Fix the nursery size calculation when -N is used with -H

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

diff --git a/rts/sm/GC.c b/rts/sm/GC.c
index 7488006..97f391c 100644 (file)
--- a/rts/sm/GC.c
+++ b/rts/sm/GC.c
@@ -11,35 +11,32 @@
  *
  * ---------------------------------------------------------------------------*/
 
-// #include "PosixSource.h"
+#include "PosixSource.h"
 #include "Rts.h"
-#include "RtsFlags.h"
+#include "HsFFI.h"
+
+#include "Storage.h"
 #include "RtsUtils.h"
 #include "Apply.h"
-#include "OSThreads.h"
-#include "LdvProfile.h"
 #include "Updates.h"
 #include "Stats.h"
 #include "Schedule.h"
 #include "Sanity.h"
 #include "BlockAlloc.h"
-#include "MBlock.h"
 #include "ProfHeap.h"
-#include "SchedAPI.h"
 #include "Weak.h"
 #include "Prelude.h"
-#include "ParTicky.h"          // ToDo: move into Rts.h
 #include "RtsSignals.h"
 #include "STM.h"
-#include "HsFFI.h"
-#include "Linker.h"
 #if defined(RTS_GTK_FRONTPANEL)
 #include "FrontPanel.h"
 #endif
 #include "Trace.h"
 #include "RetainerProfile.h"
+#include "LdvProfile.h"
 #include "RaiseAsync.h"
 #include "Papi.h"
+#include "Stable.h"
 
 #include "GC.h"
 #include "GCThread.h"
@@ -240,7 +237,8 @@ GarbageCollect (rtsBool force_major_gc,
   n = initialise_N(force_major_gc);
 
 #if defined(THREADED_RTS)
-  work_stealing = RtsFlags.ParFlags.parGcLoadBalancing;
+  work_stealing = RtsFlags.ParFlags.parGcLoadBalancingEnabled &&
+                  N >= RtsFlags.ParFlags.parGcLoadBalancingGen;
       // It's not always a good idea to do load balancing in parallel
       // GC.  In particular, for a parallel program we don't want to
       // lose locality by moving cached data into another CPU's cache
@@ -281,7 +279,7 @@ GarbageCollect (rtsBool force_major_gc,
 
 #ifdef DEBUG
   // check for memory leaks if DEBUG is on 
-  memInventory(traceClass(DEBUG_gc));
+  memInventory(DEBUG_gc);
 #endif
 
   // check stack sanity *before* GC
@@ -784,7 +782,7 @@ SET_GCT(gc_threads[0]);
 
 #ifdef DEBUG
   // check for memory leaks if DEBUG is on 
-  memInventory(traceClass(DEBUG_gc));
+  memInventory(DEBUG_gc);
 #endif
 
 #ifdef RTS_GTK_FRONTPANEL
@@ -937,6 +935,23 @@ initGcThreads (void)
     }
 }
 
+void
+freeGcThreads (void)
+{
+    if (gc_threads != NULL) {
+#if defined(THREADED_RTS)
+        nat i;
+       for (i = 0; i < RtsFlags.ParFlags.nNodes; i++) {
+            stgFree (gc_threads[i]);
+       }
+        stgFree (gc_threads);
+#else
+        stgFree (gc_threads);
+#endif
+        gc_threads = NULL;
+    }
+}
+
 /* ----------------------------------------------------------------------------
    Start GC threads
    ------------------------------------------------------------------------- */
@@ -1052,6 +1067,11 @@ loop:
 void
 gcWorkerThread (Capability *cap)
 {
+    gc_thread *saved_gct;
+
+    // necessary if we stole a callee-saves register for gct:
+    saved_gct = gct;
+
     cap->in_gc = rtsTrue;
 
     gct = gc_threads[cap->no];
@@ -1091,14 +1111,17 @@ gcWorkerThread (Capability *cap)
                gct->thread_index);
     ACQUIRE_SPIN_LOCK(&gct->mut_spin);
     debugTrace(DEBUG_gc, "GC thread %d on my way...", gct->thread_index);
+
+    SET_GCT(saved_gct);
 }
 
 #endif
 
+#if defined(THREADED_RTS)
+
 void
 waitForGcThreads (Capability *cap USED_IF_THREADS)
 {
-#if defined(THREADED_RTS)
     nat n_threads = RtsFlags.ParFlags.nNodes;
     nat me = cap->no;
     nat i, j;
@@ -1124,9 +1147,10 @@ waitForGcThreads (Capability *cap USED_IF_THREADS)
             if (!retry) break;
         }
     }
-#endif
 }
 
+#endif // THREADED_RTS
+
 static void
 start_gc_threads (void)
 {
@@ -1168,10 +1192,10 @@ shutdown_gc_threads (nat n_threads USED_IF_THREADS, nat me USED_IF_THREADS)
 #endif
 }
 
+#if defined(THREADED_RTS)
 void
 releaseGCThreads (Capability *cap USED_IF_THREADS)
 {
-#if defined(THREADED_RTS)
     nat n_threads = RtsFlags.ParFlags.nNodes;
     nat me = cap->no;
     nat i;
@@ -1184,8 +1208,8 @@ releaseGCThreads (Capability *cap USED_IF_THREADS)
         ACQUIRE_SPIN_LOCK(&gc_threads[i]->gc_spin);
         RELEASE_SPIN_LOCK(&gc_threads[i]->mut_spin);
     }
-#endif
 }
+#endif
 
 /* ----------------------------------------------------------------------------
    Initialise a generation that is to be collected 
@@ -1599,6 +1623,8 @@ resize_generations (void)
 static void
 resize_nursery (void)
 {
+    lnat min_nursery = RtsFlags.GcFlags.minAllocAreaSize * n_capabilities;
+
     if (RtsFlags.GcFlags.generations == 1)
     {   // Two-space collector:
        nat blocks;
@@ -1641,9 +1667,9 @@ resize_nursery (void)
        else
        {
            blocks *= RtsFlags.GcFlags.oldGenFactor;
-           if (blocks < RtsFlags.GcFlags.minAllocAreaSize)
+           if (blocks < min_nursery)
            {
-               blocks = RtsFlags.GcFlags.minAllocAreaSize;
+               blocks = min_nursery;
            }
        }
        resizeNurseries(blocks);
@@ -1690,8 +1716,8 @@ resize_nursery (void)
                (((long)RtsFlags.GcFlags.heapSizeSuggestion - (long)needed) * 100) /
                (100 + (long)g0s0_pcnt_kept);
            
-           if (blocks < (long)RtsFlags.GcFlags.minAllocAreaSize) {
-               blocks = RtsFlags.GcFlags.minAllocAreaSize;
+           if (blocks < (long)min_nursery) {
+               blocks = min_nursery;
            }
            
            resizeNurseries((nat)blocks);