Mercurial > hg > graal-compiler
annotate src/share/vm/gc_implementation/parNew/parCardTableModRefBS.cpp @ 1833:8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
Summary: Associate number of GC workers with the workgang as opposed to the task.
Reviewed-by: johnc, ysr
| author | jmasa |
|---|---|
| date | Mon, 20 Sep 2010 14:38:38 -0700 |
| parents | c18cbe5936b8 |
| children | f95d63e2154a |
| rev | line source |
|---|---|
| 0 | 1 /* |
|
1833
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1552
diff
changeset
|
2 * Copyright (c) 2007, 2010 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:
845
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
845
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:
845
diff
changeset
|
21 * questions. |
| 0 | 22 * |
| 23 */ | |
| 24 | |
| 25 # include "incls/_precompiled.incl" | |
| 26 # include "incls/_parCardTableModRefBS.cpp.incl" | |
| 27 | |
| 28 void CardTableModRefBS::par_non_clean_card_iterate_work(Space* sp, MemRegion mr, | |
| 29 DirtyCardToOopClosure* dcto_cl, | |
| 30 MemRegionClosure* cl, | |
| 31 bool clear, | |
| 32 int n_threads) { | |
| 33 if (n_threads > 0) { | |
| 845 | 34 assert((n_threads == 1 && ParallelGCThreads == 0) || |
| 35 n_threads <= (int)ParallelGCThreads, | |
| 36 "# worker threads != # requested!"); | |
| 37 // Make sure the LNC array is valid for the space. | |
| 0 | 38 jbyte** lowest_non_clean; |
| 39 uintptr_t lowest_non_clean_base_chunk_index; | |
| 40 size_t lowest_non_clean_chunk_size; | |
| 41 get_LNC_array_for_space(sp, lowest_non_clean, | |
| 42 lowest_non_clean_base_chunk_index, | |
| 43 lowest_non_clean_chunk_size); | |
| 44 | |
| 45 int n_strides = n_threads * StridesPerThread; | |
| 46 SequentialSubTasksDone* pst = sp->par_seq_tasks(); | |
|
1833
8b10f48633dc
6984287: Regularize how GC parallel workers are specified.
jmasa
parents:
1552
diff
changeset
|
47 pst->set_n_threads(n_threads); |
| 0 | 48 pst->set_n_tasks(n_strides); |
| 49 | |
| 50 int stride = 0; | |
| 51 while (!pst->is_task_claimed(/* reference */ stride)) { | |
| 52 process_stride(sp, mr, stride, n_strides, dcto_cl, cl, clear, | |
| 53 lowest_non_clean, | |
| 54 lowest_non_clean_base_chunk_index, | |
| 55 lowest_non_clean_chunk_size); | |
| 56 } | |
| 57 if (pst->all_tasks_completed()) { | |
| 58 // Clear lowest_non_clean array for next time. | |
| 59 intptr_t first_chunk_index = addr_to_chunk_index(mr.start()); | |
| 60 uintptr_t last_chunk_index = addr_to_chunk_index(mr.last()); | |
| 61 for (uintptr_t ch = first_chunk_index; ch <= last_chunk_index; ch++) { | |
| 62 intptr_t ind = ch - lowest_non_clean_base_chunk_index; | |
| 63 assert(0 <= ind && ind < (intptr_t)lowest_non_clean_chunk_size, | |
| 64 "Bounds error"); | |
| 65 lowest_non_clean[ind] = NULL; | |
| 66 } | |
| 67 } | |
| 68 } | |
| 69 } | |
| 70 | |
| 71 void | |
| 72 CardTableModRefBS:: | |
| 73 process_stride(Space* sp, | |
| 74 MemRegion used, | |
| 75 jint stride, int n_strides, | |
| 76 DirtyCardToOopClosure* dcto_cl, | |
| 77 MemRegionClosure* cl, | |
| 78 bool clear, | |
| 79 jbyte** lowest_non_clean, | |
| 80 uintptr_t lowest_non_clean_base_chunk_index, | |
| 81 size_t lowest_non_clean_chunk_size) { | |
| 82 // We don't have to go downwards here; it wouldn't help anyway, | |
| 83 // because of parallelism. | |
| 84 | |
| 85 // Find the first card address of the first chunk in the stride that is | |
| 86 // at least "bottom" of the used region. | |
