annotate src/fftw-3.3.8/threads/openmp.c @ 82:d0c2a83c1364

Add FFTW 3.3.8 source, and a Linux build
author Chris Cannam
date Tue, 19 Nov 2019 14:52:55 +0000
parents
children
rev   line source
Chris@82 1 /*
Chris@82 2 * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@82 3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@82 4 *
Chris@82 5 * This program is free software; you can redistribute it and/or modify
Chris@82 6 * it under the terms of the GNU General Public License as published by
Chris@82 7 * the Free Software Foundation; either version 2 of the License, or
Chris@82 8 * (at your option) any later version.
Chris@82 9 *
Chris@82 10 * This program is distributed in the hope that it will be useful,
Chris@82 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@82 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
Chris@82 13 * GNU General Public License for more details.
Chris@82 14 *
Chris@82 15 * You should have received a copy of the GNU General Public License
Chris@82 16 * along with this program; if not, write to the Free Software
Chris@82 17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
Chris@82 18 *
Chris@82 19 */
Chris@82 20
Chris@82 21 /* openmp.c: thread spawning via OpenMP */
Chris@82 22
Chris@82 23 #include "threads/threads.h"
Chris@82 24
Chris@82 25 #if !defined(_OPENMP)
Chris@82 26 #error OpenMP enabled but not using an OpenMP compiler
Chris@82 27 #endif
Chris@82 28
Chris@82 29 int X(ithreads_init)(void)
Chris@82 30 {
Chris@82 31 return 0; /* no error */
Chris@82 32 }
Chris@82 33
Chris@82 34 /* Distribute a loop from 0 to loopmax-1 over nthreads threads.
Chris@82 35 proc(d) is called to execute a block of iterations from d->min
Chris@82 36 to d->max-1. d->thr_num indicate the number of the thread
Chris@82 37 that is executing proc (from 0 to nthreads-1), and d->data is
Chris@82 38 the same as the data parameter passed to X(spawn_loop).
Chris@82 39
Chris@82 40 This function returns only after all the threads have completed. */
Chris@82 41 void X(spawn_loop)(int loopmax, int nthr, spawn_function proc, void *data)
Chris@82 42 {
Chris@82 43 int block_size;
Chris@82 44 spawn_data d;
Chris@82 45 int i;
Chris@82 46
Chris@82 47 A(loopmax >= 0);
Chris@82 48 A(nthr > 0);
Chris@82 49 A(proc);
Chris@82 50
Chris@82 51 if (!loopmax) return;
Chris@82 52
Chris@82 53 /* Choose the block size and number of threads in order to (1)
Chris@82 54 minimize the critical path and (2) use the fewest threads that
Chris@82 55 achieve the same critical path (to minimize overhead).
Chris@82 56 e.g. if loopmax is 5 and nthr is 4, we should use only 3
Chris@82 57 threads with block sizes of 2, 2, and 1. */
Chris@82 58 block_size = (loopmax + nthr - 1) / nthr;
Chris@82 59 nthr = (loopmax + block_size - 1) / block_size;
Chris@82 60
Chris@82 61 #pragma omp parallel for private(d)
Chris@82 62 for (i = 0; i < nthr; ++i) {
Chris@82 63 d.max = (d.min = i * block_size) + block_size;
Chris@82 64 if (d.max > loopmax)
Chris@82 65 d.max = loopmax;
Chris@82 66 d.thr_num = i;
Chris@82 67 d.data = data;
Chris@82 68 proc(&d);
Chris@82 69 }
Chris@82 70 }
Chris@82 71
Chris@82 72 void X(threads_cleanup)(void)
Chris@82 73 {
Chris@82 74 }
Chris@82 75
Chris@82 76 /* FIXME [Matteo Frigo 2015-05-25] What does "thread-safe"
Chris@82 77 mean for openmp? */
Chris@82 78 void X(threads_register_planner_hooks)(void)
Chris@82 79 {
Chris@82 80 }