LLVM OpenMP* Runtime Library
kmp.h
1 
2 /*
3  * kmp.h -- KPTS runtime header file.
4  */
5 
6 
7 //===----------------------------------------------------------------------===//
8 //
9 // The LLVM Compiler Infrastructure
10 //
11 // This file is dual licensed under the MIT and the University of Illinois Open
12 // Source Licenses. See LICENSE.txt for details.
13 //
14 //===----------------------------------------------------------------------===//
15 
16 
17 #ifndef KMP_H
18 #define KMP_H
19 
20 #include "kmp_config.h"
21 
22 /* #define BUILD_PARALLEL_ORDERED 1 */
23 
24 /* This fix replaces gettimeofday with clock_gettime for better scalability on
25  the Altix. Requires user code to be linked with -lrt.
26 */
27 //#define FIX_SGI_CLOCK
28 
29 /* Defines for OpenMP 3.0 tasking and auto scheduling */
30 
31 # ifndef KMP_STATIC_STEAL_ENABLED
32 # define KMP_STATIC_STEAL_ENABLED 1
33 # endif
34 
35 #define TASK_CURRENT_NOT_QUEUED 0
36 #define TASK_CURRENT_QUEUED 1
37 
38 #ifdef BUILD_TIED_TASK_STACK
39 #define TASK_STACK_EMPTY 0 // entries when the stack is empty
40 
41 #define TASK_STACK_BLOCK_BITS 5 // Used to define TASK_STACK_SIZE and TASK_STACK_MASK
42 #define TASK_STACK_BLOCK_SIZE ( 1 << TASK_STACK_BLOCK_BITS ) // Number of entries in each task stack array
43 #define TASK_STACK_INDEX_MASK ( TASK_STACK_BLOCK_SIZE - 1 ) // Mask for determining index into stack block
44 #endif // BUILD_TIED_TASK_STACK
45 
46 #define TASK_NOT_PUSHED 1
47 #define TASK_SUCCESSFULLY_PUSHED 0
48 #define TASK_TIED 1
49 #define TASK_UNTIED 0
50 #define TASK_EXPLICIT 1
51 #define TASK_IMPLICIT 0
52 #define TASK_PROXY 1
53 #define TASK_FULL 0
54 
55 #define KMP_CANCEL_THREADS
56 #define KMP_THREAD_ATTR
57 
58 // Android does not have pthread_cancel. Undefine KMP_CANCEL_THREADS if being
59 // built on Android
60 #if defined(__ANDROID__)
61 #undef KMP_CANCEL_THREADS
62 #endif
63 
64 #include <stdio.h>
65 #include <stdlib.h>
66 #include <stddef.h>
67 #include <stdarg.h>
68 #include <string.h>
69 #include <signal.h>
70 /* include <ctype.h> don't use; problems with /MD on Windows* OS NT due to bad Microsoft library */
71 /* some macros provided below to replace some of these functions */
72 #ifndef __ABSOFT_WIN
73 #include <sys/types.h>
74 #endif
75 #include <limits.h>
76 #include <time.h>
77 
78 #include <errno.h>
79 
80 #include "kmp_os.h"
81 
82 #include "kmp_safe_c_api.h"
83 
84 #if KMP_STATS_ENABLED
85 class kmp_stats_list;
86 #endif
87 
88 #if KMP_USE_HWLOC && KMP_AFFINITY_SUPPORTED
89 # include "hwloc.h"
90 #endif
91 
92 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
93 #include <xmmintrin.h>
94 #endif
95 
96 #include "kmp_version.h"
97 #include "kmp_debug.h"
98 #include "kmp_lock.h"
99 #if USE_DEBUGGER
100 #include "kmp_debugger.h"
101 #endif
102 #include "kmp_i18n.h"
103 
104 #define KMP_HANDLE_SIGNALS (KMP_OS_UNIX || KMP_OS_WINDOWS)
105 
106 #include "kmp_wrapper_malloc.h"
107 #if KMP_OS_UNIX
108 # include <unistd.h>
109 # if !defined NSIG && defined _NSIG
110 # define NSIG _NSIG
111 # endif
112 #endif
113 
114 #if KMP_OS_LINUX
115 # pragma weak clock_gettime
116 #endif
117 
118 #if OMPT_SUPPORT
119 #include "ompt-internal.h"
120 #endif
121 
122 /*Select data placement in NUMA memory */
123 #define NO_FIRST_TOUCH 0
124 #define FIRST_TOUCH 1 /* Exploit SGI's first touch page placement algo */
125 
126 /* If not specified on compile command line, assume no first touch */
127 #ifndef BUILD_MEMORY
128 #define BUILD_MEMORY NO_FIRST_TOUCH
129 #endif
130 
131 // 0 - no fast memory allocation, alignment: 8-byte on x86, 16-byte on x64.
132 // 3 - fast allocation using sync, non-sync free lists of any size, non-self free lists of limited size.
133 #ifndef USE_FAST_MEMORY
134 #define USE_FAST_MEMORY 3
135 #endif
136 
137 #ifndef KMP_NESTED_HOT_TEAMS
138 # define KMP_NESTED_HOT_TEAMS 0
139 # define USE_NESTED_HOT_ARG(x)
140 #else
141 # if KMP_NESTED_HOT_TEAMS
142 # if OMP_40_ENABLED
143 # define USE_NESTED_HOT_ARG(x) ,x
144 # else
145 // Nested hot teams feature depends on omp 4.0, disable it for earlier versions
146 # undef KMP_NESTED_HOT_TEAMS
147 # define KMP_NESTED_HOT_TEAMS 0
148 # define USE_NESTED_HOT_ARG(x)
149 # endif
150 # else
151 # define USE_NESTED_HOT_ARG(x)
152 # endif
153 #endif
154 
155 // Assume using BGET compare_exchange instruction instead of lock by default.
156 #ifndef USE_CMP_XCHG_FOR_BGET
157 #define USE_CMP_XCHG_FOR_BGET 1
158 #endif
159 
160 // Test to see if queuing lock is better than bootstrap lock for bget
161 // #ifndef USE_QUEUING_LOCK_FOR_BGET
162 // #define USE_QUEUING_LOCK_FOR_BGET
163 // #endif
164 
165 #define KMP_NSEC_PER_SEC 1000000000L
166 #define KMP_USEC_PER_SEC 1000000L
167 
173 // FIXME DOXYGEN... need to group these flags somehow (Making them an anonymous enum would do it...)
178 #define KMP_IDENT_IMB 0x01
179 
180 #define KMP_IDENT_KMPC 0x02
181 /* 0x04 is no longer used */
183 #define KMP_IDENT_AUTOPAR 0x08
184 
185 #define KMP_IDENT_ATOMIC_REDUCE 0x10
186 
187 #define KMP_IDENT_BARRIER_EXPL 0x20
188 
189 #define KMP_IDENT_BARRIER_IMPL 0x0040
190 #define KMP_IDENT_BARRIER_IMPL_MASK 0x01C0
191 #define KMP_IDENT_BARRIER_IMPL_FOR 0x0040
192 #define KMP_IDENT_BARRIER_IMPL_SECTIONS 0x00C0
193 
194 #define KMP_IDENT_BARRIER_IMPL_SINGLE 0x0140
195 #define KMP_IDENT_BARRIER_IMPL_WORKSHARE 0x01C0
196 
200 typedef struct ident {
201  kmp_int32 reserved_1;
202  kmp_int32 flags;
203  kmp_int32 reserved_2;
204 #if USE_ITT_BUILD
205  /* but currently used for storing region-specific ITT */
206  /* contextual information. */
207 #endif /* USE_ITT_BUILD */
208  kmp_int32 reserved_3;
209  char const *psource;
213 } ident_t;
218 // Some forward declarations.
219 
220 typedef union kmp_team kmp_team_t;
221 typedef struct kmp_taskdata kmp_taskdata_t;
222 typedef union kmp_task_team kmp_task_team_t;
223 typedef union kmp_team kmp_team_p;
224 typedef union kmp_info kmp_info_p;
225 typedef union kmp_root kmp_root_p;
226 
227 #ifdef __cplusplus
228 extern "C" {
229 #endif
230 
231 /* ------------------------------------------------------------------------ */
232 /* ------------------------------------------------------------------------ */
233 
234 /* Pack two 32-bit signed integers into a 64-bit signed integer */
235 /* ToDo: Fix word ordering for big-endian machines. */
236 #define KMP_PACK_64(HIGH_32,LOW_32) \
237  ( (kmp_int64) ((((kmp_uint64)(HIGH_32))<<32) | (kmp_uint64)(LOW_32)) )
238 
239 
240 /*
241  * Generic string manipulation macros.
242  * Assume that _x is of type char *
243  */
244 #define SKIP_WS(_x) { while (*(_x) == ' ' || *(_x) == '\t') (_x)++; }
245 #define SKIP_DIGITS(_x) { while (*(_x) >= '0' && *(_x) <= '9') (_x)++; }
246 #define SKIP_TO(_x,_c) { while (*(_x) != '\0' && *(_x) != (_c)) (_x)++; }
247 
248 /* ------------------------------------------------------------------------ */
249 /* ------------------------------------------------------------------------ */
250 
251 #define KMP_MAX( x, y ) ( (x) > (y) ? (x) : (y) )
252 #define KMP_MIN( x, y ) ( (x) < (y) ? (x) : (y) )
253 
254 /* ------------------------------------------------------------------------ */
255 /* ------------------------------------------------------------------------ */
256 
257 
258 /* Enumeration types */
259 
260 enum kmp_state_timer {
261  ts_stop,
262  ts_start,
263  ts_pause,
264 
265  ts_last_state
266 };
267 
268 enum dynamic_mode {
269  dynamic_default,
270 #ifdef USE_LOAD_BALANCE
271  dynamic_load_balance,
272 #endif /* USE_LOAD_BALANCE */
273  dynamic_random,
274  dynamic_thread_limit,
275  dynamic_max
276 };
277 
278 /* external schedule constants, duplicate enum omp_sched in omp.h in order to not include it here */
279 #ifndef KMP_SCHED_TYPE_DEFINED
280 #define KMP_SCHED_TYPE_DEFINED
281 typedef enum kmp_sched {
282  kmp_sched_lower = 0, // lower and upper bounds are for routine parameter check
283  // Note: need to adjust __kmp_sch_map global array in case this enum is changed
284  kmp_sched_static = 1, // mapped to kmp_sch_static_chunked (33)
285  kmp_sched_dynamic = 2, // mapped to kmp_sch_dynamic_chunked (35)
286  kmp_sched_guided = 3, // mapped to kmp_sch_guided_chunked (36)
287  kmp_sched_auto = 4, // mapped to kmp_sch_auto (38)
288  kmp_sched_upper_std = 5, // upper bound for standard schedules
289  kmp_sched_lower_ext = 100, // lower bound of Intel extension schedules
290  kmp_sched_trapezoidal = 101, // mapped to kmp_sch_trapezoidal (39)
291 #if KMP_STATIC_STEAL_ENABLED
292  kmp_sched_static_steal = 102, // mapped to kmp_sch_static_steal (44)
293 #endif
294  kmp_sched_upper,
295  kmp_sched_default = kmp_sched_static // default scheduling
296 } kmp_sched_t;
297 #endif
298 
305  kmp_sch_static_chunked = 33,
307  kmp_sch_dynamic_chunked = 35,
309  kmp_sch_runtime = 37,
311  kmp_sch_trapezoidal = 39,
312 
313  /* accessible only through KMP_SCHEDULE environment variable */
314  kmp_sch_static_greedy = 40,
315  kmp_sch_static_balanced = 41,
316  /* accessible only through KMP_SCHEDULE environment variable */
317  kmp_sch_guided_iterative_chunked = 42,
318  kmp_sch_guided_analytical_chunked = 43,
319 
322 #if OMP_45_ENABLED
323  kmp_sch_static_balanced_chunked = 45,
324 #endif
325 
326  /* accessible only through KMP_SCHEDULE environment variable */
330  kmp_ord_static_chunked = 65,
332  kmp_ord_dynamic_chunked = 67,
333  kmp_ord_guided_chunked = 68,
334  kmp_ord_runtime = 69,
336  kmp_ord_trapezoidal = 71,
339 #if OMP_40_ENABLED
340  /* Schedules for Distribute construct */
343 #endif
344 
345  /*
346  * For the "nomerge" versions, kmp_dispatch_next*() will always return
347  * a single iteration/chunk, even if the loop is serialized. For the
348  * schedule types listed above, the entire iteration vector is returned
349  * if the loop is serialized. This doesn't work for gcc/gcomp sections.
350  */
351  kmp_nm_lower = 160,
353  kmp_nm_static_chunked = (kmp_sch_static_chunked - kmp_sch_lower + kmp_nm_lower),
355  kmp_nm_dynamic_chunked = 163,
357  kmp_nm_runtime = 165,
358  kmp_nm_auto = 166,
359  kmp_nm_trapezoidal = 167,
360 
361  /* accessible only through KMP_SCHEDULE environment variable */
362  kmp_nm_static_greedy = 168,
363  kmp_nm_static_balanced = 169,
364  /* accessible only through KMP_SCHEDULE environment variable */
365  kmp_nm_guided_iterative_chunked = 170,
366  kmp_nm_guided_analytical_chunked = 171,
367  kmp_nm_static_steal = 172, /* accessible only through OMP_SCHEDULE environment variable */
368 
369  kmp_nm_ord_static_chunked = 193,
371  kmp_nm_ord_dynamic_chunked = 195,
372  kmp_nm_ord_guided_chunked = 196,
373  kmp_nm_ord_runtime = 197,
375  kmp_nm_ord_trapezoidal = 199,
376  kmp_nm_upper = 200,
378 #if OMP_45_ENABLED
379  /* Support for OpenMP 4.5 monotonic and nonmonotonic schedule modifiers.
380  * Since we need to distinguish the three possible cases (no modifier, monotonic modifier,
381  * nonmonotonic modifier), we need separate bits for each modifier.
382  * The absence of monotonic does not imply nonmonotonic, especially since 4.5 says
383  * that the behaviour of the "no modifier" case is implementation defined in 4.5,
384  * but will become "nonmonotonic" in 5.0.
385  *
386  * Since we're passing a full 32 bit value, we can use a couple of high bits for these
387  * flags; out of paranoia we avoid the sign bit.
388  *
389  * These modifiers can be or-ed into non-static schedules by the compiler to pass
390  * the additional information.
391  * They will be stripped early in the processing in __kmp_dispatch_init when setting up schedules, so
392  * most of the code won't ever see schedules with these bits set.
393  */
394  kmp_sch_modifier_monotonic = (1<<29),
395  kmp_sch_modifier_nonmonotonic = (1<<30),
397 # define SCHEDULE_WITHOUT_MODIFIERS(s) (enum sched_type)((s) & ~ (kmp_sch_modifier_nonmonotonic | kmp_sch_modifier_monotonic))
398 # define SCHEDULE_HAS_MONOTONIC(s) (((s) & kmp_sch_modifier_monotonic) != 0)
399 # define SCHEDULE_HAS_NONMONOTONIC(s) (((s) & kmp_sch_modifier_nonmonotonic) != 0)
400 # define SCHEDULE_HAS_NO_MODIFIERS(s) (((s) & (kmp_sch_modifier_nonmonotonic | kmp_sch_modifier_monotonic)) == 0)
401 #else
402  /* By doing this we hope to avoid multiple tests on OMP_45_ENABLED. Compilers can now eliminate tests on compile time
403  * constants and dead code that results from them, so we can leave code guarded by such an if in place.
404  */
405 # define SCHEDULE_WITHOUT_MODIFIERS(s) (s)
406 # define SCHEDULE_HAS_MONOTONIC(s) false
407 # define SCHEDULE_HAS_NONMONOTONIC(s) false
408 # define SCHEDULE_HAS_NO_MODIFIERS(s) true
409 #endif
410 
412 };
413 
414 /* Type to keep runtime schedule set via OMP_SCHEDULE or omp_set_schedule() */
415 typedef struct kmp_r_sched {
416  enum sched_type r_sched_type;
417  int chunk;
418 } kmp_r_sched_t;
419 
420 extern enum sched_type __kmp_sch_map[]; // map OMP 3.0 schedule types with our internal schedule types
421 
422 enum library_type {
423  library_none,
424  library_serial,
425  library_turnaround,
426  library_throughput
427 };
428 
429 #if KMP_OS_LINUX
430 enum clock_function_type {
431  clock_function_gettimeofday,
432  clock_function_clock_gettime
433 };
434 #endif /* KMP_OS_LINUX */
435 
436 #if KMP_ARCH_X86_64 && (KMP_OS_LINUX || KMP_OS_WINDOWS)
437 enum mic_type {
438  non_mic,
439  mic1,
440  mic2,
441  mic3,
442  dummy
443 };
444 #endif
445 
446 /* ------------------------------------------------------------------------ */
447 /* -- fast reduction stuff ------------------------------------------------ */
448 
449 #undef KMP_FAST_REDUCTION_BARRIER
450 #define KMP_FAST_REDUCTION_BARRIER 1
451 
452 #undef KMP_FAST_REDUCTION_CORE_DUO
453 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
454  #define KMP_FAST_REDUCTION_CORE_DUO 1
455 #endif
456 
457 enum _reduction_method {
458  reduction_method_not_defined = 0,
459  critical_reduce_block = ( 1 << 8 ),
460  atomic_reduce_block = ( 2 << 8 ),
461  tree_reduce_block = ( 3 << 8 ),
462  empty_reduce_block = ( 4 << 8 )
463 };
464 
465 // description of the packed_reduction_method variable
466 // the packed_reduction_method variable consists of two enum types variables that are packed together into 0-th byte and 1-st byte:
467 // 0: ( packed_reduction_method & 0x000000FF ) is a 'enum barrier_type' value of barrier that will be used in fast reduction: bs_plain_barrier or bs_reduction_barrier
468 // 1: ( packed_reduction_method & 0x0000FF00 ) is a reduction method that will be used in fast reduction;
469 // reduction method is of 'enum _reduction_method' type and it's defined the way so that the bits of 0-th byte are empty,
470 // so no need to execute a shift instruction while packing/unpacking
471 
472 #if KMP_FAST_REDUCTION_BARRIER
473  #define PACK_REDUCTION_METHOD_AND_BARRIER(reduction_method,barrier_type) \
474  ( ( reduction_method ) | ( barrier_type ) )
475 
476  #define UNPACK_REDUCTION_METHOD(packed_reduction_method) \
477  ( ( enum _reduction_method )( ( packed_reduction_method ) & ( 0x0000FF00 ) ) )
478 
479  #define UNPACK_REDUCTION_BARRIER(packed_reduction_method) \
480  ( ( enum barrier_type )( ( packed_reduction_method ) & ( 0x000000FF ) ) )
481 #else
482  #define PACK_REDUCTION_METHOD_AND_BARRIER(reduction_method,barrier_type) \
483  ( reduction_method )
484 
485  #define UNPACK_REDUCTION_METHOD(packed_reduction_method) \
486  ( packed_reduction_method )
487 
488  #define UNPACK_REDUCTION_BARRIER(packed_reduction_method) \
489  ( bs_plain_barrier )
490 #endif
491 
492 #define TEST_REDUCTION_METHOD(packed_reduction_method,which_reduction_block) \
493  ( ( UNPACK_REDUCTION_METHOD( packed_reduction_method ) ) == ( which_reduction_block ) )
494 
495 #if KMP_FAST_REDUCTION_BARRIER
496  #define TREE_REDUCE_BLOCK_WITH_REDUCTION_BARRIER \
497  ( PACK_REDUCTION_METHOD_AND_BARRIER( tree_reduce_block, bs_reduction_barrier ) )
498 
499  #define TREE_REDUCE_BLOCK_WITH_PLAIN_BARRIER \
500  ( PACK_REDUCTION_METHOD_AND_BARRIER( tree_reduce_block, bs_plain_barrier ) )
501 #endif
502 
503 typedef int PACKED_REDUCTION_METHOD_T;
504 
505 /* -- end of fast reduction stuff ----------------------------------------- */
506 
507 /* ------------------------------------------------------------------------ */
508 /* ------------------------------------------------------------------------ */
509 
510 #if KMP_OS_WINDOWS
511 # define USE_CBLKDATA
512 # pragma warning( push )
513 # pragma warning( disable: 271 310 )
514 # include <windows.h>
515 # pragma warning( pop )
516 #endif
517 
518 #if KMP_OS_UNIX
519 # include <pthread.h>
520 # include <dlfcn.h>
521 #endif
522 
523 /* ------------------------------------------------------------------------ */
524 /* ------------------------------------------------------------------------ */
525 
526 /*
527  * Only Linux* OS and Windows* OS support thread affinity.
528  */
529 #if KMP_AFFINITY_SUPPORTED
530 
531 // GROUP_AFFINITY is already defined for _MSC_VER>=1600 (VS2010 and later).
