diff --git a/harbour/ChangeLog b/harbour/ChangeLog
index 4786224162..2d26c8538f 100644
--- a/harbour/ChangeLog
+++ b/harbour/ChangeLog
@@ -8,6 +8,32 @@
    2009-12-31 13:59 UTC+0100 Foo Bar (foo.bar foobar.org)
 */
 
+2009-03-29 19:20 UTC+0200 Przemyslaw Czerpak (druzus/at/priv.onet.pl)
+  * harbour/include/hbapi.h
+  * harbour/include/hbstack.h
+  * harbour/source/vm/estack.c
+  * harbour/source/vm/fm.c
+    + added support for thread local memory allocators by using MSPACE
+      mechanism in DLALLOC.
+      It can be enabled by compiling Harbour with HB_FM_DLMT_ALLOC macro.
+      By default up to 16 (HB_MSPACE_COUNT macro) independent memory
+      allocators is dynamically created for new threads. When more threads
+      is used then allocators are shared between them and HVM tries to
+      balance used allocators to reduce number of threads using the
+      same allocator.
+      It should give noticeable scalability improvement in some OS-es
+      where default memory manages has pure MT performance.
+      Please try to compiler speedtst.prg with Harbour compiled with
+      and without HB_FM_DLMT_ALLOC and compare results in this test:
+         speedtst --thread=2 --scale
+      using real multiCPU machine.
+
+  * harbour/source/vm/dlmalloc.c
+    * pacified warning when FOOTERS are enabled
+
+  * harbour/source/vm/garbage.c
+    * cleanup comment
+
 2009-03-29 09:03 UTC-0800 Pritpal Bedi (pritpal@vouchcac.com)
   * harbour/contrib/gtwvg/wvgsink.c
     ! Fixed remaining non-ANSI C comments. Just had been overlooked!
diff --git a/harbour/include/hbapi.h b/harbour/include/hbapi.h
index 4251629f53..241d18956d 100644
--- a/harbour/include/hbapi.h
+++ b/harbour/include/hbapi.h
@@ -493,6 +493,10 @@ extern HB_EXPORT ULONG  hb_xsize( void * pMem );                    /* returns t
 extern HB_EXPORT ULONG  hb_xquery( USHORT uiMode );                 /* Query different types of memory information */
 extern HB_EXPORT void   hb_xsetfilename( char * szValue );
 extern HB_EXPORT void   hb_xsetinfo( char * szValue );
+#ifdef _HB_API_INTERNAL_
+extern void hb_xinit_thread( void );
+extern void hb_xexit_thread( void );
+#endif
 
 extern HB_EXPORT HB_VMHANDLE hb_xvalloc( ULONG nSize, USHORT nFlags );
 extern HB_EXPORT void        hb_xvfree( HB_VMHANDLE h );
diff --git a/harbour/include/hbstack.h b/harbour/include/hbstack.h
index 80e60c3606..d727807780 100644
--- a/harbour/include/hbstack.h
+++ b/harbour/include/hbstack.h
@@ -184,6 +184,7 @@ typedef struct
    void *     pStackLst;      /* this stack entry in stack linked list */
    HB_IOERRORS IOErrors;      /* MT safe buffer for IO errors */
    BYTE *     byDirBuffer;    /* MT safe buffer for hb_fsCurDir() results */
+   void *     allocator;      /* memory manager global struct pointer */
 #endif
 } HB_STACK, * PHB_STACK;
 
diff --git a/harbour/source/vm/dlmalloc.c b/harbour/source/vm/dlmalloc.c
index 09338558e8..e7653e7e4b 100644
--- a/harbour/source/vm/dlmalloc.c
+++ b/harbour/source/vm/dlmalloc.c
@@ -4566,6 +4566,7 @@ void mspace_free(mspace msp, void* mem) {
     mchunkptr p  = mem2chunk(mem);
 #if FOOTERS
     mstate fm = get_mstate_for(p);
+    ( void ) msp;
 #else /* FOOTERS */
     mstate fm = (mstate)msp;
 #endif /* FOOTERS */
diff --git a/harbour/source/vm/estack.c b/harbour/source/vm/estack.c
index 318fb820e2..4d2f446836 100644
--- a/harbour/source/vm/estack.c
+++ b/harbour/source/vm/estack.c
@@ -351,6 +351,7 @@ void hb_stackInit( void )
    {
       HB_STACK_TLS_PRELOAD
       hb_stack_init( &hb_stack );
+      hb_xinit_thread();
    }
 }
 
