treat the global work list as a queue rather than a stack

2 files changed, 11 insertions, 2 deletions

diff --git a/rts/sm/GC.c b/rts/sm/GC.c
index 5b166943d7..934c91f2eb 100644
--- a/rts/sm/GC.c
+++ b/rts/sm/GC.c
@@ -1197,6 +1197,7 @@ init_collected_gen (nat g, nat n_threads)
 
 	// we don't have any to-be-scavenged blocks yet
 	stp->todos = NULL;
+        stp->todos_last = NULL;
 	stp->n_todos = 0;
 
 	// initialise the large object queues.
diff --git a/rts/sm/GCUtils.c b/rts/sm/GCUtils.c
index 636f23dc23..b400744290 100644
--- a/rts/sm/GCUtils.c
+++ b/rts/sm/GCUtils.c
@@ -65,6 +65,9 @@ grab_todo_block (step_workspace *ws)
     ACQUIRE_SPIN_LOCK(&stp->sync_todo);
     if (stp->todos) {
 	bd = stp->todos;
+        if (stp->todos == stp->todos_last) {
+            stp->todos_last = NULL;
+        }
 	stp->todos = bd->link;
         stp->n_todos--;
 	bd->link = NULL;
@@ -78,8 +81,13 @@ push_todo_block (bdescr *bd, step *stp)
 {
     ASSERT(bd->link == NULL);
     ACQUIRE_SPIN_LOCK(&stp->sync_todo);
-    bd->link = stp->todos;
-    stp->todos = bd;
+    if (stp->todos_last == NULL) {
+        stp->todos_last = bd;
+        stp->todos = bd;
+    } else {
+        stp->todos_last->link = bd;
+        stp->todos_last = bd;
+    }
     stp->n_todos++;
     trace(TRACE_gc, "step %d, n_todos: %d", stp->abs_no, stp->n_todos);
     RELEASE_SPIN_LOCK(&stp->sync_todo);