kmp_lock.cpp revision 360784
1/* 2 * kmp_lock.cpp -- lock-related functions 3 */ 4 5//===----------------------------------------------------------------------===// 6// 7// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 8// See https://llvm.org/LICENSE.txt for license information. 9// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 10// 11//===----------------------------------------------------------------------===// 12 13#include <stddef.h> 14#include <atomic> 15 16#include "kmp.h" 17#include "kmp_i18n.h" 18#include "kmp_io.h" 19#include "kmp_itt.h" 20#include "kmp_lock.h" 21#include "kmp_wait_release.h" 22#include "kmp_wrapper_getpid.h" 23 24#include "tsan_annotations.h" 25 26#if KMP_USE_FUTEX 27#include <sys/syscall.h> 28#include <unistd.h> 29// We should really include <futex.h>, but that causes compatibility problems on 30// different Linux* OS distributions that either require that you include (or 31// break when you try to include) <pci/types.h>. Since all we need is the two 32// macros below (which are part of the kernel ABI, so can't change) we just 33// define the constants here and don't include <futex.h> 34#ifndef FUTEX_WAIT 35#define FUTEX_WAIT 0 36#endif 37#ifndef FUTEX_WAKE 38#define FUTEX_WAKE 1 39#endif 40#endif 41 42/* Implement spin locks for internal library use. */ 43/* The algorithm implemented is Lamport's bakery lock [1974]. */ 44 45void __kmp_validate_locks(void) { 46 int i; 47 kmp_uint32 x, y; 48 49 /* Check to make sure unsigned arithmetic does wraps properly */ 50 x = ~((kmp_uint32)0) - 2; 51 y = x - 2; 52 53 for (i = 0; i < 8; ++i, ++x, ++y) { 54 kmp_uint32 z = (x - y); 55 KMP_ASSERT(z == 2); 56 } 57 58 KMP_ASSERT(offsetof(kmp_base_queuing_lock, tail_id) % 8 == 0); 59} 60 61/* ------------------------------------------------------------------------ */ 62/* test and set locks */ 63 64// For the non-nested locks, we can only assume that the first 4 bytes were 65// allocated, since gcc only allocates 4 bytes for omp_lock_t, and the Intel 66// compiler only allocates a 4 byte pointer on IA-32 architecture. On 67// Windows* OS on Intel(R) 64, we can assume that all 8 bytes were allocated. 68// 69// gcc reserves >= 8 bytes for nested locks, so we can assume that the 70// entire 8 bytes were allocated for nested locks on all 64-bit platforms. 71 72static kmp_int32 __kmp_get_tas_lock_owner(kmp_tas_lock_t *lck) { 73 return KMP_LOCK_STRIP(KMP_ATOMIC_LD_RLX(&lck->lk.poll)) - 1; 74} 75 76static inline bool __kmp_is_tas_lock_nestable(kmp_tas_lock_t *lck) { 77 return lck->lk.depth_locked != -1; 78} 79 80__forceinline static int 81__kmp_acquire_tas_lock_timed_template(kmp_tas_lock_t *lck, kmp_int32 gtid) { 82 KMP_MB(); 83 84#ifdef USE_LOCK_PROFILE 85 kmp_uint32 curr = KMP_LOCK_STRIP(lck->lk.poll); 86 if ((curr != 0) && (curr != gtid + 1)) 87 __kmp_printf("LOCK CONTENTION: %p\n", lck); 88/* else __kmp_printf( "." );*/ 89#endif /* USE_LOCK_PROFILE */ 90 91 kmp_int32 tas_free = KMP_LOCK_FREE(tas); 92 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); 93 94 if (KMP_ATOMIC_LD_RLX(&lck->lk.poll) == tas_free && 95 __kmp_atomic_compare_store_acq(&lck->lk.poll, tas_free, tas_busy)) { 96 KMP_FSYNC_ACQUIRED(lck); 97 return KMP_LOCK_ACQUIRED_FIRST; 98 } 99 100 kmp_uint32 spins; 101 KMP_FSYNC_PREPARE(lck); 102 KMP_INIT_YIELD(spins); 103 kmp_backoff_t backoff = __kmp_spin_backoff_params; 104 do { 105 __kmp_spin_backoff(&backoff); 106 KMP_YIELD_OVERSUB_ELSE_SPIN(spins); 107 } while (KMP_ATOMIC_LD_RLX(&lck->lk.poll) != tas_free || 108 !__kmp_atomic_compare_store_acq(&lck->lk.poll, tas_free, tas_busy)); 109 KMP_FSYNC_ACQUIRED(lck); 110 return KMP_LOCK_ACQUIRED_FIRST; 111} 112 113int __kmp_acquire_tas_lock(kmp_tas_lock_t *lck, kmp_int32 gtid) { 114 int retval = __kmp_acquire_tas_lock_timed_template(lck, gtid); 115 ANNOTATE_TAS_ACQUIRED(lck); 116 return retval; 117} 118 119static int __kmp_acquire_tas_lock_with_checks(kmp_tas_lock_t *lck, 120 kmp_int32 gtid) { 121 char const *const func = "omp_set_lock"; 122 if ((sizeof(kmp_tas_lock_t) <= OMP_LOCK_T_SIZE) && 123 __kmp_is_tas_lock_nestable(lck)) { 124 KMP_FATAL(LockNestableUsedAsSimple, func); 125 } 126 if ((gtid >= 0) && (__kmp_get_tas_lock_owner(lck) == gtid)) { 127 KMP_FATAL(LockIsAlreadyOwned, func); 128 } 129 return __kmp_acquire_tas_lock(lck, gtid); 130} 131 132int __kmp_test_tas_lock(kmp_tas_lock_t *lck, kmp_int32 gtid) { 133 kmp_int32 tas_free = KMP_LOCK_FREE(tas); 134 kmp_int32 tas_busy = KMP_LOCK_BUSY(gtid + 1, tas); 135 if (KMP_ATOMIC_LD_RLX(&lck->lk.poll) == tas_free && 136 __kmp_atomic_compare_store_acq(&lck->lk.poll, tas_free, tas_busy)) { 137 KMP_FSYNC_ACQUIRED(lck); 138 return TRUE; 139 } 140 return FALSE; 141} 142 143static int __kmp_test_tas_lock_with_checks(kmp_tas_lock_t *lck, 144 kmp_int32 gtid) { 145 char const *const func = "omp_test_lock"; 146 if ((sizeof(kmp_tas_lock_t) <= OMP_LOCK_T_SIZE) && 147 __kmp_is_tas_lock_nestable(lck)) { 148 KMP_FATAL(LockNestableUsedAsSimple, func); 149 } 150 return __kmp_test_tas_lock(lck, gtid); 151} 152 153int __kmp_release_tas_lock(kmp_tas_lock_t *lck, kmp_int32 gtid) { 154 KMP_MB(); /* Flush all pending memory write invalidates. */ 155 156 KMP_FSYNC_RELEASING(lck); 157 ANNOTATE_TAS_RELEASED(lck); 158 KMP_ATOMIC_ST_REL(&lck->lk.poll, KMP_LOCK_FREE(tas)); 159 KMP_MB(); /* Flush all pending memory write invalidates. */ 160 161 KMP_YIELD_OVERSUB(); 162 return KMP_LOCK_RELEASED; 163} 164 165static int __kmp_release_tas_lock_with_checks(kmp_tas_lock_t *lck, 166 kmp_int32 gtid) { 167 char const *const func = "omp_unset_lock"; 168 KMP_MB(); /* in case another processor initialized lock */ 169 if ((sizeof(kmp_tas_lock_t) <= OMP_LOCK_T_SIZE) && 170 __kmp_is_tas_lock_nestable(lck)) { 171 KMP_FATAL(LockNestableUsedAsSimple, func); 172 } 173 if (__kmp_get_tas_lock_owner(lck) == -1) { 174 KMP_FATAL(LockUnsettingFree, func); 175 } 176 if ((gtid >= 0) && (__kmp_get_tas_lock_owner(lck) >= 0) && 177 (__kmp_get_tas_lock_owner(lck) != gtid)) { 178 KMP_FATAL(LockUnsettingSetByAnother, func); 179 } 180 return __kmp_release_tas_lock(lck, gtid); 181} 182 183void __kmp_init_tas_lock(kmp_tas_lock_t *lck) { 184 lck->lk.poll = KMP_LOCK_FREE(tas); 185} 186 187void __kmp_destroy_tas_lock(kmp_tas_lock_t *lck) { lck->lk.poll = 0; } 188 189static void __kmp_destroy_tas_lock_with_checks(kmp_tas_lock_t *lck) { 190 char const *const func = "omp_destroy_lock"; 191 if ((sizeof(kmp_tas_lock_t) <= OMP_LOCK_T_SIZE) && 192 __kmp_is_tas_lock_nestable(lck)) { 193 KMP_FATAL(LockNestableUsedAsSimple, func); 194 } 195 if (__kmp_get_tas_lock_owner(lck) != -1) { 196 KMP_FATAL(LockStillOwned, func); 197 } 198 __kmp_destroy_tas_lock(lck); 199} 200 201// nested test and set locks 202 203int __kmp_acquire_nested_tas_lock(kmp_tas_lock_t *lck, kmp_int32 gtid) { 204 KMP_DEBUG_ASSERT(gtid >= 0); 205 206 if (__kmp_get_tas_lock_owner(lck) == gtid) { 207 lck->lk.depth_locked += 1; 208 return KMP_LOCK_ACQUIRED_NEXT; 209 } else { 210 __kmp_acquire_tas_lock_timed_template(lck, gtid); 211 ANNOTATE_TAS_ACQUIRED(lck); 212 lck->lk.depth_locked = 1; 213 return KMP_LOCK_ACQUIRED_FIRST; 214 } 215} 216 217static int __kmp_acquire_nested_tas_lock_with_checks(kmp_tas_lock_t *lck, 218 kmp_int32 gtid) { 219 char const *const func = "omp_set_nest_lock"; 220 if (!__kmp_is_tas_lock_nestable(lck)) { 221 KMP_FATAL(LockSimpleUsedAsNestable, func); 222 } 223 return __kmp_acquire_nested_tas_lock(lck, gtid); 224} 225 226int __kmp_test_nested_tas_lock(kmp_tas_lock_t *lck, kmp_int32 gtid) { 227 int retval; 228 229 KMP_DEBUG_ASSERT(gtid >= 0); 230 231 if (__kmp_get_tas_lock_owner(lck) == gtid) { 232 retval = ++lck->lk.depth_locked; 233 } else if (!__kmp_test_tas_lock(lck, gtid)) { 234 retval = 0; 235 } else { 236 KMP_MB(); 237 retval = lck->lk.depth_locked = 1; 238 } 239 return retval; 240} 241 242static int __kmp_test_nested_tas_lock_with_checks(kmp_tas_lock_t *lck, 243 kmp_int32 gtid) { 244 char const *const func = "omp_test_nest_lock"; 245 if (!__kmp_is_tas_lock_nestable(lck)) { 246 KMP_FATAL(LockSimpleUsedAsNestable, func); 247 } 248 return __kmp_test_nested_tas_lock(lck, gtid); 249} 250 251int __kmp_release_nested_tas_lock(kmp_tas_lock_t *lck, kmp_int32 gtid) { 252 KMP_DEBUG_ASSERT(gtid >= 0); 253 254 KMP_MB(); 255 if (--(lck->lk.depth_locked) == 0) { 256 __kmp_release_tas_lock(lck, gtid); 257 return KMP_LOCK_RELEASED; 258 } 259 return KMP_LOCK_STILL_HELD; 260} 261 262static int __kmp_release_nested_tas_lock_with_checks(kmp_tas_lock_t *lck, 263 kmp_int32 gtid) { 264 char const *const func = "omp_unset_nest_lock"; 265 KMP_MB(); /* in case another processor initialized lock */ 266 if (!__kmp_is_tas_lock_nestable(lck)) { 267 KMP_FATAL(LockSimpleUsedAsNestable, func); 268 } 269 if (__kmp_get_tas_lock_owner(lck) == -1) { 270 KMP_FATAL(LockUnsettingFree, func); 271 } 272 if (__kmp_get_tas_lock_owner(lck) != gtid) { 273 KMP_FATAL(LockUnsettingSetByAnother, func); 274 } 275 return __kmp_release_nested_tas_lock(lck, gtid); 276} 277 278void __kmp_init_nested_tas_lock(kmp_tas_lock_t *lck) { 279 __kmp_init_tas_lock(lck); 280 lck->lk.depth_locked = 0; // >= 0 for nestable locks, -1 for simple locks 281} 282 283void __kmp_destroy_nested_tas_lock(kmp_tas_lock_t *lck) { 284 __kmp_destroy_tas_lock(lck); 285 lck->lk.depth_locked = 0; 286} 287 288static void __kmp_destroy_nested_tas_lock_with_checks(kmp_tas_lock_t *lck) { 289 char const *const func = "omp_destroy_nest_lock"; 290 if (!__kmp_is_tas_lock_nestable(lck)) { 291 KMP_FATAL(LockSimpleUsedAsNestable, func); 292 } 293 if (__kmp_get_tas_lock_owner(lck) != -1) { 294 KMP_FATAL(LockStillOwned, func); 295 } 296 __kmp_destroy_nested_tas_lock(lck); 297} 298 299#if KMP_USE_FUTEX 300 301/* ------------------------------------------------------------------------ */ 302/* futex locks */ 303 304// futex locks are really just test and set locks, with a different method 305// of handling contention. They take the same amount of space as test and 306// set locks, and are allocated the same way (i.e. use the area allocated by 307// the compiler for non-nested locks / allocate nested locks on the heap). 308 309static kmp_int32 __kmp_get_futex_lock_owner(kmp_futex_lock_t *lck) { 310 return KMP_LOCK_STRIP((TCR_4(lck->lk.poll) >> 1)) - 1; 311} 312 313static inline bool __kmp_is_futex_lock_nestable(kmp_futex_lock_t *lck) { 314 return lck->lk.depth_locked != -1; 315} 316 317__forceinline static int 318__kmp_acquire_futex_lock_timed_template(kmp_futex_lock_t *lck, kmp_int32 gtid) { 319 kmp_int32 gtid_code = (gtid + 1) << 1; 320 321 KMP_MB(); 322 323#ifdef USE_LOCK_PROFILE 324 kmp_uint32 curr = KMP_LOCK_STRIP(TCR_4(lck->lk.poll)); 325 if ((curr != 0) && (curr != gtid_code)) 326 __kmp_printf("LOCK CONTENTION: %p\n", lck); 327/* else __kmp_printf( "." );*/ 328#endif /* USE_LOCK_PROFILE */ 329 330 KMP_FSYNC_PREPARE(lck); 331 KA_TRACE(1000, ("__kmp_acquire_futex_lock: lck:%p(0x%x), T#%d entering\n", 332 lck, lck->lk.poll, gtid)); 333 334 kmp_int32 poll_val; 335 336 while ((poll_val = KMP_COMPARE_AND_STORE_RET32( 337 &(lck->lk.poll), KMP_LOCK_FREE(futex), 338 KMP_LOCK_BUSY(gtid_code, futex))) != KMP_LOCK_FREE(futex)) { 339 340 kmp_int32 cond = KMP_LOCK_STRIP(poll_val) & 1; 341 KA_TRACE( 342 1000, 343 ("__kmp_acquire_futex_lock: lck:%p, T#%d poll_val = 0x%x cond = 0x%x\n", 344 lck, gtid, poll_val, cond)); 345 346 // NOTE: if you try to use the following condition for this branch 347 // 348 // if ( poll_val & 1 == 0 ) 349 // 350 // Then the 12.0 compiler has a bug where the following block will 351 // always be skipped, regardless of the value of the LSB of poll_val. 352 if (!cond) { 353 // Try to set the lsb in the poll to indicate to the owner 354 // thread that they need to wake this thread up. 355 if (!KMP_COMPARE_AND_STORE_REL32(&(lck->lk.poll), poll_val, 356 poll_val | KMP_LOCK_BUSY(1, futex))) { 357 KA_TRACE( 358 1000, 359 ("__kmp_acquire_futex_lock: lck:%p(0x%x), T#%d can't set bit 0\n", 360 lck, lck->lk.poll, gtid)); 361 continue; 362 } 363 poll_val |= KMP_LOCK_BUSY(1, futex); 364 365 KA_TRACE(1000, 366 ("__kmp_acquire_futex_lock: lck:%p(0x%x), T#%d bit 0 set\n", lck, 367 lck->lk.poll, gtid)); 368 } 369 370 KA_TRACE( 371 1000, 372 ("__kmp_acquire_futex_lock: lck:%p, T#%d before futex_wait(0x%x)\n", 373 lck, gtid, poll_val)); 374 375 kmp_int32 rc; 376 if ((rc = syscall(__NR_futex, &(lck->lk.poll), FUTEX_WAIT, poll_val, NULL, 377 NULL, 0)) != 0) { 378 KA_TRACE(1000, ("__kmp_acquire_futex_lock: lck:%p, T#%d futex_wait(0x%x) " 379 "failed (rc=%d errno=%d)\n", 380 lck, gtid, poll_val, rc, errno)); 381 continue; 382 } 383 384 KA_TRACE(1000, 385 ("__kmp_acquire_futex_lock: lck:%p, T#%d after futex_wait(0x%x)\n", 386 lck, gtid, poll_val)); 387 // This thread has now done a successful futex wait call and was entered on 388 // the OS futex queue. We must now perform a futex wake call when releasing 389 // the lock, as we have no idea how many other threads are in the queue. 390 gtid_code |= 1; 391 } 392 393 KMP_FSYNC_ACQUIRED(lck); 394 KA_TRACE(1000, ("__kmp_acquire_futex_lock: lck:%p(0x%x), T#%d exiting\n", lck, 395 lck->lk.poll, gtid)); 396 return KMP_LOCK_ACQUIRED_FIRST; 397} 398 399int __kmp_acquire_futex_lock(kmp_futex_lock_t *lck, kmp_int32 gtid) { 400 int retval = __kmp_acquire_futex_lock_timed_template(lck, gtid); 401 ANNOTATE_FUTEX_ACQUIRED(lck); 402 return retval; 403} 404 405static int __kmp_acquire_futex_lock_with_checks(kmp_futex_lock_t *lck, 406 kmp_int32 gtid) { 407 char const *const func = "omp_set_lock"; 408 if ((sizeof(kmp_futex_lock_t) <= OMP_LOCK_T_SIZE) && 409 __kmp_is_futex_lock_nestable(lck)) { 410 KMP_FATAL(LockNestableUsedAsSimple, func); 411 } 412 if ((gtid >= 0) && (__kmp_get_futex_lock_owner(lck) == gtid)) { 413 KMP_FATAL(LockIsAlreadyOwned, func); 414 } 415 return __kmp_acquire_futex_lock(lck, gtid); 416} 417 418int __kmp_test_futex_lock(kmp_futex_lock_t *lck, kmp_int32 gtid) { 419 if (KMP_COMPARE_AND_STORE_ACQ32(&(lck->lk.poll), KMP_LOCK_FREE(futex), 420 KMP_LOCK_BUSY((gtid + 1) << 1, futex))) { 421 KMP_FSYNC_ACQUIRED(lck); 422 return TRUE; 423 } 424 return FALSE; 425} 426 427static int __kmp_test_futex_lock_with_checks(kmp_futex_lock_t *lck, 428 kmp_int32 gtid) { 429 char const *const func = "omp_test_lock"; 430 if ((sizeof(kmp_futex_lock_t) <= OMP_LOCK_T_SIZE) && 431 __kmp_is_futex_lock_nestable(lck)) { 432 KMP_FATAL(LockNestableUsedAsSimple, func); 433 } 434 return __kmp_test_futex_lock(lck, gtid); 435} 436 437int __kmp_release_futex_lock(kmp_futex_lock_t *lck, kmp_int32 gtid) { 438 KMP_MB(); /* Flush all pending memory write invalidates. */ 439 440 KA_TRACE(1000, ("__kmp_release_futex_lock: lck:%p(0x%x), T#%d entering\n", 441 lck, lck->lk.poll, gtid)); 442 443 KMP_FSYNC_RELEASING(lck); 444 ANNOTATE_FUTEX_RELEASED(lck); 445 446 kmp_int32 poll_val = KMP_XCHG_FIXED32(&(lck->lk.poll), KMP_LOCK_FREE(futex)); 447 448 KA_TRACE(1000, 449 ("__kmp_release_futex_lock: lck:%p, T#%d released poll_val = 0x%x\n", 450 lck, gtid, poll_val)); 451 452 if (KMP_LOCK_STRIP(poll_val) & 1) { 453 KA_TRACE(1000, 454 ("__kmp_release_futex_lock: lck:%p, T#%d futex_wake 1 thread\n", 455 lck, gtid)); 456 syscall(__NR_futex, &(lck->lk.poll), FUTEX_WAKE, KMP_LOCK_BUSY(1, futex), 457 NULL, NULL, 0); 458 } 459 460 KMP_MB(); /* Flush all pending memory write invalidates. */ 461 462 KA_TRACE(1000, ("__kmp_release_futex_lock: lck:%p(0x%x), T#%d exiting\n", lck, 463 lck->lk.poll, gtid)); 464 465 KMP_YIELD_OVERSUB(); 466 return KMP_LOCK_RELEASED; 467} 468 469static int __kmp_release_futex_lock_with_checks(kmp_futex_lock_t *lck, 470 kmp_int32 gtid) { 471 char const *const func = "omp_unset_lock"; 472 KMP_MB(); /* in case another processor initialized lock */ 473 if ((sizeof(kmp_futex_lock_t) <= OMP_LOCK_T_SIZE) && 474 __kmp_is_futex_lock_nestable(lck)) { 475 KMP_FATAL(LockNestableUsedAsSimple, func); 476 } 477 if (__kmp_get_futex_lock_owner(lck) == -1) { 478 KMP_FATAL(LockUnsettingFree, func); 479 } 480 if ((gtid >= 0) && (__kmp_get_futex_lock_owner(lck) >= 0) && 481 (__kmp_get_futex_lock_owner(lck) != gtid)) { 482 KMP_FATAL(LockUnsettingSetByAnother, func); 483 } 484 return __kmp_release_futex_lock(lck, gtid); 485} 486 487void __kmp_init_futex_lock(kmp_futex_lock_t *lck) { 488 TCW_4(lck->lk.poll, KMP_LOCK_FREE(futex)); 489} 490 491void __kmp_destroy_futex_lock(kmp_futex_lock_t *lck) { lck->lk.poll = 0; } 492 493static void __kmp_destroy_futex_lock_with_checks(kmp_futex_lock_t *lck) { 494 char const *const func = "omp_destroy_lock"; 495 if ((sizeof(kmp_futex_lock_t) <= OMP_LOCK_T_SIZE) && 496 __kmp_is_futex_lock_nestable(lck)) { 497 KMP_FATAL(LockNestableUsedAsSimple, func); 498 } 499 if (__kmp_get_futex_lock_owner(lck) != -1) { 500 KMP_FATAL(LockStillOwned, func); 501 } 502 __kmp_destroy_futex_lock(lck); 503} 504 505// nested futex locks 506 507int __kmp_acquire_nested_futex_lock(kmp_futex_lock_t *lck, kmp_int32 gtid) { 508 KMP_DEBUG_ASSERT(gtid >= 0); 509 510 if (__kmp_get_futex_lock_owner(lck) == gtid) { 511 lck->lk.depth_locked += 1; 512 return KMP_LOCK_ACQUIRED_NEXT; 513 } else { 514 __kmp_acquire_futex_lock_timed_template(lck, gtid); 515 ANNOTATE_FUTEX_ACQUIRED(lck); 516 lck->lk.depth_locked = 1; 517 return KMP_LOCK_ACQUIRED_FIRST; 518 } 519} 520 521static int __kmp_acquire_nested_futex_lock_with_checks(kmp_futex_lock_t *lck, 522 kmp_int32 gtid) { 523 char const *const func = "omp_set_nest_lock"; 524 if (!