GC refactoring: make evacuate() take an StgClosure**

Change the type of evacuate() from 
  StgClosure *evacuate(StgClosure *);
to
  void evacuate(StgClosure **);

So evacuate() itself writes the source pointer, rather than the
caller.  This is slightly cleaner, and avoids a few memory writes:
sometimes evacuate() doesn't move the object, and in these cases the
source pointer doesn't need to be written.  It doesn't have a
measurable impact on performance, though.

5 files changed, 276 insertions, 234 deletions

diff --git a/rts/sm/Evac.c b/rts/sm/Evac.c
index fb9f4c49b2..518b383a54 100644
--- a/rts/sm/Evac.c
+++ b/rts/sm/Evac.c
@@ -25,7 +25,7 @@
  */
 #define MAX_THUNK_SELECTOR_DEPTH 16
 
-static StgClosure * eval_thunk_selector (StgSelector * p, rtsBool);
+static void eval_thunk_selector (StgClosure **q, StgSelector * p, rtsBool);
 
 STATIC_INLINE StgPtr
 alloc_for_copy (nat size, step *stp)
@@ -99,12 +99,12 @@ alloc_for_copy_noscav (nat size, step *stp)
     return to;
 }
   
-STATIC_INLINE StgClosure *
-copy_tag(StgClosure *src, nat size, step *stp,StgWord tag)
+STATIC_INLINE void
+copy_tag(StgClosure **p, StgClosure *src, nat size, step *stp,StgWord tag)
 {
-  StgPtr to, from;
-  nat i;
-  StgWord info;
+    StgPtr to, tagged_to, from;
+    nat i;
+    StgWord info;
 
 #ifdef THREADED_RTS
     do {
@@ -121,6 +121,8 @@ copy_tag(StgClosure *src, nat size, step *stp,StgWord tag)
 #endif
 
     to = alloc_for_copy(size,stp);
+    tagged_to = (StgPtr)TAG_CLOSURE(tag,(StgClosure*)to);
+    *p = (StgClosure *)tagged_to;
     
     TICK_GC_WORDS_COPIED(size);
 
@@ -130,14 +132,14 @@ copy_tag(StgClosure *src, nat size, step *stp,StgWord tag)
 	to[i] = from[i];
     }
     
+    ((StgEvacuated*)from)->evacuee = (StgClosure *)tagged_to;
+
     // retag pointer before updating EVACUATE closure and returning
-    to = (StgPtr)TAG_CLOSURE(tag,(StgClosure*)to);
 
 //  if (to+size+2 < bd->start + BLOCK_SIZE_W) {
 //      __builtin_prefetch(to + size + 2, 1);
 //  }
 
-    ((StgEvacuated*)from)->evacuee = (StgClosure *)to;
 #ifdef THREADED_RTS
     write_barrier();
     ((StgEvacuated*)from)->header.info = &stg_EVACUATED_info;
@@ -148,17 +150,16 @@ copy_tag(StgClosure *src, nat size, step *stp,StgWord tag)
     // the profiler can guess the position of the next object later.
     SET_EVACUAEE_FOR_LDV(from, size);
 #endif
-    return (StgClosure *)to;
 }
   
 
 // Same as copy() above, except the object will be allocated in memory
 // that will not be scavenged.  Used for object that have no pointer
 // fields.
-STATIC_INLINE StgClosure *
-copy_noscav_tag(StgClosure *src, nat size, step *stp, StgWord tag)
+STATIC_INLINE void
+copy_noscav_tag(StgClosure **p, StgClosure *src, nat size, step *stp, StgWord tag)
 {
-    StgPtr to, from;
+    StgPtr to, tagged_to, from;
     nat i;
     StgWord info;
 
@@ -176,6 +177,8 @@ copy_noscav_tag(StgClosure *src, nat size, step *stp, StgWord tag)
 #endif
     
     to = alloc_for_copy_noscav(size,stp);
+    tagged_to = (StgPtr)TAG_CLOSURE(tag,(StgClosure*)to);
+    *p = (StgClosure *)tagged_to;
 
     TICK_GC_WORDS_COPIED(size);
     
@@ -185,10 +188,8 @@ copy_noscav_tag(StgClosure *src, nat size, step *stp, StgWord tag)
 	to[i] = from[i];
     }
 
-    // retag pointer before updating EVACUATE closure and returning
-    to = (StgPtr)TAG_CLOSURE(tag,(StgClosure*)to);
+    ((StgEvacuated*)from)->evacuee = (StgClosure *)tagged_to;
 
-    ((StgEvacuated*)from)->evacuee = (StgClosure *)to;
 #ifdef THREADED_RTS
     write_barrier();
     ((StgEvacuated*)from)->header.info = &stg_EVACUATED_info;
@@ -199,7 +200,6 @@ copy_noscav_tag(StgClosure *src, nat size, step *stp, StgWord tag)
     // the profiler can guess the position of the next object later.
     SET_EVACUAEE_FOR_LDV(from, size);
 #endif
-    return (StgClosure *)to;
 }
 
 
@@ -207,8 +207,8 @@ copy_noscav_tag(StgClosure *src, nat size, step *stp, StgWord tag)
  * pointer of an object, but reserve some padding after it.  This is
  * used to optimise evacuation of BLACKHOLEs.
  */
-static StgClosure *
-copyPart(StgClosure *src, nat size_to_reserve, nat size_to_copy, step *stp)
+static void
+copyPart(StgClosure **p, StgClosure *src, nat size_to_reserve, nat size_to_copy, step *stp)
 {
     StgPtr to, from;
     nat i;
@@ -228,6 +228,7 @@ copyPart(StgClosure *src, nat size_to_reserve, nat size_to_copy, step *stp)
 #endif
     
     to = alloc_for_copy(size_to_reserve, stp);
+    *p = (StgClosure *)to;
 
     TICK_GC_WORDS_COPIED(size_to_copy);
 
@@ -251,21 +252,20 @@ copyPart(StgClosure *src, nat size_to_reserve, nat size_to_copy, step *stp)
     if (size_to_reserve - size_to_copy > 0)
 	LDV_FILL_SLOP(to + size_to_copy - 1, (int)(size_to_reserve - size_to_copy)); 
 #endif
-    return (StgClosure *)to;
 }
 
 
 /* Copy wrappers that don't tag the closure after copying */
-STATIC_INLINE StgClosure *
-copy(StgClosure *src, nat size, step *stp)
+STATIC_INLINE void
+copy(StgClosure **p, StgClosure *src, nat size, step *stp)
 {
-    return copy_tag(src,size,stp,0);
+    copy_tag(p,src,size,stp,0);
 }
 
-STATIC_INLINE StgClosure *
-copy_noscav(StgClosure *src, nat size, step *stp)
+STATIC_INLINE void
+copy_noscav(StgClosure **p, StgClosure *src, nat size, step *stp)
 {
-    return copy_noscav_tag(src,size,stp,0);
+    copy_noscav_tag(p,src,size,stp,0);
 }
 
 /* -----------------------------------------------------------------------------
@@ -378,14 +378,17 @@ evacuate_large(StgPtr p)
    extra reads/writes than we save.
    -------------------------------------------------------------------------- */
 
-REGPARM1 StgClosure *
-evacuate(StgClosure *q)
+REGPARM1 void
+evacuate(StgClosure **p)
 {
   bdescr *bd = NULL;
   step *stp;
+  StgClosure *q;
   const StgInfoTable *info;
   StgWord tag;
 
