[Bug c++/106219] New: ice in vect_do_peeling, at tree-vect-loop-manip.cc:2703

[Bug c++/106219] New: ice in vect_do_peeling, at tree-vect-loop-manip.cc:2703

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

            Bug ID: 106219
           Summary: ice in vect_do_peeling, at
                    tree-vect-loop-manip.cc:2703
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: dcb314 at hotmail dot com
  Target Milestone: ---

Created attachment 53263
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53263&action=edit
gzipped C++ source code

The attached C++ code, when compiled as follows:

$ ../results/bin/g++ -g -O3 -march=native  -c   -fno-exceptions bug825.cc
during GIMPLE pass: vect
In file included from /usr/include/hwy/foreach_target.h:102,
                 from
/home/dcb36/rpmbuild/BUILD/libjxl-0.6.1/lib/jxl/gauss_blur.cc:16:
/home/dcb36/rpmbuild/BUILD/libjxl-0.6.1/lib/jxl/gauss_blur.cc: In function
‘jxl::ImageF jxl::N_AVX2::ConvolveXSa
mpleAndTranspose(const jxl::ImageF&, const std::vector<float>&, size_t)’:
/home/dcb36/rpmbuild/BUILD/libjxl-0.6.1/lib/jxl/gauss_blur.cc:392:8: internal
compiler error: in vect_do_peeling
, at tree-vect-loop-manip.cc:2703
  392 | ImageF ConvolveXSampleAndTranspose(const ImageF& in,
      |        ^~~~~~~~~~~~~~~~~~~~~~~~~~~
0x1396fb1 vect_do_peeling(_loop_vec_info*, tree_node*, tree_node*, tree_node**,
tree_node**, tree_node**, int, b
ool, bool, tree_node**)
        ../../trunk.git/gcc/tree-vect-loop-manip.cc:2702
0x13814ed vect_transform_loop(_loop_vec_info*, gimple*)
        ../../trunk.git/gcc/tree-vect-loop.cc:9877

I will have my usual go at reducing the code and finding a git hash range.

[Bug tree-optimization/106219] [13 Regression] ice in vect_do_peeling, at tree-vect-loop-manip.cc:2703

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|ice in vect_do_peeling, at  |[13 Regression] ice in
                   |tree-vect-loop-manip.cc:270 |vect_do_peeling, at
                   |3                           |tree-vect-loop-manip.cc:270
                   |                            |3
   Target Milestone|---                         |13.0

[Bug tree-optimization/106219] [13 Regression] ice in vect_do_peeling, at tree-vect-loop-manip.cc:2703