__kmp_is_futex_lock_nestable(lck)) { 525 KMP_FATAL(LockSimpleUsedAsNestable, func); 526 } 527 return __kmp_acquire_nested_futex_lock(lck, gtid); 528} 529 530int __kmp_test_nested_futex_lock(kmp_futex_lock_t *lck, kmp_int32 gtid) { 531 int retval; 532 533 KMP_DEBUG_ASSERT(gtid >= 0); 534 535 if (__kmp_get_futex_lock_owner(lck) == gtid) { 536 retval = ++lck->lk.depth_locked; 537 } else if (!__kmp_test_futex_lock(lck, gtid)) { 538 retval = 0; 539 } else { 540 KMP_MB(); 541 retval = lck->lk.depth_locked = 1; 542 } 543 return retval; 544} 545 546static int __kmp_test_nested_futex_lock_with_checks(kmp_futex_lock_t *lck, 547 kmp_int32 gtid) { 548 char const *const func = "omp_test_nest_lock"; 549 if (!__kmp_is_futex_lock_nestable(lck)) { 550 KMP_FATAL(LockSimpleUsedAsNestable, func); 551 } 552 return __kmp_test_nested_futex_lock(lck, gtid); 553} 554 555int __kmp_release_nested_futex_lock(kmp_futex_lock_t *lck, kmp_int32 gtid) { 556 KMP_DEBUG_ASSERT(gtid >= 0); 557 558 KMP_MB(); 559 if (--(lck->lk.depth_locked) == 0) { 560 __kmp_release_futex_lock(lck, gtid); 561 return KMP_LOCK_RELEASED; 562 } 563 return KMP_LOCK_STILL_HELD; 564} 565 566static int __kmp_release_nested_futex_lock_with_checks(kmp_futex_lock_t *lck, 567 kmp_int32 gtid) { 568 char const *const func = "omp_unset_nest_lock"; 569 KMP_MB(); /* in case another processor initialized lock */ 570 if (!__kmp_is_futex_lock_nestable(lck)) { 571 KMP_FATAL(LockSimpleUsedAsNestable, func); 572 } 573 if (__kmp_get_futex_lock_owner(lck) == -1) { 574 KMP_FATAL(LockUnsettingFree, func); 575 } 576 if (__kmp_get_futex_lock_owner(lck) != gtid) { 577 KMP_FATAL(LockUnsettingSetByAnother, func); 578 } 579 return __kmp_release_nested_futex_lock(lck, gtid); 580} 581 582void __kmp_init_nested_futex_lock(kmp_futex_lock_t *lck) { 583 __kmp_init_futex_lock(lck); 584 lck->lk.depth_locked = 0; // >= 0 for nestable locks, -1 for simple locks 585} 586 587void __kmp_destroy_nested_futex_lock(kmp_futex_lock_t *lck) { 588 __kmp_destroy_futex_lock(lck); 589 lck->lk.depth_locked = 0; 590} 591 592static void __kmp_destroy_nested_futex_lock_with_checks(kmp_futex_lock_t *lck) { 593 char const *const func = "omp_destroy_nest_lock"; 594 if (!__kmp_is_futex_lock_nestable(lck)) { 595 KMP_FATAL(LockSimpleUsedAsNestable, func); 596 } 597 if (__kmp_get_futex_lock_owner(lck) != -1) { 598 KMP_FATAL(LockStillOwned, func); 599 } 600 __kmp_destroy_nested_futex_lock(lck); 601} 602 603#endif // KMP_USE_FUTEX 604 605/* ------------------------------------------------------------------------ */ 606/* ticket (bakery) locks */ 607 608static kmp_int32 __kmp_get_ticket_lock_owner(kmp_ticket_lock_t *lck) { 609 return std::atomic_load_explicit(&lck->lk.owner_id, 610 std::memory_order_relaxed) - 611 1; 612} 613 614static inline bool __kmp_is_ticket_lock_nestable(kmp_ticket_lock_t *lck) { 615 return std::atomic_load_explicit(&lck->lk.depth_locked, 616 std::memory_order_relaxed) != -1; 617} 618 619static kmp_uint32 __kmp_bakery_check(void *now_serving, kmp_uint32 my_ticket) { 620 return std::atomic_load_explicit((std::atomic<unsigned> *)now_serving, 621 std::memory_order_acquire) == my_ticket; 622} 623 624__forceinline static int 625__kmp_acquire_ticket_lock_timed_template(kmp_ticket_lock_t *lck, 626 kmp_int32 gtid) { 627 kmp_uint32 my_ticket = std::atomic_fetch_add_explicit( 628 &lck->lk.next_ticket, 1U, std::memory_order_relaxed); 629 630#ifdef USE_LOCK_PROFILE 631 if (std::atomic_load_explicit(&lck->lk.now_serving, 632 std::memory_order_relaxed) != my_ticket) 633 __kmp_printf("LOCK CONTENTION: %p\n", lck); 634/* else __kmp_printf( "." );*/ 635#endif /* USE_LOCK_PROFILE */ 636 637 if (std::atomic_load_explicit(&lck->lk.now_serving, 638 std::memory_order_acquire) == my_ticket) { 639 return KMP_LOCK_ACQUIRED_FIRST; 640 } 641 KMP_WAIT_PTR(&lck->lk.now_serving, my_ticket, __kmp_bakery_check, lck); 642 return KMP_LOCK_ACQUIRED_FIRST; 643} 644 645int __kmp_acquire_ticket_lock(kmp_ticket_lock_t *lck, kmp_int32 gtid) { 646 int retval = __kmp_acquire_ticket_lock_timed_template(lck, gtid); 647 ANNOTATE_TICKET_ACQUIRED(lck); 648 return retval; 649} 650 651static int __kmp_acquire_ticket_lock_with_checks(kmp_ticket_lock_t *lck, 652 kmp_int32 gtid) { 653 char const *const func = "omp_set_lock"; 654 655 if (!std::atomic_load_explicit(&lck->lk.initialized, 656 std::memory_order_relaxed)) { 657 KMP_FATAL(LockIsUninitialized, func); 658 } 659 if (lck->lk.self != lck) { 660 KMP_FATAL(LockIsUninitialized, func); 661 } 662 if (__kmp_is_ticket_lock_nestable(lck)) { 663 KMP_FATAL(LockNestableUsedAsSimple, func); 664 } 665 if ((gtid >= 0) && (__kmp_get_ticket_lock_owner(lck) == gtid)) { 666 KMP_FATAL(LockIsAlreadyOwned, func); 667 } 668 669 __kmp_acquire_ticket_lock(lck, gtid); 670 671 std::atomic_store_explicit(&lck->lk.owner_id, gtid + 1, 672 std::memory_order_relaxed); 673 return KMP_LOCK_ACQUIRED_FIRST; 674} 675 676int __kmp_test_ticket_lock(kmp_ticket_lock_t *lck, kmp_int32 gtid) { 677 kmp_uint32 my_ticket = std::atomic_load_explicit(&lck->lk.next_ticket, 678 std::memory_order_relaxed); 679 680 if (std::atomic_load_explicit(&lck->lk.now_serving, 681 std::memory_order_relaxed) == my_ticket) { 682 kmp_uint32 next_ticket = my_ticket + 1; 683 if (std::atomic_compare_exchange_strong_explicit( 684 &lck->lk.next_ticket, &my_ticket, next_ticket, 685 std::memory_order_acquire, std::memory_order_acquire)) { 686 return TRUE; 687 } 688 } 689 return FALSE; 690} 691 692static int __kmp_test_ticket_lock_with_checks(kmp_ticket_lock_t *lck, 693 kmp_int32 gtid) { 694 char const *const func = "omp_test_lock"; 695 696 if (!std::atomic_load_explicit(&lck->lk.initialized, 697 std::memory_order_relaxed)) { 698 KMP_FATAL(LockIsUninitialized, func); 699 } 700 if (lck->lk.self != lck) { 701 KMP_FATAL(LockIsUninitialized, func); 702 } 703 if (__kmp_is_ticket_lock_nestable(lck)) { 704 KMP_FATAL(LockNestableUsedAsSimple, func); 705 } 706 707 int retval = __kmp_test_ticket_lock(lck, gtid); 708 709 if (retval) { 710 std::atomic_store_explicit(&lck->lk.owner_id, gtid + 1, 711 std::memory_order_relaxed); 712 } 713 return retval; 714} 715 716int __kmp_release_ticket_lock(kmp_ticket_lock_t *lck, kmp_int32 gtid) { 717 kmp_uint32 distance = std::atomic_load_explicit(&lck->lk.next_ticket, 718 std::memory_order_relaxed) - 719 std::atomic_load_explicit(&lck->lk.now_serving, 720 std::memory_order_relaxed); 721 722 ANNOTATE_TICKET_RELEASED(lck); 723 std::atomic_fetch_add_explicit(&lck->lk.now_serving, 1U, 724 std::memory_order_release); 725 726 KMP_YIELD(distance > 727 (kmp_uint32)(__kmp_avail_proc ? __kmp_avail_proc : __kmp_xproc)); 728 return KMP_LOCK_RELEASED; 729} 730 731static int __kmp_release_ticket_lock_with_checks(kmp_ticket_lock_t *lck, 732 kmp_int32 gtid) { 733 char const *const func = "omp_unset_lock"; 734 735 if (!std::atomic_load_explicit(&lck->lk.initialized, 736 std::memory_order_relaxed)) { 737 KMP_FATAL(LockIsUninitialized, func); 738 } 739 if (lck->lk.self != lck) { 740 KMP_FATAL(LockIsUninitialized, func); 741 } 742 if (__kmp_is_ticket_lock_nestable(lck)) { 743 KMP_FATAL(LockNestableUsedAsSimple, func); 744 } 745 if (__kmp_get_ticket_lock_owner(lck) == -1) { 746 KMP_FATAL(LockUnsettingFree, func); 747 } 748 if ((gtid >= 0) && (__kmp_get_ticket_lock_owner(lck) >= 0) && 749 (__kmp_get_ticket_lock_owner(lck) != gtid)) { 750 KMP_FATAL(LockUnsettingSetByAnother, func); 751 } 752 std::atomic_store_explicit(&lck->lk.owner_id, 0, std::memory_order_relaxed); 753 return __kmp_release_ticket_lock(lck, gtid); 754} 755 756void __kmp_init_ticket_lock(kmp_ticket_lock_t *lck) { 757 lck->lk.location = NULL; 758 lck->lk.self = lck; 759 std::atomic_store_explicit(&lck->lk.next_ticket, 0U, 760 std::memory_order_relaxed); 761 std::atomic_store_explicit(&lck->lk.now_serving, 0U, 762 std::memory_order_relaxed); 763 std::atomic_store_explicit( 764 &lck->lk.owner_id, 0, 765 std::memory_order_relaxed); // no thread owns the lock. 766 std::atomic_store_explicit( 767 &lck->lk.depth_locked, -1, 768 std::memory_order_relaxed); // -1 => not a nested lock. 769 std::atomic_store_explicit(&lck->lk.initialized, true, 770 std::memory_order_release); 771} 772 773void __kmp_destroy_ticket_lock(kmp_ticket_lock_t *lck) { 774 std::atomic_store_explicit(&lck->lk.initialized, false, 775 std::memory_order_release); 776 lck->lk.self = NULL; 777 lck->lk.location = NULL; 778 std::atomic_store_explicit(&lck->lk.next_ticket, 0U, 779 std::memory_order_relaxed); 780 std::atomic_store_explicit(&lck->lk.now_serving, 0U, 781 std::memory_order_relaxed); 782 std::atomic_store_explicit(&lck->lk.owner_id, 0, std::memory_order_relaxed); 783 std::atomic_store_explicit(&lck->lk.depth_locked, -1, 784 std::memory_order_relaxed); 785} 786 787static void __kmp_destroy_ticket_lock_with_checks(kmp_ticket_lock_t *lck) { 788 char const *const func = "omp_destroy_lock"; 789 790 if (!std::atomic_load_explicit(&lck->lk.initialized, 791 std::memory_order_relaxed)) { 792 KMP_FATAL(LockIsUninitialized, func); 793 } 794 if (lck->lk.self != lck) { 795 KMP_FATAL(LockIsUninitialized, func); 796 } 797 if (__kmp_is_ticket_lock_nestable(lck)) { 798 KMP_FATAL(LockNestableUsedAsSimple, func); 799 } 800 if (__kmp_get_ticket_lock_owner(lck) != -1) { 801 KMP_FATAL(LockStillOwned, func); 802 } 803 __kmp_destroy_ticket_lock(lck); 804} 805 806// nested ticket locks 807 808int __kmp_acquire_nested_ticket_lock(kmp_ticket_lock_t *lck, kmp_int32 gtid) { 809 KMP_DEBUG_ASSERT(gtid >= 0); 810 811 if (__kmp_get_ticket_lock_owner(lck) == gtid) { 812 std::atomic_fetch_add_explicit(&lck->lk.depth_locked, 1, 813 std::memory_order_relaxed); 814 return KMP_LOCK_ACQUIRED_NEXT; 815 } else { 816 __kmp_acquire_ticket_lock_timed_template(lck, gtid); 817 ANNOTATE_TICKET_ACQUIRED(lck); 818 std::atomic_store_explicit(&lck->lk.depth_locked, 1, 819 std::memory_order_relaxed); 820 std::atomic_store_explicit(&lck->lk.owner_id, gtid + 1, 821 std::memory_order_relaxed); 822 return KMP_LOCK_ACQUIRED_FIRST; 823 } 824} 825 826static int __kmp_acquire_nested_ticket_lock_with_checks(kmp_ticket_lock_t *lck, 827 kmp_int32 gtid) { 828 char const *const func = "omp_set_nest_lock"; 829 830 if (!std::atomic_load_explicit(&lck->lk.initialized, 831 std::memory_order_relaxed)) { 832 KMP_FATAL(LockIsUninitialized, func); 833 } 834 if (lck->lk.self != lck) { 835 KMP_FATAL(LockIsUninitialized, func); 836 } 837 if (!__kmp_is_ticket_lock_nestable(lck)) { 838 KMP_FATAL(LockSimpleUsedAsNestable, func); 839 } 840 return __kmp_acquire_nested_ticket_lock(lck, gtid); 841} 842 843int __kmp_test_nested_ticket_lock(kmp_ticket_lock_t *lck, kmp_int32 gtid) { 844 int retval; 845 846 KMP_DEBUG_ASSERT(gtid >= 0); 847 848 if (__kmp_get_ticket_lock_owner(lck) == gtid) { 849 retval = std::atomic_fetch_add_explicit(&lck->lk.depth_locked, 1, 850 std::memory_order_relaxed) + 851 1; 852 } else if (!__kmp_test_ticket_lock(lck, gtid)) { 853 retval = 0; 854 } else { 855 std::atomic_store_explicit(&lck->lk.depth_locked, 1, 856 std::memory_order_relaxed); 857 std::atomic_store_explicit(&lck->lk.owner_id, gtid + 1, 858 std::memory_order_relaxed); 859 retval = 1; 860 } 861 return retval; 862} 863 864static int __kmp_test_nested_ticket_lock_with_checks(kmp_ticket_lock_t *lck, 865 kmp_int32 gtid) { 866 char const *const func = "omp_test_nest_lock"; 867 868 if (!std::atomic_load_explicit(&lck->lk.initialized, 869 std::memory_order_relaxed)) { 870 KMP_FATAL(LockIsUninitialized, func); 871 } 872 if (lck->lk.self != lck) { 873 KMP_FATAL(LockIsUninitialized, func); 874 } 875 if (!__kmp_is_ticket_lock_nestable(lck)) { 876 KMP_FATAL(LockSimpleUsedAsNestable, func); 877 } 878 return __kmp_test_nested_ticket_lock(lck, gtid); 879} 880 881int __kmp_release_nested_ticket_lock(kmp_ticket_lock_t *lck, kmp_int32 gtid) { 882 KMP_DEBUG_ASSERT(gtid >= 0); 883 884 if ((std::atomic_fetch_add_explicit(&lck->lk.depth_locked, -1, 885 std::memory_order_relaxed) - 886 1) == 0) { 887 std::atomic_store_explicit(&lck->lk.owner_id, 0, std::memory_order_relaxed); 888 __kmp_release_ticket_lock(lck, gtid); 889 return KMP_LOCK_RELEASED; 890 } 891 return KMP_LOCK_STILL_HELD; 892} 893 894static int __kmp_release_nested_ticket_lock_with_checks(kmp_ticket_lock_t *lck, 895 kmp_int32 gtid) { 896 char const *const func = "omp_unset_nest_lock"; 897 898 if (!std::atomic_load_explicit(&lck->lk.initialized, 899 std::memory_order_relaxed)) { 900 KMP_FATAL(LockIsUninitialized, func); 901 } 902 if (lck->lk.self != lck) { 903 KMP_FATAL(LockIsUninitialized, func); 904 } 905 if (!__kmp_is_ticket_lock_nestable(lck)) { 906 KMP_FATAL(LockSimpleUsedAsNestable, func); 907 } 908 if (__kmp_get_ticket_lock_owner(lck) == -1) { 909 KMP_FATAL(LockUnsettingFree, func); 910 } 911 if (__kmp_get_ticket_lock_owner(lck) != gtid) { 912 KMP_FATAL(LockUnsettingSetByAnother, func); 913 } 914 return __kmp_release_nested_ticket_lock(lck, gtid); 915} 916 917void __kmp_init_nested_ticket_lock(kmp_ticket_lock_t *lck) { 918 __kmp_init_ticket_lock(lck); 919 std::atomic_store_explicit(&lck->lk.depth_locked, 0, 920 std::memory_order_relaxed); 921 // >= 0 for nestable locks, -1 for simple locks 922} 923 924void __kmp_destroy_nested_ticket_lock(kmp_ticket_lock_t *lck) { 925 __kmp_destroy_ticket_lock(lck); 926 std::atomic_store_explicit(&lck->lk.depth_locked, 0, 927 std::memory_order_relaxed); 928} 929 930static void 931__kmp_destroy_nested_ticket_lock_with_checks(kmp_ticket_lock_t *lck) { 932 char const *const func = "omp_destroy_nest_lock"; 933 934 if (!std::atomic_load_explicit(&lck->lk.initialized, 935 std::memory_order_relaxed)) { 936 KMP_FATAL(LockIsUninitialized, func); 937 } 938 if (lck->lk.self != lck) { 939 KMP_FATAL(LockIsUninitialized, func); 940 } 941 if (!__kmp_is_ticket_lock_nestable(lck)) { 942 KMP_FATAL(LockSimpleUsedAsNestable, func); 943 } 944 if (__kmp_get_ticket_lock_owner(lck) != -1) { 945 KMP_FATAL(LockStillOwned, func); 946 } 947 __kmp_destroy_nested_ticket_lock(lck); 948} 949 950// access functions to fields which don't exist for all lock kinds. 951 952static const ident_t *__kmp_get_ticket_lock_location(kmp_ticket_lock_t *lck) { 953 return lck->lk.location; 954} 955 956static void __kmp_set_ticket_lock_location(kmp_ticket_lock_t *lck, 957 const ident_t *loc) { 958 lck->lk.location = loc; 959} 960 961static kmp_lock_flags_t __kmp_get_ticket_lock_flags(kmp_ticket_lock_t *lck) { 962 return lck->lk.flags; 963} 964 965static void __kmp_set_ticket_lock_flags(kmp_ticket_lock_t *lck, 966 kmp_lock_flags_t flags) { 967 lck->lk.flags = flags; 968} 969 970/* ------------------------------------------------------------------------ */ 971/* queuing locks */ 972 973/* First the states 974 (head,tail) = 0, 0 means lock is unheld, nobody on queue 975 UINT_MAX or -1, 0 means lock is held, nobody on queue 976 h, h means lock held or about to transition, 977 1 element on queue 978 h, t h <> t, means lock is held or about to 979 transition, >1 elements on queue 980 981 Now the transitions 982 Acquire(0,0) = -1 ,0 983 Release(0,0) = Error 984 Acquire(-1,0) = h ,h h > 0 985 Release(-1,0) = 0 ,0 986 Acquire(h,h) = h ,t h > 0, t > 0, h <> t 987 Release(h,h) = -1 ,0 h > 0 988 Acquire(h,t) = h ,t' h > 0, t > 0, t' > 0, h <> t, h <> t', t <> t' 989 Release(h,t) = h',t h > 0, t > 0, h <> t, h <> h', h' maybe = t 990 991 And pictorially 992 993 +-----+ 994 | 0, 0|------- release -------> Error 995 +-----+ 996 | ^ 997 acquire| |release 998 | | 999 | | 1000 v | 1001 +-----+ 1002 |-1, 0| 1003 +-----+ 1004 | ^ 1005 acquire| |release 1006 | | 1007 | | 1008 v | 1009 +-----+ 1010 | h, h| 1011 +-----+ 1012 | ^ 1013 acquire| |release 1014 | | 1015 | | 1016 v | 1017 +-----+ 1018 | h, t|----- acquire, release loopback ---+ 1019 +-----+ | 1020 ^ | 1021 | | 1022 +------------------------------------+ 1023 */ 1024 1025#ifdef DEBUG_QUEUING_LOCKS 1026 1027/* Stuff for circular trace buffer */ 1028#define TRACE_BUF_ELE 1024 1029static char traces[TRACE_BUF_ELE][128] = {0}; 1030static int tc = 0; 1031#define TRACE_LOCK(X, Y) \ 1032 KMP_SNPRINTF(traces[tc++ % TRACE_BUF_ELE], 128, "t%d at %s\n", X, Y); 1033#define TRACE_LOCK_T(X, Y, Z) \ 1034 KMP_SNPRINTF(traces[tc++ % TRACE_BUF_ELE], 128, "t%d at %s%d\n", X, Y, Z); 1035#define TRACE_LOCK_HT(X, Y, Z, Q) \ 1036 KMP_SNPRINTF(traces[tc++ % TRACE_BUF_ELE], 128, "t%d at %s %d,%d\n", X, Y, \ 1037 Z, Q); 1038 1039static void __kmp_dump_queuing_lock(kmp_info_t *this_thr, kmp_int32 gtid, 1040 kmp_queuing_lock_t *lck, kmp_int32 head_id, 1041 kmp_int32 tail_id) { 1042 kmp_int32 t, i; 1043 1044 __kmp_printf_no_lock("\n__kmp_dump_queuing_lock: TRACE BEGINS HERE! \n"); 1045 1046 i = tc % TRACE_BUF_ELE; 1047 __kmp_printf_no_lock("%s\n", traces[i]); 1048 i = (i + 1) % TRACE_BUF_ELE; 1049 while (i != (tc % TRACE_BUF_ELE)) { 1050 __kmp_printf_no_lock("%s", traces[i]); 1051 i = (i + 1) % TRACE_BUF_ELE; 1052 } 1053 __kmp_printf_no_lock("\n"); 1054 1055 __kmp_printf_no_lock("\n__kmp_dump_queuing_lock: gtid+1:%d, spin_here:%d, " 1056 "next_wait:%d, head_id:%d, tail_id:%d\n", 1057 gtid + 1, this_thr->th.th_spin_here, 1058 this_thr->th.th_next_waiting, head_id, tail_id); 1059 1060 __kmp_printf_no_lock("\t\thead: %d ", lck->lk.head_id); 1061 1062 if (lck->lk.head_id >= 1) { 1063 t = __kmp_threads[lck->lk.head_id - 1]->th.th_next_waiting; 1064 while (t > 0) { 1065 __kmp_printf_no_lock("-> %d ", t); 1066 t = __kmp_threads[t - 1]->th.th_next_waiting; 1067 } 1068 } 1069 __kmp_printf_no_lock("; tail: %d ", lck->lk.tail_id); 1070 __kmp_printf_no_lock("\n\n"); 1071} 1072 1073#endif /* DEBUG_QUEUING_LOCKS */ 1074 1075static kmp_int32 __kmp_get_queuing_lock_owner(kmp_queuing_lock_t *lck) { 1076 return TCR_4(lck->lk.owner_id) - 1; 1077} 1078 1079static inline bool __kmp_is_queuing_lock_nestable(kmp_queuing_lock_t *lck) { 1080 return lck->lk.depth_locked != -1; 1081} 1082 1083/* Acquire a lock using a the queuing lock implementation */ 1084template <bool takeTime> 1085/* [TLW] The unused template above is left behind because of what BEB believes 1086 is a potential compiler problem with __forceinline. */ 1087__forceinline static int 1088__kmp_acquire_queuing_lock_timed_template(kmp_queuing_lock_t *lck, 1089 kmp_int32 gtid) { 1090 kmp_info_t *this_thr = __kmp_thread_from_gtid(gtid); 1091 volatile kmp_int32 *head_id_p = &lck->lk.head_id; 1092 volatile kmp_int32 *tail_id_p = &lck->lk.tail_id; 1093 volatile kmp_uint32 *spin_here_p; 1094 kmp_int32 need_mf = 1; 1095 1096#if OMPT_SUPPORT 1097 ompt_state_t prev_state = ompt_state_undefined; 1098#endif 1099 1100 KA_TRACE(1000, 1101 ("__kmp_acquire_queuing_lock: lck:%p, T#%d entering\n", lck, gtid)); 1102 1103 KMP_FSYNC_PREPARE(lck); 1104 KMP_DEBUG_ASSERT(this_thr != NULL); 1105 spin_here_p = &this_thr->th.th_spin_here; 1106 1107#ifdef DEBUG_QUEUING_LOCKS 1108 TRACE_LOCK(gtid + 1, "acq ent"); 1109 if (*spin_here_p) 1110 __kmp_dump_queuing_lock(this_thr, gtid, lck, *head_id_p, *tail_id_p); 1111 if (this_thr->th.th_next_waiting != 0) 1112 __kmp_dump_queuing_lock(this_thr, gtid, lck, *head_id_p, *tail_id_p); 1113#endif 1114 KMP_DEBUG_ASSERT(!