Updates from 0.2.1 into 0.2.1-inu-1.5release-0-2-1-inu-1-5

1 files changed, 808 insertions, 625 deletions

diff --git a/gc/include/private/gc_priv.h b/gc/include/private/gc_priv.h
index 5ce52a7..5135e3e 100644
--- a/gc/include/private/gc_priv.h
+++ b/gc/include/private/gc_priv.h
@@ -1,6 +1,9 @@
 /* 
  * Copyright 1988, 1989 Hans-J. Boehm, Alan J. Demers
  * Copyright (c) 1991-1994 by Xerox Corporation.  All rights reserved.
+ * Copyright (c) 1996-1999 by Silicon Graphics.  All rights reserved.
+ * Copyright (c) 1999-2001 by Hewlett-Packard Company. All rights reserved.
+ *
  *
  * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
  * OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.
@@ -11,7 +14,6 @@
  * provided the above notices are retained, and a notice that the code was
  * modified is included with the above copyright notice.
  */
-/* Boehm, February 16, 1996 2:30 pm PST */
  
 
 # ifndef GC_PRIVATE_H
@@ -38,16 +40,16 @@
 #   include "gc.h"
 # endif
 
-typedef GC_word word;
-typedef GC_signed_word signed_word;
+# ifndef GC_MARK_H
+#   include "../gc_mark.h"
+# endif
 
-# ifndef CONFIG_H
+# ifndef GCCONFIG_H
 #   include "gcconfig.h"
 # endif
 
-# ifndef HEADERS_H
-#   include "gc_hdrs.h"
-# endif
+typedef GC_word word;
+typedef GC_signed_word signed_word;
 
 typedef int GC_bool;
 # define TRUE 1
@@ -58,27 +60,54 @@ typedef char * ptr_t;	/* A generic pointer to which we can add	*/
 			/* Preferably identical to caddr_t, if it 	*/
 			/* exists.					*/
 			
+# ifndef HEADERS_H
+#   include "gc_hdrs.h"
+# endif
+
 #if defined(__STDC__)
 #   include <stdlib.h>
 #   if !(defined( sony_news ) )
 #       include <stddef.h>
 #   endif
 #   define VOLATILE volatile
-#   define CONST const
 #else
 #   ifdef MSWIN32
 #   	include <stdlib.h>
 #   endif
 #   define VOLATILE
-#   define CONST
 #endif
 
-#if 0 /* was once defined for AMIGA */
+#if 0 /* defined(__GNUC__) doesn't work yet */
+# define EXPECT(expr, outcome) __builtin_expect(expr,outcome)
+  /* Equivalent to (expr), but predict that usually (expr)==outcome. */
+#else
+# define EXPECT(expr, outcome) (expr)
+#endif /* __GNUC__ */
+
+# ifndef GC_LOCKS_H
+#   include "gc_locks.h"
+# endif
+
+# ifdef STACK_GROWS_DOWN
+#   define COOLER_THAN >
+#   define HOTTER_THAN <
+#   define MAKE_COOLER(x,y) if ((word)(x)+(y) > (word)(x)) {(x) += (y);} \
+			    else {(x) = (word)ONES;}
+#   define MAKE_HOTTER(x,y) (x) -= (y)
+# else
+#   define COOLER_THAN <
+#   define HOTTER_THAN >
+#   define MAKE_COOLER(x,y) if ((word)(x)-(y) < (word)(x)) {(x) -= (y);} else {(x) = 0;}
+#   define MAKE_HOTTER(x,y) (x) += (y)
+# endif
+
+#if defined(AMIGA) && defined(__SASC)
 #   define GC_FAR __far
 #else
 #   define GC_FAR
 #endif
 
+
 /*********************************/
 /*                               */
 /* Definitions for conservative  */
@@ -92,11 +121,9 @@ typedef char * ptr_t;	/* A generic pointer to which we can add	*/
 /*                               */
 /*********************************/
 
-#define STUBBORN_ALLOC	/* Define stubborn allocation primitives	*/
-#if defined(SRC_M3) || defined(SMALL_CONFIG)
-# undef STUBBORN_ALLOC
-#endif
-
+/* #define STUBBORN_ALLOC */
+		    /* Enable stubborm allocation, and thus a limited	*/
+		    /* form of incremental collection w/o dirty bits.	*/
 
 /* #define ALL_INTERIOR_POINTERS */
 		    /* Forces all pointers into the interior of an 	*/
@@ -123,6 +150,8 @@ typedef char * ptr_t;	/* A generic pointer to which we can add	*/
 		    /*    touched.					*/
 		    /* If you can easily avoid using this option, do.	*/
 		    /* If not, try to keep individual objects small.	*/
+		    /* This is now really controlled at startup,	*/
+		    /* through GC_all_interior_pointers.		*/
 		    
 #define PRINTSTATS  /* Print garbage collection statistics          	*/
 		    /* For less verbose output, undefine in reclaim.c 	*/
@@ -153,12 +182,12 @@ typedef char * ptr_t;	/* A generic pointer to which we can add	*/
 #   define GATHERSTATS
 #endif
 
-#ifdef FINALIZE_ON_DEMAND
-#   define GC_INVOKE_FINALIZERS()
-#else
-#   define GC_INVOKE_FINALIZERS() (void)GC_invoke_finalizers()
+#if defined(PRINTSTATS) || !defined(SMALL_CONFIG)
+#   define CONDPRINT  /* Print some things if GC_print_stats is set */
 #endif
 
+#define GC_INVOKE_FINALIZERS() GC_notify_or_invoke_finalizers()
+
 #define MERGE_SIZES /* Round up some object sizes, so that fewer distinct */
 		    /* free lists are actually maintained.  This applies  */
 		    /* only to the top level routines in misc.c, not to   */
@@ -170,13 +199,10 @@ typedef char * ptr_t;	/* A generic pointer to which we can add	*/
 		    /* May save significant amounts of space for obj_map  */
 		    /* entries.						  */
 
-#ifndef OLD_BLOCK_ALLOC
-   /* Macros controlling large block allocation strategy.	*/
-#  define EXACT_FIRST  	/* Make a complete pass through the large object */
-			/* free list before splitting a block		 */
-#  define PRESERVE_LAST /* Do not divide last allocated heap segment	 */
-			/* unless we would otherwise need to expand the	 */
-			/* heap.					 */
+#if defined(USE_MARK_BYTES) && !defined(ALIGN_DOUBLE)
+#  define ALIGN_DOUBLE
+   /* We use one byte for every 2 words, which doesn't allow for	*/
+   /* odd numbered words to have mark bits.				*/
 #endif
 
 /* ALIGN_DOUBLE requires MERGE_SIZES at present. */
@@ -184,15 +210,17 @@ typedef char * ptr_t;	/* A generic pointer to which we can add	*/
 #   define MERGE_SIZES
 # endif
 
-#if defined(ALL_INTERIOR_POINTERS) && !defined(DONT_ADD_BYTE_AT_END)
-# define ADD_BYTE_AT_END
+#if !defined(DONT_ADD_BYTE_AT_END)
+# define EXTRA_BYTES GC_all_interior_pointers
+#else
+# define EXTRA_BYTES 0
 #endif
 
 
 # ifndef LARGE_CONFIG
-#   define MINHINCR 16	/* Minimum heap increment, in blocks of HBLKSIZE  */
-			/* Must be multiple of largest page size.	  */
-#   define MAXHINCR 512	/* Maximum heap increment, in blocks              */
+#   define MINHINCR 16	 /* Minimum heap increment, in blocks of HBLKSIZE  */
+			 /* Must be multiple of largest page size.	   */
+#   define MAXHINCR 2048 /* Maximum heap increment, in blocks              */
 # else
 #   define MINHINCR 64
 #   define MAXHINCR 4096
@@ -225,17 +253,22 @@ typedef char * ptr_t;	/* A generic pointer to which we can add	*/
  * Number of frames and arguments to save in objects allocated by
  * debugging allocator.
  */
-#   define NFRAMES 6	/* Number of frames to save. Even for		*/
+#   ifndef SAVE_CALL_COUNT
+#     define NFRAMES 6	/* Number of frames to save. Even for		*/
 			/* alignment reasons.				*/
+#   else
+#     define NFRAMES ((SAVE_CALL_COUNT + 1) & ~1)
+#   endif
 #   define NARGS 2	/* Mumber of arguments to save for each call.	*/
 
 #   define NEED_CALLINFO
 
 /* Fill in the pc and argument information for up to NFRAMES of my	*/
 /* callers.  Ignore my frame and my callers frame.			*/
-void GC_save_callers (/* struct callinfo info[NFRAMES] */);
-
-void GC_print_callers (/* struct callinfo info[NFRAMES] */);
+struct callinfo;
+void GC_save_callers GC_PROTO((struct callinfo info[NFRAMES]));
+  
+void GC_print_callers GC_PROTO((struct callinfo info[NFRAMES]));
 
 #else
 
@@ -249,7 +282,7 @@ void GC_print_callers (/* struct callinfo info[NFRAMES] */);
 
 #ifdef NEED_CALLINFO
     struct callinfo {
-	word ci_pc;
+	word ci_pc;  	/* Caller, not callee, pc	*/
 #	if NARGS > 0
 	    word ci_arg[NARGS];	/* bit-wise complement to avoid retention */
 #	endif
@@ -278,6 +311,13 @@ void GC_print_callers (/* struct callinfo info[NFRAMES] */);
 #   define MS_TIME_DIFF(a,b) ((double) (a.tv_sec - b.tv_sec) * 1000.0 \
                                + (double) (a.tv_usec - b.tv_usec) / 1000.0)
 #else /* !BSD_TIME */
+# if defined(MSWIN32) || defined(MSWINCE)
+#   include <windows.h>
+#   include <winbase.h>
+#   define CLOCK_TYPE DWORD
+#   define GET_TIME(x) x = GetTickCount()
+#   define MS_TIME_DIFF(a,b) ((long)((a)-(b)))
+# else /* !MSWIN32, !MSWINCE, !BSD_TIME */
 #   include <time.h>
 #   if !defined(__STDC__) && defined(SPARC) && defined(SUNOS4)
       clock_t clock();	/* Not in time.h, where it belongs	*/
@@ -303,6 +343,7 @@ void GC_print_callers (/* struct callinfo info[NFRAMES] */);
 #   define GET_TIME(x) x = clock()
 #   define MS_TIME_DIFF(a,b) ((unsigned long) \
 		(1000.0*(double)((a)-(b))/(double)CLOCKS_PER_SEC))
+# endif /* !MSWIN32 */
 #endif /* !BSD_TIME */
 
 /* We use bzero and bcopy internally.  They may not be available.	*/
@@ -322,6 +363,9 @@ void GC_print_callers (/* struct callinfo info[NFRAMES] */);
 #   include <string.h>
 #   define BCOPY_EXISTS
 # endif
+# if defined(MACOSX)
+#   define BCOPY_EXISTS
+# endif
 
 # ifndef BCOPY_EXISTS
 #   include <string.h>
@@ -339,6 +383,7 @@ void GC_print_callers (/* struct callinfo info[NFRAMES] */);
 /* GET_MEM is currently not assumed to retrieve 0 filled space, */
 /* though we should perhaps take advantage of the case in which */
 /* does.							*/
+struct hblk;	/* See below.	*/
 # ifdef PCR
     char * real_malloc();
 #   define GET_MEM(bytes) HBLKPTR(real_malloc((size_t)bytes + GC_page_size) \
@@ -350,7 +395,8 @@ void GC_print_callers (/* struct callinfo info[NFRAMES] */);
 				    + GC_page_size) \
                                     + GC_page_size-1)
 #   else
-#     if defined(AMIGA) || defined(NEXT) || defined(MACOSX) || defined(DOS4GW)
+#     if defined(NEXT) || defined(MACOSX) || defined(DOS4GW) || \
+	 (defined(AMIGA) && !defined(GC_AMIGA_FASTALLOC))
 #       define GET_MEM(bytes) HBLKPTR((size_t) \
 				      calloc(1, (size_t)bytes + GC_page_size) \
                                       + GC_page_size-1)
@@ -371,198 +417,26 @@ void GC_print_callers (/* struct callinfo info[NFRAMES] */);
 			NewPtrClear(bytes + GC_page_size) + GC_page_size-1)
 #	    endif
 #	  else
-              extern ptr_t GC_unix_get_mem();
-#             define GET_MEM(bytes) (struct hblk *)GC_unix_get_mem(bytes)
+#	    ifdef MSWINCE
+	      extern ptr_t GC_wince_get_mem();
+#	      define GET_MEM(bytes) (struct hblk *)GC_wince_get_mem(bytes)
+#	    else
+#	      if defined(AMIGA) && defined(GC_AMIGA_FASTALLOC)
+	        extern void *GC_amiga_get_mem(size_t size);
+		define GET_MEM(bytes) HBLKPTR((size_t) \
+                  GC_amiga_get_mem((size_t)bytes + GC_page_size) \
+		  + GC_page_size-1)
+#	      else
+                extern ptr_t GC_unix_get_mem();
+#               define GET_MEM(bytes) (struct hblk *)GC_unix_get_mem(bytes)
+#	      endif
+#	    endif
 #	  endif
 #	endif
 #     endif
 #   endif
 # endif
 
