Mon, 08 Dec 2014 18:57:33 +0100
8067655: Clean up G1 remembered set oop iteration
Summary: Pass on the static type G1ParPushHeapRSClosure to allow oop_iterate devirtualization
Reviewed-by: jmasa, kbarrett
1 /*
2 * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
27 #include "gc_implementation/g1/g1OopClosures.inline.hpp"
28 #include "gc_implementation/g1/g1ParScanThreadState.inline.hpp"
29 #include "oops/oop.inline.hpp"
30 #include "oops/oop.pcgc.inline.hpp"
31 #include "runtime/prefetch.inline.hpp"
33 G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, uint queue_num, ReferenceProcessor* rp)
34 : _g1h(g1h),
35 _refs(g1h->task_queue(queue_num)),
36 _dcq(&g1h->dirty_card_queue_set()),
37 _ct_bs(g1h->g1_barrier_set()),
38 _g1_rem(g1h->g1_rem_set()),
39 _hash_seed(17), _queue_num(queue_num),
40 _term_attempts(0),
41 _age_table(false), _scanner(g1h, rp),
42 _strong_roots_time(0), _term_time(0) {
43 _scanner.set_par_scan_thread_state(this);
44 // we allocate G1YoungSurvRateNumRegions plus one entries, since
45 // we "sacrifice" entry 0 to keep track of surviving bytes for
46 // non-young regions (where the age is -1)
47 // We also add a few elements at the beginning and at the end in
48 // an attempt to eliminate cache contention
49 uint real_length = 1 + _g1h->g1_policy()->young_cset_region_length();
50 uint array_length = PADDING_ELEM_NUM +
51 real_length +
52 PADDING_ELEM_NUM;
53 _surviving_young_words_base = NEW_C_HEAP_ARRAY(size_t, array_length, mtGC);
54 if (_surviving_young_words_base == NULL)
55 vm_exit_out_of_memory(array_length * sizeof(size_t), OOM_MALLOC_ERROR,
56 "Not enough space for young surv histo.");
57 _surviving_young_words = _surviving_young_words_base + PADDING_ELEM_NUM;
58 memset(_surviving_young_words, 0, (size_t) real_length * sizeof(size_t));
60 _g1_par_allocator = G1ParGCAllocator::create_allocator(_g1h);
62 _start = os::elapsedTime();
63 }
65 G1ParScanThreadState::~G1ParScanThreadState() {
66 _g1_par_allocator->retire_alloc_buffers();
67 delete _g1_par_allocator;
68 FREE_C_HEAP_ARRAY(size_t, _surviving_young_words_base, mtGC);
69 }
71 void
72 G1ParScanThreadState::print_termination_stats_hdr(outputStream* const st)
73 {
74 st->print_raw_cr("GC Termination Stats");
75 st->print_raw_cr(" elapsed --strong roots-- -------termination-------"
76 " ------waste (KiB)------");
77 st->print_raw_cr("thr ms ms % ms % attempts"
78 " total alloc undo");
79 st->print_raw_cr("--- --------- --------- ------ --------- ------ --------"
80 " ------- ------- -------");
81 }
83 void
84 G1ParScanThreadState::print_termination_stats(int i,
85 outputStream* const st) const
86 {
87 const double elapsed_ms = elapsed_time() * 1000.0;
88 const double s_roots_ms = strong_roots_time() * 1000.0;
89 const double term_ms = term_time() * 1000.0;
90 const size_t alloc_buffer_waste = _g1_par_allocator->alloc_buffer_waste();
91 const size_t undo_waste = _g1_par_allocator->undo_waste();
92 st->print_cr("%3d %9.2f %9.2f %6.2f "
93 "%9.2f %6.2f " SIZE_FORMAT_W(8) " "
94 SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7),
95 i, elapsed_ms, s_roots_ms, s_roots_ms * 100 / elapsed_ms,
96 term_ms, term_ms * 100 / elapsed_ms, term_attempts(),
97 (alloc_buffer_waste + undo_waste) * HeapWordSize / K,
98 alloc_buffer_waste * HeapWordSize / K,
99 undo_waste * HeapWordSize / K);
100 }
102 #ifdef ASSERT
103 bool G1ParScanThreadState::verify_ref(narrowOop* ref) const {
104 assert(ref != NULL, "invariant");
105 assert(UseCompressedOops, "sanity");
106 assert(!has_partial_array_mask(ref), err_msg("ref=" PTR_FORMAT, p2i(ref)));
107 oop p = oopDesc::load_decode_heap_oop(ref);
108 assert(_g1h->is_in_g1_reserved(p),
109 err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
110 return true;
111 }
113 bool G1ParScanThreadState::verify_ref(oop* ref) const {
114 assert(ref != NULL, "invariant");
115 if (has_partial_array_mask(ref)) {
116 // Must be in the collection set--it's already been copied.