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
from the preprocessed source:
/home/dcb/gcc/results.20220705.asan.ubsan/libexec/gcc/x86_64-pc-linux-gnu/13.0.0/cc1plus
-quiet -I /home/dcb36/rpmbuild/BUILD/libjxl-0.6.1 -I
/home/dcb36/rpmbuild/BUILD/libjxl-0.6.1/lib/include -I
/home/dcb36/rpmbuild/BUILD/libjxl-0.6.1/redhat-linux-build/lib/include -MD
CMakeFiles/jxl_dec-obj.dir/jxl/gauss_blur.cc.d -MF
CMakeFiles/jxl_dec-obj.dir/jxl/gauss_blur.cc.o.d -MT
lib/CMakeFiles/jxl_dec-obj.dir/jxl/gauss_blur.cc.o -D_GNU_SOURCE -D
JPEGXL_MAJOR_VERSION=0 -D JPEGXL_MINOR_VERSION=6 -D JPEGXL_PATCH_VERSION=1 -D
JXL_INTERNAL_LIBRARY_BUILD -D __DATE__=redacted -D __TIMESTAMP__=redacted -D
__TIME__=redacted -D NDEBUG -D JPEGXL_ENABLE_SKCMS=1 -D JPEGXL_BUNDLE_SKCMS=1
-D_FORTIFY_SOURCE=2 -D_GLIBCXX_ASSERTIONS
/home/dcb36/rpmbuild/BUILD/libjxl-0.6.1/lib/jxl/gauss_blur.cc -march=bdver2
-mmmx -mpopcnt -msse -msse2 -msse3 -mssse3 -msse4.1 -msse4.2 -mavx -mno-avx2
-msse4a -mfma4 -mxop -mfma -mno-avx512f -mbmi -mno-bmi2 -maes -mpclmul
-mno-avx512vl -mno-avx512bw -mno-avx512dq -mno-avx512cd -mno-avx512er
-mno-avx512pf -mno-avx512vbmi -mno-avx512ifma -mno-avx5124vnniw
-mno-avx5124fmaps -mno-avx512vpopcntdq -mno-avx512vbmi2 -mno-gfni
-mno-vpclmulqdq -mno-avx512vnni -mno-avx512bitalg -mno-avx512bf16
-mno-avx512vp2intersect -mno-3dnow -mno-adx -mabm -mno-cldemote -mno-clflushopt
-mno-clwb -mno-clzero -mcx16 -mno-enqcmd -mf16c -mno-fsgsbase -mfxsr -mno-hle
-msahf -mlwp -mlzcnt -mno-movbe -mno-movdir64b -mno-movdiri -mno-mwaitx
-mno-pconfig -mno-pku -mno-prefetchwt1 -mprfchw -mno-ptwrite -mno-rdpid
-mno-rdrnd -mno-rdseed -mno-rtm -mno-serialize -mno-sgx -mno-sha -mno-shstk
-mtbm -mno-tsxldtrk -mno-vaes -mno-waitpkg -mno-wbnoinvd -mxsave -mno-xsavec
-mno-xsaveopt -mno-xsaves -mno-amx-tile -mno-amx-int8 -mno-amx-bf16 -mno-uintr
-mno-hreset -mno-kl -mno-widekl -mno-avxvnni -mno-avx512fp16 --param
l1-cache-size=16 --param l1-cache-line-size=64 --param l2-cache-size=2048
-mtune=bdver2 -quiet -dumpdir CMakeFiles/jxl_dec-obj.dir/jxl/ -dumpbase
gauss_blur.cc.cc -dumpbase-ext .cc -g -O3 -Werror=format-security
-Wno-builtin-macro-redefined -Wall -Wextra -Wc++11-compat -Warray-bounds
-Wformat-security -Wimplicit-fallthrough=3 -Wno-register -Wno-unused-function
-Wno-unused-parameter -Wnon-virtual-dtor -Woverloaded-virtual=2 -Wvla
-std=c++11 -ffat-lto-objects -fstack-protector-strong
-fasynchronous-unwind-tables -fstack-clash-protection -funwind-tables
-fno-omit-frame-pointer -fPIC -fvisibility=hidden -fvisibility-inlines-hidden
-fmerge-all-constants -fno-builtin-fwrite -fno-builtin-fread
-fsized-deallocation -fno-exceptions -fmath-errno -o - -frandom-seed=0
-fdump-noaddr

[Bug tree-optimization/106219] [13 Regression] ice in vect_do_peeling, at tree-vect-loop-manip.cc:2703

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

--- Comment #2 from David Binderman <dcb314 at hotmail dot com> ---
Git hash range seems to be f843bea4ca5613cb..683f11843974f0bd, which is 63
commits.

[Bug tree-optimization/106219] [13 Regression] ice in vect_do_peeling, at tree-vect-loop-manip.cc:2703

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

--- Comment #3 from David Binderman <dcb314 at hotmail dot com> ---
Created attachment 53264
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53264&action=edit
gzipped C++ source code

After about 100 minutes of computation, I provide the partly reduced code.

The cvise line I am using is:

(/usr/bin/g++ -c -w -O1 -march=native bug825.cc &&
/home/dcb/gcc/results/bin/g++ -c -w -g -O3 -march=native  -fno-exceptions
bug825.cc) 2>&1 | fgrep "internal compiler error: in vect_do_peeling, at
tree-vect-loop-manip.cc:2703"

native is bdver2. If no one else picks this up, I will continue tomorrow.

[Bug tree-optimization/106219] [13 Regression] ice in vect_do_peeling, at tree-vect-loop-manip.cc:2703

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

--- Comment #4 from David Binderman <dcb314 at hotmail dot com> ---
After another 100 minutes or so of reduction, I got this:

int max(int __b) {
  if (0 < __b)
    return __b;
  return 0;
}
struct Plane {
  using T = int;
  Plane(int, int);
  T *Row();
};
using ImageF = Plane;
#pragma GCC target "sse2,ssse3"                                               
\
                   ",avx,avx2"
float *ConvolveXSampleAndTranspose_rowp;
int ConvolveXSampleAndTranspose_res, ConvolveXSampleAndTranspose_r;
void ConvolveXSampleAndTranspose() {
  ImageF out(0, ConvolveXSampleAndTranspose_res);
  for (int y;;) {
    float sum;
    for (int i = ConvolveXSampleAndTranspose_r; i; ++i)
      sum += i;
    for (; ConvolveXSampleAndTranspose_r; ++ConvolveXSampleAndTranspose_r)
      sum +=
          ConvolveXSampleAndTranspose_rowp[max(ConvolveXSampleAndTranspose_r)]
*
          ConvolveXSampleAndTranspose_r;
    out.Row()[y] = sum;
  }
}