-/*
- * Mutual exclusion between allocator/collector routines.
- * Needed if there is more than one allocator thread.
- * FASTLOCK() is assumed to try to acquire the lock in a cheap and
- * dirty way that is acceptable for a few instructions, e.g. by
- * inhibiting preemption.  This is assumed to have succeeded only
- * if a subsequent call to FASTLOCK_SUCCEEDED() returns TRUE.
- * FASTUNLOCK() is called whether or not FASTLOCK_SUCCEEDED().
- * If signals cannot be tolerated with the FASTLOCK held, then
- * FASTLOCK should disable signals.  The code executed under
- * FASTLOCK is otherwise immune to interruption, provided it is
- * not restarted.
- * DCL_LOCK_STATE declares any local variables needed by LOCK and UNLOCK
- * and/or DISABLE_SIGNALS and ENABLE_SIGNALS and/or FASTLOCK.
- * (There is currently no equivalent for FASTLOCK.)
- */  
-# ifdef THREADS
-#  ifdef PCR_OBSOLETE	/* Faster, but broken with multiple lwp's	*/
-#    include  "th/PCR_Th.h"
-#    include  "th/PCR_ThCrSec.h"
-     extern struct PCR_Th_MLRep GC_allocate_ml;
-#    define DCL_LOCK_STATE  PCR_sigset_t GC_old_sig_mask
-#    define LOCK() PCR_Th_ML_Acquire(&GC_allocate_ml) 
-#    define UNLOCK() PCR_Th_ML_Release(&GC_allocate_ml)
-#    define FASTLOCK() PCR_ThCrSec_EnterSys()
-     /* Here we cheat (a lot): */
-#        define FASTLOCK_SUCCEEDED() (*(int *)(&GC_allocate_ml) == 0)
-		/* TRUE if nobody currently holds the lock */
-#    define FASTUNLOCK() PCR_ThCrSec_ExitSys()
-#  endif
-#  ifdef PCR
-#    include <base/PCR_Base.h>
-#    include <th/PCR_Th.h>
-     extern PCR_Th_ML GC_allocate_ml;
-#    define DCL_LOCK_STATE \
-	 PCR_ERes GC_fastLockRes; PCR_sigset_t GC_old_sig_mask
-#    define LOCK() PCR_Th_ML_Acquire(&GC_allocate_ml)
-#    define UNLOCK() PCR_Th_ML_Release(&GC_allocate_ml)
-#    define FASTLOCK() (GC_fastLockRes = PCR_Th_ML_Try(&GC_allocate_ml))
-#    define FASTLOCK_SUCCEEDED() (GC_fastLockRes == PCR_ERes_okay)
-#    define FASTUNLOCK()  {\
-        if( FASTLOCK_SUCCEEDED() ) PCR_Th_ML_Release(&GC_allocate_ml); }
-#  endif
-#  ifdef SRC_M3
-     extern word RT0u__inCritical;
-#    define LOCK() RT0u__inCritical++
-#    define UNLOCK() RT0u__inCritical--
-#  endif
-#  ifdef SOLARIS_THREADS
-#    include <thread.h>
-#    include <signal.h>
-     extern mutex_t GC_allocate_ml;
-#    define LOCK() mutex_lock(&GC_allocate_ml);
-#    define UNLOCK() mutex_unlock(&GC_allocate_ml);
-#  endif
-#  ifdef LINUX_THREADS
-#    include <pthread.h>
-#    ifdef __i386__
-       inline static int GC_test_and_set(volatile unsigned int *addr) {
-	  int oldval;
-	  /* Note: the "xchg" instruction does not need a "lock" prefix */
-	  __asm__ __volatile__("xchgl %0, %1"
-		: "=r"(oldval), "=m"(*(addr))
-		: "0"(1), "m"(*(addr)));
-	  return oldval;
-       }
-#    else
-       -- > Need implementation of GC_test_and_set()
-#    endif
-#    define GC_clear(addr) (*(addr) = 0)
-
-     extern volatile unsigned int GC_allocate_lock;
-	/* This is not a mutex because mutexes that obey the (optional)     */
-	/* POSIX scheduling rules are subject to convoys in high contention */
-	/* applications.  This is basically a spin lock.		    */
-     extern pthread_t GC_lock_holder;
-     extern void GC_lock(void);
-	/* Allocation lock holder.  Only set if acquired by client through */
-	/* GC_call_with_alloc_lock.					   */
-#    define SET_LOCK_HOLDER() GC_lock_holder = pthread_self()
-#    define NO_THREAD (pthread_t)(-1)
-#    define UNSET_LOCK_HOLDER() GC_lock_holder = NO_THREAD
-#    define I_HOLD_LOCK() (pthread_equal(GC_lock_holder, pthread_self()))
-#    ifdef UNDEFINED
-#    	define LOCK() pthread_mutex_lock(&GC_allocate_ml)
-#    	define UNLOCK() pthread_mutex_unlock(&GC_allocate_ml)
-#    else
-#	define LOCK() \
-		{ if (GC_test_and_set(&GC_allocate_lock)) GC_lock(); }
-#	define UNLOCK() \
-		GC_clear(&GC_allocate_lock)
-#    endif
-     extern GC_bool GC_collecting;
-#    define ENTER_GC() \
-		{ \
-		    GC_collecting = 1; \
-		}
-#    define EXIT_GC() GC_collecting = 0;
-#  endif /* LINUX_THREADS */
-#  if defined(IRIX_THREADS) || defined(IRIX_JDK_THREADS)
-#    include <pthread.h>
-#    include <mutex.h>
-
-#    if __mips < 3 || !(defined (_ABIN32) || defined(_ABI64)) \
-	|| !defined(_COMPILER_VERSION) || _COMPILER_VERSION < 700
-#        define GC_test_and_set(addr, v) test_and_set(addr,v)
-#    else
-#	 define GC_test_and_set(addr, v) __test_and_set(addr,v)
-#    endif
-     extern unsigned long GC_allocate_lock;
-	/* This is not a mutex because mutexes that obey the (optional) 	*/
-	/* POSIX scheduling rules are subject to convoys in high contention	*/
-	/* applications.  This is basically a spin lock.			*/
-     extern pthread_t GC_lock_holder;
-     extern void GC_lock(void);
-	/* Allocation lock holder.  Only set if acquired by client through */
-	/* GC_call_with_alloc_lock.					   */
-#    define SET_LOCK_HOLDER() GC_lock_holder = pthread_self()
-#    define NO_THREAD (pthread_t)(-1)
-#    define UNSET_LOCK_HOLDER() GC_lock_holder = NO_THREAD
-#    define I_HOLD_LOCK() (pthread_equal(GC_lock_holder, pthread_self()))
-#    ifdef UNDEFINED
-#    	define LOCK() pthread_mutex_lock(&GC_allocate_ml)
-#    	define UNLOCK() pthread_mutex_unlock(&GC_allocate_ml)
-#    else
-#	define LOCK() { if (GC_test_and_set(&GC_allocate_lock, 1)) GC_lock(); }
-#       if __mips >= 3 && (defined (_ABIN32) || defined(_ABI64)) \
-	   && defined(_COMPILER_VERSION) && _COMPILER_VERSION >= 700
-#	    define UNLOCK() __lock_release(&GC_allocate_lock)
-#	else
-	    /* The function call in the following should prevent the	*/
-	    /* compiler from moving assignments to below the UNLOCK.	*/
-	    /* This is probably not necessary for ucode or gcc 2.8.	*/
-	    /* It may be necessary for Ragnarok and future gcc		*/
-	    /* versions.						*/
-#           define UNLOCK() { GC_noop1(&GC_allocate_lock); \
-			*(volatile unsigned long *)(&GC_allocate_lock) = 0; }
-#	endif
-#    endif
-     extern GC_bool GC_collecting;
-#    define ENTER_GC() \
-		{ \
-		    GC_collecting = 1; \
-		}
-#    define EXIT_GC() GC_collecting = 0;
-#  endif /* IRIX_THREADS || IRIX_JDK_THREADS */
-#  ifdef WIN32_THREADS
-#    include <windows.h>
-     GC_API CRITICAL_SECTION GC_allocate_ml;
-#    define LOCK() EnterCriticalSection(&GC_allocate_ml);
-#    define UNLOCK() LeaveCriticalSection(&GC_allocate_ml);
-#  endif
-#  ifndef SET_LOCK_HOLDER
-#      define SET_LOCK_HOLDER()
-#      define UNSET_LOCK_HOLDER()
-#      define I_HOLD_LOCK() FALSE
-		/* Used on platforms were locks can be reacquired,	*/
-		/* so it doesn't matter if we lie.			*/
-#  endif
-# else
-#    define LOCK()
-#    define UNLOCK()
-# endif
-# ifndef SET_LOCK_HOLDER
-#   define SET_LOCK_HOLDER()
-#   define UNSET_LOCK_HOLDER()
-#   define I_HOLD_LOCK() FALSE
-		/* Used on platforms were locks can be reacquired,	*/
-		/* so it doesn't matter if we lie.			*/
-# endif
-# ifndef ENTER_GC
-#   define ENTER_GC()
-#   define EXIT_GC()
-# endif
-
-# ifndef DCL_LOCK_STATE
-#   define DCL_LOCK_STATE
-# endif
-# ifndef FASTLOCK
-#   define FASTLOCK() LOCK()
-#   define FASTLOCK_SUCCEEDED() TRUE
-#   define FASTUNLOCK() UNLOCK()
-# endif
-
 /* Delay any interrupts or signals that may abort this thread.  Data	*/
 /* structures are in a consistent state outside this pair of calls.	*/
 /* ANSI C allows both to be empty (though the standard isn't very	*/
@@ -576,9 +450,9 @@ void GC_print_callers (/* struct callinfo info[NFRAMES] */);
 		PCR_Th_SetSigMask(&GC_old_sig_mask, NIL)
 # else
 #   if defined(SRC_M3) || defined(AMIGA) || defined(SOLARIS_THREADS) \
-	|| defined(MSWIN32) || defined(MACOS) || defined(DJGPP) \
-	|| defined(NO_SIGNALS) || defined(IRIX_THREADS) \
-	|| defined(IRIX_JDK_THREADS) || defined(LINUX_THREADS) 
+	|| defined(MSWIN32) || defined(MSWINCE) || defined(MACOS) \
+	|| defined(DJGPP) || defined(NO_SIGNALS) || defined(IRIX_THREADS) \
+	|| defined(LINUX_THREADS) 
 			/* Also useful for debugging.		*/
 	/* Should probably use thr_sigsetmask for SOLARIS_THREADS. */
 #     define DISABLE_SIGNALS()
@@ -607,7 +481,7 @@ void GC_print_callers (/* struct callinfo info[NFRAMES] */);
 # else
 #   if defined(SOLARIS_THREADS) || defined(WIN32_THREADS) \
 	|| defined(IRIX_THREADS) || defined(LINUX_THREADS) \
-	|| defined(IRIX_JDK_THREADS)
+	|| defined(HPUX_THREADS)
       void GC_stop_world();
       void GC_start_world();
 #     define STOP_WORLD() GC_stop_world()
@@ -641,6 +515,13 @@ void GC_print_callers (/* struct callinfo info[NFRAMES] */);
 # define WARN(msg,arg) (*GC_current_warn_proc)(msg, (GC_word)(arg))
 extern GC_warn_proc GC_current_warn_proc;
 
