[ruby-dev:38423] longlife gc
From:
Narihiro Nakamura <authornari@...>
Date:
2009-05-10 15:20:38 UTC
List:
ruby-dev #38423
nariと申します． GW中にGCの改良を行いましたので，報告いたします． = Railsのヒープ調査 2ヶ月ほど前に，ふと気になってRailsがどんな風にヒープを使用しているか調査してみました． 以下のグラフはscaffoldしたRailsアプリを動かした際のものです． http://www.narihiro.info/resource/image/gc_count_statistic_from_rails.png 調査前は漠然と「文字列が一番多いんじゃないかなぁ」と考えていたのですが，実際に は構文木オブジェクトがヒープの半分を占めているようです． その構文木オブジェクトを更に詳しく調べたグラフが以下です． http://www.narihiro.info/resource/image/gc_count_statistic_for_node_from_rails.png このグラフにあるNODE_METHODやNODE_CFUNC，NODE_FBODYはメソッド定義の際に 使用され，NODE_WHILEやNODE_BLOCKはVMの内部で使用されているようです． = 長寿命GC 更に調査をすると，これらのオブジェクトはとても長生きな事が分かりました． そこで，これらを長寿命領域に隔離してしまおう，という改善を今回行いました． 部分的な世代別GCというイメージです． このアイデアは，まつもとさんからアドバイスしていただいたものです． = 実装メモ major gcを行うタイミングはヒープが拡張される時とGC.startの時にしています． ライトバリアの手法にはremembered_setを使いました．オブジェクトがライト バリアにあることを示す為に FL_RESERVED フラグを使用しています． = ベンチマーク scaffoldしただけのRailsアプリでGCのプロファイルを取りました． GC回数 改善前 : 35 改善後 : 31 総GC時間(ms) 改善前 : 332.02 改善後 : 284.01 平均GC時間(ms) 改善前 : 9.48 改善後 : 9.16 また，Railsアプリのヒープの状態を作り出すようなベンチマークコードを書い て，そのプロファイルも取りました． GC回数 改善前 : 7 改善後 : 6 総GC時間(ms) 改善前 : 1452.09 改善後 : 1208.07 平均GC時間(ms) 改善前 : 207.44 改善後 : 201.34 使用したコードは本メールに添付しております． = パッチ 現在の最新のrevisionへのパッチとruby1.9.1-p0へのパッチを作成しました． 本メールに添付しております． よろしくお願いいたします． -- Narihiro Nakamura
Attachments (3)
gc_partial_longlife_1_9_0_p0.patch
(26 KB, text/x-diff)
diff --git a/class.c b/class.c
index 89beccc..6e2af8b 100644
--- a/class.c
+++ b/class.c
@@ -88,11 +88,13 @@ clone_method(ID mid, NODE *body, struct clone_method_data *data)
}
st_insert(data->tbl, mid,
(st_data_t)
- NEW_FBODY(
- NEW_METHOD(fbody,
- data->klass, /* TODO */
- body->nd_body->nd_noex),
- 0));
+ NEW_NODE_LONGLIFE(
+ NODE_FBODY,
+ 0,
+ NEW_NODE_LONGLIFE(NODE_METHOD,
+ rb_gc_write_barrier(data->klass), /* TODO */
+ rb_gc_write_barrier(fbody),
+ body->nd_body->nd_noex), 0));
}
return ST_CONTINUE;
}
@@ -807,25 +809,25 @@ rb_obj_singleton_methods(int argc, VALUE *argv, VALUE obj)
void
rb_define_method_id(VALUE klass, ID name, VALUE (*func)(ANYARGS), int argc)
{
- rb_add_method(klass, name, NEW_CFUNC(func,argc), NOEX_PUBLIC);
+ rb_add_method(klass, name, NEW_NODE_LONGLIFE(NODE_CFUNC, func, argc, 0), NOEX_PUBLIC);
}
void
rb_define_method(VALUE klass, const char *name, VALUE (*func)(ANYARGS), int argc)
{
- rb_add_method(klass, rb_intern(name), NEW_CFUNC(func, argc), NOEX_PUBLIC);
+ rb_add_method(klass, rb_intern(name), NEW_NODE_LONGLIFE(NODE_CFUNC, func, argc, 0), NOEX_PUBLIC);
}
void
rb_define_protected_method(VALUE klass, const char *name, VALUE (*func)(ANYARGS), int argc)
{
- rb_add_method(klass, rb_intern(name), NEW_CFUNC(func, argc), NOEX_PROTECTED);
+ rb_add_method(klass, rb_intern(name), NEW_NODE_LONGLIFE(NODE_CFUNC, func, argc, 0), NOEX_PROTECTED);
}
void
rb_define_private_method(VALUE klass, const char *name, VALUE (*func)(ANYARGS), int argc)
{
- rb_add_method(klass, rb_intern(name), NEW_CFUNC(func, argc), NOEX_PRIVATE);
+ rb_add_method(klass, rb_intern(name), NEW_NODE_LONGLIFE(NODE_CFUNC, func, argc, 0), NOEX_PRIVATE);
}
void
diff --git a/debug.c b/debug.c
index 265e864..c677560 100644
--- a/debug.c
+++ b/debug.c
@@ -32,7 +32,7 @@ static const union {
RUBY_ENC_CODERANGE_VALID = ENC_CODERANGE_VALID,
RUBY_ENC_CODERANGE_BROKEN = ENC_CODERANGE_BROKEN,
RUBY_FL_MARK = FL_MARK,
- RUBY_FL_RESERVED = FL_RESERVED,
+ RUBY_FL_REMENBERED_SET = FL_REMEMBERED_SET,
RUBY_FL_FINALIZE = FL_FINALIZE,
RUBY_FL_TAINT = FL_TAINT,
RUBY_FL_UNTRUSTED = FL_UNTRUSTED,
diff --git a/gc.c b/gc.c
index 784949c..3abc7c1 100644
--- a/gc.c
+++ b/gc.c
@@ -22,6 +22,7 @@
#include <stdio.h>
#include <setjmp.h>
#include <sys/types.h>
+#include <assert.h>
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
@@ -93,13 +94,18 @@ typedef struct gc_profile_record {
double gc_mark_time;
double gc_sweep_time;
double gc_invoke_time;
+
size_t heap_use_slots;
+ size_t heap_longlife_use_slots;
size_t heap_live_objects;
size_t heap_free_objects;
size_t heap_total_objects;
size_t heap_use_size;
size_t heap_total_size;
+
int have_finalize;
+ int longlife_collection;
+
size_t allocate_increase;
size_t allocate_limit;
} gc_profile_record;
@@ -156,6 +162,7 @@ getrusage_time(void)
MEMZERO(&objspace->profile.record[count], gc_profile_record, 1);\
gc_time = getrusage_time();\
objspace->profile.record[count].gc_invoke_time = gc_time - objspace->profile.invoke_time;\
+ objspace->profile.record[count].longlife_collection = objspace->flags.longlife_collection;\
}\
} while(0)
@@ -210,15 +217,15 @@ getrusage_time(void)
if (objspace->profile.run) {\
size_t count = objspace->profile.count;\
objspace->profile.record[count].heap_use_slots = heaps_used;\
- objspace->profile.record[count].heap_live_objects = live;\
- objspace->profile.record[count].heap_free_objects = freed;\
+ objspace->profile.record[count].heap_longlife_use_slots = objspace->heap.longlife_used;\
+ objspace->profile.record[count].heap_live_objects = live + objspace->profile.longlife_objects;\
+ objspace->profile.record[count].heap_free_objects = freed + (objspace->heap.longlife_used * HEAP_OBJ_LIMIT - objspace->profile.longlife_objects); \
objspace->profile.record[count].heap_total_objects = heaps_used * HEAP_OBJ_LIMIT;\
objspace->profile.record[count].have_finalize = final_list ? Qtrue : Qfalse;\
- objspace->profile.record[count].heap_use_size = live * sizeof(RVALUE);\
+ objspace->profile.record[count].heap_use_size = (live + objspace->profile.longlife_objects) * sizeof(RVALUE); \
objspace->profile.record[count].heap_total_size = heaps_used * (HEAP_OBJ_LIMIT * sizeof(RVALUE));\
}\
} while(0)
-
#else
#define INIT_GC_PROF_PARAMS double gc_time = 0;\
size_t count = objspace->profile.count
@@ -231,7 +238,7 @@ getrusage_time(void)
if (objspace->profile.run) {\
size_t count = objspace->profile.count;\
objspace->profile.record[count].heap_total_objects = heaps_used * HEAP_OBJ_LIMIT;\
- objspace->profile.record[count].heap_use_size = live * sizeof(RVALUE);\
+ objspace->profile.record[count].heap_use_size = (live + objspace->profile.longlife_objects) * sizeof(RVALUE); \
objspace->profile.record[count].heap_total_size = heaps_used * HEAP_SIZE;\
}\
} while(0)
@@ -275,10 +282,16 @@ typedef struct RVALUE {
