X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=rts%2Fsm%2FGCUtils.c;h=39a79f6b9f7439bbe0a39da39e941d38ea300ea3;hb=f6013eedb4dea47afac8167dfa08561ae90454db;hp=465954fac44901047f93f2070953547984969e95;hpb=f86e7206ea94b48b94fb61007a1c5d55b8c60f45;p=ghc-hetmet.git

diff --git a/rts/sm/GCUtils.c b/rts/sm/GCUtils.c
index 465954f..39a79f6 100644
--- a/rts/sm/GCUtils.c
+++ b/rts/sm/GCUtils.c
@@ -1,6 +1,6 @@
 /* -----------------------------------------------------------------------------
  *
- * (c) The GHC Team 1998-2006
+ * (c) The GHC Team 1998-2008
  *
  * Generational garbage collector: utilities
  *
@@ -11,14 +11,19 @@
  *
  * ---------------------------------------------------------------------------*/
 
+#include "PosixSource.h"
 #include "Rts.h"
-#include "RtsFlags.h"
+
+#include "BlockAlloc.h"
 #include "Storage.h"
 #include "GC.h"
 #include "GCThread.h"
 #include "GCUtils.h"
 #include "Printer.h"
 #include "Trace.h"
+#ifdef THREADED_RTS
+#include "WSDeque.h"
+#endif
 
 #ifdef THREADED_RTS
 SpinLock gc_alloc_block_sync;
@@ -34,7 +39,18 @@ allocBlock_sync(void)
     return bd;
 }
 
+static bdescr *
+allocGroup_sync(nat n)
+{
+    bdescr *bd;
+    ACQUIRE_SPIN_LOCK(&gc_alloc_block_sync);
+    bd = allocGroup(n);
+    RELEASE_SPIN_LOCK(&gc_alloc_block_sync);
+    return bd;
+}
 
+
+#if 0
 static void
 allocBlocks_sync(nat n, bdescr **hd, bdescr **tl, 
                  nat gen_no, step *stp,
@@ -56,6 +72,7 @@ allocBlocks_sync(nat n, bdescr **hd, bdescr **tl,
     *tl = &bd[n-1];
     RELEASE_SPIN_LOCK(&gc_alloc_block_sync);
 }
+#endif
 
 void
 freeChain_sync(bdescr *bd)
@@ -70,35 +87,50 @@ freeChain_sync(bdescr *bd)
    -------------------------------------------------------------------------- */
 
 bdescr *
-grab_todo_block (step_workspace *ws)
+grab_local_todo_block (step_workspace *ws)
 {
     bdescr *bd;
     step *stp;
 
     stp = ws->step;
-    bd = NULL;
 
-    if (ws->buffer_todo_bd)
+    bd = ws->todo_overflow;
+    if (bd != NULL)
+    {
+        ws->todo_overflow = bd->link;
+        bd->link = NULL;
+        ws->n_todo_overflow--;
+	return bd;
+    }
+
+    bd = popWSDeque(ws->todo_q);
+    if (bd != NULL)
     {
-	bd = ws->buffer_todo_bd;
 	ASSERT(bd->link == NULL);
-	ws->buffer_todo_bd = NULL;
 	return bd;
     }
 
-    ACQUIRE_SPIN_LOCK(&stp->sync_todo);
-    if (stp->todos) {
-	bd = stp->todos;
-        if (stp->todos == stp->todos_last) {
-            stp->todos_last = NULL;
+    return NULL;
+}
+
+#if defined(THREADED_RTS)
+bdescr *
+steal_todo_block (nat s)
+{
+    nat n;
+    bdescr *bd;
+
+    // look for work to steal
+    for (n = 0; n < n_gc_threads; n++) {
+        if (n == gct->thread_index) continue;
+        bd = stealWSDeque(gc_threads[n]->steps[s].todo_q);
+        if (bd) {
+            return bd;
         }
-	stp->todos = bd->link;
-        stp->n_todos--;
-	bd->link = NULL;
-    }	
-    RELEASE_SPIN_LOCK(&stp->sync_todo);
-    return bd;
+    }
+    return NULL;
 }
+#endif
 
 void
 push_scanned_block (bdescr *bd, step_workspace *ws)
@@ -108,12 +140,12 @@ push_scanned_block (bdescr *bd, step_workspace *ws)
     ASSERT(bd->step == ws->step);
     ASSERT(bd->u.scan == bd->free);
 
-    if (bd->start + BLOCK_SIZE_W - bd->free > WORK_UNIT_WORDS)
+    if (bd->start + bd->blocks * BLOCK_SIZE_W - bd->free > WORK_UNIT_WORDS)
     {
         // a partially full block: put it on the part_list list.
         bd->link = ws->part_list;
         ws->part_list = bd;
-        ws->n_part_blocks++;
+        ws->n_part_blocks += bd->blocks;
         IF_DEBUG(sanity, 
                  ASSERT(countBlocks(ws->part_list) == ws->n_part_blocks));
     }
@@ -122,7 +154,7 @@ push_scanned_block (bdescr *bd, step_workspace *ws)
         // put the scan block on the ws->scavd_list.
         bd->link = ws->scavd_list;
         ws->scavd_list = bd;
-        ws->n_scavd_blocks ++;
+        ws->n_scavd_blocks += bd->blocks;
         IF_DEBUG(sanity, 
                  ASSERT(countBlocks(ws->scavd_list) == ws->n_scavd_blocks));
     }
@@ -131,8 +163,13 @@ push_scanned_block (bdescr *bd, step_workspace *ws)
 StgPtr
 todo_block_full (nat size, step_workspace *ws)
 {
+    StgPtr p;
     bdescr *bd;
 
