[Bug tree-optimization/104551] New: Wrong code with -O3 for skylake-avx512, icelake-server, and sapphirerapids

[Bug tree-optimization/104551] New: Wrong code with -O3 for skylake-avx512, icelake-server, and sapphirerapids

[vsevolod.livinskij at frtk dot ru]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104551

            Bug ID: 104551
           Summary: Wrong code with -O3 for skylake-avx512,
                    icelake-server, and sapphirerapids
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: vsevolod.livinskij at frtk dot ru
  Target Milestone: ---

Reproducer (I was not able to merge it into a single file):
// func.cpp
extern unsigned var_11, var_178;
extern unsigned char arr_492[];
void test() {
  for (bool a = 0; a < (bool)var_11; a = 2)
    for (unsigned b = 0; b < var_11;
         b += 1)
      var_178 = arr_492[b] ? arr_492[b] : (char) b;
}

// driver.cpp 
unsigned int var_11 = 16U;
unsigned int var_178 = 0;
unsigned char arr_492 [16];

void test();

int main() {
    for (int i = 0; i < 16; ++i)
        arr_492 [i] = (unsigned char)128;
    test();
    __builtin_printf("%u\n", var_178);
    if (var_178 != 128)
        __builtin_abort();
    return 0;
}

Error:
>$ g++ -O0 func.cpp driver.cpp -march=skylake-avx512 && sde -skx -- ./a.out 
128
>$ g++ -O3 func.cpp driver.cpp -march=skylake-avx512 && sde -skx -- ./a.out 
4294967168
Aborted (core dumped)

g++ (git://gcc.gnu.org/git/gcc.git:master
58aeb75d4097010ad9bb72b964265b18ab284f93) 12.0.1 20220213 (experimental)

[Bug tree-optimization/104551] [12 Regression] Wrong code with -O3 for skylake-avx512, icelake-server, and sapphirerapids

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104551

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
            Summary|Wrong code with -O3 for     |[12 Regression] Wrong code
                   |skylake-avx512,             |with -O3 for
                   |icelake-server, and         |skylake-avx512,
                   |sapphirerapids              |icelake-server, and
                   |                            |sapphirerapids
   Target Milestone|---                         |12.0
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2022-02-15

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Confirmed, single file version:
unsigned int var_11 = 16U;
unsigned int var_178 = 0;
unsigned char arr_492 [128];
[[gnu::noipa]]
void test() {
  for (bool a = 0; a < (bool)var_11; a = 2)
    for (unsigned b = 0; b < var_11;
         b += 1)
      var_178 = arr_492[b] ? arr_492[b] : (char) b;
}
int main() {
    for (int i = 0; i < 128; ++i)
        arr_492 [i] = (unsigned char)128;
    test();
    __builtin_printf("%u\n", var_178);
    if (var_178 != 128)
        __builtin_abort();
    return 0;
}

----- CUT -----

I increasing the array size to 128 allowed me to merge it, I don't know why
though; maybe because it allowed the code to be vectorize while 16 was not
going to.

[Bug tree-optimization/104551] [12 Regression] Wrong code with -O3 for skylake-avx512, icelake-server, and sapphirerapids

[crazylht at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104551

Hongtao.liu <crazylht at gmail dot com> changed:

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

--- Comment #2 from Hongtao.liu <crazylht at gmail dot com> ---
let me take a look.

[Bug tree-optimization/104551] [12 Regression] Wrong code with -O3 for skylake-avx512, icelake-server, and sapphirerapids

[crazylht at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104551

--- Comment #3 from Hongtao.liu <crazylht at gmail dot com> ---
(In reply to Hongtao.liu from comment #2)
> let me take a look.

Start with r12-7216

[Bug tree-optimization/104551] [12 Regression] Wrong code with -O3 for skylake-avx512, icelake-server, and sapphirerapids

[crazylht at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104551

--- Comment #4 from Hongtao.liu <crazylht at gmail dot com> ---
(In reply to Hongtao.liu from comment #3)
> (In reply to Hongtao.liu from comment #2)
> > let me take a look.
> 
> Start with r12-7216

  for (bool a = 0; a < (bool)var_11; a = 2)
    for (unsigned b = 0; b < var_11;
         b += 1)
      var_178 = arr_492[b] ? arr_492[b] : (char) b;

arr_492[b] is unsigned char, and (char) b is sign char,
vect_recog_cond_expr_convert_pattern shouldn't hanlde this since it's
zero_extend with sign_extend, it's not equal to

    var_178 = arr_492[b] ? (char)arr_492[b] : (char) b;

[Bug tree-optimization/104551] [12 Regression] Wrong code with -O3 for skylake-avx512, icelake-server, and sapphirerapids

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104551

--- Comment #5 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by hongtao Liu <liuhongt@gcc.gnu.org>:

https://gcc.gnu.org/g:754dce903ca28c4c2f2bc8614a8de5e631655f2e

commit r12-7276-g754dce903ca28c4c2f2bc8614a8de5e631655f2e
Author: liuhongt <hongtao.liu@intel.com>
Date:   Wed Feb 16 12:15:18 2022 +0800

    Restrict the two sources of vect_recog_cond_expr_convert_pattern to be of
the same type when convert is extension.

    It's not equal to transform

     (cond (cmp @1 @2) (convert@3 @4) (convert@5 @6))

     to

     (convert (cmp @1 @2) (convert)@4 @6)

    when(convert@3 @4) is extension because it's zero_extend vs sign_extend.

    gcc/ChangeLog:

            PR tree-optimization/104551
            PR tree-optimization/103771
            * match.pd (cond_expr_convert_p): Add types_match check when
            convert is extension.
            * tree-vect-patterns.cc
            (gimple_cond_expr_convert_p): Adjust comments.
            (vect_recog_cond_expr_convert_pattern): Ditto.

    gcc/testsuite/ChangeLog:

            * gcc.target/i386/pr104551.c: New test.

[Bug tree-optimization/104551] [12 Regression] Wrong code with -O3 for skylake-avx512, icelake-server, and sapphirerapids

[crazylht at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104551

--- Comment #6 from Hongtao.liu <crazylht at gmail dot com> ---
Fixed in GCC12.

[Bug tree-optimization/104551] [12 Regression] Wrong code with -O3 for skylake-avx512, icelake-server, and sapphirerapids

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104551

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

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

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Fixed so closing.