Hi!
The testcase in the patch ICEs, because with __builtin_unreachable ()
there is a basic block containing just notes, followed by BARRIER, and
BARRIER doesn't have BLOCK_FOR_INSN. BARRIERs are always after the bb
though, so we should handle it like BLOCK_FOR_INSN (head) != bb.
Bootstrapped/regtested on x86_64-linux and i686-linux with yes,rtl checking,
ok for trunk?
2010-08-20 Jakub Jelinek<jakub@redhat.com>
PR rtl-optimization/45353
* sel-sched-ir.c (sel_bb_head): Return NULL even if next_nonnote_insn
after bb_note is a BARRIER.
* gcc.dg/pr45353.c: New test.
--- gcc/sel-sched-ir.c.jj 2010-07-09 13:44:24.000000000 +0200
+++ gcc/sel-sched-ir.c 2010-08-20 14:14:57.000000000 +0200
@@ -4326,7 +4326,7 @@ sel_bb_head (basic_block bb)
note = bb_note (bb);
head = next_nonnote_insn (note);
- if (head&& BLOCK_FOR_INSN (head) != bb)
+ if (head&& (BARRIER_P (head) || BLOCK_FOR_INSN (head) != bb))
head = NULL_RTX;
}
--- gcc/testsuite/gcc.dg/pr45353.c.jj 2010-08-20 14:17:44.000000000 +0200
+++ gcc/testsuite/gcc.dg/pr45353.c 2010-08-20 14:17:24.000000000 +0200
@@ -0,0 +1,9 @@
+/* PR rtl-optimization/45353 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -fschedule-insns -fselective-scheduling" } */
+
+void
+foo ()
+{
+ __builtin_unreachable ();
+}
Jakub