libgomp/libgomp.h

   1 /* Copyright (C) 2005-2015 Free Software Foundation, Inc.
   2    Contributed by Richard Henderson <rth@redhat.com>.
   3
   4    This file is part of the GNU Offloading and Multi Processing Library
   5    (libgomp).
   6
   7    Libgomp is free software; you can redistribute it and/or modify it
   8    under the terms of the GNU General Public License as published by
   9    the Free Software Foundation; either version 3, or (at your option)
  10    any later version.
  11
  12    Libgomp is distributed in the hope that it will be useful, but WITHOUT ANY
  13    WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
  14    FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  15    more details.
  16
  17    Under Section 7 of GPL version 3, you are granted additional
  18    permissions described in the GCC Runtime Library Exception, version
  19    3.1, as published by the Free Software Foundation.
  20
  21    You should have received a copy of the GNU General Public License and
  22    a copy of the GCC Runtime Library Exception along with this program;
  23    see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
  24    <http://www.gnu.org/licenses/>.  */
  25
  26 /* This file contains data types and function declarations that are not
  27    part of the official OpenACC or OpenMP user interfaces.  There are
  28    declarations in here that are part of the GNU Offloading and Multi
  29    Processing ABI, in that the compiler is required to know about them
  30    and use them.
  31
  32    The convention is that the all caps prefix "GOMP" is used group items
  33    that are part of the external ABI, and the lower case prefix "gomp"
  34    is used group items that are completely private to the library.  */
  35
  36 #ifndef LIBGOMP_H
  37 #define LIBGOMP_H 1
  38
  39 #include "config.h"
  40 #include "gstdint.h"
  41 #include "libgomp-plugin.h"
  42
  43 #include <pthread.h>
  44 #include <stdbool.h>
  45 #include <stdlib.h>
  46 #include <stdarg.h>
  47
  48 #ifdef HAVE_ATTRIBUTE_VISIBILITY
  49 # pragma GCC visibility push(hidden)
  50 #endif
  51
  52 /* If we were a C++ library, we'd get this from <std/atomic>.  */
  53 enum memmodel
  54 {
  55   MEMMODEL_RELAXED = 0,
  56   MEMMODEL_CONSUME = 1,
  57   MEMMODEL_ACQUIRE = 2,
  58   MEMMODEL_RELEASE = 3,
  59   MEMMODEL_ACQ_REL = 4,
  60   MEMMODEL_SEQ_CST = 5
  61 };
  62
  63 #include "sem.h"
  64 #include "mutex.h"
  65 #include "bar.h"
  66 #include "ptrlock.h"
  67
  68
  69 /* This structure contains the data to control one work-sharing construct,
  70    either a LOOP (FOR/DO) or a SECTIONS.  */
  71
  72 enum gomp_schedule_type
  73 {
  74   GFS_RUNTIME,
  75   GFS_STATIC,
  76   GFS_DYNAMIC,
  77   GFS_GUIDED,
  78   GFS_AUTO
  79 };
  80
  81 struct gomp_work_share
  82 {
  83   /* This member records the SCHEDULE clause to be used for this construct.
  84      The user specification of "runtime" will already have been resolved.
  85      If this is a SECTIONS construct, this value will always be DYNAMIC.  */
  86   enum gomp_schedule_type sched;
  87
  88   int mode;
  89
  90   union {
  91     struct {
  92       /* This is the chunk_size argument to the SCHEDULE clause.  */
  93       long chunk_size;
  94
  95       /* This is the iteration end point.  If this is a SECTIONS construct,
  96          this is the number of contained sections.  */
  97       long end;
  98
  99       /* This is the iteration step.  If this is a SECTIONS construct, this
 100          is always 1.  */
 101       long incr;
 102     };
 103
 104     struct {
 105       /* The same as above, but for the unsigned long long loop variants.  */
 106       unsigned long long chunk_size_ull;
 107       unsigned long long end_ull;
 108       unsigned long long incr_ull;
 109     };
 110   };
 111
 112   /* This is a circular queue that details which threads will be allowed
 113      into the ordered region and in which order.  When a thread allocates
 114      iterations on which it is going to work, it also registers itself at
 115      the end of the array.  When a thread reaches the ordered region, it
 116      checks to see if it is the one at the head of the queue.  If not, it
 117      blocks on its RELEASE semaphore.  */
 118   unsigned *ordered_team_ids;
 119
 120   /* This is the number of threads that have registered themselves in
 121      the circular queue ordered_team_ids.  */
 122   unsigned ordered_num_used;
 123
 124   /* This is the team_id of the currently acknowledged owner of the ordered
