| 1 | /* |
| 2 | Copyright (c) 2005-2019 Intel Corporation |
| 3 | |
| 4 | Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | you may not use this file except in compliance with the License. |
| 6 | You may obtain a copy of the License at |
| 7 | |
| 8 | http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | |
| 10 | Unless required by applicable law or agreed to in writing, software |
| 11 | distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | See the License for the specific language governing permissions and |
| 14 | limitations under the License. |
| 15 | */ |
| 16 | |
| 17 | #include "tbb/tbb_stddef.h" |
| 18 | #include "tbb/tbb_machine.h" |
| 19 | #include "tbb/tbb_exception.h" |
| 20 | // Define required to satisfy test in internal file. |
| 21 | #define __TBB_concurrent_queue_H |
| 22 | #include "tbb/internal/_concurrent_queue_impl.h" |
| 23 | #include "concurrent_monitor.h" |
| 24 | #include "itt_notify.h" |
| 25 | #include <new> |
| 26 | #include <cstring> // for memset() |
| 27 | |
| 28 | #if defined(_MSC_VER) && defined(_Wp64) |
| 29 | // Workaround for overzealous compiler warnings in /Wp64 mode |
| 30 | #pragma warning (disable: 4267) |
| 31 | #endif |
| 32 | |
| 33 | #define RECORD_EVENTS 0 |
| 34 | |
| 35 | |
| 36 | namespace tbb { |
| 37 | |
| 38 | namespace internal { |
| 39 | |
| 40 | typedef concurrent_queue_base_v3 concurrent_queue_base; |
| 41 | |
| 42 | typedef size_t ticket; |
| 43 | |
| 44 | //! A queue using simple locking. |
| 45 | /** For efficiency, this class has no constructor. |
| 46 | The caller is expected to zero-initialize it. */ |
| 47 | struct micro_queue { |
| 48 | typedef concurrent_queue_base::page page; |
| 49 | |
| 50 | friend class micro_queue_pop_finalizer; |
| 51 | |
| 52 | atomic<page*> head_page; |
| 53 | atomic<ticket> head_counter; |
| 54 | |
| 55 | atomic<page*> tail_page; |
| 56 | atomic<ticket> tail_counter; |
| 57 | |
| 58 | spin_mutex page_mutex; |
| 59 | |
| 60 | void push( const void* item, ticket k, concurrent_queue_base& base, |
| 61 | concurrent_queue_base::copy_specifics op_type ); |
| 62 | |
| 63 | void abort_push( ticket k, concurrent_queue_base& base ); |
| 64 | |
| 65 | bool pop( void* dst, ticket k, concurrent_queue_base& base ); |
| 66 | |
| 67 | micro_queue& assign( const micro_queue& src, concurrent_queue_base& base, |
| 68 | concurrent_queue_base::copy_specifics op_type ); |
| 69 | |
| 70 | page* make_copy ( concurrent_queue_base& base, const page* src_page, size_t begin_in_page, |
| 71 | size_t end_in_page, ticket& g_index, concurrent_queue_base::copy_specifics op_type ) ; |
| 72 | |
| 73 | void make_invalid( ticket k ); |
| 74 | }; |
| 75 | |
| 76 | // we need to yank it out of micro_queue because of concurrent_queue_base::deallocate_page being virtual. |
| 77 | class micro_queue_pop_finalizer: no_copy { |
| 78 | typedef concurrent_queue_base::page page; |
| 79 | ticket my_ticket; |
| 80 | micro_queue& my_queue; |
| 81 | page* my_page; |
| 82 | concurrent_queue_base &base; |
| 83 | public: |
| 84 | micro_queue_pop_finalizer( micro_queue& queue, concurrent_queue_base& b, ticket k, page* p ) : |
| 85 | my_ticket(k), my_queue(queue), my_page(p), base(b) |
| 86 | {} |
| 87 | ~micro_queue_pop_finalizer() { |
| 88 | page* p = my_page; |
| 89 | if( p ) { |
| 90 | spin_mutex::scoped_lock lock( my_queue.page_mutex ); |
| 91 | page* q = p->next; |
| 92 | my_queue.head_page = q; |
