https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100359
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |ASSIGNED
Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot
gnu.org
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Indeed:
-Setting value number of a.0_1 to &b (changed)
+Setting value number of a.0_1 to a.0_1 (changed)
+Using extra use virtual operand .MEM_39
+Making available beyond BB2 a.0_1 for value a.0_1
...
Setting value number of _3 to &b (changed)
Value numbering stmt = _4 = *_3;
-Setting value number of _4 to 0 (changed)
+Setting value number of _4 to _4 (changed)
+Making available beyond BB2 _4 for value _4
Value numbering stmt = if (_4 != 0)
-marking known outgoing edge 2 -> 5 executable
and the IL difference is just
<bb 2> [local count: 1073741824]:
f = 0;
a = &b;
- c.2_18 = c;
- *c.2_18 = &b;
+ j ={v} {CLOBBER};
+ c.2_40 = c;
+ *c.2_40 = &b;
f = 1;
a.0_1 = a;
*a.0_1 = 0;
note that 'a' is ESCAPED and _18/_40 point to ESCAPED. So what doesn't
work with the CLOBBER is
/* If we reach a clobbering statement try to skip it and see if
we find a VN result with exactly the same value as the
possible clobber. In this case we can ignore the clobber
and return the found value. */
in vn_reference_lookup_3 because that does a non-walking lookup. I think
we have a duplicate report for this exact issue somewhere.
The clobber is removed at -O2 as part of the useless vars removal before
stdarg. DCE2 leaves us with
<bb 4> [local count: 8687547609]:
- a = &b;
+ j[0] = &b;
+ _13 = j[0];
+ a = _13;
j ={v} {CLOBBER};
so we have stray uses left with -O3 which is ultimatively because cunrolli
unrolls only the outer loop with -O3!?
Ah, the reason is try_unroll_loop_completely has
/* See if we can improve our estimate by using recorded loop bounds. */
if ((allow_peel || maxiter == 0 || ul == UL_NO_GROWTH)
&& maxiter >= 0
&& (!n_unroll_found || (unsigned HOST_WIDE_INT)maxiter < n_unroll))
{
n_unroll = maxiter;
n_unroll_found = true;
/* Loop terminates before the IV variable test, so we cannot
remove it in the last iteration. */
edge_to_cancel = NULL;
}
but with -O3 we have !allow_peel and ul == UL_ALL (and maxiter == 1),
niter is unknown. I changed this with
g:a52206ae28ed3e55d601118bedd52739456401ab but the intent there was
clearly to allow the peeling if we restrict the cases to UL_NO_GROWTH,
not to disallow it with UL_ALL.