532 # if KMP_OS_WINDOWS
533 # if _MSC_VER < 1600
534 typedef struct GROUP_AFFINITY {
535  KAFFINITY Mask;
536  WORD Group;
537  WORD Reserved[3];
538 } GROUP_AFFINITY;
539 # endif /* _MSC_VER < 1600 */
540 # if KMP_GROUP_AFFINITY
541 extern int __kmp_num_proc_groups;
542 # else
543 static const int __kmp_num_proc_groups = 1;
544 # endif /* KMP_GROUP_AFFINITY */
545 typedef DWORD (*kmp_GetActiveProcessorCount_t)(WORD);
546 extern kmp_GetActiveProcessorCount_t __kmp_GetActiveProcessorCount;
547 
548 typedef WORD (*kmp_GetActiveProcessorGroupCount_t)(void);
549 extern kmp_GetActiveProcessorGroupCount_t __kmp_GetActiveProcessorGroupCount;
550 
551 typedef BOOL (*kmp_GetThreadGroupAffinity_t)(HANDLE, GROUP_AFFINITY *);
552 extern kmp_GetThreadGroupAffinity_t __kmp_GetThreadGroupAffinity;
553 
554 typedef BOOL (*kmp_SetThreadGroupAffinity_t)(HANDLE, const GROUP_AFFINITY *, GROUP_AFFINITY *);
555 extern kmp_SetThreadGroupAffinity_t __kmp_SetThreadGroupAffinity;
556 # endif /* KMP_OS_WINDOWS */
557 
558 # if KMP_USE_HWLOC
559 extern hwloc_topology_t __kmp_hwloc_topology;
560 extern int __kmp_hwloc_error;
561 # endif
562 
563 extern size_t __kmp_affin_mask_size;
564 # define KMP_AFFINITY_CAPABLE() (__kmp_affin_mask_size > 0)
565 # define KMP_AFFINITY_DISABLE() (__kmp_affin_mask_size = 0)
566 # define KMP_AFFINITY_ENABLE(mask_size) (__kmp_affin_mask_size = mask_size)
567 # define KMP_CPU_SET_ITERATE(i,mask) \
568  for (i = (mask)->begin(); i != (mask)->end() ; i = (mask)->next(i))
569 # define KMP_CPU_SET(i,mask) (mask)->set(i)
570 # define KMP_CPU_ISSET(i,mask) (mask)->is_set(i)
571 # define KMP_CPU_CLR(i,mask) (mask)->clear(i)
572 # define KMP_CPU_ZERO(mask) (mask)->zero()
573 # define KMP_CPU_COPY(dest, src) (dest)->copy(src)
574 # define KMP_CPU_AND(dest, src) (dest)->bitwise_and(src)
575 # define KMP_CPU_COMPLEMENT(max_bit_number, mask) (mask)->bitwise_not()
576 # define KMP_CPU_UNION(dest, src) (dest)->bitwise_or(src)
577 # define KMP_CPU_ALLOC(ptr) (ptr = __kmp_affinity_dispatch->allocate_mask())
578 # define KMP_CPU_FREE(ptr) __kmp_affinity_dispatch->deallocate_mask(ptr)
579 # define KMP_CPU_ALLOC_ON_STACK(ptr) KMP_CPU_ALLOC(ptr)
580 # define KMP_CPU_FREE_FROM_STACK(ptr) KMP_CPU_FREE(ptr)
581 # define KMP_CPU_INTERNAL_ALLOC(ptr) KMP_CPU_ALLOC(ptr)
582 # define KMP_CPU_INTERNAL_FREE(ptr) KMP_CPU_FREE(ptr)
583 # define KMP_CPU_INDEX(arr,i) __kmp_affinity_dispatch->index_mask_array(arr, i)
584 # define KMP_CPU_ALLOC_ARRAY(arr, n) (arr = __kmp_affinity_dispatch->allocate_mask_array(n))
585 # define KMP_CPU_FREE_ARRAY(arr, n) __kmp_affinity_dispatch->deallocate_mask_array(arr)
586 # define KMP_CPU_INTERNAL_ALLOC_ARRAY(arr, n) KMP_CPU_ALLOC_ARRAY(arr, n)
587 # define KMP_CPU_INTERNAL_FREE_ARRAY(arr, n) KMP_CPU_FREE_ARRAY(arr, n)
588 # define __kmp_get_system_affinity(mask, abort_bool) (mask)->get_system_affinity(abort_bool)
589 # define __kmp_set_system_affinity(mask, abort_bool) (mask)->set_system_affinity(abort_bool)
590 # define __kmp_get_proc_group(mask) (mask)->get_proc_group()
591 
592 class KMPAffinity {
593 public:
594  class Mask {
595  public:
596  void* operator new(size_t n);
597  void operator delete(void* p);
598  void* operator new[](size_t n);
599  void operator delete[](void* p);
600  virtual ~Mask() {}
601  // Set bit i to 1
602  virtual void set(int i) {}
603  // Return bit i
604  virtual bool is_set(int i) const { return false; }
605  // Set bit i to 0
606  virtual void clear(int i) {}
607  // Zero out entire mask
608  virtual void zero() {}
609  // Copy src into this mask
610  virtual void copy(const Mask* src) {}
611  // this &= rhs
612  virtual void bitwise_and(const Mask* rhs) {}
613  // this |= rhs
614  virtual void bitwise_or(const Mask* rhs) {}
615  // this = ~this
616  virtual void bitwise_not() {}
617  // API for iterating over an affinity mask
618  // for (int i = mask->begin(); i != mask->end(); i = mask->next(i))
619  virtual int begin() const { return 0; }
620  virtual int end() const { return 0; }
621  virtual int next(int previous) const { return 0; }
622  // Set the system's affinity to this affinity mask's value
623  virtual int set_system_affinity(bool abort_on_error) const { return -1; }
624  // Set this affinity mask to the current system affinity
625  virtual int get_system_affinity(bool abort_on_error) { return -1; }
626  // Only 1 DWORD in the mask should have any procs set.
627  // Return the appropriate index, or -1 for an invalid mask.
628  virtual int get_proc_group() const { return -1; }
629  };
630  void* operator new(size_t n);
631  void operator delete(void* p);
632  // Determine if affinity is capable
633  virtual void determine_capable(const char* env_var) {}
634  // Bind the current thread to os proc
635  virtual void bind_thread(int proc) {}
636  // Factory functions to allocate/deallocate a mask
637  virtual Mask* allocate_mask() { return nullptr; }
638  virtual void deallocate_mask(Mask* m) { }
639  virtual Mask* allocate_mask_array(int num) { return nullptr; }
640  virtual void deallocate_mask_array(Mask* m) { }
641  virtual Mask* index_mask_array(Mask* m, int index) { return nullptr; }
642  static void pick_api();
643  static void destroy_api();
644  enum api_type {
645  NATIVE_OS
646 #if KMP_USE_HWLOC
647  , HWLOC
648 #endif
649  };
650  virtual api_type get_api_type() const { KMP_ASSERT(0); return NATIVE_OS; };
651 private:
652  static bool picked_api;
653 };
654 
655 typedef KMPAffinity::Mask kmp_affin_mask_t;
656 extern KMPAffinity* __kmp_affinity_dispatch;
657 
658 //
659 // Declare local char buffers with this size for printing debug and info
660 // messages, using __kmp_affinity_print_mask().
661 //
662 #define KMP_AFFIN_MASK_PRINT_LEN 1024
663 
664 enum affinity_type {
665  affinity_none = 0,
666  affinity_physical,
667  affinity_logical,
668  affinity_compact,
669  affinity_scatter,
670  affinity_explicit,
671  affinity_balanced,
672  affinity_disabled, // not used outsize the env var parser
673  affinity_default
674 };
675 
676 enum affinity_gran {
677  affinity_gran_fine = 0,
678  affinity_gran_thread,
679  affinity_gran_core,
680  affinity_gran_package,
681  affinity_gran_node,
682 #if KMP_GROUP_AFFINITY
683  //
684  // The "group" granularity isn't necesssarily coarser than all of the
685  // other levels, but we put it last in the enum.
686  //
687  affinity_gran_group,
688 #endif /* KMP_GROUP_AFFINITY */
689  affinity_gran_default
690 };
691 
692 enum affinity_top_method {
693  affinity_top_method_all = 0, // try all (supported) methods, in order
694 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
695  affinity_top_method_apicid,
696  affinity_top_method_x2apicid,
697 #endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
698  affinity_top_method_cpuinfo, // KMP_CPUINFO_FILE is usable on Windows* OS, too
699 #if KMP_GROUP_AFFINITY
700  affinity_top_method_group,
701 #endif /* KMP_GROUP_AFFINITY */
702  affinity_top_method_flat,
703 #if KMP_USE_HWLOC
704  affinity_top_method_hwloc,
705 #endif
706  affinity_top_method_default
707 };
708 
709 #define affinity_respect_mask_default (-1)
710 
711 extern enum affinity_type __kmp_affinity_type; /* Affinity type */
712 extern enum affinity_gran __kmp_affinity_gran; /* Affinity granularity */
713 extern int __kmp_affinity_gran_levels; /* corresponding int value */
714 extern int __kmp_affinity_dups; /* Affinity duplicate masks */
715 extern enum affinity_top_method __kmp_affinity_top_method;
716 extern int __kmp_affinity_compact; /* Affinity 'compact' value */
717 extern int __kmp_affinity_offset; /* Affinity offset value */
718 extern int __kmp_affinity_verbose; /* Was verbose specified for KMP_AFFINITY? */
719 extern int __kmp_affinity_warnings; /* KMP_AFFINITY warnings enabled ? */
720 extern int __kmp_affinity_respect_mask; /* Respect process' initial affinity mask? */
721 extern char * __kmp_affinity_proclist; /* proc ID list */
722 extern kmp_affin_mask_t *__kmp_affinity_masks;
723 extern unsigned __kmp_affinity_num_masks;
724 extern void __kmp_affinity_bind_thread(int which);
725 
726 extern kmp_affin_mask_t *__kmp_affin_fullMask;
727 extern char const * __kmp_cpuinfo_file;
728 
729 #endif /* KMP_AFFINITY_SUPPORTED */
730 
731 #if OMP_40_ENABLED
732 
733 //
734 // This needs to be kept in sync with the values in omp.h !!!
735 //
736 typedef enum kmp_proc_bind_t {
737  proc_bind_false = 0,
738  proc_bind_true,
739  proc_bind_master,
740  proc_bind_close,
741  proc_bind_spread,
742  proc_bind_intel, // use KMP_AFFINITY interface
743  proc_bind_default
744 } kmp_proc_bind_t;
745 
746 typedef struct kmp_nested_proc_bind_t {
747  kmp_proc_bind_t *bind_types;
748  int size;
749  int used;
750 } kmp_nested_proc_bind_t;
751 
752 extern kmp_nested_proc_bind_t __kmp_nested_proc_bind;
753 
754 #endif /* OMP_40_ENABLED */
755 
756 # if KMP_AFFINITY_SUPPORTED
757 # define KMP_PLACE_ALL (-1)
758 # define KMP_PLACE_UNDEFINED (-2)
759 # endif /* KMP_AFFINITY_SUPPORTED */
760 
761 extern int __kmp_affinity_num_places;
762 
763 
764 #if OMP_40_ENABLED
765 typedef enum kmp_cancel_kind_t {
766  cancel_noreq = 0,
767  cancel_parallel = 1,
768  cancel_loop = 2,
769  cancel_sections = 3,
770  cancel_taskgroup = 4
771 } kmp_cancel_kind_t;
772 #endif // OMP_40_ENABLED
773 
774 extern int __kmp_place_num_sockets;
775 extern int __kmp_place_socket_offset;
776 extern int __kmp_place_num_cores;
777 extern int __kmp_place_core_offset;
778 extern int __kmp_place_num_threads_per_core;
779 
780 /* ------------------------------------------------------------------------ */
781 /* ------------------------------------------------------------------------ */
782 
783 #define KMP_PAD(type, sz) (sizeof(type) + (sz - ((sizeof(type) - 1) % (sz)) - 1))
784 
785 //
786 // We need to avoid using -1 as a GTID as +1 is added to the gtid
787 // when storing it in a lock, and the value 0 is reserved.
788 //
789 #define KMP_GTID_DNE (-2) /* Does not exist */
790 #define KMP_GTID_SHUTDOWN (-3) /* Library is shutting down */
791 #define KMP_GTID_MONITOR (-4) /* Monitor thread ID */
792 #define KMP_GTID_UNKNOWN (-5) /* Is not known */
793 #define KMP_GTID_MIN (-6) /* Minimal gtid for low bound check in DEBUG */
794 
795 #define __kmp_get_gtid() __kmp_get_global_thread_id()
796 #define __kmp_entry_gtid() __kmp_get_global_thread_id_reg()
797 
798 #define __kmp_tid_from_gtid(gtid) ( KMP_DEBUG_ASSERT( (gtid) >= 0 ), \
799  __kmp_threads[ (gtid) ]->th.th_info.ds.ds_tid )
800 
801 #define __kmp_get_tid() ( __kmp_tid_from_gtid( __kmp_get_gtid() ) )
802 #define __kmp_gtid_from_tid(tid,team) ( KMP_DEBUG_ASSERT( (tid) >= 0 && (team) != NULL ), \
803  team -> t.t_threads[ (tid) ] -> th.th_info .ds.ds_gtid )
804 
805 #define __kmp_get_team() ( __kmp_threads[ (__kmp_get_gtid()) ]-> th.th_team )
806 #define __kmp_team_from_gtid(gtid) ( KMP_DEBUG_ASSERT( (gtid) >= 0 ), \
807  __kmp_threads[ (gtid) ]-> th.th_team )
808 
809 #define __kmp_thread_from_gtid(gtid) ( KMP_DEBUG_ASSERT( (gtid) >= 0 ), __kmp_threads[ (gtid) ] )
810 #define __kmp_get_thread() ( __kmp_thread_from_gtid( __kmp_get_gtid() ) )
811 
812  // Returns current thread (pointer to kmp_info_t). In contrast to __kmp_get_thread(), it works
813  // with registered and not-yet-registered threads.
814 #define __kmp_gtid_from_thread(thr) ( KMP_DEBUG_ASSERT( (thr) != NULL ), \
815  (thr)->th.th_info.ds.ds_gtid )
816 
817 // AT: Which way is correct?
818 // AT: 1. nproc = __kmp_threads[ ( gtid ) ] -> th.th_team -> t.t_nproc;
819 // AT: 2. nproc = __kmp_threads[ ( gtid ) ] -> th.th_team_nproc;
820 #define __kmp_get_team_num_threads(gtid) ( __kmp_threads[ ( gtid ) ] -> th.th_team -> t.t_nproc )
821 
822 
823 /* ------------------------------------------------------------------------ */
824 /* ------------------------------------------------------------------------ */
825 
826 #define KMP_UINT64_MAX (~((kmp_uint64)1<<((sizeof(kmp_uint64)*(1<<3))-1)))
827 
828 #define KMP_MIN_NTH 1
829 
830 #ifndef KMP_MAX_NTH
831 # if defined(PTHREAD_THREADS_MAX) && PTHREAD_THREADS_MAX < INT_MAX
832 # define KMP_MAX_NTH PTHREAD_THREADS_MAX
833 # else
834 # define KMP_MAX_NTH INT_MAX
835 # endif
836 #endif /* KMP_MAX_NTH */
837 
838 #ifdef PTHREAD_STACK_MIN
839 # define KMP_MIN_STKSIZE PTHREAD_STACK_MIN
840 #else
841 # define KMP_MIN_STKSIZE ((size_t)(32 * 1024))
842 #endif
843 
844 #define KMP_MAX_STKSIZE (~((size_t)1<<((sizeof(size_t)*(1<<3))-1)))
845 
846 #if KMP_ARCH_X86
847 # define KMP_DEFAULT_STKSIZE ((size_t)(2 * 1024 * 1024))
848 #elif KMP_ARCH_X86_64
849 # define KMP_DEFAULT_STKSIZE ((size_t)(4 * 1024 * 1024))
850 # define KMP_BACKUP_STKSIZE ((size_t)(2 * 1024 * 1024))
851 #else
852 # define KMP_DEFAULT_STKSIZE ((size_t)(1024 * 1024))
853 #endif
854 
855 #define KMP_DEFAULT_MALLOC_POOL_INCR ((size_t) (1024 * 1024))
856 #define KMP_MIN_MALLOC_POOL_INCR ((size_t) (4 * 1024))
857 #define KMP_MAX_MALLOC_POOL_INCR (~((size_t)1<<((sizeof(size_t)*(1<<3))-1)))
858 
859 #define KMP_MIN_STKOFFSET (0)
860 #define KMP_MAX_STKOFFSET KMP_MAX_STKSIZE
861 #if KMP_OS_DARWIN
862 # define KMP_DEFAULT_STKOFFSET KMP_MIN_STKOFFSET
863 #else
864 # define KMP_DEFAULT_STKOFFSET CACHE_LINE
865 #endif
866 
867 #define KMP_MIN_STKPADDING (0)
868 #define KMP_MAX_STKPADDING (2 * 1024 * 1024)
869 
870 #define KMP_BLOCKTIME_MULTIPLIER (1000) /* number of blocktime units per second */
871 #define KMP_MIN_BLOCKTIME (0)
872 #define KMP_MAX_BLOCKTIME (INT_MAX) /* Must be this for "infinite" setting the work */
873 #define KMP_DEFAULT_BLOCKTIME (200) /* __kmp_blocktime is in milliseconds */
874 
875 #if KMP_USE_MONITOR
876 #define KMP_DEFAULT_MONITOR_STKSIZE ((size_t)(64 * 1024))
877 #define KMP_MIN_MONITOR_WAKEUPS (1) /* min number of times monitor wakes up per second */
878 #define KMP_MAX_MONITOR_WAKEUPS (1000) /* maximum number of times monitor can wake up per second */
879 
880 /* Calculate new number of monitor wakeups for a specific block time based on previous monitor_wakeups */
881 /* Only allow increasing number of wakeups */
882 #define KMP_WAKEUPS_FROM_BLOCKTIME(blocktime, monitor_wakeups) \
883  ( ((blocktime) == KMP_MAX_BLOCKTIME) ? (monitor_wakeups) : \
884  ((blocktime) == KMP_MIN_BLOCKTIME) ? KMP_MAX_MONITOR_WAKEUPS : \
885  ((monitor_wakeups) > (KMP_BLOCKTIME_MULTIPLIER / (blocktime))) ? (monitor_wakeups) : \
886  (KMP_BLOCKTIME_MULTIPLIER) / (blocktime) )
887 
888 /* Calculate number of intervals for a specific block time based on monitor_wakeups */
889 #define KMP_INTERVALS_FROM_BLOCKTIME(blocktime, monitor_wakeups) \
890  ( ( (blocktime) + (KMP_BLOCKTIME_MULTIPLIER / (monitor_wakeups)) - 1 ) / \
891  (KMP_BLOCKTIME_MULTIPLIER / (monitor_wakeups)) )
892 #endif // KMP_USE_MONITOR
893 
894 #define KMP_MIN_STATSCOLS 40
895 #define KMP_MAX_STATSCOLS 4096
896 #define KMP_DEFAULT_STATSCOLS 80
897 
898 #define KMP_MIN_INTERVAL 0
899 #define KMP_MAX_INTERVAL (INT_MAX-1)
900 #define KMP_DEFAULT_INTERVAL 0
901 
902 #define KMP_MIN_CHUNK 1
903 #define KMP_MAX_CHUNK (INT_MAX-1)
904 #define KMP_DEFAULT_CHUNK 1
905 
906 #define KMP_MIN_INIT_WAIT 1
907 #define KMP_MAX_INIT_WAIT (INT_MAX/2)
908 #define KMP_DEFAULT_INIT_WAIT 2048U
909 
910 #define KMP_MIN_NEXT_WAIT 1
911 #define KMP_MAX_NEXT_WAIT (INT_MAX/2)
912 #define KMP_DEFAULT_NEXT_WAIT 1024U
913 
914 #define KMP_DFLT_DISP_NUM_BUFF 7
915 #define KMP_MAX_ORDERED 8
916 
917 #define KMP_MAX_FIELDS 32
918 
919 #define KMP_MAX_BRANCH_BITS 31
920 
921 #define KMP_MAX_ACTIVE_LEVELS_LIMIT INT_MAX
922 
923 #define KMP_MAX_DEFAULT_DEVICE_LIMIT INT_MAX
924 
925 #define KMP_MAX_TASK_PRIORITY_LIMIT INT_MAX
926 
927 /* Minimum number of threads before switch to TLS gtid (experimentally determined) */
928 /* josh TODO: what about OS X* tuning? */
929 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
930 # define KMP_TLS_GTID_MIN 5
931 #else
932 # define KMP_TLS_GTID_MIN INT_MAX
933 #endif
934 
935 #define KMP_MASTER_TID(tid) ( (tid) == 0 )
936 #define KMP_WORKER_TID(tid) ( (tid) != 0 )
937 
938 #define KMP_MASTER_GTID(gtid) ( __kmp_tid_from_gtid((gtid)) == 0 )
939 #define KMP_WORKER_GTID(gtid) ( __kmp_tid_from_gtid((gtid)) != 0 )
940 #define KMP_UBER_GTID(gtid) \
941  ( \
942  KMP_DEBUG_ASSERT( (gtid) >= KMP_GTID_MIN ), \
943  KMP_DEBUG_ASSERT( (gtid) < __kmp_threads_capacity ), \
944  (gtid) >= 0 && __kmp_root[(gtid)] && __kmp_threads[(gtid)] && \
945  (__kmp_threads[(gtid)] == __kmp_root[(gtid)]->r.r_uber_thread)\
946  )
947 #define KMP_INITIAL_GTID(gtid) ( (gtid) == 0 )
948 
949 #ifndef TRUE
950 #define FALSE 0
951 #define TRUE (! FALSE)
952 #endif
953 
954 /* NOTE: all of the following constants must be even */
955 
956 #if KMP_OS_WINDOWS
957 # define KMP_INIT_WAIT 64U /* initial number of spin-tests */
958 # define KMP_NEXT_WAIT 32U /* susequent number of spin-tests */
959 #elif KMP_OS_CNK
960 # define KMP_INIT_WAIT 16U /* initial number of spin-tests */
961 # define KMP_NEXT_WAIT 8U /* susequent number of spin-tests */
962 #elif KMP_OS_LINUX
963 # define KMP_INIT_WAIT 1024U /* initial number of spin-tests */
964 # define KMP_NEXT_WAIT 512U /* susequent number of spin-tests */
965 #elif KMP_OS_DARWIN
966 /* TODO: tune for KMP_OS_DARWIN */
967 # define KMP_INIT_WAIT 1024U /* initial number of spin-tests */
968 # define KMP_NEXT_WAIT 512U /* susequent number of spin-tests */
969 #elif KMP_OS_FREEBSD
970 /* TODO: tune for KMP_OS_FREEBSD */
971 # define KMP_INIT_WAIT 1024U /* initial number of spin-tests */
972 # define KMP_NEXT_WAIT 512U /* susequent number of spin-tests */
973 #elif KMP_OS_NETBSD
974 /* TODO: tune for KMP_OS_NETBSD */
975 # define KMP_INIT_WAIT 1024U /* initial number of spin-tests */
976 # define KMP_NEXT_WAIT 512U /* susequent number of spin-tests */
977 #endif
978 
979 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
980 typedef struct kmp_cpuid {
981  kmp_uint32 eax;
982  kmp_uint32 ebx;
983  kmp_uint32 ecx;
984  kmp_uint32 edx;
985 } kmp_cpuid_t;
986 extern void __kmp_x86_cpuid( int mode, int mode2, struct kmp_cpuid *p );
987 # if KMP_ARCH_X86
988  extern void __kmp_x86_pause( void );
989 # elif KMP_MIC
990  static void __kmp_x86_pause( void ) { _mm_delay_32( 100 ); }
991 # else
992  static void __kmp_x86_pause( void ) { _mm_pause(); }
993 # endif
994 # define KMP_CPU_PAUSE() __kmp_x86_pause()
995 #elif KMP_ARCH_PPC64
996 # define KMP_PPC64_PRI_LOW() __asm__ volatile ("or 1, 1, 1")
997 # define KMP_PPC64_PRI_MED() __asm__ volatile ("or 2, 2, 2")
998 # define KMP_PPC64_PRI_LOC_MB() __asm__ volatile ("" : : : "memory")
999 # define KMP_CPU_PAUSE() do { KMP_PPC64_PRI_LOW(); KMP_PPC64_PRI_MED(); KMP_PPC64_PRI_LOC_MB(); } while (0)
1000 #else
1001 # define KMP_CPU_PAUSE() /* nothing to do */
1002 #endif
1003 
1004 #define KMP_INIT_YIELD(count) { (count) = __kmp_yield_init; }
1005 
1006 #define KMP_YIELD(cond) { KMP_CPU_PAUSE(); __kmp_yield( (cond) ); }
1007 
1008 // Note the decrement of 2 in the following Macros. With KMP_LIBRARY=turnaround,
1009 // there should be no yielding since the starting value from KMP_INIT_YIELD() is odd.