+/* Get environment entry */
+#if !defined(NO_GETENV)
+#   define GETENV(name) getenv(name)
+#else
+#   define GETENV(name) 0
+#endif
+
 /*********************************/
 /*                               */
 /* Word-size-dependent defines   */
@@ -670,7 +551,7 @@ extern GC_warn_proc GC_current_warn_proc;
 #define WORDSZ ((word)CPP_WORDSZ)
 #define SIGNB  ((word)1 << (WORDSZ-1))
 #define BYTES_PER_WORD      ((word)(sizeof (word)))
-#define ONES                ((word)(-1))
+#define ONES                ((word)(signed_word)(-1))
 #define divWORDSZ(n) ((n) >> LOGWL)	   /* divide n by size of word      */
 
 /*********************/
@@ -723,6 +604,8 @@ extern GC_warn_proc GC_current_warn_proc;
 /*  max size objects supported by freelist (larger objects may be   */
 /*  allocated, but less efficiently)                                */
 
+#define CPP_MAXOBJBYTES (CPP_HBLKSIZE/2)
+#define MAXOBJBYTES ((word)CPP_MAXOBJBYTES)
 #define CPP_MAXOBJSZ    BYTES_TO_WORDS(CPP_HBLKSIZE/2)
 #define MAXOBJSZ ((word)CPP_MAXOBJSZ)
 		
@@ -743,25 +626,28 @@ extern GC_warn_proc GC_current_warn_proc;
 # define HBLKDISPL(objptr) (((word) (objptr)) & (HBLKSIZE-1))
 
 /* Round up byte allocation requests to integral number of words, etc. */
-# ifdef ADD_BYTE_AT_END
-#   define ROUNDED_UP_WORDS(n) BYTES_TO_WORDS((n) + WORDS_TO_BYTES(1))
-#   ifdef ALIGN_DOUBLE
-#       define ALIGNED_WORDS(n) (BYTES_TO_WORDS((n) + WORDS_TO_BYTES(2)) & ~1)
-#   else
-#       define ALIGNED_WORDS(n) ROUNDED_UP_WORDS(n)
-#   endif
-#   define SMALL_OBJ(bytes) ((bytes) < WORDS_TO_BYTES(MAXOBJSZ))
-#   define ADD_SLOP(bytes) ((bytes)+1)
-# else
-#   define ROUNDED_UP_WORDS(n) BYTES_TO_WORDS((n) + (WORDS_TO_BYTES(1) - 1))
-#   ifdef ALIGN_DOUBLE
+# define ROUNDED_UP_WORDS(n) \
+	BYTES_TO_WORDS((n) + (WORDS_TO_BYTES(1) - 1 + EXTRA_BYTES))
+# ifdef ALIGN_DOUBLE
 #       define ALIGNED_WORDS(n) \
-			(BYTES_TO_WORDS((n) + WORDS_TO_BYTES(2) - 1) & ~1)
-#   else
+	    (BYTES_TO_WORDS((n) + WORDS_TO_BYTES(2) - 1 + EXTRA_BYTES) & ~1)
+# else
 #       define ALIGNED_WORDS(n) ROUNDED_UP_WORDS(n)
+# endif
+# define SMALL_OBJ(bytes) ((bytes) < (MAXOBJBYTES - EXTRA_BYTES))
+# define ADD_SLOP(bytes) ((bytes) + EXTRA_BYTES)
+# ifndef MIN_WORDS
+    /* MIN_WORDS is the size of the smallest allocated object.	*/
+    /* 1 and 2 are the only valid values.			*/
+    /* 2 must be used if:					*/
+    /* - GC_gcj_malloc can be used for objects of requested 	*/
+    /*   size  smaller than 2 words, or				*/
+    /* - USE_MARK_BYTES is defined.				*/
+#   if defined(USE_MARK_BYTES) || defined(GC_GCJ_SUPPORT)
+#     define MIN_WORDS 2   	/* Smallest allocated object.	*/
+#   else
+#     define MIN_WORDS 1
 #   endif
-#   define SMALL_OBJ(bytes) ((bytes) <= WORDS_TO_BYTES(MAXOBJSZ))
-#   define ADD_SLOP(bytes) (bytes)
 # endif
 
 
@@ -772,11 +658,19 @@ extern GC_warn_proc GC_current_warn_proc;
  */
  
 # ifdef LARGE_CONFIG
-#   define LOG_PHT_ENTRIES  17
+#   define LOG_PHT_ENTRIES  19  /* Collisions likely at 512K blocks,	*/
+				/* which is >= 2GB.  Each table takes	*/
+				/* 64KB.				*/
 # else
-#   define LOG_PHT_ENTRIES  14	/* Collisions are likely if heap grows	*/
-				/* to more than 16K hblks = 64MB.	*/
-				/* Each hash table occupies 2K bytes.   */
+#   ifdef SMALL_CONFIG
+#     define LOG_PHT_ENTRIES  14 /* Collisions are likely if heap grows	*/
+				 /* to more than 16K hblks = 64MB.	*/
+				 /* Each hash table occupies 2K bytes.   */
+#   else /* default "medium" configuration */
+#     define LOG_PHT_ENTRIES  16 /* Collisions are likely if heap grows	*/
+				 /* to more than 16K hblks >= 256MB.	*/
+				 /* Each hash table occupies 8K bytes.  */
+#   endif
 # endif
 # define PHT_ENTRIES ((word)1 << LOG_PHT_ENTRIES)
 # define PHT_SIZE (PHT_ENTRIES >> LOGWL)
@@ -790,6 +684,10 @@ typedef word page_hash_table[PHT_SIZE];
 		(bl)[divWORDSZ(index)] |= (word)1 << modWORDSZ(index)
 # define clear_pht_entry_from_index(bl, index) \
 		(bl)[divWORDSZ(index)] &= ~((word)1 << modWORDSZ(index))
+/* And a dumb but thread-safe version of set_pht_entry_from_index.	*/
+/* This sets (many) extra bits.						*/
+# define set_pht_entry_from_index_safe(bl, index) \
+		(bl)[divWORDSZ(index)] = ONES
 	
 
 
@@ -802,20 +700,28 @@ typedef word page_hash_table[PHT_SIZE];
 /*  heap block header */
 #define HBLKMASK   (HBLKSIZE-1)
 
-#define BITS_PER_HBLK (HBLKSIZE * 8)
+#define BITS_PER_HBLK (CPP_HBLKSIZE * 8)
 
 #define MARK_BITS_PER_HBLK (BITS_PER_HBLK/CPP_WORDSZ)
 	   /* upper bound                                    */
-	   /* We allocate 1 bit/word.  Only the first word   */
+	   /* We allocate 1 bit/word, unless USE_MARK_BYTES  */
+	   /* is defined.  Only the first word   	     */
 	   /* in each object is actually marked.             */
 
-# ifdef ALIGN_DOUBLE
-#   define MARK_BITS_SZ (((MARK_BITS_PER_HBLK + 2*CPP_WORDSZ - 1) \
-			  / (2*CPP_WORDSZ))*2)
+# ifdef USE_MARK_BYTES
+#   define MARK_BITS_SZ (MARK_BITS_PER_HBLK/2)
+	/* Unlike the other case, this is in units of bytes.		*/
+	/* We actually allocate only every second mark bit, since we	*/
+	/* force all objects to be doubleword aligned.			*/
+	/* However, each mark bit is allocated as a byte.		*/
 # else
-#   define MARK_BITS_SZ ((MARK_BITS_PER_HBLK + CPP_WORDSZ - 1)/CPP_WORDSZ)
+#   define MARK_BITS_SZ (MARK_BITS_PER_HBLK/CPP_WORDSZ)
 # endif
-	   /* Upper bound on number of mark words per heap block  */
+
+/* We maintain layout maps for heap blocks containing objects of a given */
+/* size.  Each entry in this map describes a byte offset and has the	 */
+/* following type.							 */
+typedef unsigned char map_entry_type;
 
 struct hblkhdr {
     word hb_sz;  /* If in use, size in words, of objects in the block. */
@@ -826,7 +732,8 @@ struct hblkhdr {
     struct hblk * hb_prev;	/* Backwards link for free list.	*/
     word hb_descr;   		/* object descriptor for marking.  See	*/
     				/* mark.h.				*/
-    char* hb_map;	/* A pointer to a pointer validity map of the block. */
+    map_entry_type * hb_map;	
+    			/* A pointer to a pointer validity map of the block. */
     		      	/* See GC_obj_map.				     */
     		     	/* Valid for all blocks with headers.		     */
     		     	/* Free blocks point to GC_invalid_map.		     */
@@ -848,51 +755,49 @@ struct hblkhdr {
     				/* Value of GC_gc_no when block was	*/
     				/* last allocated or swept. May wrap.   */
 				/* For a free block, this is maintained */
-				/* unly for USE_MUNMAP, and indicates	*/
+				/* only for USE_MUNMAP, and indicates	*/
 				/* when the header was allocated, or	*/
 				/* when the size of the block last	*/
 				/* changed.				*/
-    word hb_marks[MARK_BITS_SZ];
+#   ifdef USE_MARK_BYTES
+      union {
+        char _hb_marks[MARK_BITS_SZ];
+			    /* The i'th byte is 1 if the object 	*/
+			    /* starting at word 2i is marked, 0 o.w.	*/
+	word dummy;	/* Force word alignment of mark bytes. */
+      } _mark_byte_union;
+#     define hb_marks _mark_byte_union._hb_marks
+#   else
+      word hb_marks[MARK_BITS_SZ];
 			    /* Bit i in the array refers to the             */
 			    /* object starting at the ith word (header      */
 			    /* INCLUDED) in the heap block.                 */
 			    /* The lsb of word 0 is numbered 0.		    */
+			    /* Unused bits are invalid, and are 	    */
+			    /* occasionally set, e.g for uncollectable	    */
+			    /* objects.					    */
+#   endif /* !USE_MARK_BYTES */
 };
 
 /*  heap block body */
 
-# define DISCARD_WORDS 0
-	/* Number of words to be dropped at the beginning of each block	*/
-	/* Must be a multiple of WORDSZ.  May reasonably be nonzero	*/
-	/* on machines that don't guarantee longword alignment of	*/
-	/* pointers, so that the number of false hits is minimized.	*/
-	/* 0 and WORDSZ are probably the only reasonable values.	*/
-
-# define BODY_SZ ((HBLKSIZE-WORDS_TO_BYTES(DISCARD_WORDS))/sizeof(word))
+# define BODY_SZ (HBLKSIZE/sizeof(word))
 
 struct hblk {
-#   if (DISCARD_WORDS != 0)
+#   if 0  /* DISCARDWORDS no longer supported */
         word garbage[DISCARD_WORDS];
 #   endif
     word hb_body[BODY_SZ];
 };
 
-# define HDR_WORDS ((word)DISCARD_WORDS)
-# define HDR_BYTES ((word)WORDS_TO_BYTES(DISCARD_WORDS))
-
 # define OBJ_SZ_TO_BLOCKS(sz) \
-    divHBLKSZ(HDR_BYTES + WORDS_TO_BYTES(sz) + HBLKSIZE-1)
+    divHBLKSZ(WORDS_TO_BYTES(sz) + HBLKSIZE-1)
     /* Size of block (in units of HBLKSIZE) needed to hold objects of	*/
     /* given sz (in words).						*/
 
 /* Object free list link */
 # define obj_link(p) (*(ptr_t *)(p))
 
-/* The type of mark procedures.  This really belongs in gc_mark.h.	*/
-/* But we put it here, so that we can avoid scanning the mark proc	*/
-/* table.								*/
-typedef struct ms_entry * (*mark_proc)(/* word * addr, mark_stack_ptr,
-					  mark_stack_limit, env */);
 # define LOG_MAX_MARK_PROCS 6
 # define MAX_MARK_PROCS (1 << LOG_MAX_MARK_PROCS)
 
@@ -906,12 +811,12 @@ typedef struct ms_entry * (*mark_proc)(/* word * addr, mark_stack_ptr,
 #   ifdef PCR
 #     define MAX_ROOT_SETS 1024
 #   else
-#     ifdef MSWIN32
-#	define MAX_ROOT_SETS 512
+#     if defined(MSWIN32) || defined(MSWINCE)
+#	define MAX_ROOT_SETS 1024
 	    /* Under NT, we add only written pages, which can result 	*/
 	    /* in many small root sets.					*/
 #     else
-#       define MAX_ROOT_SETS 64
+#       define MAX_ROOT_SETS 256
 #     endif
 #   endif
 # endif
@@ -934,14 +839,14 @@ struct exclusion {
 struct roots {
 	ptr_t r_start;
 	ptr_t r_end;
-#	ifndef MSWIN32
+#	if !defined(MSWIN32) && !defined(MSWINCE)
 	  struct roots * r_next;
 #	endif
 	GC_bool r_tmp;
 	  	/* Delete before registering new dynamic libraries */
 };
 