+  q = *p;
+
 loop:
   /* The tag and the pointer are split, to be merged after evacing */
   tag = GET_CLOSURE_TAG(q);
@@ -395,7 +398,7 @@ loop:
 
   if (!HEAP_ALLOCED(q)) {
 
-      if (!major_gc) return TAG_CLOSURE(tag,q);
+      if (!major_gc) return;
 
       info = get_itbl(q);
       switch (info->type) {
@@ -410,7 +413,7 @@ loop:
 	      }
 	      RELEASE_SPIN_LOCK(&static_objects_sync);
 	  }
-	  return q;
+	  return;
 	  
       case FUN_STATIC:
 	  if (info->srt_bitmap != 0 &&
@@ -422,7 +425,7 @@ loop:
 	      }
 	      RELEASE_SPIN_LOCK(&static_objects_sync);
 	  }
-	  return q;
+	  return;
 	  
       case IND_STATIC:
 	  /* If q->saved_info != NULL, then it's a revertible CAF - it'll be
@@ -437,7 +440,7 @@ loop:
 	      }
 	      RELEASE_SPIN_LOCK(&static_objects_sync);
 	  }
-	  return q;
+	  return;
 	  
       case CONSTR_STATIC:
 	  if (*STATIC_LINK(info,(StgClosure *)q) == NULL) {
@@ -451,13 +454,13 @@ loop:
 	        /* I am assuming that static_objects pointers are not
 		 * written to other objects, and thus, no need to retag. */
 	  }
-	  return TAG_CLOSURE(tag,q);
+          return;
 	  
       case CONSTR_NOCAF_STATIC:
 	  /* no need to put these on the static linked list, they don't need
 	   * to be scavenged.
 	   */
-	  return TAG_CLOSURE(tag,q);
+	  return;
 	  
       default:
 	  barf("evacuate(static): strange closure type %d", (int)(info->type));
@@ -477,7 +480,7 @@ loop:
 	  gct->failed_to_evac = rtsTrue;
 	  TICK_GC_FAILED_PROMOTION();
       }
-      return TAG_CLOSURE(tag,q);
+      return;
   }
 
   if ((bd->flags & (BF_LARGE | BF_COMPACTED | BF_EVACUATED)) != 0) {
@@ -492,7 +495,7 @@ loop:
 	      gct->failed_to_evac = rtsTrue;
 	      TICK_GC_FAILED_PROMOTION();
 	  }
-	  return TAG_CLOSURE(tag,q);
+	  return;
       }
 
       /* evacuate large objects by re-linking them onto a different list.
@@ -502,10 +505,11 @@ loop:
 	  if (info->type == TSO && 
 	      ((StgTSO *)q)->what_next == ThreadRelocated) {
 	      q = (StgClosure *)((StgTSO *)q)->link;
+              *p = q;
 	      goto loop;
 	  }
 	  evacuate_large((P_)q);
-	  return TAG_CLOSURE(tag,q);
+	  return;
       }
       
       /* If the object is in a step that we're compacting, then we
@@ -520,7 +524,7 @@ loop:
 	      }
 	      push_mark_stack((P_)q);
 	  }
-	  return TAG_CLOSURE(tag,q);
+	  return;
       }
   }
       
@@ -537,7 +541,8 @@ loop:
   case MUT_VAR_DIRTY:
   case MVAR_CLEAN:
   case MVAR_DIRTY:
-      return copy(q,sizeW_fromITBL(info),stp);
+      copy(p,q,sizeW_fromITBL(info),stp);
+      return;
 
   case CONSTR_0_1:
   { 
@@ -545,28 +550,32 @@ loop:
       if (q->header.info == Czh_con_info &&
 	  // unsigned, so always true:  (StgChar)w >= MIN_CHARLIKE &&  
 	  (StgChar)w <= MAX_CHARLIKE) {
-	  return TAG_CLOSURE(tag,
-			     (StgClosure *)CHARLIKE_CLOSURE((StgChar)w)
-			     );
+	  *p =  TAG_CLOSURE(tag,
+                            (StgClosure *)CHARLIKE_CLOSURE((StgChar)w)
+			   );
       }
       if (q->header.info == Izh_con_info &&
 	  (StgInt)w >= MIN_INTLIKE && (StgInt)w <= MAX_INTLIKE) {
-	  return TAG_CLOSURE(tag,
+	  *p = TAG_CLOSURE(tag,
 			     (StgClosure *)INTLIKE_CLOSURE((StgInt)w)
 			     );
       }
-      // else
-      return copy_noscav_tag(q,sizeofW(StgHeader)+1,stp,tag);
+      else {
+          copy_noscav_tag(p,q,sizeofW(StgHeader)+1,stp,tag);
+      }
+      return;
   }
 
   case FUN_0_1:
   case FUN_1_0:
   case CONSTR_1_0:
-    return copy_tag(q,sizeofW(StgHeader)+1,stp,tag);
+      copy_tag(p,q,sizeofW(StgHeader)+1,stp,tag);
+      return;
 
   case THUNK_1_0:
   case THUNK_0_1:
-    return copy(q,sizeofW(StgThunk)+1,stp);
+      copy(p,q,sizeofW(StgThunk)+1,stp);
+      return;
 
   case THUNK_1_1:
   case THUNK_2_0:
@@ -578,20 +587,24 @@ loop:
       stp = bd->step;
     }
 #endif
-    return copy(q,sizeofW(StgThunk)+2,stp);
+    copy(p,q,sizeofW(StgThunk)+2,stp);
+    return;
 
   case FUN_1_1:
   case FUN_2_0:
   case FUN_0_2:
   case CONSTR_1_1:
   case CONSTR_2_0:
-    return copy_tag(q,sizeofW(StgHeader)+2,stp,tag);
+      copy_tag(p,q,sizeofW(StgHeader)+2,stp,tag);
+      return;
 
   case CONSTR_0_2:
-    return copy_noscav_tag(q,sizeofW(StgHeader)+2,stp,tag);
+      copy_noscav_tag(p,q,sizeofW(StgHeader)+2,stp,tag);
+      return;
 
   case THUNK:
-    return copy(q,thunk_sizeW_fromITBL(info),stp);
+      copy(p,q,thunk_sizeW_fromITBL(info),stp);
+      return;
 
   case FUN:
   case IND_PERM:
@@ -599,24 +612,29 @@ loop:
   case WEAK:
   case STABLE_NAME:
   case CONSTR:
-    return copy_tag(q,sizeW_fromITBL(info),stp,tag);
+      copy_tag(p,q,sizeW_fromITBL(info),stp,tag);
+      return;
 
   case BCO:
-    return copy(q,bco_sizeW((StgBCO *)q),stp);
+      copy(p,q,bco_sizeW((StgBCO *)q),stp);
+      return;
 
   case CAF_BLACKHOLE:
   case SE_CAF_BLACKHOLE:
   case SE_BLACKHOLE:
   case BLACKHOLE:
-    return copyPart(q,BLACKHOLE_sizeW(),sizeofW(StgHeader),stp);
+      copyPart(p,q,BLACKHOLE_sizeW(),sizeofW(StgHeader),stp);
+      return;
 
   case THUNK_SELECTOR:
-    return eval_thunk_selector((StgSelector *)q, rtsTrue);
+      eval_thunk_selector(p, (StgSelector *)q, rtsTrue);
+      return;
 
   case IND:
   case IND_OLDGEN:
     // follow chains of indirections, don't evacuate them 
     q = ((StgInd*)q)->indirectee;
+    *p = q;
     goto loop;
 
   case RET_BCO:
@@ -633,13 +651,16 @@ loop:
     barf("evacuate: stack frame at %p\n", q);
 
   case PAP:
-      return copy(q,pap_sizeW((StgPAP*)q),stp);
+      copy(p,q,pap_sizeW((StgPAP*)q),stp);
+      return;
 
   case AP:
-      return copy(q,ap_sizeW((StgAP*)q),stp);
+      copy(p,q,ap_sizeW((StgAP*)q),stp);
+      return;
 
   case AP_STACK:
-      return copy(q,ap_stack_sizeW((StgAP_STACK*)q),stp);
+      copy(p,q,ap_stack_sizeW((StgAP_STACK*)q),stp);
+      return;
 
   case EVACUATED:
     /* Already evacuated, just return the forwarding address.
@@ -658,25 +679,30 @@ loop:
      * current object would be evacuated to, so we only do the full
      * check if stp is too low.
      */
-    if (gct->evac_gen > 0 && stp->gen_no < gct->evac_gen) {  // optimisation 
-      StgClosure *p = ((StgEvacuated*)q)->evacuee;
-      if (HEAP_ALLOCED(p) && Bdescr((P_)p)->gen_no < gct->evac_gen) {
-	gct->failed_to_evac = rtsTrue;
-	TICK_GC_FAILED_PROMOTION();
+  {
+      StgClosure *e = ((StgEvacuated*)q)->evacuee;
+      *p = e;
+      if (gct->evac_gen > 0 && stp->gen_no < gct->evac_gen) {  // optimisation 
+	  if (HEAP_ALLOCED(e) && Bdescr((P_)e)->gen_no < gct->evac_gen) {
+	      gct->failed_to_evac = rtsTrue;
+	      TICK_GC_FAILED_PROMOTION();
+	  }
       }
-    }
-    return ((StgEvacuated*)q)->evacuee;
+      return;
+  }
 
   case ARR_WORDS:
       // just copy the block 
-      return copy_noscav(q,arr_words_sizeW((StgArrWords *)q),stp);
+      copy_noscav(p,q,arr_words_sizeW((StgArrWords *)q),stp);
+      return;
 
   case MUT_ARR_PTRS_CLEAN:
   case MUT_ARR_PTRS_DIRTY:
   case MUT_ARR_PTRS_FROZEN:
   case MUT_ARR_PTRS_FROZEN0:
       // just copy the block 
-      return copy(q,mut_arr_ptrs_sizeW((StgMutArrPtrs *)q),stp);
+      copy(p,q,mut_arr_ptrs_sizeW((StgMutArrPtrs *)q),stp);
+      return;
 
   case TSO:
     {
@@ -686,6 +712,7 @@ loop:
        */
       if (tso->what_next == ThreadRelocated) {
 	q = (StgClosure *)tso->link;
+	*p = q;
 	goto loop;
       }
 