+    // todo_free has been pre-incremented by Evac.c:alloc_for_copy().  We
+    // are expected to leave it bumped when we've finished here.
+    ws->todo_free -= size;
+
     bd = ws->todo_bd;
 
     ASSERT(bd != NULL);
@@ -143,13 +180,15 @@ todo_block_full (nat size, step_workspace *ws)
     // this block to push, and there's enough room in
     // this block to evacuate the current object, then just increase
     // the limit.
-    if (ws->step->todos != NULL || 
+    if (!looksEmptyWSDeque(ws->todo_q) || 
         (ws->todo_free - bd->u.scan < WORK_UNIT_WORDS / 2)) {
-        if (ws->todo_free + size < bd->start + BLOCK_SIZE_W) {
-            ws->todo_lim = stg_min(bd->start + BLOCK_SIZE_W,
+        if (ws->todo_free + size < bd->start + bd->blocks * BLOCK_SIZE_W) {
+            ws->todo_lim = stg_min(bd->start + bd->blocks * BLOCK_SIZE_W,
                                    ws->todo_lim + stg_max(WORK_UNIT_WORDS,size));
             debugTrace(DEBUG_gc, "increasing limit for %p to %p", bd->start, ws->todo_lim);
-            return ws->todo_free;
+            p = ws->todo_free;
+            ws->todo_free += size;
+            return p;
         }
     }
     
@@ -176,19 +215,15 @@ todo_block_full (nat size, step_workspace *ws)
         {
             step *stp;
             stp = ws->step;
-            trace(TRACE_gc|DEBUG_gc, "push todo block %p (%d words), step %d, n_todos: %d", 
-                  bd->start, bd->free - bd->u.scan, stp->abs_no, stp->n_todos);
-            // ToDo: use buffer_todo
-            ACQUIRE_SPIN_LOCK(&stp->sync_todo);
-            if (stp->todos_last == NULL) {
-                stp->todos_last = bd;
-                stp->todos = bd;
-            } else {
-                stp->todos_last->link = bd;
-                stp->todos_last = bd;
+            debugTrace(DEBUG_gc, "push todo block %p (%ld words), step %d, todo_q: %ld", 
+                  bd->start, (unsigned long)(bd->free - bd->u.scan),
+                  stp->abs_no, dequeElements(ws->todo_q));
+
+            if (!pushWSDeque(ws->todo_q, bd)) {
+                bd->link = ws->todo_overflow;
+                ws->todo_overflow = bd;
+                ws->n_todo_overflow++;
             }
-            stp->n_todos++;
-            RELEASE_SPIN_LOCK(&stp->sync_todo);
         }
     }
 
@@ -198,47 +233,53 @@ todo_block_full (nat size, step_workspace *ws)
 
     alloc_todo_block(ws, size);
 
-    return ws->todo_free;
+    p = ws->todo_free;
+    ws->todo_free += size;
+    return p;
 }
 
 StgPtr
 alloc_todo_block (step_workspace *ws, nat size)
 {
-    bdescr *bd, *hd, *tl;
-    StgWord32 flags;
+    bdescr *bd/*, *hd, *tl */;
 
     // Grab a part block if we have one, and it has enough room
-    if (ws->part_list != NULL && 
-        ws->part_list->start + BLOCK_SIZE_W - ws->part_list->free > (int)size)
+    bd = ws->part_list;
+    if (bd != NULL &&
+        bd->start + bd->blocks * BLOCK_SIZE_W - bd->free > (int)size)
     {
-        bd = ws->part_list;
         ws->part_list = bd->link;
-        ws->n_part_blocks--;
+        ws->n_part_blocks -= bd->blocks;
     }
     else
     {
-        // blocks in to-space in generations up to and including N
-        // get the BF_EVACUATED flag.
-        if (ws->step->gen_no <= N) {
-            flags = BF_EVACUATED;
+        // blocks in to-space get the BF_EVACUATED flag.
+
+//        allocBlocks_sync(16, &hd, &tl, 
+//                         ws->step->gen_no, ws->step, BF_EVACUATED);
+//
+//        tl->link = ws->part_list;
+//        ws->part_list = hd->link;
+//        ws->n_part_blocks += 15;
+//
+//        bd = hd;
+
+        if (size > BLOCK_SIZE_W) {
+            bd = allocGroup_sync((lnat)BLOCK_ROUND_UP(size*sizeof(W_))
+                                 / BLOCK_SIZE);
         } else {
-            flags = 0;
+            bd = allocBlock_sync();
         }
-        allocBlocks_sync(4, &hd, &tl, 
-                         ws->step->gen_no, ws->step, flags);
-
-        tl->link = ws->part_list;
-        ws->part_list = hd->link;
-        ws->n_part_blocks += 3;
-
-        bd = hd;
+        initBdescr(bd, ws->step);
+        bd->flags = BF_EVACUATED;
+        bd->u.scan = bd->free = bd->start;
     }
 
     bd->link = NULL;
 
     ws->todo_bd = bd;
     ws->todo_free = bd->free;
-    ws->todo_lim  = stg_min(bd->start + BLOCK_SIZE_W,
+    ws->todo_lim  = stg_min(bd->start + bd->blocks * BLOCK_SIZE_W,
                             bd->free + stg_max(WORK_UNIT_WORDS,size));
 
     debugTrace(DEBUG_gc, "alloc new todo block %p for step %d",