#pragma pack(pop)
#endif
+enum lifetime {
+ lifetime_normal,
+ lifetime_longlife
+};
+
struct heaps_slot {
void *membase;
RVALUE *slot;
int limit;
+ enum lifetime lifetime;
};
#define HEAP_MIN_SLOTS 10000
@@ -291,6 +304,11 @@ struct gc_list {
#define CALC_EXACT_MALLOC_SIZE 0
+typedef struct remembered_set {
+ RVALUE *obj;
+ struct remembered_set *next;
+} remembered_set_t;
+
typedef struct rb_objspace {
struct {
size_t limit;
@@ -305,13 +323,20 @@ typedef struct rb_objspace {
struct heaps_slot *ptr;
size_t length;
size_t used;
+ size_t longlife_used;
RVALUE *freelist;
+ RVALUE *longlife_freelist;
RVALUE *range[2];
RVALUE *freed;
} heap;
struct {
+ remembered_set_t *ptr;
+ remembered_set_t *freed;
+ } remembered_set;
+ struct {
int dont_gc;
int during_gc;
+ int longlife_collection;
} flags;
struct {
st_table *table;
@@ -327,6 +352,7 @@ typedef struct rb_objspace {
gc_profile_record *record;
size_t count;
size_t size;
+ size_t longlife_objects;
double invoke_time;
} profile;
struct gc_list *global_list;
@@ -395,6 +421,7 @@ rb_objspace_alloc(void)
/*#define HEAP_SIZE 0x800 */
#define HEAP_OBJ_LIMIT (HEAP_SIZE / sizeof(struct RVALUE))
+#define NORMAL_HEAPS_USED (objspace->heap.used - objspace->heap.longlife_used)
extern VALUE rb_cMutex;
extern st_table *rb_class_tbl;
@@ -858,7 +885,7 @@ allocate_heaps(rb_objspace_t *objspace, size_t next_heaps_length)
}
static void
-assign_heap_slot(rb_objspace_t *objspace)
+assign_heap_slot(rb_objspace_t *objspace, RVALUE **list, enum lifetime lifetime)
{
RVALUE *p, *pend, *membase;
size_t hi, lo, mid;
@@ -902,15 +929,17 @@ assign_heap_slot(rb_objspace_t *objspace)
heaps[hi].membase = membase;
heaps[hi].slot = p;
heaps[hi].limit = objs;
+ heaps[hi].lifetime = lifetime;
pend = p + objs;
if (lomem == 0 || lomem > p) lomem = p;
if (himem < pend) himem = pend;
+ if (lifetime == lifetime_longlife) objspace->heap.longlife_used++;
heaps_used++;
while (p < pend) {
p->as.free.flags = 0;
- p->as.free.next = freelist;
- freelist = p;
+ p->as.free.next = *list;
+ *list = p;
p++;
}
}
@@ -927,7 +956,7 @@ init_heap(rb_objspace_t *objspace)
}
for (i = 0; i < add; i++) {
- assign_heap_slot(objspace);
+ assign_heap_slot(objspace, &freelist, lifetime_normal);
}
heaps_inc = 0;
objspace->profile.invoke_time = getrusage_time();
@@ -949,16 +978,45 @@ static int
heaps_increment(rb_objspace_t *objspace)
{
if (heaps_inc > 0) {
- assign_heap_slot(objspace);
+ assign_heap_slot(objspace, &freelist, lifetime_normal);
heaps_inc--;
return Qtrue;
}
return Qfalse;
}
+#define LONGLIFE_ALLOCATE_HEAPS_MIN 10
+
+static void
+add_longlife_heaps_slot(rb_objspace_t *objspace)
+{
+ if ((heaps_used + heaps_inc) >= heaps_length) {
+ allocate_heaps(objspace, (heaps_length + LONGLIFE_ALLOCATE_HEAPS_MIN));
+ }
+ assign_heap_slot(objspace, &objspace->heap.longlife_freelist, lifetime_longlife);
+}
+
#define RANY(o) ((RVALUE*)(o))
static VALUE
+rb_newobj_from_longlife_heap(rb_objspace_t *objspace)
+{
+ VALUE obj;
+ if (!objspace->heap.longlife_freelist) {
+ add_longlife_heaps_slot(objspace);
+ }
+
+ obj = (VALUE)objspace->heap.longlife_freelist;
+ objspace->heap.longlife_freelist = objspace->heap.longlife_freelist->as.free.next;
+
+ MEMZERO((void*)obj, RVALUE, 1);
+ FL_SET(RANY(obj), FL_MARK);
+ objspace->profile.longlife_objects++;
+
+ return obj;
+}
+
+static VALUE
rb_newobj_from_heap(rb_objspace_t *objspace)
{
VALUE obj;
@@ -1051,6 +1109,22 @@ rb_newobj(void)
#endif
}
+VALUE
+rb_newobj_longlife(void)
+{
+#if defined(ENABLE_VM_OBJSPACE) && ENABLE_VM_OBJSPACE
+ rb_objspace_t *objspace = th->vm->objspace;
+#else
+ rb_objspace_t *objspace = &rb_objspace;
+#endif
+ if (during_gc) {
+ dont_gc = 1;
+ during_gc = 0;
+ rb_bug("object allocation during garbage collection phase");
+ }
+ return rb_newobj_from_longlife_heap(objspace);
+}
+
NODE*
rb_node_newnode(enum node_type type, VALUE a0, VALUE a1, VALUE a2)
{
@@ -1066,6 +1140,21 @@ rb_node_newnode(enum node_type type, VALUE a0, VALUE a1, VALUE a2)
return n;
}
+NODE*
+rb_node_newnode_longlife(enum node_type type, VALUE a0, VALUE a1, VALUE a2)
+{
+ NODE *n = (NODE*)rb_newobj_longlife();
+
+ n->flags |= T_NODE;
+ nd_set_type(n, type);
+
+ n->u1.value = a0;
+ n->u2.value = a1;
+ n->u3.value = a2;
+
+ return n;
+}
+
VALUE
rb_data_object_alloc(VALUE klass, void *datap, RUBY_DATA_FUNC dmark, RUBY_DATA_FUNC dfree)
{
@@ -1222,6 +1311,30 @@ is_pointer_to_heap(rb_objspace_t *objspace, void *ptr)
return Qfalse;
}
+VALUE
+rb_gc_write_barrier(VALUE ptr)
+{
+ rb_objspace_t *objspace = &rb_objspace;
+ remembered_set_t *tmp;
+ RVALUE *obj = RANY(ptr);
+
+ if (!SPECIAL_CONST_P(ptr) &&
+ !(RBASIC(ptr)->flags & FL_MARK || RBASIC(ptr)->flags & FL_REMEMBERED_SET)) {
+ if (objspace->remembered_set.freed) {
+ tmp = objspace->remembered_set.freed;
+ objspace->remembered_set.freed = objspace->remembered_set.freed->next;
+ }
+ else {
+ tmp = ALLOC(remembered_set_t);
+ }
+ tmp->next = objspace->remembered_set.ptr;
+ tmp->obj = obj;
+ obj->as.basic.flags |= FL_REMEMBERED_SET;
+ objspace->remembered_set.ptr = tmp;
+ }
+ return ptr;
+}
+
static void
mark_locations_array(rb_objspace_t *objspace, register VALUE *x, register long n)
{
@@ -1645,12 +1758,12 @@ gc_mark_children(rb_objspace_t *objspace, VALUE ptr, int lev)
static int obj_free(rb_objspace_t *, VALUE);
static inline void
-add_freelist(rb_objspace_t *objspace, RVALUE *p)
+add_freelist(rb_objspace_t *objspace, RVALUE **list, RVALUE *p)
{
VALGRIND_MAKE_MEM_UNDEFINED((void*)p, sizeof(RVALUE));
p->as.free.flags = 0;
- p->as.free.next = freelist;
- freelist = p;
+ p->as.free.next = *list;
+ *list = p;
}
static void
@@ -1660,7 +1773,7 @@ finalize_list(rb_objspace_t *objspace, RVALUE *p)
RVALUE *tmp = p->as.free.next;
run_final(objspace, (VALUE)p);
if (!FL_TEST(p, FL_SINGLETON)) { /* not freeing page */
- add_freelist(objspace, p);
+ add_freelist(objspace, &freelist, p);
}
else {
struct heaps_slot *slot = (struct heaps_slot *)RDATA(p)->dmark;
@@ -1684,6 +1797,9 @@ free_unused_heaps(rb_objspace_t *objspace)
else {
free(heaps[i].membase);
}
+ if (heaps[i].lifetime == lifetime_longlife) {
+ objspace->heap.longlife_used--;
+ }
heaps_used--;
}
else {
@@ -1729,6 +1845,7 @@ gc_sweep(rb_objspace_t *objspace)