| 93 | if( !q ) { |
| 94 | my_queue.tail_page = NULL; |
| 95 | } |
| 96 | } |
| 97 | my_queue.head_counter = my_ticket; |
| 98 | if( p ) |
| 99 | base.deallocate_page( p ); |
| 100 | } |
| 101 | }; |
| 102 | |
| 103 | struct predicate_leq { |
| 104 | ticket t; |
| 105 | predicate_leq( ticket t_ ) : t(t_) {} |
| 106 | bool operator() ( uintptr_t p ) const {return (ticket)p<=t;} |
| 107 | }; |
| 108 | |
| 109 | //! Internal representation of a ConcurrentQueue. |
| 110 | /** For efficiency, this class has no constructor. |
| 111 | The caller is expected to zero-initialize it. */ |
| 112 | class concurrent_queue_rep { |
| 113 | public: |
| 114 | private: |
| 115 | friend struct micro_queue; |
| 116 | |
| 117 | //! Approximately n_queue/golden ratio |
| 118 | static const size_t phi = 3; |
| 119 | |
| 120 | public: |
| 121 | //! Must be power of 2 |
| 122 | static const size_t n_queue = 8; |
| 123 | |
| 124 | //! Map ticket to an array index |
| 125 | static size_t index( ticket k ) { |
| 126 | return k*phi%n_queue; |
| 127 | } |
| 128 | |
| 129 | atomic<ticket> head_counter; |
| 130 | concurrent_monitor items_avail; |
| 131 | atomic<size_t> n_invalid_entries; |
| 132 | char pad1[NFS_MaxLineSize-((sizeof(atomic<ticket>)+sizeof(concurrent_monitor)+sizeof(atomic<size_t>))&(NFS_MaxLineSize-1))]; |
| 133 | |
| 134 | atomic<ticket> tail_counter; |
| 135 | concurrent_monitor slots_avail; |
| 136 | char pad2[NFS_MaxLineSize-((sizeof(atomic<ticket>)+sizeof(concurrent_monitor))&(NFS_MaxLineSize-1))]; |
| 137 | micro_queue array[n_queue]; |
| 138 | |
| 139 | micro_queue& choose( ticket k ) { |
| 140 | // The formula here approximates LRU in a cache-oblivious way. |
| 141 | return array[index(k)]; |
| 142 | } |
| 143 | |
| 144 | atomic<unsigned> abort_counter; |
| 145 | |
| 146 | //! Value for effective_capacity that denotes unbounded queue. |
| 147 | static const ptrdiff_t infinite_capacity = ptrdiff_t(~size_t(0)/2); |
| 148 | }; |
| 149 | |
| 150 | #if _MSC_VER && !defined(__INTEL_COMPILER) |
| 151 | // unary minus operator applied to unsigned type, result still unsigned |
| 152 | #pragma warning( push ) |
| 153 | #pragma warning( disable: 4146 ) |
| 154 | #endif |
| 155 | |
| 156 | static void* static_invalid_page; |
| 157 | |
| 158 | //------------------------------------------------------------------------ |
| 159 | // micro_queue |
| 160 | //------------------------------------------------------------------------ |
| 161 | void micro_queue::push( const void* item, ticket k, concurrent_queue_base& base, |
| 162 | concurrent_queue_base::copy_specifics op_type ) { |
| 163 | k &= -concurrent_queue_rep::n_queue; |
| 164 | page* p = NULL; |
| 165 | // find index on page where we would put the data |
| 166 | size_t index = modulo_power_of_two( k/concurrent_queue_rep::n_queue, base.items_per_page ); |
| 167 | if( !index ) { // make a new page |
| 168 | __TBB_TRY { |
| 169 | p = base.allocate_page(); |
| 170 | } __TBB_CATCH(...) { |
| 171 | ++base.my_rep->n_invalid_entries; |
| 172 | make_invalid( k ); |
| 173 | __TBB_RETHROW(); |
| 174 | } |
| 175 | p->mask = 0; |
| 176 | p->next = NULL; |
| 177 | } |
| 178 | |
| 179 | // wait for my turn |
| 180 | if( tail_counter!=k ) // The developer insisted on keeping first check out of the backoff loop |
| 181 | for( atomic_backoff b(true);;b.pause() ) { |
| 182 | ticket tail = tail_counter; |
| 183 | if( tail==k ) break; |
| 184 | else if( tail&0x1 ) { |
