Two related changes for accurate GC with the MLDS->C back-end:

Estimated hours taken: 2
Branches: main

Two related changes for accurate GC with the MLDS->C back-end:
- Add a new runtime option --heap-margin-size for the
  threshold of free heap space before a collection is
  invoked, rather than the previous hack of reusing
  --heap-zone-size for that purpose.
- A minor optimization to reduce the cost of MR_gc_check():
  cache the value of MR_zone_end - MR_heap_margin_size.

runtime/mercury_wrapper.h:
runtime/mercury_wrapper.c:
runtime/mercury_memory.c:
	Add new variable `MR_heap_margin_size',
	and add a new runtime option `--heap-margin-size' to set it.

runtime/mercury_memory_zones.h:
	Add new field `gc_threshold' to the MR_MemoryZone struct.

runtime/mercury_memory_zones.c:
	Initialize the new field to point to the zone end minus
	MR_heap_margin_size.

runtime/mercury.h:
	Change MR_GC_check() to use the `gc_threshold' field.

runtime/mercury.h

@@ -514,8 +514,8 @@ MR_Box MR_asm_box_float(MR_Float f);
 */
 #define MR_GC_check()							\
 	do {								\
-		if ((char *) MR_hp + MR_heap_zone_size >=		\
-		    (char *) MR_ENGINE(MR_eng_heap_zone)->MR_zone_end)	\
+		if ((char *) MR_hp >=					\
+		    MR_ENGINE(MR_eng_heap_zone)->gc_threshold)		\
 		{							\
 			MR_save_registers();				\
 			MR_garbage_collect();				\

runtime/mercury_memory.c

@@ -137,6 +137,7 @@ MR_init_memory(void)
 	MR_global_heap_zone_size = 0;
 	MR_debug_heap_size	 = 0;
 	MR_debug_heap_zone_size	 = 0;
+	MR_heap_margin_size	 = 0;
 #else
 	MR_heap_size		 = MR_round_up(MR_heap_size * 1024,
 					MR_unit);
@@ -155,6 +156,8 @@ MR_init_memory(void)
 					MR_unit);
 	MR_debug_heap_zone_size	 = MR_round_up(MR_debug_heap_zone_size * 1024,
 					MR_unit);
+	/* Note that there's no need for the heap margin to be rounded up */
+	MR_heap_margin_size	 = MR_heap_margin_size * 1024;
 #endif
 	MR_detstack_size	 = MR_round_up(MR_detstack_size * 1024,
 					MR_unit);

runtime/mercury_memory_zones.c

@@ -1,5 +1,5 @@
 /*
-** Copyright (C) 1998-2000 The University of Melbourne.
+** Copyright (C) 1998-2000, 2002 The University of Melbourne.
 ** This file may only be copied under the terms of the GNU Library General
 ** Public License - see the file COPYING.LIB in the Mercury distribution.
 */
@@ -424,6 +424,9 @@ MR_construct_zone(const char *name, int id, MR_Word *base,
 	}
 #endif	/* MR_PROTECTPAGE */
 
+#if defined(NATIVE_GC) && defined(MR_HIGHLEVEL_CODE)
+	zone->gc_threshold = (char *) zone->MR_zone_end - MR_heap_margin_size;
+#endif
 
 	return zone;
 } /* end MR_construct_zone() */

runtime/mercury_memory_zones.h

@@ -1,5 +1,5 @@
 /*
-** Copyright (C) 1998-2001 University of Melbourne.
+** Copyright (C) 1998-2002 University of Melbourne.
 ** This file may only be copied under the terms of the GNU Library General
 ** Public License - see the file COPYING.LIB in the Mercury distribution.
 */
@@ -125,6 +125,15 @@ struct MR_MemoryZone_Struct {
 #else
 	#define MR_zone_end	top
 #endif
+
+#if defined(NATIVE_GC) && defined(MR_HIGHLEVEL_CODE)
+	/*
+	** This field, which is only used for heap zones,
+	** points to MR_heap_margin_size bytes before MR_zone_end.
+	** It is used to decide when to do garbage collection.
+	*/
+	char *gc_threshold;	/* == MR_zone_end - MR_heap_margin_size */
+#endif
 };
 
 /*

runtime/mercury_wrapper.c

@@ -82,6 +82,18 @@ size_t		MR_debug_heap_zone_size =	  16;
 size_t		MR_generatorstack_zone_size =	  16;
 size_t		MR_cutstack_zone_size =		  16;
 
+/*
+** MR_heap_margin_size is used for accurate GC with the MLDS->C back-end.
+** It is used to decide when to actually do a garbage collection.
+** At each call to MR_GC_check(), which is normally done on entry
+** to each C function, we check whether there is less than this
+** amount of heap space still available, and if not, we call
+** MR_garbage_collect().
+** Like the sizes above, it is measured in kilobytes
+** (but we later multiply by 1024 to convert to bytes).
+*/
+size_t		MR_heap_margin_size =		  16;
+
 /* primary cache size to optimize for, in bytes */
 size_t		MR_pcache_size =	        8192;
 
@@ -735,6 +747,7 @@ enum MR_long_option {
 	MR_NONDETSTACK_REDZONE_SIZE,
 	MR_SOLUTIONS_HEAP_REDZONE_SIZE,
 	MR_TRAIL_REDZONE_SIZE,
+	MR_HEAP_MARGIN_SIZE,
 	MR_MDB_TTY,
 	MR_MDB_IN,
 	MR_MDB_OUT,
@@ -754,6 +767,7 @@ struct MR_option MR_long_opts[] = {
 	{ "nondetstack-redzone-size", 	1, 0, MR_NONDETSTACK_REDZONE_SIZE },
 	{ "solutions-heap-redzone-size",1, 0, MR_SOLUTIONS_HEAP_REDZONE_SIZE },
 	{ "trail-redzone-size", 	1, 0, MR_TRAIL_REDZONE_SIZE },
+	{ "heap-margin-size", 		1, 0, MR_HEAP_MARGIN_SIZE },
 	{ "mdb-tty", 			1, 0, MR_MDB_TTY },
 	{ "mdb-in", 			1, 0, MR_MDB_IN },
 	{ "mdb-out", 			1, 0, MR_MDB_OUT },
@@ -845,6 +859,13 @@ process_options(int argc, char **argv)
 			MR_trail_zone_size = size;
 			break;
 
+		case MR_HEAP_MARGIN_SIZE:
+			if (sscanf(MR_optarg, "%lu", &size) != 1)
+				usage();
+
+			MR_heap_margin_size = size;
+			break;
+
 		case 'i':
 		case MR_MDB_IN:
 			MR_mdb_in_filename = MR_copy_string(MR_optarg);

runtime/mercury_wrapper.h

@@ -209,6 +209,9 @@ extern	size_t		MR_debug_heap_zone_size;
 extern	size_t		MR_generatorstack_zone_size;
 extern	size_t		MR_cutstack_zone_size;
 
+/* heap margin for MLDS->C accurate GC (documented in mercury_wrapper.c) */
+extern	size_t		MR_heap_margin_size;
+
 /* file names for the mdb debugging streams */
 extern	const char	*MR_mdb_in_filename;
 extern	const char	*MR_mdb_out_filename;