 125      section, or -1u if the ordered section has not been acknowledged by
 126      any thread.  This is distinguished from the thread that is *allowed*
 127      to take the section next.  */
 128   unsigned ordered_owner;
 129
 130   /* This is the index into the circular queue ordered_team_ids of the
 131      current thread that's allowed into the ordered reason.  */
 132   unsigned ordered_cur;
 133
 134   /* This is a chain of allocated gomp_work_share blocks, valid only
 135      in the first gomp_work_share struct in the block.  */
 136   struct gomp_work_share *next_alloc;
 137
 138   /* The above fields are written once during workshare initialization,
 139      or related to ordered worksharing.  Make sure the following fields
 140      are in a different cache line.  */
 141
 142   /* This lock protects the update of the following members.  */
 143   gomp_mutex_t lock __attribute__((aligned (64)));
 144
 145   /* This is the count of the number of threads that have exited the work
 146      share construct.  If the construct was marked nowait, they have moved on
 147      to other work; otherwise they're blocked on a barrier.  The last member
 148      of the team to exit the work share construct must deallocate it.  */
 149   unsigned threads_completed;
 150
 151   union {
 152     /* This is the next iteration value to be allocated.  In the case of
 153        GFS_STATIC loops, this the iteration start point and never changes.  */
 154     long next;
 155
 156     /* The same, but with unsigned long long type.  */
 157     unsigned long long next_ull;
 158
 159     /* This is the returned data structure for SINGLE COPYPRIVATE.  */
 160     void *copyprivate;
 161   };
 162
 163   union {
 164     /* Link to gomp_work_share struct for next work sharing construct
 165        encountered after this one.  */
 166     gomp_ptrlock_t next_ws;
 167
 168     /* gomp_work_share structs are chained in the free work share cache
 169        through this.  */
 170     struct gomp_work_share *next_free;
 171   };
 172
 173   /* If only few threads are in the team, ordered_team_ids can point
 174      to this array which fills the padding at the end of this struct.  */
 175   unsigned inline_ordered_team_ids[0];
 176 };
 177
 178 /* This structure contains all of the thread-local data associated with
 179    a thread team.  This is the data that must be saved when a thread
 180    encounters a nested PARALLEL construct.  */
 181
 182 struct gomp_team_state
 183 {
 184   /* This is the team of which the thread is currently a member.  */
 185   struct gomp_team *team;
 186
 187   /* This is the work share construct which this thread is currently
 188      processing.  Recall that with NOWAIT, not all threads may be
 189      processing the same construct.  */
 190   struct gomp_work_share *work_share;
 191
 192   /* This is the previous work share construct or NULL if there wasn't any.
 193      When all threads are done with the current work sharing construct,
 194      the previous one can be freed.  The current one can't, as its
 195      next_ws field is used.  */
 196   struct gomp_work_share *last_work_share;
 197
 198   /* This is the ID of this thread within the team.  This value is
 199      guaranteed to be between 0 and N-1, where N is the number of
 200      threads in the team.  */
 201   unsigned team_id;
 202
 203   /* Nesting level.  */
 204   unsigned level;
 205
 206   /* Active nesting level.  Only active parallel regions are counted.  */
 207   unsigned active_level;
 208
 209   /* Place-partition-var, offset and length into gomp_places_list array.  */
 210   unsigned place_partition_off;
 211   unsigned place_partition_len;
 212
 213 #ifdef HAVE_SYNC_BUILTINS
 214   /* Number of single stmts encountered.  */
 215   unsigned long single_count;
 216 #endif
 217
 218   /* For GFS_RUNTIME loops that resolved to GFS_STATIC, this is the
 219      trip number through the loop.  So first time a particular loop
 220      is encountered this number is 0, the second time through the loop
 221      is 1, etc.  This is unused when the compiler knows in advance that
 222      the loop is statically scheduled.  */
 223   unsigned long static_trip;
 224 };
 225
 226 struct target_mem_desc;
 227
 228 /* These are the OpenMP 4.0 Internal Control Variables described in
 229    section 2.3.1.  Those described as having one copy per task are
 230    stored within the structure; those described as having one copy
 231    for the whole program are (naturally) global variables.  */
 232
 233 struct gomp_task_icv