| 87 jbyte* start_card = byte_for(used.start()); | |
| 88 jbyte* end_card = byte_after(used.last()); | |
| 89 uintptr_t start_chunk = addr_to_chunk_index(used.start()); | |
| 90 uintptr_t start_chunk_stride_num = start_chunk % n_strides; | |
| 91 jbyte* chunk_card_start; | |
| 92 | |
| 93 if ((uintptr_t)stride >= start_chunk_stride_num) { | |
| 94 chunk_card_start = (jbyte*)(start_card + | |
| 95 (stride - start_chunk_stride_num) * | |
| 96 CardsPerStrideChunk); | |
| 97 } else { | |
| 98 // Go ahead to the next chunk group boundary, then to the requested stride. | |
| 99 chunk_card_start = (jbyte*)(start_card + | |
| 100 (n_strides - start_chunk_stride_num + stride) * | |
| 101 CardsPerStrideChunk); | |
| 102 } | |
| 103 | |
| 104 while (chunk_card_start < end_card) { | |
| 105 // We don't have to go downwards here; it wouldn't help anyway, | |
| 106 // because of parallelism. (We take care with "min_done"; see below.) | |
| 107 // Invariant: chunk_mr should be fully contained within the "used" region. | |
| 108 jbyte* chunk_card_end = chunk_card_start + CardsPerStrideChunk; | |
| 109 MemRegion chunk_mr = MemRegion(addr_for(chunk_card_start), | |
| 110 chunk_card_end >= end_card ? | |
| 111 used.end() : addr_for(chunk_card_end)); | |
| 112 assert(chunk_mr.word_size() > 0, "[chunk_card_start > used_end)"); | |
| 113 assert(used.contains(chunk_mr), "chunk_mr should be subset of used"); | |
| 114 | |
| 115 // Process the chunk. | |
| 116 process_chunk_boundaries(sp, | |
| 117 dcto_cl, | |
| 118 chunk_mr, | |
| 119 used, | |
| 120 lowest_non_clean, | |
| 121 lowest_non_clean_base_chunk_index, | |
| 122 lowest_non_clean_chunk_size); | |
| 123 | |
| 124 non_clean_card_iterate_work(chunk_mr, cl, clear); | |
| 125 | |
| 126 // Find the next chunk of the stride. | |
| 127 chunk_card_start += CardsPerStrideChunk * n_strides; | |
| 128 } | |
| 129 } | |
| 130 | |
| 131 void | |
| 132 CardTableModRefBS:: | |
| 133 process_chunk_boundaries(Space* sp, | |
| 134 DirtyCardToOopClosure* dcto_cl, | |
| 135 MemRegion chunk_mr, | |
| 136 MemRegion used, | |
| 137 jbyte** lowest_non_clean, | |
| 138 uintptr_t lowest_non_clean_base_chunk_index, | |
| 139 size_t lowest_non_clean_chunk_size) | |
| 140 { | |
| 141 // We must worry about the chunk boundaries. | |
| 142 | |
| 143 // First, set our max_to_do: | |
| 144 HeapWord* max_to_do = NULL; | |
| 145 uintptr_t cur_chunk_index = addr_to_chunk_index(chunk_mr.start()); | |
| 146 cur_chunk_index = cur_chunk_index - lowest_non_clean_base_chunk_index; | |
| 147 | |
| 148 if (chunk_mr.end() < used.end()) { | |
| 149 // This is not the last chunk in the used region. What is the last | |
| 150 // object? | |
| 151 HeapWord* last_block = sp->block_start(chunk_mr.end()); | |
| 152 assert(last_block <= chunk_mr.end(), "In case this property changes."); | |
| 153 if (last_block == chunk_mr.end() | |
| 154 || !sp->block_is_obj(last_block)) { | |
| 155 max_to_do = chunk_mr.end(); | |
| 156 | |
| 157 } else { | |
| 158 // It is an object and starts before the end of the current chunk. | |
| 159 // last_obj_card is the card corresponding to the start of the last object | |
| 160 // in the chunk. Note that the last object may not start in | |
| 161 // the chunk. | |
| 162 jbyte* last_obj_card = byte_for(last_block); | |
| 163 if (!card_may_have_been_dirty(*last_obj_card)) { | |
| 164 // The card containing the head is not dirty. Any marks in | |