1010 
1011 #define KMP_YIELD_WHEN(cond,count) { KMP_CPU_PAUSE(); (count) -= 2; \
1012  if (!(count)) { KMP_YIELD(cond); (count) = __kmp_yield_next; } }
1013 #define KMP_YIELD_SPIN(count) { KMP_CPU_PAUSE(); (count) -=2; \
1014  if (!(count)) { KMP_YIELD(1); (count) = __kmp_yield_next; } }
1015 
1016 /* ------------------------------------------------------------------------ */
1017 /* Support datatypes for the orphaned construct nesting checks. */
1018 /* ------------------------------------------------------------------------ */
1019 
1020 enum cons_type {
1021  ct_none,
1022  ct_parallel,
1023  ct_pdo,
1024  ct_pdo_ordered,
1025  ct_psections,
1026  ct_psingle,
1027 
1028  /* the following must be left in order and not split up */
1029  ct_taskq,
1030  ct_task, /* really task inside non-ordered taskq, considered a worksharing type */
1031  ct_task_ordered, /* really task inside ordered taskq, considered a worksharing type */
1032  /* the preceding must be left in order and not split up */
1033 
1034  ct_critical,
1035  ct_ordered_in_parallel,
1036  ct_ordered_in_pdo,
1037  ct_ordered_in_taskq,
1038  ct_master,
1039  ct_reduce,
1040  ct_barrier
1041 };
1042 
1043 /* test to see if we are in a taskq construct */
1044 # define IS_CONS_TYPE_TASKQ( ct ) ( ((int)(ct)) >= ((int)ct_taskq) && ((int)(ct)) <= ((int)ct_task_ordered) )
1045 # define IS_CONS_TYPE_ORDERED( ct ) ((ct) == ct_pdo_ordered || (ct) == ct_task_ordered)
1046 
1047 struct cons_data {
1048  ident_t const *ident;
1049  enum cons_type type;
1050  int prev;
1051  kmp_user_lock_p name; /* address exclusively for critical section name comparison */
1052 };
1053 
1054 struct cons_header {
1055  int p_top, w_top, s_top;
1056  int stack_size, stack_top;
1057  struct cons_data *stack_data;
1058 };
1059 
1060 struct kmp_region_info {
1061  char *text;
1062  int offset[KMP_MAX_FIELDS];
1063  int length[KMP_MAX_FIELDS];
1064 };
1065 
1066 
1067 /* ---------------------------------------------------------------------- */
1068 /* ---------------------------------------------------------------------- */
1069 
1070 #if KMP_OS_WINDOWS
1071  typedef HANDLE kmp_thread_t;
1072  typedef DWORD kmp_key_t;
1073 #endif /* KMP_OS_WINDOWS */
1074 
1075 #if KMP_OS_UNIX
1076  typedef pthread_t kmp_thread_t;
1077  typedef pthread_key_t kmp_key_t;
1078 #endif
1079 
1080 extern kmp_key_t __kmp_gtid_threadprivate_key;
1081 
1082 typedef struct kmp_sys_info {
1083  long maxrss; /* the maximum resident set size utilized (in kilobytes) */
1084  long minflt; /* the number of page faults serviced without any I/O */
1085  long majflt; /* the number of page faults serviced that required I/O */
1086  long nswap; /* the number of times a process was "swapped" out of memory */
1087  long inblock; /* the number of times the file system had to perform input */
1088  long oublock; /* the number of times the file system had to perform output */
1089  long nvcsw; /* the number of times a context switch was voluntarily */
1090  long nivcsw; /* the number of times a context switch was forced */
1091 } kmp_sys_info_t;
1092 
1093 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
1094 typedef struct kmp_cpuinfo {
1095  int initialized; // If 0, other fields are not initialized.
1096  int signature; // CPUID(1).EAX
1097  int family; // CPUID(1).EAX[27:20] + CPUID(1).EAX[11:8] ( Extended Family + Family )
1098  int model; // ( CPUID(1).EAX[19:16] << 4 ) + CPUID(1).EAX[7:4] ( ( Extended Model << 4 ) + Model)
1099  int stepping; // CPUID(1).EAX[3:0] ( Stepping )
1100  int sse2; // 0 if SSE2 instructions are not supported, 1 otherwise.
1101  int rtm; // 0 if RTM instructions are not supported, 1 otherwise.
1102  int cpu_stackoffset;
1103  int apic_id;
1104  int physical_id;
1105  int logical_id;
1106  kmp_uint64 frequency; // Nominal CPU frequency in Hz.
1107  char name [3*sizeof (kmp_cpuid_t)]; // CPUID(0x80000002,0x80000003,0x80000004)
1108 } kmp_cpuinfo_t;
1109 #endif
1110 
1111 #ifdef BUILD_TV
1112 
1113 struct tv_threadprivate {
1114  /* Record type #1 */
1115  void *global_addr;
1116  void *thread_addr;
1117 };
1118 
1119 struct tv_data {
1120  struct tv_data *next;
1121  void *type;
1122  union tv_union {
1123  struct tv_threadprivate tp;
1124  } u;
1125 };
1126 
1127 extern kmp_key_t __kmp_tv_key;
1128 
1129 #endif /* BUILD_TV */
1130 
1131 /* ------------------------------------------------------------------------ */
1132 
1133 #if USE_ITT_BUILD
1134 // We cannot include "kmp_itt.h" due to circular dependency. Declare the only required type here.
1135 // Later we will check the type meets requirements.
1136 typedef int kmp_itt_mark_t;
1137 #define KMP_ITT_DEBUG 0
1138 #endif /* USE_ITT_BUILD */
1139 
1140 /* ------------------------------------------------------------------------ */
1141 
1142 /*
1143  * Taskq data structures
1144  */
1145 
1146 #define HIGH_WATER_MARK(nslots) (((nslots) * 3) / 4)
1147 #define __KMP_TASKQ_THUNKS_PER_TH 1 /* num thunks that each thread can simultaneously execute from a task queue */
1148 
1149 /* flags for taskq_global_flags, kmp_task_queue_t tq_flags, kmpc_thunk_t th_flags */
1150 
1151 #define TQF_IS_ORDERED 0x0001 /* __kmpc_taskq interface, taskq ordered */
1152 #define TQF_IS_LASTPRIVATE 0x0002 /* __kmpc_taskq interface, taskq with lastprivate list */
1153 #define TQF_IS_NOWAIT 0x0004 /* __kmpc_taskq interface, end taskq nowait */
1154 #define TQF_HEURISTICS 0x0008 /* __kmpc_taskq interface, use heuristics to decide task queue size */
1155 #define TQF_INTERFACE_RESERVED1 0x0010 /* __kmpc_taskq interface, reserved for future use */
1156 #define TQF_INTERFACE_RESERVED2 0x0020 /* __kmpc_taskq interface, reserved for future use */
1157 #define TQF_INTERFACE_RESERVED3 0x0040 /* __kmpc_taskq interface, reserved for future use */
1158 #define TQF_INTERFACE_RESERVED4 0x0080 /* __kmpc_taskq interface, reserved for future use */
1159 
1160 #define TQF_INTERFACE_FLAGS 0x00ff /* all the __kmpc_taskq interface flags */
1161 
1162 #define TQF_IS_LAST_TASK 0x0100 /* internal/read by instrumentation; only used with TQF_IS_LASTPRIVATE */
1163 #define TQF_TASKQ_TASK 0x0200 /* internal use only; this thunk->th_task is the taskq_task */
1164 #define TQF_RELEASE_WORKERS 0x0400 /* internal use only; must release worker threads once ANY queued task exists (global) */
1165 #define TQF_ALL_TASKS_QUEUED 0x0800 /* internal use only; notify workers that master has finished enqueuing tasks */
1166 #define TQF_PARALLEL_CONTEXT 0x1000 /* internal use only: this queue encountered in a parallel context: not serialized */
1167 #define TQF_DEALLOCATED 0x2000 /* internal use only; this queue is on the freelist and not in use */
1168 
1169 #define TQF_INTERNAL_FLAGS 0x3f00 /* all the internal use only flags */
1170 
1171 typedef struct KMP_ALIGN_CACHE kmpc_aligned_int32_t {
1172  kmp_int32 ai_data;
1173 } kmpc_aligned_int32_t;
1174 
1175 typedef struct KMP_ALIGN_CACHE kmpc_aligned_queue_slot_t {
1176  struct kmpc_thunk_t *qs_thunk;
1177 } kmpc_aligned_queue_slot_t;
1178 
1179 typedef struct kmpc_task_queue_t {
1180  /* task queue linkage fields for n-ary tree of queues (locked with global taskq_tree_lck) */
1181  kmp_lock_t tq_link_lck; /* lock for child link, child next/prev links and child ref counts */
1182  union {
1183  struct kmpc_task_queue_t *tq_parent; /* pointer to parent taskq, not locked */
1184  struct kmpc_task_queue_t *tq_next_free; /* for taskq internal freelists, locked with global taskq_freelist_lck */
1185  } tq;
1186  volatile struct kmpc_task_queue_t *tq_first_child; /* pointer to linked-list of children, locked by tq's tq_link_lck */
1187  struct kmpc_task_queue_t *tq_next_child; /* next child in linked-list, locked by parent tq's tq_link_lck */
1188  struct kmpc_task_queue_t *tq_prev_child; /* previous child in linked-list, locked by parent tq's tq_link_lck */
1189  volatile kmp_int32 tq_ref_count; /* reference count of threads with access to this task queue */
1190  /* (other than the thread executing the kmpc_end_taskq call) */
1191  /* locked by parent tq's tq_link_lck */
1192 
1193  /* shared data for task queue */
1194  struct kmpc_aligned_shared_vars_t *tq_shareds; /* per-thread array of pointers to shared variable structures */
1195  /* only one array element exists for all but outermost taskq */
1196 
1197  /* bookkeeping for ordered task queue */
1198  kmp_uint32 tq_tasknum_queuing; /* ordered task number assigned while queuing tasks */
1199  volatile kmp_uint32 tq_tasknum_serving; /* ordered number of next task to be served (executed) */
1200 
1201  /* thunk storage management for task queue */
1202  kmp_lock_t tq_free_thunks_lck; /* lock for thunk freelist manipulation */
1203  struct kmpc_thunk_t *tq_free_thunks; /* thunk freelist, chained via th.th_next_free */
1204  struct kmpc_thunk_t *tq_thunk_space; /* space allocated for thunks for this task queue */
1205 
1206  /* data fields for queue itself */
1207  kmp_lock_t tq_queue_lck; /* lock for [de]enqueue operations: tq_queue, tq_head, tq_tail, tq_nfull */
1208  kmpc_aligned_queue_slot_t *tq_queue; /* array of queue slots to hold thunks for tasks */
1209  volatile struct kmpc_thunk_t *tq_taskq_slot; /* special slot for taskq task thunk, occupied if not NULL */
1210  kmp_int32 tq_nslots; /* # of tq_thunk_space thunks alloc'd (not incl. tq_taskq_slot space) */
1211  kmp_int32 tq_head; /* enqueue puts next item in here (index into tq_queue array) */
1212  kmp_int32 tq_tail; /* dequeue takes next item out of here (index into tq_queue array) */
1213  volatile kmp_int32 tq_nfull; /* # of occupied entries in task queue right now */
1214  kmp_int32 tq_hiwat; /* high-water mark for tq_nfull and queue scheduling */
1215  volatile kmp_int32 tq_flags; /* TQF_xxx */
1216 
1217  /* bookkeeping for outstanding thunks */
1218  struct kmpc_aligned_int32_t *tq_th_thunks; /* per-thread array for # of regular thunks currently being executed */
1219  kmp_int32 tq_nproc; /* number of thunks in the th_thunks array */
1220 
1221  /* statistics library bookkeeping */
1222  ident_t *tq_loc; /* source location information for taskq directive */
1223 } kmpc_task_queue_t;
1224 
1225 typedef void (*kmpc_task_t) (kmp_int32 global_tid, struct kmpc_thunk_t *thunk);
1226 
1227 /* sizeof_shareds passed as arg to __kmpc_taskq call */
1228 typedef struct kmpc_shared_vars_t { /* aligned during dynamic allocation */
1229  kmpc_task_queue_t *sv_queue;
1230  /* (pointers to) shared vars */
1231 } kmpc_shared_vars_t;
1232 
1233 typedef struct KMP_ALIGN_CACHE kmpc_aligned_shared_vars_t {
1234  volatile struct kmpc_shared_vars_t *ai_data;
1235 } kmpc_aligned_shared_vars_t;
1236 
1237 /* sizeof_thunk passed as arg to kmpc_taskq call */
1238 typedef struct kmpc_thunk_t { /* aligned during dynamic allocation */
1239  union { /* field used for internal freelists too */
1240  kmpc_shared_vars_t *th_shareds;
1241  struct kmpc_thunk_t *th_next_free; /* freelist of individual thunks within queue, head at tq_free_thunks */
1242  } th;
1243  kmpc_task_t th_task; /* taskq_task if flags & TQF_TASKQ_TASK */
1244  struct kmpc_thunk_t *th_encl_thunk; /* pointer to dynamically enclosing thunk on this thread's call stack */
1245  kmp_int32 th_flags; /* TQF_xxx (tq_flags interface plus possible internal flags) */
1246  kmp_int32 th_status;
1247  kmp_uint32 th_tasknum; /* task number assigned in order of queuing, used for ordered sections */
1248  /* private vars */
1249 } kmpc_thunk_t;
1250 
1251 typedef struct KMP_ALIGN_CACHE kmp_taskq {
1252  int tq_curr_thunk_capacity;
1253 
1254  kmpc_task_queue_t *tq_root;
1255  kmp_int32 tq_global_flags;
1256 
1257  kmp_lock_t tq_freelist_lck;
1258  kmpc_task_queue_t *tq_freelist;
1259 
1260  kmpc_thunk_t **tq_curr_thunk;
1261 } kmp_taskq_t;
1262 
1263 /* END Taskq data structures */
1264 /* --------------------------------------------------------------------------- */
1265 
1266 typedef kmp_int32 kmp_critical_name[8];
1267 
1276 typedef void (*kmpc_micro) ( kmp_int32 * global_tid, kmp_int32 * bound_tid, ... );
1277 typedef void (*kmpc_micro_bound) ( kmp_int32 * bound_tid, kmp_int32 * bound_nth, ... );
1278 
1283 /* --------------------------------------------------------------------------- */
1284 /* Threadprivate initialization/finalization function declarations */
1285 
1286 /* for non-array objects: __kmpc_threadprivate_register() */
1287 
1292 typedef void *(*kmpc_ctor) (void *);
1293 
1298 typedef void (*kmpc_dtor) (void * /*, size_t */); /* 2nd arg: magic number for KCC unused by Intel compiler */
1303 typedef void *(*kmpc_cctor) (void *, void *);
1304 
1305 /* for array objects: __kmpc_threadprivate_register_vec() */
1306  /* First arg: "this" pointer */
1307  /* Last arg: number of array elements */
1313 typedef void *(*kmpc_ctor_vec) (void *, size_t);
1319 typedef void (*kmpc_dtor_vec) (void *, size_t);
1325 typedef void *(*kmpc_cctor_vec) (void *, void *, size_t); /* function unused by compiler */
1326 
1332 /* ------------------------------------------------------------------------ */
1333 
1334 /* keeps tracked of threadprivate cache allocations for cleanup later */
1335 typedef struct kmp_cached_addr {
1336  void **addr; /* address of allocated cache */
1337  struct kmp_cached_addr *next; /* pointer to next cached address */
1338 } kmp_cached_addr_t;
1339 
1340 struct private_data {
1341  struct private_data *next; /* The next descriptor in the list */
1342  void *data; /* The data buffer for this descriptor */
1343  int more; /* The repeat count for this descriptor */
1344  size_t size; /* The data size for this descriptor */
1345 };
1346 
1347 struct private_common {
1348  struct private_common *next;
1349  struct private_common *link;
1350  void *gbl_addr;
1351  void *par_addr; /* par_addr == gbl_addr for MASTER thread */
1352  size_t cmn_size;
1353 };
1354 
1355 struct shared_common
1356 {
1357  struct shared_common *next;
1358  struct private_data *pod_init;
1359  void *obj_init;
1360  void *gbl_addr;
1361  union {
1362  kmpc_ctor ctor;
1363  kmpc_ctor_vec ctorv;
1364  } ct;
1365  union {
1366  kmpc_cctor cctor;
1367  kmpc_cctor_vec cctorv;
1368  } cct;
1369  union {
1370  kmpc_dtor dtor;
1371  kmpc_dtor_vec dtorv;
1372  } dt;
1373  size_t vec_len;
1374  int is_vec;
1375  size_t cmn_size;
1376 };
1377 
1378 #define KMP_HASH_TABLE_LOG2 9 /* log2 of the hash table size */
1379 #define KMP_HASH_TABLE_SIZE (1 << KMP_HASH_TABLE_LOG2) /* size of the hash table */
1380 #define KMP_HASH_SHIFT 3 /* throw away this many low bits from the address */
1381 #define KMP_HASH(x) ((((kmp_uintptr_t) x) >> KMP_HASH_SHIFT) & (KMP_HASH_TABLE_SIZE-1))
1382 
1383 struct common_table {
1384  struct private_common *data[ KMP_HASH_TABLE_SIZE ];
1385 };
1386 
1387 struct shared_table {
1388  struct shared_common *data[ KMP_HASH_TABLE_SIZE ];
1389 };
1390 /* ------------------------------------------------------------------------ */
1391 /* ------------------------------------------------------------------------ */
1392 
1393 #if KMP_STATIC_STEAL_ENABLED
1394 typedef struct KMP_ALIGN_CACHE dispatch_private_info32 {
1395  kmp_int32 count;
1396  kmp_int32 ub;
1397  /* Adding KMP_ALIGN_CACHE here doesn't help / can hurt performance */
1398  kmp_int32 lb;
1399  kmp_int32 st;
1400  kmp_int32 tc;
1401  kmp_int32 static_steal_counter; /* for static_steal only; maybe better to put after ub */
1402 
1403  // KMP_ALIGN( 16 ) ensures ( if the KMP_ALIGN macro is turned on )
1404  // a) parm3 is properly aligned and
1405  // b) all parm1-4 are in the same cache line.
1406  // Because of parm1-4 are used together, performance seems to be better
1407  // if they are in the same line (not measured though).
1408 
1409  struct KMP_ALIGN( 32 ) { // AC: changed 16 to 32 in order to simplify template
1410  kmp_int32 parm1; // structures in kmp_dispatch.cpp. This should
1411  kmp_int32 parm2; // make no real change at least while padding is off.
1412  kmp_int32 parm3;
1413  kmp_int32 parm4;
1414  };
1415 
1416  kmp_uint32 ordered_lower;
1417  kmp_uint32 ordered_upper;
1418 #if KMP_OS_WINDOWS
1419  // This var can be placed in the hole between 'tc' and 'parm1', instead of 'static_steal_counter'.
1420  // It would be nice to measure execution times.
1421  // Conditional if/endif can be removed at all.
1422  kmp_int32 last_upper;
1423 #endif /* KMP_OS_WINDOWS */
1424 } dispatch_private_info32_t;
1425 
1426 typedef struct KMP_ALIGN_CACHE dispatch_private_info64 {
1427  kmp_int64 count; /* current chunk number for static and static-steal scheduling*/
1428  kmp_int64 ub; /* upper-bound */
1429  /* Adding KMP_ALIGN_CACHE here doesn't help / can hurt performance */
1430  kmp_int64 lb; /* lower-bound */
1431  kmp_int64 st; /* stride */
1432  kmp_int64 tc; /* trip count (number of iterations) */
1433  kmp_int64 static_steal_counter; /* for static_steal only; maybe better to put after ub */
1434 
1435  /* parm[1-4] are used in different ways by different scheduling algorithms */
1436 
1437  // KMP_ALIGN( 32 ) ensures ( if the KMP_ALIGN macro is turned on )
1438  // a) parm3 is properly aligned and
1439  // b) all parm1-4 are in the same cache line.
1440  // Because of parm1-4 are used together, performance seems to be better
1441  // if they are in the same line (not measured though).
1442 
1443  struct KMP_ALIGN( 32 ) {
1444  kmp_int64 parm1;
1445  kmp_int64 parm2;
1446  kmp_int64 parm3;
1447  kmp_int64 parm4;
1448  };
1449 
1450  kmp_uint64 ordered_lower;
1451  kmp_uint64 ordered_upper;
1452 #if KMP_OS_WINDOWS
1453  // This var can be placed in the hole between 'tc' and 'parm1', instead of 'static_steal_counter'.
1454  // It would be nice to measure execution times.
1455  // Conditional if/endif can be removed at all.
1456  kmp_int64 last_upper;
1457 #endif /* KMP_OS_WINDOWS */
1458 } dispatch_private_info64_t;
1459 #else /* KMP_STATIC_STEAL_ENABLED */
1460 typedef struct KMP_ALIGN_CACHE dispatch_private_info32 {
1461  kmp_int32 lb;
1462  kmp_int32 ub;
1463  kmp_int32 st;
1464  kmp_int32 tc;
1465 
1466  kmp_int32 parm1;
1467  kmp_int32 parm2;
1468  kmp_int32 parm3;
1469  kmp_int32 parm4;
1470 
1471  kmp_int32 count;
1472 
1473  kmp_uint32 ordered_lower;
1474  kmp_uint32 ordered_upper;
1475 #if KMP_OS_WINDOWS
1476  kmp_int32 last_upper;
1477 #endif /* KMP_OS_WINDOWS */
1478 } dispatch_private_info32_t;
1479 
1480 typedef struct KMP_ALIGN_CACHE dispatch_private_info64 {
1481  kmp_int64 lb; /* lower-bound */
1482  kmp_int64 ub; /* upper-bound */
1483  kmp_int64 st; /* stride */
1484  kmp_int64 tc; /* trip count (number of iterations) */
1485 
1486  /* parm[1-4] are used in different ways by different scheduling algorithms */
1487  kmp_int64 parm1;
1488  kmp_int64 parm2;
1489  kmp_int64 parm3;
1490  kmp_int64 parm4;
1491 
1492  kmp_int64 count; /* current chunk number for static scheduling */
1493 
1494  kmp_uint64 ordered_lower;
1495  kmp_uint64 ordered_upper;
1496 #if KMP_OS_WINDOWS
1497  kmp_int64 last_upper;
1498 #endif /* KMP_OS_WINDOWS */
1499 } dispatch_private_info64_t;
1500 #endif /* KMP_STATIC_STEAL_ENABLED */
1501 
1502 typedef struct KMP_ALIGN_CACHE dispatch_private_info {
1503  union private_info {
1504  dispatch_private_info32_t p32;
1505  dispatch_private_info64_t p64;
1506  } u;
1507  enum sched_type schedule; /* scheduling algorithm */
1508  kmp_int32 ordered; /* ordered clause specified */
1509  kmp_int32 ordered_bumped;
1510  kmp_int32 ordered_dummy[KMP_MAX_ORDERED-3]; // to retain the structure size after making ordered_iteration scalar
1511  struct dispatch_private_info * next; /* stack of buffers for nest of serial regions */
1512  kmp_int32 nomerge; /* don't merge iters if serialized */
1513  kmp_int32 type_size; /* the size of types in private_info */
1514  enum cons_type pushed_ws;
1515 } dispatch_private_info_t;
1516 
1517 typedef struct dispatch_shared_info32 {
1518  /* chunk index under dynamic, number of idle threads under static-steal;
1519  iteration index otherwise */
1520  volatile kmp_uint32 iteration;
1521  volatile kmp_uint32 num_done;
1522  volatile kmp_uint32 ordered_iteration;
1523  kmp_int32 ordered_dummy[KMP_MAX_ORDERED-1]; // to retain the structure size after making ordered_iteration scalar
1524 } dispatch_shared_info32_t;
1525 
1526 typedef struct dispatch_shared_info64 {
1527  /* chunk index under dynamic, number of idle threads under static-steal;
1528  iteration index otherwise */
1529  volatile kmp_uint64 iteration;
1530  volatile kmp_uint64 num_done;
1531  volatile kmp_uint64 ordered_iteration;
1532  kmp_int64 ordered_dummy[KMP_MAX_ORDERED-3]; // to retain the structure size after making ordered_iteration scalar
1533 } dispatch_shared_info64_t;
1534 
1535 typedef struct dispatch_shared_info {
1536  union shared_info {
1537  dispatch_shared_info32_t s32;
1538  dispatch_shared_info64_t s64;
1539  } u;
1540  volatile kmp_uint32 buffer_index;
1541 #if OMP_45_ENABLED
1542  volatile kmp_int32 doacross_buf_idx; // teamwise index
1543  volatile kmp_uint32 *doacross_flags; // shared array of iteration flags (0/1)
1544  kmp_int32 doacross_num_done; // count finished threads
1545 #endif
1546 #if KMP_USE_HWLOC
1547  // When linking with libhwloc, the ORDERED EPCC test slows down on big
1548  // machines (> 48 cores). Performance analysis showed that a cache thrash
1549  // was occurring and this padding helps alleviate the problem.