RVALUE *final = final_list;
int deferred;
+ if (heaps[i].lifetime == lifetime_longlife) continue;
p = heaps[i].slot; pend = p + heaps[i].limit;
while (p < pend) {
if (!(p->as.basic.flags & FL_MARK)) {
@@ -1745,7 +1862,7 @@ gc_sweep(rb_objspace_t *objspace)
final_num++;
}
else {
- add_freelist(objspace, p);
+ add_freelist(objspace, &freelist, p);
free_num++;
}
}
@@ -1781,8 +1898,10 @@ gc_sweep(rb_objspace_t *objspace)
}
malloc_increase = 0;
if (freed < free_min) {
- set_heaps_increment(objspace);
- heaps_increment(objspace);
+ if (!heaps_inc && objspace->heap.longlife_used)
+ objspace->flags.longlife_collection = Qtrue;
+ set_heaps_increment(objspace);
+ heaps_increment(objspace);
}
during_gc = 0;
@@ -1798,11 +1917,67 @@ gc_sweep(rb_objspace_t *objspace)
}
}
+static void
+remembered_set_recycle(rb_objspace_t *objspace)
+{
+ remembered_set_t *top = 0, *rem, *next;
+
+ rem = objspace->remembered_set.ptr;
+ while (rem) {
+ next = rem->next;
+ if (RBASIC(rem->obj)->flags & FL_MARK) {
+ top = rem;
+ }
+ else {
+ if (top) {
+ top->next = next;
+ }
+ else {
+ objspace->remembered_set.ptr = next;
+ }
+ rem->obj = 0;
+ rem->next = objspace->remembered_set.freed;
+ objspace->remembered_set.freed = rem;
+ }
+ rem = next;
+ }
+}
+
+static void
+gc_sweep_for_longlife(rb_objspace_t *objspace)
+{
+ RVALUE *p, *pend;
+ size_t i, freed = 0;
+
+ objspace->heap.longlife_freelist = 0;
+ for (i = 0; i < heaps_used; i++) {
+
+ if (heaps[i].lifetime == lifetime_normal) continue;
+ p = heaps[i].slot; pend = p + heaps[i].limit;
+ while (p < pend) {
+ if (!(p->as.basic.flags & FL_MARK)) {
+ if (p->as.basic.flags) {
+ obj_free(objspace, (VALUE)p);
+ }
+ add_freelist(objspace, &objspace->heap.longlife_freelist, p);
+ freed++;
+ }
+ p++;
+ }
+ }
+
+ remembered_set_recycle(objspace);
+ objspace->flags.longlife_collection = Qfalse;
+ objspace->profile.longlife_objects = objspace->profile.longlife_objects - freed;
+}
+
void
rb_gc_force_recycle(VALUE p)
{
rb_objspace_t *objspace = &rb_objspace;
- add_freelist(objspace, (RVALUE *)p);
+ if (!(RBASIC(p)->flags & FL_MARK || RBASIC(p)->flags & FL_REMEMBERED_SET)) {
+ add_freelist(objspace, &freelist, (RVALUE *)p);
+ }
}
static inline void
@@ -1988,6 +2163,37 @@ mark_current_machine_context(rb_objspace_t *objspace, rb_thread_t *th)
void rb_gc_mark_encodings(void);
+static void
+rb_gc_mark_remembered_set(rb_objspace_t *objspace)
+{
+ remembered_set_t *rem;
+
+ rem = objspace->remembered_set.ptr;
+ while (rem) {
+ rb_gc_mark((VALUE)rem->obj);
+ rem = rem->next;
+ }
+}
+
+static void
+clear_mark_longlife_heaps(rb_objspace_t *objspace)
+{
+ int i;
+
+ for (i = 0; i < heaps_used; i++) {
+ RVALUE *p, *pend;
+
+ if (heaps[i].lifetime == lifetime_longlife) {
+ p = heaps[i].slot; pend = p + heaps[i].limit;
+ for (;p < pend; p++) {
+ if (p->as.basic.flags & FL_MARK) {
+ RBASIC(p)->flags &= ~FL_MARK;
+ }
+ }
+ }
+ }
+}
+
static int
garbage_collect(rb_objspace_t *objspace)
{
@@ -2019,6 +2225,13 @@ garbage_collect(rb_objspace_t *objspace)
init_mark_stack(objspace);
+ if (objspace->flags.longlife_collection) {
+ clear_mark_longlife_heaps(objspace);
+ }
+ else {
+ rb_gc_mark_remembered_set(objspace);
+ }
+
th->vm->self ? rb_gc_mark(th->vm->self) : rb_vm_mark(th->vm);
if (finalizer_table) {
@@ -2057,6 +2270,9 @@ garbage_collect(rb_objspace_t *objspace)
GC_PROF_MARK_TIMER_STOP;
GC_PROF_SWEEP_TIMER_START;
+ if (objspace->flags.longlife_collection) {
+ gc_sweep_for_longlife(objspace);
+ }
gc_sweep(objspace);
GC_PROF_SWEEP_TIMER_STOP;
@@ -2106,6 +2322,10 @@ rb_gc_mark_machine_stack(rb_thread_t *th)
VALUE
rb_gc_start(void)
{
+ rb_objspace_t *objspace = &rb_objspace;
+ if (objspace->heap.longlife_used) {
+ objspace->flags.longlife_collection = Qtrue;
+ }
rb_gc();
return Qnil;
}
@@ -2771,11 +2991,13 @@ gc_profile_record_get(void)
rb_hash_aset(prof, ID2SYM(rb_intern("HEAP_TOTAL_SIZE")), rb_uint2inum(objspace->profile.record[i].heap_total_size));
rb_hash_aset(prof, ID2SYM(rb_intern("HEAP_TOTAL_OBJECTS")), rb_uint2inum(objspace->profile.record[i].heap_total_objects));
#if GC_PROFILE_MORE_DETAIL
+ rb_hash_aset(prof, ID2SYM(rb_intern("LONGLIFE_COLLECTION")), objspace->profile.record[i].longlife_collection);
rb_hash_aset(prof, ID2SYM(rb_intern("GC_MARK_TIME")), DBL2NUM(objspace->profile.record[i].gc_mark_time));
rb_hash_aset(prof, ID2SYM(rb_intern("GC_SWEEP_TIME")), DBL2NUM(objspace->profile.record[i].gc_sweep_time));
rb_hash_aset(prof, ID2SYM(rb_intern("ALLOCATE_INCREASE")), rb_uint2inum(objspace->profile.record[i].allocate_increase));
rb_hash_aset(prof, ID2SYM(rb_intern("ALLOCATE_LIMIT")), rb_uint2inum(objspace->profile.record[i].allocate_limit));
rb_hash_aset(prof, ID2SYM(rb_intern("HEAP_USE_SLOTS")), rb_uint2inum(objspace->profile.record[i].heap_use_slots));
+ rb_hash_aset(prof, ID2SYM(rb_intern("HEAP_LONGLIFE_USE_SLOTS")), rb_uint2inum(objspace->profile.record[i].heap_longlife_use_slots));
rb_hash_aset(prof, ID2SYM(rb_intern("HEAP_LIVE_OBJECTS")), rb_uint2inum(objspace->profile.record[i].heap_live_objects));
rb_hash_aset(prof, ID2SYM(rb_intern("HEAP_FREE_OBJECTS")), rb_uint2inum(objspace->profile.record[i].heap_free_objects));
rb_hash_aset(prof, ID2SYM(rb_intern("HAVE_FINALIZE")), objspace->profile.record[i].have_finalize);
@@ -2817,21 +3039,25 @@ gc_profile_result(void)
NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("HEAP_USE_SIZE")))),
NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("HEAP_TOTAL_SIZE")))),
NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("HEAP_TOTAL_OBJECTS")))),
- NUM2DBL(rb_hash_aref(r, ID2SYM(rb_intern("GC_TIME"))))*1000);
+ NUM2DBL(rb_hash_aref(r, ID2SYM(rb_intern("GC_TIME"))))*1000
+ );
}
#if GC_PROFILE_MORE_DETAIL
rb_str_cat2(result, "\n\n");
rb_str_cat2(result, "More detail.\n");
- rb_str_cat2(result, "Index Allocate Increase Allocate Limit Use Slot Have Finalize Mark Time(ms) Sweep Time(ms)\n");
+ rb_str_cat2(result, "Index Allocate Increase Allocate Limit Use Slot Longlife Slot Have Finalize Collection Mark Time(ms) Sweep Time(ms)\n");
for (i = 0; i < (int)RARRAY_LEN(record); i++) {
VALUE r = RARRAY_PTR(record)[i];
- rb_str_catf(result, "%5d %17d %17d %9d %14s %25.20f %25.20f\n",
+ rb_str_catf(result, "%5d %17d %17d %9d %14d %14s %11s %25.20f %25.20f\n",
