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 /* "MPI" DFTs where all of the data is on one processor...just
22 call through to serial API. */
23
24 #include "mpi-dft.h"
25 #include "dft/dft.h"
26
27 typedef struct {
28 plan_mpi_dft super;
29 plan *cld;
30 INT roff, ioff;
31 } P;
32
apply(const plan * ego_,R * I,R * O)33 static void apply(const plan *ego_, R *I, R *O)
34 {
35 const P *ego = (const P *) ego_;
36 plan_dft *cld;
37 INT roff = ego->roff, ioff = ego->ioff;
38 cld = (plan_dft *) ego->cld;
39 cld->apply(ego->cld, I+roff, I+ioff, O+roff, O+ioff);
40 }
41
awake(plan * ego_,enum wakefulness wakefulness)42 static void awake(plan *ego_, enum wakefulness wakefulness)
43 {
44 P *ego = (P *) ego_;
45 X(plan_awake)(ego->cld, wakefulness);
46 }
47
destroy(plan * ego_)48 static void destroy(plan *ego_)
49 {
50 P *ego = (P *) ego_;
51 X(plan_destroy_internal)(ego->cld);
52 }
53
print(const plan * ego_,printer * p)54 static void print(const plan *ego_, printer *p)
55 {
56 const P *ego = (const P *) ego_;
57 p->print(p, "(mpi-dft-serial %(%p%))", ego->cld);
58 }
59
XM(dft_serial_applicable)60 int XM(dft_serial_applicable)(const problem_mpi_dft *p)
61 {
62 return (1
63 && p->flags == 0 /* TRANSPOSED/SCRAMBLED_IN/OUT not supported */
64 && ((XM(is_local)(p->sz, IB) && XM(is_local)(p->sz, OB))
65 || p->vn == 0));
66 }
67
mkplan(const solver * ego,const problem * p_,planner * plnr)68 static plan *mkplan(const solver *ego, const problem *p_, planner *plnr)
69 {
70 const problem_mpi_dft *p = (const problem_mpi_dft *) p_;
71 P *pln;
72 plan *cld;
73 int my_pe;
74 R *ri, *ii, *ro, *io;
75 static const plan_adt padt = {
76 XM(dft_solve), awake, print, destroy
77 };
78
79 UNUSED(ego);
80
81 /* check whether applicable: */
82 if (!XM(dft_serial_applicable)(p))
83 return (plan *) 0;
84
85 X(extract_reim)(p->sign, p->I, &ri, &ii);
86 X(extract_reim)(p->sign, p->O, &ro, &io);
87
88 MPI_Comm_rank(p->comm, &my_pe);
89 if (my_pe == 0 && p->vn > 0) {
90 int i, rnk = p->sz->rnk;
91 tensor *sz = X(mktensor)(p->sz->rnk);
92 sz->dims[rnk - 1].is = sz->dims[rnk - 1].os = 2 * p->vn;
93 sz->dims[rnk - 1].n = p->sz->dims[rnk - 1].n;
94 for (i = rnk - 1; i > 0; --i) {
95 sz->dims[i - 1].is = sz->dims[i - 1].os =
96 sz->dims[i].is * sz->dims[i].n;
97 sz->dims[i - 1].n = p->sz->dims[i - 1].n;
98 }
99
100 cld = X(mkplan_d)(plnr,
101 X(mkproblem_dft_d)(sz,
102 X(mktensor_1d)(p->vn, 2, 2),
103 ri, ii, ro, io));
104 }
105 else { /* idle process: make nop plan */
106 cld = X(mkplan_d)(plnr,
107 X(mkproblem_dft_d)(X(mktensor_0d)(),
108 X(mktensor_1d)(0,0,0),
109 ri, ii, ro, io));
110 }
111 if (XM(any_true)(!cld, p->comm)) return (plan *) 0;
112
113 pln = MKPLAN_MPI_DFT(P, &padt, apply);
114 pln->cld = cld;
115 pln->roff = ro - p->O;
116 pln->ioff = io - p->O;
117 X(ops_cpy)(&cld->ops, &pln->super.super.ops);
118 return &(pln->super.super);
119 }
120
mksolver(void)121 static solver *mksolver(void)
122 {
123 static const solver_adt sadt = { PROBLEM_MPI_DFT, mkplan, 0 };
124 return MKSOLVER(solver, &sadt);
125 }
126
XM(dft_serial_register)127 void XM(dft_serial_register)(planner *p)
128 {
129 REGISTER_SOLVER(p, mksolver());
130 }
131