@@ -361,7 +362,7 @@ void hb_stackFree( void )
    HB_TRACE(HB_TR_DEBUG, ("hb_stackFree()"));
 
    hb_stack_free( &hb_stack );
-
+   hb_xexit_thread();
 #if defined( HB_MT_VM )
    hb_stack_dealloc();
 #endif
diff --git a/harbour/source/vm/fm.c b/harbour/source/vm/fm.c
index 460172c604..cf137cfdc9 100644
--- a/harbour/source/vm/fm.c
+++ b/harbour/source/vm/fm.c
@@ -83,6 +83,8 @@
    #include <malloc.h>
 */
 
+#define HB_STACK_PRELOAD
+
 #include "hbvmopt.h"
 #include "hbapi.h"
 #include "hbapiitm.h"
@@ -101,7 +103,8 @@
    #undef HB_FM_DL_ALLOC
    #undef HB_FM_WIN_ALLOC
 #elif !defined( HB_FM_DL_ALLOC ) && !defined( HB_FM_WIN_ALLOC )
-   #if defined( _MSC_VER ) || defined( __BORLANDC__ ) || defined( __MINGW32__ )
+   #if defined( _MSC_VER ) || defined( __BORLANDC__ ) || defined( __MINGW32__ ) || \
+       ( defined( HB_FM_DLMT_ALLOC ) && defined( HB_MT_VM ) )
       #define HB_FM_DL_ALLOC
    #else
       /* #define HB_FM_DL_ALLOC */
@@ -124,6 +127,12 @@
 #  define REALLOC_ZERO_BYTES_FREES
 #  if defined( HB_MT_VM )
 #     define USE_LOCKS  1
+#     if defined( HB_FM_DLMT_ALLOC )
+#        define ONLY_MSPACES  1
+#        define FOOTERS       1
+#     endif
+#  else
+#     undef HB_FM_DLMT_ALLOC
 #  endif
 #  if defined( __BORLANDC__ )
 #     pragma warn -aus
@@ -136,7 +145,7 @@
 #     define ABORT TerminateProcess( GetCurrentProcess(), 0 )
 #  elif defined( __POCC__ ) && !defined( InterlockedCompareExchangePointer )
 #     define InterlockedCompareExchangePointer
-#  elif defined( _MSC_VER ) && !defined( USE_DL_PREFIX )
+#  elif defined( _MSC_VER ) && !defined( USE_DL_PREFIX ) && !defined( HB_FM_DLMT_ALLOC )
 #     define USE_DL_PREFIX
 #  endif
 #  include "dlmalloc.c"
@@ -148,27 +157,35 @@
 #     pragma warn +prc
 #     pragma warn +rch
 #  endif
-#  if defined( USE_DL_PREFIX )
+#  if defined( HB_FM_DLMT_ALLOC )
+#     define malloc( n )         mspace_malloc( hb_mspace(), ( n ) )
+#     define realloc( p, n )     mspace_realloc( NULL, ( p ), ( n ) )
+#     define free( p )           mspace_free( NULL, ( p ) )
+#  elif defined( USE_DL_PREFIX )
 #     define malloc( n )         dlmalloc( ( n ) )
 #     define realloc( p, n )     dlrealloc( ( p ), ( n ) )
 #     define free( p )           dlfree( ( p ) )
 #  endif
-#elif defined( HB_FM_WIN_ALLOC ) && defined( HB_OS_WIN )
-#  if defined( HB_FM_LOCALALLOC )
-#     define malloc( n )      ( void * ) LocalAlloc( LMEM_FIXED, ( n ) )
-#     define realloc( p, n )  ( void * ) LocalReAlloc( ( HLOCAL ) ( p ), ( n ), LMEM_MOVEABLE )
-#     define free( p )        LocalFree( ( HLOCAL ) ( p ) )
-#  else
-      static HANDLE s_hProcessHeap = NULL;
-#     define HB_FM_NEED_INIT
-#     define HB_FM_HEAP_INIT
-#     define malloc( n )      ( void * ) HeapAlloc( s_hProcessHeap, 0, ( n ) )
-#     define realloc( p, n )  ( void * ) HeapReAlloc( s_hProcessHeap, 0, ( void * ) ( p ), ( n ) )
-#     define free( p )        HeapFree( s_hProcessHeap, 0, ( void * ) ( p ) )
+#else
+#  undef HB_FM_DLMT_ALLOC
+#  if defined( HB_FM_WIN_ALLOC ) && defined( HB_OS_WIN )
+#     if defined( HB_FM_LOCALALLOC )
+#        define malloc( n )      ( void * ) LocalAlloc( LMEM_FIXED, ( n ) )
+#        define realloc( p, n )  ( void * ) LocalReAlloc( ( HLOCAL ) ( p ), ( n ), LMEM_MOVEABLE )
+#        define free( p )        LocalFree( ( HLOCAL ) ( p ) )
+#     else
+         static HANDLE s_hProcessHeap = NULL;
+#        define HB_FM_NEED_INIT
+#        define HB_FM_HEAP_INIT
+#        define malloc( n )      ( void * ) HeapAlloc( s_hProcessHeap, 0, ( n ) )
+#        define realloc( p, n )  ( void * ) HeapReAlloc( s_hProcessHeap, 0, ( void * ) ( p ), ( n ) )
+#        define free( p )        HeapFree( s_hProcessHeap, 0, ( void * ) ( p ) )
+#     endif
 #  endif
 #endif
 