| 185 | // no memory. throws an exception; assumes concurrent_queue_rep::n_queue>1 |
| 186 | ++base.my_rep->n_invalid_entries; |
| 187 | throw_exception( eid_bad_last_alloc ); |
| 188 | } |
| 189 | } |
| 190 | |
| 191 | if( p ) { // page is newly allocated; insert in micro_queue |
| 192 | spin_mutex::scoped_lock lock( page_mutex ); |
| 193 | if( page* q = tail_page ) |
| 194 | q->next = p; |
| 195 | else |
| 196 | head_page = p; |
| 197 | tail_page = p; |
| 198 | } |
| 199 | |
| 200 | if (item) { |
| 201 | p = tail_page; |
| 202 | ITT_NOTIFY( sync_acquired, p ); |
| 203 | __TBB_TRY { |
| 204 | if( concurrent_queue_base::copy == op_type ) { |
| 205 | base.copy_item( *p, index, item ); |
| 206 | } else { |
| 207 | __TBB_ASSERT( concurrent_queue_base::move == op_type, NULL ); |
| 208 | static_cast<concurrent_queue_base_v8&>(base).move_item( *p, index, item ); |
| 209 | } |
| 210 | } __TBB_CATCH(...) { |
| 211 | ++base.my_rep->n_invalid_entries; |
| 212 | tail_counter += concurrent_queue_rep::n_queue; |
| 213 | __TBB_RETHROW(); |
| 214 | } |
| 215 | ITT_NOTIFY( sync_releasing, p ); |
| 216 | // If no exception was thrown, mark item as present. |
| 217 | p->mask |= uintptr_t(1)<<index; |
| 218 | } |
| 219 | else // no item; this was called from abort_push |
| 220 | ++base.my_rep->n_invalid_entries; |
| 221 | |
| 222 | tail_counter += concurrent_queue_rep::n_queue; |
| 223 | } |
| 224 | |
| 225 | |
| 226 | void micro_queue::abort_push( ticket k, concurrent_queue_base& base ) { |
| 227 | push(NULL, k, base, concurrent_queue_base::copy); |
| 228 | } |
| 229 | |
| 230 | bool micro_queue::pop( void* dst, ticket k, concurrent_queue_base& base ) { |
| 231 | k &= -concurrent_queue_rep::n_queue; |
| 232 | spin_wait_until_eq( head_counter, k ); |
| 233 | spin_wait_while_eq( tail_counter, k ); |
| 234 | page *p = head_page; |
| 235 | __TBB_ASSERT( p, NULL ); |
| 236 | size_t index = modulo_power_of_two( k/concurrent_queue_rep::n_queue, base.items_per_page ); |
| 237 | bool success = false; |
| 238 | { |
| 239 | micro_queue_pop_finalizer finalizer( *this, base, k+concurrent_queue_rep::n_queue, index==base.items_per_page-1 ? p : NULL ); |
| 240 | if( p->mask & uintptr_t(1)<<index ) { |
| 241 | success = true; |
| 242 | ITT_NOTIFY( sync_acquired, dst ); |
| 243 | ITT_NOTIFY( sync_acquired, head_page ); |
| 244 | base.assign_and_destroy_item( dst, *p, index ); |
| 245 | ITT_NOTIFY( sync_releasing, head_page ); |
| 246 | } else { |
| 247 | --base.my_rep->n_invalid_entries; |
| 248 | } |
| 249 | } |
| 250 | return success; |
| 251 | } |
| 252 | |
| 253 | micro_queue& micro_queue::assign( const micro_queue& src, concurrent_queue_base& base, |
| 254 | concurrent_queue_base::copy_specifics op_type ) |
| 255 | { |
| 256 | head_counter = src.head_counter; |
| 257 | tail_counter = src.tail_counter; |
| 258 | |
| 259 | const page* srcp = src.head_page; |
| 260 | if( srcp ) { |
| 261 | ticket g_index = head_counter; |
| 262 | __TBB_TRY { |
| 263 | size_t n_items = (tail_counter-head_counter)/concurrent_queue_rep::n_queue; |
| 264 | size_t index = modulo_power_of_two( head_counter/concurrent_queue_rep::n_queue, base.items_per_page ); |
| 265 | size_t end_in_first_page = (index+n_items<base.items_per_page)?(index+n_items):base.items_per_page; |
| 266 | |
| 267 | head_page = make_copy( base, srcp, index, end_in_first_page, g_index, op_type ); |
| 268 | page* cur_page = head_page; |
| 269 | |
