obtained gcc-4.6.4.tar.bz2 from upstream website;upstream

verified gcc-4.6.4.tar.bz2.sig;
imported gcc-4.6.4 source tree from verified upstream tarball.

downloading a git-generated archive based on the 'upstream' tag
should provide you with a source tree that is binary identical
to the one extracted from the above tarball.

if you have obtained the source via the command 'git clone',
however, do note that line-endings of files in your working
directory might differ from line-endings of the respective
files in the upstream repository.

1 files changed, 23 insertions, 0 deletions

diff --git a/gcc/testsuite/gcc.dg/pr42631.c b/gcc/testsuite/gcc.dg/pr42631.c
new file mode 100644
index 000000000..f5ac5a53a
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr42631.c
@@ -0,0 +1,23 @@
+/* The function below expands to a loop whose latch block starts with
+   a PHI node and the corresponding debug stmt.  In RTL, there are no
+   PHI nodes, but the debug insn that references the incoming k
+   remains, even though one of the incoming edges has it
+   uninitialized.  After unrolling, however, the debug insn becomes
+   unconditional, and this exposed a problem in the webizer.  Because
+   DF doesn't combine the uses of an uninitialized pseudo into a
+   single UD chain, we created a separate web for each use.
+   Allocating separate registers or stack slots for each uninitialized
+   use is wasteful, but the problem became more apparent in
+   -fcompare-debug tests: register numbers went out of sync, and could
+   have caused codegen differences depending on whether or not the
+   debug insns were present.  The fix was to arrange for web to
+   combine uninitialized uses into a single web.  */
+
+/* { dg-do compile } */
+/* { dg-options "-g -O1 -funroll-loops -fcompare-debug" } */
+
+void foo()
+{
+  unsigned k;
+  while (--k > 0);
+}