-#ifndef MSWIN32
+#if !defined(MSWIN32) && !defined(MSWINCE)
     /* Size of hash table index to roots.	*/
 #   define LOG_RT_SIZE 6
 #   define RT_SIZE (1 << LOG_RT_SIZE) /* Power of 2, may be != MAX_ROOT_SETS */
@@ -969,16 +874,28 @@ struct roots {
 struct _GC_arrays {
   word _heapsize;
   word _max_heapsize;
+  word _requested_heapsize;	/* Heap size due to explicit expansion */
   ptr_t _last_heap_addr;
   ptr_t _prev_heap_addr;
   word _large_free_bytes;
 	/* Total bytes contained in blocks on large object free */
 	/* list.						*/
+  word _large_allocd_bytes;
+  	/* Total number of bytes in allocated large objects blocks.	*/
+  	/* For the purposes of this counter and the next one only, a 	*/
+  	/* large object is one that occupies a block of at least	*/
+  	/* 2*HBLKSIZE.							*/
+  word _max_large_allocd_bytes;
+  	/* Maximum number of bytes that were ever allocated in		*/
+  	/* large object blocks.  This is used to help decide when it	*/
+  	/* is safe to split up a large block.				*/
   word _words_allocd_before_gc;
 		/* Number of words allocated before this	*/
 		/* collection cycle.				*/
-  word _words_allocd;
+# ifndef SEPARATE_GLOBALS
+    word _words_allocd;
   	/* Number of words allocated during this collection cycle */
+# endif
   word _words_wasted;
   	/* Number of words wasted due to internal fragmentation	*/
   	/* in large objects, or due to dropping blacklisted     */
@@ -993,14 +910,21 @@ struct _GC_arrays {
   word _mem_freed;
   	/* Number of explicitly deallocated words of memory	*/
   	/* since last collection.				*/
-  mark_proc _mark_procs[MAX_MARK_PROCS];
+  ptr_t _scratch_end_ptr;
+  ptr_t _scratch_last_end_ptr;
+	/* Used by headers.c, and can easily appear to point to	*/
+	/* heap.						*/
+  GC_mark_proc _mark_procs[MAX_MARK_PROCS];
   	/* Table of user-defined mark procedures.  There is	*/
 	/* a small number of these, which can be referenced	*/
 	/* by DS_PROC mark descriptors.  See gc_mark.h.		*/
-  ptr_t _objfreelist[MAXOBJSZ+1];
+
+# ifndef SEPARATE_GLOBALS
+    ptr_t _objfreelist[MAXOBJSZ+1];
 			  /* free list for objects */
-  ptr_t _aobjfreelist[MAXOBJSZ+1];
+    ptr_t _aobjfreelist[MAXOBJSZ+1];
 			  /* free list for atomic objs 	*/
+# endif
 
   ptr_t _uobjfreelist[MAXOBJSZ+1];
 			  /* uncollectable but traced objs 	*/
@@ -1033,41 +957,28 @@ struct _GC_arrays {
     ptr_t _sobjfreelist[MAXOBJSZ+1];
 # endif
   			  /* free list for immutable objects	*/
-  ptr_t _obj_map[MAXOBJSZ+1];
+  map_entry_type * _obj_map[MAXOBJSZ+1];
                        /* If not NIL, then a pointer to a map of valid  */
     		       /* object addresses. _obj_map[sz][i] is j if the	*/
     		       /* address block_start+i is a valid pointer      */
-    		       /* to an object at				*/
-    		       /* block_start+i&~3 - WORDS_TO_BYTES(j).		*/
-    		       /* (If ALL_INTERIOR_POINTERS is defined, then	*/
-    		       /* instead ((short *)(hb_map[sz])[i] is j if	*/
-    		       /* block_start+WORDS_TO_BYTES(i) is in the	*/
-    		       /* interior of an object starting at		*/
-    		       /* block_start+WORDS_TO_BYTES(i-j)).		*/
-    		       /* It is OBJ_INVALID if				*/
-    		       /* block_start+WORDS_TO_BYTES(i) is not		*/
-    		       /* valid as a pointer to an object.              */
-    		       /* We assume all values of j <= OBJ_INVALID.	*/
-    		       /* The zeroth entry corresponds to large objects.*/
-#   ifdef ALL_INTERIOR_POINTERS
-#	define map_entry_type short
-#       define OBJ_INVALID 0x7fff
-#	define MAP_ENTRY(map, bytes) \
-		(((map_entry_type *)(map))[BYTES_TO_WORDS(bytes)])
-#	define MAP_ENTRIES BYTES_TO_WORDS(HBLKSIZE)
-#	define MAP_SIZE (MAP_ENTRIES * sizeof(map_entry_type))
-#	define OFFSET_VALID(displ) TRUE
-#	define CPP_MAX_OFFSET (HBLKSIZE - HDR_BYTES - 1)
-#	define MAX_OFFSET ((word)CPP_MAX_OFFSET)
-#   else
-#	define map_entry_type char
-#       define OBJ_INVALID 0x7f
-#	define MAP_ENTRY(map, bytes) \
-		(map)[bytes]
-#	define MAP_ENTRIES HBLKSIZE
-#	define MAP_SIZE MAP_ENTRIES
-#	define CPP_MAX_OFFSET (WORDS_TO_BYTES(OBJ_INVALID) - 1)	
-#	define MAX_OFFSET ((word)CPP_MAX_OFFSET)
+    		       /* to an object at block_start +			*/
+ 		       /* WORDS_TO_BYTES(BYTES_TO_WORDS(i) - j)		*/
+  		       /* I.e. j is a word displacement from the	*/
+  		       /* object beginning.				*/
+  		       /* The entry is OBJ_INVALID if the corresponding	*/
+  		       /* address is not a valid pointer.  It is 	*/
+  		       /* OFFSET_TOO_BIG if the value j would be too 	*/
+  		       /* large to fit in the entry.  (Note that the	*/
+  		       /* size of these entries matters, both for 	*/
+  		       /* space consumption and for cache utilization.	*/
+#   define OFFSET_TOO_BIG 0xfe
+#   define OBJ_INVALID 0xff
+#   define MAP_ENTRY(map, bytes) (map)[bytes]
+#   define MAP_ENTRIES HBLKSIZE
+#   define MAP_SIZE MAP_ENTRIES
+#   define CPP_MAX_OFFSET (OFFSET_TOO_BIG - 1)	
+#   define MAX_OFFSET ((word)CPP_MAX_OFFSET)
+    /* The following are used only if GC_all_interior_ptrs != 0 */
 # 	define VALID_OFFSET_SZ \
 	  (CPP_MAX_OFFSET > WORDS_TO_BYTES(CPP_MAXOBJSZ)? \
 	   CPP_MAX_OFFSET+1 \
@@ -1075,11 +986,11 @@ struct _GC_arrays {
   	char _valid_offsets[VALID_OFFSET_SZ];
 				/* GC_valid_offsets[i] == TRUE ==> i 	*/
 				/* is registered as a displacement.	*/
-#	define OFFSET_VALID(displ) GC_valid_offsets[displ]
+#	define OFFSET_VALID(displ) \
+	  (GC_all_interior_pointers || GC_valid_offsets[displ])
   	char _modws_valid_offsets[sizeof(word)];
 				/* GC_valid_offsets[i] ==>		  */
 				/* GC_modws_valid_offsets[i%sizeof(word)] */
-#   endif
 # ifdef STUBBORN_ALLOC
     page_hash_table _changed_pages;
         /* Stubborn object pages that were changes since last call to	*/
@@ -1106,17 +1017,25 @@ struct _GC_arrays {
 #     define MAX_HEAP_SECTS 768		/* Separately added heap sections. */
 #   endif
 # else
-#   define MAX_HEAP_SECTS 256
+#   ifdef SMALL_CONFIG
+#     define MAX_HEAP_SECTS 128		/* Roughly 1GB			*/
+#   else
+#     define MAX_HEAP_SECTS 384		/* Roughly 3GB			*/
+#   endif
 # endif
   struct HeapSect {
       ptr_t hs_start; word hs_bytes;
   } _heap_sects[MAX_HEAP_SECTS];
-# ifdef MSWIN32
+# if defined(MSWIN32) || defined(MSWINCE)
     ptr_t _heap_bases[MAX_HEAP_SECTS];
     		/* Start address of memory regions obtained from kernel. */
 # endif
+# ifdef MSWINCE
+    word _heap_lengths[MAX_HEAP_SECTS];
+    		/* Commited lengths of memory regions obtained from kernel. */
+# endif
   struct roots _static_roots[MAX_ROOT_SETS];
-# ifndef MSWIN32
+# if !defined(MSWIN32) && !defined(MSWINCE)
     struct roots * _root_index[RT_SIZE];
 # endif
   struct exclusion _excl_table[MAX_EXCLUSIONS];
@@ -1135,8 +1054,11 @@ struct _GC_arrays {
 
 GC_API GC_FAR struct _GC_arrays GC_arrays; 
 
-# define GC_objfreelist GC_arrays._objfreelist
-# define GC_aobjfreelist GC_arrays._aobjfreelist
+# ifndef SEPARATE_GLOBALS
+#   define GC_objfreelist GC_arrays._objfreelist
+#   define GC_aobjfreelist GC_arrays._aobjfreelist
+#   define GC_words_allocd GC_arrays._words_allocd
+# endif
 # define GC_uobjfreelist GC_arrays._uobjfreelist
 # ifdef ATOMIC_UNCOLLECTABLE
 #   define GC_auobjfreelist GC_arrays._auobjfreelist
@@ -1151,24 +1073,31 @@ GC_API GC_FAR struct _GC_arrays GC_arrays;
 # define GC_obj_map GC_arrays._obj_map
 # define GC_last_heap_addr GC_arrays._last_heap_addr
 # define GC_prev_heap_addr GC_arrays._prev_heap_addr
-# define GC_words_allocd GC_arrays._words_allocd
 # define GC_words_wasted GC_arrays._words_wasted
 # define GC_large_free_bytes GC_arrays._large_free_bytes
+# define GC_large_allocd_bytes GC_arrays._large_allocd_bytes
+# define GC_max_large_allocd_bytes GC_arrays._max_large_allocd_bytes
 # define GC_words_finalized GC_arrays._words_finalized
 # define GC_non_gc_bytes_at_gc GC_arrays._non_gc_bytes_at_gc
 # define GC_mem_freed GC_arrays._mem_freed
+# define GC_scratch_end_ptr GC_arrays._scratch_end_ptr
+# define GC_scratch_last_end_ptr GC_arrays._scratch_last_end_ptr
 # define GC_mark_procs GC_arrays._mark_procs
 # define GC_heapsize GC_arrays._heapsize
 # define GC_max_heapsize GC_arrays._max_heapsize
+# define GC_requested_heapsize GC_arrays._requested_heapsize
 # define GC_words_allocd_before_gc GC_arrays._words_allocd_before_gc
 # define GC_heap_sects GC_arrays._heap_sects
 # define GC_last_stack GC_arrays._last_stack
 # ifdef USE_MUNMAP
 #   define GC_unmapped_bytes GC_arrays._unmapped_bytes
 # endif
-# ifdef MSWIN32
+# if defined(MSWIN32) || defined(MSWINCE)
 #   define GC_heap_bases GC_arrays._heap_bases
 # endif
+# ifdef MSWINCE
+#   define GC_heap_lengths GC_arrays._heap_lengths
+# endif
 # define GC_static_roots GC_arrays._static_roots
 # define GC_root_index GC_arrays._root_index
 # define GC_excl_table GC_arrays._excl_table
@@ -1194,6 +1123,8 @@ GC_API GC_FAR struct _GC_arrays GC_arrays;
 # define beginGC_arrays ((ptr_t)(&GC_arrays))
 # define endGC_arrays (((ptr_t)(&GC_arrays)) + (sizeof GC_arrays))
 
+#define USED_HEAP_SIZE (GC_heapsize - GC_large_free_bytes)
+
 /* Object kinds: */
 # define MAXOBJKINDS 16
 
@@ -1213,10 +1144,27 @@ extern struct obj_kind {
    GC_bool ok_init;   /* Clear objects before putting them on the free list. */
 } GC_obj_kinds[MAXOBJKINDS];
 