 234 {
 235   unsigned long nthreads_var;
 236   enum gomp_schedule_type run_sched_var;
 237   int run_sched_modifier;
 238   int default_device_var;
 239   unsigned int thread_limit_var;
 240   bool dyn_var;
 241   bool nest_var;
 242   char bind_var;
 243   /* Internal ICV.  */
 244   struct target_mem_desc *target_data;
 245 };
 246
 247 extern struct gomp_task_icv gomp_global_icv;
 248 #ifndef HAVE_SYNC_BUILTINS
 249 extern gomp_mutex_t gomp_managed_threads_lock;
 250 #endif
 251 extern unsigned long gomp_max_active_levels_var;
 252 extern bool gomp_cancel_var;
 253 extern unsigned long long gomp_spin_count_var, gomp_throttled_spin_count_var;
 254 extern unsigned long gomp_available_cpus, gomp_managed_threads;
 255 extern unsigned long *gomp_nthreads_var_list, gomp_nthreads_var_list_len;
 256 extern char *gomp_bind_var_list;
 257 extern unsigned long gomp_bind_var_list_len;
 258 extern void **gomp_places_list;
 259 extern unsigned long gomp_places_list_len;
 260 extern int gomp_debug_var;
 261 extern int goacc_device_num;
 262 extern char *goacc_device_type;
 263
 264 enum gomp_task_kind
 265 {
 266   GOMP_TASK_IMPLICIT,
 267   GOMP_TASK_IFFALSE,
 268   GOMP_TASK_WAITING,
 269   GOMP_TASK_TIED
 270 };
 271
 272 struct gomp_task;
 273 struct gomp_taskgroup;
 274 struct htab;
 275
 276 struct gomp_task_depend_entry
 277 {
 278   void *addr;
 279   struct gomp_task_depend_entry *next;
 280   struct gomp_task_depend_entry *prev;
 281   struct gomp_task *task;
 282   bool is_in;
 283   bool redundant;
 284   bool redundant_out;
 285 };
 286
 287 struct gomp_dependers_vec
 288 {
 289   size_t n_elem;
 290   size_t allocated;
 291   struct gomp_task *elem[];
 292 };
 293
 294 /* Used when in GOMP_taskwait or in gomp_task_maybe_wait_for_dependencies.  */
 295
 296 struct gomp_taskwait
 297 {
 298   bool in_taskwait;
 299   bool in_depend_wait;
 300   size_t n_depend;
 301   struct gomp_task *last_parent_depends_on;
 302   gomp_sem_t taskwait_sem;
 303 };
 304
 305 /* This structure describes a "task" to be run by a thread.  */
 306
 307 struct gomp_task
 308 {
 309   struct gomp_task *parent;
 310   struct gomp_task *children;
 311   struct gomp_task *next_child;
 312   struct gomp_task *prev_child;
 313   struct gomp_task *next_queue;
 314   struct gomp_task *prev_queue;
 315   struct gomp_task *next_taskgroup;
 316   struct gomp_task *prev_taskgroup;
 317   struct gomp_taskgroup *taskgroup;
 318   struct gomp_dependers_vec *dependers;
 319   struct htab *depend_hash;
 320   struct gomp_taskwait *taskwait;
 321   size_t depend_count;
 322   size_t num_dependees;
 323   struct gomp_task_icv icv;
 324   void (*fn) (void *);
 325   void *fn_data;
 326   enum gomp_task_kind kind;
 327   bool in_tied_task;
 328   bool final_task;
 329   bool copy_ctors_done;
 330   bool parent_depends_on;
 331   struct gomp_task_depend_entry depend[];
 332 };
 333
 334 struct gomp_taskgroup
 335 {
 336   struct gomp_taskgroup *prev;
 337   struct gomp_task *children;
 338   bool in_taskgroup_wait;
 339   bool cancelled;
 340   gomp_sem_t taskgroup_sem;
 341   size_t num_children;
 342 };
 343
 344 /* This structure describes a "team" of threads.  These are the threads
 345    that are spawned by a PARALLEL constructs, as well as the work sharing
 346    constructs that the team encounters.  */
 347
 348 struct gomp_team
 349 {
 350   /* This is the number of threads in the current team.  */
 351   unsigned nthreads;
 352
 353   /* This is number of gomp_work_share structs that have been allocated
 354      as a block last time.  */
 355   unsigned work_share_chunk;
 356
 357   /* This is the saved team state that applied to a master thread before
 358      the current thread was created.  */
 359   struct gomp_team_state prev_ts;
 360
 361   /* This semaphore should be used by the master thread instead of its
 362      "native" semaphore in the thread structure.  Required for nested
 363      parallels, as the master is a member of two teams.  */
 364   gomp_sem_t master_release;
 365
 366   /* This points to an array with pointers to the release semaphore
 367      of the threads in the team.  */
 368   gomp_sem_t **ordered_release;
 369
 370   /* List of work shares on which gomp_fini_work_share hasn't been
 371      called yet.  If the team hasn't been cancelled, this should be
 372      equal to each thr->ts.work_share, but otherwise it can be a possibly
 373      long list of workshares.  */
 374   struct gomp_work_share *work_shares_to_free;
 375
 376   /* List of gomp_work_share structs chained through next_free fields.
