annotate src/fftw-3.3.5/mpi/rdft2-problem.c @ 84:08ae793730bd

Add null config files
author Chris Cannam
date Mon, 02 Mar 2020 14:03:47 +0000
parents 2cd0e3b3e1fd
children
rev   line source
Chris@42 1 /*
Chris@42 2 * Copyright (c) 2003, 2007-14 Matteo Frigo
Chris@42 3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
Chris@42 4 *
Chris@42 5 * This program is free software; you can redistribute it and/or modify
Chris@42 6 * it under the terms of the GNU General Public License as published by
Chris@42 7 * the Free Software Foundation; either version 2 of the License, or
Chris@42 8 * (at your option) any later version.
Chris@42 9 *
Chris@42 10 * This program is distributed in the hope that it will be useful,
Chris@42 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@42 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
Chris@42 13 * GNU General Public License for more details.
Chris@42 14 *
Chris@42 15 * You should have received a copy of the GNU General Public License
Chris@42 16 * along with this program; if not, write to the Free Software
Chris@42 17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
Chris@42 18 *
Chris@42 19 */
Chris@42 20
Chris@42 21 #include "mpi-rdft2.h"
Chris@42 22
Chris@42 23 static void destroy(problem *ego_)
Chris@42 24 {
Chris@42 25 problem_mpi_rdft2 *ego = (problem_mpi_rdft2 *) ego_;
Chris@42 26 XM(dtensor_destroy)(ego->sz);
Chris@42 27 MPI_Comm_free(&ego->comm);
Chris@42 28 X(ifree)(ego_);
Chris@42 29 }
Chris@42 30
Chris@42 31 static void hash(const problem *p_, md5 *m)
Chris@42 32 {
Chris@42 33 const problem_mpi_rdft2 *p = (const problem_mpi_rdft2 *) p_;
Chris@42 34 int i;
Chris@42 35 X(md5puts)(m, "mpi-rdft2");
Chris@42 36 X(md5int)(m, p->I == p->O);
Chris@42 37 /* don't include alignment -- may differ between processes
Chris@42 38 X(md5int)(m, X(ialignment_of)(p->I));
Chris@42 39 X(md5int)(m, X(ialignment_of)(p->O));
Chris@42 40 ... note that applicability of MPI plans does not depend
Chris@42 41 on alignment (although optimality may, in principle). */
Chris@42 42 XM(dtensor_md5)(m, p->sz);
Chris@42 43 X(md5INT)(m, p->vn);
Chris@42 44 X(md5int)(m, p->kind);
Chris@42 45 X(md5int)(m, p->flags);
Chris@42 46 MPI_Comm_size(p->comm, &i); X(md5int)(m, i);
Chris@42 47 A(XM(md5_equal)(*m, p->comm));
Chris@42 48 }
Chris@42 49
Chris@42 50 static void print(const problem *ego_, printer *p)
Chris@42 51 {
Chris@42 52 const problem_mpi_rdft2 *ego = (const problem_mpi_rdft2 *) ego_;
Chris@42 53 int i;
Chris@42 54 p->print(p, "(mpi-rdft2 %d %d %d ",
Chris@42 55 ego->I == ego->O,
Chris@42 56 X(ialignment_of)(ego->I),
Chris@42 57 X(ialignment_of)(ego->O));
Chris@42 58 XM(dtensor_print)(ego->sz, p);
Chris@42 59 p->print(p, " %D %d %d", ego->vn, (int) ego->kind, ego->flags);