| 165 // subsequent cards still in this chunk must have been made | |
| 166 // precisely; we can cap processing at the end. | |
| 167 max_to_do = chunk_mr.end(); | |
| 168 } else { | |
| 169 // The last object must be considered dirty, and extends onto the | |
| 170 // following chunk. Look for a dirty card in that chunk that will | |
| 171 // bound our processing. | |
| 172 jbyte* limit_card = NULL; | |
| 173 size_t last_block_size = sp->block_size(last_block); | |
| 174 jbyte* last_card_of_last_obj = | |
| 175 byte_for(last_block + last_block_size - 1); | |
| 176 jbyte* first_card_of_next_chunk = byte_for(chunk_mr.end()); | |
| 177 // This search potentially goes a long distance looking | |
| 178 // for the next card that will be scanned. For example, | |
| 179 // an object that is an array of primitives will not | |
| 180 // have any cards covering regions interior to the array | |
| 181 // that will need to be scanned. The scan can be terminated | |
| 182 // at the last card of the next chunk. That would leave | |
| 183 // limit_card as NULL and would result in "max_to_do" | |
| 184 // being set with the LNC value or with the end | |
| 185 // of the last block. | |
| 186 jbyte* last_card_of_next_chunk = first_card_of_next_chunk + | |
| 187 CardsPerStrideChunk; | |
| 188 assert(byte_for(chunk_mr.end()) - byte_for(chunk_mr.start()) | |
| 189 == CardsPerStrideChunk, "last card of next chunk may be wrong"); | |
| 190 jbyte* last_card_to_check = (jbyte*) MIN2(last_card_of_last_obj, | |
| 191 last_card_of_next_chunk); | |
| 192 for (jbyte* cur = first_card_of_next_chunk; | |
| 193 cur <= last_card_to_check; cur++) { | |
| 194 if (card_will_be_scanned(*cur)) { | |
| 195 limit_card = cur; break; | |
| 196 } | |
| 197 } | |
| 198 assert(0 <= cur_chunk_index+1 && | |
| 199 cur_chunk_index+1 < lowest_non_clean_chunk_size, | |
| 200 "Bounds error."); | |
| 201 // LNC for the next chunk | |
| 202 jbyte* lnc_card = lowest_non_clean[cur_chunk_index+1]; | |
| 203 if (limit_card == NULL) { | |
| 204 limit_card = lnc_card; | |
| 205 } | |
| 206 if (limit_card != NULL) { | |
| 207 if (lnc_card != NULL) { | |
| 208 limit_card = (jbyte*)MIN2((intptr_t)limit_card, | |
| 209 (intptr_t)lnc_card); | |
| 210 } | |
| 211 max_to_do = addr_for(limit_card); | |
| 212 } else { | |
| 213 max_to_do = last_block + last_block_size; | |
| 214 } | |
| 215 } | |
| 216 } | |
| 217 assert(max_to_do != NULL, "OOPS!"); | |
| 218 } else { | |
| 219 max_to_do = used.end(); | |
| 220 } | |
| 221 // Now we can set the closure we're using so it doesn't to beyond | |
| 222 // max_to_do. | |
| 223 dcto_cl->set_min_done(max_to_do); | |
| 224 #ifndef PRODUCT | |
| 225 dcto_cl->set_last_bottom(max_to_do); | |
| 226 #endif | |
| 227 | |
| 228 // Now we set *our" lowest_non_clean entry. | |
| 229 // Find the object that spans our boundary, if one exists. | |
| 230 // Nothing to do on the first chunk. | |
| 231 if (chunk_mr.start() > used.start()) { | |
| 232 // first_block is the block possibly spanning the chunk start | |
| 233 HeapWord* first_block = sp->block_start(chunk_mr.start()); | |
| 234 // Does the block span the start of the chunk and is it | |
| 235 // an object? | |
| 236 if (first_block < chunk_mr.start() && | |
| 237 sp->block_is_obj(first_block)) { | |
| 238 jbyte* first_dirty_card = NULL; | |
| 239 jbyte* last_card_of_first_obj = | |
| 240 byte_for(first_block + sp->block_size(first_block) - 1); | |
| 241 jbyte* first_card_of_cur_chunk = byte_for(chunk_mr.start()); | |
