As part of the generalized FSA optimization work we need the ability to start a jump threading path with a joiner, then later in the path have a normal jump threading block (ie, has side effects and thus requires duplication).

thread_through_normal_block needs one tweak to make that possible. Namely it should only push the EDGE_START_JUMP_THREAD marker if the jump threading path is currently empty.

Right now we don't call thread_through_normal_block after we've processed a joiner block, but we will soon :-) So at least today, this patch has no impact on the code we generate.

Bootstrapped and regression tested on x86_64-unknown-linux-gnu. Installed on the trunk.



        * tree-ssa-threadedge.c (thread_through_normal_block): Only push
        the EDGE_START_JUMP_THREAD marker if the jump threading path is
        empty.

diff --git a/gcc/tree-ssa-threadedge.c b/gcc/tree-ssa-threadedge.c
index c9b2c69..cabfc82 100644
--- a/gcc/tree-ssa-threadedge.c
+++ b/gcc/tree-ssa-threadedge.c
@@ -940,12 +940,18 @@ thread_through_normal_block (edge e,
              || bitmap_bit_p (visited, dest->index))
            return false;
 
-          jump_thread_edge *x
-           = new jump_thread_edge (e, EDGE_START_JUMP_THREAD);
-         path->safe_push (x);
-         *backedge_seen_p |= ((e->flags & EDGE_DFS_BACK) != 0);
+         /* Only push the EDGE_START_JUMP_THREAD marker if this is
+            first edge on the path.  */
+         if (path->length () == 0)
+           {
+              jump_thread_edge *x
+               = new jump_thread_edge (e, EDGE_START_JUMP_THREAD);
+             path->safe_push (x);
+             *backedge_seen_p |= ((e->flags & EDGE_DFS_BACK) != 0);
+           }
 
-         x = new jump_thread_edge (taken_edge, EDGE_COPY_SRC_BLOCK);
+         jump_thread_edge *x
+           = new jump_thread_edge (taken_edge, EDGE_COPY_SRC_BLOCK);
          path->safe_push (x);
          *backedge_seen_p |= ((taken_edge->flags & EDGE_DFS_BACK) != 0);
 
@@ -953,7 +959,7 @@ thread_through_normal_block (edge e,
             secondary effects of threading without having to re-run DOM or
             VRP.  */
          if (!*backedge_seen_p
-              || ! cond_arg_set_in_bb (taken_edge, e->dest))
+             || ! cond_arg_set_in_bb (taken_edge, e->dest))
            {
              /* We don't want to thread back to a block we have already
                 visited.  This may be overly conservative.  */

Reply via email to