Chris@10: /*
Chris@10:  * Copyright (c) 2003, 2007-11 Matteo Frigo
Chris@10:  * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
Chris@10:  *
Chris@10:  * This program is free software; you can redistribute it and/or modify
Chris@10:  * it under the terms of the GNU General Public License as published by
Chris@10:  * the Free Software Foundation; either version 2 of the License, or
Chris@10:  * (at your option) any later version.
Chris@10:  *
Chris@10:  * This program is distributed in the hope that it will be useful,
Chris@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@10:  * GNU General Public License for more details.
Chris@10:  *
Chris@10:  * You should have received a copy of the GNU General Public License
Chris@10:  * along with this program; if not, write to the Free Software
Chris@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@10:  *
Chris@10:  */
Chris@10: 
Chris@10: /* openmp.c: thread spawning via OpenMP  */
Chris@10: 
Chris@10: #include "threads.h"
Chris@10: 
Chris@10: #if !defined(_OPENMP)
Chris@10: #error OpenMP enabled but not using an OpenMP compiler
Chris@10: #endif
Chris@10: 
Chris@10: int X(ithreads_init)(void)
Chris@10: {
Chris@10:      return 0; /* no error */
Chris@10: }
Chris@10: 
Chris@10: /* Distribute a loop from 0 to loopmax-1 over nthreads threads.
Chris@10:    proc(d) is called to execute a block of iterations from d->min
Chris@10:    to d->max-1.  d->thr_num indicate the number of the thread
Chris@10:    that is executing proc (from 0 to nthreads-1), and d->data is
Chris@10:    the same as the data parameter passed to X(spawn_loop).
Chris@10: 
Chris@10:    This function returns only after all the threads have completed. */
Chris@10: void X(spawn_loop)(int loopmax, int nthr, spawn_function proc, void *data)
Chris@10: {
Chris@10:      int block_size;
Chris@10:      spawn_data d;
Chris@10:      int i;
Chris@10: 
Chris@10:      A(loopmax >= 0);
Chris@10:      A(nthr > 0);
Chris@10:      A(proc);
Chris@10: 
Chris@10:      if (!loopmax) return;
Chris@10: 
Chris@10:      /* Choose the block size and number of threads in order to (1)
Chris@10:         minimize the critical path and (2) use the fewest threads that
Chris@10:         achieve the same critical path (to minimize overhead).
Chris@10:         e.g. if loopmax is 5 and nthr is 4, we should use only 3
Chris@10:         threads with block sizes of 2, 2, and 1. */
Chris@10:      block_size = (loopmax + nthr - 1) / nthr;
Chris@10:      nthr = (loopmax + block_size - 1) / block_size;
Chris@10: 
Chris@10:      THREAD_ON; /* prevent debugging mode from failing under threads */
Chris@10: #pragma omp parallel for private(d)
Chris@10:      for (i = 0; i < nthr; ++i) {
Chris@10: 	  d.max = (d.min = i * block_size) + block_size;
Chris@10: 	  if (d.max > loopmax)
Chris@10: 	       d.max = loopmax;
Chris@10: 	  d.thr_num = i;
Chris@10: 	  d.data = data;
Chris@10: 	  proc(&d);
Chris@10:      }
Chris@10:      THREAD_OFF; /* prevent debugging mode from failing under threads */
Chris@10: }
Chris@10: 
Chris@10: void X(threads_cleanup)(void)
Chris@10: {
Chris@10: }