-# define endGC_obj_kinds (((ptr_t)(&GC_obj_kinds)) + (sizeof GC_obj_kinds))
+# define beginGC_obj_kinds ((ptr_t)(&GC_obj_kinds))
+# define endGC_obj_kinds (beginGC_obj_kinds + (sizeof GC_obj_kinds))
 
-# define end_gc_area ((ptr_t)endGC_arrays == (ptr_t)(&GC_obj_kinds) ? \
-			endGC_obj_kinds : endGC_arrays)
+/* Variables that used to be in GC_arrays, but need to be accessed by 	*/
+/* inline allocation code.  If they were in GC_arrays, the inlined 	*/
+/* allocation code would include GC_arrays offsets (as it did), which	*/
+/* introduce maintenance problems.					*/
+
+#ifdef SEPARATE_GLOBALS
+  word GC_words_allocd;
+  	/* Number of words allocated during this collection cycle */
+  ptr_t GC_objfreelist[MAXOBJSZ+1];
+			  /* free list for NORMAL objects */
+# define beginGC_objfreelist ((ptr_t)(&GC_objfreelist))
+# define endGC_objfreelist (beginGC_objfreelist + sizeof(GC_objfreelist))
+
+  ptr_t GC_aobjfreelist[MAXOBJSZ+1];
+			  /* free list for atomic (PTRFREE) objs 	*/
+# define beginGC_aobjfreelist ((ptr_t)(&GC_aobjfreelist))
+# define endGC_aobjfreelist (beginGC_aobjfreelist + sizeof(GC_aobjfreelist))
+#endif
 
 /* Predefined kinds: */
 # define PTRFREE 0
@@ -1240,8 +1188,10 @@ extern word GC_n_heap_sects;	/* Number of separately added heap	*/
 
 extern word GC_page_size;
 
-# ifdef MSWIN32
-extern word GC_n_heap_bases;	/* See GC_heap_bases.	*/
+# if defined(MSWIN32) || defined(MSWINCE)
+  struct _SYSTEM_INFO;
+  extern struct _SYSTEM_INFO GC_sysinfo;
+  extern word GC_n_heap_bases;	/* See GC_heap_bases.	*/
 # endif
 
 extern word GC_total_stack_black_listed;
@@ -1254,7 +1204,7 @@ extern word GC_black_list_spacing;
 			/* "stack-blacklisted", i.e. that are 		*/
 			/* problematic in the interior of an object.	*/
 
-extern char * GC_invalid_map;
+extern map_entry_type * GC_invalid_map;
 			/* Pointer to the nowhere valid hblk map */
 			/* Blocks pointing to this map are free. */
 
@@ -1273,7 +1223,7 @@ extern GC_bool GC_objects_are_marked;	/* There are marked objects in  */
   extern GC_bool GC_incremental;
 			/* Using incremental/generational collection. */
 #else
-# define GC_incremental TRUE
+# define GC_incremental FALSE
 			/* Hopefully allow optimizer to remove some code. */
 #endif
 
@@ -1286,10 +1236,6 @@ extern word GC_root_size;	/* Total size of registered root sections */
 
 extern GC_bool GC_debugging_started;	/* GC_debug_malloc has been called. */ 
 
-extern ptr_t GC_least_plausible_heap_addr;
-extern ptr_t GC_greatest_plausible_heap_addr;
-			/* Bounds on the heap.  Guaranteed valid	*/
-			/* Likely to include future heap expansion.	*/
 			
 /* Operations */
 # ifndef abs
@@ -1302,6 +1248,33 @@ extern ptr_t GC_greatest_plausible_heap_addr;
 /*  with it. Only those corresponding to the beginning of an */
 /*  object are used.                                         */
 
+/* Set mark bit correctly, even if mark bits may be concurrently 	*/
+/* accessed.								*/
+#ifdef PARALLEL_MARK
+# define OR_WORD(addr, bits) \
+	{ word old; \
+	  do { \
+	    old = *((volatile word *)addr); \
+	  } while (!GC_compare_and_exchange((addr), old, old | (bits))); \
+	}
+# define OR_WORD_EXIT_IF_SET(addr, bits, exit_label) \
+	{ word old; \
+	  word my_bits = (bits); \
+	  do { \
+	    old = *((volatile word *)addr); \
+	    if (old & my_bits) goto exit_label; \
+	  } while (!GC_compare_and_exchange((addr), old, old | my_bits)); \
+	}
+#else
+# define OR_WORD(addr, bits) *(addr) |= (bits)
+# define OR_WORD_EXIT_IF_SET(addr, bits, exit_label) \
+	{ \
+	  word old = *(addr); \
+	  word my_bits = (bits); \
+	  if (old & my_bits) goto exit_label; \
+	  *(addr) = (old | my_bits); \
+	}
+#endif
 
 /* Mark bit operations */
 
@@ -1313,75 +1286,87 @@ extern ptr_t GC_greatest_plausible_heap_addr;
  * relative to the beginning of the block, including unused words)
  */
 
+#ifdef USE_MARK_BYTES
+# define mark_bit_from_hdr(hhdr,n) ((hhdr)->hb_marks[(n) >> 1])
+# define set_mark_bit_from_hdr(hhdr,n) ((hhdr)->hb_marks[(n)>>1]) = 1
+# define clear_mark_bit_from_hdr(hhdr,n) ((hhdr)->hb_marks[(n)>>1]) = 0
+#else /* !USE_MARK_BYTES */
 # define mark_bit_from_hdr(hhdr,n) (((hhdr)->hb_marks[divWORDSZ(n)] \
 			    >> (modWORDSZ(n))) & (word)1)
-# define set_mark_bit_from_hdr(hhdr,n) (hhdr)->hb_marks[divWORDSZ(n)] \
-				|= (word)1 << modWORDSZ(n)
-
+# define set_mark_bit_from_hdr(hhdr,n) \
+			    OR_WORD((hhdr)->hb_marks+divWORDSZ(n), \
+				    (word)1 << modWORDSZ(n))
 # define clear_mark_bit_from_hdr(hhdr,n) (hhdr)->hb_marks[divWORDSZ(n)] \
 				&= ~((word)1 << modWORDSZ(n))
+#endif /* !USE_MARK_BYTES */
 
