[Bug tree-optimization/113574] New: wrong code with shift and _BitInt(1) at any opt level

[Bug tree-optimization/113574] New: wrong code with shift and _BitInt(1) at any opt level

[zsojka at seznam dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113574

            Bug ID: 113574
           Summary: wrong code with shift and _BitInt(1) at any opt level
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: zsojka at seznam dot cz
                CC: jakub at gcc dot gnu.org
  Target Milestone: ---
              Host: x86_64-pc-linux-gnu
            Target: x86_64-pc-linux-gnu

Created attachment 57200
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57200&action=edit
reduced testcase

Compiler output:
$ x86_64-pc-linux-gnu-gcc testcase.c && ./a.out 
Aborted

-fsanitize=undefined fixes the problem:
$ x86_64-pc-linux-gnu-gcc testcase.c -fsanitize=undefined && ./a.out
(no output)

At the ASM level, at -O1:
...
foo:
# testcase.c:7:   ub1 += (ub16 << 2) | ub8;
        or      dil, BYTE PTR ub8[rip]  # _1, ub8
        and     edi, 1  # _4,

it seems the shift is removed; changing the type of ub1 to "unsigned
_BitInt(2)" fixes this.

$ x86_64-pc-linux-gnu-gcc -v
Using built-in specs.
COLLECT_GCC=/repo/gcc-trunk/binary-latest-amd64//bin/x86_64-pc-linux-gnu-gcc
COLLECT_LTO_WRAPPER=/repo/gcc-trunk/binary-trunk-r14-8373-20240123195900-gdbc5f1f523b-checking-yes-rtl-df-extra-nobootstrap-amd64/bin/../libexec/gcc/x86_64-pc-linux-gnu/14.0.1/lto-wrapper
Target: x86_64-pc-linux-gnu
Configured with: /repo/gcc-trunk//configure --enable-languages=c,c++
--enable-valgrind-annotations --disable-nls --enable-checking=yes,rtl,df,extra
--disable-bootstrap --with-cloog --with-ppl --with-isl
--build=x86_64-pc-linux-gnu --host=x86_64-pc-linux-gnu
--target=x86_64-pc-linux-gnu --with-ld=/usr/bin/x86_64-pc-linux-gnu-ld
--with-as=/usr/bin/x86_64-pc-linux-gnu-as --disable-libstdcxx-pch
--prefix=/repo/gcc-trunk//binary-trunk-r14-8373-20240123195900-gdbc5f1f523b-checking-yes-rtl-df-extra-nobootstrap-amd64
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 14.0.1 20240123 (experimental) (GCC)

[Bug middle-end/113574] wrong code with shift and _BitInt(1) at any opt level

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113574

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2024-01-24
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Confirmed.

Either fold or convert is causing the issue.

Reduced testcase:
```
unsigned _BitInt(1)
foo(unsigned _BitInt(16) ub16)
{
  unsigned _BitInt(1)  t  = (ub16 << 2);
    // being reduced to just (unsigned _BitInt(1) ) ub16 which is wrong
  return t;
}
unsigned _BitInt(1)
foo1(unsigned _BitInt(16) ub16)
{
  unsigned _BitInt(16) t = (ub16 << 2);
  return t;
}

int
main(void)
{
  unsigned _BitInt(16) ub16 = 0xfef1;
  if (foo(ub16) != foo1(ub16))
    __builtin_abort();
  return 0;
}
```

Note match does the correct thing:
```
/* Narrow a lshift by constant.  */
(simplify
 (convert (lshift:s@0 @1 INTEGER_CST@2))
 (if (INTEGRAL_TYPE_P (type)
      && INTEGRAL_TYPE_P (TREE_TYPE (@0))
      && !integer_zerop (@2)
      && TYPE_PRECISION (type) <= TYPE_PRECISION (TREE_TYPE (@0)))
  (if (TYPE_PRECISION (type) == TYPE_PRECISION (TREE_TYPE (@0))
       || wi::ltu_p (wi::to_wide (@2), TYPE_PRECISION (type)))
   (lshift (convert @1) @2)
   (if (wi::ltu_p (wi::to_wide (@2), TYPE_PRECISION (TREE_TYPE (@0))))
    { build_zero_cst (type); }))))

```

[Bug middle-end/113574] wrong code with shift and _BitInt(1) at any opt level

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113574

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
convert uses TYPE_SIZE ....


                /* If shift count is less than the width of the truncated type,
                   really shift.  */
                if (tree_int_cst_lt (TREE_OPERAND (expr, 1), TYPE_SIZE (type)))


Really??????

That dates from the begining of git history too :).

[Bug middle-end/113574] wrong code with shift and _BitInt(1) at any opt level

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113574

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note I was thinking I might hit this with 1 sized bit field but that seems to
work:
```
struct f
{
  unsigned ub1:1;
}t;
void
foo(unsigned short ub16)
{
  t.ub1 = (ub16 << 2);
}
int
main(void)
{
  unsigned short t2 = 0xfef1;
  foo(t2);
  if (t.ub1)
    __builtin_abort();
  return 0;
}
```

[Bug middle-end/113574] wrong code with shift and _BitInt(1) at any opt level

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113574

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |jakub at gcc dot gnu.org
             Status|NEW                         |ASSIGNED

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 57207
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57207&action=edit
gcc14-pr113574.patch

Untested fix.

[Bug middle-end/113574] wrong code with shift and _BitInt(1) at any opt level

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113574

--- Comment #5 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jakub Jelinek <jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:fb1b7e2fec951ba0bf4f68fac6a16929f4f63910

commit r14-8423-gfb1b7e2fec951ba0bf4f68fac6a16929f4f63910
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Thu Jan 25 13:15:23 2024 +0100

    convert: Fix test for out of bounds shift count [PR113574]

    The following patch is miscompiled, because convert_to_integer_1 for
    LSHIFT_EXPR tests if the INTEGER_CST shift count is too high, but
    incorrectly compares it against TYPE_SIZE rather than TYPE_PRECISION.
    The type in question is unsigned _BitInt(1), which has TYPE_PRECISION 1,
    TYPE_SIZE 8, and the shift count is 2 in that case.

    2024-01-25  Jakub Jelinek  <jakub@redhat.com>

            PR middle-end/113574
            * convert.cc (convert_to_integer_1) <case LSHIFT_EXPR>: Compare
shift
            count against TYPE_PRECISION rather than TYPE_SIZE.

            * gcc.dg/torture/bitint-52.c: New test.

[Bug middle-end/113574] wrong code with shift and _BitInt(1) at any opt level

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113574

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

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

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed.