Combine sweeping and moving

This commit combines the sweep step with moving objects.  With this
commit, we can do:

```ruby
GC.start(compact: true)
```

This code will do the following 3 steps:

1. Fully mark the heap
2. Sweep + Move objects
3. Update references

By default, this will compact in order that heap pages are allocated.
In other words, objects will be packed towards older heap pages (as
opposed to heap pages with more pinned objects like `GC.compact` does).

1 files changed, 136 insertions, 11 deletions

diff --git a/gc.c b/gc.c
index 6e8005e774..74bed6be2e 100644
--- a/gc.c
+++ b/gc.c
@@ -480,6 +480,7 @@ typedef enum {
     GPR_FLAG_HAVE_FINALIZE     = 0x4000,
     GPR_FLAG_IMMEDIATE_MARK    = 0x8000,
     GPR_FLAG_FULL_MARK        = 0x10000,
+    GPR_FLAG_COMPACT          = 0x20000,
 
     GPR_DEFAULT_REASON =
         (GPR_FLAG_FULL_MARK | GPR_FLAG_IMMEDIATE_MARK |
@@ -634,6 +635,8 @@ typedef struct rb_heap_struct {
     struct heap_page *using_page;
     struct list_head pages;
     struct heap_page *sweeping_page; /* iterator for .pages */
+    struct heap_page *compact_cursor;
+    VALUE moved_list;
 #if GC_ENABLE_INCREMENTAL_MARK
     struct heap_page *pooled_pages;
 #endif
@@ -667,6 +670,7 @@ typedef struct rb_objspace {
 	unsigned int gc_stressful: 1;
 	unsigned int has_hook: 1;
 	unsigned int during_minor_gc : 1;
+	unsigned int compact : 1;
 #if GC_ENABLE_INCREMENTAL_MARK
 	unsigned int during_incremental_marking : 1;
 #endif
@@ -4178,17 +4182,71 @@ gc_setup_mark_bits(struct heap_page *page)
     memcpy(&page->mark_bits[0], &page->uncollectible_bits[0], HEAP_PAGE_BITMAP_SIZE);
 }
 
+static int gc_is_moveable_obj(rb_objspace_t *objspace, VALUE obj);
+static VALUE gc_move(rb_objspace_t *objspace, VALUE scan, VALUE free, VALUE moved_list);
+
+static short
+try_move(rb_objspace_t *objspace, rb_heap_t *heap, struct heap_page *sweep_page, VALUE vp)
+{
+    struct heap_page * cursor = heap->compact_cursor;
+
+    while(1) {
+	bits_t *mark_bits = cursor->mark_bits;
+        RVALUE * p = cursor->start;
+        RVALUE * offset = p - NUM_IN_PAGE(p);
+
+        /* Find an object to move and move it. Movable objects must be
+         * marked, so we iterate using the marking bitmap */
+        for (size_t i = 0; i < HEAP_PAGE_BITMAP_LIMIT; i++) {
+            bits_t bits = mark_bits[i];
+
+            if (bits) {
+                p = offset + i * BITS_BITLENGTH;
+
+                do {
+                    if (bits & 1) {
+                        if (gc_is_moveable_obj(objspace, (VALUE)p)) {
+                            heap->moved_list = gc_move(objspace, (VALUE)p, vp, heap->moved_list);
+                            return 1;
+                        }
+                    }
+                    p++;
+                    bits >>= 1;
+                } while (bits);
+            }
+        }
+
+        struct heap_page * next;
+
+        next = list_prev(&heap->pages, cursor, page_node);
+
+        // Cursors have met, lets quit
+        if (next == sweep_page) {
+            break;
+        } else {
+            heap->compact_cursor = next;
+            cursor = next;
+        }
+    }
+
+    return 0;
+}
+
 static inline int
 gc_page_sweep(rb_objspace_t *objspace, rb_heap_t *heap, struct heap_page *sweep_page)
 {
     int i;
-    int empty_slots = 0, freed_slots = 0, final_slots = 0;
+    int empty_slots = 0, freed_slots = 0, final_slots = 0, moved_slots = 0;
     RVALUE *p, *pend,*offset;
     bits_t *bits, bitset;
 
     gc_report(2, objspace, "page_sweep: start.\n");
 
     sweep_page->flags.before_sweep = FALSE;
+    if (heap->compact_cursor && sweep_page == heap->compact_cursor) {
+        /* The compaction cursor and sweep page met, so we need to quit compacting */
+        heap->compact_cursor = NULL;
+    }
 
     p = sweep_page->start; pend = p + sweep_page->total_slots;
     offset = p - NUM_IN_PAGE(p);
@@ -4220,20 +4278,51 @@ gc_page_sweep(rb_objspace_t *objspace, rb_heap_t *heap, struct heap_page *sweep_
 			  }
 			  else {
 			      (void)VALGRIND_MAKE_MEM_UNDEFINED((void*)p, sizeof(RVALUE));
-                              heap_page_add_freeobj(objspace, sweep_page, vp);
+
+                              if (heap->compact_cursor) {
+                                  /* If try_move couldn't compact anything, it means
+                                   * the cursors have met and there are no objects left that
+                                   * /can/ be compacted, so we need to quit. */
+                                  if (try_move(objspace, heap, sweep_page, vp)) {
+                                      moved_slots++;
+                                  } else {
+                                      heap->compact_cursor = NULL;
+                                      heap_page_add_freeobj(objspace, sweep_page, vp);
+                                  }
+                              } else {
+                                  heap_page_add_freeobj(objspace, sweep_page, vp);
+                              }
+
                               gc_report(3, objspace, "page_sweep: %s is added to freelist\n", obj_info(vp));
-			      freed_slots++;
-                              asan_poison_object(vp);
+                              freed_slots++;
 			  }
 			  break;
 		      }
 
 			/* minor cases */
+		      case T_MOVED:
+                        if (!objspace->flags.during_compacting) {
+                            /* When compaction is combined with sweeping, some of the swept pages
+                             * will have T_MOVED objects on them.  These need to be kept alive
+                             * until references are finished updating.  Once references are finished
+                             * updating, `gc_unlink_moved_list` will clear the T_MOVED slots
+                             * and release them back to the heap.  If there are T_MOVED slots
+                             * in the heap and we're _not_ compacting, then it's a bug. */
+                            rb_bug("T_MOVED shouldn't be on the heap unless compacting\n");
+                        }
+                        break;
 		      case T_ZOMBIE:
 			/* already counted */
 			break;
 		      case T_NONE:
-			empty_slots++; /* already freed */
+                        if (heap->compact_cursor) {
+                            if (try_move(objspace, heap, sweep_page, vp)) {
+                                moved_slots++;
+                            } else {
+                                heap->compact_cursor = NULL;
+                            }
+                        }
+                        empty_slots++; /* already freed */
 			break;
 		    }
 		}
@@ -4257,7 +4346,7 @@ gc_page_sweep(rb_objspace_t *objspace, rb_heap_t *heap, struct heap_page *sweep_
 		   (int)sweep_page->total_slots,
 		   freed_slots, empty_slots, final_slots);
 
-    sweep_page->free_slots = freed_slots + empty_slots;
+    sweep_page->free_slots = (freed_slots + empty_slots) - moved_slots;
     objspace->profile.total_freed_objects += freed_slots;
     heap_pages_final_slots += final_slots;
     sweep_page->final_slots += final_slots;
@@ -4271,7 +4360,7 @@ gc_page_sweep(rb_objspace_t *objspace, rb_heap_t *heap, struct heap_page *sweep_
 
     gc_report(2, objspace, "page_sweep: end.\n");
 
-    return freed_slots + empty_slots;
+    return (freed_slots + empty_slots) - moved_slots;
 }
 
