Use work-stealing for load-balancing in the GC

[ghc-hetmet.git] / rts / sm / Storage.c

diff --git a/rts/sm/Storage.c b/rts/sm/Storage.c
index 6fa90cf..cfe649a 100644 (file)
--- a/rts/sm/Storage.c
+++ b/rts/sm/Storage.c
@@ -104,7 +104,6 @@ initStep (step *stp, int g, int s)
     stp->compact = 0;
     stp->bitmap = NULL;
 #ifdef THREADED_RTS
-    initSpinLock(&stp->sync_todo);
     initSpinLock(&stp->sync_large_objects);
 #endif
     stp->threads = END_TSO_QUEUE;
@@ -272,7 +271,6 @@ initStorage( void )
 
 #ifdef THREADED_RTS
   initSpinLock(&gc_alloc_block_sync);
-  initSpinLock(&recordMutableGen_sync);
   whitehole_spin = 0;
 #endif
 
@@ -376,7 +374,7 @@ newCAF(StgClosure* caf)
     * any more and can use it as a STATIC_LINK.
     */
     ((StgIndStatic *)caf)->saved_info = NULL;
-    recordMutableGen(caf, oldest_gen);
+    recordMutableGen(caf, oldest_gen->no);
   }
   
   RELEASE_SM_LOCK;
@@ -809,7 +807,9 @@ allocatePinned( lnat n )
     // If the request is for a large object, then allocate()
     // will give us a pinned object anyway.
     if (n >= LARGE_OBJECT_THRESHOLD/sizeof(W_)) {
-       return allocate(n);
+       p = allocate(n);
+        Bdescr(p)->flags |= BF_PINNED;
+        return p;
     }
 
     ACQUIRE_SM_LOCK;
@@ -817,13 +817,6 @@ allocatePinned( lnat n )
     TICK_ALLOC_HEAP_NOCTR(n);
     CCS_ALLOC(CCCS,n);
 
-    // we always return 8-byte aligned memory.  bd->free must be
-    // 8-byte aligned to begin with, so we just round up n to
-    // the nearest multiple of 8 bytes.
-    if (sizeof(StgWord) == 4) {
-       n = (n+1) & ~1;
-    }
-
     // If we don't have a block of pinned objects yet, or the current
     // one isn't large enough to hold the new object, allocate a new one.
     if (bd == NULL || (bd->free + n) > (bd->start + BLOCK_SIZE_W)) {
@@ -1472,7 +1465,7 @@ checkSanity( void )
 
     if (RtsFlags.GcFlags.generations == 1) {
        checkHeap(g0s0->blocks);
-       checkChain(g0s0->large_objects);
+       checkLargeObjects(g0s0->large_objects);
     } else {
        
        for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
@@ -1483,10 +1476,7 @@ checkSanity( void )
                ASSERT(countBlocks(generations[g].steps[s].large_objects)
                       == generations[g].steps[s].n_large_blocks);
                checkHeap(generations[g].steps[s].blocks);
-               checkChain(generations[g].steps[s].large_objects);
-               if (g > 0) {
-                   checkMutableList(generations[g].mut_list, g);
-               }
+               checkLargeObjects(generations[g].steps[s].large_objects);
            }
        }
 
@@ -1503,9 +1493,9 @@ checkSanity( void )
 #if defined(THREADED_RTS)
     // check the stacks too in threaded mode, because we don't do a
     // full heap sanity check in this case (see checkHeap())
-    checkGlobalTSOList(rtsTrue);
+    checkMutableLists(rtsTrue);
 #else
-    checkGlobalTSOList(rtsFalse);
+    checkMutableLists(rtsFalse);
 #endif
 }