@@ -694,38 +721,42 @@ loop:
        */
       {
 	  StgTSO *new_tso;
-	  StgPtr p, q;
+	  StgPtr r, s;
 
-	  new_tso = (StgTSO *)copyPart((StgClosure *)tso,
-				       tso_sizeW(tso),
-				       sizeofW(StgTSO), stp);
+	  copyPart(p,(StgClosure *)tso, tso_sizeW(tso), sizeofW(StgTSO), stp);
+	  new_tso = (StgTSO *)*p;
 	  move_TSO(tso, new_tso);
-	  for (p = tso->sp, q = new_tso->sp;
-	       p < tso->stack+tso->stack_size;) {
-	      *q++ = *p++;
+	  for (r = tso->sp, s = new_tso->sp;
+	       r < tso->stack+tso->stack_size;) {
+	      *s++ = *r++;
 	  }
-	  
-	  return (StgClosure *)new_tso;
+	  return;
       }
     }
 
   case TREC_HEADER: 
-    return copy(q,sizeofW(StgTRecHeader),stp);
+      copy(p,q,sizeofW(StgTRecHeader),stp);
+      return;
 
   case TVAR_WATCH_QUEUE:
-    return copy(q,sizeofW(StgTVarWatchQueue),stp);
+      copy(p,q,sizeofW(StgTVarWatchQueue),stp);
+      return;
 
   case TVAR:
-    return copy(q,sizeofW(StgTVar),stp);
+      copy(p,q,sizeofW(StgTVar),stp);
+      return;
     
   case TREC_CHUNK:
-    return copy(q,sizeofW(StgTRecChunk),stp);
+      copy(p,q,sizeofW(StgTRecChunk),stp);
+      return;
 
   case ATOMIC_INVARIANT:
-    return copy(q,sizeofW(StgAtomicInvariant),stp);
+      copy(p,q,sizeofW(StgAtomicInvariant),stp);
+      return;
 
   case INVARIANT_CHECK_QUEUE:
-    return copy(q,sizeofW(StgInvariantCheckQueue),stp);
+      copy(p,q,sizeofW(StgInvariantCheckQueue),stp);
+      return;
 
   default:
     barf("evacuate: strange closure type %d", (int)(info->type));
@@ -774,8 +805,9 @@ unchain_thunk_selectors(StgSelector *p, StgClosure *val)
    evacuated.
    -------------------------------------------------------------------------- */
 
-static StgClosure *
-eval_thunk_selector (StgSelector * p, rtsBool evac)
+static void
+eval_thunk_selector (StgClosure **q, StgSelector * p, rtsBool evac)
+                 // NB. for legacy reasons, p & q are swapped around :(
 {
     nat field;
     StgInfoTable *info;
@@ -810,7 +842,8 @@ selector_chain:
         // mutable list.
         if ((bd->gen_no > N) || (bd->flags & BF_EVACUATED)) {
             unchain_thunk_selectors(prev_thunk_selector, (StgClosure *)p);
-            return (StgClosure *)p;
+            *q = (StgClosure *)p;
+            return;
         }
         // we don't update THUNK_SELECTORS in the compacted
         // generation, because compaction does not remove the INDs
@@ -820,9 +853,10 @@ selector_chain:
         // around here, test by compiling stage 3 with +RTS -c -RTS.
         if (bd->flags & BF_COMPACTED) {
             // must call evacuate() to mark this closure if evac==rtsTrue
-            if (evac) p = (StgSelector *)evacuate((StgClosure *)p);
+            *q = (StgClosure *)p;
+            if (evac) evacuate(q);
             unchain_thunk_selectors(prev_thunk_selector, (StgClosure *)p);
-            return (StgClosure *)p;
+            return;
         }
     }
 
@@ -889,12 +923,14 @@ selector_loop:
                   ((StgClosure*)p)->payload[0] = (StgClosure *)prev_thunk_selector;
                   prev_thunk_selector = p;
 
-                  if (evac) val = evacuate(val);
+                  *q = val;
+                  if (evac) evacuate(q);
+                  val = *q;
                   // evacuate() cannot recurse through
                   // eval_thunk_selector(), because we know val is not
                   // a THUNK_SELECTOR.
                   unchain_thunk_selectors(prev_thunk_selector, val);
-                  return val;
+                  return;
               }
           }
 
@@ -927,7 +963,7 @@ selector_loop:
           // rtsFalse says "don't evacuate the result".  It will,
           // however, update any THUNK_SELECTORs that are evaluated
           // along the way.
-	  val = eval_thunk_selector((StgSelector *)selectee, rtsFalse);
+	  eval_thunk_selector(&val, (StgSelector*)selectee, rtsFalse);
 	  gct->thunk_selector_depth--;
 
           // did we actually manage to evaluate it?
@@ -964,12 +1000,13 @@ bale_out:
     // pointer.  But don't forget: we still need to evacuate the thunk itself.
     SET_INFO(p, info_ptr);
     if (evac) {
-        val = copy((StgClosure *)p,THUNK_SELECTOR_sizeW(),bd->step->to);
+        copy(&val,(StgClosure *)p,THUNK_SELECTOR_sizeW(),bd->step->to);
     } else {
         val = (StgClosure *)p;
     }
+    *q = val;
     unchain_thunk_selectors(prev_thunk_selector, val);
-    return val;
+    return;
 }
 
 /* -----------------------------------------------------------------------------
diff --git a/rts/sm/Evac.h b/rts/sm/Evac.h
index beaba755c1..c73008fb1b 100644
--- a/rts/sm/Evac.h
+++ b/rts/sm/Evac.h
@@ -27,6 +27,6 @@
 #define REGPARM1
 #endif
 
-REGPARM1 StgClosure * evacuate (StgClosure *q);
+REGPARM1 void evacuate (StgClosure **p);
 
 extern lnat thunk_selector_depth;
diff --git a/rts/sm/GC.c b/rts/sm/GC.c
index eed2da7020..d064f1d656 100644
--- a/rts/sm/GC.c
+++ b/rts/sm/GC.c
@@ -183,9 +183,12 @@ GarbageCollect ( rtsBool force_major_gc )
   lnat live, allocated;
   lnat oldgen_saved_blocks = 0;
   nat n_threads; // number of threads participating in GC
-
+  gc_thread *saved_gct;
   nat g, s, t;
 
+  // necessary if we stole a callee-saves register for gct:
+  saved_gct = gct;
+
 #ifdef PROFILING
   CostCentreStack *prev_CCS;
 #endif
@@ -679,6 +682,8 @@ GarbageCollect ( rtsBool force_major_gc )
 #endif
 
   RELEASE_SM_LOCK;
+
+  gct = saved_gct;
 }
 
 /* ---------------------------------------------------------------------------
@@ -1295,7 +1300,7 @@ init_gc_thread (gc_thread *t)
 static void
 mark_root(StgClosure **root)
 {
-  *root = evacuate(*root);
+  evacuate(root);
 }
 
 /* -----------------------------------------------------------------------------
diff --git a/rts/sm/MarkWeak.c b/rts/sm/MarkWeak.c
index bfa78e5836..62ac8e0397 100644
--- a/rts/sm/MarkWeak.c
+++ b/rts/sm/MarkWeak.c
@@ -137,8 +137,8 @@ traverseWeakPtrList(void)
 	      if (new != NULL) {
 		  w->key = new;
 		  // evacuate the value and finalizer 
-		  w->value = evacuate(w->value);
-		  w->finalizer = evacuate(w->finalizer);
+		  evacuate(&w->value);
+		  evacuate(&w->finalizer);
 		  // remove this weak ptr from the old_weak_ptr list 
 		  *last_w = w->link;
 		  // and put it on the new weak ptr list 
@@ -169,7 +169,7 @@ traverseWeakPtrList(void)
        */
       if (flag == rtsFalse) {
 	  for (w = old_weak_ptr_list; w; w = w->link) {
-	      w->finalizer = evacuate(w->finalizer);
+	      evacuate(&w->finalizer);
 	  }
 