Chris@42 60 MPI_Comm_size(ego->comm, &i); p->print(p, " %d)", i);
Chris@42 61 }
Chris@42 62
Chris@42 63 static void zero(const problem *ego_)
Chris@42 64 {
Chris@42 65 const problem_mpi_rdft2 *ego = (const problem_mpi_rdft2 *) ego_;
Chris@42 66 R *I = ego->I;
Chris@42 67 dtensor *sz;
Chris@42 68 INT i, N;
Chris@42 69 int my_pe;
Chris@42 70
Chris@42 71 sz = XM(dtensor_copy)(ego->sz);
Chris@42 72 sz->dims[sz->rnk - 1].n = sz->dims[sz->rnk - 1].n / 2 + 1;
Chris@42 73 MPI_Comm_rank(ego->comm, &my_pe);
Chris@42 74 N = 2 * ego->vn * XM(total_block)(sz, IB, my_pe);
Chris@42 75 XM(dtensor_destroy)(sz);
Chris@42 76 for (i = 0; i < N; ++i) I[i] = K(0.0);
Chris@42 77 }
Chris@42 78
Chris@42 79 static const problem_adt padt =
Chris@42 80 {
Chris@42 81 PROBLEM_MPI_RDFT2,
Chris@42 82 hash,
Chris@42 83 zero,
Chris@42 84 print,
Chris@42 85 destroy
Chris@42 86 };
Chris@42 87
Chris@42 88 problem *XM(mkproblem_rdft2)(const dtensor *sz, INT vn,
Chris@42 89 R *I, R *O,
Chris@42 90 MPI_Comm comm,
Chris@42 91 rdft_kind kind,
Chris@42 92 unsigned flags)
Chris@42 93 {
Chris@42 94 problem_mpi_rdft2 *ego =
Chris@42 95 (problem_mpi_rdft2 *)X(mkproblem)(sizeof(problem_mpi_rdft2), &padt);
Chris@42 96 int n_pes;
Chris@42 97
Chris@42 98 A(XM(dtensor_validp)(sz) && FINITE_RNK(sz->rnk) && sz->rnk > 1);
Chris@42 99 MPI_Comm_size(comm, &n_pes);
Chris@42 100 A(vn >= 0);
Chris@42 101 A(kind == R2HC || kind == HC2R);
Chris@42 102
Chris@42 103 /* enforce pointer equality if untainted pointers are equal */
Chris@42 104 if (UNTAINT(I) == UNTAINT(O))
Chris@42 105 I = O = JOIN_TAINT(I, O);
Chris@42 106
Chris@42 107 ego->sz = XM(dtensor_canonical)(sz, 0);
Chris@42 108 #ifdef FFTW_DEBUG
Chris@42 109 ego->sz->dims[sz->rnk - 1].n = sz->dims[sz->rnk - 1].n / 2 + 1;
Chris@42 110 A(n_pes >= XM(num_blocks_total)(ego->sz, IB)
Chris@42 111 && n_pes >= XM(num_blocks_total)(ego->sz, OB));
Chris@42 112 ego->sz->dims[sz->rnk - 1].n = sz->dims[sz->rnk - 1].n;
Chris@42 113 #endif
Chris@42 114
Chris@42 115 ego->vn = vn;
Chris@42 116 ego->I = I;
Chris@42 117 ego->O = O;
Chris@42 118 ego->kind = kind;
Chris@42 119
Chris@42 120 /* We only support TRANSPOSED_OUT for r2c and TRANSPOSED_IN for
Chris@42 121 c2r transforms. */
Chris@42 122
Chris@42 123 ego->flags = flags;
Chris@42 124
Chris@42 125 MPI_Comm_dup(comm, &ego->comm);
Chris@42 126
Chris@42 127 return &(ego->super);
Chris@42 128 }
Chris@42 129
Chris@42 130 problem *XM(mkproblem_rdft2_d)(dtensor *sz, INT vn,
Chris@42 131 R *I, R *O,
Chris@42 132 MPI_Comm comm,
Chris@42 133 rdft_kind kind,
Chris@42 134 unsigned flags)
Chris@42 135 {
Chris@42 136 problem *p = XM(mkproblem_rdft2)(sz, vn, I, O, comm, kind, flags);
Chris@42 137 XM(dtensor_destroy)(sz);
Chris@42 138 return p;
Chris@42 139 }