 377      This is populated and taken off only by the first thread in the
 378      team encountering a new work sharing construct, in a critical
 379      section.  */
 380   struct gomp_work_share *work_share_list_alloc;
 381
 382   /* List of gomp_work_share structs freed by free_work_share.  New
 383      entries are atomically added to the start of the list, and
 384      alloc_work_share can safely only move all but the first entry
 385      to work_share_list alloc, as free_work_share can happen concurrently
 386      with alloc_work_share.  */
 387   struct gomp_work_share *work_share_list_free;
 388
 389 #ifdef HAVE_SYNC_BUILTINS
 390   /* Number of simple single regions encountered by threads in this
 391      team.  */
 392   unsigned long single_count;
 393 #else
 394   /* Mutex protecting addition of workshares to work_share_list_free.  */
 395   gomp_mutex_t work_share_list_free_lock;
 396 #endif
 397
 398   /* This barrier is used for most synchronization of the team.  */
 399   gomp_barrier_t barrier;
 400
 401   /* Initial work shares, to avoid allocating any gomp_work_share
 402      structs in the common case.  */
 403   struct gomp_work_share work_shares[8];
 404
 405   gomp_mutex_t task_lock;
 406   struct gomp_task *task_queue;
 407   /* Number of all GOMP_TASK_{WAITING,TIED} tasks in the team.  */
 408   unsigned int task_count;
 409   /* Number of GOMP_TASK_WAITING tasks currently waiting to be scheduled.  */
 410   unsigned int task_queued_count;
 411   /* Number of GOMP_TASK_{WAITING,TIED} tasks currently running
 412      directly in gomp_barrier_handle_tasks; tasks spawned
 413      from e.g. GOMP_taskwait or GOMP_taskgroup_end don't count, even when
 414      that is called from a task run from gomp_barrier_handle_tasks.
 415      task_running_count should be always <= team->nthreads,
 416      and if current task isn't in_tied_task, then it will be
 417      even < team->nthreads.  */
 418   unsigned int task_running_count;
 419   int work_share_cancelled;
 420   int team_cancelled;
 421
 422   /* This array contains structures for implicit tasks.  */
 423   struct gomp_task implicit_task[];
 424 };
 425
 426 /* This structure contains all data that is private to libgomp and is
 427    allocated per thread.  */
 428
 429 struct gomp_thread
 430 {
 431   /* This is the function that the thread should run upon launch.  */
 432   void (*fn) (void *data);
 433   void *data;
 434
 435   /* This is the current team state for this thread.  The ts.team member
 436      is NULL only if the thread is idle.  */
 437   struct gomp_team_state ts;
 438
 439   /* This is the task that the thread is currently executing.  */
 440   struct gomp_task *task;
 441
 442   /* This semaphore is used for ordered loops.  */
 443   gomp_sem_t release;
 444
 445   /* Place this thread is bound to plus one, or zero if not bound
 446      to any place.  */
 447   unsigned int place;
 448
 449   /* User pthread thread pool */
 450   struct gomp_thread_pool *thread_pool;
 451 };
 452
 453
 454 struct gomp_thread_pool
 455 {
 456   /* This array manages threads spawned from the top level, which will
 457      return to the idle loop once the current PARALLEL construct ends.  */
 458   struct gomp_thread **threads;
 459   unsigned threads_size;
 460   unsigned threads_used;
 461   /* The last team is used for non-nested teams to delay their destruction to
 462      make sure all the threads in the team move on to the pool's barrier before
 463      the team's barrier is destroyed.  */
 464   struct gomp_team *last_team;
 465   /* Number of threads running in this contention group.  */
 466   unsigned long threads_busy;
 467
 468   /* This barrier holds and releases threads waiting in threads.  */
 469   gomp_barrier_t threads_dock;
 470 };
 471
 472 enum gomp_cancel_kind
 473 {
 474   GOMP_CANCEL_PARALLEL = 1,
 475   GOMP_CANCEL_LOOP = 2,
 476   GOMP_CANCEL_FOR = GOMP_CANCEL_LOOP,
 477   GOMP_CANCEL_DO = GOMP_CANCEL_LOOP,
 478   GOMP_CANCEL_SECTIONS = 4,
 479   GOMP_CANCEL_TASKGROUP = 8
 480 };
 481
 482 /* ... and here is that TLS data.  */
 483
 484 #if defined HAVE_TLS || defined USE_EMUTLS
 485 extern __thread struct gomp_thread gomp_tls_data;
 486 static inline struct gomp_thread *gomp_thread (void)
 487 {
 488   return &gomp_tls_data;