| 270 | if( srcp != src.tail_page ) { |
| 271 | for( srcp = srcp->next; srcp!=src.tail_page; srcp=srcp->next ) { |
| 272 | cur_page->next = make_copy( base, srcp, 0, base.items_per_page, g_index, op_type ); |
| 273 | cur_page = cur_page->next; |
| 274 | } |
| 275 | |
| 276 | __TBB_ASSERT( srcp==src.tail_page, NULL ); |
| 277 | |
| 278 | size_t last_index = modulo_power_of_two( tail_counter/concurrent_queue_rep::n_queue, base.items_per_page ); |
| 279 | if( last_index==0 ) last_index = base.items_per_page; |
| 280 | |
| 281 | cur_page->next = make_copy( base, srcp, 0, last_index, g_index, op_type ); |
| 282 | cur_page = cur_page->next; |
| 283 | } |
| 284 | tail_page = cur_page; |
| 285 | } __TBB_CATCH(...) { |
| 286 | make_invalid( g_index ); |
| 287 | __TBB_RETHROW(); |
| 288 | } |
| 289 | } else { |
| 290 | head_page = tail_page = NULL; |
| 291 | } |
| 292 | return *this; |
| 293 | } |
| 294 | |
| 295 | concurrent_queue_base::page* micro_queue::make_copy( concurrent_queue_base& base, |
| 296 | const concurrent_queue_base::page* src_page, size_t begin_in_page, size_t end_in_page, |
| 297 | ticket& g_index, concurrent_queue_base::copy_specifics op_type ) |
| 298 | { |
| 299 | page* new_page = base.allocate_page(); |
| 300 | new_page->next = NULL; |
| 301 | new_page->mask = src_page->mask; |
| 302 | for( ; begin_in_page!=end_in_page; ++begin_in_page, ++g_index ) |
| 303 | if( new_page->mask & uintptr_t(1)<<begin_in_page ) { |
| 304 | if( concurrent_queue_base::copy == op_type ) { |
| 305 | base.copy_page_item( *new_page, begin_in_page, *src_page, begin_in_page ); |
| 306 | } else { |
| 307 | __TBB_ASSERT( concurrent_queue_base::move == op_type, NULL ); |
| 308 | static_cast<concurrent_queue_base_v8&>(base).move_page_item( *new_page, begin_in_page, *src_page, begin_in_page ); |
| 309 | } |
| 310 | } |
| 311 | return new_page; |
| 312 | } |
| 313 | |
| 314 | void micro_queue::make_invalid( ticket k ) |
| 315 | { |
| 316 | static concurrent_queue_base::page dummy = {static_cast<page*>((void*)1), 0}; |
| 317 | // mark it so that no more pushes are allowed. |
| 318 | static_invalid_page = &dummy; |
| 319 | { |
| 320 | spin_mutex::scoped_lock lock( page_mutex ); |
| 321 | tail_counter = k+concurrent_queue_rep::n_queue+1; |
| 322 | if( page* q = tail_page ) |
| 323 | q->next = static_cast<page*>(static_invalid_page); |
| 324 | else |
| 325 | head_page = static_cast<page*>(static_invalid_page); |
| 326 | tail_page = static_cast<page*>(static_invalid_page); |
| 327 | } |
| 328 | } |
| 329 | |
| 330 | #if _MSC_VER && !defined(__INTEL_COMPILER) |
| 331 | #pragma warning( pop ) |
| 332 | #endif // warning 4146 is back |
| 333 | |
| 334 | //------------------------------------------------------------------------ |
| 335 | // concurrent_queue_base |
| 336 | //------------------------------------------------------------------------ |
| 337 | concurrent_queue_base_v3::concurrent_queue_base_v3( size_t item_sz ) { |
| 338 | items_per_page = item_sz<= 8 ? 32 : |
| 339 | item_sz<= 16 ? 16 : |
| 340 | item_sz<= 32 ? 8 : |
| 341 | item_sz<= 64 ? 4 : |
| 342 | item_sz<=128 ? 2 : |
| 343 | 1; |
| 344 | my_capacity = size_t(-1)/(item_sz>1 ? item_sz : 2); |
| 345 | my_rep = cache_aligned_allocator<concurrent_queue_rep>().allocate(1); |
| 346 | __TBB_ASSERT( is_aligned(my_rep, NFS_GetLineSize()), "alignment error" ); |
| 347 | __TBB_ASSERT( is_aligned(&my_rep->head_counter, NFS_GetLineSize()), "alignment error" ); |
