Mercurial > hg > truffle
annotate src/share/vm/memory/threadLocalAllocBuffer.cpp @ 15177:66e3af78ea96
HSAIL: added safepoint support
Contributed-by: Eric Caspole <eric.caspole@amd.com>
author | Doug Simon <doug.simon@oracle.com> |
---|---|
date | Thu, 17 Apr 2014 00:44:32 +0200 |
parents | 4ca6dc0799b6 |
children | a29e6e7b7a86 |
rev | line source |
---|---|
0 | 1 /* |
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
1972
diff
changeset
|
2 * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved. |
0 | 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 * | |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1165
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1165
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1165
diff
changeset
|
21 * questions. |
0 | 22 * |
23 */ | |
24 | |
1972 | 25 #include "precompiled.hpp" |
26 #include "memory/genCollectedHeap.hpp" | |
27 #include "memory/resourceArea.hpp" | |
28 #include "memory/threadLocalAllocBuffer.inline.hpp" | |
29 #include "memory/universe.inline.hpp" | |
30 #include "oops/oop.inline.hpp" | |
7180
f34d701e952e
8003935: Simplify the needed includes for using Thread::current()
stefank
parents:
3960
diff
changeset
|
31 #include "runtime/thread.inline.hpp" |
1972 | 32 #include "utilities/copy.hpp" |
0 | 33 |
1972 | 34 // Thread-Local Edens support |
0 | 35 |
36 // static member initialization | |
37 unsigned ThreadLocalAllocBuffer::_target_refills = 0; | |
38 GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats = NULL; | |
39 | |
40 void ThreadLocalAllocBuffer::clear_before_allocation() { | |
41 _slow_refill_waste += (unsigned)remaining(); | |
42 make_parsable(true); // also retire the TLAB | |
43 } | |
44 | |
45 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() { | |
46 global_stats()->initialize(); | |
47 | |
14909 | 48 for(JavaThread *thread = Threads::first(); thread; thread = thread->next()) { |
0 | 49 thread->tlab().accumulate_statistics(); |
50 thread->tlab().initialize_statistics(); | |
51 } | |
52 | |
53 // Publish new stats if some allocation occurred. | |
54 if (global_stats()->allocation() != 0) { | |
55 global_stats()->publish(); | |
56 if (PrintTLAB) { | |
57 global_stats()->print(); | |
58 } | |
59 } | |
60 } | |
61 | |
62 void ThreadLocalAllocBuffer::accumulate_statistics() { | |
14909 | 63 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize; |
64 size_t unused = Universe::heap()->unsafe_max_tlab_alloc(myThread()) / HeapWordSize; | |
65 size_t used = capacity - unused; | |
66 | |
67 // Update allocation history if a reasonable amount of eden was allocated. | |
68 bool update_allocation_history = used > 0.5 * capacity; | |
0 | 69 |
70 _gc_waste += (unsigned)remaining(); | |
71 | |
72 if (PrintTLAB && (_number_of_refills > 0 || Verbose)) { | |
73 print_stats("gc"); | |
74 } | |
75 | |
76 if (_number_of_refills > 0) { | |
77 | |
78 if (update_allocation_history) { | |
79 // Average the fraction of eden allocated in a tlab by this | |
80 // thread for use in the next resize operation. | |
81 // _gc_waste is not subtracted because it's included in | |
82 // "used". | |
14909 | 83 size_t allocation = _number_of_refills * desired_size(); |
84 double alloc_frac = allocation / (double) used; | |
0 | 85 _allocation_fraction.sample(alloc_frac); |
86 } | |
87 global_stats()->update_allocating_threads(); | |
88 global_stats()->update_number_of_refills(_number_of_refills); | |
89 global_stats()->update_allocation(_number_of_refills * desired_size()); | |
90 global_stats()->update_gc_waste(_gc_waste); | |
91 global_stats()->update_slow_refill_waste(_slow_refill_waste); | |
92 global_stats()->update_fast_refill_waste(_fast_refill_waste); | |
93 | |
94 } else { | |
7236
579ee1eb0f30
fixed overly strong assertion
Doug Simon <doug.simon@oracle.com>
parents:
7180
diff
changeset
|
95 // (ds) _gc_waste can be non-zero (see above) even if _number_of_refills is 0 |
0 | 96 assert(_number_of_refills == 0 && _fast_refill_waste == 0 && |
7236
579ee1eb0f30
fixed overly strong assertion
Doug Simon <doug.simon@oracle.com>
parents:
7180
diff
changeset
|
97 _slow_refill_waste == 0/* && _gc_waste == 0*/, |
0 | 98 "tlab stats == 0"); |
99 } | |
100 global_stats()->update_slow_allocations(_slow_allocations); | |
101 } | |
102 | |
103 // Fills the current tlab with a dummy filler array to create | |
104 // an illusion of a contiguous Eden and optionally retires the tlab. | |