 489 }
 490 #else
 491 extern pthread_key_t gomp_tls_key;
 492 static inline struct gomp_thread *gomp_thread (void)
 493 {
 494   return pthread_getspecific (gomp_tls_key);
 495 }
 496 #endif
 497
 498 extern struct gomp_task_icv *gomp_new_icv (void);
 499
 500 /* Here's how to access the current copy of the ICVs.  */
 501
 502 static inline struct gomp_task_icv *gomp_icv (bool write)
 503 {
 504   struct gomp_task *task = gomp_thread ()->task;
 505   if (task)
 506     return &task->icv;
 507   else if (write)
 508     return gomp_new_icv ();
 509   else
 510     return &gomp_global_icv;
 511 }
 512
 513 /* The attributes to be used during thread creation.  */
 514 extern pthread_attr_t gomp_thread_attr;
 515
 516 extern pthread_key_t gomp_thread_destructor;
 517
 518 /* Function prototypes.  */
 519
 520 /* affinity.c */
 521
 522 extern void gomp_init_affinity (void);
 523 extern void gomp_init_thread_affinity (pthread_attr_t *, unsigned int);
 524 extern void **gomp_affinity_alloc (unsigned long, bool);
 525 extern void gomp_affinity_init_place (void *);
 526 extern bool gomp_affinity_add_cpus (void *, unsigned long, unsigned long,
 527                                     long, bool);
 528 extern bool gomp_affinity_remove_cpu (void *, unsigned long);
 529 extern bool gomp_affinity_copy_place (void *, void *, long);
 530 extern bool gomp_affinity_same_place (void *, void *);
 531 extern bool gomp_affinity_finalize_place_list (bool);
 532 extern bool gomp_affinity_init_level (int, unsigned long, bool);
 533 extern void gomp_affinity_print_place (void *);
 534
 535 /* alloc.c */
 536
 537 extern void *gomp_malloc (size_t) __attribute__((malloc));
 538 extern void *gomp_malloc_cleared (size_t) __attribute__((malloc));
 539 extern void *gomp_realloc (void *, size_t);
 540
 541 /* Avoid conflicting prototypes of alloca() in system headers by using
 542    GCC's builtin alloca().  */
 543 #define gomp_alloca(x)  __builtin_alloca(x)
 544
 545 /* error.c */
 546
 547 extern void gomp_vdebug (int, const char *, va_list);
 548 extern void gomp_debug (int, const char *, ...)
 549         __attribute__ ((format (printf, 2, 3)));
 550 #define gomp_vdebug(KIND, FMT, VALIST) \
 551   do { \
 552     if (__builtin_expect (gomp_debug_var, 0)) \
 553       (gomp_vdebug) ((KIND), (FMT), (VALIST)); \
 554   } while (0)
 555 #define gomp_debug(KIND, ...) \
 556   do { \
 557     if (__builtin_expect (gomp_debug_var, 0)) \
 558       (gomp_debug) ((KIND), __VA_ARGS__); \
 559   } while (0)
 560 extern void gomp_verror (const char *, va_list);
 561 extern void gomp_error (const char *, ...)
 562         __attribute__ ((format (printf, 1, 2)));
 563 extern void gomp_vfatal (const char *, va_list)
 564         __attribute__ ((noreturn));
 565 extern void gomp_fatal (const char *, ...)
 566         __attribute__ ((noreturn, format (printf, 1, 2)));
 567
 568 /* iter.c */
 569
 570 extern int gomp_iter_static_next (long *, long *);
 571 extern bool gomp_iter_dynamic_next_locked (long *, long *);
 572 extern bool gomp_iter_guided_next_locked (long *, long *);
 573
 574 #ifdef HAVE_SYNC_BUILTINS
 575 extern bool gomp_iter_dynamic_next (long *, long *);
 576 extern bool gomp_iter_guided_next (long *, long *);
 577 #endif
 578
 579 /* iter_ull.c */
 580
 581 extern int gomp_iter_ull_static_next (unsigned long long *,
 582                                       unsigned long long *);
 583 extern bool gomp_iter_ull_dynamic_next_locked (unsigned long long *,
 584                                                unsigned long long *);
 585 extern bool gomp_iter_ull_guided_next_locked (unsigned long long *,
 586                                               unsigned long long *);
 587
 588 #if defined HAVE_SYNC_BUILTINS && defined __LP64__
 589 extern bool gomp_iter_ull_dynamic_next (unsigned long long *,
 590                                         unsigned long long *);
 591 extern bool gomp_iter_ull_guided_next (unsigned long long *,
 592                                        unsigned long long *);
 593 #endif
 594
 595 /* ordered.c */
 596
 597 extern void gomp_ordered_first (void);
 598 extern void gomp_ordered_last (void);
 599 extern void gomp_ordered_next (void);
 600 extern void gomp_ordered_static_init (void);
 601 extern void gomp_ordered_static_next (void);