$ /home/dcb/gcc/results/bin/g++ -c -w -O3 -march=bdver2   bug825.cc
during GIMPLE pass: vect
bug825.cc: In function ‘void ConvolveXSampleAndTranspose()’:
bug825.cc:16:6: internal compiler error: in vect_do_peeling, at
tree-vect-loop-manip.cc:2703
   16 | void ConvolveXSampleAndTranspose() {
      |      ^~~~~~~~~~~~~~~~~~~~~~~~~~~
0x1396fb1 vect_do_peeling(_loop_vec_info*, tree_node*, tree_node*, tree_node**,
tree_node**, tree_node**, int, bool, bool, tree_node**)
        ../../trunk.git/gcc/tree-vect-loop-manip.cc:2702

I have a bunch of other test cases which demonstrate this bug. I will reduce
the smallest of those.

[Bug tree-optimization/106219] [13 Regression] ICE in vect_do_peeling, at tree-vect-loop-manip.cc:2703 since r13-1457-g10b502fb78351a40

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
           Priority|P3                          |P1
                 CC|                            |rguenth at gcc dot gnu.org
            Summary|[13 Regression] ice in      |[13 Regression] ICE in
                   |vect_do_peeling, at         |vect_do_peeling, at
                   |tree-vect-loop-manip.cc:270 |tree-vect-loop-manip.cc:270
                   |3                           |3 since
                   |                            |r13-1457-g10b502fb78351a40
   Last reconfirmed|                            |2022-07-07
             Status|UNCONFIRMED                 |NEW

--- Comment #5 from Martin Liška <marxin at gcc dot gnu.org> ---
Started with r13-1457-g10b502fb78351a40.

[Bug tree-optimization/106219] [13 Regression] ICE in vect_do_peeling, at tree-vect-loop-manip.cc:2703 since r13-1457-g10b502fb78351a40

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

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 #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Thanks - will have a look.

[Bug target/106219] [13 Regression] ICE in vect_do_peeling, at tree-vect-loop-manip.cc:2703 since r13-1457-g10b502fb78351a40

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-*-*
          Component|tree-optimization           |target
                 CC|                            |crazylht at gmail dot com

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
So it's __builtin_ia32_gathersiv4sf not marked as PURE which means we think it
writes to memory.

I see

  /* AVX512PF */
  def_builtin (OPTION_MASK_ISA_AVX512PF, 0, "__builtin_ia32_gatherpfdpd",
               VOID_FTYPE_QI_V8SI_PCVOID_INT_INT,
               IX86_BUILTIN_GATHERPFDPD);
  def_builtin (OPTION_MASK_ISA_AVX512PF, 0, "__builtin_ia32_gatherpfdps",
               VOID_FTYPE_HI_V16SI_PCVOID_INT_INT,
               IX86_BUILTIN_GATHERPFDPS);
  def_builtin (OPTION_MASK_ISA_AVX512PF, 0, "__builtin_ia32_gatherpfqpd",
               VOID_FTYPE_QI_V8DI_PCVOID_INT_INT,
               IX86_BUILTIN_GATHERPFQPD);
  def_builtin (OPTION_MASK_ISA_AVX512PF, 0, "__builtin_ia32_gatherpfqps",
               VOID_FTYPE_QI_V8DI_PCVOID_INT_INT,
               IX86_BUILTIN_GATHERPFQPS);

shouldn't those use def_builtin_pure as well?

Tracing down why def_builtin_pure doesn't work here.

[Bug target/106219] [13 Regression] ICE in vect_do_peeling, at tree-vect-loop-manip.cc:2703 since r13-1457-g10b502fb78351a40

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

--- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> ---
Testing a patch for the original issue.

[Bug target/106219] [13 Regression] ICE in vect_do_peeling, at tree-vect-loop-manip.cc:2703 since r13-1457-g10b502fb78351a40

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

--- Comment #9 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:8bc243943765f969c0d4625721a9944e9516d013

commit r13-1553-g8bc243943765f969c0d4625721a9944e9516d013
Author: Richard Biener <rguenther@suse.de>
Date:   Thu Jul 7 10:46:01 2022 +0200

    target/106219 - proprly mark builtins pure via ix86_add_new_builtins

    The target optimize pragma path to initialize extra target specific
    builtins missed handling of the pure_p flag which in turn causes
    extra clobber side-effects of gather builtins leading to unexpected
    issues downhill.

