This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug tree-optimization/32500] [4.2 Regression] Loop optimization limits range to size of array used inside loop
- From: "rguenth at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: 4 Jul 2007 10:16:52 -0000
- Subject: [Bug tree-optimization/32500] [4.2 Regression] Loop optimization limits range to size of array used inside loop
- References: <bug-32500-14753@http.gcc.gnu.org/bugzilla/>
- Reply-to: gcc-bugzilla at gcc dot gnu dot org
------- Comment #12 from rguenth at gcc dot gnu dot org 2007-07-04 10:16 -------
This is SCEV. From
<L2>:;
i_7 = ASSERT_EXPR <i_20, i_20 > 4>;
i.10_10 = (unsigned int) i_7;
D.2489_11 = i.10_10 - 7;
if (D.2489_11 <= 2) goto <L3>; else goto <L4>;
we have
Found new range for i.10_10: [5, 12]
Visiting statement:
D.2489_11 = i.10_10 - 7;
(analyze_scalar_evolution
(loop_nb = 1)
(scalar = D.2489_11)
(get_scalar_evolution
(scalar = D.2489_11)
(scalar_evolution = {4294967290, +, 1}_1))
(set_scalar_evolution
(scalar = D.2489_11)
(scalar_evolution = {4294967290, +, 1}_1))
)
(instantiate_parameters
(loop_nb = 1)
(chrec = {4294967290, +, 1}_1)
(res = {4294967290, +, 1}_1))
Found new range for D.2489_11: [4294967290, +INF]
which is wrong. The new range should be ~[5, 4294967290].
scev_probably_wraps_p() returns false for the above chrec because for
the loop in question estimated_nb_iterations is 4(!) which is derived
from infer_loop_bounds_from_undefined. On the trunk this is fixed
by rewriting number of iterations analysis. On the 4.2 branch we
can fix this conservatively by
Index: tree-ssa-loop-niter.c
===================================================================
--- tree-ssa-loop-niter.c (revision 126260)
+++ tree-ssa-loop-niter.c (working copy)
@@ -1747,6 +1747,12 @@ infer_loop_bounds_from_undefined (struct
{
bb = bbs[i];
+ /* If BB is not executed in each iteration of the loop, we cannot
+ use the operations in it to infer reliable upper bound on the
+ # of iterations of the loop. */
+ if (!dominated_by_p (CDI_DOMINATORS, loop->latch, bb))
+ continue;
+
for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
{
tree stmt = bsi_stmt (bsi);
I'm going to test this.
--
rguenth at gcc dot gnu dot org changed:
What |Removed |Added
----------------------------------------------------------------------------
AssignedTo|unassigned at gcc dot gnu |rguenth at gcc dot gnu dot
|dot org |org
Status|NEW |ASSIGNED
Last reconfirmed|2007-06-26 10:45:47 |2007-07-04 10:16:51
date| |
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32500