1550  char padding[64];
1551 #endif
1552 } dispatch_shared_info_t;
1553 
1554 typedef struct kmp_disp {
1555  /* Vector for ORDERED SECTION */
1556  void (*th_deo_fcn)( int * gtid, int * cid, ident_t *);
1557  /* Vector for END ORDERED SECTION */
1558  void (*th_dxo_fcn)( int * gtid, int * cid, ident_t *);
1559 
1560  dispatch_shared_info_t *th_dispatch_sh_current;
1561  dispatch_private_info_t *th_dispatch_pr_current;
1562 
1563  dispatch_private_info_t *th_disp_buffer;
1564  kmp_int32 th_disp_index;
1565 #if OMP_45_ENABLED
1566  kmp_int32 th_doacross_buf_idx; // thread's doacross buffer index
1567  volatile kmp_uint32 *th_doacross_flags; // pointer to shared array of flags
1568  union { // we can use union here because doacross cannot be used in nonmonotonic loops
1569  kmp_int64 *th_doacross_info; // info on loop bounds
1570  kmp_lock_t *th_steal_lock; // lock used for chunk stealing (8-byte variable)
1571  };
1572 #else
1573 #if KMP_STATIC_STEAL_ENABLED
1574  kmp_lock_t *th_steal_lock; // lock used for chunk stealing (8-byte variable)
1575  void* dummy_padding[1]; // make it 64 bytes on Intel(R) 64
1576 #else
1577  void* dummy_padding[2]; // make it 64 bytes on Intel(R) 64
1578 #endif
1579 #endif
1580 #if KMP_USE_INTERNODE_ALIGNMENT
1581  char more_padding[INTERNODE_CACHE_LINE];
1582 #endif
1583 } kmp_disp_t;
1584 
1585 /* ------------------------------------------------------------------------ */
1586 /* ------------------------------------------------------------------------ */
1587 
1588 /* Barrier stuff */
1589 
1590 /* constants for barrier state update */
1591 #define KMP_INIT_BARRIER_STATE 0 /* should probably start from zero */
1592 #define KMP_BARRIER_SLEEP_BIT 0 /* bit used for suspend/sleep part of state */
1593 #define KMP_BARRIER_UNUSED_BIT 1 /* bit that must never be set for valid state */
1594 #define KMP_BARRIER_BUMP_BIT 2 /* lsb used for bump of go/arrived state */
1595 
1596 #define KMP_BARRIER_SLEEP_STATE (1 << KMP_BARRIER_SLEEP_BIT)
1597 #define KMP_BARRIER_UNUSED_STATE (1 << KMP_BARRIER_UNUSED_BIT)
1598 #define KMP_BARRIER_STATE_BUMP (1 << KMP_BARRIER_BUMP_BIT)
1599 
1600 #if (KMP_BARRIER_SLEEP_BIT >= KMP_BARRIER_BUMP_BIT)
1601 # error "Barrier sleep bit must be smaller than barrier bump bit"
1602 #endif
1603 #if (KMP_BARRIER_UNUSED_BIT >= KMP_BARRIER_BUMP_BIT)
1604 # error "Barrier unused bit must be smaller than barrier bump bit"
1605 #endif
1606 
1607 // Constants for release barrier wait state: currently, hierarchical only
1608 #define KMP_BARRIER_NOT_WAITING 0 // Normal state; worker not in wait_sleep
1609 #define KMP_BARRIER_OWN_FLAG 1 // Normal state; worker waiting on own b_go flag in release
1610 #define KMP_BARRIER_PARENT_FLAG 2 // Special state; worker waiting on parent's b_go flag in release
1611 #define KMP_BARRIER_SWITCH_TO_OWN_FLAG 3 // Special state; tells worker to shift from parent to own b_go
1612 #define KMP_BARRIER_SWITCHING 4 // Special state; worker resets appropriate flag on wake-up
1613 
1614 enum barrier_type {
1615  bs_plain_barrier = 0, /* 0, All non-fork/join barriers (except reduction barriers if enabled) */
1616  bs_forkjoin_barrier, /* 1, All fork/join (parallel region) barriers */
1617  #if KMP_FAST_REDUCTION_BARRIER
1618  bs_reduction_barrier, /* 2, All barriers that are used in reduction */
1619  #endif // KMP_FAST_REDUCTION_BARRIER
1620  bs_last_barrier /* Just a placeholder to mark the end */
1621 };
1622 
1623 // to work with reduction barriers just like with plain barriers
1624 #if !KMP_FAST_REDUCTION_BARRIER
1625  #define bs_reduction_barrier bs_plain_barrier
1626 #endif // KMP_FAST_REDUCTION_BARRIER
1627 
1628 typedef enum kmp_bar_pat { /* Barrier communication patterns */
1629  bp_linear_bar = 0, /* Single level (degenerate) tree */
1630  bp_tree_bar = 1, /* Balanced tree with branching factor 2^n */
1631  bp_hyper_bar = 2, /* Hypercube-embedded tree with min branching factor 2^n */
1632  bp_hierarchical_bar = 3, /* Machine hierarchy tree */
1633  bp_last_bar = 4 /* Placeholder to mark the end */
1634 } kmp_bar_pat_e;
1635 
1636 # define KMP_BARRIER_ICV_PUSH 1
1637 
1638 /* Record for holding the values of the internal controls stack records */
1639 typedef struct kmp_internal_control {
1640  int serial_nesting_level; /* corresponds to the value of the th_team_serialized field */
1641  kmp_int8 nested; /* internal control for nested parallelism (per thread) */
1642  kmp_int8 dynamic; /* internal control for dynamic adjustment of threads (per thread) */
1643  kmp_int8 bt_set; /* internal control for whether blocktime is explicitly set */
1644  int blocktime; /* internal control for blocktime */
1645 #if KMP_USE_MONITOR
1646  int bt_intervals; /* internal control for blocktime intervals */
1647 #endif
1648  int nproc; /* internal control for #threads for next parallel region (per thread) */
1649  int max_active_levels; /* internal control for max_active_levels */
1650  kmp_r_sched_t sched; /* internal control for runtime schedule {sched,chunk} pair */
1651 #if OMP_40_ENABLED
1652  kmp_proc_bind_t proc_bind; /* internal control for affinity */
1653  kmp_int32 default_device; /* internal control for default device */
1654 #endif // OMP_40_ENABLED
1655  struct kmp_internal_control *next;
1656 } kmp_internal_control_t;
1657 
1658 static inline void
1659 copy_icvs( kmp_internal_control_t *dst, kmp_internal_control_t *src ) {
1660  *dst = *src;
1661 }
1662 
1663 /* Thread barrier needs volatile barrier fields */
1664 typedef struct KMP_ALIGN_CACHE kmp_bstate {
1665  // th_fixed_icvs is aligned by virtue of kmp_bstate being aligned (and all uses of it).
1666  // It is not explicitly aligned below, because we *don't* want it to be padded -- instead,
1667  // we fit b_go into the same cache line with th_fixed_icvs, enabling NGO cache lines
1668  // stores in the hierarchical barrier.
1669  kmp_internal_control_t th_fixed_icvs; // Initial ICVs for the thread
1670  // Tuck b_go into end of th_fixed_icvs cache line, so it can be stored with same NGO store
1671  volatile kmp_uint64 b_go; // STATE => task should proceed (hierarchical)
1672  KMP_ALIGN_CACHE volatile kmp_uint64 b_arrived; // STATE => task reached synch point.
1673  kmp_uint32 *skip_per_level;
1674  kmp_uint32 my_level;
1675  kmp_int32 parent_tid;
1676  kmp_int32 old_tid;
1677  kmp_uint32 depth;
1678  struct kmp_bstate *parent_bar;
1679  kmp_team_t *team;
1680  kmp_uint64 leaf_state;
1681  kmp_uint32 nproc;
1682  kmp_uint8 base_leaf_kids;
1683  kmp_uint8 leaf_kids;
1684  kmp_uint8 offset;
1685  kmp_uint8 wait_flag;
1686  kmp_uint8 use_oncore_barrier;
1687 #if USE_DEBUGGER
1688  // The following field is intended for the debugger solely. Only the worker thread itself accesses this
1689  // field: the worker increases it by 1 when it arrives to a barrier.
1690  KMP_ALIGN_CACHE kmp_uint b_worker_arrived;
1691 #endif /* USE_DEBUGGER */
1692 } kmp_bstate_t;
1693 
1694 union KMP_ALIGN_CACHE kmp_barrier_union {
1695  double b_align; /* use worst case alignment */
1696  char b_pad[ KMP_PAD(kmp_bstate_t, CACHE_LINE) ];
1697  kmp_bstate_t bb;
1698 };
1699 
1700 typedef union kmp_barrier_union kmp_balign_t;
1701 
1702 /* Team barrier needs only non-volatile arrived counter */
1703 union KMP_ALIGN_CACHE kmp_barrier_team_union {
1704  double b_align; /* use worst case alignment */
1705  char b_pad[ CACHE_LINE ];
1706  struct {
1707  kmp_uint64 b_arrived; /* STATE => task reached synch point. */
1708 #if USE_DEBUGGER
1709  // The following two fields are indended for the debugger solely. Only master of the team accesses
1710  // these fields: the first one is increased by 1 when master arrives to a barrier, the
1711  // second one is increased by one when all the threads arrived.
1712  kmp_uint b_master_arrived;
1713  kmp_uint b_team_arrived;
1714 #endif
1715  };
1716 };
1717 
1718 typedef union kmp_barrier_team_union kmp_balign_team_t;
1719 
1720 /*
1721  * Padding for Linux* OS pthreads condition variables and mutexes used to signal
1722  * threads when a condition changes. This is to workaround an NPTL bug
1723  * where padding was added to pthread_cond_t which caused the initialization
1724  * routine to write outside of the structure if compiled on pre-NPTL threads.
1725  */
1726 
1727 #if KMP_OS_WINDOWS
1728 typedef struct kmp_win32_mutex
1729 {
1730  /* The Lock */
1731  CRITICAL_SECTION cs;
1732 } kmp_win32_mutex_t;
1733 
1734 typedef struct kmp_win32_cond
1735 {
1736  /* Count of the number of waiters. */
1737  int waiters_count_;
1738 
1739  /* Serialize access to <waiters_count_> */
1740  kmp_win32_mutex_t waiters_count_lock_;
1741 
1742  /* Number of threads to release via a <cond_broadcast> or a */
1743  /* <cond_signal> */
1744  int release_count_;
1745 
1746  /* Keeps track of the current "generation" so that we don't allow */
1747  /* one thread to steal all the "releases" from the broadcast. */
1748  int wait_generation_count_;
1749 
1750  /* A manual-reset event that's used to block and release waiting */
1751  /* threads. */
1752  HANDLE event_;
1753 } kmp_win32_cond_t;
1754 #endif
1755 
1756 #if KMP_OS_UNIX
1757 
1758 union KMP_ALIGN_CACHE kmp_cond_union {
1759  double c_align;
1760  char c_pad[ CACHE_LINE ];
1761  pthread_cond_t c_cond;
1762 };
1763 
1764 typedef union kmp_cond_union kmp_cond_align_t;
1765 
1766 union KMP_ALIGN_CACHE kmp_mutex_union {
1767  double m_align;
1768  char m_pad[ CACHE_LINE ];
1769  pthread_mutex_t m_mutex;
1770 };
1771 
1772 typedef union kmp_mutex_union kmp_mutex_align_t;
1773 
1774 #endif /* KMP_OS_UNIX */
1775 
1776 typedef struct kmp_desc_base {
1777  void *ds_stackbase;
1778  size_t ds_stacksize;
1779  int ds_stackgrow;
1780  kmp_thread_t ds_thread;
1781  volatile int ds_tid;
1782  int ds_gtid;
1783 #if KMP_OS_WINDOWS
1784  volatile int ds_alive;
1785  DWORD ds_thread_id;
1786  /*
1787  ds_thread keeps thread handle on Windows* OS. It is enough for RTL purposes. However,
1788  debugger support (libomp_db) cannot work with handles, because they uncomparable. For
1789  example, debugger requests info about thread with handle h. h is valid within debugger
1790  process, and meaningless within debugee process. Even if h is duped by call to
1791  DuplicateHandle(), so the result h' is valid within debugee process, but it is a *new*
1792  handle which does *not* equal to any other handle in debugee... The only way to
1793  compare handles is convert them to system-wide ids. GetThreadId() function is
1794  available only in Longhorn and Server 2003. :-( In contrast, GetCurrentThreadId() is
1795  available on all Windows* OS flavours (including Windows* 95). Thus, we have to get thread id by
1796  call to GetCurrentThreadId() from within the thread and save it to let libomp_db
1797  identify threads.
1798  */
1799 #endif /* KMP_OS_WINDOWS */
1800 } kmp_desc_base_t;
1801 
1802 typedef union KMP_ALIGN_CACHE kmp_desc {
1803  double ds_align; /* use worst case alignment */
1804  char ds_pad[ KMP_PAD(kmp_desc_base_t, CACHE_LINE) ];
1805  kmp_desc_base_t ds;
1806 } kmp_desc_t;
1807 
1808 
1809 typedef struct kmp_local {
1810  volatile int this_construct; /* count of single's encountered by thread */
1811  void *reduce_data;
1812 #if KMP_USE_BGET
1813  void *bget_data;
1814  void *bget_list;
1815 #if ! USE_CMP_XCHG_FOR_BGET
1816 #ifdef USE_QUEUING_LOCK_FOR_BGET
1817  kmp_lock_t bget_lock; /* Lock for accessing bget free list */
1818 #else
1819  kmp_bootstrap_lock_t bget_lock; /* Lock for accessing bget free list */
1820  /* Must be bootstrap lock so we can use it at library shutdown */
1821 #endif /* USE_LOCK_FOR_BGET */
1822 #endif /* ! USE_CMP_XCHG_FOR_BGET */
1823 #endif /* KMP_USE_BGET */
1824 
1825 #ifdef BUILD_TV
1826  struct tv_data *tv_data;
1827 #endif
1828 
1829  PACKED_REDUCTION_METHOD_T packed_reduction_method; /* stored by __kmpc_reduce*(), used by __kmpc_end_reduce*() */
1830 
1831 } kmp_local_t;
1832 
1833 #define KMP_CHECK_UPDATE(a, b) if ((a) != (b)) (a) = (b)
1834 #define KMP_CHECK_UPDATE_SYNC(a, b) if ((a) != (b)) TCW_SYNC_PTR((a), (b))
1835 
1836 #define get__blocktime( xteam, xtid ) ((xteam)->t.t_threads[(xtid)]->th.th_current_task->td_icvs.blocktime)
1837 #define get__bt_set( xteam, xtid ) ((xteam)->t.t_threads[(xtid)]->th.th_current_task->td_icvs.bt_set)
1838 #if KMP_USE_MONITOR
1839 #define get__bt_intervals( xteam, xtid ) ((xteam)->t.t_threads[(xtid)]->th.th_current_task->td_icvs.bt_intervals)
1840 #endif
1841 
1842 #define get__nested_2(xteam,xtid) ((xteam)->t.t_threads[(xtid)]->th.th_current_task->td_icvs.nested)
1843 #define get__dynamic_2(xteam,xtid) ((xteam)->t.t_threads[(xtid)]->th.th_current_task->td_icvs.dynamic)
1844 #define get__nproc_2(xteam,xtid) ((xteam)->t.t_threads[(xtid)]->th.th_current_task->td_icvs.nproc)
1845 #define get__sched_2(xteam,xtid) ((xteam)->t.t_threads[(xtid)]->th.th_current_task->td_icvs.sched)
1846 
1847 #define set__blocktime_team( xteam, xtid, xval ) \
1848  ( ( (xteam)->t.t_threads[(xtid)]->th.th_current_task->td_icvs.blocktime ) = (xval) )
1849 
1850 #if KMP_USE_MONITOR
1851 #define set__bt_intervals_team( xteam, xtid, xval ) \
1852  ( ( (xteam)->t.t_threads[(xtid)]->th.th_current_task->td_icvs.bt_intervals ) = (xval) )
1853 #endif
1854 
1855 #define set__bt_set_team( xteam, xtid, xval ) \
1856  ( ( (xteam)->t.t_threads[(xtid)]->th.th_current_task->td_icvs.bt_set ) = (xval) )
1857 
1858 
1859 #define set__nested( xthread, xval ) \
1860  ( ( (xthread)->th.th_current_task->td_icvs.nested ) = (xval) )
1861 #define get__nested( xthread ) \
1862  ( ( (xthread)->th.th_current_task->td_icvs.nested ) ? (FTN_TRUE) : (FTN_FALSE) )
1863 
1864 #define set__dynamic( xthread, xval ) \
1865  ( ( (xthread)->th.th_current_task->td_icvs.dynamic ) = (xval) )
1866 #define get__dynamic( xthread ) \
1867  ( ( (xthread)->th.th_current_task->td_icvs.dynamic ) ? (FTN_TRUE) : (FTN_FALSE) )
1868 
1869 #define set__nproc( xthread, xval ) \
1870  ( ( (xthread)->th.th_current_task->td_icvs.nproc ) = (xval) )
1871 
1872 #define set__max_active_levels( xthread, xval ) \
1873  ( ( (xthread)->th.th_current_task->td_icvs.max_active_levels ) = (xval) )
1874 
1875 #define set__sched( xthread, xval ) \
1876  ( ( (xthread)->th.th_current_task->td_icvs.sched ) = (xval) )
1877 
1878 #if OMP_40_ENABLED
1879 
1880 #define set__proc_bind( xthread, xval ) \
1881  ( ( (xthread)->th.th_current_task->td_icvs.proc_bind ) = (xval) )
1882 #define get__proc_bind( xthread ) \
1883  ( (xthread)->th.th_current_task->td_icvs.proc_bind )
1884 
1885 #endif /* OMP_40_ENABLED */
1886 
1887 
1888 /* ------------------------------------------------------------------------ */
1889 // OpenMP tasking data structures
1890 //
1891 
1892 typedef enum kmp_tasking_mode {
1893  tskm_immediate_exec = 0,
1894  tskm_extra_barrier = 1,
1895  tskm_task_teams = 2,
1896  tskm_max = 2
1897 } kmp_tasking_mode_t;
1898 
1899 extern kmp_tasking_mode_t __kmp_tasking_mode; /* determines how/when to execute tasks */
1900 extern kmp_int32 __kmp_task_stealing_constraint;
1901 #if OMP_40_ENABLED
1902  extern kmp_int32 __kmp_default_device; // Set via OMP_DEFAULT_DEVICE if specified, defaults to 0 otherwise
1903 #endif
1904 #if OMP_45_ENABLED
1905  extern kmp_int32 __kmp_max_task_priority; // Set via OMP_MAX_TASK_PRIORITY if specified, defaults to 0 otherwise
1906 #endif
1907 
1908 /* NOTE: kmp_taskdata_t and kmp_task_t structures allocated in single block with taskdata first */
1909 #define KMP_TASK_TO_TASKDATA(task) (((kmp_taskdata_t *) task) - 1)
1910 #define KMP_TASKDATA_TO_TASK(taskdata) (kmp_task_t *) (taskdata + 1)
1911 
1912 // The tt_found_tasks flag is a signal to all threads in the team that tasks were spawned and
1913 // queued since the previous barrier release.
1914 #define KMP_TASKING_ENABLED(task_team) \
1915  (TCR_SYNC_4((task_team)->tt.tt_found_tasks) == TRUE)
1916 
1923 typedef kmp_int32 (* kmp_routine_entry_t)( kmp_int32, void * );
1924 
1925 #if OMP_40_ENABLED || OMP_45_ENABLED
1926 typedef union kmp_cmplrdata {
1927 #if OMP_45_ENABLED
1928  kmp_int32 priority;
1929 #endif // OMP_45_ENABLED
1930 #if OMP_40_ENABLED
1931  kmp_routine_entry_t destructors; /* pointer to function to invoke deconstructors of firstprivate C++ objects */
1932 #endif // OMP_40_ENABLED
1933  /* future data */
1934 } kmp_cmplrdata_t;
1935 #endif
1936 
1937 /* sizeof_kmp_task_t passed as arg to kmpc_omp_task call */
1940 typedef struct kmp_task { /* GEH: Shouldn't this be aligned somehow? */
1941  void * shareds;
1942  kmp_routine_entry_t routine;
1943  kmp_int32 part_id;
1944 #if OMP_40_ENABLED || OMP_45_ENABLED
1945  kmp_cmplrdata_t data1; /* Two known optional additions: destructors and priority */
1946  kmp_cmplrdata_t data2; /* Process destructors first, priority second */
1947  /* future data */
1948 #endif
1949  /* private vars */
1950 } kmp_task_t;
1951 
1956 #if OMP_40_ENABLED
1957 typedef struct kmp_taskgroup {
1958  kmp_uint32 count; // number of allocated and not yet complete tasks
1959  kmp_int32 cancel_request; // request for cancellation of this taskgroup
1960  struct kmp_taskgroup *parent; // parent taskgroup
1961 } kmp_taskgroup_t;
1962 
1963 
1964 // forward declarations
1965 typedef union kmp_depnode kmp_depnode_t;
1966 typedef struct kmp_depnode_list kmp_depnode_list_t;
1967 typedef struct kmp_dephash_entry kmp_dephash_entry_t;
1968 
1969 typedef struct kmp_depend_info {
1970  kmp_intptr_t base_addr;
1971  size_t len;
1972  struct {
1973  bool in:1;
1974  bool out:1;
1975  } flags;
1976 } kmp_depend_info_t;
1977 
1978 struct kmp_depnode_list {
1979  kmp_depnode_t * node;
1980  kmp_depnode_list_t * next;
1981 };
1982 
1983 typedef struct kmp_base_depnode {
1984  kmp_depnode_list_t * successors;
1985  kmp_task_t * task;
1986 
1987  kmp_lock_t lock;
1988 
1989 #if KMP_SUPPORT_GRAPH_OUTPUT
1990  kmp_uint32 id;
1991 #endif
1992 
1993  volatile kmp_int32 npredecessors;
1994  volatile kmp_int32 nrefs;
1995 } kmp_base_depnode_t;
1996 
1997 union KMP_ALIGN_CACHE kmp_depnode {
1998  double dn_align; /* use worst case alignment */
1999  char dn_pad[ KMP_PAD(kmp_base_depnode_t, CACHE_LINE) ];
2000  kmp_base_depnode_t dn;
2001 };
2002 
2003 struct kmp_dephash_entry {
2004  kmp_intptr_t addr;
2005  kmp_depnode_t * last_out;
2006  kmp_depnode_list_t * last_ins;
2007  kmp_dephash_entry_t * next_in_bucket;
2008 };
2009 
2010 typedef struct kmp_dephash {
2011  kmp_dephash_entry_t ** buckets;
2012  size_t size;
2013 #ifdef KMP_DEBUG
2014  kmp_uint32 nelements;
2015  kmp_uint32 nconflicts;
2016 #endif
2017 } kmp_dephash_t;
2018 
2019 #endif
2020 
2021 #ifdef BUILD_TIED_TASK_STACK
2022 
2023 /* Tied Task stack definitions */
2024 typedef struct kmp_stack_block {
2025  kmp_taskdata_t * sb_block[ TASK_STACK_BLOCK_SIZE ];
2026  struct kmp_stack_block * sb_next;
2027  struct kmp_stack_block * sb_prev;
2028 } kmp_stack_block_t;
2029 
2030 typedef struct kmp_task_stack {
2031  kmp_stack_block_t ts_first_block; // first block of stack entries
2032  kmp_taskdata_t ** ts_top; // pointer to the top of stack
2033  kmp_int32 ts_entries; // number of entries on the stack
2034 } kmp_task_stack_t;
2035 
2036 #endif // BUILD_TIED_TASK_STACK
2037 
2038 typedef struct kmp_tasking_flags { /* Total struct must be exactly 32 bits */
2039  /* Compiler flags */ /* Total compiler flags must be 16 bits */
2040  unsigned tiedness : 1; /* task is either tied (1) or untied (0) */
2041  unsigned final : 1; /* task is final(1) so execute immediately */
2042  unsigned merged_if0 : 1; /* no __kmpc_task_{begin/complete}_if0 calls in if0 code path */
2043 #if OMP_40_ENABLED
2044  unsigned destructors_thunk : 1; /* set if the compiler creates a thunk to invoke destructors from the runtime */
2045 #if OMP_45_ENABLED
2046  unsigned proxy : 1; /* task is a proxy task (it will be executed outside the context of the RTL) */
2047  unsigned priority_specified :1; /* set if the compiler provides priority setting for the task */
2048  unsigned reserved : 10; /* reserved for compiler use */
2049 #else
2050  unsigned reserved : 12; /* reserved for compiler use */
2051 #endif
2052 #else // OMP_40_ENABLED
2053  unsigned reserved : 13; /* reserved for compiler use */
2054 #endif // OMP_40_ENABLED
2055 
2056  /* Library flags */ /* Total library flags must be 16 bits */
2057  unsigned tasktype : 1; /* task is either explicit(1) or implicit (0) */
2058  unsigned task_serial : 1; /* this task is executed immediately (1) or deferred (0) */
2059  unsigned tasking_ser : 1; /* all tasks in team are either executed immediately (1) or may be deferred (0) */
2060  unsigned team_serial : 1; /* entire team is serial (1) [1 thread] or parallel (0) [>= 2 threads] */
2061  /* If either team_serial or tasking_ser is set, task team may be NULL */
2062  /* Task State Flags: */
2063  unsigned started : 1; /* 1==started, 0==not started */
2064  unsigned executing : 1; /* 1==executing, 0==not executing */
2065  unsigned complete : 1; /* 1==complete, 0==not complete */
2066  unsigned freed : 1; /* 1==freed, 0==allocateed */
2067  unsigned native : 1; /* 1==gcc-compiled task, 0==intel */
2068  unsigned reserved31 : 7; /* reserved for library use */
2069 
2070 } kmp_tasking_flags_t;
2071 
2072 
2073 struct kmp_taskdata { /* aligned during dynamic allocation */
2074  kmp_int32 td_task_id; /* id, assigned by debugger */
2075  kmp_tasking_flags_t td_flags; /* task flags */
2076  kmp_team_t * td_team; /* team for this task */
2077  kmp_info_p * td_alloc_thread; /* thread that allocated data structures */
2078  /* Currently not used except for perhaps IDB */
2079  kmp_taskdata_t * td_parent; /* parent task */
2080  kmp_int32 td_level; /* task nesting level */
2081  kmp_int32 td_untied_count; /* untied task active parts counter */
2082  ident_t * td_ident; /* task identifier */
2083  // Taskwait data.