| 348 | __TBB_ASSERT( is_aligned(&my_rep->tail_counter, NFS_GetLineSize()), "alignment error" ); |
| 349 | __TBB_ASSERT( is_aligned(&my_rep->array, NFS_GetLineSize()), "alignment error" ); |
| 350 | std::memset(static_cast<void*>(my_rep),0,sizeof(concurrent_queue_rep)); |
| 351 | new ( &my_rep->items_avail ) concurrent_monitor(); |
| 352 | new ( &my_rep->slots_avail ) concurrent_monitor(); |
| 353 | this->item_size = item_sz; |
| 354 | } |
| 355 | |
| 356 | concurrent_queue_base_v3::~concurrent_queue_base_v3() { |
| 357 | size_t nq = my_rep->n_queue; |
| 358 | for( size_t i=0; i<nq; i++ ) |
| 359 | __TBB_ASSERT( my_rep->array[i].tail_page==NULL, "pages were not freed properly" ); |
| 360 | cache_aligned_allocator<concurrent_queue_rep>().deallocate(my_rep,1); |
| 361 | } |
| 362 | |
| 363 | void concurrent_queue_base_v3::internal_push( const void* src ) { |
| 364 | internal_insert_item( src, copy ); |
| 365 | } |
| 366 | |
| 367 | void concurrent_queue_base_v8::internal_push_move( const void* src ) { |
| 368 | internal_insert_item( src, move ); |
| 369 | } |
| 370 | |
| 371 | void concurrent_queue_base_v3::internal_insert_item( const void* src, copy_specifics op_type ) { |
| 372 | concurrent_queue_rep& r = *my_rep; |
| 373 | unsigned old_abort_counter = r.abort_counter; |
| 374 | ticket k = r.tail_counter++; |
| 375 | ptrdiff_t e = my_capacity; |
| 376 | #if DO_ITT_NOTIFY |
| 377 | bool sync_prepare_done = false; |
| 378 | #endif |
| 379 | if( (ptrdiff_t)(k-r.head_counter)>=e ) { // queue is full |
| 380 | #if DO_ITT_NOTIFY |
| 381 | if( !sync_prepare_done ) { |
| 382 | ITT_NOTIFY( sync_prepare, &sync_prepare_done ); |
| 383 | sync_prepare_done = true; |
| 384 | } |
| 385 | #endif |
| 386 | bool slept = false; |
| 387 | concurrent_monitor::thread_context thr_ctx; |
| 388 | r.slots_avail.prepare_wait( thr_ctx, ((ptrdiff_t)(k-e)) ); |
| 389 | while( (ptrdiff_t)(k-r.head_counter)>=const_cast<volatile ptrdiff_t&>(e = my_capacity) ) { |
| 390 | __TBB_TRY { |
| 391 | if( r.abort_counter!=old_abort_counter ) { |
| 392 | r.slots_avail.cancel_wait( thr_ctx ); |
| 393 | throw_exception( eid_user_abort ); |
| 394 | } |
| 395 | slept = r.slots_avail.commit_wait( thr_ctx ); |
| 396 | } __TBB_CATCH( tbb::user_abort& ) { |
| 397 | r.choose(k).abort_push(k, *this); |
| 398 | __TBB_RETHROW(); |
| 399 | } __TBB_CATCH(...) { |
| 400 | __TBB_RETHROW(); |
| 401 | } |
| 402 | if (slept == true) break; |
| 403 | r.slots_avail.prepare_wait( thr_ctx, ((ptrdiff_t)(k-e)) ); |
| 404 | } |
| 405 | if( !slept ) |
| 406 | r.slots_avail.cancel_wait( thr_ctx ); |
| 407 | } |
| 408 | ITT_NOTIFY( sync_acquired, &sync_prepare_done ); |
| 409 | __TBB_ASSERT( (ptrdiff_t)(k-r.head_counter)<my_capacity, NULL); |
| 410 | r.choose( k ).push( src, k, *this, op_type ); |
| 411 | r.items_avail.notify( predicate_leq(k) ); |
| 412 | } |
| 413 | |
| 414 | void concurrent_queue_base_v3::internal_pop( void* dst ) { |
| 415 | concurrent_queue_rep& r = *my_rep; |
| 416 | ticket k; |
| 417 | #if DO_ITT_NOTIFY |
| 418 | bool sync_prepare_done = false; |
| 419 | #endif |
| 420 | unsigned old_abort_counter = r.abort_counter; |
| 421 | // This loop is a single pop operation; abort_counter should not be re-read inside |
| 422 | do { |
| 423 | k=r.head_counter++; |
| 424 | if ( (ptrdiff_t)(r.tail_counter-k)<=0 ) { // queue is empty |
| 425 | #if DO_ITT_NOTIFY |
| 426 | if( !sync_prepare_done ) { |