105 // Waste accounting should be done in caller as appropriate; see, | |
106 // for example, clear_before_allocation(). | |
107 void ThreadLocalAllocBuffer::make_parsable(bool retire) { | |
108 if (end() != NULL) { | |
109 invariants(); | |
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
1972
diff
changeset
|
110 |
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
1972
diff
changeset
|
111 if (retire) { |
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
1972
diff
changeset
|
112 myThread()->incr_allocated_bytes(used_bytes()); |
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
1972
diff
changeset
|
113 } |
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
1972
diff
changeset
|
114 |
1165
2dd52dea6d28
6902115: G1:assert(ignore_max_completed||thread->is_Java_thread()||SafepointSynchronize::is_at_safepoint())
johnc
parents:
579
diff
changeset
|
115 CollectedHeap::fill_with_object(top(), hard_end(), retire); |
0 | 116 |
117 if (retire || ZeroTLAB) { // "Reset" the TLAB | |
118 set_start(NULL); | |
119 set_top(NULL); | |
120 set_pf_top(NULL); | |
121 set_end(NULL); | |
122 } | |
123 } | |
124 assert(!(retire || ZeroTLAB) || | |
125 (start() == NULL && end() == NULL && top() == NULL), | |
126 "TLAB must be reset"); | |
127 } | |
128 | |
129 void ThreadLocalAllocBuffer::resize_all_tlabs() { | |
14909 | 130 for(JavaThread *thread = Threads::first(); thread; thread = thread->next()) { |
131 thread->tlab().resize(); | |
0 | 132 } |
133 } | |
134 | |
135 void ThreadLocalAllocBuffer::resize() { | |
136 | |
14909 | 137 if (ResizeTLAB) { |
138 // Compute the next tlab size using expected allocation amount | |
139 size_t alloc = (size_t)(_allocation_fraction.average() * | |
140 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize)); | |
141 size_t new_size = alloc / _target_refills; | |
142 | |
143 new_size = MIN2(MAX2(new_size, min_size()), max_size()); | |
144 | |
145 size_t aligned_new_size = align_object_size(new_size); | |
0 | 146 |
14909 | 147 if (PrintTLAB && Verbose) { |
148 gclog_or_tty->print("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]" | |
149 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT "\n", | |
150 myThread(), myThread()->osthread()->thread_id(), | |
151 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size); | |
152 } | |
153 set_desired_size(aligned_new_size); | |
0 | 154 |
14909 | 155 set_refill_waste_limit(initial_refill_waste_limit()); |
0 | 156 } |
157 } | |
158 | |
159 void ThreadLocalAllocBuffer::initialize_statistics() { | |
160 _number_of_refills = 0; | |
161 _fast_refill_waste = 0; | |
162 _slow_refill_waste = 0; | |
163 _gc_waste = 0; | |
164 _slow_allocations = 0; | |
165 } | |
166 | |
167 void ThreadLocalAllocBuffer::fill(HeapWord* start, | |
168 HeapWord* top, | |
169 size_t new_size) { | |
170 _number_of_refills++; | |
171 if (PrintTLAB && Verbose) { | |
172 print_stats("fill"); | |
173 } | |
174 assert(top <= start + new_size - alignment_reserve(), "size too small"); | |
175 initialize(start, top, start + new_size - alignment_reserve()); | |
176 | |
177 // Reset amount of internal fragmentation | |
178 set_refill_waste_limit(initial_refill_waste_limit()); | |
179 } | |
180 | |
181 void ThreadLocalAllocBuffer::initialize(HeapWord* start, | |
182 HeapWord* top, | |
183 HeapWord* end) { | |
184 set_start(start); | |
185 set_top(top); | |
186 set_pf_top(top); | |
187 set_end(end); | |
188 invariants(); | |
189 } | |
190 | |
191 void ThreadLocalAllocBuffer::initialize() { | |
192 initialize(NULL, // start | |
193 NULL, // top | |
194 NULL); // end | |
195 | |
196 set_desired_size(initial_desired_size()); | |
197 | |
198 // Following check is needed because at startup the main (primordial) | |
199 // thread is initialized before the heap is. The initialization for | |
200 // this thread is redone in startup_initialization below. | |
201 if (Universe::heap() != NULL) { | |
202 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize; | |
203 double alloc_frac = desired_size() * target_refills() / (double) capacity; | |
204 _allocation_fraction.sample(alloc_frac); | |
205 } | |
206 | |
207 set_refill_waste_limit(initial_refill_waste_limit()); | |
208 | |
209 initialize_statistics(); | |
210 } | |
211 | |
212 void ThreadLocalAllocBuffer::startup_initialization() { | |
213 | |
214 // Assuming each thread's active tlab is, on average, | |
215 // 1/2 full at a GC | |