2084  ident_t * td_taskwait_ident;
2085  kmp_uint32 td_taskwait_counter;
2086  kmp_int32 td_taskwait_thread; /* gtid + 1 of thread encountered taskwait */
2087  KMP_ALIGN_CACHE kmp_internal_control_t td_icvs; /* Internal control variables for the task */
2088  KMP_ALIGN_CACHE volatile kmp_uint32 td_allocated_child_tasks; /* Child tasks (+ current task) not yet deallocated */
2089  volatile kmp_uint32 td_incomplete_child_tasks; /* Child tasks not yet complete */
2090 #if OMP_40_ENABLED
2091  kmp_taskgroup_t * td_taskgroup; // Each task keeps pointer to its current taskgroup
2092  kmp_dephash_t * td_dephash; // Dependencies for children tasks are tracked from here
2093  kmp_depnode_t * td_depnode; // Pointer to graph node if this task has dependencies
2094 #endif
2095 #if OMPT_SUPPORT
2096  ompt_task_info_t ompt_task_info;
2097 #endif
2098 #if OMP_45_ENABLED
2099  kmp_task_team_t * td_task_team;
2100  kmp_int32 td_size_alloc; // The size of task structure, including shareds etc.
2101 #endif
2102 }; // struct kmp_taskdata
2103 
2104 // Make sure padding above worked
2105 KMP_BUILD_ASSERT( sizeof(kmp_taskdata_t) % sizeof(void *) == 0 );
2106 
2107 // Data for task team but per thread
2108 typedef struct kmp_base_thread_data {
2109  kmp_info_p * td_thr; // Pointer back to thread info
2110  // Used only in __kmp_execute_tasks_template, maybe not avail until task is queued?
2111  kmp_bootstrap_lock_t td_deque_lock; // Lock for accessing deque
2112  kmp_taskdata_t ** td_deque; // Deque of tasks encountered by td_thr, dynamically allocated
2113  kmp_int32 td_deque_size; // Size of deck
2114  kmp_uint32 td_deque_head; // Head of deque (will wrap)
2115  kmp_uint32 td_deque_tail; // Tail of deque (will wrap)
2116  kmp_int32 td_deque_ntasks; // Number of tasks in deque
2117  // GEH: shouldn't this be volatile since used in while-spin?
2118  kmp_int32 td_deque_last_stolen; // Thread number of last successful steal
2119 #ifdef BUILD_TIED_TASK_STACK
2120  kmp_task_stack_t td_susp_tied_tasks; // Stack of suspended tied tasks for task scheduling constraint
2121 #endif // BUILD_TIED_TASK_STACK
2122 } kmp_base_thread_data_t;
2123 
2124 #define TASK_DEQUE_BITS 8 // Used solely to define INITIAL_TASK_DEQUE_SIZE
2125 #define INITIAL_TASK_DEQUE_SIZE ( 1 << TASK_DEQUE_BITS )
2126 
2127 #define TASK_DEQUE_SIZE(td) ((td).td_deque_size)
2128 #define TASK_DEQUE_MASK(td) ((td).td_deque_size - 1)
2129 
2130 typedef union KMP_ALIGN_CACHE kmp_thread_data {
2131  kmp_base_thread_data_t td;
2132  double td_align; /* use worst case alignment */
2133  char td_pad[ KMP_PAD(kmp_base_thread_data_t, CACHE_LINE) ];
2134 } kmp_thread_data_t;
2135 
2136 
2137 // Data for task teams which are used when tasking is enabled for the team
2138 typedef struct kmp_base_task_team {
2139  kmp_bootstrap_lock_t tt_threads_lock; /* Lock used to allocate per-thread part of task team */
2140  /* must be bootstrap lock since used at library shutdown*/
2141  kmp_task_team_t * tt_next; /* For linking the task team free list */
2142  kmp_thread_data_t * tt_threads_data; /* Array of per-thread structures for task team */
2143  /* Data survives task team deallocation */
2144  kmp_int32 tt_found_tasks; /* Have we found tasks and queued them while executing this team? */
2145  /* TRUE means tt_threads_data is set up and initialized */
2146  kmp_int32 tt_nproc; /* #threads in team */
2147  kmp_int32 tt_max_threads; /* number of entries allocated for threads_data array */
2148 #if OMP_45_ENABLED
2149  kmp_int32 tt_found_proxy_tasks; /* Have we found proxy tasks since last barrier */
2150 #endif
2151 
2152  KMP_ALIGN_CACHE
2153  volatile kmp_uint32 tt_unfinished_threads; /* #threads still active */
2154 
2155  KMP_ALIGN_CACHE
2156  volatile kmp_uint32 tt_active; /* is the team still actively executing tasks */
2157 } kmp_base_task_team_t;
2158 
2159 union KMP_ALIGN_CACHE kmp_task_team {
2160  kmp_base_task_team_t tt;
2161  double tt_align; /* use worst case alignment */
2162  char tt_pad[ KMP_PAD(kmp_base_task_team_t, CACHE_LINE) ];
2163 };
2164 
2165 #if ( USE_FAST_MEMORY == 3 ) || ( USE_FAST_MEMORY == 5 )
2166 // Free lists keep same-size free memory slots for fast memory allocation routines
2167 typedef struct kmp_free_list {
2168  void *th_free_list_self; // Self-allocated tasks free list
2169  void *th_free_list_sync; // Self-allocated tasks stolen/returned by other threads
2170  void *th_free_list_other; // Non-self free list (to be returned to owner's sync list)
2171 } kmp_free_list_t;
2172 #endif
2173 #if KMP_NESTED_HOT_TEAMS
2174 // Hot teams array keeps hot teams and their sizes for given thread.
2175 // Hot teams are not put in teams pool, and they don't put threads in threads pool.
2176 typedef struct kmp_hot_team_ptr {
2177  kmp_team_p *hot_team; // pointer to hot_team of given nesting level
2178  kmp_int32 hot_team_nth; // number of threads allocated for the hot_team
2179 } kmp_hot_team_ptr_t;
2180 #endif
2181 #if OMP_40_ENABLED
2182 typedef struct kmp_teams_size {
2183  kmp_int32 nteams; // number of teams in a league
2184  kmp_int32 nth; // number of threads in each team of the league
2185 } kmp_teams_size_t;
2186 #endif
2187 
2188 /* ------------------------------------------------------------------------ */
2189 // OpenMP thread data structures
2190 //
2191 
2192 typedef struct KMP_ALIGN_CACHE kmp_base_info {
2193 /*
2194  * Start with the readonly data which is cache aligned and padded.
2195  * this is written before the thread starts working by the master.
2196  * (uber masters may update themselves later)
2197  * (usage does not consider serialized regions)
2198  */
2199  kmp_desc_t th_info;
2200  kmp_team_p *th_team; /* team we belong to */
2201  kmp_root_p *th_root; /* pointer to root of task hierarchy */
2202  kmp_info_p *th_next_pool; /* next available thread in the pool */
2203  kmp_disp_t *th_dispatch; /* thread's dispatch data */
2204  int th_in_pool; /* in thread pool (32 bits for TCR/TCW) */
2205 
2206  /* The following are cached from the team info structure */
2207  /* TODO use these in more places as determined to be needed via profiling */
2208  int th_team_nproc; /* number of threads in a team */
2209  kmp_info_p *th_team_master; /* the team's master thread */
2210  int th_team_serialized; /* team is serialized */
2211 #if OMP_40_ENABLED
2212  microtask_t th_teams_microtask; /* save entry address for teams construct */
2213  int th_teams_level; /* save initial level of teams construct */
2214  /* it is 0 on device but may be any on host */
2215 #endif
2216 
2217  /* The blocktime info is copied from the team struct to the thread sruct */
2218  /* at the start of a barrier, and the values stored in the team are used */
2219  /* at points in the code where the team struct is no longer guaranteed */
2220  /* to exist (from the POV of worker threads). */
2221 #if KMP_USE_MONITOR
2222  int th_team_bt_intervals;
2223 #endif
2224  int th_team_bt_set;
2225 
2226 
2227 #if KMP_AFFINITY_SUPPORTED
2228  kmp_affin_mask_t *th_affin_mask; /* thread's current affinity mask */
2229 #endif
2230 
2231 /*
2232  * The data set by the master at reinit, then R/W by the worker
2233  */
2234  KMP_ALIGN_CACHE int th_set_nproc; /* if > 0, then only use this request for the next fork */
2235 #if KMP_NESTED_HOT_TEAMS
2236  kmp_hot_team_ptr_t *th_hot_teams; /* array of hot teams */
2237 #endif
2238 #if OMP_40_ENABLED
2239  kmp_proc_bind_t th_set_proc_bind; /* if != proc_bind_default, use request for next fork */
2240  kmp_teams_size_t th_teams_size; /* number of teams/threads in teams construct */
2241 # if KMP_AFFINITY_SUPPORTED
2242  int th_current_place; /* place currently bound to */
2243  int th_new_place; /* place to bind to in par reg */
2244  int th_first_place; /* first place in partition */
2245  int th_last_place; /* last place in partition */
2246 # endif
2247 #endif
2248 #if USE_ITT_BUILD
2249  kmp_uint64 th_bar_arrive_time; /* arrival to barrier timestamp */
2250  kmp_uint64 th_bar_min_time; /* minimum arrival time at the barrier */
2251  kmp_uint64 th_frame_time; /* frame timestamp */
2252 #endif /* USE_ITT_BUILD */
2253  kmp_local_t th_local;
2254  struct private_common *th_pri_head;
2255 
2256 /*
2257  * Now the data only used by the worker (after initial allocation)
2258  */
2259  /* TODO the first serial team should actually be stored in the info_t
2260  * structure. this will help reduce initial allocation overhead */
2261  KMP_ALIGN_CACHE kmp_team_p *th_serial_team; /*serialized team held in reserve*/
2262 
2263 #if OMPT_SUPPORT
2264  ompt_thread_info_t ompt_thread_info;
2265 #endif
2266 
2267 /* The following are also read by the master during reinit */
2268  struct common_table *th_pri_common;
2269 
2270  volatile kmp_uint32 th_spin_here; /* thread-local location for spinning */
2271  /* while awaiting queuing lock acquire */
2272 
2273  volatile void *th_sleep_loc; // this points at a kmp_flag<T>
2274 
2275  ident_t *th_ident;
2276  unsigned th_x; // Random number generator data
2277  unsigned th_a; // Random number generator data
2278 
2279 /*
2280  * Tasking-related data for the thread
2281  */
2282  kmp_task_team_t * th_task_team; // Task team struct
2283  kmp_taskdata_t * th_current_task; // Innermost Task being executed
2284  kmp_uint8 th_task_state; // alternating 0/1 for task team identification
2285  kmp_uint8 * th_task_state_memo_stack; // Stack holding memos of th_task_state at nested levels
2286  kmp_uint32 th_task_state_top; // Top element of th_task_state_memo_stack
2287  kmp_uint32 th_task_state_stack_sz; // Size of th_task_state_memo_stack
2288 
2289  /*
2290  * More stuff for keeping track of active/sleeping threads
2291  * (this part is written by the worker thread)
2292  */
2293  kmp_uint8 th_active_in_pool; // included in count of
2294  // #active threads in pool
2295  int th_active; // ! sleeping
2296  // 32 bits for TCR/TCW
2297 
2298  struct cons_header * th_cons; // used for consistency check
2299 
2300 /*
2301  * Add the syncronizing data which is cache aligned and padded.
2302  */
2303  KMP_ALIGN_CACHE kmp_balign_t th_bar[ bs_last_barrier ];
2304 
2305  KMP_ALIGN_CACHE volatile kmp_int32 th_next_waiting; /* gtid+1 of next thread on lock wait queue, 0 if none */
2306 
2307 #if ( USE_FAST_MEMORY == 3 ) || ( USE_FAST_MEMORY == 5 )
2308  #define NUM_LISTS 4
2309  kmp_free_list_t th_free_lists[NUM_LISTS]; // Free lists for fast memory allocation routines
2310 #endif
2311 
2312 #if KMP_OS_WINDOWS
2313  kmp_win32_cond_t th_suspend_cv;
2314  kmp_win32_mutex_t th_suspend_mx;
2315  int th_suspend_init;
2316 #endif
2317 #if KMP_OS_UNIX
2318  kmp_cond_align_t th_suspend_cv;
2319  kmp_mutex_align_t th_suspend_mx;
2320  int th_suspend_init_count;
2321 #endif
2322 
2323 #if USE_ITT_BUILD
2324  kmp_itt_mark_t th_itt_mark_single;
2325  // alignment ???
2326 #endif /* USE_ITT_BUILD */
2327 #if KMP_STATS_ENABLED
2328  kmp_stats_list* th_stats;
2329 #endif
2330 } kmp_base_info_t;
2331 
2332 typedef union KMP_ALIGN_CACHE kmp_info {
2333  double th_align; /* use worst case alignment */
2334  char th_pad[ KMP_PAD(kmp_base_info_t, CACHE_LINE) ];
2335  kmp_base_info_t th;
2336 } kmp_info_t;
2337 
2338 /* ------------------------------------------------------------------------ */
2339 // OpenMP thread team data structures
2340 //
2341 typedef struct kmp_base_data {
2342  volatile kmp_uint32 t_value;
2343 } kmp_base_data_t;
2344 
2345 typedef union KMP_ALIGN_CACHE kmp_sleep_team {
2346  double dt_align; /* use worst case alignment */
2347  char dt_pad[ KMP_PAD(kmp_base_data_t, CACHE_LINE) ];
2348  kmp_base_data_t dt;
2349 } kmp_sleep_team_t;
2350 
2351 typedef union KMP_ALIGN_CACHE kmp_ordered_team {
2352  double dt_align; /* use worst case alignment */
2353  char dt_pad[ KMP_PAD(kmp_base_data_t, CACHE_LINE) ];
2354  kmp_base_data_t dt;
2355 } kmp_ordered_team_t;
2356 
2357 typedef int (*launch_t)( int gtid );
2358 
2359 /* Minimum number of ARGV entries to malloc if necessary */
2360 #define KMP_MIN_MALLOC_ARGV_ENTRIES 100
2361 
2362 // Set up how many argv pointers will fit in cache lines containing t_inline_argv. Historically, we
2363 // have supported at least 96 bytes. Using a larger value for more space between the master write/worker
2364 // read section and read/write by all section seems to buy more performance on EPCC PARALLEL.
2365 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
2366 # define KMP_INLINE_ARGV_BYTES ( 4 * CACHE_LINE - ( ( 3 * KMP_PTR_SKIP + 2 * sizeof(int) + 2 * sizeof(kmp_int8) + sizeof(kmp_int16) + sizeof(kmp_uint32) ) % CACHE_LINE ) )
2367 #else
2368 # define KMP_INLINE_ARGV_BYTES ( 2 * CACHE_LINE - ( ( 3 * KMP_PTR_SKIP + 2 * sizeof(int) ) % CACHE_LINE ) )
2369 #endif
2370 #define KMP_INLINE_ARGV_ENTRIES (int)( KMP_INLINE_ARGV_BYTES / KMP_PTR_SKIP )
2371 
2372 typedef struct KMP_ALIGN_CACHE kmp_base_team {
2373  // Synchronization Data ---------------------------------------------------------------------------------
2374  KMP_ALIGN_CACHE kmp_ordered_team_t t_ordered;
2375  kmp_balign_team_t t_bar[ bs_last_barrier ];
2376  volatile int t_construct; // count of single directive encountered by team
2377  kmp_lock_t t_single_lock; // team specific lock
2378 
2379  // Master only -----------------------------------------------------------------------------------------
2380  KMP_ALIGN_CACHE int t_master_tid; // tid of master in parent team
2381  int t_master_this_cons; // "this_construct" single counter of master in parent team
2382  ident_t *t_ident; // if volatile, have to change too much other crud to volatile too
2383  kmp_team_p *t_parent; // parent team
2384  kmp_team_p *t_next_pool; // next free team in the team pool
2385  kmp_disp_t *t_dispatch; // thread's dispatch data
2386  kmp_task_team_t *t_task_team[2]; // Task team struct; switch between 2
2387 #if OMP_40_ENABLED
2388  kmp_proc_bind_t t_proc_bind; // bind type for par region
2389 #endif // OMP_40_ENABLED
2390 #if USE_ITT_BUILD
2391  kmp_uint64 t_region_time; // region begin timestamp
2392 #endif /* USE_ITT_BUILD */
2393 
2394  // Master write, workers read --------------------------------------------------------------------------
2395  KMP_ALIGN_CACHE void **t_argv;
2396  int t_argc;
2397  int t_nproc; // number of threads in team
2398  microtask_t t_pkfn;
2399  launch_t t_invoke; // procedure to launch the microtask
2400 
2401 #if OMPT_SUPPORT
2402  ompt_team_info_t ompt_team_info;
2403  ompt_lw_taskteam_t *ompt_serialized_team_info;
2404 #endif
2405 
2406 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
2407  kmp_int8 t_fp_control_saved;
2408  kmp_int8 t_pad2b;
2409  kmp_int16 t_x87_fpu_control_word; // FP control regs
2410  kmp_uint32 t_mxcsr;
2411 #endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
2412 
2413  void *t_inline_argv[ KMP_INLINE_ARGV_ENTRIES ];
2414 
2415  KMP_ALIGN_CACHE kmp_info_t **t_threads;
2416  kmp_taskdata_t *t_implicit_task_taskdata; // Taskdata for the thread's implicit task
2417  int t_level; // nested parallel level
2418 
2419  KMP_ALIGN_CACHE int t_max_argc;
2420  int t_max_nproc; // maximum threads this team can handle (dynamicly expandable)
2421  int t_serialized; // levels deep of serialized teams
2422  dispatch_shared_info_t *t_disp_buffer; // buffers for dispatch system
2423  int t_id; // team's id, assigned by debugger.
2424  int t_active_level; // nested active parallel level
2425  kmp_r_sched_t t_sched; // run-time schedule for the team
2426 #if OMP_40_ENABLED && KMP_AFFINITY_SUPPORTED
2427  int t_first_place; // first & last place in parent thread's partition.
2428  int t_last_place; // Restore these values to master after par region.
2429 #endif // OMP_40_ENABLED && KMP_AFFINITY_SUPPORTED
2430  int t_size_changed; // team size was changed?: 0: no, 1: yes, -1: changed via omp_set_num_threads() call
2431 
2432  // Read/write by workers as well -----------------------------------------------------------------------
2433 #if (KMP_ARCH_X86 || KMP_ARCH_X86_64)
2434  // Using CACHE_LINE=64 reduces memory footprint, but causes a big perf regression of epcc 'parallel'
2435  // and 'barrier' on fxe256lin01. This extra padding serves to fix the performance of epcc 'parallel'
2436  // and 'barrier' when CACHE_LINE=64. TODO: investigate more and get rid if this padding.
2437  char dummy_padding[1024];
2438 #endif
2439  KMP_ALIGN_CACHE kmp_internal_control_t *t_control_stack_top; // internal control stack for additional nested teams.
2440  // for SERIALIZED teams nested 2 or more levels deep
2441 #if OMP_40_ENABLED
2442  kmp_int32 t_cancel_request; // typed flag to store request state of cancellation
2443 #endif
2444  int t_master_active; // save on fork, restore on join
2445  kmp_taskq_t t_taskq; // this team's task queue
2446  void *t_copypriv_data; // team specific pointer to copyprivate data array
2447  kmp_uint32 t_copyin_counter;
2448 #if USE_ITT_BUILD
2449  void *t_stack_id; // team specific stack stitching id (for ittnotify)
2450 #endif /* USE_ITT_BUILD */
2451 } kmp_base_team_t;
2452 
2453 union KMP_ALIGN_CACHE kmp_team {
2454  kmp_base_team_t t;
2455  double t_align; /* use worst case alignment */
2456  char t_pad[ KMP_PAD(kmp_base_team_t, CACHE_LINE) ];
2457 };
2458 
2459 
2460 typedef union KMP_ALIGN_CACHE kmp_time_global {
2461  double dt_align; /* use worst case alignment */
2462  char dt_pad[ KMP_PAD(kmp_base_data_t, CACHE_LINE) ];
2463  kmp_base_data_t dt;
2464 } kmp_time_global_t;
2465 
2466 typedef struct kmp_base_global {
2467  /* cache-aligned */
2468  kmp_time_global_t g_time;
2469 
2470  /* non cache-aligned */
2471  volatile int g_abort;
2472  volatile int g_done;
2473 
2474  int g_dynamic;
2475  enum dynamic_mode g_dynamic_mode;
2476 } kmp_base_global_t;
2477 
2478 typedef union KMP_ALIGN_CACHE kmp_global {
2479  kmp_base_global_t g;
2480  double g_align; /* use worst case alignment */
2481  char g_pad[ KMP_PAD(kmp_base_global_t, CACHE_LINE) ];
2482 } kmp_global_t;
2483 
2484 
2485 typedef struct kmp_base_root {
2486  // TODO: GEH - combine r_active with r_in_parallel then r_active == (r_in_parallel>= 0)
2487  // TODO: GEH - then replace r_active with t_active_levels if we can to reduce the synch
2488  // overhead or keeping r_active
2489 
2490  volatile int r_active; /* TRUE if some region in a nest has > 1 thread */
2491  // GEH: This is misnamed, should be r_in_parallel
2492  volatile int r_nested; // TODO: GEH - This is unused, just remove it entirely.
