[Bug tree-optimization/41454] [4.4/4.5 Regression] DOM miscompiles gcc.c-torture/execute/990513-1.c at -O2 -fno-tree-vrp

["rguenth at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

------- Comment #1 from rguenth at gcc dot gnu dot org  2009-09-23 21:46 -------
So, we have

;; basic block 3, loop depth 1, count 0
;; prev block 2, next block 7
;; pred:       7 [100.0%]  (fallthru,dfs_back) 2 [100.0%]  (fallthru,exec)
;; succ:       7 [98.5%]  (true,exec) 4 [1.5%]  (false,exec)
<bb 3>:
# BM_tab_19 = PHI <BM_tab_11(7), &BM_tab[256](2)>
MEM[base: BM_tab_19, offset: 0x0fffffffc] = 6;
MEM[base: BM_tab_19, offset: 0x0fffffff8] = 6;
MEM[base: BM_tab_19, offset: 0x0fffffff4] = 6;
D.2422_17 = (unsigned int) BM_tab_19;
D.2423_18 = D.2422_17 + 0x0fffffff0;
BM_tab_24 = (int *) D.2423_18;
BM_tab_11 = BM_tab_24;
MEM[base: BM_tab_24] = 6;
if (&BM_tab[0] != BM_tab_24)
  goto <bb 7>;
else
  goto <bb 4>;

;; basic block 7, loop depth 1, count 0
;; prev block 3, next block 4
;; pred:       3 [98.5%]  (true,exec)
;; succ:       3 [100.0%]  (fallthru,dfs_back)
<bb 7>:
goto <bb 3>;


and for block 7 do

Optimizing block #7

1>>> COND 1 = &BM_tab[0] ne_expr BM_tab_24
1>>> COND 0 = &BM_tab[0] eq_expr BM_tab_24
LKUP STMT D.2422_17 = nop_expr BM_tab_24
          D.2422_17 = (unsigned int) BM_tab_24;
LKUP STMT D.2423_18 = D.2422_17 plus_expr 0x0fffffff0
          D.2423_18 = D.2422_17 + 0x0fffffff0;
FIND: D.2423_18
LKUP STMT BM_tab_24 = nop_expr D.2423_18
          BM_tab_24 = (int *) D.2423_18;
FIND: BM_tab_24
LKUP STMT &BM_tab[0] ne_expr BM_tab_24
          if (&BM_tab[0] != BM_tab_24)
FIND: 1


oops.  And we jump-thread into an endless loop.

How is this supposed to work?


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |law at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41454