 	  // Next, move to the WeakThreads stage after fully
@@ -241,7 +241,8 @@ traverseWeakPtrList(void)
 	  StgTSO *t, *tmp, *next;
 	  for (t = old_all_threads; t != END_TSO_QUEUE; t = next) {
 	      next = t->global_link;
-	      tmp = (StgTSO *)evacuate((StgClosure *)t);
+	      tmp = t;
+	      evacuate((StgClosure **)&tmp);
 	      tmp->global_link = resurrected_threads;
 	      resurrected_threads = tmp;
 	  }
@@ -301,7 +302,8 @@ traverseBlackholeQueue (void)
                     continue;
                 }
             }
-            t = (StgTSO *)evacuate((StgClosure *)t);
+            tmp = t;
+            evacuate((StgClosure **)&tmp);
             if (prev) prev->link = t;
             flag = rtsTrue;
 	}
@@ -324,14 +326,15 @@ traverseBlackholeQueue (void)
 void
 markWeakPtrList ( void )
 {
-  StgWeak *w, **last_w;
+  StgWeak *w, **last_w, *tmp;
 
   last_w = &weak_ptr_list;
   for (w = weak_ptr_list; w; w = w->link) {
       // w might be WEAK, EVACUATED, or DEAD_WEAK (actually CON_STATIC) here
       ASSERT(w->header.info == &stg_DEAD_WEAK_info 
 	     || get_itbl(w)->type == WEAK || get_itbl(w)->type == EVACUATED);
-      w = (StgWeak *)evacuate((StgClosure *)w);
+      tmp = w;
+      evacuate((StgClosure **)&tmp);
       *last_w = w;
       last_w = &(w->link);
   }
diff --git a/rts/sm/Scav.c b/rts/sm/Scav.c
index 8d7e5825ee..080c75014e 100644
--- a/rts/sm/Scav.c
+++ b/rts/sm/Scav.c
@@ -50,7 +50,7 @@ scavenge_large_srt_bitmap( StgLargeSRT *large_srt )
     p      = (StgClosure **)large_srt->srt;
     for (i = 0; i < size; ) {
 	if ((bitmap & 1) != 0) {
-	    evacuate(*p);
+	    evacuate(p);
 	}
 	i++;
 	p++;
@@ -93,12 +93,12 @@ scavenge_srt (StgClosure **srt, nat srt_bitmap)
 	  // If the SRT entry hasn't got bit 0 set, the SRT entry points to a
 	  // closure that's fixed at link-time, and no extra magic is required.
 	  if ( (unsigned long)(*srt) & 0x1 ) {
-	      evacuate(*stgCast(StgClosure**,(stgCast(unsigned long, *srt) & ~0x1)));
+	      evacuate(stgCast(StgClosure**,(stgCast(unsigned long, *srt) & ~0x1)));
 	  } else {
-	      evacuate(*p);
+	      evacuate(p);
 	  }
 #else
-	  evacuate(*p);
+	  evacuate(p);
 #endif
       }
       p++;
@@ -140,20 +140,19 @@ scavengeTSO (StgTSO *tso)
 	|| tso->why_blocked == BlockedOnBlackHole
 	|| tso->why_blocked == BlockedOnException
 	) {
-	tso->block_info.closure = evacuate(tso->block_info.closure);
+	evacuate(&tso->block_info.closure);
     }
-    tso->blocked_exceptions = 
-	(StgTSO *)evacuate((StgClosure *)tso->blocked_exceptions);
+    evacuate((StgClosure **)&tso->blocked_exceptions);
     
     // We don't always chase the link field: TSOs on the blackhole
     // queue are not automatically alive, so the link field is a
     // "weak" pointer in that case.
     if (tso->why_blocked != BlockedOnBlackHole) {
-	tso->link = (StgTSO *)evacuate((StgClosure *)tso->link);
+	evacuate((StgClosure **)&tso->link);
     }
 
     // scavange current transaction record
-    tso->trec = (StgTRecHeader *)evacuate((StgClosure *)tso->trec);
+    evacuate((StgClosure **)&tso->trec);
     
     // scavenge this thread's stack 
     scavenge_stack(tso->sp, &(tso->stack[tso->stack_size]));
@@ -188,7 +187,7 @@ scavenge_arg_block (StgFunInfoTable *fun_info, StgClosure **args)
     small_bitmap:
 	while (size > 0) {
 	    if ((bitmap & 1) == 0) {
-		*p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+		evacuate((StgClosure **)p);
 	    }
 	    p++;
 	    bitmap = bitmap >> 1;
@@ -227,7 +226,7 @@ scavenge_PAP_payload (StgClosure *fun, StgClosure **payload, StgWord size)
     small_bitmap:
 	while (size > 0) {
 	    if ((bitmap & 1) == 0) {
-		*p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+		evacuate((StgClosure **)p);
 	    }
 	    p++;
 	    bitmap = bitmap >> 1;
@@ -241,14 +240,14 @@ scavenge_PAP_payload (StgClosure *fun, StgClosure **payload, StgWord size)
 STATIC_INLINE StgPtr
 scavenge_PAP (StgPAP *pap)
 {
-    pap->fun = evacuate(pap->fun);
+    evacuate(&pap->fun);
     return scavenge_PAP_payload (pap->fun, pap->payload, pap->n_args);
 }
 
 STATIC_INLINE StgPtr
 scavenge_AP (StgAP *ap)
 {
-    ap->fun = evacuate(ap->fun);
+    evacuate(&ap->fun);
     return scavenge_PAP_payload (ap->fun, ap->payload, ap->n_args);
 }
 
@@ -300,9 +299,9 @@ scavenge_block (bdescr *bd, StgPtr scan)
 
 	StgMVar *mvar = ((StgMVar *)p);
 	gct->eager_promotion = rtsFalse;
-	mvar->head = (StgTSO *)evacuate((StgClosure *)mvar->head);
-	mvar->tail = (StgTSO *)evacuate((StgClosure *)mvar->tail);
-	mvar->value = evacuate((StgClosure *)mvar->value);
+	evacuate((StgClosure **)&mvar->head);
+	evacuate((StgClosure **)&mvar->tail);
+	evacuate((StgClosure **)&mvar->value);
 	gct->eager_promotion = saved_eager_promotion;
 
 	if (gct->failed_to_evac) {
@@ -316,34 +315,34 @@ scavenge_block (bdescr *bd, StgPtr scan)
 
     case FUN_2_0:
 	scavenge_fun_srt(info);
-	((StgClosure *)p)->payload[1] = evacuate(((StgClosure *)p)->payload[1]);
-	((StgClosure *)p)->payload[0] = evacuate(((StgClosure *)p)->payload[0]);
+	evacuate(&((StgClosure *)p)->payload[1]);
+	evacuate(&((StgClosure *)p)->payload[0]);
 	p += sizeofW(StgHeader) + 2;
 	break;
 
     case THUNK_2_0:
 	scavenge_thunk_srt(info);
-	((StgThunk *)p)->payload[1] = evacuate(((StgThunk *)p)->payload[1]);
-	((StgThunk *)p)->payload[0] = evacuate(((StgThunk *)p)->payload[0]);
+	evacuate(&((StgThunk *)p)->payload[1]);
+	evacuate(&((StgThunk *)p)->payload[0]);
 	p += sizeofW(StgThunk) + 2;
 	break;
 
     case CONSTR_2_0:
-	((StgClosure *)p)->payload[1] = evacuate(((StgClosure *)p)->payload[1]);
-	((StgClosure *)p)->payload[0] = evacuate(((StgClosure *)p)->payload[0]);
+	evacuate(&((StgClosure *)p)->payload[1]);
+	evacuate(&((StgClosure *)p)->payload[0]);
 	p += sizeofW(StgHeader) + 2;
 	break;
 	
     case THUNK_1_0:
 	scavenge_thunk_srt(info);
-	((StgThunk *)p)->payload[0] = evacuate(((StgThunk *)p)->payload[0]);
+	evacuate(&((StgThunk *)p)->payload[0]);
 	p += sizeofW(StgThunk) + 1;
 	break;
 	
     case FUN_1_0:
 	scavenge_fun_srt(info);
     case CONSTR_1_0:
-	((StgClosure *)p)->payload[0] = evacuate(((StgClosure *)p)->payload[0]);
+	evacuate(&((StgClosure *)p)->payload[0]);
 	p += sizeofW(StgHeader) + 1;
 	break;
 	