216 _target_refills = 100 / (2 * TLABWasteTargetPercent); | |
217 _target_refills = MAX2(_target_refills, (unsigned)1U); | |
218 | |
219 _global_stats = new GlobalTLABStats(); | |
220 | |
221 // During jvm startup, the main (primordial) thread is initialized | |
222 // before the heap is initialized. So reinitialize it now. | |
223 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread"); | |
224 Thread::current()->tlab().initialize(); | |
225 | |
226 if (PrintTLAB && Verbose) { | |
227 gclog_or_tty->print("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT "\n", | |
228 min_size(), Thread::current()->tlab().initial_desired_size(), max_size()); | |
229 } | |
230 } | |
231 | |
232 size_t ThreadLocalAllocBuffer::initial_desired_size() { | |
233 size_t init_sz; | |
234 | |
235 if (TLABSize > 0) { | |
236 init_sz = MIN2(TLABSize / HeapWordSize, max_size()); | |
237 } else if (global_stats() == NULL) { | |
238 // Startup issue - main thread initialized before heap initialized. | |
239 init_sz = min_size(); | |
240 } else { | |
241 // Initial size is a function of the average number of allocating threads. | |
242 unsigned nof_threads = global_stats()->allocating_threads_avg(); | |
243 | |
244 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) / | |
245 (nof_threads * target_refills()); | |
246 init_sz = align_object_size(init_sz); | |
247 init_sz = MIN2(MAX2(init_sz, min_size()), max_size()); | |
248 } | |
249 return init_sz; | |
250 } | |
251 | |
14909 | 252 const size_t ThreadLocalAllocBuffer::max_size() { |
253 | |
254 // TLABs can't be bigger than we can fill with a int[Integer.MAX_VALUE]. | |
255 // This restriction could be removed by enabling filling with multiple arrays. | |
256 // If we compute that the reasonable way as | |
257 // header_size + ((sizeof(jint) * max_jint) / HeapWordSize) | |
258 // we'll overflow on the multiply, so we do the divide first. | |
259 // We actually lose a little by dividing first, | |
260 // but that just makes the TLAB somewhat smaller than the biggest array, | |
261 // which is fine, since we'll be able to fill that. | |
262 | |
263 size_t unaligned_max_size = typeArrayOopDesc::header_size(T_INT) + | |
264 sizeof(jint) * | |
265 ((juint) max_jint / (size_t) HeapWordSize); | |
266 return align_size_down(unaligned_max_size, MinObjAlignment); | |
267 } | |
268 | |
0 | 269 void ThreadLocalAllocBuffer::print_stats(const char* tag) { |
270 Thread* thrd = myThread(); | |
271 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste; | |
272 size_t alloc = _number_of_refills * _desired_size; | |
273 double waste_percent = alloc == 0 ? 0.0 : | |
274 100.0 * waste / alloc; | |
14909 | 275 size_t tlab_used = Universe::heap()->tlab_capacity(thrd) - |
276 Universe::heap()->unsafe_max_tlab_alloc(thrd); | |
0 | 277 gclog_or_tty->print("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]" |
278 " desired_size: " SIZE_FORMAT "KB" | |
279 " slow allocs: %d refill waste: " SIZE_FORMAT "B" | |
280 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB" | |
281 " slow: %dB fast: %dB\n", | |
282 tag, thrd, thrd->osthread()->thread_id(), | |
283 _desired_size / (K / HeapWordSize), | |
284 _slow_allocations, _refill_waste_limit * HeapWordSize, | |
285 _allocation_fraction.average(), | |
286 _allocation_fraction.average() * tlab_used / K, | |
287 _number_of_refills, waste_percent, | |
288 _gc_waste * HeapWordSize, | |
289 _slow_refill_waste * HeapWordSize, | |
290 _fast_refill_waste * HeapWordSize); | |
291 } | |
292 | |
293 void ThreadLocalAllocBuffer::verify() { | |
294 HeapWord* p = start(); | |
295 HeapWord* t = top(); | |
296 HeapWord* prev_p = NULL; | |
297 while (p < t) { | |
298 oop(p)->verify(); | |
299 prev_p = p; | |
300 p += oop(p)->size(); | |
301 } | |
302 guarantee(p == top(), "end of last object must match end of space"); | |
303 } | |
304 | |
305 Thread* ThreadLocalAllocBuffer::myThread() { | |
306 return (Thread*)(((char *)this) + | |
307 in_bytes(start_offset()) - | |
308 in_bytes(Thread::tlab_start_offset())); | |
309 } | |
310 | |
311 | |
312 GlobalTLABStats::GlobalTLABStats() : | |
313 _allocating_threads_avg(TLABAllocationWeight) { | |
314 | |
315 initialize(); | |
316 | |
317 _allocating_threads_avg.sample(1); // One allocating thread at startup | |
318 | |
319 if (UsePerfData) { | |
320 | |
321 EXCEPTION_MARK; | |
322 ResourceMark rm; | |
323 | |
324 char* cname = PerfDataManager::counter_name("tlab", "allocThreads"); | |
