[Bug fortran/115563] New: Unnecessary brackets prevent fortran vectorisation

[Bug fortran/115563] New: Unnecessary brackets prevent fortran vectorisation

[mjr19 at cam dot ac.uk]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115563

            Bug ID: 115563
           Summary: Unnecessary brackets prevent fortran vectorisation
           Product: gcc
           Version: 14.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: mjr19 at cam dot ac.uk
  Target Milestone: ---

The code

subroutine foo(a,n)

  complex (kind(1d0)) :: a(*)
  integer :: i,n

!$OMP SIMD
  do i=1,n
     a(i)=(a(i)+(6d0,1d0))
  enddo

end subroutine foo

compiled with

$ gfortran-14 -S -O3 -ffast-math -fopenmp-simd  -mavx2 t3.f90

does not vectorise.

$ grep -c ymm t3.s 
0

If the extra brackets are removed, so that the loop line reads

     a(i)=a(i)+(6d0,1d0)

then it does.

$ grep -c ymm t3.s 
3

It seems surprising that the code produced should differ depending on whether
the brackets are present, as I do not believe that they alter the meaning of
the code in any way. One can construct many similar examples, e.g. using a
complex variable in the place of the (6d0,1d0), or using a real constant, etc.

[Bug fortran/115563] Unnecessary brackets prevent fortran vectorisation

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115563

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

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

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

  _11 = ((_9));
  _19 = REALPART_EXPR <_11>;
  _20 = IMAGPART_EXPR <_11>;
  REALPART_EXPR <(*a_7(D))[_6]> = _19;
  IMAGPART_EXPR <(*a_7(D))[_6]> = _20;

[Bug fortran/115563] Unnecessary brackets prevent fortran vectorisation

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115563

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |DUPLICATE
             Status|UNCONFIRMED                 |RESOLVED

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

*** This bug has been marked as a duplicate of bug 68855 ***

[Bug fortran/115563] Unnecessary brackets prevent fortran vectorisation

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115563

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note the full IR:

  _9 = COMPLEX_EXPR <_15, _14>;
  _11 = ((_9));
  _19 = REALPART_EXPR <_11>;
  _20 = IMAGPART_EXPR <_11>;
  REALPART_EXPR <(*a_7(D))[_6]> = _19;
  IMAGPART_EXPR <(*a_7(D))[_6]> = _20;

Yes it is a dup.
// (()) is PAREN_EXPR .

[Bug fortran/115563] Unnecessary brackets prevent fortran vectorisation

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115563

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note I have a fix for this which I attached to PR 68855 and will be submitting
it after the bootstrap/test finishes. Thanks again for the testcase and the
decent bug report. It definitely was useful.

[Bug fortran/115563] Unnecessary brackets prevent fortran vectorisation

[mjr19 at cam dot ac.uk]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115563

--- Comment #5 from mjr19 at cam dot ac.uk ---
I'm glad this was useful, and thanks for the impressively rapid fix. I stumbled
across this by chance whilst trying to construct a minimal example for a rather
different missed vectorisation case.

[Bug fortran/115563] Unnecessary brackets prevent fortran vectorisation

[mjr19 at cam dot ac.uk]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115563

--- Comment #6 from mjr19 at cam dot ac.uk ---
A further comment to aid others reading this report. It is not just unnecessary
brackets which used to prevent vectorisation, but also necessary ones.

subroutine foo(a,b,c,n)
  complex (kind(1d0)) :: a(*),b,c
  integer :: i,n

  do i=1,n
     a(i)=(a(i)+b)*c
  enddo
end subroutine foo

does not vectorise with gfortran-14, but does with gfortran-15.0-20240623.

The performance increase in loops making extensive use of complex variables can
therefore be quite significant -- fifty percent or more.

The almost-equivalent C code of

void foo(_Complex double *a, _Complex double b, _Complex double c, int n){
  int i;
  for(i=0;i<n;i++) a[i]=(a[i]+b)*c;
}

does vectorise with gcc-14 (and earlier versions).

I still find gfortran to be slower than ifort for code using complex variables,
as ifort is better at flipping alternate signs in a vector and generally
alternating nops in a vector, see PR 114324 and PR 114767, but this is a very
welcome improvement. Many thanks.