/**************************************************************************\
 *
 *  This file is part of the Coin 3D visualization library.
 *  Copyright (C) by Kongsberg Oil & Gas Technologies.
 *
 *  This library is free software; you can redistribute it and/or
 *  modify it under the terms of the GNU General Public License
 *  ("GPL") version 2 as published by the Free Software Foundation.
 *  See the file LICENSE.GPL at the root directory of this source
 *  distribution for additional information about the GNU GPL.
 *
 *  For using Coin with software that can not be combined with the GNU
 *  GPL, and for taking advantage of the additional benefits of our
 *  support services, please contact Kongsberg Oil & Gas Technologies
 *  about acquiring a Coin Professional Edition License.
 *
 *  See http://www.coin3d.org/ for more information.
 *
 *  Kongsberg Oil & Gas Technologies, Bygdoy Alle 5, 0257 Oslo, NORWAY.
 *  http://www.sim.no/  sales@sim.no  coin-support@coin3d.org
 *
\**************************************************************************/

#include <Inventor/C/threads/sched.h>

#include <assert.h>

#ifdef HAVE_CONFIG_H
#include <config.h>
#endif /* HAVE_CONFIG_H */

/* FIXME: This class has changed somewhat since Coin-2. Tag this so
   we can find it when writing Coin-2 -> Coin-3 API change documentation.
   kintel 20061124. */

/* FIXME: Evaluate if some of the new functionality can be ported back 
   to Coin-2. kintel 20061124. */

#ifndef HAVE_THREADS

/* FIXME: instead of disallowing the use of these completely when
   thread support is not available (that's why there are asserts
   within them), could we perhaps implement in such a manner that they
   still work, but within only the calling thread?

   20051202 mortene. */

cc_sched * cc_sched_construct(int numthreads) { assert(FALSE); return NULL; }
void cc_sched_destruct(cc_sched * sched) { assert(FALSE); }
void cc_sched_set_num_threads(cc_sched * sched, int num) { assert(FALSE); }
int cc_sched_get_num_threads(cc_sched * sched) { assert(FALSE); return 0; }
uint32_t cc_sched_schedule(cc_sched * sched, 
                           cc_sched_f * workfunc, void * closure,
                           float priority) { assert(FALSE); }
void cc_sched_wait_all(cc_sched * sched) { assert(FALSE); }
SbBool cc_sched_unschedule(cc_sched * sched, 
                           uint32_t schedid) { assert(FALSE); }
void cc_sched_set_num_allowed(cc_sched * sched, 
                              int num)  { assert(FALSE); }
void cc_sched_change_priority(cc_sched * sched, 
                              uint32_t schedid, 
                              float priority)  { assert(FALSE); }

#else /* HAVE_THREADS */

#include <stdlib.h>

#include <Inventor/C/errors/debugerror.h>
#include <Inventor/C/threads/mutex.h>
#include <Inventor/C/threads/thread.h>
#include <Inventor/C/threads/wpool.h>

#include "threads/schedp.h"

/* ********************************************************************** */

#ifdef __cplusplus
extern "C" {
#endif /* __cplusplus */

/* private methods */

static void sched_worker_entry_point(void * userdata);

typedef struct {
  cc_sched_f * workfunc;
  void * closure;
  float priority;
  uint32_t schedid;
} sched_item;

static int
sched_item_compare(void * o1, void * o2)
{
  sched_item * i1 = (sched_item *) o1;
  sched_item * i2 = (sched_item *) o2;
  
  float diff = i1->priority - i2->priority;
  if (diff > 0) return 1;
  else if (diff < 0) return -1;
  return 0;
}

/* assumes mutex is locked */
static SbBool
sched_try_trigger(cc_sched * sched)
{
  if (cc_wpool_try_begin(sched->pool, 1)) {
    cc_wpool_start_worker(sched->pool, sched_worker_entry_point, sched);
    cc_wpool_end(sched->pool);
    return TRUE;
  }
  return FALSE;
}

void
sched_worker_entry_point(void * userdata)
{
  sched_item * item;
  cc_sched * sched = (cc_sched *)userdata;

  cc_mutex_lock(sched->mutex);
  while (!cc_heap_empty(sched->itemheap) && 
         (sched->numallowed != 0 || sched->iswaitingall)) {
    item = (sched_item *)cc_heap_extract_top(sched->itemheap);
    cc_dict_remove(sched->schedid_dict, item->schedid);
    cc_mutex_unlock(sched->mutex);
    item->workfunc(item->closure);
    cc_mutex_lock(sched->mutex);
    cc_memalloc_deallocate(sched->itemalloc, (void *)item);
    if (sched->numallowed > 0) sched->numallowed--;
  }
  cc_mutex_unlock(sched->mutex);
}

/* ********************************************************************** */
/* public api */


/*!
  Construct a scheduler that uses \a numthreads threads.
*/
cc_sched *
cc_sched_construct(int numthreads)
{
  cc_sched * sched = (cc_sched *) malloc(sizeof(cc_sched));
  assert(sched);
  sched->pool = cc_wpool_construct(numthreads);
  sched->mutex = cc_mutex_construct();
 
  sched->itemheap = cc_heap_construct(64, sched_item_compare, TRUE);
  sched->itemalloc = cc_memalloc_construct(sizeof(sched_item));
  sched->schedid_dict = cc_dict_construct(64, 0.75f);
  sched->schedid_counter = 1;
  sched->iswaitingall = FALSE;
  sched->numallowed = -1; /* Unlimited */

  return sched;
}

/*!
  Destruct the scheduler.