 602 extern void gomp_ordered_sync (void);
 603
 604 /* parallel.c */
 605
 606 extern unsigned gomp_resolve_num_threads (unsigned, unsigned);
 607
 608 /* proc.c (in config/) */
 609
 610 extern void gomp_init_num_threads (void);
 611 extern unsigned gomp_dynamic_max_threads (void);
 612
 613 /* task.c */
 614
 615 extern void gomp_init_task (struct gomp_task *, struct gomp_task *,
 616                             struct gomp_task_icv *);
 617 extern void gomp_end_task (void);
 618 extern void gomp_barrier_handle_tasks (gomp_barrier_state_t);
 619
 620 static void inline
 621 gomp_finish_task (struct gomp_task *task)
 622 {
 623   if (__builtin_expect (task->depend_hash != NULL, 0))
 624     free (task->depend_hash);
 625 }
 626
 627 /* team.c */
 628
 629 extern struct gomp_team *gomp_new_team (unsigned);
 630 extern void gomp_team_start (void (*) (void *), void *, unsigned,
 631                              unsigned, struct gomp_team *);
 632 extern void gomp_team_end (void);
 633 extern void gomp_free_thread (void *);
 634
 635 /* target.c */
 636
 637 extern void gomp_init_targets_once (void);
 638 extern int gomp_get_num_devices (void);
 639
 640 typedef struct splay_tree_node_s *splay_tree_node;
 641 typedef struct splay_tree_s *splay_tree;
 642 typedef struct splay_tree_key_s *splay_tree_key;
 643
 644 struct target_mem_desc {
 645   /* Reference count.  */
 646   uintptr_t refcount;
 647   /* All the splay nodes allocated together.  */
 648   splay_tree_node array;
 649   /* Start of the target region.  */
 650   uintptr_t tgt_start;
 651   /* End of the targer region.  */
 652   uintptr_t tgt_end;
 653   /* Handle to free.  */
 654   void *to_free;
 655   /* Previous target_mem_desc.  */
 656   struct target_mem_desc *prev;
 657   /* Number of items in following list.  */
 658   size_t list_count;
 659
 660   /* Corresponding target device descriptor.  */
 661   struct gomp_device_descr *device_descr;
 662
 663   /* List of splay keys to remove (or decrease refcount)
 664      at the end of region.  */
 665   splay_tree_key list[];
 666 };
 667
 668 struct splay_tree_key_s {
 669   /* Address of the host object.  */
 670   uintptr_t host_start;
 671   /* Address immediately after the host object.  */
 672   uintptr_t host_end;
 673   /* Descriptor of the target memory.  */
 674   struct target_mem_desc *tgt;
 675   /* Offset from tgt->tgt_start to the start of the target object.  */
 676   uintptr_t tgt_offset;
 677   /* Reference count.  */
 678   uintptr_t refcount;
 679   /* Asynchronous reference count.  */
 680   uintptr_t async_refcount;
 681   /* True if data should be copied from device to host at the end.  */
 682   bool copy_from;
 683 };
 684
 685 #include "splay-tree.h"
 686
 687 typedef struct acc_dispatch_t
 688 {
 689   /* This is a linked list of data mapped using the
 690      acc_map_data/acc_unmap_data or "acc enter data"/"acc exit data" pragmas.
 691      Unlike mapped_data in the goacc_thread struct, unmapping can
 692      happen out-of-order with respect to mapping.  */
 693   /* This is guarded by the lock in the "outer" struct gomp_device_descr.  */
 694   struct target_mem_desc *data_environ;
 695
 696   /* Execute.  */
 697   void (*exec_func) (void (*) (void *), size_t, void **, void **, size_t *,
 698                      unsigned short *, int, int, int, int, void *);
 699
 700   /* Async cleanup callback registration.  */
 701   void (*register_async_cleanup_func) (void *);
 702
 703   /* Asynchronous routines.  */
 704   int (*async_test_func) (int);
 705   int (*async_test_all_func) (void);
 706   void (*async_wait_func) (int);
 707   void (*async_wait_async_func) (int, int);
 708   void (*async_wait_all_func) (void);
 709   void (*async_wait_all_async_func) (int);
 710   void (*async_set_async_func) (int);
 711
 712   /* Create/destroy TLS data.  */
 713   void *(*create_thread_data_func) (int);
 714   void (*destroy_thread_data_func) (void *);
 715
 716   /* NVIDIA target specific routines.  */
 717   struct {
 718     void *(*get_current_device_func) (void);
 719     void *(*get_current_context_func) (void);
 720     void *(*get_stream_func) (int);
 721     int (*set_stream_func) (int, void *);
 722   } cuda;
 723 } acc_dispatch_t;
 724
 725 /* This structure describes accelerator device.
