Hello,
I've this in my tree since some time already. In-tree there is only one
user of pre_and_rev_post_order_compute{,_fn} that actually wants entry and
exit included, and that one is just a debug routine
(draw_cfg_nodes_no_loops); so this bug right now is harmless. But I've
used this for some other patches, and in particular the hsa branch now
introduces a use that's really interested in entry and exit blocks (only
for code quality, but hey).
So, to demonstrate, given such CFG:
EN -> 1 -> 2 -> EX
|--> 3 /
One pre order would be:
EN, 1, 2, 3, EX
One post order would be:
EX, 2, 3, 1, EN
And therefore a reverse post order would be:
EN, 1, 3, 2, EX
In particular in a reverse post order the entry block comes first, and the
exit block last. The routine contains a thinko in that it places the
entry block last, and the exit block first. The bug exists since the
introduction of the include_entry_exit parameter.
I'm regstrapping with this in-tree since many moons, but just started
another one; of course, given that there aren't non-debug users right now,
no changes in testsuites are expected.
Okay for trunk?
Ciao,
Michael.
------------------
* cfganal.c (pre_and_rev_post_order_compute_fn): Correctly
enter entry and exit blocks for reverse post order.
Index: cfganal.c
===================================================================
--- cfganal.c (revision 227259)
+++ cfganal.c (working copy)
@@ -925,7 +925,7 @@ pre_and_rev_post_order_compute_fn (struc
pre_order[pre_order_num] = ENTRY_BLOCK;
pre_order_num++;
if (rev_post_order)
- rev_post_order[rev_post_order_num--] = ENTRY_BLOCK;
+ rev_post_order[rev_post_order_num--] = EXIT_BLOCK;
}
else
rev_post_order_num -= NUM_FIXED_BLOCKS;
@@ -996,7 +996,7 @@ pre_and_rev_post_order_compute_fn (struc
pre_order[pre_order_num] = EXIT_BLOCK;
pre_order_num++;
if (rev_post_order)
- rev_post_order[rev_post_order_num--] = EXIT_BLOCK;
+ rev_post_order[rev_post_order_num--] = ENTRY_BLOCK;
}
return pre_order_num;
