/* Copyright (C) 2005-2026 Free Software Foundation, Inc.   Contributed by Richard Henderson <rth@redhat.com>.    This file is part of the GNU Offloading and Multi Processing Library   (libgomp).    Libgomp is free software; you can redistribute it and/or modify it   under the terms of the GNU General Public License as published by   the Free Software Foundation; either version 3, or (at your option)   any later version.    Libgomp is distributed in the hope that it will be useful, but WITHOUT ANY   WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS   FOR A PARTICULAR PURPOSE.  See the GNU General Public License for   more details.    Under Section 7 of GPL version 3, you are granted additional   permissions described in the GCC Runtime Library Exception, version   3.1, as published by the Free Software Foundation.    You should have received a copy of the GNU General Public License and   a copy of the GCC Runtime Library Exception along with this program;   see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see   <http://www.gnu.org/licenses/>.  */ /* This file handles the LOOP (FOR/DO) construct.  */ #include <limits.h>#include <stdlib.h>#include <string.h>#include "libgomp.h"  ialias (GOMP_loop_runtime_next)ialias_redirect (GOMP_taskgroup_reduction_register) /* Initialize the given work share construct from the given arguments.  */ static inline voidgomp_loop_init (struct gomp_work_share *ws, long start, long end, long incr,		enum gomp_schedule_type sched, long chunk_size){  ws->sched = sched;  ws->chunk_size = chunk_size;  /* Canonicalize loops that have zero iterations to ->next == ->end.  */  ws->end = ((incr > 0 && start > end) || (incr < 0 && start < end))	    ? start : end;  ws->incr = incr;  ws->next = start;  if (sched == GFS_DYNAMIC)    {      ws->chunk_size *= incr; #ifdef HAVE_SYNC_BUILTINS      {	/* For dynamic scheduling prepare things to make each iteration	   faster.  */	struct gomp_thread *thr = gomp_thread ();	struct gomp_team *team = thr->ts.team;	long nthreads = team ? team->nthreads : 1; 	if (__builtin_expect (incr > 0, 1))	  {	    /* Cheap overflow protection.  */	    if (__builtin_expect ((nthreads | ws->chunk_size)				  >= 1UL << (sizeof (long)					     * __CHAR_BIT__ / 2 - 1), 0))	      ws->mode = 0;	    else	      ws->mode = ws->end < (LONG_MAX				    - (nthreads + 1) * ws->chunk_size);	  }	/* Cheap overflow protection.  */	else if (__builtin_expect ((nthreads | -ws->chunk_size)				   >= 1UL << (sizeof (long)					      * __CHAR_BIT__ / 2 - 1), 0))	  ws->mode = 0;	else	  ws->mode = ws->end > (nthreads + 1) * -ws->chunk_size - LONG_MAX;      }#endif    }} /* The *_start routines are called when first encountering a loop construct   that is not bound directly to a parallel construct.  The first thread   that arrives will create the work-share construct; subsequent threads   will see the construct exists and allocate work from it.    START, END, INCR are the bounds of the loop; due to the restrictions of   OpenMP, these values must be the same in every thread.  This is not   verified (nor is it entirely verifiable, since START is not necessarily   retained intact in the work-share data structure).  CHUNK_SIZE is the   scheduling parameter; again this must be identical in all threads.    Returns true if there's any work for this thread to perform.  If so,   *ISTART and *IEND are filled with the bounds of the iteration block   allocated to this thread.  Returns false if all work was assigned to   other threads prior to this thread's arrival.  */ static boolgomp_loop_static_start (long start, long end, long incr, long chunk_size,			long *istart, long *iend){  struct gomp_thread *thr = gomp_thread ();   thr->ts.static_trip = 0;  if (gomp_work_share_start (0))    {      gomp_loop_init (thr->ts.work_share, start, end, incr,		      GFS_STATIC, chunk_size);      gomp_work_share_init_done ();    }   return !gomp_iter_static_next (istart, iend);} /* The current dynamic implementation is always monotonic.  The   entrypoints without nonmonotonic in them have to be always monotonic,   but the nonmonotonic ones could be changed to use work-stealing for   improved scalability.  */ static boolgomp_loop_dynamic_start (long start, long end, long incr, long chunk_size,			 long *istart, long *iend){  struct gomp_thread *thr = gomp_thread ();  bool ret;   if (gomp_work_share_start (0))    {      gomp_loop_init (thr->ts.work_share, start, end, incr,		      GFS_DYNAMIC, chunk_size);      gomp_work_share_init_done ();    } #ifdef HAVE_SYNC_BUILTINS  ret = gomp_iter_dynamic_next (istart, iend);#else  gomp_mutex_lock (&thr->ts.work_share->lock);  ret = gomp_iter_dynamic_next_locked (istart, iend);  gomp_mutex_unlock (&thr->ts.work_share->lock);#endif   return ret;} /* Similarly as for dynamic, though the question is how can the chunk sizes   be decreased without a central locking or atomics.  */ static boolgomp_loop_guided_start (long start, long end, long incr, long chunk_size,			long *istart, long *iend){  struct gomp_thread *thr = gomp_thread ();  bool ret;   if (gomp_work_share_start (0))    {      gomp_loop_init (thr->ts.work_share, start, end, incr,		      GFS_GUIDED, chunk_size);      gomp_work_share_init_done ();    } #ifdef HAVE_SYNC_BUILTINS  ret = gomp_iter_guided_next (istart, iend);#else  gomp_mutex_lock (&thr->ts.work_share->lock);  ret = gomp_iter_guided_next_locked (istart, iend);  gomp_mutex_unlock (&thr->ts.work_share->lock);#endif   return ret;} boolGOMP_loop_runtime_start (long start, long end, long incr,			 long *istart, long *iend){  struct gomp_task_icv *icv = gomp_icv (false);  switch (icv->run_sched_var & ~GFS_MONOTONIC)    {    case GFS_STATIC:      return gomp_loop_static_start (start, end, incr,				     icv->run_sched_chunk_size,				     istart, iend);    case GFS_DYNAMIC:      return gomp_loop_dynamic_start (start, end, incr,				      icv->run_sched_chunk_size,				      istart, iend);    case GFS_GUIDED:      return gomp_loop_guided_start (start, end, incr,				     icv->run_sched_chunk_size,				     istart, iend);    case GFS_AUTO:      /* For now map to schedule(static), later on we could play with feedback	 driven choice.  */      return gomp_loop_static_start (start, end, incr, 0, istart, iend);    default:      abort ();    }} static longgomp_adjust_sched (long sched, long *chunk_size){  sched &= ~GFS_MONOTONIC;  switch (sched)    {    case GFS_STATIC:    case GFS_DYNAMIC:    case GFS_GUIDED:      return sched;    /* GFS_RUNTIME is used for runtime schedule without monotonic       or nonmonotonic modifiers on the clause.       GFS_RUNTIME|GFS_MONOTONIC for runtime schedule with monotonic       modifier.  */    case GFS_RUNTIME:    /* GFS_AUTO is used for runtime schedule with nonmonotonic       modifier.  */    case GFS_AUTO:      {	struct gomp_task_icv *icv = gomp_icv (false);	sched = icv->run_sched_var & ~GFS_MONOTONIC;	switch (sched)	  {	  case GFS_STATIC:	  case GFS_DYNAMIC:	  case GFS_GUIDED:	    *chunk_size = icv->run_sched_chunk_size;	    break;	  case GFS_AUTO:	    sched = GFS_STATIC;	    *chunk_size = 0;	    break;	  default:	    abort ();	  }	return sched;      }    default:      abort ();    }} boolGOMP_loop_start (long start, long end, long incr, long sched,		 long chunk_size, long *istart, long *iend,		 uintptr_t *reductions, void **mem){  struct gomp_thread *thr = gomp_thread ();   thr->ts.static_trip = 0;  if (reductions)    gomp_workshare_taskgroup_start ();  if (gomp_work_share_start (0))    {      sched = gomp_adjust_sched (sched, &chunk_size);      gomp_loop_init (thr->ts.work_share, start, end, incr,		      sched, chunk_size);      if (reductions)	{	  GOMP_taskgroup_reduction_register (reductions);	  thr->task->taskgroup->workshare = true;	  thr->ts.work_share->task_reductions = reductions;	}      if (mem)	{	  uintptr_t size = (uintptr_t) *mem;#define INLINE_ORDERED_TEAM_IDS_OFF \  ((offsetof (struct gomp_work_share, inline_ordered_team_ids)		\    + __alignof__ (long long) - 1) & ~(__alignof__ (long long) - 1))	  if (sizeof (struct gomp_work_share)	      <= INLINE_ORDERED_TEAM_IDS_OFF	      || __alignof__ (struct gomp_work_share) < __alignof__ (long long)	      || size > (sizeof (struct gomp_work_share)			- INLINE_ORDERED_TEAM_IDS_OFF))	    *mem	      = (void *) (thr->ts.work_share->ordered_team_ids			  = gomp_malloc_cleared (size));	  else	    *mem = memset (((char *) thr->ts.work_share)			   + INLINE_ORDERED_TEAM_IDS_OFF, '\0', size);	}      gomp_work_share_init_done ();    }  else    {      if (reductions)	{	  uintptr_t *first_reductions = thr->ts.work_share->task_reductions;	  gomp_workshare_task_reduction_register (reductions,						  first_reductions);	}      if (mem)	{	  if ((offsetof (struct gomp_work_share, inline_ordered_team_ids)	       & (__alignof__ (long long) - 1)) == 0)	    *mem = (void *) thr->ts.work_share->ordered_team_ids;	  else	    {	      uintptr_t p = (uintptr_t) thr->ts.work_share->ordered_team_ids;	      p += __alignof__ (long long) - 1;	      p &= ~(__alignof__ (long long) - 1);	      *mem = (void *) p;	    }	}    }   if (!istart)    return true;  return ialias_call (GOMP_loop_runtime_next) (istart, iend);} /* The *_ordered_*_start routines are similar.  The only difference is that   this work-share construct is initialized to expect an ORDERED section.  */ static boolgomp_loop_ordered_static_start (long start, long end, long incr,				long chunk_size, long *istart, long *iend){  struct gomp_thread *thr = gomp_thread ();   thr->ts.static_trip = 0;  if (gomp_work_share_start (1))    {      gomp_loop_init (thr->ts.work_share, start, end, incr,		      GFS_STATIC, chunk_size);      gomp_ordered_static_init ();      gomp_work_share_init_done ();    }   return !gomp_iter_static_next (istart, iend);} static boolgomp_loop_ordered_dynamic_start (long start, long end, long incr,				 long chunk_size, long *istart, long *iend){  struct gomp_thread *thr = gomp_thread ();  bool ret;   if (gomp_work_share_start (1))    {      gomp_loop_init (thr->ts.work_share, start, end, incr,		      GFS_DYNAMIC, chunk_size);      gomp_mutex_lock (&thr->ts.work_share->lock);      gomp_work_share_init_done ();    }  else    gomp_mutex_lock (&thr->ts.work_share->lock);   ret = gomp_iter_dynamic_next_locked (istart, iend);  if (ret)    gomp_ordered_first ();  gomp_mutex_unlock (&thr->ts.work_share->lock);   return ret;} static boolgomp_loop_ordered_guided_start (long start, long end, long incr,				long chunk_size, long *istart, long *iend){  struct gomp_thread *thr = gomp_thread ();  bool ret;   if (gomp_work_share_start (1))    {      gomp_loop_init (thr->ts.work_share, start, end, incr,		      GFS_GUIDED, chunk_size);      gomp_mutex_lock (&thr->ts.work_share->lock);      gomp_work_share_init_done ();    }  else    gomp_mutex_lock (&thr->ts.work_share->lock);   ret = gomp_iter_guided_next_locked (istart, iend);  if (ret)    gomp_ordered_first ();  gomp_mutex_unlock (&thr->ts.work_share->lock);   return ret;} boolGOMP_loop_ordered_runtime_start (long start, long end, long incr,				 long *istart, long *iend){  struct gomp_task_icv *icv = gomp_icv (false);  switch (icv->run_sched_var & ~GFS_MONOTONIC)    {    case GFS_STATIC:      return gomp_loop_ordered_static_start (start, end, incr,					     icv->run_sched_chunk_size,					     istart, iend);    case GFS_DYNAMIC:      return gomp_loop_ordered_dynamic_start (start, end, incr,					      icv->run_sched_chunk_size,					      istart, iend);    case GFS_GUIDED:      return gomp_loop_ordered_guided_start (start, end, incr,					     icv->run_sched_chunk_size,					     istart, iend);    case GFS_AUTO:      /* For now map to schedule(static), later on we could play with feedback	 driven choice.  */      return gomp_loop_ordered_static_start (start, end, incr,					     0, istart, iend);    default:      abort ();    }} boolGOMP_loop_ordered_start (long start, long end, long incr, long sched,			 long chunk_size, long *istart, long *iend,			 uintptr_t *reductions, void **mem){  struct gomp_thread *thr = gomp_thread ();  size_t ordered = 1;  bool ret;   thr->ts.static_trip = 0;  if (reductions)    gomp_workshare_taskgroup_start ();  if (mem)    ordered += (uintptr_t) *mem;  if (gomp_work_share_start (ordered))    {      sched = gomp_adjust_sched (sched, &chunk_size);      gomp_loop_init (thr->ts.work_share, start, end, incr,		      sched, chunk_size);      if (reductions)	{	  GOMP_taskgroup_reduction_register (reductions);	  thr->task->taskgroup->workshare = true;	  thr->ts.work_share->task_reductions = reductions;	}      if (sched == GFS_STATIC)	gomp_ordered_static_init ();      else	gomp_mutex_lock (&thr->ts.work_share->lock);      gomp_work_share_init_done ();    }  else    {      if (reductions)	{	  uintptr_t *first_reductions = thr->ts.work_share->task_reductions;	  gomp_workshare_task_reduction_register (reductions,						  first_reductions);	}      sched = thr->ts.work_share->sched;      if (sched != GFS_STATIC)	gomp_mutex_lock (&thr->ts.work_share->lock);    }   if (mem)    {      uintptr_t p	= (uintptr_t) (thr->ts.work_share->ordered_team_ids		       + (thr->ts.team ? thr->ts.team->nthreads : 1));      p += __alignof__ (long long) - 1;      p &= ~(__alignof__ (long long) - 1);      *mem = (void *) p;    }   switch (sched)    {    case GFS_STATIC:    case GFS_AUTO:      return !gomp_iter_static_next (istart, iend);    case GFS_DYNAMIC:      ret = gomp_iter_dynamic_next_locked (istart, iend);      break;    case GFS_GUIDED:      ret = gomp_iter_guided_next_locked (istart, iend);      break;    default:      abort ();    }   if (ret)    gomp_ordered_first ();  gomp_mutex_unlock (&thr->ts.work_share->lock);  return ret;} /* The *_doacross_*_start routines are similar.  The only difference is that   this work-share construct is initialized to expect an ORDERED(N) - DOACROSS   section, and the worksharing loop iterates always from 0 to COUNTS[0] - 1   and other COUNTS array elements tell the library number of iterations   in the ordered inner loops.  */ static boolgomp_loop_doacross_static_start (unsigned ncounts, long *counts,				 long chunk_size, long *istart, long *iend){  struct gomp_thread *thr = gomp_thread ();

Lines 1–500 of 1188. Load more