 /* Important internal collector routines */
 
-ptr_t GC_approx_sp();
-
-GC_bool GC_should_collect();
-#ifdef PRESERVE_LAST
-    GC_bool GC_in_last_heap_sect(/* ptr_t */);
-	/* In last added heap section?  If so, avoid breaking up.	*/
-#endif
-void GC_apply_to_all_blocks(/*fn, client_data*/);
-			/* Invoke fn(hbp, client_data) for each 	*/
-			/* allocated heap block.			*/
-struct hblk * GC_next_used_block(/* struct hblk * h */);
-			/* Return first in-use block >= h	*/
-struct hblk * GC_prev_block(/* struct hblk * h */);
-			/* Return last block <= h.  Returned block	*/
-			/* is managed by GC, but may or may not be in	*/
+ptr_t GC_approx_sp GC_PROTO((void));
+  
+GC_bool GC_should_collect GC_PROTO((void));
+  
+void GC_apply_to_all_blocks GC_PROTO(( \
+    void (*fn) GC_PROTO((struct hblk *h, word client_data)), \
+    word client_data));
+  			/* Invoke fn(hbp, client_data) for each 	*/
+  			/* allocated heap block.			*/
+struct hblk * GC_next_used_block GC_PROTO((struct hblk * h));
+  			/* Return first in-use block >= h	*/
+struct hblk * GC_prev_block GC_PROTO((struct hblk * h));
+  			/* Return last block <= h.  Returned block	*/
+  			/* is managed by GC, but may or may not be in	*/
 			/* use.						*/
-void GC_mark_init();
-void GC_clear_marks();	/* Clear mark bits for all heap objects. */
-void GC_invalidate_mark_state();	/* Tell the marker that	marked 	   */
-					/* objects may point to	unmarked   */
-					/* ones, and roots may point to	   */
-					/* unmarked objects.		   */
-					/* Reset mark stack.		   */
-void GC_mark_from_mark_stack(); /* Mark from everything on the mark stack. */
-				/* Return after about one pages worth of   */
-				/* work.				   */
-GC_bool GC_mark_stack_empty();
-GC_bool GC_mark_some(/* cold_gc_frame */);
-			/* Perform about one pages worth of marking	*/
-			/* work of whatever kind is needed.  Returns	*/
-			/* quickly if no collection is in progress.	*/
-			/* Return TRUE if mark phase finished.		*/
-void GC_initiate_gc();		/* initiate collection.			*/
-				/* If the mark state is invalid, this	*/
-				/* becomes full colleection.  Otherwise */
-				/* it's partial.			*/
-void GC_push_all(/*b,t*/);	/* Push everything in a range 		*/
-				/* onto mark stack.			*/
-void GC_push_dirty(/*b,t*/);      /* Push all possibly changed	 	*/
-				  /* subintervals of [b,t) onto		*/
-				  /* mark stack.			*/
+void GC_mark_init GC_PROTO((void));
+void GC_clear_marks GC_PROTO((void));	/* Clear mark bits for all heap objects. */
+void GC_invalidate_mark_state GC_PROTO((void));
+					/* Tell the marker that	marked 	   */
+  					/* objects may point to	unmarked   */
+  					/* ones, and roots may point to	   */
+  					/* unmarked objects.		   */
+  					/* Reset mark stack.		   */
+GC_bool GC_mark_stack_empty GC_PROTO((void));
+GC_bool GC_mark_some GC_PROTO((ptr_t cold_gc_frame));
+  			/* Perform about one pages worth of marking	*/
+  			/* work of whatever kind is needed.  Returns	*/
+  			/* quickly if no collection is in progress.	*/
+  			/* Return TRUE if mark phase finished.		*/
+void GC_initiate_gc GC_PROTO((void));
+				/* initiate collection.			*/
+  				/* If the mark state is invalid, this	*/
+  				/* becomes full colleection.  Otherwise */
+  				/* it's partial.			*/
+void GC_push_all GC_PROTO((ptr_t bottom, ptr_t top));
+				/* Push everything in a range 		*/
+  				/* onto mark stack.			*/
+void GC_push_selected GC_PROTO(( \
+    ptr_t bottom, \
+    ptr_t top, \
+    int (*dirty_fn) GC_PROTO((struct hblk *h)), \
+    void (*push_fn) GC_PROTO((ptr_t bottom, ptr_t top)) ));
+				  /* Push all pages h in [b,t) s.t. 	*/
+				  /* select_fn(h) != 0 onto mark stack. */
 #ifndef SMALL_CONFIG
-  void GC_push_conditional(/* ptr_t b, ptr_t t, GC_bool all*/);
+  void GC_push_conditional GC_PROTO((ptr_t b, ptr_t t, GC_bool all));
 #else
 # define GC_push_conditional(b, t, all) GC_push_all(b, t)
 #endif
                                 /* Do either of the above, depending	*/
 				/* on the third arg.			*/
-void GC_push_all_stack(/*b,t*/);    /* As above, but consider		*/
+void GC_push_all_stack GC_PROTO((ptr_t b, ptr_t t));
+				    /* As above, but consider		*/
 				    /*  interior pointers as valid  	*/
-void GC_push_all_eager(/*b,t*/);    /* Same as GC_push_all_stack, but   */
+void GC_push_all_eager GC_PROTO((ptr_t b, ptr_t t));
+				    /* Same as GC_push_all_stack, but   */
 				    /* ensures that stack is scanned	*/
 				    /* immediately, not just scheduled  */
 				    /* for scanning.			*/
 #ifndef THREADS
-  void GC_push_all_stack_partially_eager(/* bottom, top, cold_gc_frame */);
+  void GC_push_all_stack_partially_eager GC_PROTO(( \
+      ptr_t bottom, ptr_t top, ptr_t cold_gc_frame ));
 			/* Similar to GC_push_all_eager, but only the	*/
 			/* part hotter than cold_gc_frame is scanned	*/
-			/* immediately.  Needed to endure that callee-	*/
+			/* immediately.  Needed to ensure that callee-	*/
 			/* save registers are not missed.		*/
 #else
   /* In the threads case, we push part of the current thread stack	*/
@@ -1390,163 +1375,240 @@ void GC_push_all_eager(/*b,t*/);    /* Same as GC_push_all_stack, but   */
   /* stacks are scheduled for scanning in *GC_push_other_roots, which	*/
   /* is thread-package-specific.					*/
 #endif
-void GC_push_current_stack(/* ptr_t cold_gc_frame */);
-			/* Push enough of the current stack eagerly to	*/
-			/* ensure that callee-save registers saved in	*/
-			/* GC frames are scanned.			*/
-			/* In the non-threads case, schedule entire	*/
-			/* stack for scanning.				*/
-void GC_push_roots(/* GC_bool all, ptr_t cold_gc_frame */);
-			/* Push all or dirty roots.	*/
-extern void (*GC_push_other_roots)();
-			/* Push system or application specific roots	*/
-			/* onto the mark stack.  In some environments	*/
-			/* (e.g. threads environments) this is		*/
-			/* predfined to be non-zero.  A client supplied */
-			/* replacement should also call the original	*/
-			/* function.					*/
-extern void (*GC_start_call_back)(/* void */);
-			/* Called at start of full collections.		*/
-			/* Not called if 0.  Called with allocation 	*/
-			/* lock held.					*/
-			/* 0 by default.				*/
-void GC_push_regs();	/* Push register contents onto mark stack.	*/
-void GC_remark();	/* Mark from all marked objects.  Used	*/
-		 	/* only if we had to drop something.	*/
-# if defined(MSWIN32)
-  void __cdecl GC_push_one();
+void GC_push_current_stack GC_PROTO((ptr_t cold_gc_frame));
+  			/* Push enough of the current stack eagerly to	*/
+  			/* ensure that callee-save registers saved in	*/
+  			/* GC frames are scanned.			*/
+  			/* In the non-threads case, schedule entire	*/
+  			/* stack for scanning.				*/
+void GC_push_roots GC_PROTO((GC_bool all, ptr_t cold_gc_frame));
+  			/* Push all or dirty roots.	*/
+extern void (*GC_push_other_roots) GC_PROTO((void));
+  			/* Push system or application specific roots	*/
+  			/* onto the mark stack.  In some environments	*/
+  			/* (e.g. threads environments) this is		*/
+  			/* predfined to be non-zero.  A client supplied */
+  			/* replacement should also call the original	*/
+  			/* function.					*/
+extern void GC_push_gc_structures GC_PROTO((void));
+			/* Push GC internal roots.  These are normally	*/
+			/* included in the static data segment, and 	*/
+			/* Thus implicitly pushed.  But we must do this	*/
+			/* explicitly if normal root processing is 	*/
+			/* disabled.  Calls the following:		*/
+	extern void GC_push_finalizer_structures GC_PROTO((void));
+	extern void GC_push_stubborn_structures GC_PROTO((void));
+#	ifdef THREADS
+	  extern void GC_push_thread_structures GC_PROTO((void));
+#	endif
+extern void (*GC_start_call_back) GC_PROTO((void));
+  			/* Called at start of full collections.		*/
+  			/* Not called if 0.  Called with allocation 	*/
+  			/* lock held.					*/
+  			/* 0 by default.				*/
+# if defined(USE_GENERIC_PUSH_REGS)
+  void GC_generic_push_regs GC_PROTO((ptr_t cold_gc_frame));
+# else
+  void GC_push_regs GC_PROTO((void));
+# endif
+			/* Push register contents onto mark stack.	*/
+  			/* If NURSERY is defined, the default push	*/
+  			/* action can be overridden with GC_push_proc	*/
+
+# ifdef NURSERY
+    extern void (*GC_push_proc)(ptr_t);
+# endif
+# if defined(MSWIN32) || defined(MSWINCE)
+  void __cdecl GC_push_one GC_PROTO((word p));
 # else
-  void GC_push_one(/*p*/);    /* If p points to an object, mark it    */
+  void GC_push_one GC_PROTO((word p));
+			      /* If p points to an object, mark it    */
                               /* and push contents on the mark stack  */
+  			      /* Pointer recognition test always      */
+  			      /* accepts interior pointers, i.e. this */
+  			      /* is appropriate for pointers found on */
+  			      /* stack.				      */
 # endif
-void GC_push_one_checked(/*p*/); /* Ditto, omits plausibility test	*/
-void GC_push_marked(/* struct hblk h, hdr * hhdr */);
+# if defined(PRINT_BLACK_LIST) || defined(KEEP_BACK_PTRS)
+  void GC_mark_and_push_stack GC_PROTO((word p, ptr_t source));
+				/* Ditto, omits plausibility test	*/
+# else
+  void GC_mark_and_push_stack GC_PROTO((word p));
+# endif
+void GC_push_marked GC_PROTO((struct hblk * h, hdr * hhdr));
 		/* Push contents of all marked objects in h onto	*/
 		/* mark stack.						*/
 #ifdef SMALL_CONFIG
 # define GC_push_next_marked_dirty(h) GC_push_next_marked(h)
 #else
-  struct hblk * GC_push_next_marked_dirty(/* h */);
+  struct hblk * GC_push_next_marked_dirty GC_PROTO((struct hblk * h));
 		/* Invoke GC_push_marked on next dirty block above h.	*/
 		/* Return a pointer just past the end of this block.	*/
 #endif /* !SMALL_CONFIG */
-struct hblk * GC_push_next_marked(/* h */);
-		/* Ditto, but also mark from clean pages.	*/
-struct hblk * GC_push_next_marked_uncollectable(/* h */);
-		/* Ditto, but mark only from uncollectable pages.	*/
-GC_bool GC_stopped_mark(); /* Stop world and mark from all roots	*/
-			/* and rescuers.			*/
-void GC_clear_hdr_marks(/* hhdr */);  /* Clear the mark bits in a header */
-void GC_set_hdr_marks(/* hhdr */);  /* Set the mark bits in a header */
-void GC_add_roots_inner();
-GC_bool GC_is_static_root(/* ptr_t p */);
-		/* Is the address p in one of the registered static	*/
+struct hblk * GC_push_next_marked GC_PROTO((struct hblk * h));
+  		/* Ditto, but also mark from clean pages.	*/
+struct hblk * GC_push_next_marked_uncollectable GC_PROTO((struct hblk * h));
+  		/* Ditto, but mark only from uncollectable pages.	*/
+GC_bool GC_stopped_mark GC_PROTO((GC_stop_func stop_func));
+ 			/* Stop world and mark from all roots	*/
+  			/* and rescuers.			*/
+void GC_clear_hdr_marks GC_PROTO((hdr * hhdr));
+				    /* Clear the mark bits in a header */
+void GC_set_hdr_marks GC_PROTO((hdr * hhdr));
+ 				    /* Set the mark bits in a header */
+void GC_set_fl_marks GC_PROTO((ptr_t p));
+				    /* Set all mark bits associated with */
+				    /* a free list.			 */
+void GC_add_roots_inner GC_PROTO((char * b, char * e, GC_bool tmp));
+GC_bool GC_is_static_root GC_PROTO((ptr_t p));
+  		/* Is the address p in one of the registered static	*/
+  		/* root sections?					*/
+# if defined(MSWIN32) || defined(_WIN32_WCE_EMULATION)
+GC_bool GC_is_tmp_root GC_PROTO((ptr_t p));
+		/* Is the address p in one of the temporary static	*/
 		/* root sections?					*/
-void GC_register_dynamic_libraries();
-		/* Add dynamic library data sections to the root set. */
-
+# endif
+void GC_register_dynamic_libraries GC_PROTO((void));
+  		/* Add dynamic library data sections to the root set. */
+  
 /* Machine dependent startup routines */
-ptr_t GC_get_stack_base();
-void GC_register_data_segments();
-
+ptr_t GC_get_stack_base GC_PROTO((void));	/* Cold end of stack */
+#ifdef IA64
+  ptr_t GC_get_register_stack_base GC_PROTO((void));
+  					/* Cold end of register stack.	*/
+#endif
+void GC_register_data_segments GC_PROTO((void));
+  
 /* Black listing: */
-void GC_bl_init(); 	
-# ifndef ALL_INTERIOR_POINTERS
-    void GC_add_to_black_list_normal(/* bits, maybe source */);
+void GC_bl_init GC_PROTO((void));
+# ifdef PRINT_BLACK_LIST
+      void GC_add_to_black_list_normal GC_PROTO((word p, ptr_t source));
 			/* Register bits as a possible future false	*/
 			/* reference from the heap or static data	*/
-#   ifdef PRINT_BLACK_LIST
 #     define GC_ADD_TO_BLACK_LIST_NORMAL(bits, source) \
-			GC_add_to_black_list_normal(bits, source)
-#   else
-#     define GC_ADD_TO_BLACK_LIST_NORMAL(bits, source) \
-			GC_add_to_black_list_normal(bits)
-#   endif
+      		if (GC_all_interior_pointers) { \
+		  GC_add_to_black_list_stack(bits, (ptr_t)(source)); \
+		} else { \
+  		  GC_add_to_black_list_normal(bits, (ptr_t)(source)); \
+		}
 # else
-#   ifdef PRINT_BLACK_LIST
-#     define GC_ADD_TO_BLACK_LIST_NORMAL(bits, source) \
-			GC_add_to_black_list_stack(bits, source)
-#   else
+      void GC_add_to_black_list_normal GC_PROTO((word p));
 #     define GC_ADD_TO_BLACK_LIST_NORMAL(bits, source) \
-			GC_add_to_black_list_stack(bits)
-#   endif
+      		if (GC_all_interior_pointers) { \
+		  GC_add_to_black_list_stack(bits); \
+		} else { \
+  		  GC_add_to_black_list_normal(bits); \
+		}
 # endif
 