*spin_here_p); 1115 KMP_DEBUG_ASSERT(this_thr->th.th_next_waiting == 0); 1116 1117 /* The following st.rel to spin_here_p needs to precede the cmpxchg.acq to 1118 head_id_p that may follow, not just in execution order, but also in 1119 visibility order. This way, when a releasing thread observes the changes to 1120 the queue by this thread, it can rightly assume that spin_here_p has 1121 already been set to TRUE, so that when it sets spin_here_p to FALSE, it is 1122 not premature. If the releasing thread sets spin_here_p to FALSE before 1123 this thread sets it to TRUE, this thread will hang. */ 1124 *spin_here_p = TRUE; /* before enqueuing to prevent race */ 1125 1126 while (1) { 1127 kmp_int32 enqueued; 1128 kmp_int32 head; 1129 kmp_int32 tail; 1130 1131 head = *head_id_p; 1132 1133 switch (head) { 1134 1135 case -1: { 1136#ifdef DEBUG_QUEUING_LOCKS 1137 tail = *tail_id_p; 1138 TRACE_LOCK_HT(gtid + 1, "acq read: ", head, tail); 1139#endif 1140 tail = 0; /* to make sure next link asynchronously read is not set 1141 accidentally; this assignment prevents us from entering the 1142 if ( t > 0 ) condition in the enqueued case below, which is not 1143 necessary for this state transition */ 1144 1145 need_mf = 0; 1146 /* try (-1,0)->(tid,tid) */ 1147 enqueued = KMP_COMPARE_AND_STORE_ACQ64((volatile kmp_int64 *)tail_id_p, 1148 KMP_PACK_64(-1, 0), 1149 KMP_PACK_64(gtid + 1, gtid + 1)); 1150#ifdef DEBUG_QUEUING_LOCKS 1151 if (enqueued) 1152 TRACE_LOCK(gtid + 1, "acq enq: (-1,0)->(tid,tid)"); 1153#endif 1154 } break; 1155 1156 default: { 1157 tail = *tail_id_p; 1158 KMP_DEBUG_ASSERT(tail != gtid + 1); 1159 1160#ifdef DEBUG_QUEUING_LOCKS 1161 TRACE_LOCK_HT(gtid + 1, "acq read: ", head, tail); 1162#endif 1163 1164 if (tail == 0) { 1165 enqueued = FALSE; 1166 } else { 1167 need_mf = 0; 1168 /* try (h,t) or (h,h)->(h,tid) */ 1169 enqueued = KMP_COMPARE_AND_STORE_ACQ32(tail_id_p, tail, gtid + 1); 1170 1171#ifdef DEBUG_QUEUING_LOCKS 1172 if (enqueued) 1173 TRACE_LOCK(gtid + 1, "acq enq: (h,t)->(h,tid)"); 1174#endif 1175 } 1176 } break; 1177 1178 case 0: /* empty queue */ 1179 { 1180 kmp_int32 grabbed_lock; 1181 1182#ifdef DEBUG_QUEUING_LOCKS 1183 tail = *tail_id_p; 1184 TRACE_LOCK_HT(gtid + 1, "acq read: ", head, tail); 1185#endif 1186 /* try (0,0)->(-1,0) */ 1187 1188 /* only legal transition out of head = 0 is head = -1 with no change to 1189 * tail */ 1190 grabbed_lock = KMP_COMPARE_AND_STORE_ACQ32(head_id_p, 0, -1); 1191 1192 if (grabbed_lock) { 1193 1194 *spin_here_p = FALSE; 1195 1196 KA_TRACE( 1197 1000, 1198 ("__kmp_acquire_queuing_lock: lck:%p, T#%d exiting: no queuing\n", 1199 lck, gtid)); 1200#ifdef DEBUG_QUEUING_LOCKS 1201 TRACE_LOCK_HT(gtid + 1, "acq exit: ", head, 0); 1202#endif 1203 1204#if OMPT_SUPPORT 1205 if (ompt_enabled.enabled && prev_state != ompt_state_undefined) { 1206 /* change the state before clearing wait_id */ 1207 this_thr->th.ompt_thread_info.state = prev_state; 1208 this_thr->th.ompt_thread_info.wait_id = 0; 1209 } 1210#endif 1211 1212 KMP_FSYNC_ACQUIRED(lck); 1213 return KMP_LOCK_ACQUIRED_FIRST; /* lock holder cannot be on queue */ 1214 } 1215 enqueued = FALSE; 1216 } break; 1217 } 1218 1219#if OMPT_SUPPORT 1220 if (ompt_enabled.enabled && prev_state == ompt_state_undefined) { 1221 /* this thread will spin; set wait_id before entering wait state */ 1222 prev_state = this_thr->th.ompt_thread_info.state; 1223 this_thr->th.ompt_thread_info.wait_id = (uint64_t)lck; 1224 this_thr->th.ompt_thread_info.state = ompt_state_wait_lock; 1225 } 1226#endif 1227 1228 if (enqueued) { 1229 if (tail > 0) { 1230 kmp_info_t *tail_thr = __kmp_thread_from_gtid(tail - 1); 1231 KMP_ASSERT(tail_thr != NULL); 1232 tail_thr->th.th_next_waiting = gtid + 1; 1233 /* corresponding wait for this write in release code */ 1234 } 1235 KA_TRACE(1000, 1236 ("__kmp_acquire_queuing_lock: lck:%p, T#%d waiting for lock\n", 1237 lck, gtid)); 1238 1239 KMP_MB(); 1240 // ToDo: Use __kmp_wait_sleep or similar when blocktime != inf 1241 KMP_WAIT(spin_here_p, FALSE, KMP_EQ, lck); 1242 1243#ifdef DEBUG_QUEUING_LOCKS 1244 TRACE_LOCK(gtid + 1, "acq spin"); 1245 1246 if (this_thr->th.th_next_waiting != 0) 1247 __kmp_dump_queuing_lock(this_thr, gtid, lck, *head_id_p, *tail_id_p); 1248#endif 1249 KMP_DEBUG_ASSERT(this_thr->th.th_next_waiting == 0); 1250 KA_TRACE(1000, ("__kmp_acquire_queuing_lock: lck:%p, T#%d exiting: after " 1251 "waiting on queue\n", 1252 lck, gtid)); 1253 1254#ifdef DEBUG_QUEUING_LOCKS 1255 TRACE_LOCK(gtid + 1, "acq exit 2"); 1256#endif 1257 1258#if OMPT_SUPPORT 1259 /* change the state before clearing wait_id */ 1260 this_thr->th.ompt_thread_info.state = prev_state; 1261 this_thr->th.ompt_thread_info.wait_id = 0; 1262#endif 1263 1264 /* got lock, we were dequeued by the thread that released lock */ 1265 return KMP_LOCK_ACQUIRED_FIRST; 1266 } 1267 1268 /* Yield if number of threads > number of logical processors */ 1269 /* ToDo: Not sure why this should only be in oversubscription case, 1270 maybe should be traditional YIELD_INIT/YIELD_WHEN loop */ 1271 KMP_YIELD_OVERSUB(); 1272 1273#ifdef DEBUG_QUEUING_LOCKS 1274 TRACE_LOCK(gtid + 1, "acq retry"); 1275#endif 1276 } 1277 KMP_ASSERT2(0, "should not get here"); 1278 return KMP_LOCK_ACQUIRED_FIRST; 1279} 1280 1281int __kmp_acquire_queuing_lock(kmp_queuing_lock_t *lck, kmp_int32 gtid) { 1282 KMP_DEBUG_ASSERT(gtid >= 0); 1283 1284 int retval = __kmp_acquire_queuing_lock_timed_template<false>(lck, gtid); 1285 ANNOTATE_QUEUING_ACQUIRED(lck); 1286 return retval; 1287} 1288 1289static int __kmp_acquire_queuing_lock_with_checks(kmp_queuing_lock_t *lck, 1290 kmp_int32 gtid) { 1291 char const *const func = "omp_set_lock"; 1292 if (lck->lk.initialized != lck) { 1293 KMP_FATAL(LockIsUninitialized, func); 1294 } 1295 if (__kmp_is_queuing_lock_nestable(lck)) { 1296 KMP_FATAL(LockNestableUsedAsSimple, func); 1297 } 1298 if (__kmp_get_queuing_lock_owner(lck) == gtid) { 1299 KMP_FATAL(LockIsAlreadyOwned, func); 1300 } 1301 1302 __kmp_acquire_queuing_lock(lck, gtid); 1303 1304 lck->lk.owner_id = gtid + 1; 1305 return KMP_LOCK_ACQUIRED_FIRST; 1306} 1307 1308int __kmp_test_queuing_lock(kmp_queuing_lock_t *lck, kmp_int32 gtid) { 1309 volatile kmp_int32 *head_id_p = &lck->lk.head_id; 1310 kmp_int32 head; 1311#ifdef KMP_DEBUG 1312 kmp_info_t *this_thr; 1313#endif 1314 1315 KA_TRACE(1000, ("__kmp_test_queuing_lock: T#%d entering\n", gtid)); 1316 KMP_DEBUG_ASSERT(gtid >= 0); 1317#ifdef KMP_DEBUG 1318 this_thr = __kmp_thread_from_gtid(gtid); 1319 KMP_DEBUG_ASSERT(this_thr != NULL); 1320 KMP_DEBUG_ASSERT(!this_thr->th.th_spin_here); 1321#endif 1322 1323 head = *head_id_p; 1324 1325 if (head == 0) { /* nobody on queue, nobody holding */ 1326 /* try (0,0)->(-1,0) */ 1327 if (KMP_COMPARE_AND_STORE_ACQ32(head_id_p, 0, -1)) { 1328 KA_TRACE(1000, 1329 ("__kmp_test_queuing_lock: T#%d exiting: holding lock\n", gtid)); 1330 KMP_FSYNC_ACQUIRED(lck); 1331 ANNOTATE_QUEUING_ACQUIRED(lck); 1332 return TRUE; 1333 } 1334 } 1335 1336 KA_TRACE(1000, 1337 ("__kmp_test_queuing_lock: T#%d exiting: without lock\n", gtid)); 1338 return FALSE; 1339} 1340 1341static int __kmp_test_queuing_lock_with_checks(kmp_queuing_lock_t *lck, 1342 kmp_int32 gtid) { 1343 char const *const func = "omp_test_lock"; 1344 if (lck->lk.initialized != lck) { 1345 KMP_FATAL(LockIsUninitialized, func); 1346 } 1347 if (__kmp_is_queuing_lock_nestable(lck)) { 1348 KMP_FATAL(LockNestableUsedAsSimple, func); 1349 } 1350 1351 int retval = __kmp_test_queuing_lock(lck, gtid); 1352 1353 if (retval) { 1354 lck->lk.owner_id = gtid + 1; 1355 } 1356 return retval; 1357} 1358 1359int __kmp_release_queuing_lock(kmp_queuing_lock_t *lck, kmp_int32 gtid) { 1360 kmp_info_t *this_thr; 1361 volatile kmp_int32 *head_id_p = &lck->lk.head_id; 1362 volatile kmp_int32 *tail_id_p = &lck->lk.tail_id; 1363 1364 KA_TRACE(1000, 1365 ("__kmp_release_queuing_lock: lck:%p, T#%d entering\n", lck, gtid)); 1366 KMP_DEBUG_ASSERT(gtid >= 0); 1367 this_thr = __kmp_thread_from_gtid(gtid); 1368 KMP_DEBUG_ASSERT(this_thr != NULL); 1369#ifdef DEBUG_QUEUING_LOCKS 1370 TRACE_LOCK(gtid + 1, "rel ent"); 1371 1372 if (this_thr->th.th_spin_here) 1373 __kmp_dump_queuing_lock(this_thr, gtid, lck, *head_id_p, *tail_id_p); 1374 if (this_thr->th.th_next_waiting != 0) 1375 __kmp_dump_queuing_lock(this_thr, gtid, lck, *head_id_p, *tail_id_p); 1376#endif 1377 KMP_DEBUG_ASSERT(!this_thr->th.th_spin_here); 1378 KMP_DEBUG_ASSERT(this_thr->th.th_next_waiting == 0); 1379 1380 KMP_FSYNC_RELEASING(lck); 1381 ANNOTATE_QUEUING_RELEASED(lck); 1382 1383 while (1) { 1384 kmp_int32 dequeued; 1385 kmp_int32 head; 1386 kmp_int32 tail; 1387 1388 head = *head_id_p; 1389 1390#ifdef DEBUG_QUEUING_LOCKS 1391 tail = *tail_id_p; 1392 TRACE_LOCK_HT(gtid + 1, "rel read: ", head, tail); 1393 if (head == 0) 1394 __kmp_dump_queuing_lock(this_thr, gtid, lck, head, tail); 1395#endif 1396 KMP_DEBUG_ASSERT(head != 1397 0); /* holding the lock, head must be -1 or queue head */ 1398 1399 if (head == -1) { /* nobody on queue */ 1400 /* try (-1,0)->(0,0) */ 1401 if (KMP_COMPARE_AND_STORE_REL32(head_id_p, -1, 0)) { 1402 KA_TRACE( 1403 1000, 1404 ("__kmp_release_queuing_lock: lck:%p, T#%d exiting: queue empty\n", 1405 lck, gtid)); 1406#ifdef DEBUG_QUEUING_LOCKS 1407 TRACE_LOCK_HT(gtid + 1, "rel exit: ", 0, 0); 1408#endif 1409 1410#if OMPT_SUPPORT 1411/* nothing to do - no other thread is trying to shift blame */ 1412#endif 1413 return KMP_LOCK_RELEASED; 1414 } 1415 dequeued = FALSE; 1416 } else { 1417 KMP_MB(); 1418 tail = *tail_id_p; 1419 if (head == tail) { /* only one thread on the queue */ 1420#ifdef DEBUG_QUEUING_LOCKS 1421 if (head <= 0) 1422 __kmp_dump_queuing_lock(this_thr, gtid, lck, head, tail); 1423#endif 1424 KMP_DEBUG_ASSERT(head > 0); 1425 1426 /* try (h,h)->(-1,0) */ 1427 dequeued = KMP_COMPARE_AND_STORE_REL64( 1428 RCAST(volatile kmp_int64 *, tail_id_p), KMP_PACK_64(head, head), 1429 KMP_PACK_64(-1, 0)); 1430#ifdef DEBUG_QUEUING_LOCKS 1431 TRACE_LOCK(gtid + 1, "rel deq: (h,h)->(-1,0)"); 1432#endif 1433 1434 } else { 1435 volatile kmp_int32 *waiting_id_p; 1436 kmp_info_t *head_thr = __kmp_thread_from_gtid(head - 1); 1437 KMP_DEBUG_ASSERT(head_thr != NULL); 1438 waiting_id_p = &head_thr->th.th_next_waiting; 1439 1440/* Does this require synchronous reads? */ 1441#ifdef DEBUG_QUEUING_LOCKS 1442 if (head <= 0 || tail <= 0) 1443 __kmp_dump_queuing_lock(this_thr, gtid, lck, head, tail); 1444#endif 1445 KMP_DEBUG_ASSERT(head > 0 && tail > 0); 1446 1447 /* try (h,t)->(h',t) or (t,t) */ 1448 KMP_MB(); 1449 /* make sure enqueuing thread has time to update next waiting thread 1450 * field */ 1451 *head_id_p = 1452 KMP_WAIT((volatile kmp_uint32 *)waiting_id_p, 0, KMP_NEQ, NULL); 1453#ifdef DEBUG_QUEUING_LOCKS 1454 TRACE_LOCK(gtid + 1, "rel deq: (h,t)->(h',t)"); 1455#endif 1456 dequeued = TRUE; 1457 } 1458 } 1459 1460 if (dequeued) { 1461 kmp_info_t *head_thr = __kmp_thread_from_gtid(head - 1); 1462 KMP_DEBUG_ASSERT(head_thr != NULL); 1463 1464/* Does this require synchronous reads? */ 1465#ifdef DEBUG_QUEUING_LOCKS 1466 if (head <= 0 || tail <= 0) 1467 __kmp_dump_queuing_lock(this_thr, gtid, lck, head, tail); 1468#endif 1469 KMP_DEBUG_ASSERT(head > 0 && tail > 0); 1470 1471 /* For clean code only. Thread not released until next statement prevents 1472 race with acquire code. */ 1473 head_thr->th.th_next_waiting = 0; 1474#ifdef DEBUG_QUEUING_LOCKS 1475 TRACE_LOCK_T(gtid + 1, "rel nw=0 for t=", head); 1476#endif 1477 1478 KMP_MB(); 1479 /* reset spin value */ 1480 head_thr->th.th_spin_here = FALSE; 1481 1482 KA_TRACE(1000, ("__kmp_release_queuing_lock: lck:%p, T#%d exiting: after " 1483 "dequeuing\n", 1484 lck, gtid)); 1485#ifdef DEBUG_QUEUING_LOCKS 1486 TRACE_LOCK(gtid + 1, "rel exit 2"); 1487#endif 1488 return KMP_LOCK_RELEASED; 1489 } 1490/* KMP_CPU_PAUSE(); don't want to make releasing thread hold up acquiring 1491 threads */ 1492 1493#ifdef DEBUG_QUEUING_LOCKS 1494 TRACE_LOCK(gtid + 1, "rel retry"); 1495#endif 1496 1497 } /* while */ 1498 KMP_ASSERT2(0, "should not get here"); 1499 return KMP_LOCK_RELEASED; 1500} 1501 1502static int __kmp_release_queuing_lock_with_checks(kmp_queuing_lock_t *lck, 1503 kmp_int32 gtid) { 1504 char const *const func = "omp_unset_lock"; 1505 KMP_MB(); /* in case another processor initialized lock */ 1506 if (lck->lk.initialized != lck) { 1507 KMP_FATAL(LockIsUninitialized, func); 1508 } 1509 if (__kmp_is_queuing_lock_nestable(lck)) { 1510 KMP_FATAL(LockNestableUsedAsSimple, func); 1511 } 1512 if (__kmp_get_queuing_lock_owner(lck) == -1) { 1513 KMP_FATAL(LockUnsettingFree, func); 1514 } 1515 if (__kmp_get_queuing_lock_owner(lck) != gtid) { 1516 KMP_FATAL(LockUnsettingSetByAnother, func); 1517 } 1518 lck->lk.owner_id = 0; 1519 return __kmp_release_queuing_lock(lck, gtid); 1520} 1521 1522void __kmp_init_queuing_lock(kmp_queuing_lock_t *lck) { 1523 lck->lk.location = NULL; 1524 lck->lk.head_id = 0; 1525 lck->lk.tail_id = 0; 1526 lck->lk.next_ticket = 0; 1527 lck->lk.now_serving = 0; 1528 lck->lk.owner_id = 0; // no thread owns the lock. 1529 lck->lk.depth_locked = -1; // >= 0 for nestable locks, -1 for simple locks. 1530 lck->lk.initialized = lck; 1531 1532 KA_TRACE(1000, ("__kmp_init_queuing_lock: lock %p initialized\n", lck)); 1533} 1534 1535void __kmp_destroy_queuing_lock(kmp_queuing_lock_t *lck) { 1536 lck->lk.initialized = NULL; 1537 lck->lk.location = NULL; 1538 lck->lk.head_id = 0; 1539 lck->lk.tail_id = 0; 1540 lck->lk.next_ticket = 0; 1541 lck->lk.now_serving = 0; 1542 lck->lk.owner_id = 0; 1543 lck->lk.depth_locked = -1; 1544} 1545 1546static void __kmp_destroy_queuing_lock_with_checks(kmp_queuing_lock_t *lck) { 1547 char const *const func = "omp_destroy_lock"; 1548 if (lck->lk.initialized != lck) { 1549 KMP_FATAL(LockIsUninitialized, func); 1550 } 1551 if (__kmp_is_queuing_lock_nestable(lck)) { 1552 KMP_FATAL(LockNestableUsedAsSimple, func); 1553 } 1554 if (__kmp_get_queuing_lock_owner(lck) != -1) { 1555 KMP_FATAL(LockStillOwned, func); 1556 } 1557 __kmp_destroy_queuing_lock(lck); 1558} 1559 1560// nested queuing locks 1561 1562int __kmp_acquire_nested_queuing_lock(kmp_queuing_lock_t *lck, kmp_int32 gtid) { 1563 KMP_DEBUG_ASSERT(gtid >= 0); 1564 1565 if (__kmp_get_queuing_lock_owner(lck) == gtid) { 1566 lck->lk.depth_locked += 1; 1567 return KMP_LOCK_ACQUIRED_NEXT; 1568 } else { 1569 __kmp_acquire_queuing_lock_timed_template<false>(lck, gtid); 1570 ANNOTATE_QUEUING_ACQUIRED(lck); 1571 KMP_MB(); 1572 lck->lk.depth_locked = 1; 1573 KMP_MB(); 1574 lck->lk.owner_id = gtid + 1; 1575 return KMP_LOCK_ACQUIRED_FIRST; 1576 } 1577} 1578 1579static int 1580__kmp_acquire_nested_queuing_lock_with_checks(kmp_queuing_lock_t *lck, 1581 kmp_int32 gtid) { 1582 char const *const func = "omp_set_nest_lock"; 1583 if (lck->lk.initialized != lck) { 1584 KMP_FATAL(LockIsUninitialized, func); 1585 } 1586 if (!