Hi!
As mentioned in the PR, the problem is that we might grow the vector
without clearing new entries, while we overwrite the last entry, there might
be gaps containing garbage pointers and cause lots of weird issues.
The testcase is unfortunately too large (creduced it down to 26100 bytes)
for the testsuite. Bootstrapped/regtested on x86_64-linux and i686-linux,
committed to trunk as obvious.
2016-03-15 Jakub Jelinek <ja...@redhat.com>
PR middle-end/70239
* tree-ssa-sccvn.c (VN_INFO_GET): Use safe_grow_cleared instead
of safe_grow.
--- gcc/tree-ssa-sccvn.c.jj 2016-02-16 16:14:43.000000000 +0100
+++ gcc/tree-ssa-sccvn.c 2016-03-15 13:02:40.876997092 +0100
@@ -407,7 +407,7 @@ VN_INFO_GET (tree name)
newinfo = XOBNEW (&vn_ssa_aux_obstack, struct vn_ssa_aux);
memset (newinfo, 0, sizeof (struct vn_ssa_aux));
if (SSA_NAME_VERSION (name) >= vn_ssa_aux_table.length ())
- vn_ssa_aux_table.safe_grow (SSA_NAME_VERSION (name) + 1);
+ vn_ssa_aux_table.safe_grow_cleared (SSA_NAME_VERSION (name) + 1);
vn_ssa_aux_table[SSA_NAME_VERSION (name)] = newinfo;
return newinfo;
}
Jakub