i+1, NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("ALLOCATE_INCREASE")))),
NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("ALLOCATE_LIMIT")))),
NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("HEAP_USE_SLOTS")))),
+ NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("HEAP_LONGLIFE_USE_SLOTS")))),
rb_hash_aref(r, ID2SYM(rb_intern("HAVE_FINALIZE")))? "true" : "false",
+ rb_hash_aref(r, ID2SYM(rb_intern("LONGLIFE_COLLECTION")))? "longlife" : "normal",
NUM2DBL(rb_hash_aref(r, ID2SYM(rb_intern("GC_MARK_TIME"))))*1000,
- NUM2DBL(rb_hash_aref(r, ID2SYM(rb_intern("GC_SWEEP_TIME"))))*1000);
+ NUM2DBL(rb_hash_aref(r, ID2SYM(rb_intern("GC_SWEEP_TIME"))))*1000
+ );
}
#endif
}
@@ -2866,7 +3092,6 @@ gc_profile_report(int argc, VALUE *argv, VALUE self)
return Qnil;
}
-
/*
* The <code>GC</code> module provides an interface to Ruby's mark and
* sweep garbage collection mechanism. Some of the underlying methods
diff --git a/id.h b/id.h
index 97faaf4..a8c4c3a 100644
--- a/id.h
+++ b/id.h
@@ -3,7 +3,7 @@
id.h -
- $Author: nobu $
+ $Author$
created at: Sun Oct 19 21:12:51 2008
Copyright (C) 2007 Koichi Sasada
diff --git a/include/ruby/intern.h b/include/ruby/intern.h
index 7ce7395..60aedc4 100644
--- a/include/ruby/intern.h
+++ b/include/ruby/intern.h
@@ -362,6 +362,7 @@ void rb_gc_mark_maybe(VALUE);
void rb_gc_mark(VALUE);
void rb_gc_force_recycle(VALUE);
void rb_gc(void);
+VALUE rb_gc_write_barrier(VALUE);
void rb_gc_copy_finalizer(VALUE,VALUE);
void rb_gc_finalize_deferred(void);
void rb_gc_call_finalizer_at_exit(void);
diff --git a/include/ruby/ruby.h b/include/ruby/ruby.h
index b887378..68436ab 100644
--- a/include/ruby/ruby.h
+++ b/include/ruby/ruby.h
@@ -804,7 +804,7 @@ struct RBignum {
#define FL_SINGLETON FL_USER0
#define FL_MARK (((VALUE)1)<<5)
-#define FL_RESERVED (((VALUE)1)<<6) /* will be used in the future GC */
+#define FL_REMEMBERED_SET (((VALUE)1)<<6)
#define FL_FINALIZE (((VALUE)1)<<7)
#define FL_TAINT (((VALUE)1)<<8)
#define FL_UNTRUSTED (((VALUE)1)<<9)
diff --git a/insns.def b/insns.def
index 2ffd0ea..5eaad52 100644
--- a/insns.def
+++ b/insns.def
@@ -1221,7 +1221,7 @@ setinlinecache
{
IC ic = GET_CONST_INLINE_CACHE(dst);
- ic->ic_value = val;
+ ic->ic_value = rb_gc_write_barrier(val);
ic->ic_vmstat = GET_VM_STATE_VERSION() - ruby_vm_const_missing_count;
ruby_vm_const_missing_count = 0;
}
diff --git a/iseq.c b/iseq.c
index 28c1a39..eb96cb5 100644
--- a/iseq.c
+++ b/iseq.c
@@ -113,12 +113,12 @@ set_relation(rb_iseq_t *iseq, const VALUE parent)
/* set class nest stack */
if (type == ISEQ_TYPE_TOP) {
/* toplevel is private */
- iseq->cref_stack = NEW_BLOCK(th->top_wrapper ? th->top_wrapper : rb_cObject);
+ iseq->cref_stack = NEW_NODE_LONGLIFE(NODE_BLOCK, th->top_wrapper ? th->top_wrapper : rb_cObject, 0, 0);
iseq->cref_stack->nd_file = 0;
iseq->cref_stack->nd_visi = NOEX_PRIVATE;
}
else if (type == ISEQ_TYPE_METHOD || type == ISEQ_TYPE_CLASS) {
- iseq->cref_stack = NEW_BLOCK(0); /* place holder */
+ iseq->cref_stack = NEW_NODE_LONGLIFE(NODE_BLOCK,0,0,0); /* place holder */
iseq->cref_stack->nd_file = 0;
}
else if (RTEST(parent)) {
diff --git a/node.h b/node.h
index 358bbca..8c30f1c 100644
--- a/node.h
+++ b/node.h
@@ -354,6 +354,7 @@ typedef struct RNode {
#define nd_visi u2.argc
#define NEW_NODE(t,a0,a1,a2) rb_node_newnode((t),(VALUE)(a0),(VALUE)(a1),(VALUE)(a2))
+#define NEW_NODE_LONGLIFE(t,a0,a1,a2) rb_node_newnode_longlife((t),(VALUE)(a0),(VALUE)(a1),(VALUE)(a2))
#define NEW_METHOD(n,x,v) NEW_NODE(NODE_METHOD,x,n,v)
#define NEW_FBODY(n,i) NEW_NODE(NODE_FBODY,i,n,0)
@@ -497,6 +498,7 @@ NODE *rb_compile_file(const char*, VALUE, int);
void rb_add_method(VALUE, ID, NODE *, int);
NODE *rb_node_newnode(enum node_type,VALUE,VALUE,VALUE);
+NODE *rb_node_newnode_longlife(enum node_type,VALUE,VALUE,VALUE);
NODE* rb_method_node(VALUE klass, ID id);
int rb_node_arity(NODE* node);
diff --git a/vm.c b/vm.c
index a549f02..dc7b088 100644
--- a/vm.c
+++ b/vm.c
@@ -1702,7 +1702,7 @@ vm_define_method(rb_thread_t *th, VALUE obj, ID id, VALUE iseqval,
COPY_CREF(miseq->cref_stack, cref);
miseq->klass = klass;
miseq->defined_method_id = id;
- newbody = NEW_NODE(RUBY_VM_METHOD_NODE, 0, miseq->self, 0);
+ newbody = NEW_NODE_LONGLIFE(RUBY_VM_METHOD_NODE, 0, rb_gc_write_barrier(miseq->self), 0);
rb_add_method(klass, id, newbody, noex);
if (!is_singleton && noex == NOEX_MODFUNC) {
diff --git a/vm_core.h b/vm_core.h
index 68f83ff..5b56e5e 100644
--- a/vm_core.h
+++ b/vm_core.h
@@ -523,7 +523,7 @@ typedef struct {
/* inline (method|const) cache */
-#define NEW_INLINE_CACHE_ENTRY() NEW_WHILE(Qundef, 0, 0)
+#define NEW_INLINE_CACHE_ENTRY() NEW_NODE_LONGLIFE(NODE_WHILE, Qundef, 0, 0)
#define ic_class u1.value
#define ic_method u2.node
#define ic_value u2.value
diff --git a/vm_insnhelper.c b/vm_insnhelper.c
index 7b4fd37..9e9153c 100644
--- a/vm_insnhelper.c
+++ b/vm_insnhelper.c
@@ -1160,8 +1160,8 @@ vm_method_search(VALUE id, VALUE klass, IC ic)
}
else {
mn = rb_method_node(klass, id);
- ic->ic_class = klass;
- ic->ic_method = mn;
+ ic->ic_class = rb_gc_write_barrier(klass);
+ ic->ic_method = (NODE *)rb_gc_write_barrier((VALUE)mn);
ic->ic_vmstat = GET_VM_STATE_VERSION();
}
}
diff --git a/vm_method.c b/vm_method.c
index 774fa1e..459652b 100644
--- a/vm_method.c
+++ b/vm_method.c
@@ -140,9 +140,12 @@ rb_add_method(VALUE klass, ID mid, NODE * node, int noex)
* nd_cnt : alias count // (3)
*/
if (node) {
- NODE *method = NEW_METHOD(node, klass, NOEX_WITH_SAFE(noex));
+ NODE *method = NEW_NODE_LONGLIFE(NODE_METHOD,
+ rb_gc_write_barrier(klass),
+ rb_gc_write_barrier((VALUE)node),
+ NOEX_WITH_SAFE(noex));
method->nd_file = (void *)mid;
- body = NEW_FBODY(method, mid);
+ body = NEW_NODE_LONGLIFE(NODE_FBODY, mid, method, 0);
}
else {
body = 0;
@@ -193,7 +196,8 @@ void
rb_define_alloc_func(VALUE klass, VALUE (*func)(VALUE))
{
Check_Type(klass, T_CLASS);
- rb_add_method(rb_singleton_class(klass), ID_ALLOCATOR, NEW_CFUNC(func, 0),
+ rb_add_method(rb_singleton_class(klass), ID_ALLOCATOR,
+ NEW_NODE_LONGLIFE(NODE_CFUNC, func, 0, 0),
NOEX_PRIVATE);
}
@@ -764,10 +768,14 @@ rb_alias(VALUE klass, ID name, ID def)
}
st_insert(RCLASS_M_TBL(klass), name,