117 oop p = clear_partial_array_mask(ref);
118 assert(_g1h->obj_in_cs(p),
119 err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
120 } else {
121 oop p = oopDesc::load_decode_heap_oop(ref);
122 assert(_g1h->is_in_g1_reserved(p),
123 err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
124 }
125 return true;
126 }
128 bool G1ParScanThreadState::verify_task(StarTask ref) const {
129 if (ref.is_narrow()) {
130 return verify_ref((narrowOop*) ref);
131 } else {
132 return verify_ref((oop*) ref);
133 }
134 }
135 #endif // ASSERT
137 void G1ParScanThreadState::trim_queue() {
138 assert(_evac_failure_cl != NULL, "not set");
140 StarTask ref;
141 do {
142 // Drain the overflow stack first, so other threads can steal.
143 while (_refs->pop_overflow(ref)) {
144 dispatch_reference(ref);
145 }
147 while (_refs->pop_local(ref)) {
148 dispatch_reference(ref);
149 }
150 } while (!_refs->is_empty());
151 }
153 oop G1ParScanThreadState::copy_to_survivor_space(oop const old) {
154 size_t word_sz = old->size();
155 HeapRegion* from_region = _g1h->heap_region_containing_raw(old);
156 // +1 to make the -1 indexes valid...
157 int young_index = from_region->young_index_in_cset()+1;
158 assert( (from_region->is_young() && young_index > 0) ||
159 (!from_region->is_young() && young_index == 0), "invariant" );
160 G1CollectorPolicy* g1p = _g1h->g1_policy();
161 markOop m = old->mark();
162 int age = m->has_displaced_mark_helper() ? m->displaced_mark_helper()->age()
163 : m->age();
164 GCAllocPurpose alloc_purpose = g1p->evacuation_destination(from_region, age,
165 word_sz);
166 AllocationContext_t context = from_region->allocation_context();
167 HeapWord* obj_ptr = _g1_par_allocator->allocate(alloc_purpose, word_sz, context);
168 #ifndef PRODUCT
169 // Should this evacuation fail?
170 if (_g1h->evacuation_should_fail()) {
171 if (obj_ptr != NULL) {
172 _g1_par_allocator->undo_allocation(alloc_purpose, obj_ptr, word_sz, context);
173 obj_ptr = NULL;
174 }
175 }
176 #endif // !PRODUCT
178 if (obj_ptr == NULL) {
179 // This will either forward-to-self, or detect that someone else has
180 // installed a forwarding pointer.
181 return _g1h->handle_evacuation_failure_par(this, old);
182 }
184 oop obj = oop(obj_ptr);
186 // We're going to allocate linearly, so might as well prefetch ahead.
187 Prefetch::write(obj_ptr, PrefetchCopyIntervalInBytes);
189 oop forward_ptr = old->forward_to_atomic(obj);
190 if (forward_ptr == NULL) {
191 Copy::aligned_disjoint_words((HeapWord*) old, obj_ptr, word_sz);
193 // alloc_purpose is just a hint to allocate() above, recheck the type of region
194 // we actually allocated from and update alloc_purpose accordingly
195 HeapRegion* to_region = _g1h->heap_region_containing_raw(obj_ptr);
196 alloc_purpose = to_region->is_young() ? GCAllocForSurvived : GCAllocForTenured;
198 if (g1p->track_object_age(alloc_purpose)) {
199 // We could simply do obj->incr_age(). However, this causes a
200 // performance issue. obj->incr_age() will first check whether
201 // the object has a displaced mark by checking its mark word;
202 // getting the mark word from the new location of the object
203 // stalls. So, given that we already have the mark word and we
204 // are about to install it anyway, it's better to increase the
205 // age on the mark word, when the object does not have a
206 // displaced mark word. We're not expecting many objects to have
207 // a displaced marked word, so that case is not optimized
208 // further (it could be...) and we simply call obj->incr_age().
210 if (m->has_displaced_mark_helper()) {
211 // in this case, we have to install the mark word first,
212 // otherwise obj looks to be forwarded (the old mark word,
213 // which contains the forward pointer, was copied)
214 obj->set_mark(m);
215 obj->incr_age();
216 } else {
217 m = m->incr_age();
218 obj->set_mark(m);
219 }
220 age_table()->add(obj, word_sz);
221 } else {
222 obj->set_mark(m);
223 }
225 if (G1StringDedup::is_enabled()) {
226 G1StringDedup::enqueue_from_evacuation(from_region->is_young(),
227 to_region->is_young(),
228 queue_num(),
229 obj);
230 }
232 size_t* surv_young_words = surviving_young_words();
233 surv_young_words[young_index] += word_sz;
235 if (obj->is_objArray() && arrayOop(obj)->length() >= ParGCArrayScanChunk) {
236 // We keep track of the next start index in the length field of
237 // the to-space object. The actual length can be found in the
238 // length field of the from-space object.
239 arrayOop(obj)->set_length(0);
240 oop* old_p = set_partial_array_mask(old);
241 push_on_queue(old_p);
242 } else {
243 // No point in using the slower heap_region_containing() method,
244 // given that we know obj is in the heap.
245 _scanner.set_region(_g1h->heap_region_containing_raw(obj));
246 obj->oop_iterate_backwards(&_scanner);
247 }
248 } else {
249 _g1_par_allocator->undo_allocation(alloc_purpose, obj_ptr, word_sz, context);
250 obj = forward_ptr;
251 }
252 return obj;
253 }