[Bug tree-optimization/57702] New: Reassoc missed optimizations

[Bug tree-optimization/57702] New: Reassoc missed optimizations

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57702

            Bug ID: 57702
           Summary: Reassoc missed optimizations
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: jakub at gcc dot gnu.org

On:

int f1 (int x, int y) { x += y; return x + 3 * y; }
unsigned int f2 (unsigned int x, unsigned int y) { x += y; return x + 3 * y; }
int f3 (int x, int y) { x += 7 * y; return x + 12 * y; }
unsigned int f4 (unsigned int x, unsigned int y) { x += 7 * y; return x + 12 *
y; }

reassoc optimizes only the f4 function into x += 19 * y; at the tree level, and
at the RTL level combiner happens to optimize it except for f3, which has more
insns than f4.  I don't see why not optimizing this even for signed types would
be problematic, as long as the multiplication is performed in the same signed
type and all terms have the same sign (with different sizes the optimization
could remove undefined overflow, but I don't see how it could introduce it).

When things are vectorized the RTL optimizations will hardly help though.

[Bug tree-optimization/57702] Reassoc missed optimizations

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57702

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Another example:
unsigned int foo (unsigned int x, unsigned int y)
{
  x += y;
  x += y;
  x += y;
  x += y;
  x += y;
  return x;
}

Seems for the x += 7 * y; x += 12 * y; case it is undistribute_ops_list that
performs this optimization.  For non-wrapping integers we give up very early,
can_reassociate_p returns false and we don't try anything, perhaps with care we
could try at least undistribution, provided that we are very careful not to
reorder anything, only undistribute adjacent operations.  For unsigned
integers,
we give up because we never consider an SSA_NAME to be 1 * SSA_NAME for the
purposes of undistribution.

[Bug tree-optimization/57702] Reassoc missed optimizations

[ktietz at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57702

Kai Tietz <ktietz at gcc dot gnu.org> changed:

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

--- Comment #3 from Kai Tietz <ktietz at gcc dot gnu.org> ---
Looks tp me like a duplicate of 45218

[Bug tree-optimization/57702] Reassoc missed optimizations

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57702

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2013-06-25 00:00:00         |2021-7-19

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
f2:
  _1 = y_3(D) * 3;
  _7 = _1 + x_2(D);
  _5 = y_3(D) + _7;

f4:
  _9 = y_3(D) * 19;
  _6 = x_4(D) + _9;

foo:
  _9 = y_2(D) * 5;
  x_7 = x_1(D) + _9;

[Bug tree-optimization/57702] Reassoc missed optimizations

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57702

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

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