@@ -371,14 +370,14 @@ scavenge_block (bdescr *bd, StgPtr scan)
 	
     case THUNK_1_1:
 	scavenge_thunk_srt(info);
-	((StgThunk *)p)->payload[0] = evacuate(((StgThunk *)p)->payload[0]);
+	evacuate(&((StgThunk *)p)->payload[0]);
 	p += sizeofW(StgThunk) + 2;
 	break;
 
     case FUN_1_1:
 	scavenge_fun_srt(info);
     case CONSTR_1_1:
-	((StgClosure *)p)->payload[0] = evacuate(((StgClosure *)p)->payload[0]);
+	evacuate(&((StgClosure *)p)->payload[0]);
 	p += sizeofW(StgHeader) + 2;
 	break;
 	
@@ -393,7 +392,7 @@ scavenge_block (bdescr *bd, StgPtr scan)
 	scavenge_thunk_srt(info);
 	end = (P_)((StgThunk *)p)->payload + info->layout.payload.ptrs;
 	for (p = (P_)((StgThunk *)p)->payload; p < end; p++) {
-	    *p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+	    evacuate((StgClosure **)p);
 	}
 	p += info->layout.payload.nptrs;
 	break;
@@ -408,7 +407,7 @@ scavenge_block (bdescr *bd, StgPtr scan)
 
 	end = (P_)((StgClosure *)p)->payload + info->layout.payload.ptrs;
 	for (p = (P_)((StgClosure *)p)->payload; p < end; p++) {
-	    *p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+	    evacuate((StgClosure **)p);
 	}
 	p += info->layout.payload.nptrs;
 	break;
@@ -416,9 +415,9 @@ scavenge_block (bdescr *bd, StgPtr scan)
 
     case BCO: {
 	StgBCO *bco = (StgBCO *)p;
-	bco->instrs = (StgArrWords *)evacuate((StgClosure *)bco->instrs);
-	bco->literals = (StgArrWords *)evacuate((StgClosure *)bco->literals);
-	bco->ptrs = (StgMutArrPtrs *)evacuate((StgClosure *)bco->ptrs);
+	evacuate((StgClosure **)&bco->instrs);
+	evacuate((StgClosure **)&bco->literals);
+	evacuate((StgClosure **)&bco->ptrs);
 	p += bco_sizeW(bco);
 	break;
     }
@@ -441,7 +440,7 @@ scavenge_block (bdescr *bd, StgPtr scan)
       }
 	// fall through 
     case IND_OLDGEN_PERM:
-	((StgInd *)p)->indirectee = evacuate(((StgInd *)p)->indirectee);
+	evacuate(&((StgInd *)p)->indirectee);
 	p += sizeofW(StgInd);
 	break;
 
@@ -450,7 +449,7 @@ scavenge_block (bdescr *bd, StgPtr scan)
 	rtsBool saved_eager_promotion = gct->eager_promotion;
 
 	gct->eager_promotion = rtsFalse;
-	((StgMutVar *)p)->var = evacuate(((StgMutVar *)p)->var);
+	evacuate(&((StgMutVar *)p)->var);
 	gct->eager_promotion = saved_eager_promotion;
 
 	if (gct->failed_to_evac) {
@@ -472,7 +471,7 @@ scavenge_block (bdescr *bd, StgPtr scan)
     case THUNK_SELECTOR:
     { 
 	StgSelector *s = (StgSelector *)p;
-	s->selectee = evacuate(s->selectee);
+	evacuate(&s->selectee);
 	p += THUNK_SELECTOR_sizeW();
 	break;
     }
@@ -482,7 +481,7 @@ scavenge_block (bdescr *bd, StgPtr scan)
     {
 	StgAP_STACK *ap = (StgAP_STACK *)p;
 
-	ap->fun = evacuate(ap->fun);
+	evacuate(&ap->fun);
 	scavenge_stack((StgPtr)ap->payload, (StgPtr)ap->payload + ap->size);
 	p = (StgPtr)ap->payload + ap->size;
 	break;
@@ -516,7 +515,7 @@ scavenge_block (bdescr *bd, StgPtr scan)
 	gct->eager_promotion = rtsFalse;
 	next = p + mut_arr_ptrs_sizeW((StgMutArrPtrs*)p);
 	for (p = (P_)((StgMutArrPtrs *)p)->payload; p < next; p++) {
-	    *p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+	    evacuate((StgClosure **)p);
 	}
 	gct->eager_promotion = saved_eager;
 
@@ -538,7 +537,7 @@ scavenge_block (bdescr *bd, StgPtr scan)
 
 	next = p + mut_arr_ptrs_sizeW((StgMutArrPtrs*)p);
 	for (p = (P_)((StgMutArrPtrs *)p)->payload; p < next; p++) {
-	    *p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+	    evacuate((StgClosure **)p);
 	}
 
 	// If we're going to put this object on the mutable list, then
@@ -575,9 +574,9 @@ scavenge_block (bdescr *bd, StgPtr scan)
       {
 	StgTVarWatchQueue *wq = ((StgTVarWatchQueue *) p);
 	gct->evac_gen = 0;
-	wq->closure = (StgClosure*)evacuate((StgClosure*)wq->closure);
-	wq->next_queue_entry = (StgTVarWatchQueue *)evacuate((StgClosure*)wq->next_queue_entry);
-	wq->prev_queue_entry = (StgTVarWatchQueue *)evacuate((StgClosure*)wq->prev_queue_entry);
+	evacuate((StgClosure **)&wq->closure);
+	evacuate((StgClosure **)&wq->next_queue_entry);
+	evacuate((StgClosure **)&wq->prev_queue_entry);
 	gct->evac_gen = saved_evac_gen;
 	gct->failed_to_evac = rtsTrue; // mutable
 	p += sizeofW(StgTVarWatchQueue);
@@ -588,8 +587,8 @@ scavenge_block (bdescr *bd, StgPtr scan)
       {
 	StgTVar *tvar = ((StgTVar *) p);
 	gct->evac_gen = 0;
-	tvar->current_value = evacuate((StgClosure*)tvar->current_value);
-	tvar->first_watch_queue_entry = (StgTVarWatchQueue *)evacuate((StgClosure*)tvar->first_watch_queue_entry);
+	evacuate((StgClosure **)&tvar->current_value);
+	evacuate((StgClosure **)&tvar->first_watch_queue_entry);
 	gct->evac_gen = saved_evac_gen;
 	gct->failed_to_evac = rtsTrue; // mutable
 	p += sizeofW(StgTVar);
@@ -600,9 +599,9 @@ scavenge_block (bdescr *bd, StgPtr scan)
       {
         StgTRecHeader *trec = ((StgTRecHeader *) p);
         gct->evac_gen = 0;
-	trec->enclosing_trec = (StgTRecHeader *)evacuate((StgClosure*)trec->enclosing_trec);
-	trec->current_chunk = (StgTRecChunk *)evacuate((StgClosure*)trec->current_chunk);
-	trec->invariants_to_check = (StgInvariantCheckQueue *)evacuate((StgClosure*)trec->invariants_to_check);
+	evacuate((StgClosure **)&trec->enclosing_trec);
+	evacuate((StgClosure **)&trec->current_chunk);
+	evacuate((StgClosure **)&trec->invariants_to_check);
 	gct->evac_gen = saved_evac_gen;
 	gct->failed_to_evac = rtsTrue; // mutable
 	p += sizeofW(StgTRecHeader);
@@ -615,11 +614,11 @@ scavenge_block (bdescr *bd, StgPtr scan)
 	StgTRecChunk *tc = ((StgTRecChunk *) p);
 	TRecEntry *e = &(tc -> entries[0]);
 	gct->evac_gen = 0;
-	tc->prev_chunk = (StgTRecChunk *)evacuate((StgClosure*)tc->prev_chunk);
+	evacuate((StgClosure **)&tc->prev_chunk);
 	for (i = 0; i < tc -> next_entry_idx; i ++, e++ ) {
-	  e->tvar = (StgTVar *)evacuate((StgClosure*)e->tvar);
-	  e->expected_value = evacuate((StgClosure*)e->expected_value);
-	  e->new_value = evacuate((StgClosure*)e->new_value);
+	  evacuate((StgClosure **)&e->tvar);
+	  evacuate((StgClosure **)&e->expected_value);
+	  evacuate((StgClosure **)&e->new_value);
 	}
 	gct->evac_gen = saved_evac_gen;
 	gct->failed_to_evac = rtsTrue; // mutable
@@ -631,8 +630,8 @@ scavenge_block (bdescr *bd, StgPtr scan)
       {
         StgAtomicInvariant *invariant = ((StgAtomicInvariant *) p);
         gct->evac_gen = 0;
-	invariant->code = (StgClosure *)evacuate(invariant->code);
-	invariant->last_execution = (StgTRecHeader *)evacuate((StgClosure*)invariant->last_execution);
+	evacuate(&invariant->code);
+	evacuate((StgClosure **)&invariant->last_execution);
 	gct->evac_gen = saved_evac_gen;
 	gct->failed_to_evac = rtsTrue; // mutable
 	p += sizeofW(StgAtomicInvariant);
@@ -643,9 +642,9 @@ scavenge_block (bdescr *bd, StgPtr scan)
       {
         StgInvariantCheckQueue *queue = ((StgInvariantCheckQueue *) p);
         gct->evac_gen = 0;
-	queue->invariant = (StgAtomicInvariant *)evacuate((StgClosure*)queue->invariant);
-	queue->my_execution = (StgTRecHeader *)evacuate((StgClosure*)queue->my_execution);
-	queue->next_queue_entry = (StgInvariantCheckQueue *)evacuate((StgClosure*)queue->next_queue_entry);
+	evacuate((StgClosure **)&queue->invariant);
+	evacuate((StgClosure **)&queue->my_execution);
+	evacuate((StgClosure **)&queue->next_queue_entry);
 	gct->evac_gen = saved_evac_gen;
 	gct->failed_to_evac = rtsTrue; // mutable
 	p += sizeofW(StgInvariantCheckQueue);
@@ -710,9 +709,9 @@ linear_scan:
             