| 427 | ITT_NOTIFY( sync_prepare, dst ); |
| 428 | sync_prepare_done = true; |
| 429 | } |
| 430 | #endif |
| 431 | bool slept = false; |
| 432 | concurrent_monitor::thread_context thr_ctx; |
| 433 | r.items_avail.prepare_wait( thr_ctx, k ); |
| 434 | while( (ptrdiff_t)(r.tail_counter-k)<=0 ) { |
| 435 | __TBB_TRY { |
| 436 | if( r.abort_counter!=old_abort_counter ) { |
| 437 | r.items_avail.cancel_wait( thr_ctx ); |
| 438 | throw_exception( eid_user_abort ); |
| 439 | } |
| 440 | slept = r.items_avail.commit_wait( thr_ctx ); |
| 441 | } __TBB_CATCH( tbb::user_abort& ) { |
| 442 | r.head_counter--; |
| 443 | __TBB_RETHROW(); |
| 444 | } __TBB_CATCH(...) { |
| 445 | __TBB_RETHROW(); |
| 446 | } |
| 447 | if (slept == true) break; |
| 448 | r.items_avail.prepare_wait( thr_ctx, k ); |
| 449 | } |
| 450 | if( !slept ) |
| 451 | r.items_avail.cancel_wait( thr_ctx ); |
| 452 | } |
| 453 | __TBB_ASSERT((ptrdiff_t)(r.tail_counter-k)>0, NULL); |
| 454 | } while( !r.choose(k).pop(dst,k,*this) ); |
| 455 | |
| 456 | // wake up a producer.. |
| 457 | r.slots_avail.notify( predicate_leq(k) ); |
| 458 | } |
| 459 | |
| 460 | void concurrent_queue_base_v3::internal_abort() { |
| 461 | concurrent_queue_rep& r = *my_rep; |
| 462 | ++r.abort_counter; |
| 463 | r.items_avail.abort_all(); |
| 464 | r.slots_avail.abort_all(); |
| 465 | } |
| 466 | |
| 467 | bool concurrent_queue_base_v3::internal_pop_if_present( void* dst ) { |
| 468 | concurrent_queue_rep& r = *my_rep; |
| 469 | ticket k; |
| 470 | do { |
| 471 | k = r.head_counter; |
| 472 | for(;;) { |
| 473 | if( (ptrdiff_t)(r.tail_counter-k)<=0 ) { |
| 474 | // Queue is empty |
| 475 | return false; |
| 476 | } |
| 477 | // Queue had item with ticket k when we looked. Attempt to get that item. |
| 478 | ticket tk=k; |
| 479 | k = r.head_counter.compare_and_swap( tk+1, tk ); |
| 480 | if( k==tk ) |
| 481 | break; |
| 482 | // Another thread snatched the item, retry. |
| 483 | } |
| 484 | } while( !r.choose( k ).pop( dst, k, *this ) ); |
| 485 | |
| 486 | r.slots_avail.notify( predicate_leq(k) ); |
| 487 | |
| 488 | return true; |
| 489 | } |
| 490 | |
| 491 | bool concurrent_queue_base_v3::internal_push_if_not_full( const void* src ) { |
| 492 | return internal_insert_if_not_full( src, copy ); |
| 493 | } |
| 494 | |
| 495 | bool concurrent_queue_base_v8::internal_push_move_if_not_full( const void* src ) { |
| 496 | return internal_insert_if_not_full( src, move ); |
| 497 | } |
| 498 | |
| 499 | bool concurrent_queue_base_v3::internal_insert_if_not_full( const void* src, copy_specifics op_type ) { |
| 500 | concurrent_queue_rep& r = *my_rep; |
| 501 | ticket k = r.tail_counter; |
| 502 | for(;;) { |
| 503 | if( (ptrdiff_t)(k-r.head_counter)>=my_capacity ) { |
| 504 | // Queue is full |
| 505 | return false; |
| 506 | } |
| 507 | // Queue had empty slot with ticket k when we looked. Attempt to claim that slot. |
| 508 | ticket tk=k; |
| 509 | k = r.tail_counter.compare_and_swap( tk+1, tk ); |
| 510 | if( k==tk ) |
| 511 | break; |
| 512 | // Another thread claimed the slot, so retry. |
| 513 | } |
| 514 | r.choose(k).push(src, k, *this, op_type); |
| 515 | r.items_avail.notify( predicate_leq(k) ); |
| 516 | return true; |
| 517 | } |
| 518 | |
| 519 | ptrdiff_t concurrent_queue_base_v3::internal_size() const { |
| 520 | __TBB_ASSERT( sizeof(ptrdiff_t)<=sizeof(size_t), NULL ); |