 #if defined( HB_MT_VM ) && ( defined( HB_FM_STATISTICS ) || \
+    defined( HB_FM_DLMT_ALLOC ) || \
     !defined( HB_ATOM_INC ) || !defined( HB_ATOM_DEC ) )
 
    static HB_CRITICAL_NEW( s_fmMtx );
@@ -307,6 +324,116 @@ typedef void * PHB_MEMINFO;
 #endif
 
 
+#if defined( HB_FM_DLMT_ALLOC )
+
+#  if !defined( HB_MSPACE_COUNT )
+#     define HB_MSPACE_COUNT  16
+#  endif
+
+typedef struct
+{
+   int      count;
+   mspace * ms;
+} HB_MSPACE, * PHB_MSPACE;
+
+static mspace s_gm = NULL;
+static HB_MSPACE s_mspool[ HB_MSPACE_COUNT ];
+
+static mspace hb_mspace( void )
+{
+   HB_STACK_TLS_PRELOAD
+
+   if( hb_stackId() && hb_stack.allocator )
+      return ( ( PHB_MSPACE ) hb_stack.allocator )->ms;
+
+   if( !s_gm )
+      s_gm = create_mspace( 0, 1 );
+
+   return s_gm;
+}
+
+static void hb_mspace_cleanup( void )
+{
+   int i;
+
+   s_gm = NULL;
+   for( i = 0; i < HB_MSPACE_COUNT; ++i )
+   {
+      if( s_mspool[ i ].ms )
+      {
+         destroy_mspace( s_mspool[ i ].ms );
+         s_mspool[ i ].ms = NULL;
+         s_mspool[ i ].count = 0;
+      }
+   }
+}
+
+#elif defined( HB_FM_DL_ALLOC ) && defined( USE_DL_PREFIX )
+
+static void dlmalloc_destroy( void )
+{
+   if( ok_magic(gm) )
+   {
+      msegmentptr sp = &gm->seg;
+      while(sp != 0 )
+      {
+         char* base = sp->base;
+         size_t size = sp->size;
+         flag_t flag = sp->sflags;
+         sp = sp->next;
+         if( (flag & IS_MMAPPED_BIT) && !(flag & EXTERN_BIT) )
+            CALL_MUNMAP(base, size);
+      }
+   }
+}
+
+#endif
+
+void hb_xinit_thread( void )
+{
+#if defined( HB_FM_DLMT_ALLOC )
+   HB_STACK_TLS_PRELOAD
+
+   if( hb_stack.allocator == NULL )
+   {
+      HB_FM_LOCK
+      if( s_mspool[ 0 ].ms == NULL && s_gm )
+      {
+         s_mspool[ 0 ].count = 1;
+         s_mspool[ 0 ].ms = s_gm;
+         hb_stack.allocator = ( void * ) &s_mspool[ 0 ];
+      }
+      else
+      {
+         int i, imin = 0;
+         for( i = 1; i < HB_MSPACE_COUNT; ++i )
+         {
+            if( s_mspool[ i ].count < s_mspool[ imin ].count )
+               imin = i;
+         }
+         if( s_mspool[ imin ].ms == NULL )
+            s_mspool[ imin ].ms = create_mspace( 0, 1 );
+         s_mspool[ imin ].count++;
+         hb_stack.allocator = ( void * ) &s_mspool[ imin ];
+      }
+      HB_FM_UNLOCK
+   }
+#endif
+}
+
+void hb_xexit_thread( void )
+{
+#if defined( HB_FM_DLMT_ALLOC )
+   if( hb_stack.allocator != NULL )
+   {
+      HB_FM_LOCK
+      ( ( PHB_MSPACE ) hb_stack.allocator )->count--;
+      hb_stack.allocator = NULL;
+      HB_FM_UNLOCK
+   }
+#endif
+}
+
 void hb_xsetfilename( char * szValue )
 {
 #ifdef HB_FM_STATISTICS
@@ -795,25 +922,6 @@ void hb_xinit( void ) /* Initialize fixed memory subsystem */
 #endif /* HB_FM_NEED_INIT */
 }
 
