[Bug tree-optimization/46009] New: ?: vectorized, very similar if is not

[Bug tree-optimization/46009] New: ?: vectorized, very similar if is not

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46009

           Summary: ?: vectorized, very similar if is not
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: jakub@gcc.gnu.org
                CC: irar@gcc.gnu.org


void bar (int *);
void foo (int *a, int *b, int *c, int *d)
{
  int e[1024], i, g;
  for (i = 0; i < 1024; i++)
    {
      g = a[i] + b[i] + c[i] * d[i];;
#ifdef WORKS
      e[i] = g < 10 ? 1 : g;
#else
      if (g < 10)
        e[i] = 1;
      else
        e[i] = g;
#endif
    }
  bar (e);
}

is vectorized only if -DWORKS -O3, but not for -O3, although the code is really
identical.  Not sure if ifcvt should handle this, or instead some earlier pass
should detect that both bbs have the same memory destination and change
  if (g_25 <= 9)
    goto <bb 4>;
  else
    goto <bb 5>;

<bb 4>:
  e[i_14] = 1;
  goto <bb 6>;

<bb 5>:
  e[i_14] = g_25;

<bb 6>:

into an unconditional assignment of a PHI result into e[i].

[Bug tree-optimization/46009] ?: vectorized, very similar if is not

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46009

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> 2010-10-13 16:59:28 UTC ---
Related to PR 21998.

[Bug tree-optimization/46009] ?: vectorized, very similar if is not

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46009

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2010.11.03 11:11:40
         AssignedTo|unassigned at gcc dot       |jakub at gcc dot gnu.org
                   |gnu.org                     |
     Ever Confirmed|0                           |1

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-11-03 11:11:40 UTC ---
Created attachment 22245
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22245
gcc46-pr46009.patch

Untested fix.

As discussed on IRC, this is something that can be handled either in store
sinking pass, or in cselim (which has some infrastructure for it), or both.

This patch implements it in cselim.

[Bug tree-optimization/46009] ?: vectorized, very similar if is not

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46009

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #22245|0                           |1
        is obsolete|                            |

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-11-03 12:01:18 UTC ---
Created attachment 22247
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22247
gcc46-pr46009.patch

Updated patch for richi's comments on IRC.

[Bug tree-optimization/46009] ?: vectorized, very similar if is not

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46009

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> 2010-11-03 15:18:55 UTC ---
Author: jakub
Date: Wed Nov  3 15:18:50 2010
New Revision: 166251

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=166251
Log:
    PR tree-optimization/46009
    * tree-ssa-phiopt.c (tree_ssa_phiopt_worker): Call
    cond_if_else_store_replacement if bb1 and bb2 have the same
    single successor.
    (cond_store_replacement): Use gimple_assign_single_p, don't
    check if rhs is SSA_NAME or invariant.  Call release_defs for
    assign.
    (cond_if_else_store_replacement): New function.

    * gcc.dg/vect/pr46009.c: New function.

Added:
    trunk/gcc/testsuite/gcc.dg/vect/pr46009.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-ssa-phiopt.c

[Bug tree-optimization/46009] ?: vectorized, very similar if is not

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46009

H.J. Lu <hjl.tools at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.6.0

--- Comment #5 from H.J. Lu <hjl.tools at gmail dot com> 2011-08-25 15:37:02 UTC ---
Fixed.