| 521 | return ptrdiff_t(my_rep->tail_counter-my_rep->head_counter-my_rep->n_invalid_entries); |
| 522 | } |
| 523 | |
| 524 | bool concurrent_queue_base_v3::internal_empty() const { |
| 525 | ticket tc = my_rep->tail_counter; |
| 526 | ticket hc = my_rep->head_counter; |
| 527 | // if tc!=r.tail_counter, the queue was not empty at some point between the two reads. |
| 528 | return ( tc==my_rep->tail_counter && ptrdiff_t(tc-hc-my_rep->n_invalid_entries)<=0 ); |
| 529 | } |
| 530 | |
| 531 | void concurrent_queue_base_v3::internal_set_capacity( ptrdiff_t capacity, size_t /*item_sz*/ ) { |
| 532 | my_capacity = capacity<0 ? concurrent_queue_rep::infinite_capacity : capacity; |
| 533 | } |
| 534 | |
| 535 | void concurrent_queue_base_v3::internal_finish_clear() { |
| 536 | size_t nq = my_rep->n_queue; |
| 537 | for( size_t i=0; i<nq; ++i ) { |
| 538 | page* tp = my_rep->array[i].tail_page; |
| 539 | __TBB_ASSERT( my_rep->array[i].head_page==tp, "at most one page should remain" ); |
| 540 | if( tp!=NULL) { |
| 541 | if( tp!=static_invalid_page ) deallocate_page( tp ); |
| 542 | my_rep->array[i].tail_page = NULL; |
| 543 | } |
| 544 | } |
| 545 | } |
| 546 | |
| 547 | void concurrent_queue_base_v3::internal_throw_exception() const { |
| 548 | throw_exception( eid_bad_alloc ); |
| 549 | } |
| 550 | |
| 551 | void concurrent_queue_base_v3::internal_assign( const concurrent_queue_base& src, copy_specifics op_type ) { |
| 552 | items_per_page = src.items_per_page; |
| 553 | my_capacity = src.my_capacity; |
| 554 | |
| 555 | // copy concurrent_queue_rep. |
| 556 | my_rep->head_counter = src.my_rep->head_counter; |
| 557 | my_rep->tail_counter = src.my_rep->tail_counter; |
| 558 | my_rep->n_invalid_entries = src.my_rep->n_invalid_entries; |
| 559 | my_rep->abort_counter = src.my_rep->abort_counter; |
| 560 | |
| 561 | // copy micro_queues |
| 562 | for( size_t i = 0; i<my_rep->n_queue; ++i ) |
| 563 | my_rep->array[i].assign( src.my_rep->array[i], *this, op_type ); |
| 564 | |
| 565 | __TBB_ASSERT( my_rep->head_counter==src.my_rep->head_counter && my_rep->tail_counter==src.my_rep->tail_counter, |
| 566 | "the source concurrent queue should not be concurrently modified." ); |
| 567 | } |
| 568 | |
| 569 | void concurrent_queue_base_v3::assign( const concurrent_queue_base& src ) { |
| 570 | internal_assign( src, copy ); |
| 571 | } |
| 572 | |
| 573 | void concurrent_queue_base_v8::move_content( concurrent_queue_base_v8& src ) { |
| 574 | internal_assign( src, move ); |
| 575 | } |
| 576 | |
| 577 | //------------------------------------------------------------------------ |
| 578 | // concurrent_queue_iterator_rep |
| 579 | //------------------------------------------------------------------------ |
| 580 | class concurrent_queue_iterator_rep: no_assign { |
| 581 | public: |
| 582 | ticket head_counter; |
| 583 | const concurrent_queue_base& my_queue; |
| 584 | const size_t offset_of_last; |
| 585 | concurrent_queue_base::page* array[concurrent_queue_rep::n_queue]; |
| 586 | concurrent_queue_iterator_rep( const concurrent_queue_base& queue, size_t offset_of_last_ ) : |
| 587 | head_counter(queue.my_rep->head_counter), |
| 588 | my_queue(queue), |
| 589 | offset_of_last(offset_of_last_) |
| 590 | { |
| 591 | const concurrent_queue_rep& rep = *queue.my_rep; |
| 592 | for( size_t k=0; k<concurrent_queue_rep::n_queue; ++k ) |
| 593 | array[k] = rep.array[k].head_page; |
| 594 | } |