2493  int r_in_parallel; /* keeps a count of active parallel regions per root */
2494  // GEH: This is misnamed, should be r_active_levels
2495  kmp_team_t *r_root_team;
2496  kmp_team_t *r_hot_team;
2497  kmp_info_t *r_uber_thread;
2498  kmp_lock_t r_begin_lock;
2499  volatile int r_begin;
2500  int r_blocktime; /* blocktime for this root and descendants */
2501 } kmp_base_root_t;
2502 
2503 typedef union KMP_ALIGN_CACHE kmp_root {
2504  kmp_base_root_t r;
2505  double r_align; /* use worst case alignment */
2506  char r_pad[ KMP_PAD(kmp_base_root_t, CACHE_LINE) ];
2507 } kmp_root_t;
2508 
2509 struct fortran_inx_info {
2510  kmp_int32 data;
2511 };
2512 
2513 /* ------------------------------------------------------------------------ */
2514 
2515 /* ------------------------------------------------------------------------ */
2516 /* ------------------------------------------------------------------------ */
2517 
2518 extern int __kmp_settings;
2519 extern int __kmp_duplicate_library_ok;
2520 #if USE_ITT_BUILD
2521 extern int __kmp_forkjoin_frames;
2522 extern int __kmp_forkjoin_frames_mode;
2523 #endif
2524 extern PACKED_REDUCTION_METHOD_T __kmp_force_reduction_method;
2525 extern int __kmp_determ_red;
2526 
2527 #ifdef KMP_DEBUG
2528 extern int kmp_a_debug;
2529 extern int kmp_b_debug;
2530 extern int kmp_c_debug;
2531 extern int kmp_d_debug;
2532 extern int kmp_e_debug;
2533 extern int kmp_f_debug;
2534 #endif /* KMP_DEBUG */
2535 
2536 /* For debug information logging using rotating buffer */
2537 #define KMP_DEBUG_BUF_LINES_INIT 512
2538 #define KMP_DEBUG_BUF_LINES_MIN 1
2539 
2540 #define KMP_DEBUG_BUF_CHARS_INIT 128
2541 #define KMP_DEBUG_BUF_CHARS_MIN 2
2542 
2543 extern int __kmp_debug_buf; /* TRUE means use buffer, FALSE means print to stderr */
2544 extern int __kmp_debug_buf_lines; /* How many lines of debug stored in buffer */
2545 extern int __kmp_debug_buf_chars; /* How many characters allowed per line in buffer */
2546 extern int __kmp_debug_buf_atomic; /* TRUE means use atomic update of buffer entry pointer */
2547 
2548 extern char *__kmp_debug_buffer; /* Debug buffer itself */
2549 extern int __kmp_debug_count; /* Counter for number of lines printed in buffer so far */
2550 extern int __kmp_debug_buf_warn_chars; /* Keep track of char increase recommended in warnings */
2551 /* end rotating debug buffer */
2552 
2553 #ifdef KMP_DEBUG
2554 extern int __kmp_par_range; /* +1 => only go par for constructs in range */
2555 
2556 #define KMP_PAR_RANGE_ROUTINE_LEN 1024
2557 extern char __kmp_par_range_routine[KMP_PAR_RANGE_ROUTINE_LEN];
2558 #define KMP_PAR_RANGE_FILENAME_LEN 1024
2559 extern char __kmp_par_range_filename[KMP_PAR_RANGE_FILENAME_LEN];
2560 extern int __kmp_par_range_lb;
2561 extern int __kmp_par_range_ub;
2562 #endif
2563 
2564 /* For printing out dynamic storage map for threads and teams */
2565 extern int __kmp_storage_map; /* True means print storage map for threads and teams */
2566 extern int __kmp_storage_map_verbose; /* True means storage map includes placement info */
2567 extern int __kmp_storage_map_verbose_specified;
2568 
2569 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
2570 extern kmp_cpuinfo_t __kmp_cpuinfo;
2571 #endif
2572 
2573 extern volatile int __kmp_init_serial;
2574 extern volatile int __kmp_init_gtid;
2575 extern volatile int __kmp_init_common;
2576 extern volatile int __kmp_init_middle;
2577 extern volatile int __kmp_init_parallel;
2578 #if KMP_USE_MONITOR
2579 extern volatile int __kmp_init_monitor;
2580 #endif
2581 extern volatile int __kmp_init_user_locks;
2582 extern int __kmp_init_counter;
2583 extern int __kmp_root_counter;
2584 extern int __kmp_version;
2585 
2586 /* list of address of allocated caches for commons */
2587 extern kmp_cached_addr_t *__kmp_threadpriv_cache_list;
2588 
2589 /* Barrier algorithm types and options */
2590 extern kmp_uint32 __kmp_barrier_gather_bb_dflt;
2591 extern kmp_uint32 __kmp_barrier_release_bb_dflt;
2592 extern kmp_bar_pat_e __kmp_barrier_gather_pat_dflt;
2593 extern kmp_bar_pat_e __kmp_barrier_release_pat_dflt;
2594 extern kmp_uint32 __kmp_barrier_gather_branch_bits [ bs_last_barrier ];
2595 extern kmp_uint32 __kmp_barrier_release_branch_bits [ bs_last_barrier ];
2596 extern kmp_bar_pat_e __kmp_barrier_gather_pattern [ bs_last_barrier ];
2597 extern kmp_bar_pat_e __kmp_barrier_release_pattern [ bs_last_barrier ];
2598 extern char const *__kmp_barrier_branch_bit_env_name [ bs_last_barrier ];
2599 extern char const *__kmp_barrier_pattern_env_name [ bs_last_barrier ];
2600 extern char const *__kmp_barrier_type_name [ bs_last_barrier ];
2601 extern char const *__kmp_barrier_pattern_name [ bp_last_bar ];
2602 
2603 /* Global Locks */
2604 extern kmp_bootstrap_lock_t __kmp_initz_lock; /* control initialization */
2605 extern kmp_bootstrap_lock_t __kmp_forkjoin_lock; /* control fork/join access */
2606 extern kmp_bootstrap_lock_t __kmp_exit_lock; /* exit() is not always thread-safe */
2607 #if KMP_USE_MONITOR
2608 extern kmp_bootstrap_lock_t __kmp_monitor_lock; /* control monitor thread creation */
2609 #endif
2610 extern kmp_bootstrap_lock_t __kmp_tp_cached_lock; /* used for the hack to allow threadprivate cache and __kmp_threads expansion to co-exist */
2611 
2612 extern kmp_lock_t __kmp_global_lock; /* control OS/global access */
2613 extern kmp_queuing_lock_t __kmp_dispatch_lock; /* control dispatch access */
2614 extern kmp_lock_t __kmp_debug_lock; /* control I/O access for KMP_DEBUG */
2615 
2616 /* used for yielding spin-waits */
2617 extern unsigned int __kmp_init_wait; /* initial number of spin-tests */
2618 extern unsigned int __kmp_next_wait; /* susequent number of spin-tests */
2619 
2620 extern enum library_type __kmp_library;
2621 
2622 extern enum sched_type __kmp_sched; /* default runtime scheduling */
2623 extern enum sched_type __kmp_static; /* default static scheduling method */
2624 extern enum sched_type __kmp_guided; /* default guided scheduling method */
2625 extern enum sched_type __kmp_auto; /* default auto scheduling method */
2626 extern int __kmp_chunk; /* default runtime chunk size */
2627 
2628 extern size_t __kmp_stksize; /* stack size per thread */
2629 #if KMP_USE_MONITOR
2630 extern size_t __kmp_monitor_stksize;/* stack size for monitor thread */
2631 #endif
2632 extern size_t __kmp_stkoffset; /* stack offset per thread */
2633 extern int __kmp_stkpadding; /* Should we pad root thread(s) stack */
2634 
2635 extern size_t __kmp_malloc_pool_incr; /* incremental size of pool for kmp_malloc() */
2636 extern int __kmp_env_chunk; /* was KMP_CHUNK specified? */
2637 extern int __kmp_env_stksize; /* was KMP_STACKSIZE specified? */
2638 extern int __kmp_env_omp_stksize;/* was OMP_STACKSIZE specified? */
2639 extern int __kmp_env_all_threads; /* was KMP_ALL_THREADS or KMP_MAX_THREADS specified? */
2640 extern int __kmp_env_omp_all_threads;/* was OMP_THREAD_LIMIT specified? */
2641 extern int __kmp_env_blocktime; /* was KMP_BLOCKTIME specified? */
2642 extern int __kmp_env_checks; /* was KMP_CHECKS specified? */
2643 extern int __kmp_env_consistency_check; /* was KMP_CONSISTENCY_CHECK specified? */
2644 extern int __kmp_generate_warnings; /* should we issue warnings? */
2645 extern int __kmp_reserve_warn; /* have we issued reserve_threads warning? */
2646 
2647 #ifdef DEBUG_SUSPEND
2648 extern int __kmp_suspend_count; /* count inside __kmp_suspend_template() */
2649 #endif
2650 
2651 extern kmp_uint32 __kmp_yield_init;
2652 extern kmp_uint32 __kmp_yield_next;
2653 
2654 #if KMP_USE_MONITOR
2655 extern kmp_uint32 __kmp_yielding_on;
2656 extern kmp_uint32 __kmp_yield_cycle;
2657 extern kmp_int32 __kmp_yield_on_count;
2658 extern kmp_int32 __kmp_yield_off_count;
2659 #endif
2660 
2661 /* ------------------------------------------------------------------------- */
2662 extern int __kmp_allThreadsSpecified;
2663 
2664 extern size_t __kmp_align_alloc;
2665 /* following data protected by initialization routines */
2666 extern int __kmp_xproc; /* number of processors in the system */
2667 extern int __kmp_avail_proc; /* number of processors available to the process */
2668 extern size_t __kmp_sys_min_stksize; /* system-defined minimum stack size */
2669 extern int __kmp_sys_max_nth; /* system-imposed maximum number of threads */
2670 extern int __kmp_max_nth; /* maximum total number of concurrently-existing threads */
2671 extern int __kmp_threads_capacity; /* capacity of the arrays __kmp_threads and __kmp_root */
2672 extern int __kmp_dflt_team_nth; /* default number of threads in a parallel region a la OMP_NUM_THREADS */
2673 extern int __kmp_dflt_team_nth_ub; /* upper bound on "" determined at serial initialization */
2674 extern int __kmp_tp_capacity; /* capacity of __kmp_threads if threadprivate is used (fixed) */
2675 extern int __kmp_tp_cached; /* whether threadprivate cache has been created (__kmpc_threadprivate_cached()) */
2676 extern int __kmp_dflt_nested; /* nested parallelism enabled by default a la OMP_NESTED */
2677 extern int __kmp_dflt_blocktime; /* number of milliseconds to wait before blocking (env setting) */
2678 #if KMP_USE_MONITOR
2679 extern int __kmp_monitor_wakeups;/* number of times monitor wakes up per second */
2680 extern int __kmp_bt_intervals; /* number of monitor timestamp intervals before blocking */
2681 #endif
2682 #ifdef KMP_ADJUST_BLOCKTIME
2683 extern int __kmp_zero_bt; /* whether blocktime has been forced to zero */
2684 #endif /* KMP_ADJUST_BLOCKTIME */
2685 #ifdef KMP_DFLT_NTH_CORES
2686 extern int __kmp_ncores; /* Total number of cores for threads placement */
2687 #endif
2688 extern int __kmp_abort_delay; /* Number of millisecs to delay on abort for VTune */
2689 
2690 extern int __kmp_need_register_atfork_specified;
2691 extern int __kmp_need_register_atfork;/* At initialization, call pthread_atfork to install fork handler */
2692 extern int __kmp_gtid_mode; /* Method of getting gtid, values:
2693  0 - not set, will be set at runtime
2694  1 - using stack search
2695  2 - dynamic TLS (pthread_getspecific(Linux* OS/OS X*) or TlsGetValue(Windows* OS))
2696  3 - static TLS (__declspec(thread) __kmp_gtid), Linux* OS .so only.
2697  */
2698 extern int __kmp_adjust_gtid_mode; /* If true, adjust method based on #threads */
2699 #ifdef KMP_TDATA_GTID
2700 #if KMP_OS_WINDOWS
2701 extern __declspec(thread) int __kmp_gtid; /* This thread's gtid, if __kmp_gtid_mode == 3 */
2702 #else
2703 extern __thread int __kmp_gtid;
2704 #endif /* KMP_OS_WINDOWS - workaround because Intel(R) Many Integrated Core compiler 20110316 doesn't accept __declspec */
2705 #endif
2706 extern int __kmp_tls_gtid_min; /* #threads below which use sp search for gtid */
2707 extern int __kmp_foreign_tp; /* If true, separate TP var for each foreign thread */
2708 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
2709 extern int __kmp_inherit_fp_control; /* copy fp creg(s) parent->workers at fork */
2710 extern kmp_int16 __kmp_init_x87_fpu_control_word; /* init thread's FP control reg */
2711 extern kmp_uint32 __kmp_init_mxcsr; /* init thread's mxscr */
2712 #endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
2713 
2714 extern int __kmp_dflt_max_active_levels; /* max_active_levels for nested parallelism enabled by default a la OMP_MAX_ACTIVE_LEVELS */
2715 extern int __kmp_dispatch_num_buffers; /* max possible dynamic loops in concurrent execution per team */
2716 #if KMP_NESTED_HOT_TEAMS
2717 extern int __kmp_hot_teams_mode;
2718 extern int __kmp_hot_teams_max_level;
2719 #endif
2720 
2721 # if KMP_OS_LINUX
2722 extern enum clock_function_type __kmp_clock_function;
2723 extern int __kmp_clock_function_param;
2724 # endif /* KMP_OS_LINUX */
2725 
2726 #if KMP_ARCH_X86_64 && (KMP_OS_LINUX || KMP_OS_WINDOWS)
2727 extern enum mic_type __kmp_mic_type;
2728 #endif
2729 
2730 # ifdef USE_LOAD_BALANCE
2731 extern double __kmp_load_balance_interval; /* Interval for the load balance algorithm */
2732 # endif /* USE_LOAD_BALANCE */
2733 
2734 // OpenMP 3.1 - Nested num threads array
2735 typedef struct kmp_nested_nthreads_t {
2736  int * nth;
2737  int size;
2738  int used;
2739 } kmp_nested_nthreads_t;
2740 
2741 extern kmp_nested_nthreads_t __kmp_nested_nth;
2742 
2743 #if KMP_USE_ADAPTIVE_LOCKS
2744 
2745 // Parameters for the speculative lock backoff system.
2746 struct kmp_adaptive_backoff_params_t {
2747  // Number of soft retries before it counts as a hard retry.
2748  kmp_uint32 max_soft_retries;
2749  // Badness is a bit mask : 0,1,3,7,15,... on each hard failure we move one to the right
2750  kmp_uint32 max_badness;
2751 };
2752 
2753 extern kmp_adaptive_backoff_params_t __kmp_adaptive_backoff_params;
2754 
2755 #if KMP_DEBUG_ADAPTIVE_LOCKS
2756 extern char * __kmp_speculative_statsfile;
2757 #endif
2758 
2759 #endif // KMP_USE_ADAPTIVE_LOCKS
2760 
2761 #if OMP_40_ENABLED
2762 extern int __kmp_display_env; /* TRUE or FALSE */
2763 extern int __kmp_display_env_verbose; /* TRUE if OMP_DISPLAY_ENV=VERBOSE */
2764 extern int __kmp_omp_cancellation; /* TRUE or FALSE */
2765 #endif
2766 
2767 /* ------------------------------------------------------------------------- */
2768 
2769 /* --------------------------------------------------------------------------- */
2770 /* the following are protected by the fork/join lock */
2771 /* write: lock read: anytime */
2772 extern kmp_info_t **__kmp_threads; /* Descriptors for the threads */
2773 /* read/write: lock */
2774 extern volatile kmp_team_t * __kmp_team_pool;
2775 extern volatile kmp_info_t * __kmp_thread_pool;
2776 
2777 /* total number of threads reachable from some root thread including all root threads*/
2778 extern volatile int __kmp_nth;
2779 /* total number of threads reachable from some root thread including all root threads,
2780  and those in the thread pool */
2781 extern volatile int __kmp_all_nth;
2782 extern int __kmp_thread_pool_nth;
2783 extern volatile int __kmp_thread_pool_active_nth;
2784 
2785 extern kmp_root_t **__kmp_root; /* root of thread hierarchy */
2786 /* end data protected by fork/join lock */
2787 /* --------------------------------------------------------------------------- */
2788 
2789 extern kmp_global_t __kmp_global; /* global status */
2790 
2791 extern kmp_info_t __kmp_monitor;
2792 extern volatile kmp_uint32 __kmp_team_counter; // Used by Debugging Support Library.
2793 extern volatile kmp_uint32 __kmp_task_counter; // Used by Debugging Support Library.
2794 
2795 #if USE_DEBUGGER
2796 
2797 #define _KMP_GEN_ID( counter ) \
2798  ( \
2799  __kmp_debugging \
2800  ? \
2801  KMP_TEST_THEN_INC32( (volatile kmp_int32 *) & counter ) + 1 \
2802  : \
2803  ~ 0 \
2804  )
2805 #else
2806 #define _KMP_GEN_ID( counter ) \
2807  ( \
2808  ~ 0 \
2809  )
2810 #endif /* USE_DEBUGGER */
2811 
2812 #define KMP_GEN_TASK_ID() _KMP_GEN_ID( __kmp_task_counter )
2813 #define KMP_GEN_TEAM_ID() _KMP_GEN_ID( __kmp_team_counter )
2814 
2815 /* ------------------------------------------------------------------------ */
2816 /* ------------------------------------------------------------------------ */
2817 
2818 extern void __kmp_print_storage_map_gtid( int gtid, void *p1, void* p2, size_t size, char const *format, ... );
2819 
2820 extern void __kmp_serial_initialize( void );
2821 extern void __kmp_middle_initialize( void );
2822 extern void __kmp_parallel_initialize( void );
2823 
2824 extern void __kmp_internal_begin( void );
2825 extern void __kmp_internal_end_library( int gtid );
2826 extern void __kmp_internal_end_thread( int gtid );
2827 extern void __kmp_internal_end_atexit( void );
2828 extern void __kmp_internal_end_fini( void );
2829 extern void __kmp_internal_end_dtor( void );
2830 extern void __kmp_internal_end_dest( void* );
2831 
2832 extern int __kmp_register_root( int initial_thread );
2833 extern void __kmp_unregister_root( int gtid );
2834 
2835 extern int __kmp_ignore_mppbeg( void );
2836 extern int __kmp_ignore_mppend( void );
2837 
2838 extern int __kmp_enter_single( int gtid, ident_t *id_ref, int push_ws );
2839 extern void __kmp_exit_single( int gtid );
2840 
2841 extern void __kmp_parallel_deo( int *gtid_ref, int *cid_ref, ident_t *loc_ref );
2842 extern void __kmp_parallel_dxo( int *gtid_ref, int *cid_ref, ident_t *loc_ref );
2843 
2844 #ifdef USE_LOAD_BALANCE
2845 extern int __kmp_get_load_balance( int );
2846 #endif
2847 
2848 #ifdef BUILD_TV
2849 extern void __kmp_tv_threadprivate_store( kmp_info_t *th, void *global_addr, void *thread_addr );
2850 #endif
2851 
2852 extern int __kmp_get_global_thread_id( void );
2853 extern int __kmp_get_global_thread_id_reg( void );
2854 extern void __kmp_exit_thread( int exit_status );
2855 extern void __kmp_abort( char const * format, ... );
2856 extern void __kmp_abort_thread( void );
2857 extern void __kmp_abort_process( void );
2858 extern void __kmp_warn( char const * format, ... );
2859 
2860 extern void __kmp_set_num_threads( int new_nth, int gtid );
2861 
2862 // Returns current thread (pointer to kmp_info_t). Current thread *must* be registered.