 726    It contains name of the corresponding libgomp plugin, function handlers for
 727    interaction with the device, ID-number of the device, and information about
 728    mapped memory.  */
 729 struct gomp_device_descr
 730 {
 731   /* Immutable data, which is only set during initialization, and which is not
 732      guarded by the lock.  */
 733
 734   /* The name of the device.  */
 735   const char *name;
 736
 737   /* Capabilities of device (supports OpenACC, OpenMP).  */
 738   unsigned int capabilities;
 739
 740   /* This is the ID number of device among devices of the same type.  */
 741   int target_id;
 742
 743   /* This is the TYPE of device.  */
 744   enum offload_target_type type;
 745
 746   /* Function handlers.  */
 747   const char *(*get_name_func) (void);
 748   unsigned int (*get_caps_func) (void);
 749   int (*get_type_func) (void);
 750   int (*get_num_devices_func) (void);
 751   void (*init_device_func) (int);
 752   void (*fini_device_func) (int);
 753   unsigned (*version_func) (void);
 754   int (*load_image_func) (int, unsigned, const void *, struct addr_pair **);
 755   void (*unload_image_func) (int, unsigned, const void *);
 756   void *(*alloc_func) (int, size_t);
 757   void (*free_func) (int, void *);
 758   void *(*dev2host_func) (int, void *, const void *, size_t);
 759   void *(*host2dev_func) (int, void *, const void *, size_t);
 760   void (*run_func) (int, void *, void *);
 761
 762   /* Splay tree containing information about mapped memory regions.  */
 763   struct splay_tree_s mem_map;
 764
 765   /* Mutex for the mutable data.  */
 766   gomp_mutex_t lock;
 767
 768   /* Set to true when device is initialized.  */
 769   bool is_initialized;
 770
 771   /* OpenACC-specific data and functions.  */
 772   /* This is mutable because of its mutable data_environ and target_data
 773      members.  */
 774   acc_dispatch_t openacc;
 775 };
 776
 777 extern void gomp_acc_insert_pointer (size_t, void **, size_t *, void *);
 778 extern void gomp_acc_remove_pointer (void *, bool, int, int);
 779
 780 extern struct target_mem_desc *gomp_map_vars (struct gomp_device_descr *,
 781                                               size_t, void **, void **,
 782                                               size_t *, void *, bool, bool);
 783 extern void gomp_copy_from_async (struct target_mem_desc *);
 784 extern void gomp_unmap_vars (struct target_mem_desc *, bool);
 785 extern void gomp_init_device (struct gomp_device_descr *);
 786 extern void gomp_free_memmap (struct splay_tree_s *);
 787 extern void gomp_fini_device (struct gomp_device_descr *);
 788 extern void gomp_unload_device (struct gomp_device_descr *);
 789
 790 /* work.c */
 791
 792 extern void gomp_init_work_share (struct gomp_work_share *, bool, unsigned);
 793 extern void gomp_fini_work_share (struct gomp_work_share *);
 794 extern bool gomp_work_share_start (bool);
 795 extern void gomp_work_share_end (void);
 796 extern bool gomp_work_share_end_cancel (void);
 797 extern void gomp_work_share_end_nowait (void);
 798
 799 static inline void
 800 gomp_work_share_init_done (void)
 801 {
 802   struct gomp_thread *thr = gomp_thread ();
 803   if (__builtin_expect (thr->ts.last_work_share != NULL, 1))
 804     gomp_ptrlock_set (&thr->ts.last_work_share->next_ws, thr->ts.work_share);
 805 }
 806
 807 #ifdef HAVE_ATTRIBUTE_VISIBILITY
 808 # pragma GCC visibility pop
 809 #endif
 810
 811 /* Now that we're back to default visibility, include the globals.  */
 812 #include "libgomp_g.h"
