[Bug tree-optimization/40168] missing unrolling/scalarization/reassoc/free

[Bug tree-optimization/40168] missing unrolling/scalarization/reassoc/free

[Joost.VandeVondele at mat dot ethz.ch]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40168

Joost VandeVondele <Joost.VandeVondele at mat dot ethz.ch> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2009-12-18 14:45:13         |2013-03-29
                 CC|                            |Joost.VandeVondele at mat
                   |                            |dot ethz.ch

--- Comment #21 from Joost VandeVondele <Joost.VandeVondele at mat dot ethz.ch> 2013-03-29 10:50:49 UTC ---
So, the testcase in comment #14 is indeed still (4.9.0) yielding the 324
multiplies for subroutine S2, instead of the more optimal 192 as shown in S1.
Ifort also results in 324 multiplies, but is able to do a couple of them with
mulpd instead of mulsd. So the common subexpressions are still not found.

[Bug tree-optimization/40168] finding common subexpressions

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=40168

--- Comment #22 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
With the trunk on the second testcase, SLP works but we get stuff like:
  _2 = (*b_420(D))[80];
  _4 = (*b_420(D))[79];
  _538 = {_2, _4};
  _980 = _538 * vect__1.182_559;

Shouldn't that just be loading a vector from (*b_420(D))[79] and then doing an
VEC_PERM?
In a reduced C testcase we get the correct thing:
typedef double array[1000];

void f(array  *a, array *b, array *c)
{
    double t = (*a)[1] * (*b)[0];
    double t1 = (*a)[0] * (*b)[0];
    double t2 = (*a)[3] * (*b)[1];
    double t3 = (*a)[2] * (*b)[1];
    (*c)[0] = t;
    (*c)[1] = t1;
    (*c)[2] = t2;
    (*c)[3] = t3;
}

[Bug tree-optimization/40168] missing unrolling/scalarization/reassoc/free

[rguenth at gcc dot gnu dot org]

------- Comment #20 from rguenth at gcc dot gnu dot org  2009-12-18 14:45 -------
There is still the issue about the 2nd testcase.  It needs to be re-analyzed
and possibly simplified.  But it's now a pure optimization issue, not a
frontend issue anymore.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
          Component|fortran                     |tree-optimization
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2009-12-18 14:45:13
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=40168