include/thread_mpi/barrier.h

   1 /*
   2 This source code file is part of thread_mpi.
   3 Written by Sander Pronk, Erik Lindahl, and possibly others.
   4
   5 Copyright (c) 2009, Sander Pronk, Erik Lindahl.
   6 All rights reserved.
   7
   8 Redistribution and use in source and binary forms, with or without
   9 modification, are permitted provided that the following conditions are met:
  10 1) Redistributions of source code must retain the above copyright
  11    notice, this list of conditions and the following disclaimer.
  12 2) Redistributions in binary form must reproduce the above copyright
  13    notice, this list of conditions and the following disclaimer in the
  14    documentation and/or other materials provided with the distribution.
  15 3) Neither the name of the copyright holders nor the
  16    names of its contributors may be used to endorse or promote products
  17    derived from this software without specific prior written permission.
  18
  19 THIS SOFTWARE IS PROVIDED BY US ''AS IS'' AND ANY
  20 EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
  21 WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  22 DISCLAIMED. IN NO EVENT SHALL WE BE LIABLE FOR ANY
  23 DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
  24 (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  25 LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
  26 ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  27 (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  28 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  29
  30 If you want to redistribute modifications, please consider that
  31 scientific software is very special. Version control is crucial -
  32 bugs must be traceable. We will be happy to consider code for
  33 inclusion in the official distribution, but derived work should not
  34 be called official thread_mpi. Details are found in the README & COPYING
  35 files.
  36 */
  37
  38 #ifndef _TMPI_BARRIER_H_
  39 #define _TMPI_BARRIER_H_
  40
  41 #include "atomic.h"
  42 #include "wait.h"
  43
  44 /** Fast (possibly busy-wait-based) barrier type
  45  *
  46  *  This barrier has the same functionality as the standard
  47  *  tMPI_Thread_barrier_t, but since it is based on spinlocks that yield
  48  *  to the scheduler in case of waiting, it provides faster synchronization
  49  *  at the cost of busy-waiting, while still behaving relatively nicely
  50  *  to other processes/threads. This is therefore the preferred type of
  51  *  barrier for when waits are expected to be reasonably short.
  52  *
  53  *  Variables of this type should be initialized by calling
  54  *  tMPI_Barrier_init() to set the number of threads
  55  *  that should be synchronized.
  56  *
  57  * \see
  58  * - tMPI_Barrier_init
  59  * - tMPI_Barrier_wait
  60  */
  61 typedef struct tMPI_Barrier_t tMPI_Barrier_t;
  62 struct tMPI_Barrier_t
  63 {
  64     tMPI_Atomic_t     count;     /*!< Number of threads remaining     */
  65     int               threshold; /*!< Total number of threads         */
  66     volatile int      cycle;     /*!< Current cycle (alternating 0/1) */
  67     TMPI_YIELD_WAIT_DATA
  68 };
  69
  70
  71
  72 /** Initialize barrier
  73  *
  74  *  \param barrier  Pointer to _spinlock_ barrier. Note that this is not
  75  *                  the same datatype as the full, thread based, barrier.
  76  *  \param count    Number of threads to synchronize. All threads
  77  *                  will be released after \a count calls to
  78  *                  tMPI_Barrier_wait().
  79  */
  80 void tMPI_Barrier_init(tMPI_Barrier_t *barrier, int count);
  81
  82
  83 /** Perform yielding, busy-waiting barrier synchronization
  84   *
  85   *  This function blocks until it has been called N times,
  86   *  where N is the count value the barrier was initialized with.
  87   *  After N total calls all threads return. The barrier automatically
  88   *  cycles, and thus requires another N calls to unblock another time.
  89   *
  90   *  \param barrier  Pointer to previously created barrier.
  91   *
  92   *  \return The last thread returns -1, all the others 0.
  93   */
  94 int tMPI_Barrier_wait(tMPI_Barrier_t *barrier);
  95
  96
  97 #ifdef DOXYGEN
  98 /** Get the number of threads to synchronize for a barrier
  99   *
 100   *  This function returns the total number of threads the barrier
 101   *  synchronizes.
 102   *
 103   *  \param barrier  Pointer to barrier.
 104   *
 105   *  \return the number of threads to synchronize
 106   */
 107 int tMPI_Barrier_N(tMPI_Barrier_t *barrier);
 108 #else
 109 #define tMPI_Barrier_N(barrier)  ((barrier)->threshold)
 110 #endif
 111
 112 #endif