 813
 814 /* Include omp.h by parts.  */
 815 #include "omp-lock.h"
 816 #define _LIBGOMP_OMP_LOCK_DEFINED 1
 817 #include "omp.h.in"
 818
 819 #if !defined (HAVE_ATTRIBUTE_VISIBILITY) \
 820     || !defined (HAVE_ATTRIBUTE_ALIAS) \
 821     || !defined (HAVE_AS_SYMVER_DIRECTIVE) \
 822     || !defined (PIC) \
 823     || !defined (HAVE_SYMVER_SYMBOL_RENAMING_RUNTIME_SUPPORT)
 824 # undef LIBGOMP_GNU_SYMBOL_VERSIONING
 825 #endif
 826
 827 #ifdef LIBGOMP_GNU_SYMBOL_VERSIONING
 828 extern void gomp_init_lock_30 (omp_lock_t *) __GOMP_NOTHROW;
 829 extern void gomp_destroy_lock_30 (omp_lock_t *) __GOMP_NOTHROW;
 830 extern void gomp_set_lock_30 (omp_lock_t *) __GOMP_NOTHROW;
 831 extern void gomp_unset_lock_30 (omp_lock_t *) __GOMP_NOTHROW;
 832 extern int gomp_test_lock_30 (omp_lock_t *) __GOMP_NOTHROW;
 833 extern void gomp_init_nest_lock_30 (omp_nest_lock_t *) __GOMP_NOTHROW;
 834 extern void gomp_destroy_nest_lock_30 (omp_nest_lock_t *) __GOMP_NOTHROW;
 835 extern void gomp_set_nest_lock_30 (omp_nest_lock_t *) __GOMP_NOTHROW;
 836 extern void gomp_unset_nest_lock_30 (omp_nest_lock_t *) __GOMP_NOTHROW;
 837 extern int gomp_test_nest_lock_30 (omp_nest_lock_t *) __GOMP_NOTHROW;
 838
 839 extern void gomp_init_lock_25 (omp_lock_25_t *) __GOMP_NOTHROW;
 840 extern void gomp_destroy_lock_25 (omp_lock_25_t *) __GOMP_NOTHROW;
 841 extern void gomp_set_lock_25 (omp_lock_25_t *) __GOMP_NOTHROW;
 842 extern void gomp_unset_lock_25 (omp_lock_25_t *) __GOMP_NOTHROW;
 843 extern int gomp_test_lock_25 (omp_lock_25_t *) __GOMP_NOTHROW;
 844 extern void gomp_init_nest_lock_25 (omp_nest_lock_25_t *) __GOMP_NOTHROW;
 845 extern void gomp_destroy_nest_lock_25 (omp_nest_lock_25_t *) __GOMP_NOTHROW;
 846 extern void gomp_set_nest_lock_25 (omp_nest_lock_25_t *) __GOMP_NOTHROW;
 847 extern void gomp_unset_nest_lock_25 (omp_nest_lock_25_t *) __GOMP_NOTHROW;
 848 extern int gomp_test_nest_lock_25 (omp_nest_lock_25_t *) __GOMP_NOTHROW;
 849
 850 # define strong_alias(fn, al) \
 851   extern __typeof (fn) al __attribute__ ((alias (#fn)));
 852 # define omp_lock_symver(fn) \
 853   __asm (".symver g" #fn "_30, " #fn "@@OMP_3.0"); \
 854   __asm (".symver g" #fn "_25, " #fn "@OMP_1.0");
 855 #else
 856 # define gomp_init_lock_30 omp_init_lock
 857 # define gomp_destroy_lock_30 omp_destroy_lock
 858 # define gomp_set_lock_30 omp_set_lock
 859 # define gomp_unset_lock_30 omp_unset_lock
 860 # define gomp_test_lock_30 omp_test_lock
 861 # define gomp_init_nest_lock_30 omp_init_nest_lock
 862 # define gomp_destroy_nest_lock_30 omp_destroy_nest_lock
 863 # define gomp_set_nest_lock_30 omp_set_nest_lock
 864 # define gomp_unset_nest_lock_30 omp_unset_nest_lock
 865 # define gomp_test_nest_lock_30 omp_test_nest_lock
 866 #endif
 867
 868 #ifdef HAVE_ATTRIBUTE_VISIBILITY
 869 # define attribute_hidden __attribute__ ((visibility ("hidden")))
 870 #else
 871 # define attribute_hidden
 872 #endif
 873
 874 #ifdef HAVE_ATTRIBUTE_ALIAS
 875 # define ialias_ulp     ialias_str1(__USER_LABEL_PREFIX__)
 876 # define ialias_str1(x) ialias_str2(x)
 877 # define ialias_str2(x) #x
 878 # define ialias(fn) \
 879   extern __typeof (fn) gomp_ialias_##fn \
 880     __attribute__ ((alias (#fn))) attribute_hidden;
 881 # define ialias_redirect(fn) \
 882   extern __typeof (fn) fn __asm__ (ialias_ulp "gomp_ialias_" #fn) attribute_hidden;
 883 # define ialias_call(fn) gomp_ialias_ ## fn
 884 #else
 885 # define ialias(fn)
 886 # define ialias_redirect(fn)
 887 # define ialias_call(fn) fn
 888 #endif
 889
 890 #endif /* LIBGOMP_H */