[Bug fortran/33341] New: unnecessary stores for array constructor

[Bug fortran/33341] New: unnecessary stores for array constructor

[tkoenig at gcc dot gnu dot org]

The following idiom causes unnecessary runtime
overhead:

$ cat compare.f90
function foo(a,b,c,d)
  logical :: foo
  integer, intent(in):: a,b,c,d

  foo = all((/ a, b, c /) /= d)
end function foo
$ gfortran -fdump-tree-optimized -O3 -S compare.f90

The *.optimized file shows

<bb 2>:
  D.521 = (int4[0:] *) &A.2;
  D.522 = *a;
  (*D.521)[0] = D.522;
  D.523 = *b;
  (*D.521)[1] = D.523;
  D.524 = *c;
  (*D.521)[2] = D.524;
  D.516 = *d;

The stores are unneeded, as the array itself could
be eliminated.


-- 
           Summary: unnecessary stores for array constructor
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: tkoenig at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33341

[Bug fortran/33341] unnecessary stores for array constructor

[tkoenig at gcc dot gnu dot org]

-- 

tkoenig at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33341

[Bug fortran/33341] unnecessary stores for array constructor

[tkoenig at gcc dot gnu dot org]

------- Comment #1 from tkoenig at gcc dot gnu dot org  2007-09-08 11:32 -------
An equivalent C testcase eliminates the array stores:

int foo(int *a,int *b,int *c,int *d)
{
    int val[3];
    val[0] = *a;
    val[1] = *b;
    val[2] = *c;
    return (val[0] != *d) && (val[1] != *d) && (val[2] != *d);
}
$ gcc -O3 -fdump-tree-optimized -S compare.c 
$ cat compare.c.116t.optimized 

;; Function foo (foo)

Analyzing Edge Insertions.

Opportunities in BB 5 for stmt/block reduction:
Splitting BB 5 for Common stmt list.  Original block is now BB6.
D.1138 = 0;
  Edge (2->5) lands here.
  Edge (3->5) lands here.
foo (a, b, c, d)
{
  int D.1143;
  int D.1138;
  int D.1137;
  int D.1136;

<bb 2>:
  D.1136 = *b;
  D.1137 = *c;
  D.1143 = *d;
  if (*a == D.1143)
    goto <bb 5>;
  else
    goto <bb 3>;

<bb 3>:
  if (D.1136 == D.1143)
    goto <bb 5>;
  else
    goto <bb 4>;

<bb 4>:
  (void) 0;
  D.1138 = (int) !(D.1137 == D.1143);
  goto <bb 6>;

<bb 5>:
  D.1138 = 0;

<bb 6>:
  return D.1138;

}

$ gcc -O3 -fdump-tree-optimized -S compare.c 
$ cat compare.c.116t.optimized 

;; Function foo (foo)

Analyzing Edge Insertions.

Opportunities in BB 5 for stmt/block reduction:
Splitting BB 5 for Common stmt list.  Original block is now BB6.
D.1138 = 0;
  Edge (2->5) lands here.
  Edge (3->5) lands here.
foo (a, b, c, d)
{
  int D.1143;
  int D.1138;
  int D.1137;
  int D.1136;

<bb 2>:
  D.1136 = *b;
  D.1137 = *c;
  D.1143 = *d;
  if (*a == D.1143)
    goto <bb 5>;
  else
    goto <bb 3>;

<bb 3>:
  if (D.1136 == D.1143)
    goto <bb 5>;
  else
    goto <bb 4>;

<bb 4>:
  (void) 0;
  D.1138 = (int) !(D.1137 == D.1143);
  goto <bb 6>;

<bb 5>:
  D.1138 = 0;

<bb 6>:
  return D.1138;

}


-- 

tkoenig at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |4.3.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33341

[Bug fortran/33341] unnecessary stores for array constructor

[fxcoudert at gcc dot gnu dot org]

-- 

fxcoudert at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2007-10-01 13:15:26
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33341

[Bug fortran/33341] array temporaries for "all intrinsic" (unnecessary stores for array constructor)

[tkoenig at gcc dot gnu dot org]