2863 static inline kmp_info_t * __kmp_entry_thread()
2864 {
2865  int gtid = __kmp_entry_gtid();
2866 
2867  return __kmp_threads[gtid];
2868 }
2869 
2870 extern void __kmp_set_max_active_levels( int gtid, int new_max_active_levels );
2871 extern int __kmp_get_max_active_levels( int gtid );
2872 extern int __kmp_get_ancestor_thread_num( int gtid, int level );
2873 extern int __kmp_get_team_size( int gtid, int level );
2874 extern void __kmp_set_schedule( int gtid, kmp_sched_t new_sched, int chunk );
2875 extern void __kmp_get_schedule( int gtid, kmp_sched_t * sched, int * chunk );
2876 
2877 extern unsigned short __kmp_get_random( kmp_info_t * thread );
2878 extern void __kmp_init_random( kmp_info_t * thread );
2879 
2880 extern kmp_r_sched_t __kmp_get_schedule_global( void );
2881 extern void __kmp_adjust_num_threads( int new_nproc );
2882 
2883 extern void * ___kmp_allocate( size_t size KMP_SRC_LOC_DECL );
2884 extern void * ___kmp_page_allocate( size_t size KMP_SRC_LOC_DECL );
2885 extern void ___kmp_free( void * ptr KMP_SRC_LOC_DECL );
2886 #define __kmp_allocate( size ) ___kmp_allocate( (size) KMP_SRC_LOC_CURR )
2887 #define __kmp_page_allocate( size ) ___kmp_page_allocate( (size) KMP_SRC_LOC_CURR )
2888 #define __kmp_free( ptr ) ___kmp_free( (ptr) KMP_SRC_LOC_CURR )
2889 
2890 #if USE_FAST_MEMORY
2891 extern void * ___kmp_fast_allocate( kmp_info_t *this_thr, size_t size KMP_SRC_LOC_DECL );
2892 extern void ___kmp_fast_free( kmp_info_t *this_thr, void *ptr KMP_SRC_LOC_DECL );
2893 extern void __kmp_free_fast_memory( kmp_info_t *this_thr );
2894 extern void __kmp_initialize_fast_memory( kmp_info_t *this_thr );
2895 #define __kmp_fast_allocate( this_thr, size ) ___kmp_fast_allocate( (this_thr), (size) KMP_SRC_LOC_CURR )
2896 #define __kmp_fast_free( this_thr, ptr ) ___kmp_fast_free( (this_thr), (ptr) KMP_SRC_LOC_CURR )
2897 #endif
2898 
2899 extern void * ___kmp_thread_malloc( kmp_info_t *th, size_t size KMP_SRC_LOC_DECL );
2900 extern void * ___kmp_thread_calloc( kmp_info_t *th, size_t nelem, size_t elsize KMP_SRC_LOC_DECL );
2901 extern void * ___kmp_thread_realloc( kmp_info_t *th, void *ptr, size_t size KMP_SRC_LOC_DECL );
2902 extern void ___kmp_thread_free( kmp_info_t *th, void *ptr KMP_SRC_LOC_DECL );
2903 #define __kmp_thread_malloc( th, size ) ___kmp_thread_malloc( (th), (size) KMP_SRC_LOC_CURR )
2904 #define __kmp_thread_calloc( th, nelem, elsize ) ___kmp_thread_calloc( (th), (nelem), (elsize) KMP_SRC_LOC_CURR )
2905 #define __kmp_thread_realloc( th, ptr, size ) ___kmp_thread_realloc( (th), (ptr), (size) KMP_SRC_LOC_CURR )
2906 #define __kmp_thread_free( th, ptr ) ___kmp_thread_free( (th), (ptr) KMP_SRC_LOC_CURR )
2907 
2908 #define KMP_INTERNAL_MALLOC(sz) malloc(sz)
2909 #define KMP_INTERNAL_FREE(p) free(p)
2910 #define KMP_INTERNAL_REALLOC(p,sz) realloc((p),(sz))
2911 #define KMP_INTERNAL_CALLOC(n,sz) calloc((n),(sz))
2912 
2913 extern void __kmp_push_num_threads( ident_t *loc, int gtid, int num_threads );
2914 
2915 #if OMP_40_ENABLED
2916 extern void __kmp_push_proc_bind( ident_t *loc, int gtid, kmp_proc_bind_t proc_bind );
2917 extern void __kmp_push_num_teams( ident_t *loc, int gtid, int num_teams, int num_threads );
2918 #endif
2919 
2920 extern void __kmp_yield( int cond );
2921 
2922 extern void __kmpc_dispatch_init_4( ident_t *loc, kmp_int32 gtid,
2923  enum sched_type schedule, kmp_int32 lb, kmp_int32 ub, kmp_int32 st,
2924  kmp_int32 chunk );
2925 extern void __kmpc_dispatch_init_4u( ident_t *loc, kmp_int32 gtid,
2926  enum sched_type schedule, kmp_uint32 lb, kmp_uint32 ub, kmp_int32 st,
2927  kmp_int32 chunk );
2928 extern void __kmpc_dispatch_init_8( ident_t *loc, kmp_int32 gtid,
2929  enum sched_type schedule, kmp_int64 lb, kmp_int64 ub, kmp_int64 st,
2930  kmp_int64 chunk );
2931 extern void __kmpc_dispatch_init_8u( ident_t *loc, kmp_int32 gtid,
2932  enum sched_type schedule, kmp_uint64 lb, kmp_uint64 ub, kmp_int64 st,
2933  kmp_int64 chunk );
2934 
2935 extern int __kmpc_dispatch_next_4( ident_t *loc, kmp_int32 gtid,
2936  kmp_int32 *p_last, kmp_int32 *p_lb, kmp_int32 *p_ub, kmp_int32 *p_st );
2937 extern int __kmpc_dispatch_next_4u( ident_t *loc, kmp_int32 gtid,
2938  kmp_int32 *p_last, kmp_uint32 *p_lb, kmp_uint32 *p_ub, kmp_int32 *p_st );
2939 extern int __kmpc_dispatch_next_8( ident_t *loc, kmp_int32 gtid,
2940  kmp_int32 *p_last, kmp_int64 *p_lb, kmp_int64 *p_ub, kmp_int64 *p_st );
2941 extern int __kmpc_dispatch_next_8u( ident_t *loc, kmp_int32 gtid,
2942  kmp_int32 *p_last, kmp_uint64 *p_lb, kmp_uint64 *p_ub, kmp_int64 *p_st );
2943 
2944 extern void __kmpc_dispatch_fini_4( ident_t *loc, kmp_int32 gtid );
2945 extern void __kmpc_dispatch_fini_8( ident_t *loc, kmp_int32 gtid );
2946 extern void __kmpc_dispatch_fini_4u( ident_t *loc, kmp_int32 gtid );
2947 extern void __kmpc_dispatch_fini_8u( ident_t *loc, kmp_int32 gtid );
2948 
2949 
2950 #ifdef KMP_GOMP_COMPAT
2951 
2952 extern void __kmp_aux_dispatch_init_4( ident_t *loc, kmp_int32 gtid,
2953  enum sched_type schedule, kmp_int32 lb, kmp_int32 ub, kmp_int32 st,
2954  kmp_int32 chunk, int push_ws );
2955 extern void __kmp_aux_dispatch_init_4u( ident_t *loc, kmp_int32 gtid,
2956  enum sched_type schedule, kmp_uint32 lb, kmp_uint32 ub, kmp_int32 st,
2957  kmp_int32 chunk, int push_ws );
2958 extern void __kmp_aux_dispatch_init_8( ident_t *loc, kmp_int32 gtid,
2959  enum sched_type schedule, kmp_int64 lb, kmp_int64 ub, kmp_int64 st,
2960  kmp_int64 chunk, int push_ws );
2961 extern void __kmp_aux_dispatch_init_8u( ident_t *loc, kmp_int32 gtid,
2962  enum sched_type schedule, kmp_uint64 lb, kmp_uint64 ub, kmp_int64 st,
2963  kmp_int64 chunk, int push_ws );
2964 extern void __kmp_aux_dispatch_fini_chunk_4( ident_t *loc, kmp_int32 gtid );
2965 extern void __kmp_aux_dispatch_fini_chunk_8( ident_t *loc, kmp_int32 gtid );
2966 extern void __kmp_aux_dispatch_fini_chunk_4u( ident_t *loc, kmp_int32 gtid );
2967 extern void __kmp_aux_dispatch_fini_chunk_8u( ident_t *loc, kmp_int32 gtid );
2968 
2969 #endif /* KMP_GOMP_COMPAT */
2970 
2971 
2972 extern kmp_uint32 __kmp_eq_4( kmp_uint32 value, kmp_uint32 checker );
2973 extern kmp_uint32 __kmp_neq_4( kmp_uint32 value, kmp_uint32 checker );
2974 extern kmp_uint32 __kmp_lt_4( kmp_uint32 value, kmp_uint32 checker );
2975 extern kmp_uint32 __kmp_ge_4( kmp_uint32 value, kmp_uint32 checker );
2976 extern kmp_uint32 __kmp_le_4( kmp_uint32 value, kmp_uint32 checker );
2977 extern kmp_uint32 __kmp_wait_yield_4( kmp_uint32 volatile * spinner, kmp_uint32 checker, kmp_uint32 (*pred) (kmp_uint32, kmp_uint32), void * obj );
2978 extern void __kmp_wait_yield_4_ptr( void * spinner, kmp_uint32 checker, kmp_uint32 (* pred)( void *, kmp_uint32 ), void * obj );
2979 
2980 class kmp_flag_32;
2981 class kmp_flag_64;
2982 class kmp_flag_oncore;
2983 extern void __kmp_wait_64(kmp_info_t *this_thr, kmp_flag_64 *flag, int final_spin
2984 #if USE_ITT_BUILD
2985  , void * itt_sync_obj
2986 #endif
2987  );
2988 extern void __kmp_release_64(kmp_flag_64 *flag);
2989 
2990 extern void __kmp_infinite_loop( void );
2991 
2992 extern void __kmp_cleanup( void );
2993 
2994 #if KMP_HANDLE_SIGNALS
2995  extern int __kmp_handle_signals;
2996  extern void __kmp_install_signals( int parallel_init );
2997  extern void __kmp_remove_signals( void );
2998 #endif
2999 
3000 extern void __kmp_clear_system_time( void );
3001 extern void __kmp_read_system_time( double *delta );
3002 
3003 extern void __kmp_check_stack_overlap( kmp_info_t *thr );
3004 
3005 extern void __kmp_expand_host_name( char *buffer, size_t size );
3006 extern void __kmp_expand_file_name( char *result, size_t rlen, char *pattern );
3007 
3008 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
3009 extern void __kmp_initialize_system_tick( void ); /* Initialize timer tick value */
3010 #endif
3011 
3012 extern void __kmp_runtime_initialize( void ); /* machine specific initialization */
3013 extern void __kmp_runtime_destroy( void );
3014 
3015 #if KMP_AFFINITY_SUPPORTED
3016 extern char *__kmp_affinity_print_mask(char *buf, int buf_len, kmp_affin_mask_t *mask);
3017 extern void __kmp_affinity_initialize(void);
3018 extern void __kmp_affinity_uninitialize(void);
3019 extern void __kmp_affinity_set_init_mask(int gtid, int isa_root); /* set affinity according to KMP_AFFINITY */
3020 #if OMP_40_ENABLED
3021 extern void __kmp_affinity_set_place(int gtid);
3022 #endif
3023 extern void __kmp_affinity_determine_capable( const char *env_var );
3024 extern int __kmp_aux_set_affinity(void **mask);
3025 extern int __kmp_aux_get_affinity(void **mask);
3026 extern int __kmp_aux_get_affinity_max_proc();
3027 extern int __kmp_aux_set_affinity_mask_proc(int proc, void **mask);
3028 extern int __kmp_aux_unset_affinity_mask_proc(int proc, void **mask);
3029 extern int __kmp_aux_get_affinity_mask_proc(int proc, void **mask);
3030 extern void __kmp_balanced_affinity( int tid, int team_size );
3031 #endif /* KMP_AFFINITY_SUPPORTED */
3032 
3033 extern void __kmp_cleanup_hierarchy();
3034 extern void __kmp_get_hierarchy(kmp_uint32 nproc, kmp_bstate_t *thr_bar);
3035 
3036 #if KMP_USE_FUTEX
3037 
3038 extern int __kmp_futex_determine_capable( void );
3039 
3040 #endif // KMP_USE_FUTEX
3041 
3042 extern void __kmp_gtid_set_specific( int gtid );
3043 extern int __kmp_gtid_get_specific( void );
3044 
3045 extern double __kmp_read_cpu_time( void );
3046 
3047 extern int __kmp_read_system_info( struct kmp_sys_info *info );
3048 
3049 #if KMP_USE_MONITOR
3050 extern void __kmp_create_monitor( kmp_info_t *th );
3051 #endif
3052 
3053 extern void *__kmp_launch_thread( kmp_info_t *thr );
3054 
3055 extern void __kmp_create_worker( int gtid, kmp_info_t *th, size_t stack_size );
3056 
3057 #if KMP_OS_WINDOWS
3058 extern int __kmp_still_running(kmp_info_t *th);
3059 extern int __kmp_is_thread_alive( kmp_info_t * th, DWORD *exit_val );
3060 extern void __kmp_free_handle( kmp_thread_t tHandle );
3061 #endif
3062 
3063 #if KMP_USE_MONITOR
3064 extern void __kmp_reap_monitor( kmp_info_t *th );
3065 #endif
3066 extern void __kmp_reap_worker( kmp_info_t *th );
3067 extern void __kmp_terminate_thread( int gtid );
3068 
3069 extern void __kmp_suspend_32( int th_gtid, kmp_flag_32 *flag );
3070 extern void __kmp_suspend_64( int th_gtid, kmp_flag_64 *flag );
3071 extern void __kmp_suspend_oncore( int th_gtid, kmp_flag_oncore *flag );
3072 extern void __kmp_resume_32( int target_gtid, kmp_flag_32 *flag );
3073 extern void __kmp_resume_64( int target_gtid, kmp_flag_64 *flag );
3074 extern void __kmp_resume_oncore( int target_gtid, kmp_flag_oncore *flag );
3075 
3076 extern void __kmp_elapsed( double * );
3077 extern void __kmp_elapsed_tick( double * );
3078 
3079 extern void __kmp_enable( int old_state );
3080 extern void __kmp_disable( int *old_state );
3081 
3082 extern void __kmp_thread_sleep( int millis );
3083 
3084 extern void __kmp_common_initialize( void );
3085 extern void __kmp_common_destroy( void );
3086 extern void __kmp_common_destroy_gtid( int gtid );
3087 
3088 #if KMP_OS_UNIX
3089 extern void __kmp_register_atfork( void );
3090 #endif
3091 extern void __kmp_suspend_initialize( void );
3092 extern void __kmp_suspend_uninitialize_thread( kmp_info_t *th );
3093 
3094 extern kmp_info_t * __kmp_allocate_thread( kmp_root_t *root,
3095  kmp_team_t *team, int tid);
3096 #if OMP_40_ENABLED
3097 extern kmp_team_t * __kmp_allocate_team( kmp_root_t *root, int new_nproc, int max_nproc,
3098 #if OMPT_SUPPORT
3099  ompt_parallel_id_t ompt_parallel_id,
3100 #endif
3101  kmp_proc_bind_t proc_bind,
3102  kmp_internal_control_t *new_icvs,
3103  int argc USE_NESTED_HOT_ARG(kmp_info_t *thr) );
3104 #else
3105 extern kmp_team_t * __kmp_allocate_team( kmp_root_t *root, int new_nproc, int max_nproc,
3106 #if OMPT_SUPPORT
3107  ompt_parallel_id_t ompt_parallel_id,
3108 #endif
3109  kmp_internal_control_t *new_icvs,
3110  int argc USE_NESTED_HOT_ARG(kmp_info_t *thr) );
3111 #endif // OMP_40_ENABLED
3112 extern void __kmp_free_thread( kmp_info_t * );
3113 extern void __kmp_free_team( kmp_root_t *, kmp_team_t * USE_NESTED_HOT_ARG(kmp_info_t *) );
3114 extern kmp_team_t * __kmp_reap_team( kmp_team_t * );
3115 
3116 /* ------------------------------------------------------------------------ */
3117 
3118 extern void __kmp_initialize_bget( kmp_info_t *th );
3119 extern void __kmp_finalize_bget( kmp_info_t *th );
3120 
3121 KMP_EXPORT void *kmpc_malloc( size_t size );
3122 KMP_EXPORT void *kmpc_aligned_malloc( size_t size, size_t alignment );
3123 KMP_EXPORT void *kmpc_calloc( size_t nelem, size_t elsize );
3124 KMP_EXPORT void *kmpc_realloc( void *ptr, size_t size );
3125 KMP_EXPORT void kmpc_free( void *ptr );
3126 
3127 /* ------------------------------------------------------------------------ */
3128 /* declarations for internal use */
3129 
3130 extern int __kmp_barrier( enum barrier_type bt, int gtid, int is_split,
3131  size_t reduce_size, void *reduce_data, void (*reduce)(void *, void *) );
3132 extern void __kmp_end_split_barrier ( enum barrier_type bt, int gtid );
3133 
3137 enum fork_context_e
3138 {
3139  fork_context_gnu,
3140  fork_context_intel,
3141  fork_context_last
3142 };
3143 extern int __kmp_fork_call( ident_t *loc, int gtid, enum fork_context_e fork_context,
3144  kmp_int32 argc,
3145 #if OMPT_SUPPORT
3146  void *unwrapped_task,
3147 #endif
3148  microtask_t microtask, launch_t invoker,
3149 /* TODO: revert workaround for Intel(R) 64 tracker #96 */
3150 #if (KMP_ARCH_ARM || KMP_ARCH_X86_64 || KMP_ARCH_AARCH64) && KMP_OS_LINUX
3151  va_list *ap
3152 #else
3153  va_list ap
3154 #endif
3155  );
3156 
3157 extern void __kmp_join_call( ident_t *loc, int gtid
3158 #if OMPT_SUPPORT
3159  , enum fork_context_e fork_context
3160 #endif
3161 #if OMP_40_ENABLED
3162  , int exit_teams = 0
3163 #endif
3164  );
3165 
3166 extern void __kmp_serialized_parallel(ident_t *id, kmp_int32 gtid);
3167 extern void __kmp_internal_fork( ident_t *id, int gtid, kmp_team_t *team );
3168 extern void __kmp_internal_join( ident_t *id, int gtid, kmp_team_t *team );
3169 extern int __kmp_invoke_task_func( int gtid );
3170 extern void __kmp_run_before_invoked_task( int gtid, int tid, kmp_info_t *this_thr, kmp_team_t *team );
3171 extern void __kmp_run_after_invoked_task( int gtid, int tid, kmp_info_t *this_thr, kmp_team_t *team );
3172 
3173 // should never have been exported
3174 KMP_EXPORT int __kmpc_invoke_task_func( int gtid );
3175 #if OMP_40_ENABLED
3176 extern int __kmp_invoke_teams_master( int gtid );
3177 extern void __kmp_teams_master( int gtid );
3178 #endif
3179 extern void __kmp_save_internal_controls( kmp_info_t * thread );
3180 extern void __kmp_user_set_library (enum library_type arg);
3181 extern void __kmp_aux_set_library (enum library_type arg);
3182 extern void __kmp_aux_set_stacksize( size_t arg);
3183 extern void __kmp_aux_set_blocktime (int arg, kmp_info_t *thread, int tid);
3184 extern void __kmp_aux_set_defaults( char const * str, int len );
3185 
3186 /* Functions called from __kmp_aux_env_initialize() in kmp_settings.cpp */
3187 void kmpc_set_blocktime (int arg);
3188 void ompc_set_nested( int flag );
3189 void ompc_set_dynamic( int flag );
3190 void ompc_set_num_threads( int arg );
3191 
3192 extern void __kmp_push_current_task_to_thread( kmp_info_t *this_thr,
3193  kmp_team_t *team, int tid );
3194 extern void __kmp_pop_current_task_from_thread( kmp_info_t *this_thr );
3195 extern kmp_task_t* __kmp_task_alloc( ident_t *loc_ref, kmp_int32 gtid,
3196  kmp_tasking_flags_t *flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds,
3197  kmp_routine_entry_t task_entry );
3198 extern void __kmp_init_implicit_task( ident_t *loc_ref, kmp_info_t *this_thr,
3199  kmp_team_t *team, int tid, int set_curr_task );
3200 extern void __kmp_finish_implicit_task(kmp_info_t *this_thr);
3201 extern void __kmp_free_implicit_task(kmp_info_t *this_thr);
3202 
3203 int __kmp_execute_tasks_32(kmp_info_t *thread, kmp_int32 gtid, kmp_flag_32 *flag, int final_spin,
3204  int *thread_finished,
3205 #if USE_ITT_BUILD
3206  void * itt_sync_obj,
3207 #endif /* USE_ITT_BUILD */
3208  kmp_int32 is_constrained);
3209 int __kmp_execute_tasks_64(kmp_info_t *thread, kmp_int32 gtid, kmp_flag_64 *flag, int final_spin,
3210  int *thread_finished,
3211 #if USE_ITT_BUILD
3212  void * itt_sync_obj,
3213 #endif /* USE_ITT_BUILD */
3214  kmp_int32 is_constrained);
3215 int __kmp_execute_tasks_oncore(kmp_info_t *thread, kmp_int32 gtid, kmp_flag_oncore *flag, int final_spin,
3216  int *thread_finished,
3217 #if USE_ITT_BUILD
3218  void * itt_sync_obj,
3219 #endif /* USE_ITT_BUILD */
3220  kmp_int32 is_constrained);
3221 
3222 extern void __kmp_free_task_team( kmp_info_t *thread, kmp_task_team_t *task_team );
3223 extern void __kmp_reap_task_teams( void );
3224 extern void __kmp_wait_to_unref_task_teams( void );
3225 extern void __kmp_task_team_setup ( kmp_info_t *this_thr, kmp_team_t *team, int always );
3226 extern void __kmp_task_team_sync ( kmp_info_t *this_thr, kmp_team_t *team );
3227 extern void __kmp_task_team_wait ( kmp_info_t *this_thr, kmp_team_t *team
3228 #if USE_ITT_BUILD
3229  , void * itt_sync_obj
3230 #endif /* USE_ITT_BUILD */
3231  , int wait=1
3232 );
3233 extern void __kmp_tasking_barrier( kmp_team_t *team, kmp_info_t *thread, int gtid );
3234 
3235 extern int __kmp_is_address_mapped( void *addr );
3236 extern kmp_uint64 __kmp_hardware_timestamp(void);
3237 
3238 #if KMP_OS_UNIX
3239 extern int __kmp_read_from_file( char const *path, char const *format, ... );
3240 #endif
3241 
3242 /* ------------------------------------------------------------------------ */
3243 //
3244 // Assembly routines that have no compiler intrinsic replacement
3245 //
3246 
3247 #if KMP_ARCH_X86 || KMP_ARCH_X86_64
3248 
3249 extern void __kmp_query_cpuid( kmp_cpuinfo_t *p );
3250 
3251 #define __kmp_load_mxcsr(p) _mm_setcsr(*(p))
3252 static inline void __kmp_store_mxcsr( kmp_uint32 *p ) { *p = _mm_getcsr(); }
3253 
3254 extern void __kmp_load_x87_fpu_control_word( kmp_int16 *p );
3255 extern void __kmp_store_x87_fpu_control_word( kmp_int16 *p );
3256 extern void __kmp_clear_x87_fpu_status_word();
3257 # define KMP_X86_MXCSR_MASK 0xffffffc0 /* ignore status flags (6 lsb) */
3258 
3259 #endif /* KMP_ARCH_X86 || KMP_ARCH_X86_64 */
3260 
3261 extern int __kmp_invoke_microtask( microtask_t pkfn, int gtid, int npr, int argc, void *argv[]
3262 #if OMPT_SUPPORT
3263  , void **exit_frame_ptr
3264 #endif
3265 );
3266 
3267 
3268 /* ------------------------------------------------------------------------ */
3269 
3270 KMP_EXPORT void __kmpc_begin ( ident_t *, kmp_int32 flags );
3271 KMP_EXPORT void __kmpc_end ( ident_t * );
3272 
3273 KMP_EXPORT void __kmpc_threadprivate_register_vec ( ident_t *, void * data, kmpc_ctor_vec ctor,
3274  kmpc_cctor_vec cctor, kmpc_dtor_vec dtor, size_t vector_length );