            PR target/106219
            * config/i386/i386-builtins.cc (ix86_add_new_builtins): Properly
            set DECL_PURE_P.

            * g++.dg/pr106219.C: New testcase.

[Bug target/106219] [13 Regression] ICE in vect_do_peeling, at tree-vect-loop-manip.cc:2703 since r13-1457-g10b502fb78351a40

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

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

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

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.

[Bug target/106219] [13 Regression] ICE in vect_do_peeling, at tree-vect-loop-manip.cc:2703 since r13-1457-g10b502fb78351a40

[seurer at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

seurer at gcc dot gnu.org changed:

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

--- Comment #11 from seurer at gcc dot gnu.org ---
This may be the same problem we are seeing which is causing a bootstrap issue
on power 7 BE.  

g:10b502fb78351a4073b6682c026a92c82d3da6c5, r13-1457-g10b502fb78351a 

libtool: compile:  /home/seurer/gcc/git/build/gcc-trunk/./gcc/xgcc
-B/home/seurer/gcc/git/build/gcc-trunk/./gcc/
-B/home/seurer/gcc/git/install/gcc-trunk/powerpc64-unknown-linux-gnu/bin/
-B/home/seurer/gcc/git/install/gcc-trunk/powerpc64-unknown-linux-gnu/lib/
-isystem
/home/seurer/gcc/git/install/gcc-trunk/powerpc64-unknown-linux-gnu/include
-isystem
/home/seurer/gcc/git/install/gcc-trunk/powerpc64-unknown-linux-gnu/sys-include
-DHAVE_CONFIG_H -I. -I/home/seurer/gcc/git/gcc-trunk/libgfortran
-iquote/home/seurer/gcc/git/gcc-trunk/libgfortran/io
-I/home/seurer/gcc/git/gcc-trunk/libgfortran/../gcc
-I/home/seurer/gcc/git/gcc-trunk/libgfortran/../gcc/config
-I/home/seurer/gcc/git/gcc-trunk/libgfortran/../libquadmath -I../.././gcc
-I/home/seurer/gcc/git/gcc-trunk/libgfortran/../libgcc -I../libgcc
-I/home/seurer/gcc/git/gcc-trunk/libgfortran/../libbacktrace -I../libbacktrace
-I../libbacktrace -std=gnu11 -Wall -Wstrict-prototypes -Wmissing-prototypes
-Wold-style-definition -Wextra -Wwrite-strings
-Werror=implicit-function-declaration -Werror=vla -fcx-fortran-rules
-ffunction-sections -fdata-sections -ffast-math -ftree-vectorize -funroll-loops
--param max-unroll-times=4 -g -O2 -MT matmul_i1.lo -MD -MP -MF
.deps/matmul_i1.Tpo -c
/home/seurer/gcc/git/gcc-trunk/libgfortran/generated/matmul_i1.c  -fPIC -DPIC
-o .libs/matmul_i1.o
during GIMPLE pass: vect
/home/seurer/gcc/git/gcc-trunk/libgfortran/generated/matmul_i1.c: In function
'matmul_i1':
/home/seurer/gcc/git/gcc-trunk/libgfortran/generated/matmul_i1.c:2450:1:
internal compiler error: in vect_do_peeling, at tree-vect-loop-manip.cc:2690
 2450 | matmul_i1 (gfc_array_i1 * const restrict retarray,
      | ^~~~~~~~~
0x10fbc6e3 vect_do_peeling(_loop_vec_info*, tree_node*, tree_node*,
tree_node**, tree_node**, tree_node**, int, bool, bool, tree_node**)
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vect-loop-manip.cc:2690
0x10faafff vect_transform_loop(_loop_vec_info*, gimple*)
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vect-loop.cc:9747
0x10ff354b vect_transform_loops
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vectorizer.cc:1006
0x10ff3de3 try_vectorize_loop_1
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vectorizer.cc:1136
0x10ff3de3 try_vectorize_loop
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vectorizer.cc:1165
0x10ff4553 execute
        /home/seurer/gcc/git/gcc-trunk/gcc/tree-vectorizer.cc:1281



commit 10b502fb78351a4073b6682c026a92c82d3da6c5
Author: Richard Biener <rguenther@suse.de>
Date:   Mon Jul 4 12:36:05 2022 +0200

    Keep virtual SSA up-to-date in vectorizer

[Bug target/106219] [13 Regression] ICE in vect_do_peeling, at tree-vect-loop-manip.cc:2703 since r13-1457-g10b502fb78351a40

[seurer at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106219

--- Comment #12 from seurer at gcc dot gnu.org ---
Note: The fix did not take care of the problem on power 7 but it does trace to
the same commit.