Chris@10
|
1 /*
|
Chris@10
|
2 * Copyright (c) 2003, 2007-11 Matteo Frigo
|
Chris@10
|
3 * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
|
Chris@10
|
4 *
|
Chris@10
|
5 * This program is free software; you can redistribute it and/or modify
|
Chris@10
|
6 * it under the terms of the GNU General Public License as published by
|
Chris@10
|
7 * the Free Software Foundation; either version 2 of the License, or
|
Chris@10
|
8 * (at your option) any later version.
|
Chris@10
|
9 *
|
Chris@10
|
10 * This program is distributed in the hope that it will be useful,
|
Chris@10
|
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
Chris@10
|
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
Chris@10
|
13 * GNU General Public License for more details.
|
Chris@10
|
14 *
|
Chris@10
|
15 * You should have received a copy of the GNU General Public License
|
Chris@10
|
16 * along with this program; if not, write to the Free Software
|
Chris@10
|
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
Chris@10
|
18 *
|
Chris@10
|
19 */
|
Chris@10
|
20
|
Chris@10
|
21 /* "MPI" DFTs where all of the data is on one processor...just
|
Chris@10
|
22 call through to serial API. */
|
Chris@10
|
23
|
Chris@10
|
24 #include "mpi-dft.h"
|
Chris@10
|
25 #include "dft.h"
|
Chris@10
|
26
|
Chris@10
|
27 typedef struct {
|
Chris@10
|
28 plan_mpi_dft super;
|
Chris@10
|
29 plan *cld;
|
Chris@10
|
30 INT roff, ioff;
|
Chris@10
|
31 } P;
|
Chris@10
|
32
|
Chris@10
|
33 static void apply(const plan *ego_, R *I, R *O)
|
Chris@10
|
34 {
|
Chris@10
|
35 const P *ego = (const P *) ego_;
|
Chris@10
|
36 plan_dft *cld;
|
Chris@10
|
37 INT roff = ego->roff, ioff = ego->ioff;
|
Chris@10
|
38 cld = (plan_dft *) ego->cld;
|
Chris@10
|
39 cld->apply(ego->cld, I+roff, I+ioff, O+roff, O+ioff);
|
Chris@10
|
40 }
|
Chris@10
|
41
|
Chris@10
|
42 static void awake(plan *ego_, enum wakefulness wakefulness)
|
Chris@10
|
43 {
|
Chris@10
|
44 P *ego = (P *) ego_;
|
Chris@10
|
45 X(plan_awake)(ego->cld, wakefulness);
|
Chris@10
|
46 }
|
Chris@10
|
47
|
Chris@10
|
48 static void destroy(plan *ego_)
|
Chris@10
|
49 {
|
Chris@10
|
50 P *ego = (P *) ego_;
|
Chris@10
|
51 X(plan_destroy_internal)(ego->cld);
|
Chris@10
|
52 }
|
Chris@10
|
53
|
Chris@10
|
54 static void print(const plan *ego_, printer *p)
|
Chris@10
|
55 {
|
Chris@10
|
56 const P *ego = (const P *) ego_;
|
Chris@10
|
57 p->print(p, "(mpi-dft-serial %(%p%))", ego->cld);
|
Chris@10
|
58 }
|
Chris@10
|
59
|
Chris@10
|
60 int XM(dft_serial_applicable)(const problem_mpi_dft *p)
|
Chris@10
|
61 {
|
Chris@10
|
62 return (1
|
Chris@10
|
63 && p->flags == 0 /* TRANSPOSED/SCRAMBLED_IN/OUT not supported */
|
Chris@10
