[Bug middle-end/60418] [4.9 Regression] 435.gromacs in SPEC CPU 2006 is miscompiled

[hjl.tools at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60418

--- Comment #11 from H.J. Lu <hjl.tools at gmail dot com> ---
sort_by_operand_rank in reassoc.cin has

  /* Lastly, make sure the versions that are the same go next to each
     other.  We use SSA_NAME_VERSION because it's stable.  */
  if ((oeb->rank - oea->rank == 0)
      && TREE_CODE (oea->op) == SSA_NAME
      && TREE_CODE (oeb->op) == SSA_NAME)
    {    
      if (SSA_NAME_VERSION (oeb->op) != SSA_NAME_VERSION (oea->op))
        return SSA_NAME_VERSION (oeb->op) - SSA_NAME_VERSION (oea->op);
      else 
        return oeb->id - oea->id;
    }    

It will treat

  powmult_80 = dx11_70 * dx11_70;
  powmult_241 = dy11_71 * dy11_71;
  _75 = powmult_80 + powmult_241;
  powmult_240 = dz11_72 * dz11_72;
  rsq11_77 = _75 + powmult_240;

where powmult_240 is between powmult_80 and powmult_241, differently from

  powmult_80 = dx11_70 * dx11_70;
  powmult_213 = dy11_71 * dy11_71;
  _75 = powmult_80 + powmult_213;
  powmult_244 = dz11_72 * dz11_72;
  rsq11_77 = _75 + powmult_244;

where powmult_244 > powmult_80 and powmult_213.