| 242 jbyte* last_card_of_cur_chunk = byte_for(chunk_mr.last()); | |
| 243 jbyte* last_card_to_check = | |
| 244 (jbyte*) MIN2((intptr_t) last_card_of_cur_chunk, | |
| 245 (intptr_t) last_card_of_first_obj); | |
| 246 for (jbyte* cur = first_card_of_cur_chunk; | |
| 247 cur <= last_card_to_check; cur++) { | |
| 248 if (card_will_be_scanned(*cur)) { | |
| 249 first_dirty_card = cur; break; | |
| 250 } | |
| 251 } | |
| 252 if (first_dirty_card != NULL) { | |
| 253 assert(0 <= cur_chunk_index && | |
| 254 cur_chunk_index < lowest_non_clean_chunk_size, | |
| 255 "Bounds error."); | |
| 256 lowest_non_clean[cur_chunk_index] = first_dirty_card; | |
| 257 } | |
| 258 } | |
| 259 } | |
| 260 } | |
| 261 | |
| 262 void | |
| 263 CardTableModRefBS:: | |
| 264 get_LNC_array_for_space(Space* sp, | |
| 265 jbyte**& lowest_non_clean, | |
| 266 uintptr_t& lowest_non_clean_base_chunk_index, | |
| 267 size_t& lowest_non_clean_chunk_size) { | |
| 268 | |
| 269 int i = find_covering_region_containing(sp->bottom()); | |
| 270 MemRegion covered = _covered[i]; | |
| 271 size_t n_chunks = chunks_to_cover(covered); | |
| 272 | |
| 273 // Only the first thread to obtain the lock will resize the | |
| 274 // LNC array for the covered region. Any later expansion can't affect | |
| 275 // the used_at_save_marks region. | |
| 276 // (I observed a bug in which the first thread to execute this would | |
| 277 // resize, and then it would cause "expand_and_allocates" that would | |
| 278 // Increase the number of chunks in the covered region. Then a second | |
| 279 // thread would come and execute this, see that the size didn't match, | |
| 280 // and free and allocate again. So the first thread would be using a | |
| 281 // freed "_lowest_non_clean" array.) | |
| 282 | |
| 283 // Do a dirty read here. If we pass the conditional then take the rare | |
| 284 // event lock and do the read again in case some other thread had already | |
| 285 // succeeded and done the resize. | |
| 286 int cur_collection = Universe::heap()->total_collections(); | |
| 287 if (_last_LNC_resizing_collection[i] != cur_collection) { | |
| 288 MutexLocker x(ParGCRareEvent_lock); | |
| 289 if (_last_LNC_resizing_collection[i] != cur_collection) { | |
| 290 if (_lowest_non_clean[i] == NULL || | |
| 291 n_chunks != _lowest_non_clean_chunk_size[i]) { | |
| 292 | |
| 293 // Should we delete the old? | |
| 294 if (_lowest_non_clean[i] != NULL) { | |
| 295 assert(n_chunks != _lowest_non_clean_chunk_size[i], | |
| 296 "logical consequence"); | |
| 297 FREE_C_HEAP_ARRAY(CardPtr, _lowest_non_clean[i]); | |
| 298 _lowest_non_clean[i] = NULL; | |
| 299 } | |
| 300 // Now allocate a new one if necessary. | |
| 301 if (_lowest_non_clean[i] == NULL) { | |
| 302 _lowest_non_clean[i] = NEW_C_HEAP_ARRAY(CardPtr, n_chunks); | |
| 303 _lowest_non_clean_chunk_size[i] = n_chunks; | |
| 304 _lowest_non_clean_base_chunk_index[i] = addr_to_chunk_index(covered.start()); | |
| 305 for (int j = 0; j < (int)n_chunks; j++) | |
| 306 _lowest_non_clean[i][j] = NULL; | |
| 307 } | |
| 308 } | |
| 309 _last_LNC_resizing_collection[i] = cur_collection; | |
| 310 } | |
| 311 } | |
| 312 // In any case, now do the initialization. | |
| 313 lowest_non_clean = _lowest_non_clean[i]; | |
| 314 lowest_non_clean_base_chunk_index = _lowest_non_clean_base_chunk_index[i]; | |
| 315 lowest_non_clean_chunk_size = _lowest_non_clean_chunk_size[i]; | |
| 316 } |