             StgMVar *mvar = ((StgMVar *)p);
             gct->eager_promotion = rtsFalse;
-            mvar->head = (StgTSO *)evacuate((StgClosure *)mvar->head);
-            mvar->tail = (StgTSO *)evacuate((StgClosure *)mvar->tail);
-            mvar->value = evacuate((StgClosure *)mvar->value);
+            evacuate((StgClosure **)&mvar->head);
+            evacuate((StgClosure **)&mvar->tail);
+            evacuate((StgClosure **)&mvar->value);
             gct->eager_promotion = saved_eager_promotion;
             
             if (gct->failed_to_evac) {
@@ -725,36 +724,36 @@ linear_scan:
 
 	case FUN_2_0:
 	    scavenge_fun_srt(info);
-	    ((StgClosure *)p)->payload[1] = evacuate(((StgClosure *)p)->payload[1]);
-	    ((StgClosure *)p)->payload[0] = evacuate(((StgClosure *)p)->payload[0]);
+	    evacuate(&((StgClosure *)p)->payload[1]);
+	    evacuate(&((StgClosure *)p)->payload[0]);
 	    break;
 
 	case THUNK_2_0:
 	    scavenge_thunk_srt(info);
-	    ((StgThunk *)p)->payload[1] = evacuate(((StgThunk *)p)->payload[1]);
-	    ((StgThunk *)p)->payload[0] = evacuate(((StgThunk *)p)->payload[0]);
+	    evacuate(&((StgThunk *)p)->payload[1]);
+	    evacuate(&((StgThunk *)p)->payload[0]);
 	    break;
 
 	case CONSTR_2_0:
-	    ((StgClosure *)p)->payload[1] = evacuate(((StgClosure *)p)->payload[1]);
-	    ((StgClosure *)p)->payload[0] = evacuate(((StgClosure *)p)->payload[0]);
+	    evacuate(&((StgClosure *)p)->payload[1]);
+	    evacuate(&((StgClosure *)p)->payload[0]);
 	    break;
 	
 	case FUN_1_0:
 	case FUN_1_1:
 	    scavenge_fun_srt(info);
-	    ((StgClosure *)p)->payload[0] = evacuate(((StgClosure *)p)->payload[0]);
+	    evacuate(&((StgClosure *)p)->payload[0]);
 	    break;
 
 	case THUNK_1_0:
 	case THUNK_1_1:
 	    scavenge_thunk_srt(info);
-	    ((StgThunk *)p)->payload[0] = evacuate(((StgThunk *)p)->payload[0]);
+	    evacuate(&((StgThunk *)p)->payload[0]);
 	    break;
 
 	case CONSTR_1_0:
 	case CONSTR_1_1:
-	    ((StgClosure *)p)->payload[0] = evacuate(((StgClosure *)p)->payload[0]);
+	    evacuate(&((StgClosure *)p)->payload[0]);
 	    break;
 	
 	case FUN_0_1:
@@ -782,7 +781,7 @@ linear_scan:
 	    scavenge_thunk_srt(info);
 	    end = (P_)((StgThunk *)p)->payload + info->layout.payload.ptrs;
 	    for (p = (P_)((StgThunk *)p)->payload; p < end; p++) {
-		*p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+		evacuate((StgClosure **)p);
 	    }
 	    break;
 	}
@@ -796,16 +795,16 @@ linear_scan:
 	    
 	    end = (P_)((StgClosure *)p)->payload + info->layout.payload.ptrs;
 	    for (p = (P_)((StgClosure *)p)->payload; p < end; p++) {
-		*p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+		evacuate((StgClosure **)p);
 	    }
 	    break;
 	}
 
 	case BCO: {
 	    StgBCO *bco = (StgBCO *)p;
-	    bco->instrs = (StgArrWords *)evacuate((StgClosure *)bco->instrs);
-	    bco->literals = (StgArrWords *)evacuate((StgClosure *)bco->literals);
-	    bco->ptrs = (StgMutArrPtrs *)evacuate((StgClosure *)bco->ptrs);
+	    evacuate((StgClosure **)&bco->instrs);
+	    evacuate((StgClosure **)&bco->literals);
+	    evacuate((StgClosure **)&bco->ptrs);
 	    break;
 	}
 
@@ -817,8 +816,7 @@ linear_scan:
 
 	case IND_OLDGEN:
 	case IND_OLDGEN_PERM:
-	    ((StgInd *)p)->indirectee = 
-		evacuate(((StgInd *)p)->indirectee);
+	    evacuate(&((StgInd *)p)->indirectee);
 	    break;
 
 	case MUT_VAR_CLEAN:
@@ -826,7 +824,7 @@ linear_scan:
 	    rtsBool saved_eager_promotion = gct->eager_promotion;
 	    
 	    gct->eager_promotion = rtsFalse;
-	    ((StgMutVar *)p)->var = evacuate(((StgMutVar *)p)->var);
+	    evacuate(&((StgMutVar *)p)->var);
 	    gct->eager_promotion = saved_eager_promotion;
 	    
 	    if (gct->failed_to_evac) {
@@ -847,7 +845,7 @@ linear_scan:
 	case THUNK_SELECTOR:
 	{ 
 	    StgSelector *s = (StgSelector *)p;
-	    s->selectee = evacuate(s->selectee);
+	    evacuate(&s->selectee);
 	    break;
 	}
 
@@ -856,7 +854,7 @@ linear_scan:
 	{
 	    StgAP_STACK *ap = (StgAP_STACK *)p;
 	    
-	    ap->fun = evacuate(ap->fun);
+	    evacuate(&ap->fun);
 	    scavenge_stack((StgPtr)ap->payload, (StgPtr)ap->payload + ap->size);
 	    break;
 	}
@@ -884,7 +882,7 @@ linear_scan:
 	    gct->eager_promotion = rtsFalse;
 	    next = p + mut_arr_ptrs_sizeW((StgMutArrPtrs*)p);
 	    for (p = (P_)((StgMutArrPtrs *)p)->payload; p < next; p++) {
-		*p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+		evacuate((StgClosure **)p);
 	    }
 	    gct->eager_promotion = saved_eager;
 
@@ -906,7 +904,7 @@ linear_scan:
 	    
 	    next = p + mut_arr_ptrs_sizeW((StgMutArrPtrs*)p);
 	    for (p = (P_)((StgMutArrPtrs *)p)->payload; p < next; p++) {
-		*p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+		evacuate((StgClosure **)p);
 	    }
 
