[Bug c++/106247] New: GCC12 warning in Eigen: array subscript is partly outside array bounds

[Bug c++/106247] New: GCC12 warning in Eigen: array subscript is partly outside array bounds

[fchelnokov at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106247

            Bug ID: 106247
           Summary: GCC12 warning in Eigen: array subscript is partly
                    outside array bounds
           Product: gcc
           Version: 12.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: fchelnokov at gmail dot com
  Target Milestone: ---

If one compiles the following program using Eigen 3.4.0:

#include <Eigen/Dense>

int main(){
    Eigen::Array<double, 2, 2> a;
    a.setRandom();
    Eigen::Array<double, 3, 2> b;
    b.col(0).tail(2) = a.col(1);
}


in GCC12 with command-line options: -std=c++20 -Wall -Wextra -pedantic-errors
-O2 -march=haswell
then a large error occurs ending with

/opt/compiler-explorer/gcc-12.1.0/lib/gcc/x86_64-linux-gnu/12.1.0/include/avxintrin.h:893:24:
warning: array subscript '__m256d_u[0]' is partly outside array bounds of
'Eigen::Array<double, 2, 2> [1]' [-Warray-bounds]
  893 |   return *(__m256d_u *)__P;
      |                        ^~~

Similar errors occur with other -march options or other code using Eigen
library.

Everything is fine in GCC 11 or other compilers. Online demo:
https://gcc.godbolt.org/z/hT348GYo9

Is it Eigen or GCC bug? Related discussion:
https://stackoverflow.com/q/72871100/7325599

[Bug tree-optimization/106247] GCC12 warning in Eigen: array subscript is partly outside array bounds

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106247

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|                            |x86_64-linux-gnu
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2022-07-10
           Keywords|                            |diagnostic
          Component|c++                         |tree-optimization

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Can you attach the preprocessed source?

[Bug tree-optimization/106247] GCC12 warning in Eigen: array subscript is partly outside array bounds

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106247

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
the warning is correct in the sense the load is there in IR, though it looks
like it is dead (but only because b and a are unused):
  array.7_137 = (long unsigned int) &MEM <double[6]> [(void *)&b + 8B];
  _139 = array.7_137 >> 3;
  _140 = -_139;
  _141 = _140 & 3;
  first_142 = (long int) _141;
  _143 = MIN_EXPR <first_142, 2>;
  _145 = 2 - _143;
  alignedEnd_148 = _143;
  goto <bb 4>; [100.00%]

  <bb 3> [local count: 8655976563]:
  index.12_166 = (long unsigned int) index_165;
  _167 = index.12_166 * 8;
  _168 = &MEM <double[4]> [(void *)&a + 16B] + _167;
  _171 = &MEM <double[6]> [(void *)&b + 8B] + _167;
  _172 = *_168;
  MEM[(double &)_171] = _172;
  index_173 = index_165 + 1;

  <bb 4> [local count: 9725816346]:
  # index_165 = PHI <0(2), index_173(3)>
  if (_143 > index_165)
    goto <bb 3>; [89.00%]
  else
    goto <bb 6>; [11.00%]

  <bb 5> [local count: 8655976563]:
  index.10_152 = (long unsigned int) index_151;
  _153 = index.10_152 * 8;
  _154 = &MEM <double[4]> [(void *)&a + 16B] + _153;
  _155 = MEM[(__m256d_u * {ref-all})_154];
  _158 = &MEM <double[6]> [(void *)&b + 8B] + _153;
  MEM[(__m256d * {ref-all})_158] = _155;
  index_159 = index_151 + 4;

  <bb 6> [local count: 9725816346]:
  # index_151 = PHI <_143(4), index_159(5)>
  if (alignedEnd_148 > index_151)
    goto <bb 5>; [89.00%]
  else
    goto <bb 8>; [11.00%]

[Bug tree-optimization/106247] GCC12 warning in Eigen: array subscript is partly outside array bounds

[fchelnokov at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106247

Fedor Chelnokov <fchelnokov at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |fchelnokov at gmail dot com

--- Comment #3 from Fedor Chelnokov <fchelnokov at gmail dot com> ---
Created attachment 53284
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53284&action=edit
Preprocessed source

Preprocessed source is attached.