__kmp_is_queuing_lock_nestable(lck)) { 1587 KMP_FATAL(LockSimpleUsedAsNestable, func); 1588 } 1589 return __kmp_acquire_nested_queuing_lock(lck, gtid); 1590} 1591 1592int __kmp_test_nested_queuing_lock(kmp_queuing_lock_t *lck, kmp_int32 gtid) { 1593 int retval; 1594 1595 KMP_DEBUG_ASSERT(gtid >= 0); 1596 1597 if (__kmp_get_queuing_lock_owner(lck) == gtid) { 1598 retval = ++lck->lk.depth_locked; 1599 } else if (!__kmp_test_queuing_lock(lck, gtid)) { 1600 retval = 0; 1601 } else { 1602 KMP_MB(); 1603 retval = lck->lk.depth_locked = 1; 1604 KMP_MB(); 1605 lck->lk.owner_id = gtid + 1; 1606 } 1607 return retval; 1608} 1609 1610static int __kmp_test_nested_queuing_lock_with_checks(kmp_queuing_lock_t *lck, 1611 kmp_int32 gtid) { 1612 char const *const func = "omp_test_nest_lock"; 1613 if (lck->lk.initialized != lck) { 1614 KMP_FATAL(LockIsUninitialized, func); 1615 } 1616 if (!__kmp_is_queuing_lock_nestable(lck)) { 1617 KMP_FATAL(LockSimpleUsedAsNestable, func); 1618 } 1619 return __kmp_test_nested_queuing_lock(lck, gtid); 1620} 1621 1622int __kmp_release_nested_queuing_lock(kmp_queuing_lock_t *lck, kmp_int32 gtid) { 1623 KMP_DEBUG_ASSERT(gtid >= 0); 1624 1625 KMP_MB(); 1626 if (--(lck->lk.depth_locked) == 0) { 1627 KMP_MB(); 1628 lck->lk.owner_id = 0; 1629 __kmp_release_queuing_lock(lck, gtid); 1630 return KMP_LOCK_RELEASED; 1631 } 1632 return KMP_LOCK_STILL_HELD; 1633} 1634 1635static int 1636__kmp_release_nested_queuing_lock_with_checks(kmp_queuing_lock_t *lck, 1637 kmp_int32 gtid) { 1638 char const *const func = "omp_unset_nest_lock"; 1639 KMP_MB(); /* in case another processor initialized lock */ 1640 if (lck->lk.initialized != lck) { 1641 KMP_FATAL(LockIsUninitialized, func); 1642 } 1643 if (!__kmp_is_queuing_lock_nestable(lck)) { 1644 KMP_FATAL(LockSimpleUsedAsNestable, func); 1645 } 1646 if (__kmp_get_queuing_lock_owner(lck) == -1) { 1647 KMP_FATAL(LockUnsettingFree, func); 1648 } 1649 if (__kmp_get_queuing_lock_owner(lck) != gtid) { 1650 KMP_FATAL(LockUnsettingSetByAnother, func); 1651 } 1652 return __kmp_release_nested_queuing_lock(lck, gtid); 1653} 1654 1655void __kmp_init_nested_queuing_lock(kmp_queuing_lock_t *lck) { 1656 __kmp_init_queuing_lock(lck); 1657 lck->lk.depth_locked = 0; // >= 0 for nestable locks, -1 for simple locks 1658} 1659 1660void __kmp_destroy_nested_queuing_lock(kmp_queuing_lock_t *lck) { 1661 __kmp_destroy_queuing_lock(lck); 1662 lck->lk.depth_locked = 0; 1663} 1664 1665static void 1666__kmp_destroy_nested_queuing_lock_with_checks(kmp_queuing_lock_t *lck) { 1667 char const *const func = "omp_destroy_nest_lock"; 1668 if (lck->lk.initialized != lck) { 1669 KMP_FATAL(LockIsUninitialized, func); 1670 } 1671 if (!__kmp_is_queuing_lock_nestable(lck)) { 1672 KMP_FATAL(LockSimpleUsedAsNestable, func); 1673 } 1674 if (__kmp_get_queuing_lock_owner(lck) != -1) { 1675 KMP_FATAL(LockStillOwned, func); 1676 } 1677 __kmp_destroy_nested_queuing_lock(lck); 1678} 1679 1680// access functions to fields which don't exist for all lock kinds. 1681 1682static const ident_t *__kmp_get_queuing_lock_location(kmp_queuing_lock_t *lck) { 1683 return lck->lk.location; 1684} 1685 1686static void __kmp_set_queuing_lock_location(kmp_queuing_lock_t *lck, 1687 const ident_t *loc) { 1688 lck->lk.location = loc; 1689} 1690 1691static kmp_lock_flags_t __kmp_get_queuing_lock_flags(kmp_queuing_lock_t *lck) { 1692 return lck->lk.flags; 1693} 1694 1695static void __kmp_set_queuing_lock_flags(kmp_queuing_lock_t *lck, 1696 kmp_lock_flags_t flags) { 1697 lck->lk.flags = flags; 1698} 1699 1700#if KMP_USE_ADAPTIVE_LOCKS 1701 1702/* RTM Adaptive locks */ 1703 1704#if (KMP_COMPILER_ICC && __INTEL_COMPILER >= 1300) || \ 1705 (KMP_COMPILER_MSVC && _MSC_VER >= 1700) || \ 1706 (KMP_COMPILER_CLANG && KMP_MSVC_COMPAT) 1707 1708#include <immintrin.h> 1709#define SOFT_ABORT_MASK (_XABORT_RETRY | _XABORT_CONFLICT | _XABORT_EXPLICIT) 1710 1711#else 1712 1713// Values from the status register after failed speculation. 1714#define _XBEGIN_STARTED (~0u) 1715#define _XABORT_EXPLICIT (1 << 0) 1716#define _XABORT_RETRY (1 << 1) 1717#define _XABORT_CONFLICT (1 << 2) 1718#define _XABORT_CAPACITY (1 << 3) 1719#define _XABORT_DEBUG (1 << 4) 1720#define _XABORT_NESTED (1 << 5) 1721#define _XABORT_CODE(x) ((unsigned char)(((x) >> 24) & 0xFF)) 1722 1723// Aborts for which it's worth trying again immediately 1724#define SOFT_ABORT_MASK (_XABORT_RETRY | _XABORT_CONFLICT | _XABORT_EXPLICIT) 1725 1726#define STRINGIZE_INTERNAL(arg) #arg 1727#define STRINGIZE(arg) STRINGIZE_INTERNAL(arg) 1728 1729// Access to RTM instructions 1730/*A version of XBegin which returns -1 on speculation, and the value of EAX on 1731 an abort. This is the same definition as the compiler intrinsic that will be 1732 supported at some point. */ 1733static __inline int _xbegin() { 1734 int res = -1; 1735 1736#if KMP_OS_WINDOWS 1737#if KMP_ARCH_X86_64 1738 _asm { 1739 _emit 0xC7 1740 _emit 0xF8 1741 _emit 2 1742 _emit 0 1743 _emit 0 1744 _emit 0 1745 jmp L2 1746 mov res, eax 1747 L2: 1748 } 1749#else /* IA32 */ 1750 _asm { 1751 _emit 0xC7 1752 _emit 0xF8 1753 _emit 2 1754 _emit 0 1755 _emit 0 1756 _emit 0 1757 jmp L2 1758 mov res, eax 1759 L2: 1760 } 1761#endif // KMP_ARCH_X86_64 1762#else 1763 /* Note that %eax must be noted as killed (clobbered), because the XSR is 1764 returned in %eax(%rax) on abort. Other register values are restored, so 1765 don't need to be killed. 1766 1767 We must also mark 'res' as an input and an output, since otherwise 1768 'res=-1' may be dropped as being dead, whereas we do need the assignment on 1769 the successful (i.e., non-abort) path. */ 1770 __asm__ volatile("1: .byte 0xC7; .byte 0xF8;\n" 1771 " .long 1f-1b-6\n" 1772 " jmp 2f\n" 1773 "1: movl %%eax,%0\n" 1774 "2:" 1775 : "+r"(res)::"memory", "%eax"); 1776#endif // KMP_OS_WINDOWS 1777 return res; 1778} 1779 1780/* Transaction end */ 1781static __inline void _xend() { 1782#if KMP_OS_WINDOWS 1783 __asm { 1784 _emit 0x0f 1785 _emit 0x01 1786 _emit 0xd5 1787 } 1788#else 1789 __asm__ volatile(".byte 0x0f; .byte 0x01; .byte 0xd5" ::: "memory"); 1790#endif 1791} 1792 1793/* This is a macro, the argument must be a single byte constant which can be 1794 evaluated by the inline assembler, since it is emitted as a byte into the 1795 assembly code. */ 1796// clang-format off 1797#if KMP_OS_WINDOWS 1798#define _xabort(ARG) _asm _emit 0xc6 _asm _emit 0xf8 _asm _emit ARG 1799#else 1800#define _xabort(ARG) \ 1801 __asm__ volatile(".byte 0xC6; .byte 0xF8; .byte " STRINGIZE(ARG):::"memory"); 1802#endif 1803// clang-format on 1804#endif // KMP_COMPILER_ICC && __INTEL_COMPILER >= 1300 1805 1806// Statistics is collected for testing purpose 1807#if KMP_DEBUG_ADAPTIVE_LOCKS 1808 1809// We accumulate speculative lock statistics when the lock is destroyed. We 1810// keep locks that haven't been destroyed in the liveLocks list so that we can 1811// grab their statistics too. 1812static kmp_adaptive_lock_statistics_t destroyedStats; 1813 1814// To hold the list of live locks. 1815static kmp_adaptive_lock_info_t liveLocks; 1816 1817// A lock so we can safely update the list of locks. 1818static kmp_bootstrap_lock_t chain_lock = 1819 KMP_BOOTSTRAP_LOCK_INITIALIZER(chain_lock); 1820 1821// Initialize the list of stats. 1822void __kmp_init_speculative_stats() { 1823 kmp_adaptive_lock_info_t *lck = &liveLocks; 1824 1825 memset(CCAST(kmp_adaptive_lock_statistics_t *, &(lck->stats)), 0, 1826 sizeof(lck->stats)); 1827 lck->stats.next = lck; 1828 lck->stats.prev = lck; 1829 1830 KMP_ASSERT(lck->stats.next->stats.prev == lck); 1831 KMP_ASSERT(lck->stats.prev->stats.next == lck); 1832 1833 __kmp_init_bootstrap_lock(&chain_lock); 1834} 1835 1836// Insert the lock into the circular list 1837static void __kmp_remember_lock(kmp_adaptive_lock_info_t *lck) { 1838 __kmp_acquire_bootstrap_lock(&chain_lock); 1839 1840 lck->stats.next = liveLocks.stats.next; 1841 lck->stats.prev = &liveLocks; 1842 1843 liveLocks.stats.next = lck; 1844 lck->stats.next->stats.prev = lck; 1845 1846 KMP_ASSERT(lck->stats.next->stats.prev == lck); 1847 KMP_ASSERT(lck->stats.prev->stats.next == lck); 1848 1849 __kmp_release_bootstrap_lock(&chain_lock); 1850} 1851 1852static void __kmp_forget_lock(kmp_adaptive_lock_info_t *lck) { 1853 KMP_ASSERT(lck->stats.next->stats.prev == lck); 1854 KMP_ASSERT(lck->stats.prev->stats.next == lck); 1855 1856 kmp_adaptive_lock_info_t *n = lck->stats.next; 1857 kmp_adaptive_lock_info_t *p = lck->stats.prev; 1858 1859 n->stats.prev = p; 1860 p->stats.next = n; 1861} 1862 1863static void __kmp_zero_speculative_stats(kmp_adaptive_lock_info_t *lck) { 1864 memset(CCAST(kmp_adaptive_lock_statistics_t *, &lck->stats), 0, 1865 sizeof(lck->stats)); 1866 __kmp_remember_lock(lck); 1867} 1868 1869static void __kmp_add_stats(kmp_adaptive_lock_statistics_t *t, 1870 kmp_adaptive_lock_info_t *lck) { 1871 kmp_adaptive_lock_statistics_t volatile *s = &lck->stats; 1872 1873 t->nonSpeculativeAcquireAttempts += lck->acquire_attempts; 1874 t->successfulSpeculations += s->successfulSpeculations; 1875 t->hardFailedSpeculations += s->hardFailedSpeculations; 1876 t->softFailedSpeculations += s->softFailedSpeculations; 1877 t->nonSpeculativeAcquires += s->nonSpeculativeAcquires; 1878 t->lemmingYields += s->lemmingYields; 1879} 1880 1881static void __kmp_accumulate_speculative_stats(kmp_adaptive_lock_info_t *lck) { 1882 __kmp_acquire_bootstrap_lock(&chain_lock); 1883 1884 __kmp_add_stats(&destroyedStats, lck); 1885 __kmp_forget_lock(lck); 1886 1887 __kmp_release_bootstrap_lock(&chain_lock); 1888} 1889 1890static float percent(kmp_uint32 count, kmp_uint32 total) { 1891 return (total == 0) ? 0.0 : (100.0 * count) / total; 1892} 1893 1894static FILE *__kmp_open_stats_file() { 1895 if (strcmp(__kmp_speculative_statsfile, "-") == 0) 1896 return stdout; 1897 1898 size_t buffLen = KMP_STRLEN(__kmp_speculative_statsfile) + 20; 1899 char buffer[buffLen]; 1900 KMP_SNPRINTF(&buffer[0], buffLen, __kmp_speculative_statsfile, 1901 (kmp_int32)getpid()); 1902 FILE *result = fopen(&buffer[0], "w"); 1903 1904 // Maybe we should issue a warning here... 1905 return result ? result : stdout; 1906} 1907 1908void __kmp_print_speculative_stats() { 1909 kmp_adaptive_lock_statistics_t total = destroyedStats; 1910 kmp_adaptive_lock_info_t *lck; 1911 1912 for (lck = liveLocks.stats.next; lck != &liveLocks; lck = lck->stats.next) { 1913 __kmp_add_stats(&total, lck); 1914 } 1915 kmp_adaptive_lock_statistics_t *t = &total; 1916 kmp_uint32 totalSections = 1917 t->nonSpeculativeAcquires + t->successfulSpeculations; 1918 kmp_uint32 totalSpeculations = t->successfulSpeculations + 1919 t->hardFailedSpeculations + 1920 t->softFailedSpeculations; 1921 if (totalSections <= 0) 1922 return; 1923 1924 FILE *statsFile = __kmp_open_stats_file(); 1925 1926 fprintf(statsFile, "Speculative lock statistics (all approximate!)\n"); 1927 fprintf(statsFile, " Lock parameters: \n" 1928 " max_soft_retries : %10d\n" 1929 " max_badness : %10d\n", 1930 __kmp_adaptive_backoff_params.max_soft_retries, 1931 __kmp_adaptive_backoff_params.max_badness); 1932 fprintf(statsFile, " Non-speculative acquire attempts : %10d\n", 1933 t->nonSpeculativeAcquireAttempts); 1934 fprintf(statsFile, " Total critical sections : %10d\n", 1935 totalSections); 1936 fprintf(statsFile, " Successful speculations : %10d (%5.1f%%)\n", 1937 t->successfulSpeculations, 1938 percent(t->successfulSpeculations, totalSections)); 1939 fprintf(statsFile, " Non-speculative acquires : %10d (%5.1f%%)\n", 1940 t->nonSpeculativeAcquires, 1941 percent(t->nonSpeculativeAcquires, totalSections)); 1942 fprintf(statsFile, " Lemming yields : %10d\n\n", 1943 t->lemmingYields); 1944 1945 fprintf(statsFile, " Speculative acquire attempts : %10d\n", 1946 totalSpeculations); 1947 fprintf(statsFile, " Successes : %10d (%5.1f%%)\n", 1948 t->successfulSpeculations, 1949 percent(t->successfulSpeculations, totalSpeculations)); 1950 fprintf(statsFile, " Soft failures : %10d (%5.1f%%)\n", 1951 t->softFailedSpeculations, 1952 percent(t->softFailedSpeculations, totalSpeculations)); 1953 fprintf(statsFile, " Hard failures : %10d (%5.1f%%)\n", 1954 t->hardFailedSpeculations, 1955 percent(t->hardFailedSpeculations, totalSpeculations)); 1956 1957 if (statsFile != stdout) 1958 fclose(statsFile); 1959} 1960 1961#define KMP_INC_STAT(lck, stat) (lck->lk.adaptive.stats.stat++) 1962#else 1963#define KMP_INC_STAT(lck, stat) 1964 1965#endif // KMP_DEBUG_ADAPTIVE_LOCKS 1966 1967static inline bool __kmp_is_unlocked_queuing_lock(kmp_queuing_lock_t *lck) { 1968 // It is enough to check that the head_id is zero. 1969 // We don't also need to check the tail. 1970 bool res = lck->lk.head_id == 0; 1971 1972// We need a fence here, since we must ensure that no memory operations 1973// from later in this thread float above that read. 1974#if KMP_COMPILER_ICC 1975 _mm_mfence(); 1976#else 1977 __sync_synchronize(); 1978#endif 1979 1980 return res; 1981} 1982 1983// Functions for manipulating the badness 1984static __inline void 1985__kmp_update_badness_after_success(kmp_adaptive_lock_t *lck) { 1986 // Reset the badness to zero so we eagerly try to speculate again 1987 lck->lk.adaptive.badness = 0; 1988 KMP_INC_STAT(lck, successfulSpeculations); 1989} 1990 1991// Create a bit mask with one more set bit. 1992static __inline void __kmp_step_badness(kmp_adaptive_lock_t *lck) { 1993 kmp_uint32 newBadness = (lck->lk.adaptive.badness << 1) | 1; 1994 if (newBadness > lck->lk.adaptive.max_badness) { 1995 return; 1996 } else { 1997 lck->lk.adaptive.badness = newBadness; 1998 } 1999} 2000 2001// Check whether speculation should be attempted. 2002static __inline int __kmp_should_speculate(kmp_adaptive_lock_t *lck, 2003 kmp_int32 gtid) { 2004 kmp_uint32 badness = lck->lk.adaptive.badness; 2005 kmp_uint32 attempts = lck->lk.adaptive.acquire_attempts; 2006 int res = (attempts & badness) == 0; 2007 return res; 2008} 2009 2010// Attempt to acquire only the speculative lock. 2011// Does not back off to the non-speculative lock. 2012static int __kmp_test_adaptive_lock_only(kmp_adaptive_lock_t *lck, 2013 kmp_int32 gtid) { 2014 int retries = lck->lk.adaptive.max_soft_retries; 2015 2016 // We don't explicitly count the start of speculation, rather we record the 2017 // results (success, hard fail, soft fail). The sum of all of those is the 2018 // total number of times we started speculation since all speculations must 2019 // end one of those ways. 2020 do { 2021 kmp_uint32 status = _xbegin(); 2022 // Switch this in to disable actual speculation but exercise at least some 2023 // of the rest of the code. Useful for debugging... 2024 // kmp_uint32 status = _XABORT_NESTED; 2025 2026 if (status == _XBEGIN_STARTED) { 2027 /* We have successfully started speculation. Check that no-one acquired 2028 the lock for real between when we last looked and now. This also gets 2029 the lock cache line into our read-set, which we need so that we'll 2030 abort if anyone later claims it for real. */ 2031 if (!__kmp_is_unlocked_queuing_lock(GET_QLK_PTR(lck))) { 2032 // Lock is now visibly acquired, so someone beat us to it. Abort the 2033 // transaction so we'll restart from _xbegin with the failure status. 2034 _xabort(0x01); 2035 KMP_ASSERT2(0, "should not get here"); 2036 } 2037 return 1; // Lock has been acquired (speculatively) 2038 } else { 2039 // We have aborted, update the statistics 2040 if (status & SOFT_ABORT_MASK) { 2041 KMP_INC_STAT(lck, softFailedSpeculations); 2042 // and loop round to retry. 2043 } else { 2044 KMP_INC_STAT(lck, hardFailedSpeculations); 2045 // Give up if we had a hard failure. 2046 break; 2047 } 2048 } 2049 } while (retries--); // Loop while we have retries, and didn't fail hard. 2050 2051 // Either we had a hard failure or we didn't succeed softly after 2052 // the full set of attempts, so back off the badness. 2053 __kmp_step_badness(lck); 2054 return 0; 2055} 2056 2057// Attempt to acquire the speculative lock, or back off to the non-speculative 2058// one if the speculative lock cannot be acquired. 2059// We can succeed speculatively, non-speculatively, or fail. 2060static int __kmp_test_adaptive_lock(kmp_adaptive_lock_t *lck, kmp_int32 gtid) { 2061 // First try to acquire the lock speculatively 2062 if (__kmp_should_speculate(lck, gtid) && 2063 __kmp_test_adaptive_lock_only(lck, gtid)) 2064 return 1; 2065 2066 // Speculative acquisition failed, so try to acquire it non-speculatively. 2067 // Count the non-speculative acquire attempt 2068 lck->lk.adaptive.acquire_attempts++; 2069 2070 // Use base, non-speculative lock. 2071 if (__kmp_test_queuing_lock(GET_QLK_PTR(lck), gtid)) { 2072 KMP_INC_STAT(lck, nonSpeculativeAcquires); 2073 return 1; // Lock is acquired (non-speculatively) 2074 } else { 2075 return 0; // Failed to acquire the lock, it's already visibly locked. 