Mercurial > hg > sv-dependency-builds

annotate src/fftw-3.3.5/mpi/rdft2-serial.c @ 43:5ea0608b923f

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Current zlib source
author Chris Cannam
date Tue, 18 Oct 2016 14:33:52 +0100
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 /* "MPI" DFTs where all of the data is on one processor...just
Chris@42 22 call through to serial API. */
Chris@42 23
Chris@42 24 #include "mpi-rdft2.h"
Chris@42 25 #include "rdft.h"
Chris@42 26
Chris@42 27 typedef struct {
Chris@42 28 plan_mpi_rdft2 super;
Chris@42 29 plan *cld;
Chris@42 30 INT vn;
Chris@42 31 } P;
Chris@42 32
Chris@42 33 static void apply_r2c(const plan *ego_, R *I, R *O)
Chris@42 34 {
Chris@42 35 const P *ego = (const P *) ego_;
Chris@42 36 plan_rdft2 *cld;
Chris@42 37 cld = (plan_rdft2 *) ego->cld;
Chris@42 38 cld->apply(ego->cld, I, I+ego->vn, O, O+1);
Chris@42 39 }
Chris@42 40
Chris@42 41 static void apply_c2r(const plan *ego_, R *I, R *O)
Chris@42 42 {
Chris@42 43 const P *ego = (const P *) ego_;
Chris@42 44 plan_rdft2 *cld;
Chris@42 45 cld = (plan_rdft2 *) ego->cld;
Chris@42 46 cld->apply(ego->cld, O, O+ego->vn, I, I+1);
Chris@42 47 }
Chris@42 48
Chris@42 49 static void awake(plan *ego_, enum wakefulness wakefulness)
Chris@42 50 {
Chris@42 51 P *ego = (P *) ego_;
Chris@42 52 X(plan_awake)(ego->cld, wakefulness);
Chris@42 53 }
Chris@42 54
Chris@42 55 static void destroy(plan *ego_)
Chris@42 56 {
Chris@42 57 P *ego = (P *) ego_;
Chris@42 58 X(plan_destroy_internal)(ego->cld);
Chris@42 59 }
Chris@42 60
Chris@42 61 static void print(const plan *ego_, printer *p)
Chris@42 62 {
Chris@42 63 const P *ego = (const P *) ego_;
Chris@42 64 p->print(p, "(mpi-rdft2-serial %(%p%))", ego->cld);
Chris@42 65 }
Chris@42 66
Chris@42 67 int XM(rdft2_serial_applicable)(const problem_mpi_rdft2 *p)
Chris@42 68 {
Chris@42 69 return (1
Chris@42 70 && p->flags == 0 /* TRANSPOSED/SCRAMBLED_IN/OUT not supported */
Chris@42 71 && ((XM(is_local)(p->sz, IB) && XM(is_local)(p->sz, OB))
Chris@42 72 || p->vn == 0));
Chris@42 73 }
Chris@42 74
Chris@42 75 static plan *mkplan(const solver *ego, const problem *p_, planner *plnr)
Chris@42 76 {
Chris@42 77 const problem_mpi_rdft2 *p = (const problem_mpi_rdft2 *) p_;
Chris@42 78 P *pln;
Chris@42 79 plan *cld;
Chris@42 80 int my_pe;
Chris@42 81 R *r0, *r1, *cr, *ci;
Chris@42 82 static const plan_adt padt = {
Chris@42 83 XM(rdft2_solve), awake, print, destroy
Chris@42 84 };
Chris@42 85
Chris@42 86 UNUSED(ego);
Chris@42 87
Chris@42 88 /* check whether applicable: */
Chris@42 89 if (!XM(rdft2_serial_applicable)(p))
Chris@42 90 return (plan *) 0;
Chris@42 91
Chris@42 92 if (p->kind == R2HC) {
Chris@42 93 r1 = (r0 = p->I) + p->vn;
Chris@42 94 ci = (cr = p->O) + 1;
Chris@42 95 }
Chris@42 96 else {
Chris@42 97 r1 = (r0 = p->O) + p->vn;
Chris@42 98 ci = (cr = p->I) + 1;
Chris@42 99 }
Chris@42 100
Chris@42 101 MPI_Comm_rank(p->comm, &my_pe);
Chris@42 102 if (my_pe == 0 && p->vn > 0) {
Chris@42 103 INT ivs = 1 + (p->kind == HC2R), ovs = 1 + (p->kind == R2HC);
Chris@42 104 int i, rnk = p->sz->rnk;
Chris@42 105 tensor *sz = X(mktensor)(p->sz->rnk);
Chris@42 106 sz->dims[rnk - 1].is = sz->dims[rnk - 1].os = 2 * p->vn;
Chris@42 107 sz->dims[rnk - 1].n = p->sz->dims[rnk - 1].n / 2 + 1;
Chris@42 108 for (i = rnk - 1; i > 0; --i) {
Chris@42 109 sz->dims[i - 1].is = sz->dims[i - 1].os =
Chris@42 110 sz->dims[i].is * sz->dims[i].n;
Chris@42 111 sz->dims[i - 1].n = p->sz->dims[i - 1].n;
Chris@42 112 }
Chris@42 113 sz->dims[rnk - 1].n = p->sz->dims[rnk - 1].n;
Chris@42 114
Chris@42 115 cld = X(mkplan_d)(plnr,
Chris@42 116 X(mkproblem_rdft2_d)(sz,
Chris@42 117 X(mktensor_1d)(p->vn,ivs,ovs),
Chris@42 118 r0, r1, cr, ci, p->kind));
Chris@42 119 }
Chris@42 120 else { /* idle process: make nop plan */
Chris@42 121 cld = X(mkplan_d)(plnr,
Chris@42 122 X(mkproblem_rdft2_d)(X(mktensor_0d)(),
Chris@42 123 X(mktensor_1d)(0,0,0),
Chris@42 124 cr, ci, cr, ci, HC2R));
Chris@42 125 }
Chris@42 126 if (XM(any_true)(!cld, p->comm)) return (plan *) 0;
Chris@42 127
Chris@42 128 pln = MKPLAN_MPI_RDFT2(P, &padt, p->kind == R2HC ? apply_r2c : apply_c2r);
Chris@42 129 pln->cld = cld;
Chris@42 130 pln->vn = p->vn;
Chris@42 131 X(ops_cpy)(&cld->ops, &pln->super.super.ops);
Chris@42 132 return &(pln->super.super);
Chris@42 133 }
Chris@42 134
Chris@42 135 static solver *mksolver(void)
Chris@42 136 {
Chris@42 137 static const solver_adt sadt = { PROBLEM_MPI_RDFT2, mkplan, 0 };
Chris@42 138 return MKSOLVER(solver, &sadt);
Chris@42 139 }
Chris@42 140
Chris@42 141 void XM(rdft2_serial_register)(planner *p)
Chris@42 142 {
Chris@42 143 REGISTER_SOLVER(p, mksolver());
Chris@42 144 }