 	    // If we're going to put this object on the mutable list, then
@@ -942,9 +940,9 @@ linear_scan:
 	  {
 	    StgTVarWatchQueue *wq = ((StgTVarWatchQueue *) p);
 	    gct->evac_gen = 0;
-            wq->closure = (StgClosure*)evacuate((StgClosure*)wq->closure);
-	    wq->next_queue_entry = (StgTVarWatchQueue *)evacuate((StgClosure*)wq->next_queue_entry);
-	    wq->prev_queue_entry = (StgTVarWatchQueue *)evacuate((StgClosure*)wq->prev_queue_entry);
+            evacuate((StgClosure **)&wq->closure);
+	    evacuate((StgClosure **)&wq->next_queue_entry);
+	    evacuate((StgClosure **)&wq->prev_queue_entry);
 	    gct->evac_gen = saved_evac_gen;
 	    gct->failed_to_evac = rtsTrue; // mutable
 	    break;
@@ -954,8 +952,8 @@ linear_scan:
 	  {
 	    StgTVar *tvar = ((StgTVar *) p);
 	    gct->evac_gen = 0;
-	    tvar->current_value = evacuate((StgClosure*)tvar->current_value);
-	    tvar->first_watch_queue_entry = (StgTVarWatchQueue *)evacuate((StgClosure*)tvar->first_watch_queue_entry);
+	    evacuate((StgClosure **)&tvar->current_value);
+	    evacuate((StgClosure **)&tvar->first_watch_queue_entry);
 	    gct->evac_gen = saved_evac_gen;
 	    gct->failed_to_evac = rtsTrue; // mutable
 	    break;
@@ -967,11 +965,11 @@ linear_scan:
 	    StgTRecChunk *tc = ((StgTRecChunk *) p);
 	    TRecEntry *e = &(tc -> entries[0]);
 	    gct->evac_gen = 0;
-	    tc->prev_chunk = (StgTRecChunk *)evacuate((StgClosure*)tc->prev_chunk);
+	    evacuate((StgClosure **)&tc->prev_chunk);
 	    for (i = 0; i < tc -> next_entry_idx; i ++, e++ ) {
-	      e->tvar = (StgTVar *)evacuate((StgClosure*)e->tvar);
-	      e->expected_value = evacuate((StgClosure*)e->expected_value);
-	      e->new_value = evacuate((StgClosure*)e->new_value);
+	      evacuate((StgClosure **)&e->tvar);
+	      evacuate((StgClosure **)&e->expected_value);
+	      evacuate((StgClosure **)&e->new_value);
 	    }
 	    gct->evac_gen = saved_evac_gen;
 	    gct->failed_to_evac = rtsTrue; // mutable
@@ -982,9 +980,9 @@ linear_scan:
 	  {
 	    StgTRecHeader *trec = ((StgTRecHeader *) p);
 	    gct->evac_gen = 0;
-	    trec->enclosing_trec = (StgTRecHeader *)evacuate((StgClosure*)trec->enclosing_trec);
-	    trec->current_chunk = (StgTRecChunk *)evacuate((StgClosure*)trec->current_chunk);
-  	    trec->invariants_to_check = (StgInvariantCheckQueue *)evacuate((StgClosure*)trec->invariants_to_check);
+	    evacuate((StgClosure **)&trec->enclosing_trec);
+	    evacuate((StgClosure **)&trec->current_chunk);
+  	    evacuate((StgClosure **)&trec->invariants_to_check);
 	    gct->evac_gen = saved_evac_gen;
 	    gct->failed_to_evac = rtsTrue; // mutable
 	    break;
@@ -994,8 +992,8 @@ linear_scan:
           {
             StgAtomicInvariant *invariant = ((StgAtomicInvariant *) p);
             gct->evac_gen = 0;
-	    invariant->code = (StgClosure *)evacuate(invariant->code);
-    	    invariant->last_execution = (StgTRecHeader *)evacuate((StgClosure*)invariant->last_execution);
+	    evacuate(&invariant->code);
+    	    evacuate((StgClosure **)&invariant->last_execution);
 	    gct->evac_gen = saved_evac_gen;
 	    gct->failed_to_evac = rtsTrue; // mutable
             break;
@@ -1005,9 +1003,9 @@ linear_scan:
           {
             StgInvariantCheckQueue *queue = ((StgInvariantCheckQueue *) p);
             gct->evac_gen = 0;
-    	    queue->invariant = (StgAtomicInvariant *)evacuate((StgClosure*)queue->invariant);
-	    queue->my_execution = (StgTRecHeader *)evacuate((StgClosure*)queue->my_execution);
-            queue->next_queue_entry = (StgInvariantCheckQueue *)evacuate((StgClosure*)queue->next_queue_entry);
+    	    evacuate((StgClosure **)&queue->invariant);
+	    evacuate((StgClosure **)&queue->my_execution);
+            evacuate((StgClosure **)&queue->next_queue_entry);
 	    gct->evac_gen = saved_evac_gen;
 	    gct->failed_to_evac = rtsTrue; // mutable
             break;
@@ -1096,9 +1094,9 @@ scavenge_one(StgPtr p)
 
 	StgMVar *mvar = ((StgMVar *)p);
 	gct->eager_promotion = rtsFalse;
-	mvar->head = (StgTSO *)evacuate((StgClosure *)mvar->head);
-	mvar->tail = (StgTSO *)evacuate((StgClosure *)mvar->tail);
-	mvar->value = evacuate((StgClosure *)mvar->value);
+	evacuate((StgClosure **)&mvar->head);
+	evacuate((StgClosure **)&mvar->tail);
+	evacuate((StgClosure **)&mvar->value);
 	gct->eager_promotion = saved_eager_promotion;
 
 	if (gct->failed_to_evac) {
@@ -1120,7 +1118,7 @@ scavenge_one(StgPtr p)
 	
 	end = (StgPtr)((StgThunk *)p)->payload + info->layout.payload.ptrs;
 	for (q = (StgPtr)((StgThunk *)p)->payload; q < end; q++) {
-	    *q = (StgWord)(StgPtr)evacuate((StgClosure *)*q);
+	    evacuate((StgClosure **)q);
 	}
 	break;
     }
@@ -1144,7 +1142,7 @@ scavenge_one(StgPtr p)
 	
 	end = (StgPtr)((StgClosure *)p)->payload + info->layout.payload.ptrs;
 	for (q = (StgPtr)((StgClosure *)p)->payload; q < end; q++) {
-	    *q = (StgWord)(StgPtr)evacuate((StgClosure *)*q);
+	    evacuate((StgClosure **)q);
 	}
 	break;
     }
@@ -1155,7 +1153,7 @@ scavenge_one(StgPtr p)
 	rtsBool saved_eager_promotion = gct->eager_promotion;
 
 	gct->eager_promotion = rtsFalse;
-	((StgMutVar *)p)->var = evacuate(((StgMutVar *)p)->var);
+	evacuate(&((StgMutVar *)p)->var);
 	gct->eager_promotion = saved_eager_promotion;
 
 	if (gct->failed_to_evac) {
@@ -1175,7 +1173,7 @@ scavenge_one(StgPtr p)
     case THUNK_SELECTOR:
     { 
 	StgSelector *s = (StgSelector *)p;
-	s->selectee = evacuate(s->selectee);
+	evacuate(&s->selectee);
 	break;
     }
     
@@ -1183,7 +1181,7 @@ scavenge_one(StgPtr p)
     {
 	StgAP_STACK *ap = (StgAP_STACK *)p;
 
-	ap->fun = evacuate(ap->fun);
+	evacuate(&ap->fun);
 	scavenge_stack((StgPtr)ap->payload, (StgPtr)ap->payload + ap->size);
 	p = (StgPtr)ap->payload + ap->size;
 	break;
@@ -1216,7 +1214,7 @@ scavenge_one(StgPtr p)
 	q = p;
 	next = p + mut_arr_ptrs_sizeW((StgMutArrPtrs*)p);
 	for (p = (P_)((StgMutArrPtrs *)p)->payload; p < next; p++) {
-	    *p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+	    evacuate((StgClosure **)p);
 	}
 	gct->eager_promotion = saved_eager;
 
@@ -1238,7 +1236,7 @@ scavenge_one(StgPtr p)
       
 	next = p + mut_arr_ptrs_sizeW((StgMutArrPtrs*)p);
 	for (p = (P_)((StgMutArrPtrs *)p)->payload; p < next; p++) {
-	    *p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+	    evacuate((StgClosure **)p);
 	}
 