| 595 | //! Set item to point to kth element. Return true if at end of queue or item is marked valid; false otherwise. |
| 596 | bool get_item( void*& item, size_t k ) { |
| 597 | if( k==my_queue.my_rep->tail_counter ) { |
| 598 | item = NULL; |
| 599 | return true; |
| 600 | } else { |
| 601 | concurrent_queue_base::page* p = array[concurrent_queue_rep::index(k)]; |
| 602 | __TBB_ASSERT(p,NULL); |
| 603 | size_t i = modulo_power_of_two( k/concurrent_queue_rep::n_queue, my_queue.items_per_page ); |
| 604 | item = static_cast<unsigned char*>(static_cast<void*>(p)) + offset_of_last + my_queue.item_size*i; |
| 605 | return (p->mask & uintptr_t(1)<<i)!=0; |
| 606 | } |
| 607 | } |
| 608 | }; |
| 609 | |
| 610 | //------------------------------------------------------------------------ |
| 611 | // concurrent_queue_iterator_base |
| 612 | //------------------------------------------------------------------------ |
| 613 | |
| 614 | void concurrent_queue_iterator_base_v3::initialize( const concurrent_queue_base& queue, size_t offset_of_last ) { |
| 615 | my_rep = cache_aligned_allocator<concurrent_queue_iterator_rep>().allocate(1); |
| 616 | new( my_rep ) concurrent_queue_iterator_rep(queue,offset_of_last); |
| 617 | size_t k = my_rep->head_counter; |
| 618 | if( !my_rep->get_item(my_item, k) ) advance(); |
| 619 | } |
| 620 | |
| 621 | concurrent_queue_iterator_base_v3::concurrent_queue_iterator_base_v3( const concurrent_queue_base& queue ) { |
| 622 | initialize(queue,0); |
| 623 | } |
| 624 | |
| 625 | concurrent_queue_iterator_base_v3::concurrent_queue_iterator_base_v3( const concurrent_queue_base& queue, size_t offset_of_last ) { |
| 626 | initialize(queue,offset_of_last); |
| 627 | } |
| 628 | |
| 629 | void concurrent_queue_iterator_base_v3::assign( const concurrent_queue_iterator_base& other ) { |
| 630 | if( my_rep!=other.my_rep ) { |
| 631 | if( my_rep ) { |
| 632 | cache_aligned_allocator<concurrent_queue_iterator_rep>().deallocate(my_rep, 1); |
| 633 | my_rep = NULL; |
| 634 | } |
| 635 | if( other.my_rep ) { |
| 636 | my_rep = cache_aligned_allocator<concurrent_queue_iterator_rep>().allocate(1); |
| 637 | new( my_rep ) concurrent_queue_iterator_rep( *other.my_rep ); |
| 638 | } |
| 639 | } |
| 640 | my_item = other.my_item; |
| 641 | } |
| 642 | |
| 643 | void concurrent_queue_iterator_base_v3::advance() { |
| 644 | __TBB_ASSERT( my_item, "attempt to increment iterator past end of queue" ); |
| 645 | size_t k = my_rep->head_counter; |
| 646 | const concurrent_queue_base& queue = my_rep->my_queue; |
| 647 | #if TBB_USE_ASSERT |
| 648 | void* tmp; |
| 649 | my_rep->get_item(tmp,k); |
| 650 | __TBB_ASSERT( my_item==tmp, NULL ); |
| 651 | #endif /* TBB_USE_ASSERT */ |
| 652 | size_t i = modulo_power_of_two( k/concurrent_queue_rep::n_queue, queue.items_per_page ); |
| 653 | if( i==queue.items_per_page-1 ) { |
| 654 | concurrent_queue_base::page*& root = my_rep->array[concurrent_queue_rep::index(k)]; |
| 655 | root = root->next; |
| 656 | } |
| 657 | // advance k |
| 658 | my_rep->head_counter = ++k; |
| 659 | if( !my_rep->get_item(my_item, k) ) advance(); |
| 660 | } |
| 661 | |
| 662 | concurrent_queue_iterator_base_v3::~concurrent_queue_iterator_base_v3() { |
| 663 | //delete my_rep; |
| 664 | cache_aligned_allocator<concurrent_queue_iterator_rep>().deallocate(my_rep, 1); |
| 665 | my_rep = NULL; |
| 666 | } |
| 667 | |
| 668 | } // namespace internal |
| 669 | |
| 670 | } // namespace tbb |
| 671 | |