- (st_data_t) NEW_FBODY(
- method = NEW_METHOD(orig_fbody->nd_body->nd_body,
- orig_fbody->nd_body->nd_clss,
- NOEX_WITH_SAFE(orig_fbody->nd_body->nd_noex)), def));
+ (st_data_t) NEW_NODE_LONGLIFE(
+ NODE_FBODY,
+ def,
+ method = NEW_NODE_LONGLIFE(NODE_METHOD,
+ rb_gc_write_barrier((VALUE)orig_fbody->nd_body->nd_clss),
+ rb_gc_write_barrier((VALUE)orig_fbody->nd_body->nd_body),
+ NOEX_WITH_SAFE(orig_fbody->nd_body->nd_noex)),
+ 0));
method->nd_file = (void *)def;
rb_clear_cache_by_id(name);
gc_partial_longlife_r23386.patch
(25.7 KB, text/x-diff)
diff --git a/class.c b/class.c
index 2a5dc61..386188d 100644
--- a/class.c
+++ b/class.c
@@ -88,11 +88,13 @@ clone_method(ID mid, NODE *body, struct clone_method_data *data)
}
st_insert(data->tbl, mid,
(st_data_t)
- NEW_FBODY(
- NEW_METHOD(fbody,
- data->klass, /* TODO */
- body->nd_body->nd_noex),
- 0));
+ NEW_NODE_LONGLIFE(
+ NODE_FBODY,
+ 0,
+ NEW_NODE_LONGLIFE(NODE_METHOD,
+ rb_gc_write_barrier(data->klass), /* TODO */
+ rb_gc_write_barrier(fbody),
+ body->nd_body->nd_noex), 0));
}
return ST_CONTINUE;
}
@@ -810,7 +812,7 @@ rb_define_method_id(VALUE klass, ID name, VALUE (*func)(ANYARGS), int argc)
if (func == rb_f_notimplement)
rb_define_notimplement_method_id(klass, name, NOEX_PUBLIC);
else
- rb_add_method(klass, name, NEW_CFUNC(func,argc), NOEX_PUBLIC);
+ rb_add_method(klass, name, NEW_NODE_LONGLIFE(NODE_CFUNC, func, argc, 0), NOEX_PUBLIC);
}
void
@@ -826,7 +828,7 @@ rb_define_protected_method(VALUE klass, const char *name, VALUE (*func)(ANYARGS)
if (func == rb_f_notimplement)
rb_define_notimplement_method_id(klass, id, NOEX_PROTECTED);
else
- rb_add_method(klass, id, NEW_CFUNC(func, argc), NOEX_PROTECTED);
+ rb_add_method(klass, id, NEW_NODE_LONGLIFE(NODE_CFUNC, func, argc, 0), NOEX_PROTECTED);
}
void
@@ -836,7 +838,7 @@ rb_define_private_method(VALUE klass, const char *name, VALUE (*func)(ANYARGS),
if (func == rb_f_notimplement)
rb_define_notimplement_method_id(klass, id, NOEX_PRIVATE);
else
- rb_add_method(klass, id, NEW_CFUNC(func, argc), NOEX_PRIVATE);
+ rb_add_method(klass, id, NEW_NODE_LONGLIFE(NODE_CFUNC, func, argc, 0), NOEX_PRIVATE);
}
void
diff --git a/debug.c b/debug.c
index 3b4b76e..b5f2805 100644
--- a/debug.c
+++ b/debug.c
@@ -32,7 +32,7 @@ static const union {
RUBY_ENC_CODERANGE_VALID = ENC_CODERANGE_VALID,
RUBY_ENC_CODERANGE_BROKEN = ENC_CODERANGE_BROKEN,
RUBY_FL_MARK = FL_MARK,
- RUBY_FL_RESERVED = FL_RESERVED,
+ RUBY_FL_REMENBERED_SET = FL_REMEMBERED_SET,
RUBY_FL_FINALIZE = FL_FINALIZE,
RUBY_FL_TAINT = FL_TAINT,
RUBY_FL_UNTRUSTED = FL_UNTRUSTED,
diff --git a/gc.c b/gc.c
index 8a4075a..434dbd4 100644
--- a/gc.c
+++ b/gc.c
@@ -93,13 +93,18 @@ typedef struct gc_profile_record {
double gc_mark_time;
double gc_sweep_time;
double gc_invoke_time;
+
size_t heap_use_slots;
+ size_t heap_longlife_use_slots;
size_t heap_live_objects;
size_t heap_free_objects;
size_t heap_total_objects;
size_t heap_use_size;
size_t heap_total_size;
+
int have_finalize;
+ int longlife_collection;
+
size_t allocate_increase;
size_t allocate_limit;
} gc_profile_record;
@@ -156,6 +161,7 @@ getrusage_time(void)
MEMZERO(&objspace->profile.record[count], gc_profile_record, 1);\
gc_time = getrusage_time();\
objspace->profile.record[count].gc_invoke_time = gc_time - objspace->profile.invoke_time;\
+ objspace->profile.record[count].longlife_collection = objspace->flags.longlife_collection;\
}\
} while(0)
@@ -210,15 +216,15 @@ getrusage_time(void)
if (objspace->profile.run) {\
size_t count = objspace->profile.count;\
objspace->profile.record[count].heap_use_slots = heaps_used;\
- objspace->profile.record[count].heap_live_objects = live;\
- objspace->profile.record[count].heap_free_objects = freed;\
+ objspace->profile.record[count].heap_longlife_use_slots = objspace->heap.longlife_used;\
+ objspace->profile.record[count].heap_live_objects = live + objspace->profile.longlife_objects;\
+ objspace->profile.record[count].heap_free_objects = freed + (objspace->heap.longlife_used * HEAP_OBJ_LIMIT - objspace->profile.longlife_objects); \
objspace->profile.record[count].heap_total_objects = heaps_used * HEAP_OBJ_LIMIT;\
objspace->profile.record[count].have_finalize = final_list ? Qtrue : Qfalse;\
- objspace->profile.record[count].heap_use_size = live * sizeof(RVALUE);\
+ objspace->profile.record[count].heap_use_size = (live + objspace->profile.longlife_objects) * sizeof(RVALUE); \
objspace->profile.record[count].heap_total_size = heaps_used * (HEAP_OBJ_LIMIT * sizeof(RVALUE));\
}\
} while(0)
-
#else
#define INIT_GC_PROF_PARAMS double gc_time = 0;\
size_t count = objspace->profile.count
@@ -231,7 +237,7 @@ getrusage_time(void)
if (objspace->profile.run) {\
size_t count = objspace->profile.count;\
objspace->profile.record[count].heap_total_objects = heaps_used * HEAP_OBJ_LIMIT;\
- objspace->profile.record[count].heap_use_size = live * sizeof(RVALUE);\
+ objspace->profile.record[count].heap_use_size = (live + objspace->profile.longlife_objects) * sizeof(RVALUE); \
objspace->profile.record[count].heap_total_size = heaps_used * HEAP_SIZE;\
}\
} while(0)
@@ -275,10 +281,16 @@ typedef struct RVALUE {
#pragma pack(pop)
#endif
+enum lifetime {
+ lifetime_normal,
+ lifetime_longlife
+};
+
struct heaps_slot {
void *membase;
RVALUE *slot;
int limit;
+ enum lifetime lifetime;
};
#define HEAP_MIN_SLOTS 10000
@@ -291,6 +303,11 @@ struct gc_list {
#define CALC_EXACT_MALLOC_SIZE 0
+typedef struct remembered_set {
+ RVALUE *obj;
+ struct remembered_set *next;
+} remembered_set_t;
+
typedef struct rb_objspace {
struct {
size_t limit;
@@ -305,13 +322,20 @@ typedef struct rb_objspace {
struct heaps_slot *ptr;
size_t length;
size_t used;
+ size_t longlife_used;
RVALUE *freelist;
+ RVALUE *longlife_freelist;
RVALUE *range[2];
RVALUE *freed;
} heap;
struct {
+ remembered_set_t *ptr;
+ remembered_set_t *freed;
+ } remembered_set;
+ struct {
int dont_gc;
int during_gc;
+ int longlife_collection;
} flags;
struct {
st_table *table;
@@ -327,6 +351,7 @@ typedef struct rb_objspace {
gc_profile_record *record;
size_t count;
size_t size;
+ size_t longlife_objects;
double invoke_time;
} profile;
struct gc_list *global_list;
@@ -395,6 +420,7 @@ rb_objspace_alloc(void)