 	// If we're going to put this object on the mutable list, then
@@ -1274,9 +1272,9 @@ scavenge_one(StgPtr p)
       {
 	StgTVarWatchQueue *wq = ((StgTVarWatchQueue *) p);
 	gct->evac_gen = 0;
-        wq->closure = (StgClosure*)evacuate((StgClosure*)wq->closure);
-        wq->next_queue_entry = (StgTVarWatchQueue *)evacuate((StgClosure*)wq->next_queue_entry);
-        wq->prev_queue_entry = (StgTVarWatchQueue *)evacuate((StgClosure*)wq->prev_queue_entry);
+        evacuate((StgClosure **)&wq->closure);
+        evacuate((StgClosure **)&wq->next_queue_entry);
+        evacuate((StgClosure **)&wq->prev_queue_entry);
 	gct->evac_gen = saved_evac_gen;
 	gct->failed_to_evac = rtsTrue; // mutable
 	break;
@@ -1286,8 +1284,8 @@ scavenge_one(StgPtr p)
       {
 	StgTVar *tvar = ((StgTVar *) p);
 	gct->evac_gen = 0;
-	tvar->current_value = evacuate((StgClosure*)tvar->current_value);
-        tvar->first_watch_queue_entry = (StgTVarWatchQueue *)evacuate((StgClosure*)tvar->first_watch_queue_entry);
+	evacuate((StgClosure **)&tvar->current_value);
+        evacuate((StgClosure **)&tvar->first_watch_queue_entry);
 	gct->evac_gen = saved_evac_gen;
 	gct->failed_to_evac = rtsTrue; // mutable
 	break;
@@ -1297,9 +1295,9 @@ scavenge_one(StgPtr p)
       {
         StgTRecHeader *trec = ((StgTRecHeader *) p);
         gct->evac_gen = 0;
-	trec->enclosing_trec = (StgTRecHeader *)evacuate((StgClosure*)trec->enclosing_trec);
-	trec->current_chunk = (StgTRecChunk *)evacuate((StgClosure*)trec->current_chunk);
-        trec->invariants_to_check = (StgInvariantCheckQueue *)evacuate((StgClosure*)trec->invariants_to_check);
+	evacuate((StgClosure **)&trec->enclosing_trec);
+	evacuate((StgClosure **)&trec->current_chunk);
+        evacuate((StgClosure **)&trec->invariants_to_check);
 	gct->evac_gen = saved_evac_gen;
 	gct->failed_to_evac = rtsTrue; // mutable
         break;
@@ -1311,11 +1309,11 @@ scavenge_one(StgPtr p)
 	StgTRecChunk *tc = ((StgTRecChunk *) p);
 	TRecEntry *e = &(tc -> entries[0]);
 	gct->evac_gen = 0;
-	tc->prev_chunk = (StgTRecChunk *)evacuate((StgClosure*)tc->prev_chunk);
+	evacuate((StgClosure **)&tc->prev_chunk);
 	for (i = 0; i < tc -> next_entry_idx; i ++, e++ ) {
-	  e->tvar = (StgTVar *)evacuate((StgClosure*)e->tvar);
-	  e->expected_value = evacuate((StgClosure*)e->expected_value);
-	  e->new_value = evacuate((StgClosure*)e->new_value);
+	  evacuate((StgClosure **)&e->tvar);
+	  evacuate((StgClosure **)&e->expected_value);
+	  evacuate((StgClosure **)&e->new_value);
 	}
 	gct->evac_gen = saved_evac_gen;
 	gct->failed_to_evac = rtsTrue; // mutable
@@ -1326,8 +1324,8 @@ scavenge_one(StgPtr p)
     {
       StgAtomicInvariant *invariant = ((StgAtomicInvariant *) p);
       gct->evac_gen = 0;
-      invariant->code = (StgClosure *)evacuate(invariant->code);
-      invariant->last_execution = (StgTRecHeader *)evacuate((StgClosure*)invariant->last_execution);
+      evacuate(&invariant->code);
+      evacuate((StgClosure **)&invariant->last_execution);
       gct->evac_gen = saved_evac_gen;
       gct->failed_to_evac = rtsTrue; // mutable
       break;
@@ -1337,9 +1335,9 @@ scavenge_one(StgPtr p)
     {
       StgInvariantCheckQueue *queue = ((StgInvariantCheckQueue *) p);
       gct->evac_gen = 0;
-      queue->invariant = (StgAtomicInvariant *)evacuate((StgClosure*)queue->invariant);
-      queue->my_execution = (StgTRecHeader *)evacuate((StgClosure*)queue->my_execution);
-      queue->next_queue_entry = (StgInvariantCheckQueue *)evacuate((StgClosure*)queue->next_queue_entry);
+      evacuate((StgClosure **)&queue->invariant);
+      evacuate((StgClosure **)&queue->my_execution);
+      evacuate((StgClosure **)&queue->next_queue_entry);
       gct->evac_gen = saved_evac_gen;
       gct->failed_to_evac = rtsTrue; // mutable
       break;
@@ -1360,7 +1358,7 @@ scavenge_one(StgPtr p)
 	if (HEAP_ALLOCED(q) && Bdescr((StgPtr)q)->flags & BF_EVACUATED) {
 	    break;
 	}
-	((StgInd *)p)->indirectee = evacuate(q);
+	evacuate(&((StgInd *)p)->indirectee);
     }
 
 #if 0 && defined(DEBUG)
@@ -1460,7 +1458,7 @@ scavenge_mutable_list(generation *gen)
 		    // we don't want to have to mark a TSO dirty just
 		    // because we put it on a different queue.
 		    if (tso->why_blocked != BlockedOnBlackHole) {
-			tso->link = (StgTSO *)evacuate((StgClosure *)tso->link);
+			evacuate((StgClosure **)&tso->link);
 		    }
 		    recordMutableGen_GC((StgClosure *)p,gen);
 		    continue;
@@ -1540,7 +1538,7 @@ scavenge_static(void)
     case IND_STATIC:
       {
 	StgInd *ind = (StgInd *)p;
-	ind->indirectee = evacuate(ind->indirectee);
+	evacuate(&ind->indirectee);
 
 	/* might fail to evacuate it, in which case we have to pop it
 	 * back on the mutable list of the oldest generation.  We
@@ -1569,7 +1567,7 @@ scavenge_static(void)
 	next = (P_)p->payload + info->layout.payload.ptrs;
 	// evacuate the pointers 
 	for (q = (P_)p->payload; q < next; q++) {
-	    *q = (StgWord)(StgPtr)evacuate((StgClosure *)*q);
+	    evacuate((StgClosure **)q);
 	}
 	break;
       }
@@ -1596,7 +1594,7 @@ scavenge_large_bitmap( StgPtr p, StgLargeBitmap *large_bitmap, nat size )
     bitmap = large_bitmap->bitmap[b];
     for (i = 0; i < size; ) {
 	if ((bitmap & 1) == 0) {
-	    *p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+	    evacuate((StgClosure **)p);
 	}
 	i++;
 	p++;
@@ -1614,7 +1612,7 @@ scavenge_small_bitmap (StgPtr p, nat size, StgWord bitmap)
 {
     while (size > 0) {
 	if ((bitmap & 1) == 0) {
-	    *p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+	    evacuate((StgClosure **)p);
 	}
 	p++;
 	bitmap = bitmap >> 1;
@@ -1679,8 +1677,7 @@ scavenge_stack(StgPtr p, StgPtr stack_end)
 	    ((StgUpdateFrame *)p)->updatee->header.info = 
 		(StgInfoTable *)&stg_IND_OLDGEN_PERM_info;
         }            
-	((StgUpdateFrame *)p)->updatee 
-	    = evacuate(((StgUpdateFrame *)p)->updatee);
+	evacuate(&((StgUpdateFrame *)p)->updatee);
 	p += sizeofW(StgUpdateFrame);
 	continue;
     }
@@ -1709,7 +1706,7 @@ scavenge_stack(StgPtr p, StgPtr stack_end)
 	nat size;
 
 	p++;
-	*p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+	evacuate((StgClosure **)p);
 	bco = (StgBCO *)*p;
 	p++;
 	size = BCO_BITMAP_SIZE(bco);
@@ -1751,7 +1748,7 @@ scavenge_stack(StgPtr p, StgPtr stack_end)
 	
 	// follow the ptr words
 	for (size = RET_DYN_PTRS(dyn); size > 0; size--) {
-	    *p = (StgWord)(StgPtr)evacuate((StgClosure *)*p);
+	    evacuate((StgClosure **)p);
 	    p++;
 	}
 	continue;
@@ -1762,7 +1759,7 @@ scavenge_stack(StgPtr p, StgPtr stack_end)
 	StgRetFun *ret_fun = (StgRetFun *)p;
 	StgFunInfoTable *fun_info;
 
-	ret_fun->fun = evacuate(ret_fun->fun);
+	evacuate(&ret_fun->fun);
  	fun_info = get_fun_itbl(UNTAG_CLOSURE(ret_fun->fun));
 	p = scavenge_arg_block(fun_info, ret_fun->payload);
 	goto follow_srt;