-void GC_add_to_black_list_stack(/* bits, maybe source */);
-struct hblk * GC_is_black_listed(/* h, len */);
-			/* If there are likely to be false references	*/
-			/* to a block starting at h of the indicated    */
-			/* length, then return the next plausible	*/
-			/* starting location for h that might avoid	*/
-			/* these false references.			*/
-void GC_promote_black_lists();
-			/* Declare an end to a black listing phase.	*/
-void GC_unpromote_black_lists();
-			/* Approximately undo the effect of the above.	*/
-			/* This actually loses some information, but	*/
-			/* only in a reasonably safe way.		*/
-word GC_number_stack_black_listed(/*struct hblk *start, struct hblk *endp1 */);
-			/* Return the number of (stack) blacklisted	*/
-			/* blocks in the range for statistical		*/
-			/* purposes.					*/
-		 	
-ptr_t GC_scratch_alloc(/*bytes*/);
-				/* GC internal memory allocation for	*/
-				/* small objects.  Deallocation is not  */
-				/* possible.				*/
-	
+# ifdef PRINT_BLACK_LIST
+    void GC_add_to_black_list_stack GC_PROTO((word p, ptr_t source));
+# else
+    void GC_add_to_black_list_stack GC_PROTO((word p));
+# endif
+struct hblk * GC_is_black_listed GC_PROTO((struct hblk * h, word len));
+  			/* If there are likely to be false references	*/
+  			/* to a block starting at h of the indicated    */
+  			/* length, then return the next plausible	*/
+  			/* starting location for h that might avoid	*/
+  			/* these false references.			*/
+void GC_promote_black_lists GC_PROTO((void));
+  			/* Declare an end to a black listing phase.	*/
+void GC_unpromote_black_lists GC_PROTO((void));
+  			/* Approximately undo the effect of the above.	*/
+  			/* This actually loses some information, but	*/
+  			/* only in a reasonably safe way.		*/
+word GC_number_stack_black_listed GC_PROTO(( \
+	struct hblk *start, struct hblk *endp1));
+  			/* Return the number of (stack) blacklisted	*/
+  			/* blocks in the range for statistical		*/
+  			/* purposes.					*/
+  		 	
+ptr_t GC_scratch_alloc GC_PROTO((word bytes));
+  				/* GC internal memory allocation for	*/
+  				/* small objects.  Deallocation is not  */
+  				/* possible.				*/
+  	
 /* Heap block layout maps: */			
-void GC_invalidate_map(/* hdr */);
-				/* Remove the object map associated	*/
-				/* with the block.  This identifies	*/
-				/* the block as invalid to the mark	*/
-				/* routines.				*/
-GC_bool GC_add_map_entry(/*sz*/);
-				/* Add a heap block map for objects of	*/
-				/* size sz to obj_map.			*/
-				/* Return FALSE on failure.		*/
-void GC_register_displacement_inner(/*offset*/);
-				/* Version of GC_register_displacement	*/
-				/* that assumes lock is already held	*/
-				/* and signals are already disabled.	*/
-
+void GC_invalidate_map GC_PROTO((hdr * hhdr));
+  				/* Remove the object map associated	*/
+  				/* with the block.  This identifies	*/
+  				/* the block as invalid to the mark	*/
+  				/* routines.				*/
+GC_bool GC_add_map_entry GC_PROTO((word sz));
+  				/* Add a heap block map for objects of	*/
+  				/* size sz to obj_map.			*/
+  				/* Return FALSE on failure.		*/
+void GC_register_displacement_inner GC_PROTO((word offset));
+  				/* Version of GC_register_displacement	*/
+  				/* that assumes lock is already held	*/
+  				/* and signals are already disabled.	*/
+  
 /*  hblk allocation: */		
-void GC_new_hblk(/*size_in_words, kind*/);
-				/* Allocate a new heap block, and build */
-				/* a free list in it.			*/				
-struct hblk * GC_allochblk(/*size_in_words, kind*/);
-				/* Allocate a heap block, clear it if	*/
-				/* for composite objects, inform	*/
+void GC_new_hblk GC_PROTO((word size_in_words, int kind));
+  				/* Allocate a new heap block, and build */
+  				/* a free list in it.			*/				
+
+ptr_t GC_build_fl GC_PROTO((struct hblk *h, word sz,
+			   GC_bool clear,  ptr_t list));
+				/* Build a free list for objects of 	*/
+				/* size sz in block h.  Append list to	*/
+				/* end of the free lists.  Possibly	*/
+				/* clear objects on the list.  Normally	*/
+				/* called by GC_new_hblk, but also	*/
+				/* called explicitly without GC lock.	*/
+
+struct hblk * GC_allochblk GC_PROTO(( \
+	word size_in_words, int kind, unsigned flags));
+				/* Allocate a heap block, inform	*/
 				/* the marker that block is valid	*/
 				/* for objects of indicated size.	*/
-				/* sz < 0 ==> atomic.			*/ 
-void GC_freehblk();		/* Deallocate a heap block and mark it  */
-				/* as invalid.				*/
-				
+
+ptr_t GC_alloc_large GC_PROTO((word lw, int k, unsigned flags));
+			/* Allocate a large block of size lw words.	*/
+			/* The block is not cleared.			*/
+			/* Flags is 0 or IGNORE_OFF_PAGE.		*/
+			/* Calls GC_allchblk to do the actual 		*/
+			/* allocation, but also triggers GC and/or	*/
+			/* heap expansion as appropriate.		*/
+			/* Does not update GC_words_allocd, but does	*/
+			/* other accounting.				*/
+
+ptr_t GC_alloc_large_and_clear GC_PROTO((word lw, int k, unsigned flags));
+			/* As above, but clear block if appropriate	*/
+			/* for kind k.					*/
+
+void GC_freehblk GC_PROTO((struct hblk * p));
+				/* Deallocate a heap block and mark it  */
+  				/* as invalid.				*/
+  				
 /*  Misc GC: */
-void GC_init_inner();
-GC_bool GC_expand_hp_inner();
-void GC_start_reclaim(/*abort_if_found*/);
-				/* Restore unmarked objects to free	*/
-				/* lists, or (if abort_if_found is	*/
-				/* TRUE) report them.			*/
-				/* Sweeping of small object pages is	*/
-				/* largely deferred.			*/
-void GC_continue_reclaim(/*size, kind*/);
-				/* Sweep pages of the given size and	*/
-				/* kind, as long as possible, and	*/
-				/* as long as the corr. free list is    */
-				/* empty.				*/
-void GC_reclaim_or_delete_all();
-				/* Arrange for all reclaim lists to be	*/
-				/* empty.  Judiciously choose between	*/
-				/* sweeping and discarding each page.	*/
-GC_bool GC_reclaim_all(/* GC_stop_func f*/);
-				/* Reclaim all blocks.  Abort (in a	*/
-				/* consistent state) if f returns TRUE. */
-GC_bool GC_block_empty(/* hhdr */); /* Block completely unmarked? 	*/
-GC_bool GC_never_stop_func();	/* Returns FALSE.		*/
-GC_bool GC_try_to_collect_inner(/* GC_stop_func f */);
+void GC_init_inner GC_PROTO((void));
+GC_bool GC_expand_hp_inner GC_PROTO((word n));
+void GC_start_reclaim GC_PROTO((int abort_if_found));
+  				/* Restore unmarked objects to free	*/
+  				/* lists, or (if abort_if_found is	*/
+  				/* TRUE) report them.			*/
+  				/* Sweeping of small object pages is	*/
+  				/* largely deferred.			*/
+void GC_continue_reclaim GC_PROTO((word sz, int kind));
+  				/* Sweep pages of the given size and	*/
+  				/* kind, as long as possible, and	*/
+  				/* as long as the corr. free list is    */
+  				/* empty.				*/
+void GC_reclaim_or_delete_all GC_PROTO((void));
+  				/* Arrange for all reclaim lists to be	*/
+  				/* empty.  Judiciously choose between	*/
+  				/* sweeping and discarding each page.	*/
+GC_bool GC_reclaim_all GC_PROTO((GC_stop_func stop_func, GC_bool ignore_old));
+  				/* Reclaim all blocks.  Abort (in a	*/
+  				/* consistent state) if f returns TRUE. */
+GC_bool GC_block_empty GC_PROTO((hdr * hhdr));
+ 				/* Block completely unmarked? 	*/
+GC_bool GC_never_stop_func GC_PROTO((void));
+				/* Returns FALSE.		*/
+GC_bool GC_try_to_collect_inner GC_PROTO((GC_stop_func f));
+
 				/* Collect; caller must have acquired	*/
 				/* lock and disabled signals.		*/
 				/* Collection is aborted if f returns	*/
@@ -1554,87 +1616,133 @@ GC_bool GC_try_to_collect_inner(/* GC_stop_func f */);
 				/* successfully.			*/
 # define GC_gcollect_inner() \
 	(void) GC_try_to_collect_inner(GC_never_stop_func)
-void GC_finish_collection();	/* Finish collection.  Mark bits are	*/
-				/* consistent and lock is still held.	*/
-GC_bool GC_collect_or_expand(/* needed_blocks */);
-				/* Collect or expand heap in an attempt */
-				/* make the indicated number of free	*/
-				/* blocks available.  Should be called	*/
-				/* until the blocks are available or	*/
-				/* until it fails by returning FALSE.	*/
-GC_API void GC_init();		/* Initialize collector.		*/
-void GC_collect_a_little_inner(/* int n */);
-				/* Do n units worth of garbage 		*/
-				/* collection work, if appropriate.	*/
-				/* A unit is an amount appropriate for  */
-				/* HBLKSIZE bytes of allocation.	*/
-ptr_t GC_generic_malloc(/* bytes, kind */);
-				/* Allocate an object of the given	*/
-				/* kind.  By default, there are only	*/
-				/* a few kinds: composite(pointerfree), */
+void GC_finish_collection GC_PROTO((void));
+ 				/* Finish collection.  Mark bits are	*/
+  				/* consistent and lock is still held.	*/
+GC_bool GC_collect_or_expand GC_PROTO(( \
+	word needed_blocks, GC_bool ignore_off_page));
+  				/* Collect or expand heap in an attempt */
+  				/* make the indicated number of free	*/
+  				/* blocks available.  Should be called	*/
+  				/* until the blocks are available or	*/
+  				/* until it fails by returning FALSE.	*/
+GC_API void GC_init GC_PROTO((void)); /* Initialize collector.		*/
+
+#if defined(MSWIN32) || defined(MSWINCE)
+  void GC_deinit GC_PROTO((void));
+                                /* Free any resources allocated by      */
+                                /* GC_init                              */
+#endif
+
+void GC_collect_a_little_inner GC_PROTO((int n));
+  				/* Do n units worth of garbage 		*/
+  				/* collection work, if appropriate.	*/
+  				/* A unit is an amount appropriate for  */
+  				/* HBLKSIZE bytes of allocation.	*/
+ptr_t GC_generic_malloc GC_PROTO((word lb, int k));
+  				/* Allocate an object of the given	*/
+  				/* kind.  By default, there are only	*/
+  				/* a few kinds: composite(pointerfree), */
 				/* atomic, uncollectable, etc.		*/
 				/* We claim it's possible for clever	*/
 				/* client code that understands GC	*/
 				/* internals to add more, e.g. to	*/
 				/* communicate object layout info	*/
 				/* to the collector.			*/
-ptr_t GC_generic_malloc_ignore_off_page(/* bytes, kind */);
-				/* As above, but pointers past the 	*/
-				/* first page of the resulting object	*/
-				/* are ignored.				*/
-ptr_t GC_generic_malloc_inner(/* bytes, kind */);
-				/* Ditto, but I already hold lock, etc.	*/
-ptr_t GC_generic_malloc_words_small GC_PROTO((size_t words, int kind));
-				/* As above, but size in units of words */
-				/* Bypasses MERGE_SIZES.  Assumes	*/
-				/* words <= MAXOBJSZ.			*/
-ptr_t GC_generic_malloc_inner_ignore_off_page(/* bytes, kind */);
-				/* Allocate an object, where		*/
-				/* the client guarantees that there	*/
-				/* will always be a pointer to the 	*/
-				/* beginning of the object while the	*/
-				/* object is live.			*/
-ptr_t GC_allocobj(/* sz_inn_words, kind */);
-				/* Make the indicated 			*/
-				/* free list nonempty, and return its	*/
-				/* head.				*/
-
-void GC_init_headers();
-GC_bool GC_install_header(/*h*/);
-				/* Install a header for block h.	*/
-				/* Return FALSE on failure.		*/
-GC_bool GC_install_counts(/*h, sz*/);
-				/* Set up forwarding counts for block	*/
-				/* h of size sz.			*/
-				/* Return FALSE on failure.		*/
-void GC_remove_header(/*h*/);
-				/* Remove the header for block h.	*/
-void GC_remove_counts(/*h, sz*/);
-				/* Remove forwarding counts for h.	*/
-hdr * GC_find_header(/*p*/);	/* Debugging only.			*/
-
-void GC_finalize();	/* Perform all indicated finalization actions	*/
-			/* on unmarked objects.				*/
-			/* Unreachable finalizable objects are enqueued	*/
-			/* for processing by GC_invoke_finalizers.	*/
-			/* Invoked with lock.				*/
-			
-void GC_add_to_heap(/*p, bytes*/);
-			/* Add a HBLKSIZE aligned chunk to the heap.	*/
-
-void GC_print_obj(/* ptr_t p */);
-			/* P points to somewhere inside an object with	*/
-			/* debugging info.  Print a human readable	*/
-			/* description of the object to stderr.		*/
-extern void (*GC_check_heap)();
-			/* Check that all objects in the heap with 	*/
-			/* debugging info are intact.  Print 		*/
-			/* descriptions of any that are not.		*/
-extern void (*GC_print_heap_obj)(/* ptr_t p */);
-			/* If possible print s followed by a more	*/
-			/* detailed description of the object 		*/
-			/* referred to by p.				*/
-			
+ptr_t GC_generic_malloc_ignore_off_page GC_PROTO((size_t b, int k));
+  				/* As above, but pointers past the 	*/
+  				/* first page of the resulting object	*/
+  				/* are ignored.				*/
+ptr_t GC_generic_malloc_inner GC_PROTO((word lb, int k));
+  				/* Ditto, but I already hold lock, etc.	*/
+ptr_t GC_generic_malloc_words_small GC_PROTO((size_t lw, int k));
+  				/* As above, but size in units of words */
+  				/* Bypasses MERGE_SIZES.  Assumes	*/
+  				/* words <= MAXOBJSZ.			*/
+ptr_t GC_generic_malloc_inner_ignore_off_page GC_PROTO((size_t lb, int k));
+  				/* Allocate an object, where		*/
+  				/* the client guarantees that there	*/
+  				/* will always be a pointer to the 	*/
+  				/* beginning of the object while the	*/
+  				/* object is live.			*/
+ptr_t GC_allocobj GC_PROTO((word sz, int kind));
+  				/* Make the indicated 			*/
+  				/* free list nonempty, and return its	*/
+  				/* head.				*/
+  
+void GC_init_headers GC_PROTO((void));
+struct hblkhdr * GC_install_header GC_PROTO((struct hblk *h));
+  				/* Install a header for block h.	*/
+  				/* Return 0 on failure, or the header	*/
+  				/* otherwise.				*/
+GC_bool GC_install_counts GC_PROTO((struct hblk * h, word sz));
+  				/* Set up forwarding counts for block	*/
+  				/* h of size sz.			*/
+  				/* Return FALSE on failure.		*/
+void GC_remove_header GC_PROTO((struct hblk * h));
+  				/* Remove the header for block h.	*/
+void GC_remove_counts GC_PROTO((struct hblk * h, word sz));
+  				/* Remove forwarding counts for h.	*/
+hdr * GC_find_header GC_PROTO((ptr_t h)); /* Debugging only.		*/
+  
+void GC_finalize GC_PROTO((void));
+ 			/* Perform all indicated finalization actions	*/
+  			/* on unmarked objects.				*/
+  			/* Unreachable finalizable objects are enqueued	*/
+  			/* for processing by GC_invoke_finalizers.	*/
+  			/* Invoked with lock.				*/
+
+void GC_notify_or_invoke_finalizers GC_PROTO((void));
+			/* If GC_finalize_on_demand is not set, invoke	*/
+			/* eligible finalizers. Otherwise:		*/
+			/* Call *GC_finalizer_notifier if there are	*/
+			/* finalizers to be run, and we haven't called	*/
+			/* this procedure yet this GC cycle.		*/
+  			
+void GC_add_to_heap GC_PROTO((struct hblk *p, word bytes));
+  			/* Add a HBLKSIZE aligned chunk to the heap.	*/
+  
+void GC_print_obj GC_PROTO((ptr_t p));
+  			/* P points to somewhere inside an object with	*/
+  			/* debugging info.  Print a human readable	*/
+  			/* description of the object to stderr.		*/
+extern void (*GC_check_heap) GC_PROTO((void));
+  			/* Check that all objects in the heap with 	*/
+  			/* debugging info are intact.  Print 		*/
+  			/* descriptions of any that are not.		*/
+extern void (*GC_print_heap_obj) GC_PROTO((ptr_t p));
+  			/* If possible print s followed by a more	*/
+  			/* detailed description of the object 		*/
+  			/* referred to by p.				*/
+
+extern GC_bool GC_print_stats;	/* Produce at least some logging output	*/
+				/* Set from environment variable.	*/
+
+/* Macros used for collector internal allocation.	*/
+/* These assume the collector lock is held.		*/
+#ifdef DBG_HDRS_ALL
+    extern GC_PTR GC_debug_generic_malloc_inner(size_t lb, int k);
+    extern GC_PTR GC_debug_generic_malloc_inner_ignore_off_page(size_t lb,
+								int k);
+#   define GC_INTERNAL_MALLOC GC_debug_generic_malloc_inner
+#   define GC_INTERNAL_MALLOC_IGNORE_OFF_PAGE \
+		 GC_debug_generic_malloc_inner_ignore_off_page
+#   ifdef THREADS
+#       define GC_INTERNAL_FREE GC_debug_free_inner
+#   else
+#       define GC_INTERNAL_FREE GC_debug_free
+#   endif
+#else
+#   define GC_INTERNAL_MALLOC GC_generic_malloc_inner
+#   define GC_INTERNAL_MALLOC_IGNORE_OFF_PAGE \
+		 GC_generic_malloc_inner_ignore_off_page
+#   ifdef THREADS
+#       define GC_INTERNAL_FREE GC_free_inner
+#   else
+#       define GC_INTERNAL_FREE GC_free
+#   endif
+#endif
+
 /* Memory unmapping: */
 #ifdef USE_MUNMAP
   void GC_unmap_old(void);
@@ -1646,35 +1754,38 @@ extern void (*GC_print_heap_obj)(/* ptr_t p */);
 
