[Bug rtl-optimization/25130] [4.1/4.2 Regression] miscompilation in GCSE

["steven at gcc dot gnu dot org" <gcc-bugzilla@gcc.gnu.org>]

------- Comment #8 from steven at gcc dot gnu dot org  2005-12-17 15:14 -------
It looks like we're missing a memory modification.  Yes, making this a load PRE
problem after all, despite it failing for me even with -fno-gcse-lm, but oh
well.

I have these expressions in the table:
Expression hash table (11 buckets, 3 entries)
Index 0 (hash value 7)
  (mem/s/j:SI (plus:SI (reg/f:SI 20 frame)
        (const_int -8 [0xfffffff8])) [0 <variable>._M_impl._M_start_node+0 S4
A32])
Index 1 (hash value 10)
  (mem/s/j/c:SI (plus:SI (reg/f:SI 20 frame)
        (const_int -16 [0xfffffff0])) [0 clusters.D.1846._M_impl._M_start_cur+0
S4 A32])
Index 2 (hash value 3)
  (mem/s/j:SI (plus:SI (reg/f:SI 20 frame)
        (const_int -12 [0xfffffff4])) [0 <variable>._M_impl._M_start_node+0 S4
A32])

The second and third expressions don't do much.  For the first expression we
apply load PRE.  But the local data flow sets look wrong.  For TRANSP I have,

transp
 0
  000
 1
  111
 2
  000
 3
  111
 4
  000

and block 3 for me looks like this:

;; Start of basic block 3, registers live: (nil)
(code_label 57 94 58 3 8 "" [1 uses])

(note 58 57 59 3 [bb 3] NOTE_INSN_BASIC_BLOCK)

(note 59 58 60 3 ("t.C") 55)

(insn 60 59 96 3 t.C:55 (parallel [
            (set (mem/s/j/c:SI (plus:SI (reg/f:SI 20 frame)
                        (const_int -8 [0xfffffff8])) [0
clusters.D.1846._M_impl._M_finish_cur+0 S4 A32])
                (plus:SI (mem/s/j/c:SI (plus:SI (reg/f:SI 20 frame)
                            (const_int -8 [0xfffffff8])) [0
clusters.D.1846._M_impl._M_finish_cur+0 S4 A32])
                    (const_int 4 [0x4])))
            (clobber (reg:CC 17 flags))
        ]) 148 {*addsi_1} (nil)
    (nil))

(jump_insn 96 60 97 3 (set (pc)
        (label_ref 61)) -1 (nil)
    (nil))
;; End of basic block 3, registers live:
 (nil)

Looks to me that insn 60 should cause expression 1 to not be in TRANSP for this
basic block.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25130