[Bug tree-optimization/18178] Missed opportunity for removing bounds checking

[steven at gcc dot gnu dot org]

------- Additional Comments From steven at gcc dot gnu dot org  2005-02-08 
21:07 -------
We have this hunk in the .vrp dump: 
 
  # BLOCK 1 
  # PRED: 4 (true,exec) 
<L0>:; 
  iD.1901_30 = iD.1901_1; 
  D.1909_31 = D.1909_4; 
  i.0D.1911_5 = (unsigned intD.6) iD.1901_30; 
  D.1909_6 = D.1909_31; 
  D.1912_7 = (unsigned intD.6) D.1909_6; 
  if (i.0D.1911_5 >= D.1912_7) goto <L1>; else goto <L2>; 
  # SUCC: 2 (true,exec) 3 (false,exec) 
 
  # BLOCK 2 
  # PRED: 1 (true,exec) 
<L1>:; 
  #   TMT.3D.1926_25 = V_MAY_DEF <TMT.3D.1926_20>; 
  #   TMT.2D.1925_26 = V_MAY_DEF <TMT.2D.1925_19>; 
  #   VUSE <_ZTIiD.1905_14>; 
  D.1904_16 = __cxa_allocate_exception (4); 
  D.1914_18 = (intD.2 *) D.1904_16; 
  #   TMT.3D.1926_27 = V_MAY_DEF <TMT.3D.1926_25>; 
  *D.1914_18 = 5; 
  D.1914_2 = D.1914_18; 
  #   VUSE <_ZTIiD.1905_14>; 
  #   VUSE <TMT.3D.1926_27>; 
  #   VUSE <TMT.2D.1925_26>; 
  __cxa_throw (D.1904_16, &_ZTIiD.1905, 0B); 
  # SUCC: 
 
  # BLOCK 3 
  # PRED: 1 (false,exec) 
<L2>:; 
 
and: 
i.0_5: VARYING 
D.1909_6: [-2147483648, i_30 - 1] 
D.1912_7: VARYING 
 
 
If WORK_WORK_WORK is defined, we have: 
  # BLOCK 1 
  # PRED: 4 (true,exec) 
<L0>:; 
  iD.1901_30 = iD.1901_1; 
  D.1909_31 = D.1909_4; 
  D.1909_5 = D.1909_31; 
  if (iD.1901_30 >= D.1909_5) goto <L1>; else goto <L2>; 
  # SUCC: 2 (true,exec) 3 (false,exec) 
 
  # BLOCK 2 
  # PRED: 1 (true,exec) 
<L1>:; 
  #   TMT.2D.1925_24 = V_MAY_DEF <TMT.2D.1925_19>; 
  #   TMT.1D.1924_25 = V_MAY_DEF <TMT.1D.1924_18>; 
  #   VUSE <_ZTIiD.1905_13>; 
  D.1904_15 = __cxa_allocate_exception (4); 
  D.1912_17 = (intD.2 *) D.1904_15; 
  #   TMT.2D.1925_26 = V_MAY_DEF <TMT.2D.1925_24>; 
  *D.1912_17 = 5; 
  D.1912_2 = D.1912_17; 
  #   VUSE <_ZTIiD.1905_13>; 
  #   VUSE <TMT.2D.1925_26>; 
  #   VUSE <TMT.1D.1924_25>; 
  __cxa_throw (D.1904_15, &_ZTIiD.1905, 0B); 
  # SUCC: 
 
  # BLOCK 3 
  # PRED: 1 (false,exec) 
<L2>:; 
 
and 
i_30: [0, D.1909_4 - 1] 
D.1909_5: [0, i_30 - 1] 
 
Note that even though (i_30 >= D.1909_5), VRP does not clean this up. 
The exception stays there until DOM1. 
 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=18178