2076 } 2077} 2078 2079static int __kmp_test_adaptive_lock_with_checks(kmp_adaptive_lock_t *lck, 2080 kmp_int32 gtid) { 2081 char const *const func = "omp_test_lock"; 2082 if (lck->lk.qlk.initialized != GET_QLK_PTR(lck)) { 2083 KMP_FATAL(LockIsUninitialized, func); 2084 } 2085 2086 int retval = __kmp_test_adaptive_lock(lck, gtid); 2087 2088 if (retval) { 2089 lck->lk.qlk.owner_id = gtid + 1; 2090 } 2091 return retval; 2092} 2093 2094// Block until we can acquire a speculative, adaptive lock. We check whether we 2095// should be trying to speculate. If we should be, we check the real lock to see 2096// if it is free, and, if not, pause without attempting to acquire it until it 2097// is. Then we try the speculative acquire. This means that although we suffer 2098// from lemmings a little (because all we can't acquire the lock speculatively 2099// until the queue of threads waiting has cleared), we don't get into a state 2100// where we can never acquire the lock speculatively (because we force the queue 2101// to clear by preventing new arrivals from entering the queue). This does mean 2102// that when we're trying to break lemmings, the lock is no longer fair. However 2103// OpenMP makes no guarantee that its locks are fair, so this isn't a real 2104// problem. 2105static void __kmp_acquire_adaptive_lock(kmp_adaptive_lock_t *lck, 2106 kmp_int32 gtid) { 2107 if (__kmp_should_speculate(lck, gtid)) { 2108 if (__kmp_is_unlocked_queuing_lock(GET_QLK_PTR(lck))) { 2109 if (__kmp_test_adaptive_lock_only(lck, gtid)) 2110 return; 2111 // We tried speculation and failed, so give up. 2112 } else { 2113 // We can't try speculation until the lock is free, so we pause here 2114 // (without suspending on the queueing lock, to allow it to drain, then 2115 // try again. All other threads will also see the same result for 2116 // shouldSpeculate, so will be doing the same if they try to claim the 2117 // lock from now on. 2118 while (!__kmp_is_unlocked_queuing_lock(GET_QLK_PTR(lck))) { 2119 KMP_INC_STAT(lck, lemmingYields); 2120 KMP_YIELD(TRUE); 2121 } 2122 2123 if (__kmp_test_adaptive_lock_only(lck, gtid)) 2124 return; 2125 } 2126 } 2127 2128 // Speculative acquisition failed, so acquire it non-speculatively. 2129 // Count the non-speculative acquire attempt 2130 lck->lk.adaptive.acquire_attempts++; 2131 2132 __kmp_acquire_queuing_lock_timed_template<FALSE>(GET_QLK_PTR(lck), gtid); 2133 // We have acquired the base lock, so count that. 2134 KMP_INC_STAT(lck, nonSpeculativeAcquires); 2135 ANNOTATE_QUEUING_ACQUIRED(lck); 2136} 2137 2138static void __kmp_acquire_adaptive_lock_with_checks(kmp_adaptive_lock_t *lck, 2139 kmp_int32 gtid) { 2140 char const *const func = "omp_set_lock"; 2141 if (lck->lk.qlk.initialized != GET_QLK_PTR(lck)) { 2142 KMP_FATAL(LockIsUninitialized, func); 2143 } 2144 if (__kmp_get_queuing_lock_owner(GET_QLK_PTR(lck)) == gtid) { 2145 KMP_FATAL(LockIsAlreadyOwned, func); 2146 } 2147 2148 __kmp_acquire_adaptive_lock(lck, gtid); 2149 2150 lck->lk.qlk.owner_id = gtid + 1; 2151} 2152 2153static int __kmp_release_adaptive_lock(kmp_adaptive_lock_t *lck, 2154 kmp_int32 gtid) { 2155 if (__kmp_is_unlocked_queuing_lock(GET_QLK_PTR( 2156 lck))) { // If the lock doesn't look claimed we must be speculating. 2157 // (Or the user's code is buggy and they're releasing without locking; 2158 // if we had XTEST we'd be able to check that case...) 2159 _xend(); // Exit speculation 2160 __kmp_update_badness_after_success(lck); 2161 } else { // Since the lock *is* visibly locked we're not speculating, 2162 // so should use the underlying lock's release scheme. 2163 __kmp_release_queuing_lock(GET_QLK_PTR(lck), gtid); 2164 } 2165 return KMP_LOCK_RELEASED; 2166} 2167 2168static int __kmp_release_adaptive_lock_with_checks(kmp_adaptive_lock_t *lck, 2169 kmp_int32 gtid) { 2170 char const *const func = "omp_unset_lock"; 2171 KMP_MB(); /* in case another processor initialized lock */ 2172 if (lck->lk.qlk.initialized != GET_QLK_PTR(lck)) { 2173 KMP_FATAL(LockIsUninitialized, func); 2174 } 2175 if (__kmp_get_queuing_lock_owner(GET_QLK_PTR(lck)) == -1) { 2176 KMP_FATAL(LockUnsettingFree, func); 2177 } 2178 if (__kmp_get_queuing_lock_owner(GET_QLK_PTR(lck)) != gtid) { 2179 KMP_FATAL(LockUnsettingSetByAnother, func); 2180 } 2181 lck->lk.qlk.owner_id = 0; 2182 __kmp_release_adaptive_lock(lck, gtid); 2183 return KMP_LOCK_RELEASED; 2184} 2185 2186static void __kmp_init_adaptive_lock(kmp_adaptive_lock_t *lck) { 2187 __kmp_init_queuing_lock(GET_QLK_PTR(lck)); 2188 lck->lk.adaptive.badness = 0; 2189 lck->lk.adaptive.acquire_attempts = 0; // nonSpeculativeAcquireAttempts = 0; 2190 lck->lk.adaptive.max_soft_retries = 2191 __kmp_adaptive_backoff_params.max_soft_retries; 2192 lck->lk.adaptive.max_badness = __kmp_adaptive_backoff_params.max_badness; 2193#if KMP_DEBUG_ADAPTIVE_LOCKS 2194 __kmp_zero_speculative_stats(&lck->lk.adaptive); 2195#endif 2196 KA_TRACE(1000, ("__kmp_init_adaptive_lock: lock %p initialized\n", lck)); 2197} 2198 2199static void __kmp_destroy_adaptive_lock(kmp_adaptive_lock_t *lck) { 2200#if KMP_DEBUG_ADAPTIVE_LOCKS 2201 __kmp_accumulate_speculative_stats(&lck->lk.adaptive); 2202#endif 2203 __kmp_destroy_queuing_lock(GET_QLK_PTR(lck)); 2204 // Nothing needed for the speculative part. 2205} 2206 2207static void __kmp_destroy_adaptive_lock_with_checks(kmp_adaptive_lock_t *lck) { 2208 char const *const func = "omp_destroy_lock"; 2209 if (lck->lk.qlk.initialized != GET_QLK_PTR(lck)) { 2210 KMP_FATAL(LockIsUninitialized, func); 2211 } 2212 if (__kmp_get_queuing_lock_owner(GET_QLK_PTR(lck)) != -1) { 2213 KMP_FATAL(LockStillOwned, func); 2214 } 2215 __kmp_destroy_adaptive_lock(lck); 2216} 2217 2218#endif // KMP_USE_ADAPTIVE_LOCKS 2219 2220/* ------------------------------------------------------------------------ */ 2221/* DRDPA ticket locks */ 2222/* "DRDPA" means Dynamically Reconfigurable Distributed Polling Area */ 2223 2224static kmp_int32 __kmp_get_drdpa_lock_owner(kmp_drdpa_lock_t *lck) { 2225 return lck->lk.owner_id - 1; 2226} 2227 2228static inline bool __kmp_is_drdpa_lock_nestable(kmp_drdpa_lock_t *lck) { 2229 return lck->lk.depth_locked != -1; 2230} 2231 2232__forceinline static int 2233__kmp_acquire_drdpa_lock_timed_template(kmp_drdpa_lock_t *lck, kmp_int32 gtid) { 2234 kmp_uint64 ticket = KMP_ATOMIC_INC(&lck->lk.next_ticket); 2235 kmp_uint64 mask = lck->lk.mask; // atomic load 2236 std::atomic<kmp_uint64> *polls = lck->lk.polls; 2237 2238#ifdef USE_LOCK_PROFILE 2239 if (polls[ticket & mask] != ticket) 2240 __kmp_printf("LOCK CONTENTION: %p\n", lck); 2241/* else __kmp_printf( "." );*/ 2242#endif /* USE_LOCK_PROFILE */ 2243 2244 // Now spin-wait, but reload the polls pointer and mask, in case the 2245 // polling area has been reconfigured. Unless it is reconfigured, the 2246 // reloads stay in L1 cache and are cheap. 2247 // 2248 // Keep this code in sync with KMP_WAIT, in kmp_dispatch.cpp !!! 2249 // The current implementation of KMP_WAIT doesn't allow for mask 2250 // and poll to be re-read every spin iteration. 2251 kmp_uint32 spins; 2252 KMP_FSYNC_PREPARE(lck); 2253 KMP_INIT_YIELD(spins); 2254 while (polls[ticket & mask] < ticket) { // atomic load 2255 KMP_YIELD_OVERSUB_ELSE_SPIN(spins); 2256 // Re-read the mask and the poll pointer from the lock structure. 2257 // 2258 // Make certain that "mask" is read before "polls" !!! 2259 // 2260 // If another thread picks reconfigures the polling area and updates their 2261 // values, and we get the new value of mask and the old polls pointer, we 2262 // could access memory beyond the end of the old polling area. 2263 mask = lck->lk.mask; // atomic load 2264 polls = lck->lk.polls; // atomic load 2265 } 2266 2267 // Critical section starts here 2268 KMP_FSYNC_ACQUIRED(lck); 2269 KA_TRACE(1000, ("__kmp_acquire_drdpa_lock: ticket #%lld acquired lock %p\n", 2270 ticket, lck)); 2271 lck->lk.now_serving = ticket; // non-volatile store 2272 2273 // Deallocate a garbage polling area if we know that we are the last 2274 // thread that could possibly access it. 2275 // 2276 // The >= check is in case __kmp_test_drdpa_lock() allocated the cleanup 2277 // ticket. 2278 if ((lck->lk.old_polls != NULL) && (ticket >= lck->lk.cleanup_ticket)) { 2279 __kmp_free(lck->lk.old_polls); 2280 lck->lk.old_polls = NULL; 2281 lck->lk.cleanup_ticket = 0; 2282 } 2283 2284 // Check to see if we should reconfigure the polling area. 2285 // If there is still a garbage polling area to be deallocated from a 2286 // previous reconfiguration, let a later thread reconfigure it. 2287 if (lck->lk.old_polls == NULL) { 2288 bool reconfigure = false; 2289 std::atomic<kmp_uint64> *old_polls = polls; 2290 kmp_uint32 num_polls = TCR_4(lck->lk.num_polls); 2291 2292 if (TCR_4(__kmp_nth) > 2293 (__kmp_avail_proc ? __kmp_avail_proc : __kmp_xproc)) { 2294 // We are in oversubscription mode. Contract the polling area 2295 // down to a single location, if that hasn't been done already. 2296 if (num_polls > 1) { 2297 reconfigure = true; 2298 num_polls = TCR_4(lck->lk.num_polls); 2299 mask = 0; 2300 num_polls = 1; 2301 polls = (std::atomic<kmp_uint64> *)__kmp_allocate(num_polls * 2302 sizeof(*polls)); 2303 polls[0] = ticket; 2304 } 2305 } else { 2306 // We are in under/fully subscribed mode. Check the number of 2307 // threads waiting on the lock. The size of the polling area 2308 // should be at least the number of threads waiting. 2309 kmp_uint64 num_waiting = TCR_8(lck->lk.next_ticket) - ticket - 1; 2310 if (num_waiting > num_polls) { 2311 kmp_uint32 old_num_polls = num_polls; 2312 reconfigure = true; 2313 do { 2314 mask = (mask << 1) | 1; 2315 num_polls *= 2; 2316 } while (num_polls <= num_waiting); 2317 2318 // Allocate the new polling area, and copy the relevant portion 2319 // of the old polling area to the new area. __kmp_allocate() 2320 // zeroes the memory it allocates, and most of the old area is 2321 // just zero padding, so we only copy the release counters. 2322 polls = (std::atomic<kmp_uint64> *)__kmp_allocate(num_polls * 2323 sizeof(*polls)); 2324 kmp_uint32 i; 2325 for (i = 0; i < old_num_polls; i++) { 2326 polls[i].store(old_polls[i]); 2327 } 2328 } 2329 } 2330 2331 if (reconfigure) { 2332 // Now write the updated fields back to the lock structure. 2333 // 2334 // Make certain that "polls" is written before "mask" !!! 2335 // 2336 // If another thread picks up the new value of mask and the old polls 2337 // pointer , it could access memory beyond the end of the old polling 2338 // area. 2339 // 2340 // On x86, we need memory fences. 2341 KA_TRACE(1000, ("__kmp_acquire_drdpa_lock: ticket #%lld reconfiguring " 2342 "lock %p to %d polls\n", 2343 ticket, lck, num_polls)); 2344 2345 lck->lk.old_polls = old_polls; 2346 lck->lk.polls = polls; // atomic store 2347 2348 KMP_MB(); 2349 2350 lck->lk.num_polls = num_polls; 2351 lck->lk.mask = mask; // atomic store 2352 2353 KMP_MB(); 2354 2355 // Only after the new polling area and mask have been flushed 2356 // to main memory can we update the cleanup ticket field. 2357 // 2358 // volatile load / non-volatile store 2359 lck->lk.cleanup_ticket = lck->lk.next_ticket; 2360 } 2361 } 2362 return KMP_LOCK_ACQUIRED_FIRST; 2363} 2364 2365int __kmp_acquire_drdpa_lock(kmp_drdpa_lock_t *lck, kmp_int32 gtid) { 2366 int retval = __kmp_acquire_drdpa_lock_timed_template(lck, gtid); 2367 ANNOTATE_DRDPA_ACQUIRED(lck); 2368 return retval; 2369} 2370 2371static int __kmp_acquire_drdpa_lock_with_checks(kmp_drdpa_lock_t *lck, 2372 kmp_int32 gtid) { 2373 char const *const func = "omp_set_lock"; 2374 if (lck->lk.initialized != lck) { 2375 KMP_FATAL(LockIsUninitialized, func); 2376 } 2377 if (__kmp_is_drdpa_lock_nestable(lck)) { 2378 KMP_FATAL(LockNestableUsedAsSimple, func); 2379 } 2380 if ((gtid >= 0) && (__kmp_get_drdpa_lock_owner(lck) == gtid)) { 2381 KMP_FATAL(LockIsAlreadyOwned, func); 2382 } 2383 2384 __kmp_acquire_drdpa_lock(lck, gtid); 2385 2386 lck->lk.owner_id = gtid + 1; 2387 return KMP_LOCK_ACQUIRED_FIRST; 2388} 2389 2390int __kmp_test_drdpa_lock(kmp_drdpa_lock_t *lck, kmp_int32 gtid) { 2391 // First get a ticket, then read the polls pointer and the mask. 2392 // The polls pointer must be read before the mask!!! (See above) 2393 kmp_uint64 ticket = lck->lk.next_ticket; // atomic load 2394 std::atomic<kmp_uint64> *polls = lck->lk.polls; 2395 kmp_uint64 mask = lck->lk.mask; // atomic load 2396 if (polls[ticket & mask] == ticket) { 2397 kmp_uint64 next_ticket = ticket + 1; 2398 if (__kmp_atomic_compare_store_acq(&lck->lk.next_ticket, ticket, 2399 next_ticket)) { 2400 KMP_FSYNC_ACQUIRED(lck); 2401 KA_TRACE(1000, ("__kmp_test_drdpa_lock: ticket #%lld acquired lock %p\n", 2402 ticket, lck)); 2403 lck->lk.now_serving = ticket; // non-volatile store 2404 2405 // Since no threads are waiting, there is no possibility that we would 2406 // want to reconfigure the polling area. We might have the cleanup ticket 2407 // value (which says that it is now safe to deallocate old_polls), but 2408 // we'll let a later thread which calls __kmp_acquire_lock do that - this 2409 // routine isn't supposed to block, and we would risk blocks if we called 2410 // __kmp_free() to do the deallocation. 2411 return TRUE; 2412 } 2413 } 2414 return FALSE; 2415} 2416 2417static int __kmp_test_drdpa_lock_with_checks(kmp_drdpa_lock_t *lck, 2418 kmp_int32 gtid) { 2419 char const *const func = "omp_test_lock"; 2420 if (lck->lk.initialized != lck) { 2421 KMP_FATAL(LockIsUninitialized, func); 2422 } 2423 if (__kmp_is_drdpa_lock_nestable(lck)) { 2424 KMP_FATAL(LockNestableUsedAsSimple, func); 2425 } 2426 2427 int retval = __kmp_test_drdpa_lock(lck, gtid); 2428 2429 if (retval) { 2430 lck->lk.owner_id = gtid + 1; 2431 } 2432 return retval; 2433} 2434 2435int __kmp_release_drdpa_lock(kmp_drdpa_lock_t *lck, kmp_int32 gtid) { 2436 // Read the ticket value from the lock data struct, then the polls pointer and 2437 // the mask. The polls pointer must be read before the mask!!! (See above) 2438 kmp_uint64 ticket = lck->lk.now_serving + 1; // non-atomic load 2439 std::atomic<kmp_uint64> *polls = lck->lk.polls; // atomic load 2440 kmp_uint64 mask = lck->lk.mask; // atomic load 2441 KA_TRACE(1000, ("__kmp_release_drdpa_lock: ticket #%lld released lock %p\n", 2442 ticket - 1, lck)); 2443 KMP_FSYNC_RELEASING(lck); 2444 ANNOTATE_DRDPA_RELEASED(lck); 2445 polls[ticket & mask] = ticket; // atomic store 2446 return KMP_LOCK_RELEASED; 2447} 2448 2449static int __kmp_release_drdpa_lock_with_checks(kmp_drdpa_lock_t *lck, 2450 kmp_int32 gtid) { 2451 char const *const func = "omp_unset_lock"; 2452 KMP_MB(); /* in case another processor initialized lock */ 2453 if (lck->lk.initialized != lck) { 2454 KMP_FATAL(LockIsUninitialized, func); 2455 } 2456 if (__kmp_is_drdpa_lock_nestable(lck)) { 2457 KMP_FATAL(LockNestableUsedAsSimple, func); 2458 } 2459 if (__kmp_get_drdpa_lock_owner(lck) == -1) { 2460 KMP_FATAL(LockUnsettingFree, func); 2461 } 2462 if ((gtid >= 0) && (__kmp_get_drdpa_lock_owner(lck) >= 0) && 2463 (__kmp_get_drdpa_lock_owner(lck) != gtid)) { 2464 KMP_FATAL(LockUnsettingSetByAnother, func); 2465 } 2466 lck->lk.owner_id = 0; 2467 return __kmp_release_drdpa_lock(lck, gtid); 2468} 2469 2470void __kmp_init_drdpa_lock(kmp_drdpa_lock_t *lck) { 2471 lck->lk.location = NULL; 2472 lck->lk.mask = 0; 2473 lck->lk.num_polls = 1; 2474 lck->lk.polls = (std::atomic<kmp_uint64> *)__kmp_allocate( 2475 lck->lk.num_polls * sizeof(*(lck->lk.polls))); 2476 lck->lk.cleanup_ticket = 0; 2477 lck->lk.old_polls = NULL; 2478 lck->lk.next_ticket = 0; 2479 lck->lk.now_serving = 0; 2480 lck->lk.owner_id = 0; // no thread owns the lock. 2481 lck->lk.depth_locked = -1; // >= 0 for nestable locks, -1 for simple locks. 2482 lck->lk.initialized = lck; 2483 2484 KA_TRACE(1000, ("__kmp_init_drdpa_lock: lock %p initialized\n", lck)); 2485} 2486 2487void __kmp_destroy_drdpa_lock(kmp_drdpa_lock_t *lck) { 2488 lck->lk.initialized = NULL; 2489 lck->lk.location = NULL; 2490 if (lck->lk.polls.load() != NULL) { 2491 __kmp_free(lck->lk.polls.load()); 2492 lck->lk.polls = NULL; 2493 } 2494 if (lck->lk.old_polls != NULL) { 2495 __kmp_free(lck->lk.old_polls); 2496 lck->lk.old_polls = NULL; 2497 } 2498 lck->lk.mask = 0; 2499 lck->lk.num_polls = 0; 2500 lck->lk.cleanup_ticket = 0; 2501 lck->lk.next_ticket = 0; 2502 lck->lk.now_serving = 0; 2503 lck->lk.owner_id = 0; 2504 lck->lk.depth_locked = -1; 2505} 2506 2507static void __kmp_destroy_drdpa_lock_with_checks(kmp_drdpa_lock_t *lck) { 2508 char const *const func = "omp_destroy_lock"; 2509 if (lck->lk.initialized != lck) { 2510 KMP_FATAL(LockIsUninitialized, func); 2511 } 2512 if (__kmp_is_drdpa_lock_nestable(lck)) { 2513 KMP_FATAL(LockNestableUsedAsSimple, func); 2514 } 2515 if (__kmp_get_drdpa_lock_owner(lck) != -1) { 2516 KMP_FATAL(LockStillOwned, func); 2517 } 2518 __kmp_destroy_drdpa_lock(lck); 2519} 2520 2521// nested drdpa ticket locks 2522 2523int __kmp_acquire_nested_drdpa_lock(kmp_drdpa_lock_t *lck, kmp_int32 gtid) { 2524 KMP_DEBUG_ASSERT(gtid >= 0); 2525 2526 if (__kmp_get_drdpa_lock_owner(lck) == gtid) { 2527 lck->lk.depth_locked += 1; 2528 return KMP_LOCK_ACQUIRED_NEXT; 2529 } else { 2530 __kmp_acquire_drdpa_lock_timed_template(lck, gtid); 2531 ANNOTATE_DRDPA_ACQUIRED(lck); 2532 KMP_MB(); 2533 lck->lk.depth_locked = 1; 2534 KMP_MB(); 2535 lck->lk.owner_id = gtid + 1; 2536 return KMP_LOCK_ACQUIRED_FIRST; 2537 } 2538} 2539 2540static void __kmp_acquire_nested_drdpa_lock_with_checks(kmp_drdpa_lock_t *lck, 2541 kmp_int32 gtid) { 2542 char const *const func = "omp_set_nest_lock"; 2543 if (lck->lk.initialized != lck) { 2544 KMP_FATAL(LockIsUninitialized, func); 2545 } 2546 if (!