|
64 && ((XM(is_local)(p->sz, IB) && XM(is_local)(p->sz, OB))
|
Chris@10
|
65 || p->vn == 0));
|
Chris@10
|
66 }
|
Chris@10
|
67
|
Chris@10
|
68 static plan *mkplan(const solver *ego, const problem *p_, planner *plnr)
|
Chris@10
|
69 {
|
Chris@10
|
70 const problem_mpi_dft *p = (const problem_mpi_dft *) p_;
|
Chris@10
|
71 P *pln;
|
Chris@10
|
72 plan *cld;
|
Chris@10
|
73 int my_pe;
|
Chris@10
|
74 R *ri, *ii, *ro, *io;
|
Chris@10
|
75 static const plan_adt padt = {
|
Chris@10
|
76 XM(dft_solve), awake, print, destroy
|
Chris@10
|
77 };
|
Chris@10
|
78
|
Chris@10
|
79 UNUSED(ego);
|
Chris@10
|
80
|
Chris@10
|
81 /* check whether applicable: */
|
Chris@10
|
82 if (!XM(dft_serial_applicable)(p))
|
Chris@10
|
83 return (plan *) 0;
|
Chris@10
|
84
|
Chris@10
|
85 X(extract_reim)(p->sign, p->I, &ri, &ii);
|
Chris@10
|
86 X(extract_reim)(p->sign, p->O, &ro, &io);
|
Chris@10
|
87
|
Chris@10
|
88 MPI_Comm_rank(p->comm, &my_pe);
|
Chris@10
|
89 if (my_pe == 0 && p->vn > 0) {
|
Chris@10
|
90 int i, rnk = p->sz->rnk;
|
Chris@10
|
91 tensor *sz = X(mktensor)(p->sz->rnk);
|
Chris@10
|
92 sz->dims[rnk - 1].is = sz->dims[rnk - 1].os = 2 * p->vn;
|
Chris@10
|
93 sz->dims[rnk - 1].n = p->sz->dims[rnk - 1].n;
|
Chris@10
|
94 for (i = rnk - 1; i > 0; --i) {
|
Chris@10
|
95 sz->dims[i - 1].is = sz->dims[i - 1].os =
|
Chris@10
|
96 sz->dims[i].is * sz->dims[i].n;
|
Chris@10
|
97 sz->dims[i - 1].n = p->sz->dims[i - 1].n;
|
Chris@10
|
98 }
|
Chris@10
|
99
|
Chris@10
|
100 cld = X(mkplan_d)(plnr,
|
Chris@10
|
101 X(mkproblem_dft_d)(sz,
|
Chris@10
|
102 X(mktensor_1d)(p->vn, 2, 2),
|
Chris@10
|
103 ri, ii, ro, io));
|
Chris@10
|
104 }
|
Chris@10
|
105 else { /* idle process: make nop plan */
|
Chris@10
|
106 cld = X(mkplan_d)(plnr,
|
Chris@10
|
107 X(mkproblem_dft_d)(X(mktensor_0d)(),
|
Chris@10
|
108 X(mktensor_1d)(0,0,0),
|
Chris@10
|
109 ri, ii, ro, io));
|
Chris@10
|
110 }
|
Chris@10
|
111 if (XM(any_true)(!cld, p->comm)) return (plan *) 0;
|
Chris@10
|
112
|
Chris@10
|
113 pln = MKPLAN_MPI_DFT(P, &padt, apply);
|
Chris@10
|
114 pln->cld = cld;
|
Chris@10
|
115 pln->roff = ro - p->O;
|
Chris@10
|
116 pln->ioff = io - p->O;
|
Chris@10
|
117 X(ops_cpy)(&cld->ops, &pln->super.super.ops);
|
Chris@10
|
118 return &(pln->super.super);
|
Chris@10
|
119 }
|
Chris@10
|
120
|
Chris@10
|
121 static solver *mksolver(void)
|
Chris@10
|
122 {
|
Chris@10
|
123 static const solver_adt sadt = { PROBLEM_MPI_DFT, mkplan, 0 };
|
Chris@10
|
124 return MKSOLVER(solver, &sadt);
|
Chris@10
|
125 }
|
Chris@10
|
126
|
Chris@10
|
127 void XM(dft_serial_register)(planner *p)
|
Chris@10
|
128 {
|
Chris@10
|
129 REGISTER_SOLVER(p, mksolver());
|
Chris@10
|
130 }
|