------- Comment #2 from tkoenig at gcc dot gnu dot org  2008-07-27 19:28 -------
The test case now correctly complains about
an unneeded temporary:

$ gfortran -Warray-temporaries foo.f90
foo.f90:5.8:

  foo = all((/ a, b, c /) /= d)
       1
Warning: Creating array temporary at (1)


-- 

tkoenig at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|unnecessary stores for array|array temporaries for "all
                   |constructor                 |intrinsic" (unnecessary
                   |                            |stores for array
                   |                            |constructor)


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33341

[Bug fortran/33341] array temporaries for array constructors (unnecessary stores)

[dfranke at gcc dot gnu dot org]

------- Comment #3 from dfranke at gcc dot gnu dot org  2009-01-03 21:39 -------
Not specific to the ALL intrinsic.

$> cat ctor.f90
  INTEGER :: i, n, a(5)
  n = 5
  a = (/ (i, i = 1, n) /)
END

$> gfortran-svn -Warray-temporaries ctor.f90
ctor.f90:3.9:

  a = (/ (i, i = 1, n) /)
         1
Warning: Creating array temporary at (1)

If N is not a variable, the simplifier takes over and no runtime construction
of arrays is necessary. Not applicable example in #0, of course.


-- 

dfranke at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dfranke at gcc dot gnu dot
                   |                            |org
            Summary|array temporaries for "all  |array temporaries for array
                   |intrinsic" (unnecessary     |constructors (unnecessary
                   |stores for array            |stores)
                   |constructor)                |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33341

[Bug fortran/33341] array temporaries for array constructors (unnecessary stores)

[dfranke at gcc dot gnu dot org]

------- Comment #4 from dfranke at gcc dot gnu dot org  2009-01-04 00:20 -------
*** Bug 36935 has been marked as a duplicate of this bug. ***


-- 

dfranke at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jv244 at cam dot ac dot uk


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33341

[Bug fortran/33341] array temporaries for array constructors (unnecessary stores)

[dfranke at gcc dot gnu dot org]

------- Comment #5 from dfranke at gcc dot gnu dot org  2010-05-09 18:31 -------
This improved at some point. We still create the temporary array, but the
optimizer got smarter:

$> gfortran-svn -O3 -fdump-tree-optimized -Warray-temporaries -c pr33341.f90 
pr33341.f90:5.8:

  foo = all((/ a, b, c /) /= d)
        1
Warning: Creating array temporary at (1)

$> cat  pr33341.f90.142t.optimized 

;; Function foo (foo_)

foo (integer(kind=4) & restrict a, integer(kind=4) & restrict b,
integer(kind=4) & restrict c, integer(kind=4) & restrict d)
{
  logical(kind=4) test.0;
  integer(kind=4) D.1525;
  integer(kind=4) D.1533;
  integer(kind=4) D.1532;
  integer(kind=4) D.1531;

<bb 2>:
  D.1531_7 = *a_6(D);
  D.1532_11 = *b_10(D);
  D.1533_15 = *c_14(D);
  D.1525_17 = *d_16(D);
  if (D.1531_7 == D.1525_17)
    goto <bb 5>;
  else
    goto <bb 3>;

<bb 3>:
  if (D.1532_11 == D.1525_17)
    goto <bb 5>;
  else
    goto <bb 4>;

<bb 4>:
  test.0_32 = D.1533_15 != D.1525_17;

<bb 5>:
  # test.0_1 = PHI <0(2), 0(3), test.0_32(4)>
  return test.0_1;

}

Should we keep the PR open to (eventually) eliminate the temporary or does this
count as fixed?


-- 

dfranke at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |WAITING


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33341

[Bug fortran/33341] array temporaries for array constructors (unnecessary stores)

[dominiq at lps dot ens dot fr]

------- Comment #6 from dominiq at lps dot ens dot fr  2010-06-09 20:57 -------
*** Bug 44442 has been marked as a duplicate of this bug. ***


-- 

dominiq at lps dot ens dot fr changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dominiq at lps dot ens dot
                   |                            |fr


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33341