-#if defined( HB_FM_DL_ALLOC ) && defined( USE_DL_PREFIX )
-static void dlmalloc_destroy( void )
-{
-   if( ok_magic(gm) )
-   {
-      msegmentptr sp = &gm->seg;
-      while(sp != 0 )
-      {
-         char* base = sp->base;
-         size_t size = sp->size;
-         flag_t flag = sp->sflags;
-         sp = sp->next;
-         if( (flag & IS_MMAPPED_BIT) && !(flag & EXTERN_BIT) )
-            CALL_MUNMAP(base, size);
-      }
-   }
-}
-#endif
-
 /* Returns pointer to string containing printable version
    of pMem memory block */
 
@@ -931,7 +1039,9 @@ void hb_xexit( void ) /* Deinitialize fixed memory subsystem */
    }
 
 #if defined( HB_FM_DL_ALLOC )
-#  if defined( USE_DL_PREFIX )
+#  if defined( HB_FM_DLMT_ALLOC )
+      hb_mspace_cleanup();
+#  elif defined( USE_DL_PREFIX )
       dlmalloc_destroy();
 #  else
       malloc_trim( 0 );
@@ -946,7 +1056,9 @@ void hb_xexit( void ) /* Deinitialize fixed memory subsystem */
    HB_TRACE(HB_TR_DEBUG, ("hb_xexit()"));
 
 #if defined( HB_FM_DL_ALLOC )
-#  if defined( USE_DL_PREFIX )
+#  if defined( HB_FM_DLMT_ALLOC )
+      hb_mspace_cleanup();
+#  elif defined( USE_DL_PREFIX )
       dlmalloc_destroy();
 #  else
       malloc_trim( 0 );
diff --git a/harbour/source/vm/garbage.c b/harbour/source/vm/garbage.c
index badd13cf7d..0afb3139a3 100644
--- a/harbour/source/vm/garbage.c
+++ b/harbour/source/vm/garbage.c
@@ -72,9 +72,9 @@
 #  include "hbthread.h"
 #  include "hbatomic.h"
 
-/* Use spinlock instead of mutex in OS2 builds */
+/* Use spinlock instead of mutex */
 
-#  if defined( HB_SPINLOCK_INIT )
+#  if defined( HB_SPINLOCK_INIT ) && 1
 
       HB_SPINLOCK_T s_gcSpinLock = HB_SPINLOCK_INIT;
 #     define HB_GC_LOCK       HB_SPINLOCK_ACQUIRE( &s_gcSpinLock );