  This method will block until all currently executing jobs have finished.
  Any remaining scheduled jobs will be cancelled.

  Note that this differs from Coin-2. To emulate Coin-2 behavior, call
  cc_sched_wait_all() before calling this method.
*/
void
cc_sched_destruct(cc_sched * sched)
{
  cc_sched_set_num_allowed(sched, 0); // Exit inner scheduler loop faster
  cc_wpool_wait_all(sched->pool); // Make sure all worker threads are finished

  cc_dict_destruct(sched->schedid_dict);
  cc_heap_destruct(sched->itemheap);
  cc_memalloc_destruct(sched->itemalloc);
  cc_mutex_destruct(sched->mutex);
  cc_wpool_destruct(sched->pool);
  free(sched);
}

/*!
  Set/change the number of threads used by the scheduler.
*/
void
cc_sched_set_num_threads(cc_sched * sched, int num)
{
  cc_sched_wait_all(sched);
  cc_wpool_set_num_workers(sched->pool, num);
}

/*!
  Returns the number of threads used by the scheduler.
*/
int
cc_sched_get_num_threads(cc_sched * sched)
{
  return cc_wpool_get_num_workers(sched->pool);
}

/*! 
  Schedule a new job. A thread calls \a workfunc with the \a closure
  argument when a thread becomes available. larger \a priority values will
  be scheduled first.

  Returns a schedid that can be used to unschedule the job. schedid is
  guaranteed to be != 0.

  Note that jobs are automatically unscheduled when triggered, just before
  calling the work function.
*/
uint32_t
cc_sched_schedule(cc_sched * sched,
                  cc_sched_f * workfunc, void * closure,
                  float priority)
{
  sched_item * item;

  cc_mutex_lock(sched->mutex);
  item = (sched_item *)cc_memalloc_allocate(sched->itemalloc);
  
  item->workfunc = workfunc;
  item->closure = closure;
  item->priority = priority;
  item->schedid = sched->schedid_counter++;
  // avoid schedid == 0
  if (item->schedid == 0) {
    item->schedid = sched->schedid_counter++;
  }
  cc_heap_add(sched->itemheap, (void *)item);
  cc_dict_put(sched->schedid_dict, item->schedid, (void *)item);
  if (cc_dict_get_num_elements(sched->schedid_dict) == 1) {
    sched_try_trigger(sched);
  }

  cc_mutex_unlock(sched->mutex);

  return item->schedid;
}

/*!
  Attempt to unschedule a job. \a schedid must be an id returned
  from cc_sched_schedule().

  Note that jobs are automatically unscheduled when triggered, just before
  calling the work function.

  Returns TRUE if job was successfully removed, FALSE if job wasn't found
  in the internal dict.
*/
SbBool
cc_sched_unschedule(cc_sched * sched, uint32_t schedid)
{
  SbBool didremove = FALSE;
  void * item = NULL;
  cc_mutex_lock(sched->mutex);

  if (cc_dict_get(sched->schedid_dict, schedid, &item)) {
    cc_heap_remove(sched->itemheap, item);
    cc_dict_remove(sched->schedid_dict, schedid);
    cc_memalloc_deallocate(sched->itemalloc, item);
    didremove = TRUE;
  }
  cc_mutex_unlock(sched->mutex);
  return didremove;
}

/*!
  Returns the number of remaining scheduled jobs, not counting jobs that
  are currently being executed (i.e. are in their workfuncs).
*/
int 
cc_sched_get_num_remaining(cc_sched * sched)
{
  int num;
  cc_mutex_lock(sched->mutex);
  num = (int)cc_heap_elements(sched->itemheap);
  cc_mutex_unlock(sched->mutex);
  return num;
}

/*!
  Wait for all scheduled jobs to finish.
*/
void
cc_sched_wait_all(cc_sched * sched)
{
  cc_mutex_lock(sched->mutex);
  sched->iswaitingall = TRUE;
  /* Make sure all workers are doing something */
  while (!cc_heap_empty(sched->itemheap) && sched_try_trigger(sched)) { }

  cc_mutex_unlock(sched->mutex);
  cc_wpool_wait_all(sched->pool);

  cc_mutex_lock(sched->mutex);
  sched->iswaitingall = FALSE;
  cc_mutex_unlock(sched->mutex);
}

/*!
  Sets the number of allowed jobs per "batch". The scheduler will execute 
  at max the \e num jobs. To continue executing jobs, call this function again.
  This is typically done to limit the number of I/O or CPU intensive jobs to
  a few per frame to avoid starving the main thread.

  If \e num is -1 (the default), the number of jobs per batch is not limited.
*/
void 
cc_sched_set_num_allowed(cc_sched * sched, int num)
{
  cc_mutex_lock(sched->mutex);
  sched->numallowed = num;
  sched_try_trigger(sched);
  cc_mutex_unlock(sched->mutex);
}

/*!
  Changes the priority of the given scheduled item.
*/
void 
cc_sched_change_priority(cc_sched * sched, 
                         uint32_t schedid, float priority)
{
  void * item;
  cc_mutex_lock(sched->mutex);
  
  if (cc_dict_get(sched->schedid_dict, schedid, &item)) {
    cc_heap_remove(sched->itemheap, item);
    ((sched_item *)item)->priority = priority;
    cc_heap_add(sched->itemheap, item);
  }

  cc_mutex_unlock(sched->mutex);
}

/* ********************************************************************** */

#ifdef __cplusplus
} /* extern "C" */
#endif /* __cplusplus */

#endif /* HAVE_THREADS */