/*#define HEAP_SIZE 0x800 */
#define HEAP_OBJ_LIMIT (HEAP_SIZE / sizeof(struct RVALUE))
+#define NORMAL_HEAPS_USED (objspace->heap.used - objspace->heap.longlife_used)
extern VALUE rb_cMutex;
extern st_table *rb_class_tbl;
@@ -858,7 +884,7 @@ allocate_heaps(rb_objspace_t *objspace, size_t next_heaps_length)
}
static void
-assign_heap_slot(rb_objspace_t *objspace)
+assign_heap_slot(rb_objspace_t *objspace, RVALUE **list, enum lifetime lifetime)
{
RVALUE *p, *pend, *membase;
size_t hi, lo, mid;
@@ -902,15 +928,17 @@ assign_heap_slot(rb_objspace_t *objspace)
heaps[hi].membase = membase;
heaps[hi].slot = p;
heaps[hi].limit = objs;
+ heaps[hi].lifetime = lifetime;
pend = p + objs;
if (lomem == 0 || lomem > p) lomem = p;
if (himem < pend) himem = pend;
+ if (lifetime == lifetime_longlife) objspace->heap.longlife_used++;
heaps_used++;
while (p < pend) {
p->as.free.flags = 0;
- p->as.free.next = freelist;
- freelist = p;
+ p->as.free.next = *list;
+ *list = p;
p++;
}
}
@@ -927,11 +955,11 @@ init_heap(rb_objspace_t *objspace)
}
if ((heaps_used + add) > heaps_length) {
- allocate_heaps(objspace, heaps_used + add);
+ allocate_heaps(objspace, heaps_used + add);
}
for (i = 0; i < add; i++) {
- assign_heap_slot(objspace);
+ assign_heap_slot(objspace, &freelist, lifetime_normal);
}
heaps_inc = 0;
objspace->profile.invoke_time = getrusage_time();
@@ -958,16 +986,45 @@ static int
heaps_increment(rb_objspace_t *objspace)
{
if (heaps_inc > 0) {
- assign_heap_slot(objspace);
+ assign_heap_slot(objspace, &freelist, lifetime_normal);
heaps_inc--;
return Qtrue;
}
return Qfalse;
}
+#define LONGLIFE_ALLOCATE_HEAPS_MIN 10
+
+static void
+add_longlife_heaps_slot(rb_objspace_t *objspace)
+{
+ if ((heaps_used + heaps_inc) >= heaps_length) {
+ allocate_heaps(objspace, (heaps_length + LONGLIFE_ALLOCATE_HEAPS_MIN));
+ }
+ assign_heap_slot(objspace, &objspace->heap.longlife_freelist, lifetime_longlife);
+}
+
#define RANY(o) ((RVALUE*)(o))
static VALUE
+rb_newobj_from_longlife_heap(rb_objspace_t *objspace)
+{
+ VALUE obj;
+ if (!objspace->heap.longlife_freelist) {
+ add_longlife_heaps_slot(objspace);
+ }
+
+ obj = (VALUE)objspace->heap.longlife_freelist;
+ objspace->heap.longlife_freelist = objspace->heap.longlife_freelist->as.free.next;
+
+ MEMZERO((void*)obj, RVALUE, 1);
+ FL_SET(RANY(obj), FL_MARK);
+ objspace->profile.longlife_objects++;
+
+ return obj;
+}
+
+static VALUE
rb_newobj_from_heap(rb_objspace_t *objspace)
{
VALUE obj;
@@ -1060,6 +1117,22 @@ rb_newobj(void)
#endif
}
+VALUE
+rb_newobj_longlife(void)
+{
+#if defined(ENABLE_VM_OBJSPACE) && ENABLE_VM_OBJSPACE
+ rb_objspace_t *objspace = th->vm->objspace;
+#else
+ rb_objspace_t *objspace = &rb_objspace;
+#endif
+ if (during_gc) {
+ dont_gc = 1;
+ during_gc = 0;
+ rb_bug("object allocation during garbage collection phase");
+ }
+ return rb_newobj_from_longlife_heap(objspace);
+}
+
NODE*
rb_node_newnode(enum node_type type, VALUE a0, VALUE a1, VALUE a2)
{
@@ -1075,6 +1148,21 @@ rb_node_newnode(enum node_type type, VALUE a0, VALUE a1, VALUE a2)
return n;
}
+NODE*
+rb_node_newnode_longlife(enum node_type type, VALUE a0, VALUE a1, VALUE a2)
+{
+ NODE *n = (NODE*)rb_newobj_longlife();
+
+ n->flags |= T_NODE;
+ nd_set_type(n, type);
+
+ n->u1.value = a0;
+ n->u2.value = a1;
+ n->u3.value = a2;
+
+ return n;
+}
+
VALUE
rb_data_object_alloc(VALUE klass, void *datap, RUBY_DATA_FUNC dmark, RUBY_DATA_FUNC dfree)
{
@@ -1231,6 +1319,30 @@ is_pointer_to_heap(rb_objspace_t *objspace, void *ptr)
return Qfalse;
}
+VALUE
+rb_gc_write_barrier(VALUE ptr)
+{
+ rb_objspace_t *objspace = &rb_objspace;
+ remembered_set_t *tmp;
+ RVALUE *obj = RANY(ptr);
+
+ if (!SPECIAL_CONST_P(ptr) &&
+ !(RBASIC(ptr)->flags & FL_MARK || RBASIC(ptr)->flags & FL_REMEMBERED_SET)) {
+ if (objspace->remembered_set.freed) {
+ tmp = objspace->remembered_set.freed;
+ objspace->remembered_set.freed = objspace->remembered_set.freed->next;
+ }
+ else {
+ tmp = ALLOC(remembered_set_t);
+ }
+ tmp->next = objspace->remembered_set.ptr;
+ tmp->obj = obj;
+ obj->as.basic.flags |= FL_REMEMBERED_SET;
+ objspace->remembered_set.ptr = tmp;
+ }
+ return ptr;
+}
+
static void
mark_locations_array(rb_objspace_t *objspace, register VALUE *x, register long n)
{
@@ -1654,12 +1766,12 @@ gc_mark_children(rb_objspace_t *objspace, VALUE ptr, int lev)
static int obj_free(rb_objspace_t *, VALUE);
static inline void
-add_freelist(rb_objspace_t *objspace, RVALUE *p)
+add_freelist(rb_objspace_t *objspace, RVALUE **list, RVALUE *p)
{
VALGRIND_MAKE_MEM_UNDEFINED((void*)p, sizeof(RVALUE));
p->as.free.flags = 0;
- p->as.free.next = freelist;
- freelist = p;
+ p->as.free.next = *list;
+ *list = p;
}
static void
@@ -1669,7 +1781,7 @@ finalize_list(rb_objspace_t *objspace, RVALUE *p)
RVALUE *tmp = p->as.free.next;
run_final(objspace, (VALUE)p);
if (!FL_TEST(p, FL_SINGLETON)) { /* not freeing page */
- add_freelist(objspace, p);
+ add_freelist(objspace, &freelist, p);
}
else {
struct heaps_slot *slot = (struct heaps_slot *)RDATA(p)->dmark;
@@ -1693,6 +1805,9 @@ free_unused_heaps(rb_objspace_t *objspace)
else {
free(heaps[i].membase);
}
+ if (heaps[i].lifetime == lifetime_longlife) {
+ objspace->heap.longlife_used--;
+ }
heaps_used--;
}
else {
@@ -1738,6 +1853,7 @@ gc_sweep(rb_objspace_t *objspace)
RVALUE *final = final_list;
int deferred;
+ if (heaps[i].lifetime == lifetime_longlife) continue;
p = heaps[i].slot; pend = p + heaps[i].limit;
while (p < pend) {
if (!(p->as.basic.flags & FL_MARK)) {
@@ -1754,7 +1870,7 @@ gc_sweep(rb_objspace_t *objspace)
final_num++;
}
else {
- add_freelist(objspace, p);
+ add_freelist(objspace, &freelist, p);
free_num++;
}
}
@@ -1790,8 +1906,10 @@ gc_sweep(rb_objspace_t *objspace)
}
malloc_increase = 0;
if (freed < free_min) {
- set_heaps_increment(objspace);
- heaps_increment(objspace);
+ if (!heaps_inc && objspace->heap.longlife_used)
+ objspace->flags.longlife_collection = Qtrue;
+ set_heaps_increment(objspace);
+ heaps_increment(objspace);
}
during_gc = 0;
@@ -1807,11 +1925,67 @@ gc_sweep(rb_objspace_t *objspace)
}
}
+static void
+remembered_set_recycle(rb_objspace_t *objspace)
+{
+ remembered_set_t *top = 0, *rem, *next;
+
+ rem = objspace->remembered_set.ptr;
+ while (rem) {
+ next = rem->next;