[Bug tree-optimization/106247] GCC12 warning in Eigen: array subscript is partly outside array bounds

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106247

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|1                           |0
             Status|WAITING                     |UNCONFIRMED

[Bug tree-optimization/106247] GCC12 warning in Eigen: array subscript is partly outside array bounds

[skunk at iskunk dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106247

Daniel Richard G. <skunk at iskunk dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |skunk at iskunk dot org

--- Comment #4 from Daniel Richard G. <skunk at iskunk dot org> ---
I am seeing what may be the same issue in a different context.

This is occurring in a proprietary C source codebase newly being compiled with
12.1.0, and unfortunately the problem goes away if I attempt to cut it down to
a minimal test case. Nevertheless, the actual issue is quite simple, and IMO
clear-cut.

(some names edited to protect the guilty)

In file included from /usr/include/string.h:375,
                 from foo.c:7:
In function 'function3',
    inlined from 'function2' at foo.c:1206:5,
    inlined from 'function1' at foo.c:2840:7:
foo.c:415:3: error: array subscript 'union <anonymous>[1]' is partly outside
array bounds of 'unsigned char[5]' [-Werror=array-bounds]
  415 |   memset(constraints,0,sizeof(char)*NB_CONSTRAINTS);
      |   ^~~~~~
foo.c: In function 'function1':
foo.c:380:17: note: at offset 4 into object 'constraints' of size 5
  380 |   unsigned char constraints[NB_CONSTRAINTS];
      |                 ^~~~~~~~
cc1: all warnings being treated as errors


Curiously enough, I only see this issue in an i686 build. I don't get it on
x86-64. (Dropping -march=i686 does not help.)

[Bug tree-optimization/106247] GCC12 warning in Eigen: array subscript is partly outside array bounds

[mwd at md5i dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106247

Michael Duggan <mwd at md5i dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mwd at md5i dot com

--- Comment #7 from Michael Duggan <mwd at md5i dot com> ---
I am working with some code that exhibits a similar problem.  I've run creduce
on the preprocessed code to create an exemplar, which I have attached as
bug-mwd.cpp.  Here's the relevant warnings from g++ 12.2.0 using '-std=c++14
-Wall -O2':

In member function ‘void i< <template-parameter-1-1> >::m(k*) [with ab = s]’,
    inlined from ‘void i< <template-parameter-1-1> >::m(k*) [with ab = I]’ at
bug.cpp:25:26,
    inlined from ‘i< <template-parameter-1-1> >::i(n) [with n = I*; ab = I]’ at
bug.cpp:23:39,
    inlined from ‘am I::u()’ at bug.cpp:48:3:
bug.cpp:27:5: warning: array subscript ‘i<s>::k {aka s}[0]’ is partly outside
array bounds of ‘unsigned char [40]’ [-Warray-bounds]
   27 |     e(p->aa);
      |     ^~~~~~~~
bug.cpp: In member function ‘am I::u()’:
bug.cpp:48:9: note: object of size 40 allocated by ‘operator new’
   48 |   o(new I);
      |         ^

The example is convoluted, I am afraid.  One very strange bit is that if I
change the 'o's in 'class I' to, say, 'p', I get a completely different error.

If you think this is unrelated, let me know, and I will submit a new bug
report.

[Bug tree-optimization/106247] GCC12 warning in Eigen: array subscript is partly outside array bounds

[mwd at md5i dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106247

--- Comment #8 from Michael Duggan <mwd at md5i dot com> ---
Created attachment 53533
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53533&action=edit
Reduced bug exemplar