[Bug tree-optimization/98443] New: Failure to optimize out vector operations into a constant when possible

[Bug tree-optimization/98443] New: Failure to optimize out vector operations into a constant when possible

[gabravier at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98443

            Bug ID: 98443
           Summary: Failure to optimize out vector operations into a
                    constant when possible
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gabravier at gmail dot com
  Target Milestone: ---

typedef char A __attribute__((vector_size(32)));

int foo(A w)
{
    w *= (A){0, 1};
    return w[0];
}

This can be optimized into `return 0;`. This transformation is done by LLVM,
but not by GCC.

[Bug tree-optimization/98443] Failure to optimize out vector operations into a constant when possible

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98443

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

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

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
We'd probably need VRP on the vector elements for this.

[Bug tree-optimization/98443] Failure to optimize out vector operations into a constant when possible

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98443

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2021-01-05

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
More like value numbering vector components and/or known-zero/one-bits.
Possibly the CCP bit handling can be reasonably easy extended to handle this
particular case (integer vectors).

[Bug tree-optimization/98443] Failure to optimize out vector operations into a constant when possible

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98443

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Or we could push the BIT_FIELD_REF down the chain:
  _1 = w_3(D) & { 0, -1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 };
  _2 = BIT_FIELD_REF <_1, 8, 0>;

Which is exactly what LLVM does.
Take:
typedef char A __attribute__((vector_size(32)));

int foo(A w)
{
    w *= (A){2, 3,4,5};
    return w[0];
}

LLVM gets:
        vmovd   %xmm0, %eax
        addb    %al, %al
        movsbl  %al, %eax
        vzeroupper
        retq