+ if (RBASIC(rem->obj)->flags & FL_MARK) {
+ top = rem;
+ }
+ else {
+ if (top) {
+ top->next = next;
+ }
+ else {
+ objspace->remembered_set.ptr = next;
+ }
+ rem->obj = 0;
+ rem->next = objspace->remembered_set.freed;
+ objspace->remembered_set.freed = rem;
+ }
+ rem = next;
+ }
+}
+
+static void
+gc_sweep_for_longlife(rb_objspace_t *objspace)
+{
+ RVALUE *p, *pend;
+ size_t i, freed = 0;
+
+ objspace->heap.longlife_freelist = 0;
+ for (i = 0; i < heaps_used; i++) {
+
+ if (heaps[i].lifetime == lifetime_normal) continue;
+ p = heaps[i].slot; pend = p + heaps[i].limit;
+ while (p < pend) {
+ if (!(p->as.basic.flags & FL_MARK)) {
+ if (p->as.basic.flags) {
+ obj_free(objspace, (VALUE)p);
+ }
+ add_freelist(objspace, &objspace->heap.longlife_freelist, p);
+ freed++;
+ }
+ p++;
+ }
+ }
+
+ remembered_set_recycle(objspace);
+ objspace->flags.longlife_collection = Qfalse;
+ objspace->profile.longlife_objects = objspace->profile.longlife_objects - freed;
+}
+
void
rb_gc_force_recycle(VALUE p)
{
rb_objspace_t *objspace = &rb_objspace;
- add_freelist(objspace, (RVALUE *)p);
+ if (!(RBASIC(p)->flags & FL_MARK || RBASIC(p)->flags & FL_REMEMBERED_SET)) {
+ add_freelist(objspace, &freelist, (RVALUE *)p);
+ }
}
static inline void
@@ -1997,6 +2171,37 @@ mark_current_machine_context(rb_objspace_t *objspace, rb_thread_t *th)
void rb_gc_mark_encodings(void);
+static void
+rb_gc_mark_remembered_set(rb_objspace_t *objspace)
+{
+ remembered_set_t *rem;
+
+ rem = objspace->remembered_set.ptr;
+ while (rem) {
+ rb_gc_mark((VALUE)rem->obj);
+ rem = rem->next;
+ }
+}
+
+static void
+clear_mark_longlife_heaps(rb_objspace_t *objspace)
+{
+ int i;
+
+ for (i = 0; i < heaps_used; i++) {
+ RVALUE *p, *pend;
+
+ if (heaps[i].lifetime == lifetime_longlife) {
+ p = heaps[i].slot; pend = p + heaps[i].limit;
+ for (;p < pend; p++) {
+ if (p->as.basic.flags & FL_MARK) {
+ RBASIC(p)->flags &= ~FL_MARK;
+ }
+ }
+ }
+ }
+}
+
static int
garbage_collect(rb_objspace_t *objspace)
{
@@ -2028,6 +2233,13 @@ garbage_collect(rb_objspace_t *objspace)
init_mark_stack(objspace);
+ if (objspace->flags.longlife_collection) {
+ clear_mark_longlife_heaps(objspace);
+ }
+ else {
+ rb_gc_mark_remembered_set(objspace);
+ }
+
th->vm->self ? rb_gc_mark(th->vm->self) : rb_vm_mark(th->vm);
if (finalizer_table) {
@@ -2066,6 +2278,9 @@ garbage_collect(rb_objspace_t *objspace)
GC_PROF_MARK_TIMER_STOP;
GC_PROF_SWEEP_TIMER_START;
+ if (objspace->flags.longlife_collection) {
+ gc_sweep_for_longlife(objspace);
+ }
gc_sweep(objspace);
GC_PROF_SWEEP_TIMER_STOP;
@@ -2115,6 +2330,10 @@ rb_gc_mark_machine_stack(rb_thread_t *th)
VALUE
rb_gc_start(void)
{
+ rb_objspace_t *objspace = &rb_objspace;
+ if (objspace->heap.longlife_used) {
+ objspace->flags.longlife_collection = Qtrue;
+ }
rb_gc();
return Qnil;
}
@@ -2783,11 +3002,13 @@ gc_profile_record_get(void)
rb_hash_aset(prof, ID2SYM(rb_intern("HEAP_TOTAL_SIZE")), rb_uint2inum(objspace->profile.record[i].heap_total_size));
rb_hash_aset(prof, ID2SYM(rb_intern("HEAP_TOTAL_OBJECTS")), rb_uint2inum(objspace->profile.record[i].heap_total_objects));
#if GC_PROFILE_MORE_DETAIL
+ rb_hash_aset(prof, ID2SYM(rb_intern("LONGLIFE_COLLECTION")), objspace->profile.record[i].longlife_collection);
rb_hash_aset(prof, ID2SYM(rb_intern("GC_MARK_TIME")), DBL2NUM(objspace->profile.record[i].gc_mark_time));
rb_hash_aset(prof, ID2SYM(rb_intern("GC_SWEEP_TIME")), DBL2NUM(objspace->profile.record[i].gc_sweep_time));
rb_hash_aset(prof, ID2SYM(rb_intern("ALLOCATE_INCREASE")), rb_uint2inum(objspace->profile.record[i].allocate_increase));
rb_hash_aset(prof, ID2SYM(rb_intern("ALLOCATE_LIMIT")), rb_uint2inum(objspace->profile.record[i].allocate_limit));
rb_hash_aset(prof, ID2SYM(rb_intern("HEAP_USE_SLOTS")), rb_uint2inum(objspace->profile.record[i].heap_use_slots));
+ rb_hash_aset(prof, ID2SYM(rb_intern("HEAP_LONGLIFE_USE_SLOTS")), rb_uint2inum(objspace->profile.record[i].heap_longlife_use_slots));
rb_hash_aset(prof, ID2SYM(rb_intern("HEAP_LIVE_OBJECTS")), rb_uint2inum(objspace->profile.record[i].heap_live_objects));
rb_hash_aset(prof, ID2SYM(rb_intern("HEAP_FREE_OBJECTS")), rb_uint2inum(objspace->profile.record[i].heap_free_objects));
rb_hash_aset(prof, ID2SYM(rb_intern("HAVE_FINALIZE")), objspace->profile.record[i].have_finalize);
@@ -2829,21 +3050,25 @@ gc_profile_result(void)
NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("HEAP_USE_SIZE")))),
NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("HEAP_TOTAL_SIZE")))),
NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("HEAP_TOTAL_OBJECTS")))),
- NUM2DBL(rb_hash_aref(r, ID2SYM(rb_intern("GC_TIME"))))*1000);
+ NUM2DBL(rb_hash_aref(r, ID2SYM(rb_intern("GC_TIME"))))*1000
+ );
}
#if GC_PROFILE_MORE_DETAIL
rb_str_cat2(result, "\n\n");
rb_str_cat2(result, "More detail.\n");
- rb_str_cat2(result, "Index Allocate Increase Allocate Limit Use Slot Have Finalize Mark Time(ms) Sweep Time(ms)\n");
+ rb_str_cat2(result, "Index Allocate Increase Allocate Limit Use Slot Longlife Slot Have Finalize Collection Mark Time(ms) Sweep Time(ms)\n");
for (i = 0; i < (int)RARRAY_LEN(record); i++) {
VALUE r = RARRAY_PTR(record)[i];
- rb_str_catf(result, "%5d %17d %17d %9d %14s %25.20f %25.20f\n",
+ rb_str_catf(result, "%5d %17d %17d %9d %14d %14s %11s %25.20f %25.20f\n",
i+1, NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("ALLOCATE_INCREASE")))),
NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("ALLOCATE_LIMIT")))),
NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("HEAP_USE_SLOTS")))),
+ NUM2INT(rb_hash_aref(r, ID2SYM(rb_intern("HEAP_LONGLIFE_USE_SLOTS")))),
rb_hash_aref(r, ID2SYM(rb_intern("HAVE_FINALIZE")))? "true" : "false",
+ rb_hash_aref(r, ID2SYM(rb_intern("LONGLIFE_COLLECTION")))? "longlife" : "normal",
NUM2DBL(rb_hash_aref(r, ID2SYM(rb_intern("GC_MARK_TIME"))))*1000,
- NUM2DBL(rb_hash_aref(r, ID2SYM(rb_intern("GC_SWEEP_TIME"))))*1000);
+ NUM2DBL(rb_hash_aref(r, ID2SYM(rb_intern("GC_SWEEP_TIME"))))*1000
+ );
}
#endif
}
@@ -2878,7 +3103,6 @@ gc_profile_report(int argc, VALUE *argv, VALUE self)
return Qnil;
}
-
/*
* The <code>GC</code> module provides an interface to Ruby's mark and
* sweep garbage collection mechanism. Some of the underlying methods
