generation that annotates the AST with relevant information for loop nodes. 

I think that is possible to close this problem.

[Bug tree-optimization/56145] [4.8/4.9 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |5.0
            Summary|[4.8/4.9/5 Regression] Use  |[4.8/4.9 Regression] Use of
                   |of too much optimizations   |too much optimizations -O2
                   |-O2 -ffast-math             |-ffast-math
                   |-floop-parallelize-all      |-floop-parallelize-all

--- Comment #15 from Richard Biener <rguenth at gcc dot gnu.org> ---
Yes.  I think this has been fixed on trunk but not the branches with

2014-06-29  Roman Gareev  <gareevroman@gmail.com>

        * graphite-dependences.c (subtract_commutative_associative_deps):
        Add NULL checking of the following variables: must_raw_no_source,
        may_raw_no_source, must_war_no_source, may_war_no_source,
        must_waw_no_source, may_waw_no_source, must_raw, may_raw,
        must_war, may_war, must_waw, may_waw.

that change should probably be backported to the 4.9 and 4.8 branches.

[Bug tree-optimization/56145] [4.8/4.9 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[mircea.namolaru at inria dot fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

--- Comment #16 from Mircea Namolaru <mircea.namolaru at inria dot fr> ---
Right, the NULL check fixed this for previous versions of GCC.

For the current version, it works without these NULL checks (the NULL paths are
not followed). The relevant scop fields are always initialized to empty maps,
so they are never NULL. The problematic call to compute_deps where the NULL
problem occured was from parallelization check code
(compute_loop_level_carries_dependencies), code no longer in GCC.

[Bug tree-optimization/56145] [4.8/4.9 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.8.5                       |4.9.3

--- Comment #17 from Richard Biener <rguenth at gcc dot gnu.org> ---
The gcc-4_8-branch is being closed, re-targeting regressions to 4.9.3.

[Bug tree-optimization/56145] [4.9 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

--- Comment #18 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 4.9.3 has been released.

[Bug tree-optimization/56145] [4.9 Regression] Use of too much optimizations -O2 -ffast-math -floop-parallelize-all

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56145

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.9.3                       |4.9.4