325 _perf_allocating_threads = | |
326 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); | |
327 | |
328 cname = PerfDataManager::counter_name("tlab", "fills"); | |
329 _perf_total_refills = | |
330 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); | |
331 | |
332 cname = PerfDataManager::counter_name("tlab", "maxFills"); | |
333 _perf_max_refills = | |
334 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); | |
335 | |
336 cname = PerfDataManager::counter_name("tlab", "alloc"); | |
337 _perf_allocation = | |
338 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); | |
339 | |
340 cname = PerfDataManager::counter_name("tlab", "gcWaste"); | |
341 _perf_gc_waste = | |
342 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); | |
343 | |
344 cname = PerfDataManager::counter_name("tlab", "maxGcWaste"); | |
345 _perf_max_gc_waste = | |
346 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); | |
347 | |
348 cname = PerfDataManager::counter_name("tlab", "slowWaste"); | |
349 _perf_slow_refill_waste = | |
350 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); | |
351 | |
352 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste"); | |
353 _perf_max_slow_refill_waste = | |
354 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); | |
355 | |
356 cname = PerfDataManager::counter_name("tlab", "fastWaste"); | |
357 _perf_fast_refill_waste = | |
358 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); | |
359 | |
360 cname = PerfDataManager::counter_name("tlab", "maxFastWaste"); | |
361 _perf_max_fast_refill_waste = | |
362 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); | |
363 | |
364 cname = PerfDataManager::counter_name("tlab", "slowAlloc"); | |
365 _perf_slow_allocations = | |
366 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); | |
367 | |
368 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc"); | |
369 _perf_max_slow_allocations = | |
370 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); | |
371 } | |
372 } | |
373 | |
374 void GlobalTLABStats::initialize() { | |
375 // Clear counters summarizing info from all threads | |
376 _allocating_threads = 0; | |
377 _total_refills = 0; | |
378 _max_refills = 0; | |
379 _total_allocation = 0; | |
380 _total_gc_waste = 0; | |
381 _max_gc_waste = 0; | |
382 _total_slow_refill_waste = 0; | |
383 _max_slow_refill_waste = 0; | |
384 _total_fast_refill_waste = 0; | |
385 _max_fast_refill_waste = 0; | |
386 _total_slow_allocations = 0; | |
387 _max_slow_allocations = 0; | |
388 } | |
389 | |
390 void GlobalTLABStats::publish() { | |
391 _allocating_threads_avg.sample(_allocating_threads); | |
392 if (UsePerfData) { | |
393 _perf_allocating_threads ->set_value(_allocating_threads); | |
394 _perf_total_refills ->set_value(_total_refills); | |
395 _perf_max_refills ->set_value(_max_refills); | |
396 _perf_allocation ->set_value(_total_allocation); | |
397 _perf_gc_waste ->set_value(_total_gc_waste); | |
398 _perf_max_gc_waste ->set_value(_max_gc_waste); | |
399 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste); | |
400 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste); | |
401 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste); | |
402 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste); | |
403 _perf_slow_allocations ->set_value(_total_slow_allocations); | |
404 _perf_max_slow_allocations ->set_value(_max_slow_allocations); | |
405 } | |
406 } | |
407 | |
408 void GlobalTLABStats::print() { | |
409 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste; | |
410 double waste_percent = _total_allocation == 0 ? 0.0 : | |
411 100.0 * waste / _total_allocation; | |
412 gclog_or_tty->print("TLAB totals: thrds: %d refills: %d max: %d" | |
413 " slow allocs: %d max %d waste: %4.1f%%" | |
414 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B" | |
415 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B" | |
416 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B\n", | |
417 _allocating_threads, | |
418 _total_refills, _max_refills, | |
419 _total_slow_allocations, _max_slow_allocations, | |
420 waste_percent, | |
421 _total_gc_waste * HeapWordSize, | |
422 _max_gc_waste * HeapWordSize, | |
423 _total_slow_refill_waste * HeapWordSize, | |
424 _max_slow_refill_waste * HeapWordSize, | |
425 _total_fast_refill_waste * HeapWordSize, | |
426 _max_fast_refill_waste * HeapWordSize); | |
427 } |