diff --git a/include/ruby/intern.h b/include/ruby/intern.h
index 899a7b3..ae44385 100644
--- a/include/ruby/intern.h
+++ b/include/ruby/intern.h
@@ -367,6 +367,7 @@ void rb_gc_mark_maybe(VALUE);
void rb_gc_mark(VALUE);
void rb_gc_force_recycle(VALUE);
void rb_gc(void);
+VALUE rb_gc_write_barrier(VALUE);
void rb_gc_copy_finalizer(VALUE,VALUE);
void rb_gc_finalize_deferred(void);
void rb_gc_call_finalizer_at_exit(void);
diff --git a/include/ruby/ruby.h b/include/ruby/ruby.h
index 7a83887..a277bb2 100644
--- a/include/ruby/ruby.h
+++ b/include/ruby/ruby.h
@@ -809,7 +809,7 @@ struct RBignum {
#define FL_SINGLETON FL_USER0
#define FL_MARK (((VALUE)1)<<5)
-#define FL_RESERVED (((VALUE)1)<<6) /* will be used in the future GC */
+#define FL_REMEMBERED_SET (((VALUE)1)<<6)
#define FL_FINALIZE (((VALUE)1)<<7)
#define FL_TAINT (((VALUE)1)<<8)
#define FL_UNTRUSTED (((VALUE)1)<<9)
diff --git a/insns.def b/insns.def
index bc93010..e1e77c6 100644
--- a/insns.def
+++ b/insns.def
@@ -1222,7 +1222,7 @@ setinlinecache
{
IC ic = GET_CONST_INLINE_CACHE(dst);
- ic->ic_value = val;
+ ic->ic_value = rb_gc_write_barrier(val);
ic->ic_vmstat = GET_VM_STATE_VERSION() - ruby_vm_const_missing_count;
ruby_vm_const_missing_count = 0;
}
diff --git a/iseq.c b/iseq.c
index cc2ffaa..a102941 100644
--- a/iseq.c
+++ b/iseq.c
@@ -131,12 +131,12 @@ set_relation(rb_iseq_t *iseq, const VALUE parent)
/* set class nest stack */
if (type == ISEQ_TYPE_TOP) {
/* toplevel is private */
- iseq->cref_stack = NEW_BLOCK(th->top_wrapper ? th->top_wrapper : rb_cObject);
+ iseq->cref_stack = NEW_NODE_LONGLIFE(NODE_BLOCK, th->top_wrapper ? th->top_wrapper : rb_cObject, 0, 0);
iseq->cref_stack->nd_file = 0;
iseq->cref_stack->nd_visi = NOEX_PRIVATE;
}
else if (type == ISEQ_TYPE_METHOD || type == ISEQ_TYPE_CLASS) {
- iseq->cref_stack = NEW_BLOCK(0); /* place holder */
+ iseq->cref_stack = NEW_NODE_LONGLIFE(NODE_BLOCK,0,0,0); /* place holder */
iseq->cref_stack->nd_file = 0;
}
else if (RTEST(parent)) {
diff --git a/node.h b/node.h
index 8de2082..c6a0fbb 100644
--- a/node.h
+++ b/node.h
@@ -354,6 +354,7 @@ typedef struct RNode {
#define nd_visi u2.argc
#define NEW_NODE(t,a0,a1,a2) rb_node_newnode((t),(VALUE)(a0),(VALUE)(a1),(VALUE)(a2))
+#define NEW_NODE_LONGLIFE(t,a0,a1,a2) rb_node_newnode_longlife((t),(VALUE)(a0),(VALUE)(a1),(VALUE)(a2))
#define NEW_METHOD(n,x,v) NEW_NODE(NODE_METHOD,x,n,v)
#define NEW_FBODY(n,i) NEW_NODE(NODE_FBODY,i,n,0)
@@ -497,6 +498,7 @@ NODE *rb_compile_file(const char*, VALUE, int);
void rb_add_method(VALUE, ID, NODE *, int);
NODE *rb_node_newnode(enum node_type,VALUE,VALUE,VALUE);
+NODE *rb_node_newnode_longlife(enum node_type,VALUE,VALUE,VALUE);
NODE* rb_method_node(VALUE klass, ID id);
int rb_node_arity(NODE* node);
diff --git a/vm.c b/vm.c
index 17c2697..0b05de1 100644
--- a/vm.c
+++ b/vm.c
@@ -1702,7 +1702,7 @@ vm_define_method(rb_thread_t *th, VALUE obj, ID id, VALUE iseqval,
COPY_CREF(miseq->cref_stack, cref);
miseq->klass = klass;
miseq->defined_method_id = id;
- newbody = NEW_NODE(RUBY_VM_METHOD_NODE, 0, miseq->self, 0);
+ newbody = NEW_NODE_LONGLIFE(RUBY_VM_METHOD_NODE, 0, rb_gc_write_barrier(miseq->self), 0);
rb_add_method(klass, id, newbody, noex);
if (!is_singleton && noex == NOEX_MODFUNC) {
diff --git a/vm_core.h b/vm_core.h
index 511c1e3..ed96392 100644
--- a/vm_core.h
+++ b/vm_core.h
@@ -523,7 +523,7 @@ typedef struct {
/* inline (method|const) cache */
-#define NEW_INLINE_CACHE_ENTRY() NEW_WHILE(Qundef, 0, 0)
+#define NEW_INLINE_CACHE_ENTRY() NEW_NODE_LONGLIFE(NODE_WHILE, Qundef, 0, 0)
#define ic_class u1.value
#define ic_method u2.node
#define ic_value u2.value
diff --git a/vm_insnhelper.c b/vm_insnhelper.c
index 5293af3..13bbcf7 100644
--- a/vm_insnhelper.c
+++ b/vm_insnhelper.c
@@ -1170,8 +1170,8 @@ vm_method_search(VALUE id, VALUE klass, IC ic)
}
else {
mn = rb_method_node(klass, id);
- ic->ic_class = klass;
- ic->ic_method = mn;
+ ic->ic_class = rb_gc_write_barrier(klass);
+ ic->ic_method = (NODE *)rb_gc_write_barrier((VALUE)mn);
ic->ic_vmstat = GET_VM_STATE_VERSION();
}
}
diff --git a/vm_method.c b/vm_method.c
index aabe75d..25d2e6d 100644
--- a/vm_method.c
+++ b/vm_method.c
@@ -142,9 +142,12 @@ rb_add_method(VALUE klass, ID mid, NODE * node, int noex)
* nd_cnt : alias count // (3)
*/
if (node) {
- NODE *method = NEW_METHOD(node, klass, NOEX_WITH_SAFE(noex));
+ NODE *method = NEW_NODE_LONGLIFE(NODE_METHOD,
+ rb_gc_write_barrier(klass),
+ rb_gc_write_barrier((VALUE)node),
+ NOEX_WITH_SAFE(noex));
method->nd_file = (void *)mid;
- body = NEW_FBODY(method, mid);
+ body = NEW_NODE_LONGLIFE(NODE_FBODY, mid, method, 0);
}
else {
body = 0;
@@ -195,7 +198,8 @@ void
rb_define_alloc_func(VALUE klass, VALUE (*func)(VALUE))
{
Check_Type(klass, T_CLASS);
- rb_add_method(rb_singleton_class(klass), ID_ALLOCATOR, NEW_CFUNC(func, 0),
+ rb_add_method(rb_singleton_class(klass), ID_ALLOCATOR,
+ NEW_NODE_LONGLIFE(NODE_CFUNC, func, 0, 0),
NOEX_PRIVATE);
}
@@ -774,10 +778,14 @@ rb_alias(VALUE klass, ID name, ID def)
}
st_insert(RCLASS_M_TBL(klass), name,
- (st_data_t) NEW_FBODY(
- method = NEW_METHOD(orig_fbody->nd_body->nd_body,
- orig_fbody->nd_body->nd_clss,
- NOEX_WITH_SAFE(orig_fbody->nd_body->nd_noex)), def));
+ (st_data_t) NEW_NODE_LONGLIFE(
+ NODE_FBODY,
+ def,
+ method = NEW_NODE_LONGLIFE(NODE_METHOD,
+ rb_gc_write_barrier((VALUE)orig_fbody->nd_body->nd_clss),
+ rb_gc_write_barrier((VALUE)orig_fbody->nd_body->nd_body),
+ NOEX_WITH_SAFE(orig_fbody->nd_body->nd_noex)),
+ 0));
method->nd_file = (void *)def;
rb_clear_cache_by_id(name);
bm_gc_app_heapstyle_maker.rb
(636 Bytes, text/x-ruby)
require 'erb'
class_cnt = 10000
method_cnt = 5
puts ERB.new(DATA.read).result(binding)
__END__
$: << File.dirname(__FILE__)
YOUNG_OBJ_CNT = 500
LONGLIFE_OBJ_CNT = 3
LONGLIFE_OBJS = []
<% class_cnt.times do |i| %>
class AppHeapStyle<%= i %>
<% method_cnt.times do |j| %>
def self.method_<%= j %>
YOUNG_OBJ_CNT.times{|i| i.to_s}
LONGLIFE_OBJ_CNT.times{|i| LONGLIFE_OBJS << i.to_s}
return "class_<%= i %>_method_<%= j %>"
end
<% end %>
end
<% end %>
GC::Profiler.enable
<% class_cnt.times do |i| %>
<% method_cnt.times do |j| %>
AppHeapStyle<%= i %>.method_<%= j %>
<% end %>
<% end %>
GC::Profiler.report