__kmp_is_drdpa_lock_nestable(lck)) { 2547 KMP_FATAL(LockSimpleUsedAsNestable, func); 2548 } 2549 __kmp_acquire_nested_drdpa_lock(lck, gtid); 2550} 2551 2552int __kmp_test_nested_drdpa_lock(kmp_drdpa_lock_t *lck, kmp_int32 gtid) { 2553 int retval; 2554 2555 KMP_DEBUG_ASSERT(gtid >= 0); 2556 2557 if (__kmp_get_drdpa_lock_owner(lck) == gtid) { 2558 retval = ++lck->lk.depth_locked; 2559 } else if (!__kmp_test_drdpa_lock(lck, gtid)) { 2560 retval = 0; 2561 } else { 2562 KMP_MB(); 2563 retval = lck->lk.depth_locked = 1; 2564 KMP_MB(); 2565 lck->lk.owner_id = gtid + 1; 2566 } 2567 return retval; 2568} 2569 2570static int __kmp_test_nested_drdpa_lock_with_checks(kmp_drdpa_lock_t *lck, 2571 kmp_int32 gtid) { 2572 char const *const func = "omp_test_nest_lock"; 2573 if (lck->lk.initialized != lck) { 2574 KMP_FATAL(LockIsUninitialized, func); 2575 } 2576 if (!__kmp_is_drdpa_lock_nestable(lck)) { 2577 KMP_FATAL(LockSimpleUsedAsNestable, func); 2578 } 2579 return __kmp_test_nested_drdpa_lock(lck, gtid); 2580} 2581 2582int __kmp_release_nested_drdpa_lock(kmp_drdpa_lock_t *lck, kmp_int32 gtid) { 2583 KMP_DEBUG_ASSERT(gtid >= 0); 2584 2585 KMP_MB(); 2586 if (--(lck->lk.depth_locked) == 0) { 2587 KMP_MB(); 2588 lck->lk.owner_id = 0; 2589 __kmp_release_drdpa_lock(lck, gtid); 2590 return KMP_LOCK_RELEASED; 2591 } 2592 return KMP_LOCK_STILL_HELD; 2593} 2594 2595static int __kmp_release_nested_drdpa_lock_with_checks(kmp_drdpa_lock_t *lck, 2596 kmp_int32 gtid) { 2597 char const *const func = "omp_unset_nest_lock"; 2598 KMP_MB(); /* in case another processor initialized lock */ 2599 if (lck->lk.initialized != lck) { 2600 KMP_FATAL(LockIsUninitialized, func); 2601 } 2602 if (!__kmp_is_drdpa_lock_nestable(lck)) { 2603 KMP_FATAL(LockSimpleUsedAsNestable, func); 2604 } 2605 if (__kmp_get_drdpa_lock_owner(lck) == -1) { 2606 KMP_FATAL(LockUnsettingFree, func); 2607 } 2608 if (__kmp_get_drdpa_lock_owner(lck) != gtid) { 2609 KMP_FATAL(LockUnsettingSetByAnother, func); 2610 } 2611 return __kmp_release_nested_drdpa_lock(lck, gtid); 2612} 2613 2614void __kmp_init_nested_drdpa_lock(kmp_drdpa_lock_t *lck) { 2615 __kmp_init_drdpa_lock(lck); 2616 lck->lk.depth_locked = 0; // >= 0 for nestable locks, -1 for simple locks 2617} 2618 2619void __kmp_destroy_nested_drdpa_lock(kmp_drdpa_lock_t *lck) { 2620 __kmp_destroy_drdpa_lock(lck); 2621 lck->lk.depth_locked = 0; 2622} 2623 2624static void __kmp_destroy_nested_drdpa_lock_with_checks(kmp_drdpa_lock_t *lck) { 2625 char const *const func = "omp_destroy_nest_lock"; 2626 if (lck->lk.initialized != lck) { 2627 KMP_FATAL(LockIsUninitialized, func); 2628 } 2629 if (!__kmp_is_drdpa_lock_nestable(lck)) { 2630 KMP_FATAL(LockSimpleUsedAsNestable, func); 2631 } 2632 if (__kmp_get_drdpa_lock_owner(lck) != -1) { 2633 KMP_FATAL(LockStillOwned, func); 2634 } 2635 __kmp_destroy_nested_drdpa_lock(lck); 2636} 2637 2638// access functions to fields which don't exist for all lock kinds. 2639 2640static const ident_t *__kmp_get_drdpa_lock_location(kmp_drdpa_lock_t *lck) { 2641 return lck->lk.location; 2642} 2643 2644static void __kmp_set_drdpa_lock_location(kmp_drdpa_lock_t *lck, 2645 const ident_t *loc) { 2646 lck->lk.location = loc; 2647} 2648 2649static kmp_lock_flags_t __kmp_get_drdpa_lock_flags(kmp_drdpa_lock_t *lck) { 2650 return lck->lk.flags; 2651} 2652 2653static void __kmp_set_drdpa_lock_flags(kmp_drdpa_lock_t *lck, 2654 kmp_lock_flags_t flags) { 2655 lck->lk.flags = flags; 2656} 2657 2658// Time stamp counter 2659#if KMP_ARCH_X86 || KMP_ARCH_X86_64 2660#define __kmp_tsc() __kmp_hardware_timestamp() 2661// Runtime's default backoff parameters 2662kmp_backoff_t __kmp_spin_backoff_params = {1, 4096, 100}; 2663#else 2664// Use nanoseconds for other platforms 2665extern kmp_uint64 __kmp_now_nsec(); 2666kmp_backoff_t __kmp_spin_backoff_params = {1, 256, 100}; 2667#define __kmp_tsc() __kmp_now_nsec() 2668#endif 2669 2670// A useful predicate for dealing with timestamps that may wrap. 2671// Is a before b? Since the timestamps may wrap, this is asking whether it's 2672// shorter to go clockwise from a to b around the clock-face, or anti-clockwise. 2673// Times where going clockwise is less distance than going anti-clockwise 2674// are in the future, others are in the past. e.g. a = MAX-1, b = MAX+1 (=0), 2675// then a > b (true) does not mean a reached b; whereas signed(a) = -2, 2676// signed(b) = 0 captures the actual difference 2677static inline bool before(kmp_uint64 a, kmp_uint64 b) { 2678 return ((kmp_int64)b - (kmp_int64)a) > 0; 2679} 2680 2681// Truncated binary exponential backoff function 2682void __kmp_spin_backoff(kmp_backoff_t *boff) { 2683 // We could flatten this loop, but making it a nested loop gives better result 2684 kmp_uint32 i; 2685 for (i = boff->step; i > 0; i--) { 2686 kmp_uint64 goal = __kmp_tsc() + boff->min_tick; 2687 do { 2688 KMP_CPU_PAUSE(); 2689 } while (before(__kmp_tsc(), goal)); 2690 } 2691 boff->step = (boff->step << 1 | 1) & (boff->max_backoff - 1); 2692} 2693 2694#if KMP_USE_DYNAMIC_LOCK 2695 2696// Direct lock initializers. It simply writes a tag to the low 8 bits of the 2697// lock word. 2698static void __kmp_init_direct_lock(kmp_dyna_lock_t *lck, 2699 kmp_dyna_lockseq_t seq) { 2700 TCW_4(*lck, KMP_GET_D_TAG(seq)); 2701 KA_TRACE( 2702 20, 2703 ("__kmp_init_direct_lock: initialized direct lock with type#%d\n", seq)); 2704} 2705 2706#if KMP_USE_TSX 2707 2708// HLE lock functions - imported from the testbed runtime. 2709#define HLE_ACQUIRE ".byte 0xf2;" 2710#define HLE_RELEASE ".byte 0xf3;" 2711 2712static inline kmp_uint32 swap4(kmp_uint32 volatile *p, kmp_uint32 v) { 2713 __asm__ volatile(HLE_ACQUIRE "xchg %1,%0" : "+r"(v), "+m"(*p) : : "memory"); 2714 return v; 2715} 2716 2717static void __kmp_destroy_hle_lock(kmp_dyna_lock_t *lck) { TCW_4(*lck, 0); } 2718 2719static void __kmp_destroy_hle_lock_with_checks(kmp_dyna_lock_t *lck) { 2720 TCW_4(*lck, 0); 2721} 2722 2723static void __kmp_acquire_hle_lock(kmp_dyna_lock_t *lck, kmp_int32 gtid) { 2724 // Use gtid for KMP_LOCK_BUSY if necessary 2725 if (swap4(lck, KMP_LOCK_BUSY(1, hle)) != KMP_LOCK_FREE(hle)) { 2726 int delay = 1; 2727 do { 2728 while (*(kmp_uint32 volatile *)lck != KMP_LOCK_FREE(hle)) { 2729 for (int i = delay; i != 0; --i) 2730 KMP_CPU_PAUSE(); 2731 delay = ((delay << 1) | 1) & 7; 2732 } 2733 } while (swap4(lck, KMP_LOCK_BUSY(1, hle)) != KMP_LOCK_FREE(hle)); 2734 } 2735} 2736 2737static void __kmp_acquire_hle_lock_with_checks(kmp_dyna_lock_t *lck, 2738 kmp_int32 gtid) { 2739 __kmp_acquire_hle_lock(lck, gtid); // TODO: add checks 2740} 2741 2742static int __kmp_release_hle_lock(kmp_dyna_lock_t *lck, kmp_int32 gtid) { 2743 __asm__ volatile(HLE_RELEASE "movl %1,%0" 2744 : "=m"(*lck) 2745 : "r"(KMP_LOCK_FREE(hle)) 2746 : "memory"); 2747 return KMP_LOCK_RELEASED; 2748} 2749 2750static int __kmp_release_hle_lock_with_checks(kmp_dyna_lock_t *lck, 2751 kmp_int32 gtid) { 2752 return __kmp_release_hle_lock(lck, gtid); // TODO: add checks 2753} 2754 2755static int __kmp_test_hle_lock(kmp_dyna_lock_t *lck, kmp_int32 gtid) { 2756 return swap4(lck, KMP_LOCK_BUSY(1, hle)) == KMP_LOCK_FREE(hle); 2757} 2758 2759static int __kmp_test_hle_lock_with_checks(kmp_dyna_lock_t *lck, 2760 kmp_int32 gtid) { 2761 return __kmp_test_hle_lock(lck, gtid); // TODO: add checks 2762} 2763 2764static void __kmp_init_rtm_lock(kmp_queuing_lock_t *lck) { 2765 __kmp_init_queuing_lock(lck); 2766} 2767 2768static void __kmp_destroy_rtm_lock(kmp_queuing_lock_t *lck) { 2769 __kmp_destroy_queuing_lock(lck); 2770} 2771 2772static void __kmp_destroy_rtm_lock_with_checks(kmp_queuing_lock_t *lck) { 2773 __kmp_destroy_queuing_lock_with_checks(lck); 2774} 2775 2776static void __kmp_acquire_rtm_lock(kmp_queuing_lock_t *lck, kmp_int32 gtid) { 2777 unsigned retries = 3, status; 2778 do { 2779 status = _xbegin(); 2780 if (status == _XBEGIN_STARTED) { 2781 if (__kmp_is_unlocked_queuing_lock(lck)) 2782 return; 2783 _xabort(0xff); 2784 } 2785 if ((status & _XABORT_EXPLICIT) && _XABORT_CODE(status) == 0xff) { 2786 // Wait until lock becomes free 2787 while (!__kmp_is_unlocked_queuing_lock(lck)) { 2788 KMP_YIELD(TRUE); 2789 } 2790 } else if (!(status & _XABORT_RETRY)) 2791 break; 2792 } while (retries--); 2793 2794 // Fall-back non-speculative lock (xchg) 2795 __kmp_acquire_queuing_lock(lck, gtid); 2796} 2797 2798static void __kmp_acquire_rtm_lock_with_checks(kmp_queuing_lock_t *lck, 2799 kmp_int32 gtid) { 2800 __kmp_acquire_rtm_lock(lck, gtid); 2801} 2802 2803static int __kmp_release_rtm_lock(kmp_queuing_lock_t *lck, kmp_int32 gtid) { 2804 if (__kmp_is_unlocked_queuing_lock(lck)) { 2805 // Releasing from speculation 2806 _xend(); 2807 } else { 2808 // Releasing from a real lock 2809 __kmp_release_queuing_lock(lck, gtid); 2810 } 2811 return KMP_LOCK_RELEASED; 2812} 2813 2814static int __kmp_release_rtm_lock_with_checks(kmp_queuing_lock_t *lck, 2815 kmp_int32 gtid) { 2816 return __kmp_release_rtm_lock(lck, gtid); 2817} 2818 2819static int __kmp_test_rtm_lock(kmp_queuing_lock_t *lck, kmp_int32 gtid) { 2820 unsigned retries = 3, status; 2821 do { 2822 status = _xbegin(); 2823 if (status == _XBEGIN_STARTED && __kmp_is_unlocked_queuing_lock(lck)) { 2824 return 1; 2825 } 2826 if (!(status & _XABORT_RETRY)) 2827 break; 2828 } while (retries--); 2829 2830 return (__kmp_is_unlocked_queuing_lock(lck)) ? 1 : 0; 2831} 2832 2833static int __kmp_test_rtm_lock_with_checks(kmp_queuing_lock_t *lck, 2834 kmp_int32 gtid) { 2835 return __kmp_test_rtm_lock(lck, gtid); 2836} 2837 2838#endif // KMP_USE_TSX 2839 2840// Entry functions for indirect locks (first element of direct lock jump tables) 2841static void __kmp_init_indirect_lock(kmp_dyna_lock_t *l, 2842 kmp_dyna_lockseq_t tag); 2843static void __kmp_destroy_indirect_lock(kmp_dyna_lock_t *lock); 2844static int __kmp_set_indirect_lock(kmp_dyna_lock_t *lock, kmp_int32); 2845static int __kmp_unset_indirect_lock(kmp_dyna_lock_t *lock, kmp_int32); 2846static int __kmp_test_indirect_lock(kmp_dyna_lock_t *lock, kmp_int32); 2847static int __kmp_set_indirect_lock_with_checks(kmp_dyna_lock_t *lock, 2848 kmp_int32); 2849static int __kmp_unset_indirect_lock_with_checks(kmp_dyna_lock_t *lock, 2850 kmp_int32); 2851static int __kmp_test_indirect_lock_with_checks(kmp_dyna_lock_t *lock, 2852 kmp_int32); 2853 2854// Lock function definitions for the union parameter type 2855#define KMP_FOREACH_LOCK_KIND(m, a) m(ticket, a) m(queuing, a) m(drdpa, a) 2856 2857#define expand1(lk, op) \ 2858 static void __kmp_##op##_##lk##_##lock(kmp_user_lock_p lock) { \ 2859 __kmp_##op##_##lk##_##lock(&lock->lk); \ 2860 } 2861#define expand2(lk, op) \ 2862 static int __kmp_##op##_##lk##_##lock(kmp_user_lock_p lock, \ 2863 kmp_int32 gtid) { \ 2864 return __kmp_##op##_##lk##_##lock(&lock->lk, gtid); \ 2865 } 2866#define expand3(lk, op) \ 2867 static void __kmp_set_##lk##_##lock_flags(kmp_user_lock_p lock, \ 2868 kmp_lock_flags_t flags) { \ 2869 __kmp_set_##lk##_lock_flags(&lock->lk, flags); \ 2870 } 2871#define expand4(lk, op) \ 2872 static void __kmp_set_##lk##_##lock_location(kmp_user_lock_p lock, \ 2873 const ident_t *loc) { \ 2874 __kmp_set_##lk##_lock_location(&lock->lk, loc); \ 2875 } 2876 2877KMP_FOREACH_LOCK_KIND(expand1, init) 2878KMP_FOREACH_LOCK_KIND(expand1, init_nested) 2879KMP_FOREACH_LOCK_KIND(expand1, destroy) 2880KMP_FOREACH_LOCK_KIND(expand1, destroy_nested) 2881KMP_FOREACH_LOCK_KIND(expand2, acquire) 2882KMP_FOREACH_LOCK_KIND(expand2, acquire_nested) 2883KMP_FOREACH_LOCK_KIND(expand2, release) 2884KMP_FOREACH_LOCK_KIND(expand2, release_nested) 2885KMP_FOREACH_LOCK_KIND(expand2, test) 2886KMP_FOREACH_LOCK_KIND(expand2, test_nested) 2887KMP_FOREACH_LOCK_KIND(expand3, ) 2888KMP_FOREACH_LOCK_KIND(expand4, ) 2889 2890#undef expand1 2891#undef expand2 2892#undef expand3 2893#undef expand4 2894 2895// Jump tables for the indirect lock functions 2896// Only fill in the odd entries, that avoids the need to shift out the low bit 2897 2898// init functions 2899#define expand(l, op) 0, __kmp_init_direct_lock, 2900void (*__kmp_direct_init[])(kmp_dyna_lock_t *, kmp_dyna_lockseq_t) = { 2901 __kmp_init_indirect_lock, 0, KMP_FOREACH_D_LOCK(expand, init)}; 2902#undef expand 2903 2904// destroy functions 2905#define expand(l, op) 0, (void (*)(kmp_dyna_lock_t *))__kmp_##op##_##l##_lock, 2906static void (*direct_destroy[])(kmp_dyna_lock_t *) = { 2907 __kmp_destroy_indirect_lock, 0, KMP_FOREACH_D_LOCK(expand, destroy)}; 2908#undef expand 2909#define expand(l, op) \ 2910 0, (void (*)(kmp_dyna_lock_t *))__kmp_destroy_##l##_lock_with_checks, 2911static void (*direct_destroy_check[])(kmp_dyna_lock_t *) = { 2912 __kmp_destroy_indirect_lock, 0, KMP_FOREACH_D_LOCK(expand, destroy)}; 2913#undef expand 2914 2915// set/acquire functions 2916#define expand(l, op) \ 2917 0, (int (*)(kmp_dyna_lock_t *, kmp_int32))__kmp_##op##_##l##_lock, 2918static int (*direct_set[])(kmp_dyna_lock_t *, kmp_int32) = { 2919 __kmp_set_indirect_lock, 0, KMP_FOREACH_D_LOCK(expand, acquire)}; 2920#undef expand 2921#define expand(l, op) \ 2922 0, (int (*)(kmp_dyna_lock_t *, kmp_int32))__kmp_##op##_##l##_lock_with_checks, 2923static int (*direct_set_check[])(kmp_dyna_lock_t *, kmp_int32) = { 2924 __kmp_set_indirect_lock_with_checks, 0, 2925 KMP_FOREACH_D_LOCK(expand, acquire)}; 2926#undef expand 2927 2928// unset/release and test functions 2929#define expand(l, op) \ 2930 0, (int (*)(kmp_dyna_lock_t *, kmp_int32))__kmp_##op##_##l##_lock, 2931static int (*direct_unset[])(kmp_dyna_lock_t *, kmp_int32) = { 2932 __kmp_unset_indirect_lock, 0, KMP_FOREACH_D_LOCK(expand, release)}; 2933static int (*direct_test[])(kmp_dyna_lock_t *, kmp_int32) = { 2934 __kmp_test_indirect_lock, 0, KMP_FOREACH_D_LOCK(expand, test)}; 2935#undef expand 2936#define expand(l, op) \ 2937 0, (int (*)(kmp_dyna_lock_t *, kmp_int32))__kmp_##op##_##l##_lock_with_checks, 2938static int (*direct_unset_check[])(kmp_dyna_lock_t *, kmp_int32) = { 2939 __kmp_unset_indirect_lock_with_checks, 0, 2940 KMP_FOREACH_D_LOCK(expand, release)}; 2941static int (*direct_test_check[])(kmp_dyna_lock_t *, kmp_int32) = { 2942 __kmp_test_indirect_lock_with_checks, 0, KMP_FOREACH_D_LOCK(expand, test)}; 2943#undef expand 2944 2945// Exposes only one set of jump tables (*lock or *lock_with_checks). 2946void (**__kmp_direct_destroy)(kmp_dyna_lock_t *) = 0; 2947int (**__kmp_direct_set)(kmp_dyna_lock_t *, kmp_int32) = 0; 2948int (**__kmp_direct_unset)(kmp_dyna_lock_t *, kmp_int32) = 0; 2949int (**__kmp_direct_test)(kmp_dyna_lock_t *, kmp_int32) = 0; 2950 2951// Jump tables for the indirect lock functions 2952#define expand(l, op) (void (*)(kmp_user_lock_p)) __kmp_##op##_##l##_##lock, 2953void (*__kmp_indirect_init[])(kmp_user_lock_p) = { 2954 KMP_FOREACH_I_LOCK(expand, init)}; 2955#undef expand 2956 2957#define expand(l, op) (void (*)(kmp_user_lock_p)) __kmp_##op##_##l##_##lock, 2958static void (*indirect_destroy[])(kmp_user_lock_p) = { 2959 KMP_FOREACH_I_LOCK(expand, destroy)}; 2960#undef expand 2961#define expand(l, op) \ 2962 (void (*)(kmp_user_lock_p)) __kmp_##op##_##l##_##lock_with_checks, 2963static void (*indirect_destroy_check[])(kmp_user_lock_p) = { 2964 KMP_FOREACH_I_LOCK(expand, destroy)}; 2965#undef expand 2966 2967// set/acquire functions 2968#define expand(l, op) \ 2969 (int (*)(kmp_user_lock_p, kmp_int32)) __kmp_##op##_##l##_##lock, 2970static int (*indirect_set[])(kmp_user_lock_p, 2971 kmp_int32) = {KMP_FOREACH_I_LOCK(expand, acquire)}; 2972#undef expand 2973#define expand(l, op) \ 2974 (int (*)(kmp_user_lock_p, kmp_int32)) __kmp_##op##_##l##_##lock_with_checks, 2975static int (*indirect_set_check[])(kmp_user_lock_p, kmp_int32) = { 2976 KMP_FOREACH_I_LOCK(expand, acquire)}; 2977#undef expand 2978 2979// unset/release and test functions 2980#define expand(l, op) \ 2981 (int (*)(kmp_user_lock_p, kmp_int32)) __kmp_##op##_##l##_##lock, 2982static int (*indirect_unset[])(kmp_user_lock_p, kmp_int32) = { 2983 KMP_FOREACH_I_LOCK(expand, release)}; 2984static int (*indirect_test[])(kmp_user_lock_p, 2985 kmp_int32) = {KMP_FOREACH_I_LOCK(expand, test)}; 2986#undef expand 2987#define expand(l, op) \ 2988 (int (*)(kmp_user_lock_p, kmp_int32)) __kmp_##op##_##l##_##lock_with_checks, 2989static int (*indirect_unset_check[])(kmp_user_lock_p, kmp_int32) = { 2990 KMP_FOREACH_I_LOCK(expand, release)}; 2991static int (*indirect_test_check[])(kmp_user_lock_p, kmp_int32) = { 2992 KMP_FOREACH_I_LOCK(expand, test)}; 2993#undef expand 2994 2995// Exposes only one jump tables (*lock or *lock_with_checks). 2996void (**__kmp_indirect_destroy)(kmp_user_lock_p) = 0; 2997int (**__kmp_indirect_set)(kmp_user_lock_p, kmp_int32) = 0; 2998int (**__kmp_indirect_unset)(kmp_user_lock_p, kmp_int32) = 0; 2999int (**__kmp_indirect_test)(kmp_user_lock_p, kmp_int32) = 0; 3000 3001// Lock index table. 3002kmp_indirect_lock_table_t __kmp_i_lock_table; 3003 3004// Size of indirect locks. 3005static kmp_uint32 __kmp_indirect_lock_size[KMP_NUM_I_LOCKS] = {0}; 3006 3007// Jump tables for lock accessor/modifier. 