 /* Virtual dirty bit implementation:		*/
 /* Each implementation exports the following:	*/
-void GC_read_dirty();	/* Retrieve dirty bits.	*/
-GC_bool GC_page_was_dirty(/* struct hblk * h  */);
-			/* Read retrieved dirty bits.	*/
-GC_bool GC_page_was_ever_dirty(/* struct hblk * h  */);
-			/* Could the page contain valid heap pointers?	*/
-void GC_is_fresh(/* struct hblk * h, word number_of_blocks  */);
-			/* Assert the region currently contains no	*/
-			/* valid pointers.				*/
-void GC_write_hint(/* struct hblk * h  */);
-			/* h is about to be written.	*/
-void GC_dirty_init();
-
+void GC_read_dirty GC_PROTO((void));
+			/* Retrieve dirty bits.	*/
+GC_bool GC_page_was_dirty GC_PROTO((struct hblk *h));
+  			/* Read retrieved dirty bits.	*/
+GC_bool GC_page_was_ever_dirty GC_PROTO((struct hblk *h));
+  			/* Could the page contain valid heap pointers?	*/
+void GC_is_fresh GC_PROTO((struct hblk *h, word n));
+  			/* Assert the region currently contains no	*/
+  			/* valid pointers.				*/
+void GC_write_hint GC_PROTO((struct hblk *h));
+  			/* h is about to be written.	*/
+void GC_dirty_init GC_PROTO((void));
+  
 /* Slow/general mark bit manipulation: */
-GC_API GC_bool GC_is_marked();
-void GC_clear_mark_bit();
-void GC_set_mark_bit();
-
+GC_API GC_bool GC_is_marked GC_PROTO((ptr_t p));
+void GC_clear_mark_bit GC_PROTO((ptr_t p));
+void GC_set_mark_bit GC_PROTO((ptr_t p));
+  
 /* Stubborn objects: */
-void GC_read_changed();	/* Analogous to GC_read_dirty */
-GC_bool GC_page_was_changed(/* h */);	/* Analogous to GC_page_was_dirty */
-void GC_clean_changing_list();	/* Collect obsolete changing list entries */
-void GC_stubborn_init();
-
+void GC_read_changed GC_PROTO((void));	/* Analogous to GC_read_dirty */
+GC_bool GC_page_was_changed GC_PROTO((struct hblk * h));
+ 				/* Analogous to GC_page_was_dirty */
+void GC_clean_changing_list GC_PROTO((void));
+ 				/* Collect obsolete changing list entries */
+void GC_stubborn_init GC_PROTO((void));
+  
 /* Debugging print routines: */
-void GC_print_block_list();
-void GC_print_hblkfreelist();
-void GC_print_heap_sects();
-void GC_print_static_roots();
-void GC_dump();
+void GC_print_block_list GC_PROTO((void));
+void GC_print_hblkfreelist GC_PROTO((void));
+void GC_print_heap_sects GC_PROTO((void));
+void GC_print_static_roots GC_PROTO((void));
+void GC_dump GC_PROTO((void));
 
 #ifdef KEEP_BACK_PTRS
    void GC_store_back_pointer(ptr_t source, ptr_t dest);
@@ -1688,15 +1799,19 @@ void GC_dump();
 
 /* Make arguments appear live to compiler */
 # ifdef __WATCOMC__
-  void GC_noop(void*, ...);
+    void GC_noop(void*, ...);
 # else
-  GC_API void GC_noop();
+#   ifdef __DMC__
+      GC_API void GC_noop(...);
+#   else
+      GC_API void GC_noop();
+#   endif
 # endif
 
-void GC_noop1(/* word arg */);
+void GC_noop1 GC_PROTO((word));
 
 /* Logging and diagnostic output: 	*/
-GC_API void GC_printf GC_PROTO((char * format, long, long, long, long, long, long));
+GC_API void GC_printf GC_PROTO((GC_CONST char * format, long, long, long, long, long, long));
 			/* A version of printf that doesn't allocate,	*/
 			/* is restricted to long arguments, and		*/
 			/* (unfortunately) doesn't use varargs for	*/
@@ -1715,7 +1830,7 @@ GC_API void GC_printf GC_PROTO((char * format, long, long, long, long, long, lon
 # define GC_printf6(f,a,b,c,d,e,g) GC_printf(f, (long)a, (long)b, (long)c, \
 						(long)d, (long)e, (long)g)
 
-void GC_err_printf(/* format, a, b, c, d, e, f */);
+GC_API void GC_err_printf GC_PROTO((GC_CONST char * format, long, long, long, long, long, long));
 # define GC_err_printf0(f) GC_err_puts(f)
 # define GC_err_printf1(f,a) GC_err_printf(f, (long)a, 0l, 0l, 0l, 0l, 0l)
 # define GC_err_printf2(f,a,b) GC_err_printf(f, (long)a, (long)b, 0l, 0l, 0l, 0l)
@@ -1731,18 +1846,86 @@ void GC_err_printf(/* format, a, b, c, d, e, f */);
 							(long)e, (long)g)
 			/* Ditto, writes to stderr.			*/
 			
-void GC_err_puts(/* char *s */);
+void GC_err_puts GC_PROTO((GC_CONST char *s));
 			/* Write s to stderr, don't buffer, don't add	*/
 			/* newlines, don't ...				*/
 
+#if defined(LINUX) && !defined(SMALL_CONFIG)
+  void GC_err_write GC_PROTO((GC_CONST char *buf, size_t len));
+ 			/* Write buf to stderr, don't buffer, don't add	*/
+  			/* newlines, don't ...				*/
+#endif
 
-#   ifdef GC_ASSERTIONS
+
+# ifdef GC_ASSERTIONS
 #	define GC_ASSERT(expr) if(!(expr)) {\
 		GC_err_printf2("Assertion failure: %s:%ld\n", \
 				__FILE__, (unsigned long)__LINE__); \
 		ABORT("assertion failure"); }
-#   else 
+# else 
 #	define GC_ASSERT(expr)
+# endif
+
+# if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
+    /* We need additional synchronization facilities from the thread	*/
+    /* support.  We believe these are less performance critical		*/
+    /* than the main garbage collector lock; standard pthreads-based	*/
+    /* implementations should be sufficient.				*/
+
+    /* The mark lock and condition variable.  If the GC lock is also 	*/
+    /* acquired, the GC lock must be acquired first.  The mark lock is	*/
+    /* used to both protect some variables used by the parallel		*/
+    /* marker, and to protect GC_fl_builder_count, below.		*/
+    /* GC_notify_all_marker() is called when				*/ 
+    /* the state of the parallel marker changes				*/
+    /* in some significant way (see gc_mark.h for details).  The	*/
+    /* latter set of events includes incrementing GC_mark_no.		*/
+    /* GC_notify_all_builder() is called when GC_fl_builder_count	*/
+    /* reaches 0.							*/
+
+     extern void GC_acquire_mark_lock();
+     extern void GC_release_mark_lock();
+     extern void GC_notify_all_builder();
+     /* extern void GC_wait_builder(); */
+     extern void GC_wait_for_reclaim();
+
+     extern word GC_fl_builder_count;	/* Protected by mark lock.	*/
+# endif /* PARALLEL_MARK || THREAD_LOCAL_ALLOC */
+# ifdef PARALLEL_MARK
+     extern void GC_notify_all_marker();
+     extern void GC_wait_marker();
+     extern word GC_mark_no;		/* Protected by mark lock.	*/
+
+     extern void GC_help_marker(word my_mark_no);
+		/* Try to help out parallel marker for mark cycle 	*/
+		/* my_mark_no.  Returns if the mark cycle finishes or	*/
+		/* was already done, or there was nothing to do for	*/
+		/* some other reason.					*/
+# endif /* PARALLEL_MARK */
+
+# if defined(LINUX_THREADS) || defined(IRIX_THREADS) \
+     || defined(HPUX_THREADS) || defined(OSF1_THREADS)
+  /* We define the thread suspension signal here, so that we can refer	*/
+  /* to it in the dirty bit implementation, if necessary.  Ideally we	*/
+  /* would allocate a (real-time ?) signal using the standard mechanism.*/
+  /* unfortunately, there is no standard mechanism.  (There is one 	*/
+  /* in Linux glibc, but it's not exported.)  Thus we continue to use	*/
+  /* the same hard-coded signals we've always used.			*/
+#  if !defined(SIG_SUSPEND)
+#   if defined(LINUX_THREADS)
+#    if defined(SPARC) && !defined(SIGPWR)
+       /* SPARC/Linux doesn't properly define SIGPWR in <signal.h>.
+        * It is aliased to SIGLOST in asm/signal.h, though.		*/
+#      define SIG_SUSPEND SIGLOST
+#    else
+       /* Linuxthreads uses SIGUSR1 and SIGUSR2.			*/
+#      define SIG_SUSPEND SIGPWR
+#    endif
+#   else  /* !LINUX_THREADS */
+#    define SIG_SUSPEND _SIGRTMIN + 6
 #   endif
+#  endif /* !SIG_SUSPEND */
+  
+# endif
 
 # endif /* GC_PRIVATE_H */