commit 0e5c74a36e7368e0f55196f5e0cf5d640a051e63
Author: Kyrylo Tkachov <kyrylo.tkachov@arm.com>
Date: Wed Nov 18 16:49:21 2015 +0000
PR middle-end/68375: Restructure get_loop_body_in_bfs_order to handle loops with only a header
@@ -913,37 +913,32 @@ get_loop_body_in_bfs_order (const struct loop *loop)
basic_block *blocks;
basic_block bb;
bitmap visited;
- unsigned int i = 0;
- unsigned int vc = 1;
+ unsigned int i = 1;
+ unsigned int vc = 0;
gcc_assert (loop->num_nodes);
gcc_assert (loop->latch != EXIT_BLOCK_PTR_FOR_FN (cfun));
blocks = XNEWVEC (basic_block, loop->num_nodes);
visited = BITMAP_ALLOC (NULL);
-
- bb = loop->header;
+ blocks[0] = loop->header;
+ bitmap_set_bit (visited, loop->header->index);
while (i < loop->num_nodes)
{
edge e;
edge_iterator ei;
-
- if (bitmap_set_bit (visited, bb->index))
- /* This basic block is now visited */
- blocks[i++] = bb;
+ gcc_assert (i > vc);
+ bb = blocks[vc++];
FOR_EACH_EDGE (e, ei, bb->succs)
{
if (flow_bb_inside_loop_p (loop, e->dest))
{
+ /* This bb is now visited. */
if (bitmap_set_bit (visited, e->dest->index))
blocks[i++] = e->dest;
}
}
-
- gcc_assert (i > vc);
-
- bb = blocks[vc++];
}
BITMAP_FREE (visited);
new file mode 100644
@@ -0,0 +1,39 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-optimized-graph" } */
+
+int a, c, d, e, g, h;
+short f;
+
+int
+foo ()
+{
+}
+
+short
+fn1 (void)
+{
+ int j[2];
+ for (; e; e++)
+ if (j[0])
+ for (;;)
+ ;
+ if (!g)
+ return f;
+}
+
+int
+main (void)
+{
+ for (; a < 1; a++)
+ {
+ for (c = 0; c < 2; c++)
+ {
+ d && (f = 0);
+ h = fn1 ();
+ }
+ __builtin_printf ("%d\n", (char) f);
+ }
+
+ return 0;
+}
+
Hi all, The ICE in this PR occurs when trying to dump the CFG to a .dot file. The i > vc assert at the bottom of the loop in get_loop_body_in_bfs_order gets triggered when i == 1 and vc == 1. This is not a problem for that loop because it contains only a header and it's not going to go around the loop again to access uninitialized memory in the blocks array. Richi suggested restructuring the loop to initialise the blocks array with the header early on and moving the assert to the beginning of the loop. This fix works as far as I can test. Bootstrapped and tested on arm, aarch64, x86_64. Ok for trunk? Thanks, Kyrill 2015-11-24 Richard Biener <rguenther@suse.de> Kyrylo Tkachov <kyrylo.tkachov@arm.com> PR middle-end/68375 * cfgloop.c (get_loop_body_in_bfs_order): Restructure loop to avoid bogus assertion. 2015-11-24 Kyrylo Tkachov <kyrylo.tkachov@arm.com> PR middle-end/68375 * gcc.dg/pr68375.c: New test.