[Bug tree-optimization/94878] New: Failure to optimize div with bls/or pattern

[Bug tree-optimization/94878] New: Failure to optimize div with bls/or pattern

[gabravier at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94878

            Bug ID: 94878
           Summary: Failure to optimize div with bls/or pattern
           Product: gcc
           Version: 10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gabravier at gmail dot com
  Target Milestone: ---

unsigned y(unsigned x)
{
    unsigned s = x & -x;
    return s | (x % s);
}

With -O3, LLVM outputs : 

y(unsigned int):
  blsi ecx, edi
  lea eax, [rcx - 1]
  and eax, edi
  or eax, ecx
  ret

GCC outputs :

y(unsigned int):
  mov eax, edi
  xor edx, edx
  blsi ecx, edi
  div ecx
  mov eax, edx
  or eax, ecx
  ret

If a single change to this function is applied, changing it to this :

unsigned y(unsigned x)
{
    unsigned s = x & -x;
    return x | (x % s);
}

It's equivalent to `return x`, and LLVM recognises this : 

y(unsigned int):
  mov eax, edi
  ret

While GCC does not :

y(unsigned int):
  mov eax, edi
  xor edx, edx
  blsi ecx, edi
  div ecx
  mov eax, edx
  or eax, edi
  ret

[Bug tree-optimization/94878] Failure to optimize div with bls/or pattern

[gabravier at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94878

--- Comment #1 from Gabriel Ravier <gabravier at gmail dot com> ---
Also, the assembly outputs are for when compiling with with `-mbmi` but that
should not affect the bug itself

[Bug tree-optimization/94878] Failure to optimize div with bls/or pattern

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94878

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
           Keywords|                            |missed-optimization
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2020-04-30
             Target|                            |x86_64-*-* i?86-*-*

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.

[Bug tree-optimization/94878] Failure to optimize div with bls/or pattern

[gabravier at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94878

Gabriel Ravier <gabravier at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|x86_64-*-* i?86-*-*         |

--- Comment #3 from Gabriel Ravier <gabravier at gmail dot com> ---
I don't think this is x86-specific. Division is expensive on all processors
(except maybe 1 or two exceptions), so this should apply to all processors, and
the second transformation should apply to literally all all processors. Or do
you have confirmed that this only occurs on x86 ?

[Bug tree-optimization/94878] Failure to optimize div with bls/or pattern

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94878

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement