cannam@95: /*
cannam@95:  * Copyright (c) 2003, 2007-11 Matteo Frigo
cannam@95:  * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
cannam@95:  *
cannam@95:  * This program is free software; you can redistribute it and/or modify
cannam@95:  * it under the terms of the GNU General Public License as published by
cannam@95:  * the Free Software Foundation; either version 2 of the License, or
cannam@95:  * (at your option) any later version.
cannam@95:  *
cannam@95:  * This program is distributed in the hope that it will be useful,
cannam@95:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
cannam@95:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
cannam@95:  * GNU General Public License for more details.
cannam@95:  *
cannam@95:  * You should have received a copy of the GNU General Public License
cannam@95:  * along with this program; if not, write to the Free Software
cannam@95:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
cannam@95:  *
cannam@95:  */
cannam@95: 
cannam@95: /* openmp.c: thread spawning via OpenMP  */
cannam@95: 
cannam@95: #include "threads.h"
cannam@95: 
cannam@95: #if !defined(_OPENMP)
cannam@95: #error OpenMP enabled but not using an OpenMP compiler
cannam@95: #endif
cannam@95: 
cannam@95: int X(ithreads_init)(void)
cannam@95: {
cannam@95:      return 0; /* no error */
cannam@95: }
cannam@95: 
cannam@95: /* Distribute a loop from 0 to loopmax-1 over nthreads threads.
cannam@95:    proc(d) is called to execute a block of iterations from d->min
cannam@95:    to d->max-1.  d->thr_num indicate the number of the thread
cannam@95:    that is executing proc (from 0 to nthreads-1), and d->data is
cannam@95:    the same as the data parameter passed to X(spawn_loop).
cannam@95: 
cannam@95:    This function returns only after all the threads have completed. */
cannam@95: void X(spawn_loop)(int loopmax, int nthr, spawn_function proc, void *data)
cannam@95: {
cannam@95:      int block_size;
cannam@95:      spawn_data d;
cannam@95:      int i;
cannam@95: 
cannam@95:      A(loopmax >= 0);
cannam@95:      A(nthr > 0);
cannam@95:      A(proc);
cannam@95: 
cannam@95:      if (!loopmax) return;
cannam@95: 
cannam@95:      /* Choose the block size and number of threads in order to (1)
cannam@95:         minimize the critical path and (2) use the fewest threads that
cannam@95:         achieve the same critical path (to minimize overhead).
cannam@95:         e.g. if loopmax is 5 and nthr is 4, we should use only 3
cannam@95:         threads with block sizes of 2, 2, and 1. */
cannam@95:      block_size = (loopmax + nthr - 1) / nthr;
cannam@95:      nthr = (loopmax + block_size - 1) / block_size;
cannam@95: 
cannam@95:      THREAD_ON; /* prevent debugging mode from failing under threads */
cannam@95: #pragma omp parallel for private(d)
cannam@95:      for (i = 0; i < nthr; ++i) {
cannam@95: 	  d.max = (d.min = i * block_size) + block_size;
cannam@95: 	  if (d.max > loopmax)
cannam@95: 	       d.max = loopmax;
cannam@95: 	  d.thr_num = i;
cannam@95: 	  d.data = data;
cannam@95: 	  proc(&d);
cannam@95:      }
cannam@95:      THREAD_OFF; /* prevent debugging mode from failing under threads */
cannam@95: }
cannam@95: 
cannam@95: void X(threads_cleanup)(void)
cannam@95: {
cannam@95: }