 /* allocate additional minimum page to work */
@@ -4457,6 +4546,29 @@ gc_sweep_continue(rb_objspace_t *objspace, rb_heap_t *heap)
 }
 
 static void
+gc_compact_start(rb_objspace_t *objspace, rb_heap_t *heap)
+{
+    heap->compact_cursor = list_tail(&heap->pages, struct heap_page, page_node);
+    objspace->profile.compact_count++;
+    heap->moved_list = Qfalse;
+}
+
+static void gc_update_references(rb_objspace_t * objspace);
+static void gc_unlink_moved_list(rb_objspace_t *objspace, VALUE moved_list_head);
+
+static void
+gc_compact_finish(rb_objspace_t *objspace, rb_heap_t *heap)
+{
+    heap->compact_cursor = NULL;
+    gc_update_references(objspace);
+    rb_clear_constant_cache();
+    gc_unlink_moved_list(objspace, heap->moved_list);
+    heap->moved_list = Qfalse;
+    objspace->flags.compact = 0;
+    objspace->flags.during_compacting = FALSE;
+}
+
+static void
 gc_sweep(rb_objspace_t *objspace)
 {
     const unsigned int immediate_sweep = objspace->flags.immediate_sweep;
@@ -4468,7 +4580,13 @@ gc_sweep(rb_objspace_t *objspace)
 	gc_prof_sweep_timer_start(objspace);
 #endif
 	gc_sweep_start(objspace);
+        if (objspace->flags.compact) {
+            gc_compact_start(objspace, heap_eden);
+        }
 	gc_sweep_rest(objspace);
+        if (objspace->flags.compact) {
+            gc_compact_finish(objspace, heap_eden);
+        }
 #if !GC_ENABLE_LAZY_SWEEP
 	gc_prof_sweep_timer_stop(objspace);
 #endif
@@ -7283,6 +7401,8 @@ gc_start(rb_objspace_t *objspace, int reason)
 
     /* reason may be clobbered, later, so keep set immediate_sweep here */
     objspace->flags.immediate_sweep = !!((unsigned)reason & GPR_FLAG_IMMEDIATE_SWEEP);
+    objspace->flags.compact = !!((unsigned)reason & GPR_FLAG_COMPACT);
+    objspace->flags.during_compacting = TRUE;
 
     if (!heap_allocated_pages) return FALSE; /* heap is not ready */
     if (!(reason & GPR_FLAG_METHOD) && !ready_to_gc(objspace)) return TRUE; /* GC is not allowed */
@@ -7544,7 +7664,7 @@ garbage_collect_with_gvl(rb_objspace_t *objspace, int reason)
 }
 
 static VALUE
-gc_start_internal(rb_execution_context_t *ec, VALUE self, VALUE full_mark, VALUE immediate_mark, VALUE immediate_sweep)
+gc_start_internal(rb_execution_context_t *ec, VALUE self, VALUE full_mark, VALUE immediate_mark, VALUE immediate_sweep, VALUE compact)
 {
     rb_objspace_t *objspace = &rb_objspace;
     int reason = GPR_FLAG_FULL_MARK |
@@ -7552,9 +7672,14 @@ gc_start_internal(rb_execution_context_t *ec, VALUE self, VALUE full_mark, VALUE
                  GPR_FLAG_IMMEDIATE_SWEEP |
                  GPR_FLAG_METHOD;
 
-    if (!RTEST(full_mark))       reason &= ~GPR_FLAG_FULL_MARK;
-    if (!RTEST(immediate_mark))  reason &= ~GPR_FLAG_IMMEDIATE_MARK;
-    if (!RTEST(immediate_sweep)) reason &= ~GPR_FLAG_IMMEDIATE_SWEEP;
+    /* For now, compact implies full mark / sweep, so ignore other flags */
+    if (RTEST(compact)) {
+        reason |= GPR_FLAG_COMPACT;
+    } else {
+        if (!RTEST(full_mark))       reason &= ~GPR_FLAG_FULL_MARK;
+        if (!RTEST(immediate_mark))  reason &= ~GPR_FLAG_IMMEDIATE_MARK;
+        if (!RTEST(immediate_sweep)) reason &= ~GPR_FLAG_IMMEDIATE_SWEEP;
+    }
 
     garbage_collect(objspace, reason);
     gc_finalize_deferred(objspace);