This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[PATCH][PR tree-optimization/59221] Fix temporary equivalence handling
- From: Jeff Law <law at redhat dot com>
- To: gcc-patches <gcc-patches at gcc dot gnu dot org>
- Date: Thu, 21 Nov 2013 12:56:06 -0700
- Subject: [PATCH][PR tree-optimization/59221] Fix temporary equivalence handling
- Authentication-results: sourceware.org; auth=none
A recent change to the thread discovery code was missing a small, but
important hunk. Namely management of the temporary equivalences.
Amazingly, this didn't cause any bootstrapping problems on x86, but
Zhendong has a good testcase and it may be causing problems on Sparc.
In a nutshell, we were failing to wipe temporary equivalences when
processing the successors of a joiner block. Egad. It'd actually been
like this for a while, but the new code makes the oversight more obvious
and painful.
Bootstrapped and regression tested on x86_64-unknown-linux-gnu.
Obviously it fixes Zhendong's tesetcase as well. Installed on the
trunk. Sorry for the breakage.
Jeff
PR tree-optimization/59221
* tree-ssa-threadedge.c (thread_across_edge): Properly manage
temporary equivalences when threading through joiner blocks.
PR tree-optimization/59221
* gcc.c-torture/execute/pr59221.c: New test.
diff --git a/gcc/testsuite/gcc.c-torture/execute/pr59221.c b/gcc/testsuite/gcc.c-torture/execute/pr59221.c
new file mode 100644
index 0000000..0cd4259
--- /dev/null
+++ b/gcc/testsuite/gcc.c-torture/execute/pr59221.c
@@ -0,0 +1,19 @@
+
+
+int a = 1, b, d;
+short e;
+
+int
+main ()
+{
+ for (; b; b++)
+ ;
+ short f = a;
+ int g = 15;
+ e = f ? f : 1 << g;
+ int h = e;
+ d = h == 83647 ? 0 : h;
+ if (d != 1)
+ __builtin_abort ();
+ return 0;
+}
diff --git a/gcc/tree-ssa-threadedge.c b/gcc/tree-ssa-threadedge.c
index 7bb8829..a144875 100644
--- a/gcc/tree-ssa-threadedge.c
+++ b/gcc/tree-ssa-threadedge.c
@@ -1072,6 +1072,10 @@ thread_across_edge (gimple dummy_cond,
/* Look at each successor of E->dest to see if we can thread through it. */
FOR_EACH_EDGE (taken_edge, ei, e->dest->succs)
{
+ /* Push a fresh marker so we can unwind the equivalences created
+ for each of E->dest's successors. */
+ stack->safe_push (NULL_TREE);
+
/* Avoid threading to any block we have already visited. */
bitmap_clear (visited);
bitmap_set_bit (visited, taken_edge->dest->index);
@@ -1118,6 +1122,9 @@ thread_across_edge (gimple dummy_cond,
{
delete_jump_thread_path (path);
}
+
+ /* And unwind the equivalence table. */
+ remove_temporary_equivalences (stack);
}
BITMAP_FREE (visited);
}