1 /*
2 * Copyright (c) 2003, 2007-14 Matteo Frigo
3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18 *
19 */
20
21 #include "mpi-rdft2.h"
22
destroy(problem * ego_)23 static void destroy(problem *ego_)
24 {
25 problem_mpi_rdft2 *ego = (problem_mpi_rdft2 *) ego_;
26 XM(dtensor_destroy)(ego->sz);
27 MPI_Comm_free(&ego->comm);
28 X(ifree)(ego_);
29 }
30
hash(const problem * p_,md5 * m)31 static void hash(const problem *p_, md5 *m)
32 {
33 const problem_mpi_rdft2 *p = (const problem_mpi_rdft2 *) p_;
34 int i;
35 X(md5puts)(m, "mpi-rdft2");
36 X(md5int)(m, p->I == p->O);
37 /* don't include alignment -- may differ between processes
38 X(md5int)(m, X(ialignment_of)(p->I));
39 X(md5int)(m, X(ialignment_of)(p->O));
40 ... note that applicability of MPI plans does not depend
41 on alignment (although optimality may, in principle). */
42 XM(dtensor_md5)(m, p->sz);
43 X(md5INT)(m, p->vn);
44 X(md5int)(m, p->kind);
45 X(md5int)(m, p->flags);
46 MPI_Comm_size(p->comm, &i); X(md5int)(m, i);
47 A(XM(md5_equal)(*m, p->comm));
48 }
49
print(const problem * ego_,printer * p)50 static void print(const problem *ego_, printer *p)
51 {
52 const problem_mpi_rdft2 *ego = (const problem_mpi_rdft2 *) ego_;
53 int i;
54 p->print(p, "(mpi-rdft2 %d %d %d ",
55 ego->I == ego->O,
56 X(ialignment_of)(ego->I),
57 X(ialignment_of)(ego->O));
58 XM(dtensor_print)(ego->sz, p);
59 p->print(p, " %D %d %d", ego->vn, (int) ego->kind, ego->flags);
60 MPI_Comm_size(ego->comm, &i); p->print(p, " %d)", i);
61 }
62
zero(const problem * ego_)63 static void zero(const problem *ego_)
64 {
65 const problem_mpi_rdft2 *ego = (const problem_mpi_rdft2 *) ego_;
66 R *I = ego->I;
67 dtensor *sz;
68 INT i, N;
69 int my_pe;
70
71 sz = XM(dtensor_copy)(ego->sz);
72 sz->dims[sz->rnk - 1].n = sz->dims[sz->rnk - 1].n / 2 + 1;
73 MPI_Comm_rank(ego->comm, &my_pe);
74 N = 2 * ego->vn * XM(total_block)(sz, IB, my_pe);
75 XM(dtensor_destroy)(sz);
76 for (i = 0; i < N; ++i) I[i] = K(0.0);
77 }
78
79 static const problem_adt padt =
80 {
81 PROBLEM_MPI_RDFT2,
82 hash,
83 zero,
84 print,
85 destroy
86 };
87
XM(mkproblem_rdft2)88 problem *XM(mkproblem_rdft2)(const dtensor *sz, INT vn,
89 R *I, R *O,
90 MPI_Comm comm,
91 rdft_kind kind,
92 unsigned flags)
93 {
94 problem_mpi_rdft2 *ego =
95 (problem_mpi_rdft2 *)X(mkproblem)(sizeof(problem_mpi_rdft2), &padt);
96 int n_pes;
97
98 A(XM(dtensor_validp)(sz) && FINITE_RNK(sz->rnk) && sz->rnk > 1);
99 MPI_Comm_size(comm, &n_pes);
100 A(vn >= 0);
101 A(kind == R2HC || kind == HC2R);
102
103 /* enforce pointer equality if untainted pointers are equal */
104 if (UNTAINT(I) == UNTAINT(O))
105 I = O = JOIN_TAINT(I, O);
106
107 ego->sz = XM(dtensor_canonical)(sz, 0);
108 #ifdef FFTW_DEBUG
109 ego->sz->dims[sz->rnk - 1].n = sz->dims[sz->rnk - 1].n / 2 + 1;
110 A(n_pes >= XM(num_blocks_total)(ego->sz, IB)
111 && n_pes >= XM(num_blocks_total)(ego->sz, OB));
112 ego->sz->dims[sz->rnk - 1].n = sz->dims[sz->rnk - 1].n;
113 #endif
114
115 ego->vn = vn;
116 ego->I = I;
117 ego->O = O;
118 ego->kind = kind;
119
120 /* We only support TRANSPOSED_OUT for r2c and TRANSPOSED_IN for
121 c2r transforms. */
122
123 ego->flags = flags;
124
125 MPI_Comm_dup(comm, &ego->comm);
126
127 return &(ego->super);
128 }
129
XM(mkproblem_rdft2_d)130 problem *XM(mkproblem_rdft2_d)(dtensor *sz, INT vn,
131 R *I, R *O,
132 MPI_Comm comm,
133 rdft_kind kind,
134 unsigned flags)
135 {
136 problem *p = XM(mkproblem_rdft2)(sz, vn, I, O, comm, kind, flags);
137 XM(dtensor_destroy)(sz);
138 return p;
139 }
140