3275 KMP_EXPORT void __kmpc_threadprivate_register ( ident_t *, void * data, kmpc_ctor ctor, kmpc_cctor cctor, kmpc_dtor dtor );
3276 KMP_EXPORT void * __kmpc_threadprivate ( ident_t *, kmp_int32 global_tid, void * data, size_t size );
3277 
3278 KMP_EXPORT kmp_int32 __kmpc_global_thread_num ( ident_t * );
3279 KMP_EXPORT kmp_int32 __kmpc_global_num_threads ( ident_t * );
3280 KMP_EXPORT kmp_int32 __kmpc_bound_thread_num ( ident_t * );
3281 KMP_EXPORT kmp_int32 __kmpc_bound_num_threads ( ident_t * );
3282 
3283 KMP_EXPORT kmp_int32 __kmpc_ok_to_fork ( ident_t * );
3284 KMP_EXPORT void __kmpc_fork_call ( ident_t *, kmp_int32 nargs, kmpc_micro microtask, ... );
3285 
3286 KMP_EXPORT void __kmpc_serialized_parallel ( ident_t *, kmp_int32 global_tid );
3287 KMP_EXPORT void __kmpc_end_serialized_parallel ( ident_t *, kmp_int32 global_tid );
3288 
3289 KMP_EXPORT void __kmpc_flush ( ident_t *);
3290 KMP_EXPORT void __kmpc_barrier ( ident_t *, kmp_int32 global_tid );
3291 KMP_EXPORT kmp_int32 __kmpc_master ( ident_t *, kmp_int32 global_tid );
3292 KMP_EXPORT void __kmpc_end_master ( ident_t *, kmp_int32 global_tid );
3293 KMP_EXPORT void __kmpc_ordered ( ident_t *, kmp_int32 global_tid );
3294 KMP_EXPORT void __kmpc_end_ordered ( ident_t *, kmp_int32 global_tid );
3295 KMP_EXPORT void __kmpc_critical ( ident_t *, kmp_int32 global_tid, kmp_critical_name * );
3296 KMP_EXPORT void __kmpc_end_critical ( ident_t *, kmp_int32 global_tid, kmp_critical_name * );
3297 
3298 #if OMP_45_ENABLED
3299 KMP_EXPORT void __kmpc_critical_with_hint ( ident_t *, kmp_int32 global_tid, kmp_critical_name *, uintptr_t hint );
3300 #endif
3301 
3302 KMP_EXPORT kmp_int32 __kmpc_barrier_master ( ident_t *, kmp_int32 global_tid );
3303 KMP_EXPORT void __kmpc_end_barrier_master ( ident_t *, kmp_int32 global_tid );
3304 
3305 KMP_EXPORT kmp_int32 __kmpc_barrier_master_nowait ( ident_t *, kmp_int32 global_tid );
3306 
3307 KMP_EXPORT kmp_int32 __kmpc_single ( ident_t *, kmp_int32 global_tid );
3308 KMP_EXPORT void __kmpc_end_single ( ident_t *, kmp_int32 global_tid );
3309 
3310 KMP_EXPORT void KMPC_FOR_STATIC_INIT ( ident_t *loc, kmp_int32 global_tid, kmp_int32 schedtype, kmp_int32 *plastiter,
3311  kmp_int *plower, kmp_int *pupper, kmp_int *pstride, kmp_int incr, kmp_int chunk );
3312 
3313 KMP_EXPORT void __kmpc_for_static_fini ( ident_t *loc, kmp_int32 global_tid );
3314 
3315 KMP_EXPORT void __kmpc_copyprivate( ident_t *loc, kmp_int32 global_tid, size_t cpy_size, void *cpy_data, void(*cpy_func)(void*,void*), kmp_int32 didit );
3316 
3317 extern void KMPC_SET_NUM_THREADS ( int arg );
3318 extern void KMPC_SET_DYNAMIC ( int flag );
3319 extern void KMPC_SET_NESTED ( int flag );
3320 
3321 /* --------------------------------------------------------------------------- */
3322 
3323 /*
3324  * Taskq interface routines
3325  */
3326 
3327 KMP_EXPORT kmpc_thunk_t * __kmpc_taskq (ident_t *loc, kmp_int32 global_tid, kmpc_task_t taskq_task, size_t sizeof_thunk,
3328  size_t sizeof_shareds, kmp_int32 flags, kmpc_shared_vars_t **shareds);
3329 KMP_EXPORT void __kmpc_end_taskq (ident_t *loc, kmp_int32 global_tid, kmpc_thunk_t *thunk);
3330 KMP_EXPORT kmp_int32 __kmpc_task (ident_t *loc, kmp_int32 global_tid, kmpc_thunk_t *thunk);
3331 KMP_EXPORT void __kmpc_taskq_task (ident_t *loc, kmp_int32 global_tid, kmpc_thunk_t *thunk, kmp_int32 status);
3332 KMP_EXPORT void __kmpc_end_taskq_task (ident_t *loc, kmp_int32 global_tid, kmpc_thunk_t *thunk);
3333 KMP_EXPORT kmpc_thunk_t * __kmpc_task_buffer (ident_t *loc, kmp_int32 global_tid, kmpc_thunk_t *taskq_thunk, kmpc_task_t task);
3334 
3335 /* ------------------------------------------------------------------------ */
3336 
3337 /*
3338  * OMP 3.0 tasking interface routines
3339  */
3340 
3341 KMP_EXPORT kmp_int32
3342 __kmpc_omp_task( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task );
3343 KMP_EXPORT kmp_task_t*
3344 __kmpc_omp_task_alloc( ident_t *loc_ref, kmp_int32 gtid, kmp_int32 flags,
3345  size_t sizeof_kmp_task_t, size_t sizeof_shareds,
3346  kmp_routine_entry_t task_entry );
3347 KMP_EXPORT void
3348 __kmpc_omp_task_begin_if0( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * task );
3349 KMP_EXPORT void
3350 __kmpc_omp_task_complete_if0( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t *task );
3351 KMP_EXPORT kmp_int32
3352 __kmpc_omp_task_parts( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * new_task );
3353 KMP_EXPORT kmp_int32
3354 __kmpc_omp_taskwait( ident_t *loc_ref, kmp_int32 gtid );
3355 
3356 KMP_EXPORT kmp_int32
3357 __kmpc_omp_taskyield( ident_t *loc_ref, kmp_int32 gtid, int end_part );
3358 
3359 #if TASK_UNUSED
3360 void __kmpc_omp_task_begin( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t * task );
3361 void __kmpc_omp_task_complete( ident_t *loc_ref, kmp_int32 gtid, kmp_task_t *task );
3362 #endif // TASK_UNUSED
3363 
3364 /* ------------------------------------------------------------------------ */
3365 
3366 #if OMP_40_ENABLED
3367 
3368 KMP_EXPORT void __kmpc_taskgroup( ident_t * loc, int gtid );
3369 KMP_EXPORT void __kmpc_end_taskgroup( ident_t * loc, int gtid );
3370 
3371 KMP_EXPORT kmp_int32 __kmpc_omp_task_with_deps(
3372  ident_t *loc_ref, kmp_int32 gtid, kmp_task_t *new_task, kmp_int32 ndeps,
3373  kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias,
3374  kmp_depend_info_t *noalias_dep_list);
3375 KMP_EXPORT void __kmpc_omp_wait_deps(ident_t *loc_ref, kmp_int32 gtid,
3376  kmp_int32 ndeps,
3377  kmp_depend_info_t *dep_list,
3378  kmp_int32 ndeps_noalias,
3379  kmp_depend_info_t *noalias_dep_list);
3380 extern void __kmp_release_deps(kmp_int32 gtid, kmp_taskdata_t *task);
3381 extern void __kmp_dephash_free_entries(kmp_info_t *thread, kmp_dephash_t *h);
3382 extern void __kmp_dephash_free(kmp_info_t *thread, kmp_dephash_t *h);
3383 
3384 extern kmp_int32 __kmp_omp_task( kmp_int32 gtid, kmp_task_t * new_task, bool serialize_immediate );
3385 
3386 KMP_EXPORT kmp_int32 __kmpc_cancel(ident_t* loc_ref, kmp_int32 gtid, kmp_int32 cncl_kind);
3387 KMP_EXPORT kmp_int32 __kmpc_cancellationpoint(ident_t* loc_ref, kmp_int32 gtid, kmp_int32 cncl_kind);
3388 KMP_EXPORT kmp_int32 __kmpc_cancel_barrier(ident_t* loc_ref, kmp_int32 gtid);
3389 KMP_EXPORT int __kmp_get_cancellation_status(int cancel_kind);
3390 
3391 #if OMP_45_ENABLED
3392 
3393 KMP_EXPORT void __kmpc_proxy_task_completed( kmp_int32 gtid, kmp_task_t *ptask );
3394 KMP_EXPORT void __kmpc_proxy_task_completed_ooo ( kmp_task_t *ptask );
3395 KMP_EXPORT void __kmpc_taskloop(ident_t *loc, kmp_int32 gtid, kmp_task_t *task, kmp_int32 if_val,
3396  kmp_uint64 *lb, kmp_uint64 *ub, kmp_int64 st,
3397  kmp_int32 nogroup, kmp_int32 sched, kmp_uint64 grainsize, void * task_dup );
3398 #endif
3399 
3400 #endif
3401 
3402 
3403 /*
3404  * Lock interface routines (fast versions with gtid passed in)
3405  */
3406 KMP_EXPORT void __kmpc_init_lock( ident_t *loc, kmp_int32 gtid, void **user_lock );
3407 KMP_EXPORT void __kmpc_init_nest_lock( ident_t *loc, kmp_int32 gtid, void **user_lock );
3408 KMP_EXPORT void __kmpc_destroy_lock( ident_t *loc, kmp_int32 gtid, void **user_lock );
3409 KMP_EXPORT void __kmpc_destroy_nest_lock( ident_t *loc, kmp_int32 gtid, void **user_lock );
3410 KMP_EXPORT void __kmpc_set_lock( ident_t *loc, kmp_int32 gtid, void **user_lock );
3411 KMP_EXPORT void __kmpc_set_nest_lock( ident_t *loc, kmp_int32 gtid, void **user_lock );
3412 KMP_EXPORT void __kmpc_unset_lock( ident_t *loc, kmp_int32 gtid, void **user_lock );
3413 KMP_EXPORT void __kmpc_unset_nest_lock( ident_t *loc, kmp_int32 gtid, void **user_lock );
3414 KMP_EXPORT int __kmpc_test_lock( ident_t *loc, kmp_int32 gtid, void **user_lock );
3415 KMP_EXPORT int __kmpc_test_nest_lock( ident_t *loc, kmp_int32 gtid, void **user_lock );
3416 
3417 #if OMP_45_ENABLED
3418 KMP_EXPORT void __kmpc_init_lock_with_hint( ident_t *loc, kmp_int32 gtid, void **user_lock, uintptr_t hint );
3419 KMP_EXPORT void __kmpc_init_nest_lock_with_hint( ident_t *loc, kmp_int32 gtid, void **user_lock, uintptr_t hint );
3420 #endif
3421 
3422 /* ------------------------------------------------------------------------ */
3423 
3424 /*
3425  * Interface to fast scalable reduce methods routines
3426  */
3427 
3428 KMP_EXPORT kmp_int32 __kmpc_reduce_nowait( ident_t *loc, kmp_int32 global_tid,
3429  kmp_int32 num_vars, size_t reduce_size,
3430  void *reduce_data, void (*reduce_func)(void *lhs_data, void *rhs_data),
3431  kmp_critical_name *lck );
3432 KMP_EXPORT void __kmpc_end_reduce_nowait( ident_t *loc, kmp_int32 global_tid, kmp_critical_name *lck );
3433 KMP_EXPORT kmp_int32 __kmpc_reduce( ident_t *loc, kmp_int32 global_tid,
3434  kmp_int32 num_vars, size_t reduce_size,
3435  void *reduce_data, void (*reduce_func)(void *lhs_data, void *rhs_data),
3436  kmp_critical_name *lck );
3437 KMP_EXPORT void __kmpc_end_reduce( ident_t *loc, kmp_int32 global_tid, kmp_critical_name *lck );
3438 
3439 /*
3440  * internal fast reduction routines
3441  */
3442 
3443 extern PACKED_REDUCTION_METHOD_T
3444 __kmp_determine_reduction_method( ident_t *loc, kmp_int32 global_tid,
3445  kmp_int32 num_vars, size_t reduce_size,
3446  void *reduce_data, void (*reduce_func)(void *lhs_data, void *rhs_data),
3447  kmp_critical_name *lck );
3448 
3449 // this function is for testing set/get/determine reduce method
3450 KMP_EXPORT kmp_int32 __kmp_get_reduce_method( void );
3451 
3452 KMP_EXPORT kmp_uint64 __kmpc_get_taskid();
3453 KMP_EXPORT kmp_uint64 __kmpc_get_parent_taskid();
3454 
3455 // this function exported for testing of KMP_PLACE_THREADS functionality
3456 KMP_EXPORT void __kmpc_place_threads(int,int,int,int,int);
3457 
3458 /* ------------------------------------------------------------------------ */
3459 /* ------------------------------------------------------------------------ */
3460 
3461 // C++ port
3462 // missing 'extern "C"' declarations
3463 
3464 KMP_EXPORT kmp_int32 __kmpc_in_parallel( ident_t *loc );
3465 KMP_EXPORT void __kmpc_pop_num_threads( ident_t *loc, kmp_int32 global_tid );
3466 KMP_EXPORT void __kmpc_push_num_threads( ident_t *loc, kmp_int32 global_tid, kmp_int32 num_threads );
3467 
3468 #if OMP_40_ENABLED
3469 KMP_EXPORT void __kmpc_push_proc_bind( ident_t *loc, kmp_int32 global_tid, int proc_bind );
3470 KMP_EXPORT void __kmpc_push_num_teams( ident_t *loc, kmp_int32 global_tid, kmp_int32 num_teams, kmp_int32 num_threads );
3471 KMP_EXPORT void __kmpc_fork_teams(ident_t *loc, kmp_int32 argc, kmpc_micro microtask, ...);
3472 #endif
3473 #if OMP_45_ENABLED
3474 struct kmp_dim { // loop bounds info casted to kmp_int64
3475  kmp_int64 lo; // lower
3476  kmp_int64 up; // upper
3477  kmp_int64 st; // stride
3478 };
3479 KMP_EXPORT void __kmpc_doacross_init(ident_t *loc, kmp_int32 gtid, kmp_int32 num_dims, struct kmp_dim * dims);
3480 KMP_EXPORT void __kmpc_doacross_wait(ident_t *loc, kmp_int32 gtid, kmp_int64 *vec);
3481 KMP_EXPORT void __kmpc_doacross_post(ident_t *loc, kmp_int32 gtid, kmp_int64 *vec);
3482 KMP_EXPORT void __kmpc_doacross_fini(ident_t *loc, kmp_int32 gtid);
3483 #endif
3484 
3485 KMP_EXPORT void*
3486 __kmpc_threadprivate_cached( ident_t * loc, kmp_int32 global_tid,
3487  void * data, size_t size, void *** cache );
3488 
3489 // Symbols for MS mutual detection.
3490 extern int _You_must_link_with_exactly_one_OpenMP_library;
3491 extern int _You_must_link_with_Intel_OpenMP_library;
3492 #if KMP_OS_WINDOWS && ( KMP_VERSION_MAJOR > 4 )
3493  extern int _You_must_link_with_Microsoft_OpenMP_library;
3494 #endif
3495 
3496 
3497 // The routines below are not exported.
3498 // Consider making them 'static' in corresponding source files.
3499 void
3500 kmp_threadprivate_insert_private_data( int gtid, void *pc_addr, void *data_addr, size_t pc_size );
3501 struct private_common *
3502 kmp_threadprivate_insert( int gtid, void *pc_addr, void *data_addr, size_t pc_size );
3503 
3504 //
3505 // ompc_, kmpc_ entries moved from omp.h.
3506 //
3507 #if KMP_OS_WINDOWS
3508 # define KMPC_CONVENTION __cdecl
3509 #else
3510 # define KMPC_CONVENTION
3511 #endif
3512 
3513 #ifndef __OMP_H
3514 typedef enum omp_sched_t {
3515  omp_sched_static = 1,
3516  omp_sched_dynamic = 2,
3517  omp_sched_guided = 3,
3518  omp_sched_auto = 4
3519 } omp_sched_t;
3520 typedef void * kmp_affinity_mask_t;
3521 #endif
3522 
3523 KMP_EXPORT void KMPC_CONVENTION ompc_set_max_active_levels(int);
3524 KMP_EXPORT void KMPC_CONVENTION ompc_set_schedule(omp_sched_t, int);
3525 KMP_EXPORT int KMPC_CONVENTION ompc_get_ancestor_thread_num(int);
3526 KMP_EXPORT int KMPC_CONVENTION ompc_get_team_size(int);
3527 KMP_EXPORT int KMPC_CONVENTION kmpc_set_affinity_mask_proc(int, kmp_affinity_mask_t *);
3528 KMP_EXPORT int KMPC_CONVENTION kmpc_unset_affinity_mask_proc(int, kmp_affinity_mask_t *);
3529 KMP_EXPORT int KMPC_CONVENTION kmpc_get_affinity_mask_proc(int, kmp_affinity_mask_t *);
3530 
3531 KMP_EXPORT void KMPC_CONVENTION kmpc_set_stacksize(int);
3532 KMP_EXPORT void KMPC_CONVENTION kmpc_set_stacksize_s(size_t);
3533 KMP_EXPORT void KMPC_CONVENTION kmpc_set_library(int);
3534 KMP_EXPORT void KMPC_CONVENTION kmpc_set_defaults(char const *);
3535 KMP_EXPORT void KMPC_CONVENTION kmpc_set_disp_num_buffers(int);
3536 
3537 #ifdef __cplusplus
3538 }
3539 #endif
3540 
3541 #endif /* KMP_H */
3542 
KMP_EXPORT kmp_int32 __kmpc_master(ident_t *, kmp_int32 global_tid)
KMP_EXPORT kmp_int32 __kmpc_barrier_master(ident_t *, kmp_int32 global_tid)
kmp_int32 reserved_2
Definition: kmp.h:203
void __kmpc_dispatch_fini_4(ident_t *loc, kmp_int32 gtid)
KMP_EXPORT void __kmpc_end_single(ident_t *, kmp_int32 global_tid)
void(* kmpc_dtor)(void *)
Definition: kmp.h:1298
void __kmpc_dispatch_init_4(ident_t *loc, kmp_int32 gtid, enum sched_type schedule, kmp_int32 lb, kmp_int32 ub, kmp_int32 st, kmp_int32 chunk)
KMP_EXPORT kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars, size_t reduce_size, void *reduce_data, void(*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name *lck)
KMP_EXPORT kmp_int32 __kmpc_global_thread_num(ident_t *)
int __kmpc_dispatch_next_4u(ident_t *loc, kmp_int32 gtid, kmp_int32 *p_last, kmp_uint32 *p_lb, kmp_uint32 *p_ub, kmp_int32 *p_st)
void(* kmpc_dtor_vec)(void *, size_t)
Definition: kmp.h:1319
KMP_EXPORT void __kmpc_for_static_fini(ident_t *loc, kmp_int32 global_tid)
KMP_EXPORT void __kmpc_omp_wait_deps(ident_t *loc_ref, kmp_int32 gtid, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list)
kmp_int32 reserved_1
Definition: kmp.h:201
void *(* kmpc_ctor_vec)(void *, size_t)
Definition: kmp.h:1313
KMP_EXPORT void * __kmpc_threadprivate_cached(ident_t *loc, kmp_int32 global_tid, void *data, size_t size, void ***cache)
kmp_int32 reserved_3
Definition: kmp.h:208
void *(* kmpc_cctor_vec)(void *, void *, size_t)
Definition: kmp.h:1325
KMP_EXPORT void __kmpc_flush(ident_t *)
void __kmpc_dispatch_init_8u(ident_t *loc, kmp_int32 gtid, enum sched_type schedule, kmp_uint64 lb, kmp_uint64 ub, kmp_int64 st, kmp_int64 chunk)
KMP_EXPORT kmp_int32 __kmpc_single(ident_t *, kmp_int32 global_tid)
int __kmpc_dispatch_next_4(ident_t *loc, kmp_int32 gtid, kmp_int32 *p_last, kmp_int32 *p_lb, kmp_int32 *p_ub, kmp_int32 *p_st)
KMP_EXPORT void __kmpc_end(ident_t *)
KMP_EXPORT void __kmpc_end_ordered(ident_t *, kmp_int32 global_tid)
KMP_EXPORT void __kmpc_end_serialized_parallel(ident_t *, kmp_int32 global_tid)
void *(* kmpc_cctor)(void *, void *)
Definition: kmp.h:1303
KMP_EXPORT void __kmpc_threadprivate_register(ident_t *, void *data, kmpc_ctor ctor, kmpc_cctor cctor, kmpc_dtor dtor)
KMP_EXPORT kmp_int32 __kmpc_omp_task_with_deps(ident_t *loc_ref, kmp_int32 gtid, kmp_task_t *new_task, kmp_int32 ndeps, kmp_depend_info_t *dep_list, kmp_int32 ndeps_noalias, kmp_depend_info_t *noalias_dep_list)
KMP_EXPORT void __kmpc_begin(ident_t *, kmp_int32 flags)
KMP_EXPORT kmp_int32 __kmpc_bound_thread_num(ident_t *)
KMP_EXPORT kmp_int32 __kmpc_reduce_nowait(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_vars, size_t reduce_size, void *reduce_data, void(*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name *lck)
int __kmpc_dispatch_next_8(ident_t *loc, kmp_int32 gtid, kmp_int32 *p_last, kmp_int64 *p_lb, kmp_int64 *p_ub, kmp_int64 *p_st)
KMP_EXPORT void __kmpc_copyprivate(ident_t *loc, kmp_int32 global_tid, size_t cpy_size, void *cpy_data, void(*cpy_func)(void *, void *), kmp_int32 didit)
KMP_EXPORT void __kmpc_ordered(ident_t *, kmp_int32 global_tid)
KMP_EXPORT void __kmpc_critical(ident_t *, kmp_int32 global_tid, kmp_critical_name *)
Definition: kmp.h:200
KMP_EXPORT void __kmpc_end_barrier_master(ident_t *, kmp_int32 global_tid)
KMP_EXPORT void __kmpc_end_master(ident_t *, kmp_int32 global_tid)
KMP_EXPORT void __kmpc_push_num_threads(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_threads)
KMP_EXPORT void __kmpc_fork_teams(ident_t *loc, kmp_int32 argc, kmpc_micro microtask,...)
KMP_EXPORT kmp_int32 __kmpc_in_parallel(ident_t *loc)
KMP_EXPORT kmp_int32 __kmpc_ok_to_fork(ident_t *)
KMP_EXPORT kmp_int32 __kmpc_global_num_threads(ident_t *)
void __kmpc_dispatch_fini_8u(ident_t *loc, kmp_int32 gtid)
KMP_EXPORT kmp_int32 __kmpc_bound_num_threads(ident_t *)
KMP_EXPORT void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *lck)
void __kmpc_dispatch_fini_4u(ident_t *loc, kmp_int32 gtid)
KMP_EXPORT void __kmpc_barrier(ident_t *, kmp_int32 global_tid)
KMP_EXPORT void __kmpc_end_reduce_nowait(ident_t *loc, kmp_int32 global_tid, kmp_critical_name *lck)
KMP_EXPORT void __kmpc_end_critical(ident_t *, kmp_int32 global_tid, kmp_critical_name *)
void *(* kmpc_ctor)(void *)
Definition: kmp.h:1292
KMP_EXPORT void __kmpc_push_num_teams(ident_t *loc, kmp_int32 global_tid, kmp_int32 num_teams, kmp_int32 num_threads)
void __kmpc_dispatch_fini_8(ident_t *loc, kmp_int32 gtid)
void __kmpc_dispatch_init_4u(ident_t *loc, kmp_int32 gtid, enum sched_type schedule, kmp_uint32 lb, kmp_uint32 ub, kmp_int32 st, kmp_int32 chunk)
void(* kmpc_micro)(kmp_int32 *global_tid, kmp_int32 *bound_tid,...)
Definition: kmp.h:1276
KMP_EXPORT kmp_int32 __kmpc_barrier_master_nowait(ident_t *, kmp_int32 global_tid)
int __kmpc_dispatch_next_8u(ident_t *loc, kmp_int32 gtid, kmp_int32 *p_last, kmp_uint64 *p_lb, kmp_uint64 *p_ub, kmp_int64 *p_st)
KMP_EXPORT void __kmpc_serialized_parallel(ident_t *, kmp_int32 global_tid)
KMP_EXPORT void __kmpc_fork_call(ident_t *, kmp_int32 nargs, kmpc_micro microtask,...)
KMP_EXPORT void __kmpc_threadprivate_register_vec(ident_t *, void *data, kmpc_ctor_vec ctor, kmpc_cctor_vec cctor, kmpc_dtor_vec dtor, size_t vector_length)
sched_type
Definition: kmp.h:303
char const * psource
Definition: kmp.h:209
void __kmpc_dispatch_init_8(ident_t *loc, kmp_int32 gtid, enum sched_type schedule, kmp_int64 lb, kmp_int64 ub, kmp_int64 st, kmp_int64 chunk)
kmp_int32 flags
Definition: kmp.h:202
struct ident ident_t