Hi,
In VRP we can generate new SSA variables while performing the
propagation, at least in EVRP. Therefore, when we access array vr_values
we have to check num_vr_values. This caused ICE for the PR77862.
Attached patch fixes this.
Bootstrapped and regression tested on x86_64-linux-gnu with no new
regressions. Is this OK for trunk?
Thanks,
Kugan
gcc/testsuite/ChangeLog:
2016-10-06 Kugan Vivekanandarajah <kug...@linaro.org>
PR tree-optimization/77862
* gcc.dg/pr77862.c: New test.
gcc/ChangeLog:
2016-10-06 Kugan Vivekanandarajah <kug...@linaro.org>
PR tree-optimization/77862
* tree-vrp.c (add_equivalence): Use get_value_range so that
num_vr_values is checked before accessing vr_values.
diff --git a/gcc/testsuite/gcc.dg/pr77862.c b/gcc/testsuite/gcc.dg/pr77862.c
index e69de29..e9096d2 100644
--- a/gcc/testsuite/gcc.dg/pr77862.c
+++ b/gcc/testsuite/gcc.dg/pr77862.c
@@ -0,0 +1,19 @@
+
+/* PR tree-optimization/77862 */
+/* { dg-do compile } */
+/* { dg-options "-O2 -fPIC -w" } */
+
+struct paramed_type
+{
+};
+int a;
+
+int memcmp ( int*, void *, long );
+
+void
+cur_token (void)
+{
+ if ( 1 != a || memcmp ( cur_token, "", a ) )
+ return;
+ struct paramed_type b = { };
+}
diff --git a/gcc/tree-vrp.c b/gcc/tree-vrp.c
index bfb6b2f..db9dc23 100644
--- a/gcc/tree-vrp.c
+++ b/gcc/tree-vrp.c
@@ -843,7 +843,7 @@ static void
add_equivalence (bitmap *equiv, const_tree var)
{
unsigned ver = SSA_NAME_VERSION (var);
- value_range *vr = vr_value[ver];
+ value_range *vr = get_value_range (var);
if (*equiv == NULL)
*equiv = BITMAP_ALLOC (&vrp_equiv_obstack);