https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61677
Manuel López-Ibáñez <manu at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |missed-optimization
Status|WAITING |NEW
--- Comment #5 from Manuel López-Ibáñez <manu at gcc dot gnu.org> ---
(In reply to Josh Triplett from comment #4)
> Created attachment 33067 [details]
> Reduced testcase
>
> Here's a reduced testcase.
[CHECK] Found def edge 1 in jump_23 = PHI <jump_16(D)(13), jump_18(11)>
[BEFORE SIMPLICATION -- [USE]:
jump_23->offset = 42;
is guarded by :
i_19 != -1(OVF) (.AND.) _8 != 0
[BEFORE NORMALIZATION --[USE]:
jump_23->offset = 42;
is guarded by :
i_19 != -1(OVF) (.AND.) _8 != 0
[AFTER NORMALIZATION -- [USE]:
jump_23->offset = 42;
is guarded by :
menu_7(D) != 0B (.AND.) i_19 != -1(OVF)
[BEFORE SIMPLICATION -- [DEF]:
jump_23 = PHI <jump_16(D)(13), jump_18(11)>
is guarded by :
_8 != 0 (.AND.) (.NOT.) _12 != 0
[BEFORE NORMALIZATION --[DEF]:
jump_23 = PHI <jump_16(D)(13), jump_18(11)>
is guarded by :
_8 != 0 (.AND.) (.NOT.) _12 != 0
[AFTER NORMALIZATION -- [DEF]:
jump_23 = PHI <jump_16(D)(13), jump_18(11)>
is guarded by :
(.NOT.) _12 != 0 (.AND.) menu_7(D) != 0B
[CHECK]: Found unguarded use: jump_23->offset = 42;
There is something wrong in the computation of the guard-conditions. Perhaps
the computation is not powerful enough to realize that "(.NOT.) _12 != 0" has
to appear also in the use.
It seems also that VRP is not powerful enough here:
# RANGE [0, 2147483645] NONZERO 15
# i_3 = PHI <i_19(8), i_34(5)>
# .MEM_6 = PHI <.MEM_13(8), .MEM_24(5)>
[test.c : 14:11] # RANGE [-1, 2147483644]
i_19 = i_3 + -1;
[test.c : 14:11] if (i_19 != -1(OVF))
The range should be [0, 8].