3008void (*__kmp_indirect_set_location[KMP_NUM_I_LOCKS])(kmp_user_lock_p, 3009 const ident_t *) = {0}; 3010void (*__kmp_indirect_set_flags[KMP_NUM_I_LOCKS])(kmp_user_lock_p, 3011 kmp_lock_flags_t) = {0}; 3012const ident_t *(*__kmp_indirect_get_location[KMP_NUM_I_LOCKS])( 3013 kmp_user_lock_p) = {0}; 3014kmp_lock_flags_t (*__kmp_indirect_get_flags[KMP_NUM_I_LOCKS])( 3015 kmp_user_lock_p) = {0}; 3016 3017// Use different lock pools for different lock types. 3018static kmp_indirect_lock_t *__kmp_indirect_lock_pool[KMP_NUM_I_LOCKS] = {0}; 3019 3020// User lock allocator for dynamically dispatched indirect locks. Every entry of 3021// the indirect lock table holds the address and type of the allocated indrect 3022// lock (kmp_indirect_lock_t), and the size of the table doubles when it is 3023// full. A destroyed indirect lock object is returned to the reusable pool of 3024// locks, unique to each lock type. 3025kmp_indirect_lock_t *__kmp_allocate_indirect_lock(void **user_lock, 3026 kmp_int32 gtid, 3027 kmp_indirect_locktag_t tag) { 3028 kmp_indirect_lock_t *lck; 3029 kmp_lock_index_t idx; 3030 3031 __kmp_acquire_lock(&__kmp_global_lock, gtid); 3032 3033 if (__kmp_indirect_lock_pool[tag] != NULL) { 3034 // Reuse the allocated and destroyed lock object 3035 lck = __kmp_indirect_lock_pool[tag]; 3036 if (OMP_LOCK_T_SIZE < sizeof(void *)) 3037 idx = lck->lock->pool.index; 3038 __kmp_indirect_lock_pool[tag] = (kmp_indirect_lock_t *)lck->lock->pool.next; 3039 KA_TRACE(20, ("__kmp_allocate_indirect_lock: reusing an existing lock %p\n", 3040 lck)); 3041 } else { 3042 idx = __kmp_i_lock_table.next; 3043 // Check capacity and double the size if it is full 3044 if (idx == __kmp_i_lock_table.size) { 3045 // Double up the space for block pointers 3046 int row = __kmp_i_lock_table.size / KMP_I_LOCK_CHUNK; 3047 kmp_indirect_lock_t **new_table = (kmp_indirect_lock_t **)__kmp_allocate( 3048 2 * row * sizeof(kmp_indirect_lock_t *)); 3049 KMP_MEMCPY(new_table, __kmp_i_lock_table.table, 3050 row * sizeof(kmp_indirect_lock_t *)); 3051 kmp_indirect_lock_t **old_table = __kmp_i_lock_table.table; 3052 __kmp_i_lock_table.table = new_table; 3053 __kmp_free(old_table); 3054 // Allocate new objects in the new blocks 3055 for (int i = row; i < 2 * row; ++i) 3056 *(__kmp_i_lock_table.table + i) = (kmp_indirect_lock_t *)__kmp_allocate( 3057 KMP_I_LOCK_CHUNK * sizeof(kmp_indirect_lock_t)); 3058 __kmp_i_lock_table.size = 2 * idx; 3059 } 3060 __kmp_i_lock_table.next++; 3061 lck = KMP_GET_I_LOCK(idx); 3062 // Allocate a new base lock object 3063 lck->lock = (kmp_user_lock_p)__kmp_allocate(__kmp_indirect_lock_size[tag]); 3064 KA_TRACE(20, 3065 ("__kmp_allocate_indirect_lock: allocated a new lock %p\n", lck)); 3066 } 3067 3068 __kmp_release_lock(&__kmp_global_lock, gtid); 3069 3070 lck->type = tag; 3071 3072 if (OMP_LOCK_T_SIZE < sizeof(void *)) { 3073 *((kmp_lock_index_t *)user_lock) = idx 3074 << 1; // indirect lock word must be even 3075 } else { 3076 *((kmp_indirect_lock_t **)user_lock) = lck; 3077 } 3078 3079 return lck; 3080} 3081 3082// User lock lookup for dynamically dispatched locks. 3083static __forceinline kmp_indirect_lock_t * 3084__kmp_lookup_indirect_lock(void **user_lock, const char *func) { 3085 if (__kmp_env_consistency_check) { 3086 kmp_indirect_lock_t *lck = NULL; 3087 if (user_lock == NULL) { 3088 KMP_FATAL(LockIsUninitialized, func); 3089 } 3090 if (OMP_LOCK_T_SIZE < sizeof(void *)) { 3091 kmp_lock_index_t idx = KMP_EXTRACT_I_INDEX(user_lock); 3092 if (idx >= __kmp_i_lock_table.size) { 3093 KMP_FATAL(LockIsUninitialized, func); 3094 } 3095 lck = KMP_GET_I_LOCK(idx); 3096 } else { 3097 lck = *((kmp_indirect_lock_t **)user_lock); 3098 } 3099 if (lck == NULL) { 3100 KMP_FATAL(LockIsUninitialized, func); 3101 } 3102 return lck; 3103 } else { 3104 if (OMP_LOCK_T_SIZE < sizeof(void *)) { 3105 return KMP_GET_I_LOCK(KMP_EXTRACT_I_INDEX(user_lock)); 3106 } else { 3107 return *((kmp_indirect_lock_t **)user_lock); 3108 } 3109 } 3110} 3111 3112static void __kmp_init_indirect_lock(kmp_dyna_lock_t *lock, 3113 kmp_dyna_lockseq_t seq) { 3114#if KMP_USE_ADAPTIVE_LOCKS 3115 if (seq == lockseq_adaptive && !__kmp_cpuinfo.rtm) { 3116 KMP_WARNING(AdaptiveNotSupported, "kmp_lockseq_t", "adaptive"); 3117 seq = lockseq_queuing; 3118 } 3119#endif 3120#if KMP_USE_TSX 3121 if (seq == lockseq_rtm && !__kmp_cpuinfo.rtm) { 3122 seq = lockseq_queuing; 3123 } 3124#endif 3125 kmp_indirect_locktag_t tag = KMP_GET_I_TAG(seq); 3126 kmp_indirect_lock_t *l = 3127 __kmp_allocate_indirect_lock((void **)lock, __kmp_entry_gtid(), tag); 3128 KMP_I_LOCK_FUNC(l, init)(l->lock); 3129 KA_TRACE( 3130 20, ("__kmp_init_indirect_lock: initialized indirect lock with type#%d\n", 3131 seq)); 3132} 3133 3134static void __kmp_destroy_indirect_lock(kmp_dyna_lock_t *lock) { 3135 kmp_uint32 gtid = __kmp_entry_gtid(); 3136 kmp_indirect_lock_t *l = 3137 __kmp_lookup_indirect_lock((void **)lock, "omp_destroy_lock"); 3138 KMP_I_LOCK_FUNC(l, destroy)(l->lock); 3139 kmp_indirect_locktag_t tag = l->type; 3140 3141 __kmp_acquire_lock(&__kmp_global_lock, gtid); 3142 3143 // Use the base lock's space to keep the pool chain. 3144 l->lock->pool.next = (kmp_user_lock_p)__kmp_indirect_lock_pool[tag]; 3145 if (OMP_LOCK_T_SIZE < sizeof(void *)) { 3146 l->lock->pool.index = KMP_EXTRACT_I_INDEX(lock); 3147 } 3148 __kmp_indirect_lock_pool[tag] = l; 3149 3150 __kmp_release_lock(&__kmp_global_lock, gtid); 3151} 3152 3153static int __kmp_set_indirect_lock(kmp_dyna_lock_t *lock, kmp_int32 gtid) { 3154 kmp_indirect_lock_t *l = KMP_LOOKUP_I_LOCK(lock); 3155 return KMP_I_LOCK_FUNC(l, set)(l->lock, gtid); 3156} 3157 3158static int __kmp_unset_indirect_lock(kmp_dyna_lock_t *lock, kmp_int32 gtid) { 3159 kmp_indirect_lock_t *l = KMP_LOOKUP_I_LOCK(lock); 3160 return KMP_I_LOCK_FUNC(l, unset)(l->lock, gtid); 3161} 3162 3163static int __kmp_test_indirect_lock(kmp_dyna_lock_t *lock, kmp_int32 gtid) { 3164 kmp_indirect_lock_t *l = KMP_LOOKUP_I_LOCK(lock); 3165 return KMP_I_LOCK_FUNC(l, test)(l->lock, gtid); 3166} 3167 3168static int __kmp_set_indirect_lock_with_checks(kmp_dyna_lock_t *lock, 3169 kmp_int32 gtid) { 3170 kmp_indirect_lock_t *l = 3171 __kmp_lookup_indirect_lock((void **)lock, "omp_set_lock"); 3172 return KMP_I_LOCK_FUNC(l, set)(l->lock, gtid); 3173} 3174 3175static int __kmp_unset_indirect_lock_with_checks(kmp_dyna_lock_t *lock, 3176 kmp_int32 gtid) { 3177 kmp_indirect_lock_t *l = 3178 __kmp_lookup_indirect_lock((void **)lock, "omp_unset_lock"); 3179 return KMP_I_LOCK_FUNC(l, unset)(l->lock, gtid); 3180} 3181 3182static int __kmp_test_indirect_lock_with_checks(kmp_dyna_lock_t *lock, 3183 kmp_int32 gtid) { 3184 kmp_indirect_lock_t *l = 3185 __kmp_lookup_indirect_lock((void **)lock, "omp_test_lock"); 3186 return KMP_I_LOCK_FUNC(l, test)(l->lock, gtid); 3187} 3188 3189kmp_dyna_lockseq_t __kmp_user_lock_seq = lockseq_queuing; 3190 3191// This is used only in kmp_error.cpp when consistency checking is on. 3192kmp_int32 __kmp_get_user_lock_owner(kmp_user_lock_p lck, kmp_uint32 seq) { 3193 switch (seq) { 3194 case lockseq_tas: 3195 case lockseq_nested_tas: 3196 return __kmp_get_tas_lock_owner((kmp_tas_lock_t *)lck); 3197#if KMP_USE_FUTEX 3198 case lockseq_futex: 3199 case lockseq_nested_futex: 3200 return __kmp_get_futex_lock_owner((kmp_futex_lock_t *)lck); 3201#endif 3202 case lockseq_ticket: 3203 case lockseq_nested_ticket: 3204 return __kmp_get_ticket_lock_owner((kmp_ticket_lock_t *)lck); 3205 case lockseq_queuing: 3206 case lockseq_nested_queuing: 3207#if KMP_USE_ADAPTIVE_LOCKS 3208 case lockseq_adaptive: 3209#endif 3210 return __kmp_get_queuing_lock_owner((kmp_queuing_lock_t *)lck); 3211 case lockseq_drdpa: 3212 case lockseq_nested_drdpa: 3213 return __kmp_get_drdpa_lock_owner((kmp_drdpa_lock_t *)lck); 3214 default: 3215 return 0; 3216 } 3217} 3218 3219// Initializes data for dynamic user locks. 3220void __kmp_init_dynamic_user_locks() { 3221 // Initialize jump table for the lock functions 3222 if (__kmp_env_consistency_check) { 3223 __kmp_direct_set = direct_set_check; 3224 __kmp_direct_unset = direct_unset_check; 3225 __kmp_direct_test = direct_test_check; 3226 __kmp_direct_destroy = direct_destroy_check; 3227 __kmp_indirect_set = indirect_set_check; 3228 __kmp_indirect_unset = indirect_unset_check; 3229 __kmp_indirect_test = indirect_test_check; 3230 __kmp_indirect_destroy = indirect_destroy_check; 3231 } else { 3232 __kmp_direct_set = direct_set; 3233 __kmp_direct_unset = direct_unset; 3234 __kmp_direct_test = direct_test; 3235 __kmp_direct_destroy = direct_destroy; 3236 __kmp_indirect_set = indirect_set; 3237 __kmp_indirect_unset = indirect_unset; 3238 __kmp_indirect_test = indirect_test; 3239 __kmp_indirect_destroy = indirect_destroy; 3240 } 3241 // If the user locks have already been initialized, then return. Allow the 3242 // switch between different KMP_CONSISTENCY_CHECK values, but do not allocate 3243 // new lock tables if they have already been allocated. 3244 if (__kmp_init_user_locks) 3245 return; 3246 3247 // Initialize lock index table 3248 __kmp_i_lock_table.size = KMP_I_LOCK_CHUNK; 3249 __kmp_i_lock_table.table = 3250 (kmp_indirect_lock_t **)__kmp_allocate(sizeof(kmp_indirect_lock_t *)); 3251 *(__kmp_i_lock_table.table) = (kmp_indirect_lock_t *)__kmp_allocate( 3252 KMP_I_LOCK_CHUNK * sizeof(kmp_indirect_lock_t)); 3253 __kmp_i_lock_table.next = 0; 3254 3255 // Indirect lock size 3256 __kmp_indirect_lock_size[locktag_ticket] = sizeof(kmp_ticket_lock_t); 3257 __kmp_indirect_lock_size[locktag_queuing] = sizeof(kmp_queuing_lock_t); 3258#if KMP_USE_ADAPTIVE_LOCKS 3259 __kmp_indirect_lock_size[locktag_adaptive] = sizeof(kmp_adaptive_lock_t); 3260#endif 3261 __kmp_indirect_lock_size[locktag_drdpa] = sizeof(kmp_drdpa_lock_t); 3262#if KMP_USE_TSX 3263 __kmp_indirect_lock_size[locktag_rtm] = sizeof(kmp_queuing_lock_t); 3264#endif 3265 __kmp_indirect_lock_size[locktag_nested_tas] = sizeof(kmp_tas_lock_t); 3266#if KMP_USE_FUTEX 3267 __kmp_indirect_lock_size[locktag_nested_futex] = sizeof(kmp_futex_lock_t); 3268#endif 3269 __kmp_indirect_lock_size[locktag_nested_ticket] = sizeof(kmp_ticket_lock_t); 3270 __kmp_indirect_lock_size[locktag_nested_queuing] = sizeof(kmp_queuing_lock_t); 3271 __kmp_indirect_lock_size[locktag_nested_drdpa] = sizeof(kmp_drdpa_lock_t); 3272 3273// Initialize lock accessor/modifier 3274#define fill_jumps(table, expand, sep) \ 3275 { \ 3276 table[locktag##sep##ticket] = expand(ticket); \ 3277 table[locktag##sep##queuing] = expand(queuing); \ 3278 table[locktag##sep##drdpa] = expand(drdpa); \ 3279 } 3280 3281#if KMP_USE_ADAPTIVE_LOCKS 3282#define fill_table(table, expand) \ 3283 { \ 3284 fill_jumps(table, expand, _); \ 3285 table[locktag_adaptive] = expand(queuing); \ 3286 fill_jumps(table, expand, _nested_); \ 3287 } 3288#else 3289#define fill_table(table, expand) \ 3290 { \ 3291 fill_jumps(table, expand, _); \ 3292 fill_jumps(table, expand, _nested_); \ 3293 } 3294#endif // KMP_USE_ADAPTIVE_LOCKS 3295 3296#define expand(l) \ 3297 (void (*)(kmp_user_lock_p, const ident_t *)) __kmp_set_##l##_lock_location 3298 fill_table(__kmp_indirect_set_location, expand); 3299#undef expand 3300#define expand(l) \ 3301 (void (*)(kmp_user_lock_p, kmp_lock_flags_t)) __kmp_set_##l##_lock_flags 3302 fill_table(__kmp_indirect_set_flags, expand); 3303#undef expand 3304#define expand(l) \ 3305 (const ident_t *(*)(kmp_user_lock_p)) __kmp_get_##l##_lock_location 3306 fill_table(__kmp_indirect_get_location, expand); 3307#undef expand 3308#define expand(l) \ 3309 (kmp_lock_flags_t(*)(kmp_user_lock_p)) __kmp_get_##l##_lock_flags 3310 fill_table(__kmp_indirect_get_flags, expand); 3311#undef expand 3312 3313 __kmp_init_user_locks = TRUE; 3314} 3315 3316// Clean up the lock table. 3317void __kmp_cleanup_indirect_user_locks() { 3318 kmp_lock_index_t i; 3319 int k; 3320 3321 // Clean up locks in the pools first (they were already destroyed before going 3322 // into the pools). 3323 for (k = 0; k < KMP_NUM_I_LOCKS; ++k) { 3324 kmp_indirect_lock_t *l = __kmp_indirect_lock_pool[k]; 3325 while (l != NULL) { 3326 kmp_indirect_lock_t *ll = l; 3327 l = (kmp_indirect_lock_t *)l->lock->pool.next; 3328 KA_TRACE(20, ("__kmp_cleanup_indirect_user_locks: freeing %p from pool\n", 3329 ll)); 3330 __kmp_free(ll->lock); 3331 ll->lock = NULL; 3332 } 3333 __kmp_indirect_lock_pool[k] = NULL; 3334 } 3335 // Clean up the remaining undestroyed locks. 3336 for (i = 0; i < __kmp_i_lock_table.next; i++) { 3337 kmp_indirect_lock_t *l = KMP_GET_I_LOCK(i); 3338 if (l->lock != NULL) { 3339 // Locks not destroyed explicitly need to be destroyed here. 3340 KMP_I_LOCK_FUNC(l, destroy)(l->lock); 3341 KA_TRACE( 3342 20, 3343 ("__kmp_cleanup_indirect_user_locks: destroy/freeing %p from table\n", 3344 l)); 3345 __kmp_free(l->lock); 3346 } 3347 } 3348 // Free the table 3349 for (i = 0; i < __kmp_i_lock_table.size / KMP_I_LOCK_CHUNK; i++) 3350 __kmp_free(__kmp_i_lock_table.table[i]); 3351 __kmp_free(__kmp_i_lock_table.table); 3352 3353 __kmp_init_user_locks = FALSE; 3354} 3355 3356enum kmp_lock_kind __kmp_user_lock_kind = lk_default; 3357int __kmp_num_locks_in_block = 1; // FIXME - tune this value 3358 3359#else // KMP_USE_DYNAMIC_LOCK 3360 3361static void __kmp_init_tas_lock_with_checks(kmp_tas_lock_t *lck) { 3362 __kmp_init_tas_lock(lck); 3363} 3364 3365static void __kmp_init_nested_tas_lock_with_checks(kmp_tas_lock_t *lck) { 3366 __kmp_init_nested_tas_lock(lck); 3367} 3368 3369#if KMP_USE_FUTEX 3370static void __kmp_init_futex_lock_with_checks(kmp_futex_lock_t *lck) { 3371 __kmp_init_futex_lock(lck); 3372} 3373 3374static void __kmp_init_nested_futex_lock_with_checks(kmp_futex_lock_t *lck) { 3375 __kmp_init_nested_futex_lock(lck); 3376} 3377#endif 3378 3379static int __kmp_is_ticket_lock_initialized(kmp_ticket_lock_t *lck) { 3380 return lck == lck->lk.self; 3381} 3382 3383static void __kmp_init_ticket_lock_with_checks(kmp_ticket_lock_t *lck) { 3384 __kmp_init_ticket_lock(lck); 3385} 3386 3387static void __kmp_init_nested_ticket_lock_with_checks(kmp_ticket_lock_t *lck) { 3388 __kmp_init_nested_ticket_lock(lck); 3389} 3390 3391static int __kmp_is_queuing_lock_initialized(kmp_queuing_lock_t *lck) { 3392 return lck == lck->lk.initialized; 3393} 3394 3395static void __kmp_init_queuing_lock_with_checks(kmp_queuing_lock_t *lck) { 3396 __kmp_init_queuing_lock(lck); 3397} 3398 3399static void 3400__kmp_init_nested_queuing_lock_with_checks(kmp_queuing_lock_t *lck) { 3401 __kmp_init_nested_queuing_lock(lck); 3402} 3403 3404#if KMP_USE_ADAPTIVE_LOCKS 3405static void __kmp_init_adaptive_lock_with_checks(kmp_adaptive_lock_t *lck) { 3406 __kmp_init_adaptive_lock(lck); 3407} 3408#endif 3409 3410static int __kmp_is_drdpa_lock_initialized(kmp_drdpa_lock_t *lck) { 3411 return lck == lck->lk.initialized; 3412} 3413 3414static void __kmp_init_drdpa_lock_with_checks(kmp_drdpa_lock_t *lck) { 3415 __kmp_init_drdpa_lock(lck); 3416} 3417 3418static void __kmp_init_nested_drdpa_lock_with_checks(kmp_drdpa_lock_t *lck) { 3419 __kmp_init_nested_drdpa_lock(lck); 3420} 3421 3422/* user locks 3423 * They are implemented as a table of function pointers which are set to the 3424 * lock functions of the appropriate kind, once that has been determined. */ 3425 3426enum kmp_lock_kind __kmp_user_lock_kind = lk_default; 3427 3428size_t __kmp_base_user_lock_size = 0; 3429size_t __kmp_user_lock_size = 0; 3430 3431kmp_int32 (*__kmp_get_user_lock_owner_)(kmp_user_lock_p lck) = NULL; 3432int (*__kmp_acquire_user_lock_with_checks_)(kmp_user_lock_p lck, 3433 kmp_int32 gtid) = NULL; 3434 3435int (*__kmp_test_user_lock_with_checks_)(kmp_user_lock_p lck, 3436 kmp_int32 gtid) = NULL; 3437int (*__kmp_release_user_lock_with_checks_)(kmp_user_lock_p lck, 3438 kmp_int32 gtid) = NULL; 3439void (*__kmp_init_user_lock_with_checks_)(kmp_user_lock_p lck) = NULL; 3440void (*__kmp_destroy_user_lock_)(kmp_user_lock_p lck) = NULL; 3441void (*__kmp_destroy_user_lock_with_checks_)(kmp_user_lock_p lck) = NULL; 3442int (*__kmp_acquire_nested_user_lock_with_checks_)(kmp_user_lock_p lck, 3443 kmp_int32 gtid) = NULL; 3444 3445int (*__kmp_test_nested_user_lock_with_checks_)(kmp_user_lock_p lck, 3446 kmp_int32 gtid) = NULL; 3447int (*__kmp_release_nested_user_lock_with_checks_)(kmp_user_lock_p lck, 3448 kmp_int32 gtid) = NULL; 3449void (*__kmp_init_nested_user_lock_with_checks_)(kmp_user_lock_p lck) = NULL; 3450void (*__kmp_destroy_nested_user_lock_with_checks_)(kmp_user_lock_p lck) = NULL; 3451 3452int (*__kmp_is_user_lock_initialized_)(kmp_user_lock_p lck) = NULL; 3453const ident_t *(*__kmp_get_user_lock_location_)(kmp_user_lock_p lck) = NULL; 3454void (*__kmp_set_user_lock_location_)(kmp_user_lock_p lck, 3455 const ident_t *loc) = NULL; 3456kmp_lock_flags_t (*__kmp_get_user_lock_flags_)(kmp_user_lock_p lck) = NULL; 3457void (*__kmp_set_user_lock_flags_)(kmp_user_lock_p lck, 3458 kmp_lock_flags_t flags) = NULL; 3459 3460void __kmp_set_user_lock_vptrs(kmp_lock_kind_t user_lock_kind) { 3461 switch (user_lock_kind) { 3462 case lk_default: 3463 default: 3464 KMP_ASSERT(0); 3465 3466 case lk_tas: { 3467 __kmp_base_user_lock_size = sizeof(kmp_base_tas_lock_t); 3468 __kmp_user_lock_size = sizeof(kmp_tas_lock_t); 3469 3470 __kmp_get_user_lock_owner_ = 3471 (kmp_int32(*)(kmp_user_lock_p))(&__kmp_get_tas_lock_owner); 3472 3473 if (__kmp_env_consistency_check) { 3474 KMP_BIND_USER_LOCK_WITH_CHECKS(tas); 3475 KMP_BIND_NESTED_USER_LOCK_WITH_CHECKS(tas); 3476 } else { 3477 KMP_BIND_USER_LOCK(tas); 3478 KMP_BIND_NESTED_USER_LOCK(tas); 3479 } 3480 3481 __kmp_destroy_user_lock_ = 3482 (void (*)(kmp_user_lock_p))(&__kmp_destroy_tas_lock); 3483 3484 __kmp_is_user_lock_initialized_ = (int (*)(kmp_user_lock_p))NULL; 3485 3486 __kmp_get_user_lock_location_ = (const ident_t *(*)(kmp_user_lock_p))NULL; 3487 3488 __kmp_set_user_lock_location_ = 3489 (void (*)(kmp_user_lock_p, const ident_t *))NULL; 3490 3491 __kmp_get_user_lock_flags_ = (kmp_lock_flags_t(*)(kmp_user_lock_p))NULL; 3492 3493 __kmp_set_user_lock_flags_ = 3494 (void (*)(kmp_user_lock_p, kmp_lock_flags_t))NULL; 3495 } break; 3496 3497#if KMP_USE_FUTEX 3498 3499 case lk_futex: { 3500 __kmp_base_user_lock_size = sizeof(kmp_base_futex_lock_t); 3501 __kmp_user_lock_size = sizeof(kmp_futex_lock_t); 3502 3503 __kmp_get_user_lock_owner_ = 3504 (kmp_int32(*)(kmp_user_lock_p))(&__kmp_get_futex_lock_owner); 3505 3506 if (__kmp_env_consistency_check) { 3507 KMP_BIND_USER_LOCK_WITH_CHECKS(futex); 3508 KMP_BIND_NESTED_USER_LOCK_WITH_CHECKS(futex); 3509 } else { 3510 KMP_BIND_USER_LOCK(futex); 3511 KMP_BIND_NESTED_USER_LOCK(futex); 3512 } 3513 3514 __kmp_destroy_user_lock_ = 3515 (void (*)(kmp_user_lock_p))(&__kmp_destroy_futex_lock); 3516 3517 __kmp_is_user_lock_initialized_ = (int (*)(kmp_user_lock_p))NULL; 3518 3519 __kmp_get_user_lock_location_ = (const ident_t *(*)(kmp_user_lock_p))NULL; 3520 3521 __kmp_set_user_lock_location_ = 3522 (void (*)(kmp_user_lock_p, const ident_t *))NULL; 3523 3524 __kmp_get_user_lock_flags_ = (kmp_lock_flags_t(*)(kmp_user_lock_p))NULL; 3525 3526 __kmp_set_user_lock_flags_ = 3527 (void (*)(kmp_user_lock_p, kmp_lock_flags_t))NULL; 3528 } break; 3529 3530#endif // KMP_USE_FUTEX 3531 3532 case lk_ticket: { 3533 __kmp_base_user_lock_size = sizeof(kmp_base_ticket_lock_t); 3534 __kmp_user_lock_size = sizeof(kmp_ticket_lock_t); 3535 3536 __kmp_get_user_lock_owner_ = 3537 (kmp_int32(*)(kmp_user_lock_p))(&__kmp_get_ticket_lock_owner); 3538 3539 if (__kmp_env_consistency_check) { 3540 KMP_BIND_USER_LOCK_WITH_CHECKS(ticket); 3541 KMP_BIND_NESTED_USER_LOCK_WITH_CHECKS(ticket); 3542 } else { 3543 KMP_BIND_USER_LOCK(ticket); 3544 KMP_BIND_NESTED_USER_LOCK(ticket); 3545 } 3546 3547 __kmp_destroy_user_lock_ = 3548 (void (*)(kmp_user_lock_p))(&__kmp_destroy_ticket_lock); 3549 3550 __kmp_is_user_lock_initialized_ = 3551 (int (*)(kmp_user_lock_p))(&__kmp_is_ticket_lock_initialized); 3552 3553 __kmp_get_user_lock_location_ = 3554 (const ident_t *(*)(kmp_user_lock_p))(&__kmp_get_ticket_lock_location); 3555 3556 __kmp_set_user_lock_location_ = (void (*)( 3557 kmp_user_lock_p, const ident_t *))(&__kmp_set_ticket_lock_location); 3558 3559 __kmp_get_user_lock_flags_ = 3560 (kmp_lock_flags_t(*)(kmp_user_lock_p))(&__kmp_get_ticket_lock_flags); 3561 3562 __kmp_set_user_lock_flags_ = (void (*)(kmp_user_lock_p, kmp_lock_flags_t))( 3563 &__kmp_set_ticket_lock_flags); 3564 } break; 3565 3566 case lk_queuing: { 3567 __kmp_base_user_lock_size = sizeof(kmp_base_queuing_lock_t); 3568 __kmp_user_lock_size = sizeof(kmp_queuing_lock_t); 3569 3570 __kmp_get_user_lock_owner_ = 3571 (kmp_int32(*)(kmp_user_lock_p))(&__kmp_get_queuing_lock_owner); 3572 3573 if (__kmp_env_consistency_check) { 3574 KMP_BIND_USER_LOCK_WITH_CHECKS(queuing); 3575 KMP_BIND_NESTED_USER_LOCK_WITH_CHECKS(queuing); 3576 } else { 3577 KMP_BIND_USER_LOCK(queuing); 3578 KMP_BIND_NESTED_USER_LOCK(queuing); 3579 } 3580 3581 __kmp_destroy_user_lock_ = 3582 (void (*)(kmp_user_lock_p))(&__kmp_destroy_queuing_lock); 3583 3584 __kmp_is_user_lock_initialized_ = 3585 (int (*)(kmp_user_lock_p))(&__kmp_is_queuing_lock_initialized); 3586 3587 __kmp_get_user_lock_location_ = 3588 (const ident_t *(*)(kmp_user_lock_p))(&__kmp_get_queuing_lock_location); 3589 3590 __kmp_set_user_lock_location_ = (void (*)( 3591 kmp_user_lock_p, const ident_t *))(&__kmp_set_queuing_lock_location); 3592 3593 __kmp_get_user_lock_flags_ = 3594 (kmp_lock_flags_t(*)(kmp_user_lock_p))(&__kmp_get_queuing_lock_flags); 3595 3596 __kmp_set_user_lock_flags_ = (void (*)(kmp_user_lock_p, kmp_lock_flags_t))( 3597 &__kmp_set_queuing_lock_flags); 3598 } break; 3599 3600#if KMP_USE_ADAPTIVE_LOCKS 3601 case lk_adaptive: { 3602 __kmp_base_user_lock_size = sizeof(kmp_base_adaptive_lock_t); 3603 __kmp_user_lock_size = sizeof(kmp_adaptive_lock_t); 3604 3605 __kmp_get_user_lock_owner_ = 3606 (kmp_int32(*)(kmp_user_lock_p))(&__kmp_get_queuing_lock_owner); 3607 3608 if (__kmp_env_consistency_check) { 3609 KMP_BIND_USER_LOCK_WITH_CHECKS(adaptive); 3610 } else { 3611 KMP_BIND_USER_LOCK(adaptive); 3612 } 3613 3614 __kmp_destroy_user_lock_ = 3615 (void (*)(kmp_user_lock_p))(&__kmp_destroy_adaptive_lock); 3616 3617 __kmp_is_user_lock_initialized_ = 3618 (int (*)(kmp_user_lock_p))(&__kmp_is_queuing_lock_initialized); 3619 3620 __kmp_get_user_lock_location_ = 3621 (const ident_t *(*)(kmp_user_lock_p))(&__kmp_get_queuing_lock_location); 3622 3623 __kmp_set_user_lock_location_ = (void (*)( 3624 kmp_user_lock_p, const ident_t *))(&__kmp_set_queuing_lock_location); 3625 3626 __kmp_get_user_lock_flags_ = 3627 (kmp_lock_flags_t(*)(kmp_user_lock_p))(&__kmp_get_queuing_lock_flags); 3628 3629 __kmp_set_user_lock_flags_ = (void (*)(kmp_user_lock_p, kmp_lock_flags_t))( 3630 &__kmp_set_queuing_lock_flags); 3631 3632 } break; 3633#endif // KMP_USE_ADAPTIVE_LOCKS 3634 3635 case lk_drdpa: { 3636 __kmp_base_user_lock_size = sizeof(kmp_base_drdpa_lock_t); 3637 __kmp_user_lock_size = sizeof(kmp_drdpa_lock_t); 3638 3639 __kmp_get_user_lock_owner_ = 3640 (kmp_int32(*)(kmp_user_lock_p))(&__kmp_get_drdpa_lock_owner); 3641 3642 if (__kmp_env_consistency_check) { 3643 KMP_BIND_USER_LOCK_WITH_CHECKS(drdpa); 3644 KMP_BIND_NESTED_USER_LOCK_WITH_CHECKS(drdpa); 3645 } else { 3646 KMP_BIND_USER_LOCK(drdpa); 3647 KMP_BIND_NESTED_USER_LOCK(drdpa); 3648 } 3649 3650 __kmp_destroy_user_lock_ = 3651 (void (*)(kmp_user_lock_p))(&__kmp_destroy_drdpa_lock); 3652 3653 __kmp_is_user_lock_initialized_ = 3654 (int (*)(kmp_user_lock_p))(&__kmp_is_drdpa_lock_initialized); 3655 3656 __kmp_get_user_lock_location_ = 3657 (const ident_t *(*)(kmp_user_lock_p))(&__kmp_get_drdpa_lock_location); 3658 3659 __kmp_set_user_lock_location_ = (void (*)( 3660 kmp_user_lock_p, const ident_t *))(&__kmp_set_drdpa_lock_location); 3661 3662 __kmp_get_user_lock_flags_ = 3663 (kmp_lock_flags_t(*)(kmp_user_lock_p))(&__kmp_get_drdpa_lock_flags); 3664 3665 __kmp_set_user_lock_flags_ = (void (*)(kmp_user_lock_p, kmp_lock_flags_t))( 3666 &__kmp_set_drdpa_lock_flags); 3667 } break; 3668 } 3669} 3670 3671// ---------------------------------------------------------------------------- 3672// User lock table & lock allocation 3673 3674kmp_lock_table_t __kmp_user_lock_table = {1, 0, NULL}; 3675kmp_user_lock_p __kmp_lock_pool = NULL; 3676 3677// Lock block-allocation support. 3678kmp_block_of_locks *__kmp_lock_blocks = NULL; 3679int __kmp_num_locks_in_block = 1; // FIXME - tune this value 3680 3681static kmp_lock_index_t __kmp_lock_table_insert(kmp_user_lock_p lck) { 3682 // Assume that kmp_global_lock is held upon entry/exit. 3683 kmp_lock_index_t index; 3684 if (__kmp_user_lock_table.used >= __kmp_user_lock_table.allocated) { 3685 kmp_lock_index_t size; 3686 kmp_user_lock_p *table; 3687 // Reallocate lock table. 3688 if (__kmp_user_lock_table.allocated == 0) { 3689 size = 1024; 3690 } else { 3691 size = __kmp_user_lock_table.allocated * 2; 3692 } 3693 table = (kmp_user_lock_p *)__kmp_allocate(sizeof(kmp_user_lock_p) * size); 3694 KMP_MEMCPY(table + 1, __kmp_user_lock_table.table + 1, 3695 sizeof(kmp_user_lock_p) * (__kmp_user_lock_table.used - 1)); 3696 table[0] = (kmp_user_lock_p)__kmp_user_lock_table.table; 3697 // We cannot free the previous table now, since it may be in use by other 3698 // threads. So save the pointer to the previous table in in the first 3699 // element of the new table. All the tables will be organized into a list, 3700 // and could be freed when library shutting down. 3701 __kmp_user_lock_table.table = table; 3702 __kmp_user_lock_table.allocated = size; 3703 } 3704 KMP_DEBUG_ASSERT(__kmp_user_lock_table.used < 3705 __kmp_user_lock_table.allocated); 3706 index = __kmp_user_lock_table.used; 3707 __kmp_user_lock_table.table[index] = lck; 3708 ++__kmp_user_lock_table.used; 3709 return index; 3710} 3711 3712static kmp_user_lock_p __kmp_lock_block_allocate() { 3713 // Assume that kmp_global_lock is held upon entry/exit. 3714 static int last_index = 0; 3715 if ((last_index >= __kmp_num_locks_in_block) || (__kmp_lock_blocks == NULL)) { 3716 // Restart the index. 3717 last_index = 0; 3718 // Need to allocate a new block. 3719 KMP_DEBUG_ASSERT(__kmp_user_lock_size > 0); 3720 size_t space_for_locks = __kmp_user_lock_size * __kmp_num_locks_in_block; 3721 char *buffer = 3722 (char *)__kmp_allocate(space_for_locks + sizeof(kmp_block_of_locks)); 3723 // Set up the new block. 3724 kmp_block_of_locks *new_block = 3725 (kmp_block_of_locks *)(&buffer[space_for_locks]); 3726 new_block->next_block = __kmp_lock_blocks; 3727 new_block->locks = (void *)buffer; 3728 // Publish the new block. 3729 KMP_MB(); 3730 __kmp_lock_blocks = new_block; 3731 } 3732 kmp_user_lock_p ret = (kmp_user_lock_p)(&( 3733 ((char *)(__kmp_lock_blocks->locks))[last_index * __kmp_user_lock_size])); 3734 last_index++; 3735 return ret; 3736} 3737 3738// Get memory for a lock. It may be freshly allocated memory or reused memory 3739// from lock pool. 3740kmp_user_lock_p __kmp_user_lock_allocate(void **user_lock, kmp_int32 gtid, 3741 kmp_lock_flags_t flags) { 3742 kmp_user_lock_p lck; 3743 kmp_lock_index_t index; 3744 KMP_DEBUG_ASSERT(user_lock); 3745 3746 __kmp_acquire_lock(&__kmp_global_lock, gtid); 3747 3748 if (__kmp_lock_pool == NULL) { 3749 // Lock pool is empty. Allocate new memory. 3750 3751 // ANNOTATION: Found no good way to express the syncronisation 3752 // between allocation and usage, so ignore the allocation 3753 ANNOTATE_IGNORE_WRITES_BEGIN(); 3754 if (__kmp_num_locks_in_block <= 1) { // Tune this cutoff point. 3755 lck = (kmp_user_lock_p)__kmp_allocate(__kmp_user_lock_size); 3756 } else { 3757 lck = __kmp_lock_block_allocate(); 3758 } 3759 ANNOTATE_IGNORE_WRITES_END(); 3760 3761 // Insert lock in the table so that it can be freed in __kmp_cleanup, 3762 // and debugger has info on all allocated locks. 3763 index = __kmp_lock_table_insert(lck); 3764 } else { 3765 // Pick up lock from pool. 3766 lck = __kmp_lock_pool; 3767 index = __kmp_lock_pool->pool.index; 3768 __kmp_lock_pool = __kmp_lock_pool->pool.next; 3769 } 3770 3771 // We could potentially differentiate between nested and regular locks 3772 // here, and do the lock table lookup for regular locks only. 3773 if (OMP_LOCK_T_SIZE < sizeof(void *)) { 3774 *((kmp_lock_index_t *)user_lock) = index; 3775 } else { 3776 *((kmp_user_lock_p *)user_lock) = lck; 3777 } 3778 3779 // mark the lock if it is critical section lock. 3780 __kmp_set_user_lock_flags(lck, flags); 3781 3782 __kmp_release_lock(&__kmp_global_lock, gtid); // AC: TODO move this line upper 3783 3784 return lck; 3785} 3786 3787// Put lock's memory to pool for reusing. 3788void __kmp_user_lock_free(void **user_lock, kmp_int32 gtid, 3789 kmp_user_lock_p lck) { 3790 KMP_DEBUG_ASSERT(user_lock != NULL); 3791 KMP_DEBUG_ASSERT(lck != NULL); 3792 3793 __kmp_acquire_lock(&__kmp_global_lock, gtid); 3794 3795 lck->pool.next = __kmp_lock_pool; 3796 __kmp_lock_pool = lck; 3797 if (OMP_LOCK_T_SIZE < sizeof(void *)) { 3798 kmp_lock_index_t index = *((kmp_lock_index_t *)user_lock); 3799 KMP_DEBUG_ASSERT(0 < index && index <= __kmp_user_lock_table.used); 3800 lck->pool.index = index; 3801 } 3802 3803 __kmp_release_lock(&__kmp_global_lock, gtid); 3804} 3805 3806kmp_user_lock_p __kmp_lookup_user_lock(void **user_lock, char const *func) { 3807 kmp_user_lock_p lck = NULL; 3808 3809 if (__kmp_env_consistency_check) { 3810 if (user_lock == NULL) { 3811 KMP_FATAL(LockIsUninitialized, func); 3812 } 3813 } 3814 3815 if (OMP_LOCK_T_SIZE < sizeof(void *)) { 3816 kmp_lock_index_t index = *((kmp_lock_index_t *)user_lock); 3817 if (__kmp_env_consistency_check) { 3818 if (!(0 < index && index < __kmp_user_lock_table.used)) { 3819 KMP_FATAL(LockIsUninitialized, func); 3820 } 3821 } 3822 KMP_DEBUG_ASSERT(0 < index && index < __kmp_user_lock_table.used); 3823 KMP_DEBUG_ASSERT(__kmp_user_lock_size > 0); 3824 lck = __kmp_user_lock_table.table[index]; 3825 } else { 3826 lck = *((kmp_user_lock_p *)user_lock); 3827 } 3828 3829 if (__kmp_env_consistency_check) { 3830 if (lck == NULL) { 3831 KMP_FATAL(LockIsUninitialized, func); 3832 } 3833 } 3834 3835 return lck; 3836} 3837 3838void __kmp_cleanup_user_locks(void) { 3839 // Reset lock pool. Don't worry about lock in the pool--we will free them when 3840 // iterating through lock table (it includes all the locks, dead or alive). 3841 __kmp_lock_pool = NULL; 3842 3843#define IS_CRITICAL(lck) \ 3844 ((__kmp_get_user_lock_flags_ != NULL) && \ 3845 ((*__kmp_get_user_lock_flags_)(lck)&kmp_lf_critical_section)) 3846 3847 // Loop through lock table, free all locks. 3848 // Do not free item [0], it is reserved for lock tables list. 3849 // 3850 // FIXME - we are iterating through a list of (pointers to) objects of type 3851 // union kmp_user_lock, but we have no way of knowing whether the base type is 3852 // currently "pool" or whatever the global user lock type is. 3853 // 3854 // We are relying on the fact that for all of the user lock types 3855 // (except "tas"), the first field in the lock struct is the "initialized" 3856 // field, which is set to the address of the lock object itself when 3857 // the lock is initialized. When the union is of type "pool", the 3858 // first field is a pointer to the next object in the free list, which 3859 // will not be the same address as the object itself. 3860 // 3861 // This means that the check (*__kmp_is_user_lock_initialized_)(lck) will fail 3862 // for "pool" objects on the free list. This must happen as the "location" 3863 // field of real user locks overlaps the "index" field of "pool" objects. 3864 // 3865 // It would be better to run through the free list, and remove all "pool" 3866 // objects from the lock table before executing this loop. However, 3867 // "pool" objects do not always have their index field set (only on 3868 // lin_32e), and I don't want to search the lock table for the address 3869 // of every "pool" object on the free list. 3870 while (__kmp_user_lock_table.used > 1) { 3871 const ident *loc; 3872 3873 // reduce __kmp_user_lock_table.used before freeing the lock, 3874 // so that state of locks is consistent 3875 kmp_user_lock_p lck = 3876 __kmp_user_lock_table.table[--__kmp_user_lock_table.used]; 3877 3878 if ((__kmp_is_user_lock_initialized_ != NULL) && 3879 (*__kmp_is_user_lock_initialized_)(lck)) { 3880 // Issue a warning if: KMP_CONSISTENCY_CHECK AND lock is initialized AND 3881 // it is NOT a critical section (user is not responsible for destroying 3882 // criticals) AND we know source location to report. 3883 if (__kmp_env_consistency_check && (!IS_CRITICAL(lck)) && 3884 ((loc = __kmp_get_user_lock_location(lck)) != NULL) && 3885 (loc->psource != NULL)) { 3886 kmp_str_loc_t str_loc = __kmp_str_loc_init(loc->psource, 0); 3887 KMP_WARNING(CnsLockNotDestroyed, str_loc.file, str_loc.line); 3888 __kmp_str_loc_free(&str_loc); 3889 } 3890 3891#ifdef KMP_DEBUG 3892 if (IS_CRITICAL(lck)) { 3893 KA_TRACE( 3894 20, 3895 ("__kmp_cleanup_user_locks: free critical section lock %p (%p)\n", 3896 lck, *(void **)lck)); 3897 } else { 3898 KA_TRACE(20, ("__kmp_cleanup_user_locks: free lock %p (%p)\n", lck, 3899 *(void **)lck)); 3900 } 3901#endif // KMP_DEBUG 3902 3903 // Cleanup internal lock dynamic resources (for drdpa locks particularly). 3904 __kmp_destroy_user_lock(lck); 3905 } 3906 3907 // Free the lock if block allocation of locks is not used. 3908 if (__kmp_lock_blocks == NULL) { 3909 __kmp_free(lck); 3910 } 3911 } 3912 3913#undef IS_CRITICAL 3914 3915 // delete lock table(s). 3916 kmp_user_lock_p *table_ptr = __kmp_user_lock_table.table; 3917 __kmp_user_lock_table.table = NULL; 3918 __kmp_user_lock_table.allocated = 0; 3919 3920 while (table_ptr != NULL) { 3921 // In the first element we saved the pointer to the previous 3922 // (smaller) lock table. 3923 kmp_user_lock_p *next = (kmp_user_lock_p *)(table_ptr[0]); 3924 __kmp_free(table_ptr); 3925 table_ptr = next; 3926 } 3927 3928 // Free buffers allocated for blocks of locks. 3929 kmp_block_of_locks_t *block_ptr = __kmp_lock_blocks; 3930 __kmp_lock_blocks = NULL; 3931 3932 while (block_ptr != NULL) { 3933 kmp_block_of_locks_t *next = block_ptr->next_block; 3934 __kmp_free(block_ptr->locks); 3935 // *block_ptr itself was allocated at the end of the locks vector. 3936 block_ptr = next; 3937 } 3938 3939 TCW_4(__kmp_init_user_locks, FALSE); 3940} 3941 3942#endif // KMP_USE_DYNAMIC_LOCK 3943